governance/goals/selected/consistent-and-secure-rbac.rst

Consistent and Secure Default RBAC

Problem Summary

OpenStack's initial goal to be a multi-tenant platform drove the idea that users operate within the confines of one project at a time. Early versions of the authorization system, which included keystone and various middleware, fulfilled this requirement.

However, OpenStack's explosive growth and adoption added services and API surface area to the ecosystem. This growth quickly outpaced the authorization engine. This allowed the community to develop rich APIs, across services, that operate on different layers of the infrastructure. For example, OpenStack has APIs that manage compute hosts, services, endpoints, domains, physical networks, and storage pools. All of these resources require knowledge about the underlying hardware, deployment architecture, and usage within a given organization. These APIs are clearly targeted at different users from APIs that expose resources, like instance, block storage devices, or virtual networks.

The authorization strategy didn't age gracefully with the rest of OpenStack. This means we used the best available tools at the time to protect the API we were developing across OpenStack.

This led to the following problems:

1. By default, users were either average end-users or operators, which is far too restrictive for real-world clouds
2. The design violated the principle of least privilege
3. Inconsistent authorization behavior across services, resulting in some default policies being completely open to any user
4. Operators need to be intimately familiar with the policy implementation to supply overrides for valid use cases (read-only privileges)
5. Auditing OpenStack APIs requires administrative access
6. Having no role hierarchy makes it hard to establish any low-level collection permission collection, like a role for read-only access, which is implemented inconsistently across deployments

The above issues aren't a complete set of all problems related to authorization in OpenStack, but they are pain points we, as the upstream community, know about. They also prohibit the adoption of OpenStack by:

• Requiring operators to understand and configure policy for any compliance target
• Aggregating all authoritative power into select users, violating the constrained RBAC model
• Not providing a role hierarchy that allows for easy authorization management
• Not providing a granular set of permissions
• Not providing an easy way for operators to audit what a particular user can do within the deployment

Where are we today?

The following initiatives are in progress or complete:

1. Moved policy and documentation into code
2. Created a default role hierarchy in keystone (default roles specification)
3. Added a new scope to keystone (system scope specification)
4. Updated all libraries to understand the new scope
5. Documented the idea of personas
6. Created a policy popup team to enable this work across more OpenStack services.
7. Proposed and implemented several testing strategies using tempest, functional, and unit tests for projects to use as a reference
8. Audited every active OpenStack project API and mapped administrative functionality into the system-scope personas
9. Applied the reader and member role consistently to project-scoped resources
10. Converted each service policy file from using JSON to YAML

To date, the work to audit each API, propose new default policies, and implement unit, functional, or tempest tests has accumulated more than 130,000 lines of code change across 41 repositories.

Direction change

Throughout this process we've communicated with operators and end users about the changes to implement a new scope type. Early feedback on the approach to isolate system-level APIs behind a new authorization target alluded to the ability for operators to continue supporting their users by interacting with project-owned resources. A good example of this use case is evacuating an instance from a host.

Based on the initial discussions of how system-scope would be used, we decided to allow operators to interact with project-owned resources using system-scoped tokens.

It wasn't until we started applying this idea to various services that we realized it was going to cause issues with service-to-service communication and require significant refactoring in each service. This is due to the fact that OpenStack services have been developed with the assumption that project IDs will always be present, and it's rare to interact with a resource without a project ID associated to the request.

For example, if an operator uses a system-scoped token to create an instance for a user in a specific project, they need to specify the project ID that owns the instance and they need to pass their system-scoped token to the service. Each service would need to understand the concept of system-scope and make sure to use the correct project ID. This approach is error prone, especially since each OpenStack service can have multiple clients to other services. We worked through the design and uncovered these issues while implementing a specification we targeted for the Xena release that allowed system users to pass-through project IDs with a system-scoped token. This work would have required significant refactoring and non-trivial changes to multiple projects, increasing the risk in implementing the functionality consistently and safely.

We spent significant amount of time during the Yoga PTG revisiting the discussion (etherpad). Ultimately, we stepped back and realized that the primary use case for allowing system users to operate on project-owned resources with a system-scoped token was to allow for backwards compatibility.

