Role-based access control for UML: the TextUML-Cloudfier approach

Cloudfier is an approach for building business applications, and since role-based access control (RBAC) is such an important thing for any business application, Cloudfier is bound to provide support for modeling what users can do to application objects. Here is our plan.

[There are several proposals by third-parties on how to do security with UML (remember, TextUML is UML), but the OMG itself has not officially adopted any so far. So, since there is no standard and no clear 3rd-party winner, I decided I might as well make up my own approach, tailor-made for the needs of business applications.]

The gist of the idea:

  1. classes, attributes and operations may declare “access constraints”, which are UML constraints specialized on describing how accessible an element (in practice, classes, attributes and operations);
  2. those constraints allow specific capabilities (like create and destroy instances of a classes, read and write attribute values or associations links, and call operations);
  3. access constraints are defined for one or more user roles – one element may have multiple constraints, matching different roles;
  4. the constraint specification may be an expression that (typically) will take the current logged-in user/actor in account in order to decide access should be allowed.

Profile changes

TextUML models, by default, can count on a UML profile called mdd_extensions. This profile defines several extensions to the base UML standard, for things such as allowing one to mark a class as a test class, elements as debuggable, activities as closures, blocks as initialization blocks, type casts, attribute derivations etc. The mdd_extensions profile was enhanced to support the idea of role classes and access constraints

Role classes

A role class is a class that represents a role an actor can play. Here is the definition of the stereotype:


(* A role class is a class that represents the role for a user. *)
stereotype Role extends UML::Class
end;

Access constraints

Access constraints are constraints with the “Access” stereotype applied to them. The stereotype allows setting:

  • the roles affected – user roles that are covered by this constraint;
  • the capabilities allowed – what users with one of the roles can do. Values are values of the AccessCapability enumeration.

The constraint specification, which is a boolean value specification (could be just a constant, or a complex expression), determines (in addition to the user roles) whether the constraint applies. See the source that defines those extensions:

stereotype Invariant extends UML::Constraint
end;

(* You can declare access constraints on any element. *)
stereotype Access specializes Invariant
    property roles : Class[1,*];
    property allowed : AccessCapability[*];
end;

enumeration AccessCapability
    /* Objects */
    Create; 
    Delete;
    /* Attributes/links */
    Read;
    Update;
    /* Operations */
    Call;
end;

Notation changes

The notation changes were:

  • you can now declare a class as a role class by using the ‘role‘ modifier (or you can apply the Role stereotype directly if you prefer);
  • you can now use the ‘allow‘ keyword to declare access constraints on objects, attributes and operations.

Here is an example that should make things clear:


model banking;

  role class AccountOwner end;
  role class BranchManager end;
  role class AccountManager end;
  role class SecurityOfficer end;

  class Branch
    reference manager : BranchManager;
  end;

  class BankAccount
    allow BranchManager, AccountManager create, delete;
    reference owner : AccountOwner;
    reference branch : Branch;
    attribute balance : Double
        allow AccountOwner read { System#user() == self.owner } 
        allow BranchManager read { System#user() == self.branch.manager } 
        allow SecurityOfficer read;
    operation withdraw(amount : Double)
        allow AccountOwner { System#user() == self.owner }; 
    operation deposit(amount : Double)
        allow AccountOwner { System#user() == self.owner }; 
  end;
end.

For simplicity, I omitted details on the role classes, but they are normal classes otherwise and can have any attributes you may want, may specialize other classes etc.

But the use of the “allow” keyword to declare access constraints should be clear. If it isn’t to you, please provide feedback, here or on the project issue tracker.

Great, when can I use this?

TextUML Toolkit users: you should be able to use this if you update the plug-in in Eclipse now.

Cloudfier users: Cloudfier needs to honor the new access control features when running the model natively, or generating code, and some work is required before that happens, so you will need to wait a little longer to use this feature in your Cloudfier applications.

No matter which tool you use (or even if you don’t use any of them), if you have any opinion on the choices made, your feedback is really quite welcome.

Email this to someoneShare on FacebookShare on LinkedInShare on Google+Tweet about this on Twitter

7 thoughts on “Role-based access control for UML: the TextUML-Cloudfier approach

  1. rad9k

    March 19, 2016 at 9:50am

    - need to have roles inheritance
    - need to have “deny” as well as allow (and thus some more complicated semantics on access when having multiple roles)

    General comment: I would feel more comfortable with re use of actor (there is inheritance already there) and possibly even use case semantics. Use cases might seem a bit high level but I believe that there is place for good practical solution based on use cases

    • Rafael Chaves

      March 19, 2016 at 1:43pm

      Hi Rad, thanks for your feedback!

      need to have roles inheritance

      That is supported, as role classes are just normal classes, and are free to specialize other (role or non-role) classes.

      need to have “deny” as well

      Indeed. The profile actually does have that but I am planning to only actually support it later, when the basic use cases are well understood, and the cases for “deny” become more obvious.

      General comment: I would feel more comfortable with re use of actor (there is inheritance already there) and possibly even use case semantics. Use cases might seem a bit high level but I believe that there is place for good practical solution based on use cases

      I did want to use actors instead of classes initially, but in the UML metamodel, actors cannot have properties, so I gave up on that (it was not something I could fix with a profile). I did consider use cases as well, but there was no way to relate use cases to classes, attributes or operations (although for this I could have used a profile as well).

  2. rad9k

    March 19, 2016 at 5:43pm

    Not possible to extend actors: is it emf or textuml limitation?

    Bdw: I have tendencies to treat uml as a reference not a religious book.

    Bdw: uml is logically incoherent so… its limits will show up from time to time.

    • Rafael Chaves

      March 19, 2016 at 5:56pm

      You can extend actors in UML (via stereotypes), but turning them into something that can have properties (a structured classifier) would go too much against the spirit of the language for my tastes.

      Remember that TextUML is not a new language, it is just a notation for an existing language, UML. UML is actually quite flexible, but it still has some limits/directions in how you can extend it.

  3. rad9k

    March 19, 2016 at 6:59pm

    sure :)

    bdw: why do you need attributes on roles? maybe can go with this?

    why I’m pushing so much into use cases for rights management? in few analysis projects I was using use cases for rights modeling and it was ok. not sure if use cases will always work (depends on the work flow for sure, but there might be some other limitations as well)

    • Rafael Chaves

      March 19, 2016 at 8:34pm

      I believe we do want to keep different information on each role – like a bank customer, a branch manager, a security officer, there all sorts of particular information attributes that apply to each of them. An option I considered when looking into actors was to mirror each actor with a class, for the sake of storing information about the users, but the forced duplication annoyed me.

  4. rad9k

    March 19, 2016 at 7:00pm

    ” maybe can go with this?” > “maybe can go without this?” :)

Comments are closed.