Skip to content

asseco-voice/laravel-json-authorization

Repository files navigation

Laravel JSON authorization

This package enables authorization via JSON objects imposed on each model which can be authorized.

Package is developed mainly for the purpose of multiple Laravel microservices authorization having in mind to avoiding the additional trips to authorization service.

This also makes non-auth services self-contained. Authentication service should provide roles ( or any other form of authorization), while services should provide limits that are imposed on any of the roles. Should auth service ever need to be replaced, the only responsibility is to re-map roles on a new auth service, and role limits will stay intact.

Why this approach?

This package offers a great flexibility for imposing rights on Eloquent models. What makes the package unique is the concept switch in a way that you do not want to protect your routes, but rather protecting the resource itself.

This in turn results in two great benefits which the route approach doesn't have out-of-the-box:

  • calling a single endpoint doesn't mean that it operates on a single model, making it impossible for the route approach to do the underlying protection for something which you meant to stay protected.
  • calling a relation on a protected model doesn't protect the related model, so if you're eager/lazy loading something through Eloquent relations you have no way of protecting what is being resolved.

Resource protection here imposes limits you provided independently of where your request comes from. We are doing that by taking advantage of Laravel scopes and Eloquent events.

Of course, there are also some limitations:

  • relation will not be protected if you manually forward a relation_id to model I.e. ContactType has many Contacts. If you impose the right to only update contacts with contact type ID 1, the following will still pass as valid: Contacts::create([... 'contact_type_id' = 2 ...])
  • package will try to authorize early based on the limitations provided, however on complex limits imposed package will make a select on a DB which in some cases may prove to be a heavy action. This mostly affects create/update/delete rights, not read ones.

Installation

Install the package through composer. It is automatically registered as a Laravel service provider, so no additional actions are required to register the package.

composer require asseco-voice/laravel-json-authorization

Terminology

  • calling something authorizable means it is capable of being authorized
  • authorizable set - collection of authorizable user properties (i.e. a collection of roles classifies as an authorizable set)
  • authorizable set value - single object within an authorizable set (i.e. a single role - example_role_1)
  • authorizable set type - logical authorizable sets separation (i.e. you can have a set of roles, set of groups... which would classify as an authorizable set type)
  • authorizable model - model upon which the authorization can be enforced
  • right - a single CRUD right for a single authorizable model, and a single authorizable set value (i.e. having a create right for some model)
  • rule - set of rights for a single authorizable model, and a single authorizable set value

Usage

Package initialization requires few steps to set up:

  1. Pick authorizable models
  2. Migrate tables
  3. Modify User
  4. Attach rules
  5. Flush cache

Pick authorizable models

Models you want protected MUST implement Asseco\JsonAuthorization\App\Traits\Authorizable trait.

After this is done, be sure to run php artisan asseco:sync-authorizable-models to sync models which implement Authorizable trait with the DB.

Run this command each time you add or remove Authorizable trait from a model.

If model already has relation to some rules, the command will throw an exception. This is purposely done to make you manually delete rules for the models you're about to delete, so that it doesn't happen by accident.

Migrate tables

Running php artisan migrate will publish 3 tables:

    authorization_rules ----M:1--- authorizable_models
          |
          |
          M
          -
          1
          |
          |
 authorizable_set_types

authorizable_models - a list of full Eloquent (namespaced) models for authorizable models. This table is filled out automatically upon package usage but is not deleted automatically if you remove the trait after it is already written in the DB. Only models within app folder are scanned. In case you have a different folder structure, or need to implement external models, modify the config models_path variable to include what you need.

authorization_rules - a list of authorizable set values and rules imposed on them.

authorizable_set_types - types represent different sets of things to authorize by. If you are authorizing only by roles, then it makes sense to have only roles there, however there may be cases where you'd like to merge authorizable set values from different authorizable set types in which case you will add those as well.

With regard to the performance, everything is cached to the great extent, invalidated and re-cached upon change.

Seeders are available to use by including AuthorizationSeeder (wrapper for several seeders) within your app DatabaseSeeder. If needed, you can include single seeders from that class as well.

Modify User

User should implement AuthorizesUsers interface which requires you to implement a single method.