While we certainly want to make things as easy as possible for operators to use, we're not sure the additional overhead required to teach each OpenStack service about system-scope in this way would be beneficial. This is especially true when we considered the fact that a single user account, or bearer token, carries a significant amount of authorization. We're really just pushing the problem from a user with the admin role on a project to anyone with the admin role on the system.

Instead, we decided to remove the assumption that anyone using a system-scoped token should automatically be able to access any OpenStack API.

To clarify, we did agree that system administrators (e.g., operators) should be able to manage resources within a project, but we don't want to conflate that use case into the system-scope construct for the reasons described above. System administrators have the ability to grant themselves authorization to projects, domains, and the deployment system itself. A few extra steps would allow them to get the correct authorization to the intended project and perform the necessary operations using a token flow that's already supported. Additionally, it provides a very clear audit trail.

In yoga cycle, we redefined this goal with the changes mentioned above so that allowing system administrators to access system level resources APIs only and allow project users to access project-level resource APIs. These changes have been done for nova and neutron.

This was not the end of the RBAC design discussion. After knowing the operators' use cases we used the feedback to redefine the direction in the Zed cycle.

The issues we are facing with scope concept:

1. It breaks OpenStack existing NFV use case and orchestration tooling:

   When the deployment project started consuming the nova new policy defaults with scope enabled, we got to know that the scope enable will break heat (orchestration tooling), Tacker (NFV deployment) users or any operators script interacting all the OpenStack interfaces with administrators user.

   Heat 'create stack' API uses the user credentials (admin) to create project and system-level resources in backend services. For example, it creates project users in keystone (system level resource), flavors in nova (system level resource), servers in nova (project level resource), and networks in neutron (project level resource). If we enable the scope checking in services, then the user calling heat 'create stack' APIs which are scoped to either project (existing way) or system (if we change that) will not be able to call the system and project scoped APIs on the service side. We discussed the possible solutions in Zed PTG , openstack-discuss ML, and in policy popup meetings but none of those are good solutions and end up breaking the existing stack.

   Enabling scope checking also breaks Tacker (NFV Orchestration service) deployment as they use heat 'create stack' to build OpenStack infrastructure.

2. Operator feedback on `scope`:

   We collected the operators' feedback on scope and how OpenStack APIs will be accessed with scope enabled.

   First feedback is taken in ops meetup, Berlin where it was clear that scope things are difficult to understand for most of the operators. It will break their use case of 'accessing everything with a single token'. 'Admin' is already a confusing concept for many of them and admin with scope combination makes it more confusing. The operators agreed with postponing the scope implementation to be able to land the project persona first.

   KDDI, japanese telco company shared the feedback about their use case and how the scope will break their use case also. An "OpenStack Administrator" who is created by the "keystone-manage bootstrap" command, should be able to operate the complete stack even that is project-level or system-level resources. Dividing the permissions for project and system level resources may have an impact on echosystems or scripts outside OpenStack. Another point they raised is that there should be a way that the operator can configure the policy permissions in policy.json and with the scope that cannot be done as the scope is not the configurable thing.

Due to the above feedback and use case, we decided to postpone the scope implementation. That is the way forward to at least implement the project personas which is asked by many operators. Basically, we define the boundaries of this goal:

• Finish delivering project personas This is to introduce the member and reader roles to operate things within their project. By default, any other project role like foo will not be allowed to do anything in the project.

• Change the scope implementation to be project only

  Services with project resources that have already implemented scope (or have yet to) should make all policy rules set scope_types=['project']. This will help ensure that any API operations performed with a system-scoped token will fail early, with a 403, instead of later in the process when a project_id is required. One exception here is Ironic, which has implemented scope and has some users adopting it. We must not break these users so it is okay to keep the scope implementation as-is.

So, where do we go from here?

We have a set of OpenStack services that have implemented or over-extended the usage of system-scope and applied it to project-specific resources. Other services have yet to adopt the system-scope feature.

Currently, none of the policy work we've done since Queens is widely usable by default since it's not applied consistently across services. The idea of this community goal is to define the absolute minimum amount of work required to allow operators to opt into the new authorization behavior and start using the personas we've been developing since Queens.

We should defer any policy work including scope that isn't absolutely necessary to the criteria of this goal for future improvements. Otherwise we risk delaying the functionality another release. Instead, we can acknowledge the gaps, order them on a timeline for future improvements, and at least deliver something useful to operators sooner rather than later. At least we have a clear understanding on project persona from developer as well from operator side and if we again delay implementing it, there is high possibility that developers involved in this work will loose the motivation and we will never ship the usable project persona in OpenStack RBAC. Let's accept all the challenges we have with scope concept and be ready to revert the scope implemented even that is already implemented in your project.

Phase 1

Change in scope implementation

There are some projects like nova, neutron, ironic and octavia that have already implemented the scope_type in their policy. This section will provide a clear direction for such project as well as if any new projects want to implement the scope.

• As mentioned above, Ironic will keep scope implementation as-is.
• Other projects who have already implemented scope for example, Nova, Neutron, Octavia etc or any project who has not yet implemented it, should make everything scoped to project (scope_type to project only). Keeping everything as project scoped will make sure to fail the operations performed with a system scoped token (which does not have project_id) early with 403 instead of failing it with 500 in the lower layer.
• Keystone will continue supporting the scope implementation for deployment moved/can move to system scope enable for example, ironic + keystone. But we need to make sure it also works for deployments that do not use system scope token means continue working with the project scoped token. For that we need to do two changes in keystone:
  1. Remove the scope string (system:all) from the policy rule check_str.
  2. Add the project in scope_type in every policy rule.

Example:

policy.DocumentedRuleDefault(
    name='os_compute_api:os-hypervisors:list',
    check_str='role:admin',
    scope_types=['project']
)

Implement support for project-reader and project-member personas

The project-reader and project-member changes will make sure that by default any other role for example foo in that project will not be able to do anything.

Legacy admin will be unchanged and continue to work same way as it does today._

`project-reader`:

project-reader is denoted by someone with the reader role on a project. It is intended to be used by end users for read-only access within a project.

project-reader persona in the policy check string:

policy.RuleDefault(
    name="project_reader",
    check_str="role:reader and project_id:%(project_id)s",
    description="Default rule for Project level read only APIs."
)

Using it in policy rule (with admin + reader access): (because we want to keep legacy admin behavior the same we need to give access of reader APIs to admin role too.)

policy.DocumentedRuleDefault(
    name='os_compute_api:servers:show',
    check_str='role:admin or (' + 'role:reader and project_id:%(project_id)s)',
    description="Show a server",
    operations=[
        {
            'method': 'GET',
            'path': '/servers/{server_id}'
        }
    ],
    scope_types=['project'],
)

OR

policy.RuleDefault(
    name="admin_api",
    check_str="role:admin",
    description="Default rule for administrative APIs."
)

policy.DocumentedRuleDefault(
    name='os_compute_api:servers:show',
    check_str='rule:admin_api or rule:project_reader',
    description='Show a server',
    operations=[
        {
            'method': 'GET',
            'path': '/servers/{server_id}'
        }
    ],
    scope_types=['project'],
)

`project-member`:

project-member is denoted by someone with the member role on a project. It is intended to be used by end users who consume resources within a project. It inherits all the permissions of a project-reader.

project-member persona in the policy check string:

policy.RuleDefault(
    name="project_member",
    check_str="role:member and project_id:%(project_id)s",
    description="Default rule for Project level non admin APIs."
)

Using it in policy rule (with admin + member access): (because we want to keep legacy admin behavior same we need to give access of member APIs to admin role too.)

policy.DocumentedRuleDefault(
    name='os_compute_api:servers:create',
    check_str='role:admin or (' + 'role:member and project_id:%(project_id)s)',
    description='Create a server',
    operations=[
        {
            'method': 'POST',
            'path': '/servers'
        }
    ],
    scope_types=['project'],
)

OR

policy.RuleDefault(
    name="admin_api",
    check_str="role:admin",
    description="Default rule for administrative APIs."
)

policy.DocumentedRuleDefault(
    name='os_compute_api:servers:create',
    check_str='rule:admin_api or rule:project_member',
    description='Create a server',
    operations=[
        {
            'method': 'POST',
            'path': '/servers'
        }
    ],
    scope_types=['project'],
)

'project_id:%(project_id)s' in the check_str is important to restrict the access within the requested project.

This would push the functionality even closer to end users, making the API more self-serviceable.

Legacy admin continues to work as it is

During the operator feedback, it is clear that we need to keep the legacy admin working as it is currently. We will not do any change in legacy admin behavior and access information. In Phase 2, we will introduce the project manager persona who will be able to do the more privileged operation within the project than project member. More details in Phase 2 section.

policy.DocumentedRuleDefault(
    name='os_compute_api:os-hypervisors:list',
    check_str='role:admin',
    scope_types=['project']
)

Managed volumes:

policy.DocumentedRuleDefault(
    name='volume_extension:volume_manage',
    check_str='role:admin',
    scope_types=['project'],
)

policy.DocumentedRuleDefault(
    name='identity:delete_service',
    check_str='role:admin',
     scope_types=['project'],
)
policy.DocumentedRuleDefault(
    name='identity:create_endpoint',
    check_str='role:admin',
)

Listing project resources across the deployment

As we are keeping the legacy admin same as it is currently, legacy admin (meaning anyone with the admin role on a project) will continue to be able to list all the resources across the deployment (for applicable APIs only.) The following is an example of what a policy would look like using this approach:

policy.DocumentedRuleDefault(
    name='os_compute_api:servers:detail:get_all_tenants',
    check_str='role:admin',
     scope_types=['project']
),

This functionality is important for operators finding resources, especially for support cases, like rebooting or live migrating an instance.

How operators opt into the new functionality

If we can complete each item above in the Zed release, operators will be able to configure each service to opt into the new defaults across all services, securely implementing the same personas across the deployment:

[oslo_policy]
enforce_new_defaults=True

This configuration enables the following personas:

• Admin

  • Denoted by someone with the admin role on a project
  • This is existing admin we have in OpenStack policy.
  • Intended for operators who need elevated privilege on complete deployement
  • Not intended for end users
  • List Hypervisors detail
  • Forcibly reset the state of an instance
  • Forcibly deleting an application stack
  • Making an image public to the entire deployment
  • Create physical provider networks
  • Add or delete services and endpoints
  • Create new volume types
  • Move pre-existing volumes in and out of projects
  • Create or delete HSM transport keys

• Project Member

  • Denoted by someone with the member role on a project
  • Operate within their own project resource
  • Intended to be used by end users who consume resources within a project
  • Create, delete, or update an instance
  • Create, delete, or update a volume
  • Create, delete, or update a network
  • Can get or list the instances from its own project
  • Cannot create, delete, or delete the instance, volume, or network of other project
  • Cannot get or list instances, volumes, or networks of other project

• Project Reader

  • Denoted by someone with the reader role on a project
  • Operate within the own project resource
  • Intended to be used by end users for read-only access within a project
  • Not allowed to make any writable changes to project-owned resources
  • List and get instances
  • List and get volumes
  • List and get images, including private images within the project
  • List and get networks
  • Cannot get or list instances, volumes, or networks of other project

These new personas fix the existing issue where any user having any role within project (for example 'foo' role) can create or delete the resources in that project. It also provides the ability for the operator to assign the read-only role for cloud auditing the project resources/activities. This does not directly solve the case of doing global audit with single role.

Phase 2

Isolate service-to-service APIs to the service role

Any API developed for machines to communicate with each other should use the service role. This is an important part in reducing authorization for each service. For example, neutron needs to inform nova about network changes, but it shouldn't need the ability to create new users and groups in keystone, which it currently has.

Phase 3

Implement support for project-manager personas

`project-manager`:

A project-manager can use project-level management APIs and is denoted by someone with the manager role on a project. It is intended to perform more privileged operations than project-member on its project resources. A project-manager can also perform any operations allowed to a project-member or project-reader.

An example of a project-level management API is the Block Storage default-types API, which allows a default volume type to be set for a particular project. Since the change affects only that project, it makes sense to allow a responsible person within the project to set the default type, rather than require them to contact an administrator to do it. Implementing the project-manager persona will make this possible.

It is up to each service to define which API calls (if any) should be considered as project-level management APIs.

