openstack team mailing list archive

["Nguyen, Liem Manh" <liem_m_nguyen@xxxxxx>]

Hi Joe/Dolph,

>>We do need to come up with a suggested default implementation - given that users are not segmented into domains, what would you suggest?
I think it is good that we can encapsulate the concept of isolation (or lack thereof) within RBAC rather than via API (+1 for that  :)).  That said, I think it is worth-while to demonstrate (e.g.,  via a sample RBAC rule) how we can achieve isolation (for tenant, we do already; but also for users).  An isolation use-case would be where one would want a user created in one domain to be not modifiable (or readable)  by a domain-admin in another domain.

Thanks,
Liem

From: Joseph Heck [mailto:heckj@xxxxxx]
Sent: Sunday, June 10, 2012 1:37 PM
To: Nguyen, Liem Manh
Cc: openstack@xxxxxxxxxxxxxxxxxxx (openstack@xxxxxxxxxxxxxxxxxxx); Dolph Mathews
Subject: Re: [Openstack] [openstack][keystone] v3 API question


On Jun 8, 2012, at 8:38 PM, Dolph Mathews wrote:
On Jun 8, 2012, at 6:47 PM, "Nguyen, Liem Manh" <liem_m_nguyen@xxxxxx<mailto:liem_m_nguyen@xxxxxx>> wrote:

Hi Joe/Dolph,

I have a few questions on the v3 API's create_user (sorry the comments section in the Google docs is getting pretty cluttered now):

(POST) /users ==> create_user
{
" tenant_id": ...
"name": ...
"password": ...
"enabled": ...
"email": ...
"description": ...
}


1.       Does this tenant_id option establish the default tenancy of the created user?

Yes.



2.       If it does, is this default tenancy immutable or mutable?  If it is mutable, who (what role) can change it and via what API?

Should be mutable by admins, via the admin API, as it's just a regular attribute of the user and the keystone "admin" concept currently applies to the entire deployment.



3.       What is the intended purpose of a user's default tenancy?  Is the default tenancy association intended to link a user to a given domain (rather than the normal user-tenant role association)?

To be a little more explicit about the "default tenant" thing - it's used so that the user can give a "username/password" pair to the POST to /token and get a token scoped to a tenant without the user asserting which tenant it wishes to use.



"Auto-scoping" the user's context, when a tenant is not explicitly specified during auth.

I can't fairly answer the second question because the idea of domains wasn't around at the time. However, if you replace the term "domain" with "tenant", I'd say yes.



The reason I am asking this is that I would like to know what level of isolation (if any) we can establish for users that are homed to different domains...  So, for example, an isolation would be that a user A with a default tenancy in domain X may not be modified or deleted by a domain-admin in domain Y, even when user A has tenant membership in domain Y.

I think that's an issue best solved per-deployment by robust RBAC, rather than being hardcoded either way.

I agree with Dolph - since domains don't segment users, only tenants - then the choice is really driven by RBAC, which can be encapsulated in rules. None of the API defined in the v1 draft currently even hints at what RBAC might be wrapped around each of the calls, but all of those calls can and should be wrapped as actions and defined in the policy.json file.

We do need to come up with a suggested default implementation - given that users are not segmented into domains, what would you suggest?