package Maypole::Plugin::Authorization; use strict; use warnings; # This module provides role-based authorization for Maypole our $VERSION = '0.02'; # We can determine whether a given user_id is authorized to invoke a # particular method in a model_class using the following SQL query: my $check_auth_sql = <model_class; my $method = $r->action; my $userid = $r->user->id; # The caller needs to deal with requests with no model class before # calling us because the response is application-specific. If such a # request gets this far, we just turn it down return undef unless $class; # Find a class that can run SQL queries for us and make sure the SQL # query has been prepared my $cdbi_class = $r->config->auth->{user_class}; $cdbi_class->set_sql(check_authorization => $check_auth_sql) unless $cdbi_class->can('sql_check_authorization'); # Check the permissions return $cdbi_class->sql_check_authorization ->select_val($userid, $class, $method); } # Auxiliary methods for finding lists of authorized classes and methods my $get_auth_classes_sql = <user->id; my $cdbi_class = $r->config->auth->{user_class}; $cdbi_class->set_sql(get_authorized_classes => $get_auth_classes_sql) unless $cdbi_class->can('sql_get_authorized_classes'); return $cdbi_class->sql_get_authorized_classes->execute($userid); } my $get_auth_methods_sql = <model_class; my $userid = $r->user->id; my $cdbi_class = $r->config->auth->{user_class}; $cdbi_class->set_sql(get_authorized_methods => $get_auth_methods_sql) unless $cdbi_class->can('sql_get_authorized_methods'); return $cdbi_class->sql_get_authorized_methods->execute($userid, $class); } 1; __END__ =head1 NAME Maypole::Plugin::Authorization - Provide role-based authorization for Maypole applications =head1 SYNOPSIS package BeerDB; use Maypole::Application qw(Authentication::UserSessionCookie Authorization); sub authenticate { my ($self, $r) = @_; ... if $self->authorize($r) { return OK; } else { # take application-specific authorization failure action ... } ... } =head1 DESCRIPTION This module provides simple role-based authorization for Maypole. It uses the database to store permissions, which fits well with Maypole. It determines whether I are authorized to invoke specific I in I. Normally these will be I in model classes. Permission to invoke methods is not granted directly; it is assigned to I, and each user may be assigned one or more roles. =head1 Database structure The module depends on four database tables to store the necessary data. =over =item users The C table records details of each individual who has an account on the system. It is also used by L to do user authentication and session management. Additional columns can be added to suit whatever other needs you have. =item auth_roles Users are not given permissions directly because that causes an explosion in the table size and an administrative headache. Instead roles are given permissions and users acquire those permissions by being assigned to roles. The C table just records the name of the role. You could add things like a description if you wish. The table is not called C so that the name is left free for your application to use. =item role_assignments C is a classic many-many link table. Records contain the id of a user and of a role which the user has been assigned. =item permissions The C table authorizes a specific role to execute a particular method in a particular class. The classes are expected to be the model subclasses and the methods will be the actions, but the scheme will also work in other situations. To reduce administrative burden and table size, it is allowed to use a '*' wildcard instead of a method name; this grants permission to all methods in the class. It would be possible to add a similar wildcard for classes but there's probably no action that you want to allow on B classes! =back The table definitions to implement this scheme look like this: CREATE TABLE users ( id INT NOT NULL AUTO_INCREMENT, name VARCHAR(100) NOT NULL, UID VARCHAR(20) NOT NULL, password VARCHAR(20) NOT NULL, PRIMARY KEY (id), UNIQUE (UID), ) TYPE=InnoDB; CREATE TABLE auth_roles ( id INT NOT NULL AUTO_INCREMENT, name VARCHAR(40) NOT NULL, PRIMARY KEY (id), ) TYPE=InnoDB; CREATE TABLE role_assignments ( id INT NOT NULL AUTO_INCREMENT, user_id INT NOT NULL, auth_role_id INT NOT NULL, PRIMARY KEY (id), UNIQUE (user_id, auth_role_id), INDEX (auth_role_id), FOREIGN KEY (user_id) REFERENCES users (id), FOREIGN KEY (auth_role_id) REFERENCES auth_roles (id), ) TYPE=InnoDB; CREATE