← Back to team overview

openstack team mailing list archive

Re: OpenStack Identity: Keystone API Proposal

 

Agreed. I also can't think of a better name for 'thingies' than what we have now; role. Open to changing it if we find one soon. If not, we could always change it in the next version of the API.

It's also starting to sound like thingies are policies (a la XACML and other AuthZ constructs) or parts of policies… might be a good topic to test in the next summit (did I hear it was going to be in Boston?)

Given the current scope of supporting Nova, the granularity is defined for us in http://nova.openstack.org/runnova/managing.users.html . Five roles:

  *   Cloud Administrator. (cloudadmin) Users of this class enjoy complete system access.
  *   IT Security. (itsec) This role is limited to IT security personnel. It permits role holders to quarantine instances.
  *   System Administrator. (sysadmin) The default for project owners, this role affords users the ability to add other users to a project, interact with project images, and launch and terminate instances.
  *   Network Administrator. (netadmin) Users with this role are permitted to allocate and assign publicly accessible IP addresses as well as create and modify firewall rules.
  *   Developer. (developer) This is a general purpose role that is assigned to users by default.

But I grant you verily that this does not address the very valid future use case you bring up of having more granular roles or larger, more complex policies… we'll have to address pretty soon after Diablo though; PEPs, PDPs, COPS, MAIDs, etc…

Z

From: andi abes <andi.abes@xxxxxxxxx<mailto:andi.abes@xxxxxxxxx>>
Date: Wed, 13 Jul 2011 23:22:32 -0400
To: Ziad Sawalha <ziad.sawalha@xxxxxxxxxxxxx<mailto:ziad.sawalha@xxxxxxxxxxxxx>>
Cc: "Rouault, Jason (Cloud Services)" <jason.rouault@xxxxxx<mailto:jason.rouault@xxxxxx>>, "openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>" <openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>>
Subject: Re: [Openstack] OpenStack Identity: Keystone API Proposal

Dropped off the thread for a while... sorry.

Ziad, I think this sounds very reasonable. I think the only hiccup might be with the use of the term "role" which might connote some "bigger" meaning to folks with AAAA backgrounds.

If I understand your proposal, then a service can decide what is the granularity of ""thingies""  it registers with keystone(sorry.. no better term comes to mind - it's what you refer to in #2). These thingies can be "roles" or "actions" or whatever makes sense to the authorizing middleware.

In Jason's model, these thingies can be in the form of: <service-namespace>:<object-type>:<action>. In this case, it could be convenient to have a keystone api to check if a given thingy is bestowed on a user, rather than returning the full list of thingies the user possess. The rational here is that if services do end up defining very granular thingies (please do come up with a better name...) then the list might get somewhat lengthy. On some systems I've worked on we had 100's of action-tokens (more or less the name we used) - returning the full list, even filtered by the service's name-space, might be a bit much.


On Wed, Jul 13, 2011 at 12:45 AM, Ziad Sawalha <ziad.sawalha@xxxxxxxxxxxxx<mailto:ziad.sawalha@xxxxxxxxxxxxx>> wrote:
Here's a possible use case we can implement to address this:

  1.  A service 'registers' itself with Keystone and reserves a name (Ex. Swift, or nova). Keystone will guarantee uniqueness.
  2.  Registered services can then create roles for the service (Ex. swift:admin or nova:netadmin) or tuples as suggested below (nova:delete:volume)
  3.  On token validation, Keystone returns these roles and a service can apply it's own policies based on them.

This is super-simplified and we can expand on it.

Other benefits:

  *   Registration would also be handy to allow services to add and manage endpoints as well.
  *   We can also tie this with the concept of a ClientID so services can identify themselves as well with a long-lived token (see https://github.com/rackspace/keystone/issues/84)
  *   Common names for services could be implemented as shareable among different implementations (Ex: compute:admin)

Thoughts?

And comments inline ZNS>>


From: "Rouault, Jason (Cloud Services)" <jason.rouault@xxxxxx<mailto:jason.rouault@xxxxxx>>
Date: Thu, 16 Jun 2011 19:54:22 +0000
To: andi abes <andi.abes@xxxxxxxxx<mailto:andi.abes@xxxxxxxxx>>
Cc: Ziad Sawalha <ziad.sawalha@xxxxxxxxxxxxx<mailto:ziad.sawalha@xxxxxxxxxxxxx>>, "openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>" <openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>>
Subject: RE: [Openstack] OpenStack Identity: Keystone API Proposal

See inline…

Jason

From: andi abes [mailto:andi.abes@xxxxxxxxx]
Sent: Wednesday, June 15, 2011 5:04 PM
To: Rouault, Jason (Cloud Services)
Cc: Ziad Sawalha; openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Subject: Re: [Openstack] OpenStack Identity: Keystone API Proposal

Jason,
  Sounds like the model you're proposing could be achieved by  something like this:

On Keystone:
- Roles are identified by name and contain tuples in the form of:
 -- the service to which this permission applies (e.g. service nova, swift). Including the service is meant to side track attempts to normalize actions across very different types of services
 --  the type of target this action applies to - e.g.  volume, network port etc.
 -- action this permission allows - e.g. start vm, create volume.

- An authorize API call which accepts:
 - the service requesting the authorization
 -  user token (from a previous authentication)
 - tenant ID (to resolve the realm of the user token)
 - a target type
 -  the attempted action.

 This API would lookup the token, and if its present combine a set of the relevant permissions from all the roles the token is referencing. If the requested tuple exists in this combine set, the request is authorized.

A few caveats remain:

a) the above description doesn't include Resource Groups... as Ziad mentioned, that is currently differed. When those are introduced, the service should probably pass the instance-id of the target, and Keystone would have to take that into account.
JLR>>  I think there are a number of ways to account for this if we leveraged a hierarchical (URI) structure and allowed for wild carding.
ZNS>> We start to get in the world of policies here (lookup XACML)

b) the current API's in keystone allows a service to perform actions on multiple instances across tenants (containers) efficiently - a service could obtain a list of accessible tenants and cache it. If only the 'authorize' API is available, the service would need to perform a check with keystone for every instance
JLR>> Please explain the model for Tenant, Accounts, Projects, Groups, Roles…. I have not been able to discern how tenant will map to accounts and projects.  In any case, there are things that can be done to improve the potential overhead of authorization calls… however, you will not eliminate it completely.  That is the price you pay for increased security.  If authorization is pluggable, operators can determine if and how they want to use it.
ZNS>>tenants are the same as account in swift and project in Nova. The term tenant is not as overloaded as account, so we chose it.

c) In this model it is required to populate role definitions into keystone, for all services. Since keystone should be independent of other services,  the set of actions/targets should probably be considered as ""data"" for it - requiring a deployment step of sorts to make keystone aware of these roles.
This could be avoided if the authorization decision is looked at as 2 separate steps:
 1. figure out what roles a user posses.
 2. expand the set of roles to set of actions allowed
 3. determine if the action attempted is allowed

JLR>> Each service would need to document its target structure and actions (this would be very similar to their published API’s).  I think there would be a default set of roles and permissions pre-populated to satisfy the base set of use cases.
ZNS>>Maybe there are some built-in or reserved ones for Keystone and/or OpenStack (Ex keystone:admin and compute:admin). I'd suggest they were configurable if we did do that… but would help simplify a fast install and canonical deployment of OpenStack.

it is obviously debatable where keystone ends and the services begin. In the model above, keystone is responsible for all 3 steps via the authorize API. I *think* the current API provides a very similar model, with the line drawn at 1 - i.e. keystone provides to roles, and there is a separate middleware piece to perform 2 & 3, executing in the request pipleline of the service. Where this middleware executes (i.e. what is the API boundary to keystone) doesn't necessarily change the overall model.

