The declarative authorization plugin offers an authorization mechanism inspired by RBAC. The most notable distinction to other authorization plugins is the declarative approach. That is, authorization rules are not defined programmatically in between business logic but in an authorization configuration.

With programmatic authorization rules, the developer needs to specify which roles are allowed to access a specific controller action or a part of a view, which is not DRY. With a growing application code base roles’ permissions often change and new roles are introduced. Then, at several places of the source code the changes have to be implemented, possibly leading to omissions and thus hard to find errors. In these cases, a declarative approach as offered by decl_auth increases the development and maintenance efficiency.

Plugin features

• Authorization at controller action level

• Authorization helpers for Views

• Authorization at model level

  • Authorize CRUD (Create, Read, Update, Delete) activities

  • Query rewriting to automatically only fetch authorized records

• DSL for specifying Authorization rules in an authorization configuration

Requirements

• An authentication mechanism

  • User object in Controller#current_user

  • (For model security) Setting Authorization.current_user

• User objects need to respond to a method :role_symbols that returns an array of role symbols

See below for installation instructions.

There is a decl_auth screencast by Ryan Bates, nicely introducing the main concepts: railscasts.com/episodes/188-declarative-authorization

----- App domain ----|-------- Authorization conf ---------|------- App domain ------

                      includes                   includes
                       .--.                        .---.
                       |  v                        |   v
 .------.  can_play  .------.  has_permission  .------------.  requires  .----------.
 | User |----------->| Role |----------------->| Permission |<-----------| Activity |
 '------' *        * '------' *              * '------------' 1        * '----------'
                                                     |
                                             .-------+------.
                                          1 /        | 1     \ *
                                .-----------.   .---------.  .-----------.
                                | Privilege |   | Context |  | Attribute |
                                '-----------'   '---------'  '-----------'

In the application domain, each User may be assigned to Roles that should define the users’ job in the application, such as Administrator. On the right-hand side of this diagram, application developers specify which Permissions are necessary for users to perform activities, such as calling a controller action, viewing parts of a View or acting on records in the database. Note that Permissions consist of an Privilege that is to be performed, such as read, and a Context in that the Operation takes place, such as companies.

In the authorization configuration, Permissions are assigned to Roles and Role and Permission hierarchies are defined. Attributes may be employed to allow authorization according to dynamic information about the context and the current user, e.g. “only allow access on employees that belong to the current user’s branch.”

A fully functional example application can be found at github.com/stffn/decl_auth_demo_app

Details on the demonstrated methods can be found in the API docs, either generated by yourself or at www.tzi.org/~sbartsch/declarative_authorization

If authentication is in place, there are two ways to enable user-specific access control on controller actions. For resource controllers, which more or less follow the CRUD pattern, filter_resource_access is the simplest approach. It sets up instance variables in before filters and calls filter_access_to with the appropriate parameters to protect the CRUD methods.

class EmployeesController < ApplicationController
  filter_resource_access
  ...
end

See Authorization::AuthorizationInController::ClassMethods for options on nested resources and custom member and collection actions.

If you prefer less magic or your controller has no resemblance with the resource controllers, directly calling filter_access_to may be the better option. Examples are given in the following. E.g. the privilege index users is required for action index. This works as a first default configuration for RESTful controllers, with these privileges easily handled in the authorization configuration, which will be described below.

class EmployeesController < ApplicationController
  filter_access_to :all
  def index
    ...
  end
  ...
end

When custom actions are added to such a controller, it helps to define more clearly which privileges are the respective requirements. That is when the filter_access_to call may become more verbose:

class EmployeesController < ApplicationController
  filter_access_to :all
  # this one would be included in :all, but :read seems to be
  # a more suitable privilege than :auto_complete_for_user_name
  filter_access_to :auto_complete_for_employee_name, :require => :read
  def auto_complete_for_employee_name
    ...
  end
  ...
end

For some actions it might be necessary to check certain attributes of the object the action is to be acting on. Then, the object needs to be loaded before the action’s access control is evaluated. On the other hand, some actions might prefer the authorization to ignore specific attribute checks as the object is unknown at checking time, so attribute checks and thus automatic loading of objects needs to be enabled explicitly.

class EmployeesController < ApplicationController
  filter_access_to :update, :attribute_check => true
  def update
    # @employee is already loaded from param[:id] because of :attribute_check
  end
end

You can provide the needed object through before_filters. This way, you have full control over the object that the conditions are checked against. Just make sure, your before_filters occur before any of the filter_access_to calls.

class EmployeesController < ApplicationController
  before_filter :new_employee_from_params, :only => :create
  before_filter :new_employee, :only => [:index, :new]
  filter_access_to :all, :attribute_check => true

  def create
    @employee.save!
  end

  protected
  def new_employee_from_params
    @employee = Employee.new(params[:employee])
  end
end

If the access is denied, a permission_denied method is called on the current_controller, if defined, and the issue is logged. For further customization of the filters and object loading, have a look at the complete API documentation of filter_access_to in Authorization::AuthorizationInController::ClassMethods.

In views, a simple permitted_to? helper makes showing blocks according to the current user’s privileges easy:

<% permitted_to? :create, :employees do %>
<%= link_to 'New', new_employee_path %>
<% end %>

Only giving a symbol :employees as context prevents any checks of attributes as there is no object to check against. For example, in case of nested resources a new object may come in handy:

<% permitted_to? :create, Branch.new(:company => @company) do
        # or @company.branches.new
        # or even @company.branches %>
<%= link_to 'New', new_company_branch_path(@company) %>
<% end %>

Lists are straight-forward:

<% for employee in @employees %>
<%= link_to 'Edit', edit_employee_path(employee) if permitted_to? :update, employee %>
<% end %>

See also Authorization::AuthorizationHelper.

There are two destinct features for model security built into this plugin: authorizing CRUD operations on objects as well as query rewriting to limit results according to certain privileges.

See also Authorization::AuthorizationInModel.

To activate model security, all it takes is an explicit enabling for each model that model security should be enforced on, i.e.

class Employee < ActiveRecord::Base
  using_access_control
  ...
end

Thus,

Employee.create(...)

fails, if the current user is not allowed to :create :employees according to the authorization rules. For the application to find out about what happened if an operation is denied, the filters throw Authorization::NotAuthorized exceptions.

As access control on read are costly, with possibly lots of objects being loaded at a time in one query, checks on read need to be actived explicitly by adding the :include_read option.

When retrieving large sets of records from databases, any authorization needs to be integrated into the query in order to prevent inefficient filtering afterwards and to use LIMIT and OFFSET in SQL statements. To keep authorization rules out of the source code, this plugin offers query rewriting mechanisms through named scopes. Thus,

Employee.with_permissions_to(:read)

returns all employee records that the current user is authorized to read. In addition, just like normal named scopes, query rewriting may be chained with the usual find method:

Employee.with_permissions_to(:read).find(:all, :conditions => ...)

If the current user is completely missing the permissions, an Authorization::NotAuthorized exception is raised. Through Model.obligation_conditions, application developers may retrieve the conditions for manual rewrites.

Authorization rules are defined in config/authorization_rules.rb (Or redefine rules files path via Authorization::AUTH_RULE_FILES). E.g.

authorization do
  role :admin do
    has_permission_on :employees, :to => [:create, :read, :update, :delete]
  end
end

There is a default role :guest that is used if a request is not associated with any user or with a user without any roles. So, if your application has public pages, :guest can be used to allow access for users that are not logged in. All other roles are application defined and need to be associated with users by the application.

Privileges, such as :create, may be put into hierarchies to simplify maintenance. So the example above has the same meaning as

authorization do
  role :admin do
    has_permission_on :employees, :to => :manage
  end
end

privileges do
  privilege :manage do
    includes :create, :read, :update, :delete
  end
end

Privilege hierarchies may be context-specific, e.g. applicable to :employees.

privileges do
  privilege :manage, :employees, :includes => :increase_salary
end

For more complex use cases, authorizations need to be based on attributes. E.g. if a branch admin should manage only employees of his branch (see Authorization::Reader in the API docs for a full list of available operators):

authorization do
  role :branch_admin do
    has_permission_on :employees do
      to :manage
      # user refers to the current_user when evaluating
      if_attribute :branch => is {user.branch}
    end
  end
end

To reduce redundancy in has_permission_on blocks, a rule may depend on permissions on associated objects:

authorization do
  role :branch_admin do
    has_permission_on :branches, :to => :manage do
      if_attribute :managers => contains {user}
    end

    has_permission_on :employees, :to => :manage do
      if_permitted_to :manage, :branch
      # instead of
      #if_attribute :branch => {:managers => contains {user}}
    end
  end
end

Lastly, not only privileges may be organized in a hierarchy but roles as well. Here, project manager inherit the permissions of employees.

role :project_manager do
  includes :employee
end

See also Authorization::Reader.

declarative_authorization provides a few helpers to ease the testing with authorization in mind.