The method should return an array of authorizable sets and their values for currently authenticated user.

This needs to reflect names from authorizable_set_types table as array keys, and authorizable set values for each authorizable set type set.

Example:

authorizable_set_types

ID Name
1  roles
2  groups
3  id
public function getAuthorizableSets(): array
{
    return [
        'roles'  => Auth::user()->roles,
        'groups' => Auth::user()->groups,
        'id'     => Auth::user()->id,
    ];
}

You don't need to implement all of these though. This is valid as well (as long as roles are under authorizable_set_types table):

public function getAuthorizableSets(): array
{
    return [
        'roles'  => Auth::user()->roles
    ];
}

Depending on where the set is coming from, you can give it any method which will return an array of things to authorize by:

public function getAuthorizableSets(): array
{
    return [
        'roles'  => $someClass->methodCall(Auth::user()->id, 'https://my-external-service')
    ];
}

Once resolved, function should return for example:

return [
    'roles' => ['role1', 'role2'...],
    'groups' => ['group1', 'group2'...],
    ...
]

It is worth mentioning that final product is merge of role rules.

Example:

role 1: "read" right for IDs 1, 2 and 3
role 2: "read" right for IDs 4, 5 and 6

Final "read" right for that user are IDs 1, 2, 3, 4, 5 and 6

Attach rules

If a model is authorizable, and no limit is present within authorization_rules table for the currently logged in user, we are assuming that user has no rights to operate on the model. You are obligated to explicitly say who has the right for what.

Possible rights are:

  • create
  • read
  • update
  • delete

Each authorizable set value will have a set of rules (in JSON format) for a single model.

Package is built on top of JSON query builder where you can check query logic in depth, with the addition of an absolute right *.

To use the absolute right, you can do:

{
    "read": "*"
}

Giving you a read right to all rows for the given model.

In case you need some sort of admin available which has absolute rights to everything, publish the configuration and add it to the absolute_rights key, and you will not need to give the explicit CRUD rights for it.

Virtual role

If you have the need to protect resources globally or give the permission for a single resource to all users across the system, you can do so by utilizing a virtual role. By default, that role is voice-all-mighty, but can be overridden with .env value VIRTUAL_ROLE.

A virtual role MUST NOT exist as a standard role within your auth service. It will conflict with this and will not work well.

This works in a way that you will i.e. give a read right for some resource to virtual role which will then be inherited by all other users.

Example:

ID  Role               Authorization model ID
1   voice-all-mighty   1                      

Rules
{
	"read": {
		"search": {
			"id": "=1"
		}
	}
} 

Will give a read right to model 1 to all users across the system independently of their system roles.

Flush cache

Due to the heavy workload this package has to do, everything is cached with 1 day TTL. Be sure to flush the cache after each manual code update (i.e. you add Asseco\JsonAuthorization\App\Traits\Authorizable trait on a model).

You can flush the cache the Laravel way, or if you're using Redis as your cache driver you may use one of our packages to enable a wildcard Redis flush.

Example

Let's assume we have the following model protected:

authorizable_models

ID Name
1  App\Contact

Let's impose the rights for a role called agent

authorization_rules

ID  Role    Authorization model ID
1   agent   1                      

Rules
{
	"create": "*",
	"read": {
		"search": {
			"id": "=1;2;3;4;5"
		}
	},
	"update": {
		"search": {
			"id": "=!2"
		}
	}
} 

These rights can be roughly translated as follows:

  • you can create a contact without limitations
  • you can only read contacts with IDs 1, 2, 3, 4 and 5. This means that calling Contact::all() will return only 5 records. Also, calling Contact::find(6) will not return the record.
  • you can update any contact with ID != 2. It is important to say that read right is a top-level right which will in start limit the possible output to 1, 2, 3, 4 and 5 effectively saying that you can update IDs 1, 3, 4, and 5. Others are forbidden through an imposed read right.
  • since delete option is omitted, you have no right for deleting any contact

Extending the package & other

Publishing the configuration will enable you to change package models as well as controlling how migrations behave. If extending the model, make sure you're extending the original model in your implementation.

For dev purposes, you can disable authorization completely by adding this to your .env file:

OVERRIDE_AUTHORIZATION=true