On 9/10/07, Alex Boisvert <[EMAIL PROTECTED]> wrote:
>
>
> That's not the point.  You may still want to have only JohnDoe or any HR
> personnel invoke a specific operation, irrespective of whether the
> operation
> is a workflow task.


.. or fail the activity?  I'm totally missing how the activity expects to
behave.


Asserting that a user belongs to a role on invoke is pointless, the receiver
> > > > should never trust your assertion and should make up their own
> > > > determination
> > > > upon receiving the message.
> > >
> > >
> > > An important RBAC concept is that of role activation.  If you happen
> to
> > > have
> > > administrator role, you may not want to assert this authority at all
> > times
> > > because you might unknowingly use a privilege that you don't intent to
> > > use,
> > > or you don't want to delegate all your authority to a process that
> does
> > > who-knows-what.   Explicitly asserting roles allows you to control how
> > > much
> > > authority you're passing on.
> >
> > Except that concept brings with it the weight of a tightly-coupled
> > centralized architecture for which it was designed, and I'm not sure
> it's
> > a
> > good idea to do that.
>
>
> No it doesn't.  How did you come up with that conclusion?  RBAC with role
> activation can work in a distributed environment with, e.g., a web of
> trust
> without central root to the system.


Loosely coupled is different from distributed.  In a loosely coupled
architecture you
a) never trust the client inputs but validate them yourself, and b) never
provide more information than you want a service to act upon.  So the only
thing you're passing around are credentials.

That part we know works very well: HTTP basic/digest, WSSE security token,
SAML, JDBC, FTP, SSH, POP3, etc.  How do we send roles with assertions
around?  Can I send root (but treat as user) to a service that might then
end up being a JDBC call or SSH invocation?

Assaf

alex
>

Reply via email to