The project-manager needs to be added in the role implication so that the admin role implies manager, the manager role implies member, the member role implies reader. This needs the modification in the already merged keystone specification.

project-manager persona in the policy check string:

policy.RuleDefault(
    name="project_manager",
    check_str="role:manager and project_id:%(project_id)s",
    description="Default rule for  project-level management APIs."
)

Using it in policy rule (with admin + manager access): (because we want to keep legacy admin behavior same we need to give access of project-level management APIs to admin role too.)

policy.DocumentedRuleDefault(
    name='os_compute_api:os-migrate-server:migrate_live',
    check_str='role:admin or (' + 'role:manager and project_id:%(project_id)s)',
    description="Live migrate a server to a new host without a reboot",
    operations=[
        {
            'method': 'POST',
            'path': '/servers/{server_id}/action (os-migrateLive)'
        }
    ],
)

OR

policy.RuleDefault(
    name="admin_api",
    check_str="role:admin",
    description="Default rule for administrative APIs."
)

policy.DocumentedRuleDefault(
    name='os_compute_api:os-migrate-server:migrate_live',
    check_str='rule:admin_api or rule:project_manager',
    description="Live migrate a server to a new host without a reboot",
    operations=[
        {
            'method': 'POST',
            'path': '/servers/{server_id}/action (os-migrateLive)'
        }
    ],
)

'project_id:%(project_id)s' in the check_str is important to restrict the manager role access within the requested project and 'role:admin or' in check_str will make sure the legacy admin continues working as it is.

This will provide a way for the operator to configure a user to give the more privileged access within a project but no access to system-level resources or cross-project operations.

The project-manager persona is described as follows:

• Project Manager (project-level management)

  • Denoted by someone with the manager role on a project
  • Intended for responsible end-users to give them slightly elevated privileges that affect only their own project's resources
  • Can perform more privileged than project-members on a project
  • Forcibly reset the state of an instance
  • Forcibly deleting an application stack
  • Locking and unlocking an instance
  • Setting the default volume type for a project
  • Setting the default secret store for a project

Tracking

Etherpad: https://etherpad.opendev.org/p/rbac-goal-tracking

Champion

1. Lance Bragstad <lbragstad@redhat.com> (lbragstad)
2. Ghanshyam Mann <gmann@ghanshyammann.com> (gmann)

Gerrit Topic

To facilitate tracking, commits related to this goal should use the gerrit topic:

secure-rbac

Completion Date & Criteria

Work completed by Yoga Timeline (7th Mar 2022)

1. Keystone has the project persona (admin, member, reader) ready to be used by the services.
2. Few projects like keystone, nova, neutron, octavia etc adopted the project persona but along with scope type which needs to be modified to drop the scope type.

Z-Release Timeline

1. Convert the scope_type of every policy rule to project, or specify no scope_type, as appropriate

   Some standalone services like Ironic can still have their existing scope implementation as long as it does not break any cross service communication. Nova, Neutron, Keystone and any other projects who have already implemented the scope as system, domain and project in their policy default need to make everything scoped to project (scope_type to project only).

2. Services start implementing Phase 1

   At this point, all services are free to start implementing project-member and project-reader personas as described above in Phase 1. By the end of the Zed release, at least all the base services must have Phase 1 complete. Phase 1 introduces the new personas but allows operators to opt into the new behavior for services that complete Phase 1, allowing operators to upgrade smoothly to the new permission model on a per-service basis.

   It's important that we have an OpenStack-wide release note or statement that explicitly states the status of this work and how permissions behave across OpenStack services.

3. Keystone implements a new default role called manager

   The manager role will be a part of the role hierarchy so that the admin role implies manager, the manager role implies member. This need the modification in already merged keystone specification

4. Keystone implements a new default role called service

   The service will standardize a role that's already required in some default policies across OpenStack. This role must be built outside the existing role hierarchy, where admin implies manager implies member implies reader. This work requires a keystone specification.

5. OpenStack-wide Personas Documentation

   We need very clear documentation that describes all the potential personas, what they mean, who they were designed for, and how to use them. By the end of the Zed release, this document should include each persona and what its support is across OpenStack services.

   Engineers should use this documentation to determine what the default policy should be for APIs they're developing and maintaining. Operators should use it to understand what personas are the most appropriate for their users based on the permissions they need.

