Open Stack

Replying inline.

On Wed, 2018-06-27 at 16:39 -0400, Zane Bitter wrote:
> We're looking at using Barbican to implement a feature in Heat[1]
> and 
> ran into some questions about how secrets are identified in the
> client.
> 
> With most openstack clients, resources are identified by a UUID. You 
> pass the UUID on the command line (or via the Python API or
> whatever) 
> and the client combines that with the endpoint of the service
> obtained 
> from the service catalog and a path to the resource type to generate
> the 
> URL used to access the resource.
> 
> While there appears to be no technical reason that barbicanclient 
> couldn't also do this, instead of just the UUID it uses the full URL
> as 
> the identifier for the resource. This is extremely cumbersome for
> the 
> user, and invites confused-deputy attacks where if the attacker can 
> control the URL, they can get barbicanclient to send a token to an 
> arbitrary URL. What is the rationale for doing it this way?
> 

IIRC, using URIs instead of UUIDs was a federation pre-optimization
done many years ago when Barbican was brand new and we knew we wanted
federation but had no idea how it would work.  The rationale was that
the URI would contain both the ID of the secret as well as the location
of where it was stored.

In retrospect, that was a terrible idea, and using UUIDs for
consistency with the rest of OpenStack would have been a better choice.
 I've added a story to the python-barbicanclient storyboard to enable
usage of UUIDs instead of URLs:

https://storyboard.openstack.org/#!/story/2002754

I'm sure you've noticed, but the URI that identifies the secret
includes the UUID that Barbican uses to identify the secret internally:

http://{barbican-host}:9311/v1/secrets/{UUID}

So you don't actually need to store the URI, since it can be
reconstructed by just saving the UUID and then using whatever URL
Barbican has in the service catalog.

> 
> In a tangentially related question, since secrets are immutable once 
> they've been uploaded, what's the best way to handle a case where
> you 
> need to rotate a secret without causing a temporary condition where 
> there is no version of the secret available? (The fact that there's
> no 
> way to do this for Nova keypairs is a perpetual problem for people,
> and 
> I'd anticipate similar use cases for Barbican.) I'm going to guess
> it's:
> 
> * Create a new secret with the same name
> * GET /v1/secrets/?name=<name>&sort=created:desc&limit=1 to find out
> the 
> URL for the newest secret with that name
> * Use that URL when accessing the secret
> * Once the new secret is created, delete the old one
> 
> Should this, or whatever the actual recommended way of doing it is,
> be 
> baked in to the client somehow so that not every user needs to 
> reimplement it?
> 

When you store a secret (e.g. using POST /v1/secrets), the response
includes the URI both in the JSON body and in the Location: header. 
 
There is no need for you to mess around with searching by name, since
Barbican does not use the name to identify a secret.  You should just
save the URI (or UUID) from the response, and then update the resource
using the old secret to point to the new secret instead.

> 
> Bottom line: how should Heat expect/require a user to refer to a 
> Barbican secret in a property of a Heat resource, given that:
> - We don't want Heat to become the deputy in "confused deputy
> attack".
> - We shouldn't do things radically differently to the way Barbican
> does 
> them, because users will need to interact with Barbican first to
> store 
> the secret.
> - Many services will likely end up implementing integration with 
> Barbican and we'd like them all to have similar user interfaces.
> - Users will need to rotate credentials without downtime.
> 
> cheers,
> Zane.
> 
> BTW the user documentation for Barbican is really hard to find.
> Y'all 
> might want to look in to cross-linking all of the docs you have 
> together. e.g. there is no link from the Barbican docs to the 
> python-barbicanclient docs or vice-versa.
> 
> [1] https://storyboard.openstack.org/#!/story/2002126
> 
> _____________________________________________________________________
> _____
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubs
> cribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev