>Casper.Dik at Sun.COM wrote:
>> What I would like is:
>> 
>>      - a role which you can assume by typing your own password, not the 
>> role
>
>Which is basically what sudo does.

Can we pretend that the people who post things here "know things" and that 
we don't require a "sudo 101"?  Thanks much.

>>      - but I would like it to keep the same uid as the user
>>        (making the role a different uid in the credential)
>
>Though why do we need to have a role for this ?  Requiring a role makes 
>OpenSolaris RBAC more complex to setup that sudo.

The current implementation has issues.  However, RBAC was designed to put 
the authentication at the "su role" step. Putting dangerous profiles 
into a user's hand is abusing RBAC.

>I believe the requirement is really - needs authentication.  So why not 
>update pfexec to call PAM ?

Because we use it for tasks where we just need to use a "set-uid helper"  and
where we absolutely do not want authentication.

Pfexec is called too late; if you have a role with a profile shell, you 
have already authenticated.  (Don't forget that pfexec was invented 
because we felt that the initial RBAC implementation which used set-uid 
root pfcsh/pfsh was dangerous.)

>We could even mimic what sudo does and provide the ability on a per 
>profile basis to determine if the authentication step is necessary or not.
>
>I also believe it would be useful if pfexec when it called PAM had the 
>ability to have a embedded_pfexec(1M) version like we have 
>embedded_su(1M).  Though maybe that isn't as necessary given gksu(1M) ?
>
>There are existing RFEs already logged for the above functionality.

pfexec is used for authenticated users and pfexec should not authenticate.
I believe that such RFEs are wrong.

That doesn't mean that there's something wrong with how we use RBAC; 
clearly the giving users "Primary Administrator" it not how RBAC was
intended to be used.

Casper


Reply via email to