TABLE permissions ( id INT NOT NULL AUTO_INCREMENT, auth_role_id INT NOT NULL, model_class VARCHAR(100) NOT NULL, method VARCHAR(100) NOT NULL, PRIMARY KEY (id), UNIQUE (auth_role_id, model_class, method), INDEX (model_class(20)), INDEX (method(20)), FOREIGN KEY (auth_role_id) REFERENCES auth_roles (id), ) TYPE=InnoDB; =head1 Processing We can determine whether a given C is authorized to invoke a particular C in a C using the following SQL query: SELECT p.id FROM permissions AS p, role_assignments AS r WHERE r.user_id = ? AND p.model_class = ? AND (p.method = ? OR p.method = '*') AND p.auth_role_id = r.auth_role_id LIMIT 1 This query is executed in the C method which is called from the driver's C method. (Maypole's terminology is a little confused about authentication and authorization but the code works the same either way!) =head2 authorize The C method is called in the driver's authenticate method, though it is explicitly passed the request object and so can in be called from elsewhere if desired. package BeerDB; sub authenticate { my ($self, $r) = @_; ... if $self->authorize($r) { return OK; } else { # take application-specific auth failure action } ... } It returns a true value if authorization is granted and C if not. =head2 administration The permissions database can be maintained by any person who is assigned to the I role. Most administration is performed using normal Maypole actions and templates such as list, search, addnew, view, edit and delete. User administration is separated out to a I role. I don't yet know whether this will prove beneficial but these people are the only ones who can access passwords and personal details. There needs to be special code to allow users to edit their own passwords, since that is a data-dependent permission as opposed to the metadata-dependent nature of the authorizations scheme. Such code is part of the application's authentication scheme. There is a I role that should be assigned to every user. Perhaps it should be hardwired in the SQL so that users don't have to be actually added to the role? =head1 Use Cases =over =item Create new user User administration mechanisms belong in the domain of the authentication system, though this authorization modules imposes a few additional requirements. This action should be permitted to the user-admin role. Newly created users should automatically be assigned to the 'default' role. =item User changes password Should be permitted to the individual user only and perhaps to the user-admin role. =item Grant/change/revoke user privileges =item Create/delete role =item Alter actions permitted to role People assigned to the admin role can edit the role_assignments table and the auth_roles table in the normal Maypole way. =item Update list of classes =item Update list of methods Presently, administrators need to type in the names of the model subclasses and the actions. The methods C and C could be used to build a specialized template to populate the relevant form elements. =item Determine list of classes This is the C method. Given a user ID, find the list of classes for which s/he has some permissions. This can be used to build the list of tabs in the navbar. =item Determine list of methods This is the C method. Given a user ID and class name, find the list of methods that the user is entitled to invoke. This can be used to build a menu of permitted actions. =back =head1 Alternatives and futures There are several alternative possibilities for authorizable entities and permission checking in addition to the example implementation provided: 1/ Authorize all actions (i.e. methods with the Exported attribute). Permission could be enforced in the model's process method just before calling the action. PRO: simple to implement, uniform and easy-to-understand CON: not as flexible as alternatives 2/ Explicit call to authorize() at the beginning of every method that needs to be authorized. PRO: Flexible. Very simple to implement initially. Obvious in code where auth occurs. Auth can be done at points other than method entry if needed. CON: Error-prone and awkward to maintain. Increases code complexity. 3/ Provide some other attribute that can be attached to methods to require them to be authorized, or perhaps in combination with Exported. For example, the Exported attribute could automatically invoke authorization as would a new 'Auth' attribute, while a new 'NoAuth' attribute would declare that the action could proceed without authorization. =cut