Enforce access control for Fuel UI

https://blueprints.launchpad.net/fuel/+spec/access-control-master-node

Currently, there is no enforced access control to the Fuel UI.

Problem description

Anyone with access to network can create, change and delete environment.

At a minimum, this requirement could be fulfilled by a login/password when connecting to the Fuel UI. If implemented in this manner, additional requirements will be needed:

• Ability for a user to set / change their own password

• Default admin/admin
• Should be configurable in fuelmenu

More advanced options:

• Secure/Encrypted storage of passwords (potentially on the Fuel Master Node)

• A more advanced feature would be integration with an external authentication source (e.g. Active Directory, LDAP)

• A “super user” account that can create additional accounts - but these additional accounts cannot create more users
• A better implementation would be to have Role Based Access Control and have “super user” as a role that can be assigned to one or more users

• This may lead, in the future, to a more granular RBAC - e.g. ability to view but not take actions, restriction to specific environments, etc.

• Ability for a “super user” to change a user’s password and/or disable/remove an account
• Read only mode

Proposed change

Use Keystone as authorization tool.

Advantages:

• it can be used with LDAP or AD
• supports authorization via tokens
• support scopes, and groups of users
• has good written api with many functions like getting accessible endpoints for user
• has api easy for consumption
• has implemented events system that we can use in future for additional monitoring
• has implemented multifactor authentication (can be used with external systems)
• all apis that we need for future managing groups, roles, users and project are created
• for UI we can base our solution on horizon solution
• keystone will be also used by Ironic project

Disadvantages:

• need to run in separate container/process
• next external dependency
• may be overkill

We tested keystone with postgresql, it’s working. We tested it via console: create (user, role, tenant, endpoint) add role, get token, list(role, user, tenant) and also we used api to: get user info, get token, operations via v3 api

Blueprint will be implemented in several stages:

Stage I

1. Keystone installation

• install in nailgun container
• run script for creating separate db in postgresql container
• make available ports for keystone
• create base configuration
• in nginx allow connection to keystone

2. GUI

• login page
• logout button

3. After installation set password via fuelmenu (it will be stored in astute.yaml)

Stage II(API protection)

1. Keystone

• create new container for keystone
• create service user for OSTF

2. Nailgun

• Try to use keystone middleware, for api and api v1
• for node agent we should run separate webpy app without middleware

3. GUI

• Add authorization credentials to all requests
  • use keystone token in auth header
  • add handling of 401

4. Change OSTF and Fuel-client and Fuel CLI to use authorization
5. Make authorization optional(flag to enable/disable authorization)

Stage III(all public services protection)

1. Node agent authorization

2.GUI

• change password page

3. Keystone.

• create backup script for db

Stage IV(in unknown future)

1. Many users, groups/roles and api access based on groups/roles (i.e. read-only, network-admin)
2. External authentication (LDAP, AD)

Alternatives

Write everything by yourself or use some existing components:

we need to write user model and apis for creating and managing: user, groups etc oauth, in this case we can reuse some existing libs like oauth2 for creating and consuming tokens. Oauth will be easy to use with clients and node authorization Maybe we can also use sessions for UI to persistence user token

Advantages:

• full control
• possibilities to write good oauth2 authorization easy to use also with nodes

Disadvantages:

• a lot of work on stuff that is already implemented in keystone

Use basic auth in nginx

Advantages:

• really simple to implement, requires only changes in nginx configuration

Disadvantages:

• It shows login page from browser. On every browser it will look little different.
• We can not create custom login page.
• It is still required to implement handlers and tab for password change.
• It’s not extensible. If we want to implement non minimal requirements we need to start from beginning.

Data model impact

New database for keystone is required

REST API impact

Keystone API will be used

Security impact

Fuel will be safer now. It will protect users against unauthorized access. All actions will require authorization.

Notifications impact

Keystone can log all requests to log file.

Other end user impact

• before performing any actions user have to login.
• python-fuelclient should be adjusted to use authorization
• fuel cli should be adjusted to use authorization Password file for fuel-cli? (like .openrc but .fuelrc)

Performance Impact

None

Other deployer impact

Password for postgresql should be generated and access from remote locations should be blocked.

External connections to cobbler and rabbitmq should be allowed. But passwords should be changed to the same as for API even in first version, if possible. In future versions we’ll be able to transfer options for bootstrap node. So we should generate bootstrap ssh key during master node installation. And use password-protected API for nailgun agents.

Developer impact

None

Upgrade impact

There will be new container with keystone installed.

Implementation

Assignee(s)

Primary assignee:

loles@mirantis.com ksambor@mirantis.com

Work Items

Dependencies

None

Testing

Unit tests and functional tests are required.

Acceptance Criteria

1. Stage I
   • After installation user should be able to set password in fuelmenu. and it will be stored in astute.yaml
   • User should be able to login/logut to fuel UI with credentials he set in fuelmenu
   • If user token timeout he should be logged out.
2. Stage II
   • User should see keystone running in separate container. Using command: docker ps
   • All requests, except node agent requests, should be authenticated.
   • User should be able to run nailgun with disabled authorization. It should be done via settings or command line.
   • All requests to ostf should be authenticated
   • all tests should run without any problems
   • fuelclient should use authentication
3. Stage III
   • User should be able to change password via UI page.
   • node agent should use authentication to register in Fuel
4. Stage IV is just group of ideas. No need for acceptance criteria yet.

Documentation Impact

It should be described how to change password and where it’s required.

References

None