In your test_helper.rb, to enable the helpers add

require 'declarative_authorization/maintenance'

class Test::Unit::TestCase
  include Authorization::TestHelper
  ...
end

Now, in unit tests, you may deactivate authorization if needed e.g. for test setup and assume certain identities for tests:

class EmployeeTest < ActiveSupport::TestCase
  def test_should_read
    without_access_control do
      Employee.create(...)
    end
    assert_nothing_raised do
      with_user(admin) do
        Employee.find(:first)
      end
    end
  end
end

In functional tests, get, posts, etc. may be tested in the name of certain users:

get_with admin, :index
post_with admin, :update, :employee => {...}

See Authorization::TestHelper for more information.

One of three options to install the plugin:

• Install by Gem: Add to your environment.rb in the initializer block:

  config.gem "declarative_authorization"
  

  Note: you need gemcutter support in place, i.e. call

  gem install gemcutter
  gem tumble
  

  And call from your application’s root directory

  rake gems:install
  

• Alternatively, to install from github, execute in your application’s root directory

  cd vendor/plugins && git clone git://github.com/stffn/declarative_authorization.git

• Or, download one of the released versions from Github at github.com/stffn/declarative_authorization/downloads

Then,

• provide the requirements as noted below,

• create a basic config/authorization_rules.rb–you might want to take the provided example authorization_rules.dist.rb in the plugin root as a starting point,

• add filter_access_to, permitted_to? and model security as needed.

The requirements are

• Rails >= 2.2, including 3 and Ruby >= 1.8.6, including 1.9

• An authentication mechanism

• A user object returned by Controller#current_user

• An array of role symbols returned by User#role_symbols

• (For model security) Setting Authorization.current_user to the request’s user

Of the various ways to provide these requirements, here is one way employing restful_authentication.

• Install restful_authentication

  cd vendor/plugins && git clone git://github.com/technoweenie/restful-authentication.git restful_authentication
  cd ../.. && ruby script/generate authenticated user sessions

• Move “include AuthenticatedSystem” to ApplicationController

• Add filter_access_to calls as described above.

• If you’d like to use model security, add a before_filter that sets the user globally to your ApplicationController. This is thread-safe.

  before_filter :set_current_user
  protected
  def set_current_user
    Authorization.current_user = current_user
  end
  

• Add roles field to the User model through a :has_many association (this is just one possible approach; you could just as easily use :has_many :through or a serialized roles array):

  • create a migration for table roles

    class CreateRoles < ActiveRecord::Migration
      def self.up
        create_table "roles" do |t|
          t.column :title, :string
          t.references :user
        end
      end
    
      def self.down
        drop_table "roles"
      end
    end
    

  • create a model Role,

    class Role < ActiveRecord::Base
      belongs_to :user
    end
    

  • add has_many :roles to the User model and a roles method that returns the roles as an Array of Symbols, e.g.

    class User < ActiveRecord::Base
      has_many :roles
      def role_symbols
        (roles || []).map {|r| r.title.to_sym}
      end
    end
    

  • add roles to your User objects using e.g.

    user.roles.create(:title => "admin")
    

Note: If you choose to generate an Account model for restful_authentication instead of a User model as described above, you have to customize the examples and create a ApplicationController#current_user method.

Currently, the main means of debugging authorization decisions is logging and exceptions. Denied access to actions is logged to warn or info, including some hints about what went wrong.

All bang methods throw exceptions which may be used to retrieve more information about a denied access than a Boolean value.

If your authorization rules become more complex, you might be glad to use the authorization rules browser that comes with declarative_authorization. It has a syntax-highlighted and a graphical view with filtering of the current authorization rules.

By default, it will only be available in development mode. To use it, add the following lines to your authorization_rules.rb for the appropriate role:

has_permission_on :authorization_rules, :to => :read

Then, point your browser to

http://localhost/authorization_rules

The browser needs Rails 2.3 (for Engine support). The graphical view requires Graphviz (which e.g. can be installed through the graphviz package under Debian and Ubuntu) and has only been tested under Linux. Note: for Change Support you’ll need to have a User#login method that returns a non-ambiguous user name for identification.

We have an issue tracker for bugs and feature requests as well as a Google Group for discussions on the usage of the plugin. You are very welcome to contribute. Just fork the git repository and create a new issue, send a pull request or contact me personally.

Maintained by

Steffen Bartsch TZI, Universität Bremen, Germany sbartsch at tzi.org

Thanks to John Joseph Bachir, Eike Carls, Kai Chen, Erik Dahlstrand, Jeroen van Dijk, Sebastian Dyck, Jeremy Friesen, Tim Harper, Daniel Kristensen, Brian Langenfeld, Georg Ledermann, Geoff Longman, Olly Lylo, Mark Mansour, Thomas Maurer, Mike Vincent

Copyright © 2008 Steffen Bartsch, TZI, Universität Bremen, Germany released under the MIT license