Native SAML in keystone

Blueprint native-saml

We have had the ability for keystone to use a federated identity provider since the Icehouse release. This is done using Apache modules like mod_shib or mod_auth_mellon. Unfortunately they have some key limitations:

  1. No REST API to manage IdPs or other configuration
  2. Must restart/reload Apache to reload configuration

This works fine for small clouds where a single IdP is configured at bootstrap time, but falls down in a larger cloud where those IdPs may be a little more dynamic.

Problem Description

Usecase:

  1. As a public/private cloud operator I do not want to setup and manage the federation configuration for every domain. I would like to delegate that responsibility to the domain administrator. The domain admin can then make their domain members use whatever IdP they desire.

Proposed Change

The following things would be changed in keystone:

  1. Create APIs to manage IdPs in a backend
  2. Create APIs to manage SAML attribute mapping in a backend
  3. Create a new endpoint to do the SAML negotiation with an IdP

The code to interpret SAML2 would not be written by the keystone team. Instead we will use the pysaml2 library that we already have as a dependency.

Alternatives

There are a couple of alternatives:

  1. Do nothing, but this doesn’t make anyone happy.

  2. Fix mod_shib/mod_auth_mellon to be more dynamic.

    It is certainly possible for a C expert to make the modules reload their configuration more dynamically. It’s having APIs to manage the config that makes this much more difficult. keystone would need to write the XML config files and distribute then to all keystone nodes.

Security Impact

This code will be managing the negotiation between keystone and the IdP. There is certainly an impact if we get it wrong. In the worst case we are opening a new attack vector.

We are also making use of the pysaml2 library to do the heavy lifting. There is extra onus on us to make sure that code is correct and have a good process for dealing with security issues.

We will be using x509 certificates to validate and sign SAML documents. Those certificates will need to be managed and rotated by the operator. (This is no different for operators that are currently using an Apache module for federation.)

Notifications Impact

We may possibly have notifications for some IdP interactions. Nothing concrete yet.

Other End User Impact

End users should not be able to tell that we are providing the federation bits. It will all be transparent to them.

Performance Impact

This should have no direct significant performance impact. Signing/verifying documents may block a process a little longer, but that shouldn’t impact keystone all that much.

Other Deployer Impact

  1. Certificate rotations documented in the Security Impact section.
  2. Middleware will have to be utilized to take advantage of the feature.
  3. Deployment processes will have to use APIs to manage IdP configuration instead of managing the existing XML files.

Developer Impact

Nothing significant.

Implementation

Assignee(s)

Primary assignee:
dstanek (David Stanek <dstanek@dstanek.com>)

Work Items

Work items are documented in Proposed Change

Dependencies

Just the pysaml2 library.

Documentation Impact

  1. We’ll need to change/augment federation documentation to show how to setup and use this feature.
  2. Updates will need to be made to the keystone API documentation

References

None