JLR>>  Without a pluggable authorization system, operators will never be able to provide fine-grained access control without mucking with hardcoding in the API layer.  That is not a good path
>>ZNS: Indeed. A slippery slope. We want to provide core functionality out of the box, but we don't want to end up writing a complete IDM solution…

I *think*.








On Wed, Jun 15, 2011 at 11:52 AM, Rouault, Jason (Cloud Services) <jason.rouault@xxxxxx<mailto:jason.rouault@xxxxxx>> wrote:

In my opinion the services (and their developers) should not need to interpret roles thus resulting in varying semantics.  Roles should be defined by a set of configurable privileges to perform certain actions on specific targets for particular services.   The API should only need to know to check with an authorization subsystem whether the incoming request is allowed based on the who is making the request and the 3-tuple mentioned previously.

Jason


From: andi abes [mailto:andi.abes@xxxxxxxxx<mailto:andi.abes@xxxxxxxxx>]
Sent: Wednesday, June 15, 2011 9:18 AM
To: Rouault, Jason (Cloud Services)
Cc: Ziad Sawalha; openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Subject: Re: [Openstack] OpenStack Identity: Keystone API Proposal

I would expect that the API of each service would have to interpret the role assigned to a user in the context of that service - roles for swift nova glance quantum etc would probably carry very different semantics.

So, to my understanding, key stone provides authentication and user information - what tenants the user has access to, and what roles the user is assigned. The mapping of these to what the user can do on what instances in each service are left for the service to determine.

On Wed, Jun 15, 2011 at 10:32 AM, Rouault, Jason (Cloud Services) <jason.rouault@xxxxxx<mailto:jason.rouault@xxxxxx>> wrote:
Is there a plan to also have Keystone be the centralizing framework around authorization?   Right now it looks like policy enforcement is left to the API layer.

Thanks,

Jason

From: openstack-bounces+jason.rouault=hp.com<http://hp.com>@lists.launchpad.net<http://lists.launchpad.net> [mailto:openstack-bounces+jason.rouault<mailto:openstack-bounces%2Bjason.rouault>=hp.com<http://hp.com>@lists.launchpad.net<http://lists.launchpad.net>] On Behalf Of Ziad Sawalha
Sent: Friday, June 10, 2011 5:24 PM
To: openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Subject: [Openstack] OpenStack Identity: Keystone API Proposal

Time flies! It's June 10th already. In my last email to this community I had proposed today as the day to lock down the Keystone API so we can finalize implementation by Diablo-D2 (June 30th).

We've been working on this feverishly over the past couple of weeks and have just pushed out a proposed API here: https://github.com/rackspace/keystone/raw/master/keystone/content/identitydevguide.pdf

For any and all interested, the original source and code is on Github (https://github.com/rackspace/keystone<https://github.com/rackspace/keystone/raw/master/keystone/content/identitydevguide.pdf>), along with the current implementation of Keystone, examples, sample data, tests, instructions, and all the goodies we could muster to put together. The project also lives on Launchpad at http://launchpad.net/keystone.

The API we just put out there is still a proposal. We're going to be focusing on the implementation, but would still love to get community input, feedback, and participation.

Have a great weekend and regards to all,

Ziad







Confidentiality Notice: This e-mail message (including any attached or

embedded documents) is intended for the exclusive and confidential use of the

individual or entity to which this message is addressed, and unless otherwise

expressly indicated, is confidential and privileged information of Rackspace.

Any dissemination, distribution or copying of the enclosed material is prohibited.

If you receive this transmission in error, please notify us immediately by e-mail

at abuse@xxxxxxxxxxxxx<mailto:abuse@xxxxxxxxxxxxx>, and delete the original message.

Your cooperation is appreciated.

_______________________________________________
Mailing list: https://launchpad.net/~openstack
Post to     : openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Unsubscribe : https://launchpad.net/~openstack
More help   : https://help.launchpad.net/ListHelp


This email may include confidential information. If you received it in error, please delete it.

This email may include confidential information. If you received it in error, please delete it.

References