blokje5 / iam-authorize Goto Github PK

As the IAM-Authorize server has the concept of namespaces explicitly defined, they should be validated on policy creation and on authorization attempts.

Invalid policies with a non-existent namespace should return an error when a user attempts to create the policy.

Any authorization attempts with an invalid namespace should be blocked. A log message should be used to report the issues, but the error should be kept generic (Forbidden).

There is no PUT endpoint for policies. The PUT endpoint should do the following:

• Update the policy + the associated audit metadata
• Remove all old statements associated with the policy and override them with a the new statements

In order to determine whether a user is authorized to perform an action, we need to evaluate the policies that are loaded in the IAM system. An authorization is evaluated on a tuple containing the user, the action and the object the action is attempted on: <user, action, object>.

For the user we will fetch all associated policies. Each of those policies contain statements that determine whether a certain action is allowed or not. The engine should implicitly deny any attempts. Only if there is an explicit allow will the authorization pass.

If there is an explicit deny it will override an explicit allow.

https://github.com/chef/automate/blob/8535c9a3980a65cfb071e39ed7e6e4f2b8f62d89/components/authz-service/server/policy_refresher.go

Policy Structure:

{
  "version": "v1",
  "statements": [
    {
      "actions": ["iam:CreatePolicy"],
      "resources": "*"
    }
  ]
}

As of yet there is no mechanism to bind a user to a policy. This is needed in order to find which policies apply to which user.