Shadow users: Unified identity for multiple authentication sources¶
Locally managed users are handled slightly differently than users backed by LDAP, which are handled significantly differently than users backed by federation. Available APIs, relevant APIs, and token validation responses all vary. For example, users receive different types of IDs, passwords may or may not be stored in keystone, and in the case of federation, may not be able to receive direct role assignments. Future additional authentication methods pose a risk of complicating things further.
Instead of continuing down this path, we can refactor our user persistence to separate identities from their locally-managed credentials, if any. The result will be a unified experience for both end users and operators.
Problem Description¶
Use cases:
A consistent user experience. A federated user should have the same user experience as a locally-managed user. For example, federated users should be able to consume local role assignments just like locally-managed users can. Today, federated users can only be mapped into user groups, and receive tokens with larger payloads to accommodate their ephemeral nature.
One user, multiple credentials. In the real world, a single person has multiple means of authenticating themselves. For example, a person might carry a driver’s license, have a passport, and know a secret, such as a credit card verification value (CVV). In terms of OpenStack, a user could have a password in keystone, a private identity provider (such as a corporate LDAP), a public identity provider (such as a social media profile), an X.509 certificate, and an existing account on a remote cloud. If all of these authentication sources are equally valid, then the resulting user and operator experience should not vary based on which is chosen. All of these means of authentications should be tied to the same user identity, rather than resulting in distinct identities.
Note: We discussed the potential future need to tie privileges to authentication methods. This is out-of-scope for this spec, however the spec does not prevent or conflict from this feature being added in the future.
Additionally: Facilitates multi-factor authentication & account linking. With multiple types of credentials tied to the same user identity, users can authenticate themselves using more than one credential at a time. Keystone can then make stronger assertions about the identity of the user and the path to viable multi-factor authentication (MFA) is shortened. Likewise, the proposed changes support linking multiple accounts for the same user to a single account, as well as simplifies auditing around federated users. While this specification does not solve MFA or account linking, the refactoring done for this spec will make the development for those features easier.
Proposed Change¶
Separate user identities from their local-managed credentials. Refactor user table into an identity table and a locally-managed password table. Migrate data from the user table to these new tables and ultimately remove the user table. Modify backend code to utilize the new tables.
Shadow LDAP and federated users. Create new shadow tables for mapping LDAP and federated users to local identities. Federated users have a idp_id, protocol_id, display name, and a unique ID asserted by the identity provider. These are the minimal pieces of data required to identify returning users and provide them with a consistent identity. Likewise, LDAP users have a ‘domain_id’ and ‘dn’ used to identify the user. That being said, there may be an opportunity to generalize LDAP and federated users a single table. This will be solved during development.
Alternatives¶
We can continue treating federated users as ephemeral. In the long run, that’s either going to result in additional metadata being included in token payloads, ultimately bloating Fernet tokens beyond their intended capacity, or increasingly disparate user experiences depending on the user’s authentication source.
Security Impact¶
User IDs may be used as a source of authorization in OpenStack, and this spec will impact how they are controlled. We need to ensure that two unique users cannot be accidentally mapped to the same identity. The simplest solution to this is to always assign random UUIDs to serve as the OpenStack-facing user identity. We’ll have to carefully consider the implications of using any other source.
We’ll gain the ability to remove information about federated users from tokens, thus eliminating the concept of “federated tokens” altogether.
Federated user group membership will become persistent rather than ephemeral per authentication. A federated user might automatically receive a concrete role assignment as the result of the federation mapping and receive a Fernet token which reflects those roles. During the lifetime of that token, an operator might assign or revoke additional roles which will be immediately reflected in the user’s existing token. This behavior is already the case for locally-managed users, but will be new to federated users.
Notifications Impact¶
Auditing notifications which made use of federated identity information in tokens will no longer have access to that information, beyond multiple authentication methods presented by v3 tokens. All auditing notifications will effectively refer to a local user identity.
Other End User Impact¶
All end users will receive a consistent API experience: that of a locally-managed one.
Performance Impact¶
All user identities will need to be shadowed in keystone’s local backend. This means that if you have millions of users authenticating with keystone, even if it’s via identity federation, each of them will have a record in keystone.
The code path for token validations will be simplified, because they will always refer to local users.
Token formats requiring persistence will require less space on disk or in memory.
Federated users will be able to consume local role assignments, slightly increasing the response time of token validation.
Other Deployer Impact¶
No new configuration options.
Local identity persistence will need to be provided, even for deployments exclusively using identity federation.
During the upgrade, a database migration will be required to split locally-managed users. At runtime, shadow records will be created automatically as LDAP and federated users successfully authenticate with keystone.
Developer Impact¶
Developers supporting alternative authentication methods will be able to reference the common user identity, and create shadow records themselves. It’s important to note that existing CRUD APIs will not be impacted; only the backend is being refactored.
With federation being a more viable option, deployers supporting custom code to bootstrap new users with role assignments can take advantage of federated mappings to accomplish the same thing, if they move to federation.
Implementation¶
Assignee(s)¶
- Primary assignee:
dolph
- Other contributors:
dstanek derosenet
Work Items¶
Separate user identities from their local-managed credentials.
Shadow federated and LDAP users in new database tables.
Make concrete role assignments after federated mapping is evaluated.
Notifications emitted when a federated users is authenticated will need to be updated.
The mapping engine and other related backend logic will likely need some refactoring.
Dependencies¶
None.
Documentation Impact¶
Deployers need to understand the new local user persistence requirements, even in the case of federation.
Documentation that suggests that federated users cannot receive local role assignments needs to be revised.
References¶
Etherpad notes from the Mitaka summit federation session.