At this point, operators can choose to enable the new defaults for services that have completed Phase 1. This will require the operator to configure the service to use enforce_new_defaults=True if they chose to adopt the new behavior for services that support it.

2023.1 Release Timeline

1. All services must implement Phase 1
2. Services start implementing Phase 2
3. Services start implementing Phase 3 and updates policies to include the manager role where applicable.
4. Any service that completed Phase 1 in Zed can set enforce_new_defaults=True by default. It means new defaults will be enabled by default but operator will have way to disable it with enforce_new_defaults=False for that service. Also make enforce_scope=True to make sure project scope is enforced.

At this point, every OpenStack service will have completed Phase 1, which allows operators to opt into using project-member and project-reader across their entire deployment.

To summarize, operators will need to update every service configuration file where they want to use project-member and project-reader. For example:

1. Set glance-api.conf [DEFAULT] enforce_secure_defaults=True
2. Set glance-api.conf [oslo_policy] enforce_new_defaults=True
3. Set neutron.conf [oslo_policy] enforce_new_defaults=True
4. Set cinder.conf [oslo_policy] enforce_new_defaults=True
5. Set ironic.conf [oslo_policy] enforce_new_defaults=True
6. Set barbican.conf [oslo_policy] enforce_new_defaults=True

2023.2 Release Timeline

1. All services must implement Phase 2

2. All services must implement Phase 3

3. Update oslo.policy enforce_new_defaults=True

   Since all services have completed Phase 1, we can update the default in oslo.policy so that enforcement checks new default by default. This will allow each service to remove code to override the enforce_new_defaults=True and use the upstream default from oslo.policy.

4. Update oslo.policy enforce_scope=True

   Since all services have completed Phase 1, we can update the default in oslo.policy so that scope enforcement is checked by default. This will allow each service to remove code to override the enforce_scope=True and use the upstream default from oslo.policy.

5. Any service that implemented Phase 1 in Zed and enabled enforce_new_defaults in 2023.1 release can remove deprecated policies used to implement Phase 1.

Operators consuming the 2023.1 release will have the personas delivered in Phase 1 available and enabled by default. This includes project-member for common end-user interactions, and project-reader for a read-only variant of project-member.

2024.1-Release Timeline

1. Any service that implemented Phase 1 in 2023.1 and enabled enforce_secure_defaults in the 2023.2 release (non SLURP) needs to keep the old deprecated policies for the 2024.1 release (SLURP) also and can remove them after that. The Idea here is to have at least one SLURP release between the point when the new defaults are enabled and the old policies are removed.

2. Remove the oslo.policy enforce_scope config flag

   Since all services have completed Phase 1, and have enforce_scope=True by default in oslo.policy for every service, we can remove this configuration flag itself and have scope checks enable by default.

Operators consuming the 2024.1 release will have full support for project-manager, project-member, project-reader, and service role dedicated for service-to-service communication. There will not be support for deprecated policies in this release.

2024.2-Release Timeline

1. Update oslo.policy enforce_new_defaults=True

   At this stage, all services are supposed to complete Phase 1 and have the enforce_new_defaults flag enabled at service level. Now we can update the default value of config option enforce_new_defaults in oslo.policy to True. This will allow each service to remove code to override the enforce_new_defaults=True and use the upstream default from oslo.policy. If any service still needs to keep the default value as False then they can do it by overriding the default.

2. Update oslo.policy enforce_scope=True

   At this stage, all services are supposed to complete Phase 1 and have the enforce_scope flag enabled at service level. Now we can update the default value of config option enforce_scope in oslo.policy to True. This will allow each service to remove code to override the enforce_scope=True and use the upstream default from oslo.policy. If any service still needs to keep the default value as False then they can do it by overriding the default.

2025.2-Release Timeline

1. Remove the oslo.policy enforce_scope config flag

   The config option enforce_scope was added temporarily to migrate to the new RBAC. This is enabled by default in the 2024.2 release (and also in 2025.2 SLURP release) means all deployments get the scope enabled by default. Now we can remove this config flag and have scope checks enabled by default.

References

• Policy Pop-Up Team wiki
• https://etherpad.opendev.org/p/policy-popup-yoga-ptg

Current State / Anticipated Impact

Current progress is maintained on the tracking etherpad page.