[openstack-dev] [keystone] multiple federated keystones with single Identity Provider

Adrian Turjak adriant at catalyst.net.nz
Tue Dec 12 22:35:12 UTC 2017

On 08/12/17 11:47, Lance Bragstad wrote:
> On 12/07/2017 12:27 PM, Colleen Murphy wrote:
>> On Thu, Dec 7, 2017 at 5:37 PM, Pavlo Shchelokovskyy
>> <pshchelokovskyy at mirantis.com> wrote:
>>> Hi all,
>>> We have a following use case - several independent keystones (say KeyA and
>>> KeyB), using fernet tokens and synchronized fernet keys, and single external
>>> IdP for federated auth.
>>> Is it generally possible to configure both KeyA and KeyB such that scoped
>>> token issued by KeyA for a federated user is valid on KeyB?
>>> Currently we have the next problem - although domains/projects where
>>> keystone's mapping engine assigns federated users are equal by name between
>>> KeyA and KeyB, the UUIDs of projects/domains in KeyA and KeyB  are
>>> different, which seems to invalidate the scoped token issued by KeyA when
>>> trying to use it for KeyB. And it is not possible to create projects/domains
>>> with specific UUIDs via keystone API (which would probably solve this
>>> problem for non-autoprovisioned projects).
>>> Is such usage scenario supported? Or one should always use the unscoped
>>> token first to list projects/domains available on a specific keystone
>>> instance and then get a scoped token for usage o this instance only?
>> No, it is not currently possible to use the same token on projects in
>> different keystones, for the reasons you gave. You might be interested
>> in following https://review.openstack.org/#/c/323499/ if you're not
>> already aware of it, which has the goal of solving that problem.
>> It's also been brought up before:
>> https://review.openstack.org/#/c/403866/
>> http://lists.openstack.org/pipermail/openstack-dev/2016-December/108466.html
>> And we talked about it a lot at the last Forum (sorry my brief summary
>> does not really do the discussion justice):
>> http://www.gazlene.net/sydney-summit.html#keystone-operator-and-user-feedback
> I had a snippet about it in my recap under the "Other Feedback" section
> [0]. The TL;DR in my opinion is that we originally thought we could
> solve the problem with federation 100%, and if we couldn't we wanted to
> try and improve the parts of federation that would make that possible.
> The interesting bit we came up with during the feedback session in
> Sydney is what happens if a user no longer has a role on a project. For
> example;
> - A user has a role on Project A and in the us-east region and the
> us-west region, each region has it's own keystone deployment, but let's
> assume the ID for Project A are the same in each region
> - A user authenticates for a token scoped to Project A and starts
> creating instances in both regions
> - The user has their role from Project A removed in us-east, but not in
> us-west
> - The user isn't able to do anything within us-east since they no longer
> have a role assignment on Project A in that region, but they can still
> take the invalid token from the us-east region and effectively use it in
> the us-west region
> Without replicating revocation events, or syncing the assignment table,
> this will lead to security concerns.

Also worth noting is that this assumes both keystones have the same
fernet keys, so as to be able to generate tokens that the other can read.

From memory, the whole point of this exercise from the regulations side
was to make it so that data isn't 'replicated' and if one keystone was
compromised, the other wouldn't be, but if the fernet keys are the same,
then its still very bad if the host is compromised, since with the
fernet keys from the compromised keystone, you can now make bogus tokens
that the other keystone would trust. So that still doesn't solve the
problem and still probably falls out of regulations. In truth most of
this sounds like a loop hole around the regulations anyway rather than
honoring what they might intend (would be interesting to find out more
about that).

Unless I'm misremembering, this is all so that users in both 'regions'
can pretend it's all part of the same cloud, when in truth it isn't
really, and the regulations require that they are separate. Making that
fact clear, and that users have to swap between clouds, or generate a
token for each isn't that bad, and is overall much much safer. The user
can have the same project name in both, and getting a token from either
is as simple as just changing the auth url. Writing code to account for
this difference is probably easier than trying to solve this problem in
keystone and introducing weird potential security problems. :(
>> Lance mentioned today that we'd likely try to discuss it at our next
>> weekly meeting: http://eavesdrop.openstack.org/#Keystone_Team_Meeting
> Yep, I have it on the agenda for the next meeting [1].
> [0] https://www.lbragstad.com/blog/openstack-summit-sydney-recap
> [1] https://etherpad.openstack.org/p/keystone-weekly-meeting
>> Colleen
>> __________________________________________________________________________
>> OpenStack Development Mailing List (not for usage questions)
>> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20171213/d6ba499b/attachment.html>

More information about the OpenStack-dev mailing list