Skip to main content

Declarative access policies/permissions modeled after AWS' IAM policies.

Project description

Django REST - Access Policy

Package version Python versions

This project brings a declaritive, organized approach to managing access control in Django REST Framework projects. Each ViewSet or function-based view can be assigned an explicit policy for the exposed resource(s). No more digging through views or seralizers to understand access logic -- it's all in one place in a format that less technical stakeholders can understand. If you're familiar with other declaritive access models, such as AWS' IAM, the syntax will be familiar.

In short, you can start expressing your access rules like this:

class ArticleAccessPolicy(AccessPolicy):
    statements = [
        {
            "action": ["list", "retrieve"],
            "principal": "*",
            "effect": "allow"
        },
        {
            "action": ["publish", "unpublish"],
            "principal": ["group:editor"],
            "effect": "allow"            
        }
    ]

This project has complete test coverage and the base AccessPolicy class is only ~150 lines of code: there's no magic here.

Table of Contents:

Setup

pip install drf-access-policy

To define a policy, import AccessPolicy and subclass it:

from rest_access_policy import AccessPolicy


class ShoppingCartAccessPolicy(AccessPolicy):
    statements = [] # Now read on...

Example #1: Policy for ViewSet

In a nutshell, a policy is comprised of "statements" that declare what "actions" a "principal" can or cannot perform on the resource, with optional custom checks that can examine any detail of the current request.

Here are two more key points to remember going forward:

  • all access is implicitly denied by default
  • any statement with the "deny" effect overrides any and all "allow" statement

Now let's look at the policy below an articles endpoint, provided through a view set.

class ArticleAccessPolicy(AccessPolicy):
    statements = [
        {
            "action": ["list", "retrieve"],
            "principal": "*",
            "effect": "allow"
        },
        {
            "action": ["publish", "unpublish"],
            "principal": ["group:editor"],
            "effect": "allow"            
        },
        {
            "action": ["delete"],
            "principal": ["*"],
            "effect": "allow",
            "condition": "is_author"         
        },
        {
            "action": ["*"],
            "principal": ["*"],
            "effect": "deny",
            "condition": "is_happy_hour"
        }
    ]

    def is_author(self, request, view, action) -> bool:
        article = view.get_object()
        return request.user == article.author 

    def is_happy_hour(self, request, view, action) -> bool:
        now = datetime.datetime.now()
        return now.hour >= 17 and now.hour <= 18:

    @classmethod
    def scope_queryset(cls, request, queryset):
        if request.user.groups.filter(name='editor').exists():
            return queryset

        return queryset.filter(status='published')


class ArticleViewSet(ModelViewSet):
    # Just stick the policy here, as you would do with
    # regular DRF "permissions"
    permission_classes = (ArticleAccessPolicy, )

    # Helper property here to make get_queryset logic
    # more explicit
    @property
    def access_policy(self):
        return self.permission_classes[0]

    # Ensure that current user can only see the models 
    # they are allowed to see
    def get_queryset(self):
        return self.access_policy.scope_queryset(
            self.request, Articles.objects.all()
        )
    
    @action(method="POST")
    def publish(self, request, *args, **kwargs):
        pass

    @action(method="POST")
    def unpublish(self, request, *args, **kwargs):
        pass

    # the rest of you view set definition...

The actions correspond to the names of methods on the ViewSet.

In the example above, the following rules are put in place:

  • anyone is allowed to list and retrieve articles
  • users in the editor group are allowed to publish and unpublish articles
  • in order to delete an article, the user must be the author of the article. Notice how the condition method is_author calls get_object() on the view to get the current article.
  • if the condition is_happy_hour, evaluates to True, then no one is allowed to do anything.

Additionally, we have some logic in the scope_queryset method for filtering which models are visible to the current user. Here, we want users to only see published articles, unless they are an editor, in which case they case see articles with any status. You have to remember to call this method from the view, so I'd suggest reviewing this as part of a security audit checklist.

Example #2: Policy for Function-Based View

You can also you policies with function-based views. The action to reference in your policy statements is the name of the function. You can also bundle multiple functions into the same policy as the example below shows.

class AuditLogsAccessPolicy(AccessPolicy):
    statements = [
        {
            "action": ["search_logs"],
            "principal": "group:it_staff",
            "effect": "allow"
        },
        {
            "action": ["download_logs"],
            "principal": ["group:it_admin"],
            "effect": "allow"            
        }
    ]


@api_view(["GET"])
@permission_classes((AuditLogsAccessPolicy,))
def search_logs(request):
    ## you logic here...
    pass


@api_view(["GET"])
@permission_classes((AuditLogsAccessPolicy,))
def download_logs(request):
    ## you logic here...
    pass

Documentation

Statement Elements

principal

Description Should match the user of the current request by identifying a group they belong to or their user ID.
Special Values "*" (any user)
"authenticated" (any authenticated user)
"anonymous" (any non-authenticated user)
Type Union[str, List[str]]
Format Match by group with "group:{name}"
Match by ID with "id:{id}"
Examples ["group:admins", "id:9322"]
["id:5352"]
["anonymous"]
"*"

action

Description The action or actions that the statement applies to. The value should match the name of a view set method or the name of the view function.
Type Union[str, List[str]]
Special Values "*" (any action)
"<safe_methods>" (a read-only HTTP request: HEAD, GET, OPTIONS)
Examples ["list", "delete", "create]
["*"]
["<safe_methods>"]

effect

Description Whether the statement, if it is in effect, should allow or deny access. All access is denied by default, so use deny when you'd like to override an allow statement that will also be in effect.
Type str
Values Either "allow" or "deny"

condition

Description The name of a method on the policy that returns a boolean. The method signature is condition(request, view, action: str, custom_arg: str=None). If you want to pass a custom argument to the condition's method, format the value as {method_name}:{value}, e.g. user_must_be:owner will call a method named user_must_be, passing it the string "owner" as the final argument. If true, the policy will be in effect. Useful for enforcing object-level permissions. If list of conditions is given, all conditions must evaluate to True.
Type Union[str, List[str]]
Examples "is_manager_of_account"
"is_author_of_post"
["balance_is_positive", "account_is_not_frozen"]`
"user_must_be:account_manager"

Policy Evaluation Logic

To determine whether access to a request is granted, two steps are applied: (1) finding out which statements apply to the request (2) denying or allowing the request based on those statements.

  • Filtering statements: A statement is applicable to the current request if all of the following are true (a) the request user matches one of the statement's principals, (b) the name of the method/function matches one of its actions, and (c) all custom conditions evaluate to true.
  • Allow or deny: The request is allowed if any of the statements have an effect of "allow", and none have an effect of "deny". By default, all requests are denied. Requests are implicitly denied if no Allow statements are found, and they are explicitly denied if any Deny statements are found. Deny statements trump Allow statements.

Object-Level Permissions/Custom Conditions

What object-level permissions? You can easily check object-level access in a custom condition that's evaluated to determine whether the statement takes effect. This condition is passed the view instance, so you can get the model instance with a call to view.get_object(). You can even reference multiple conditions, to keep your access methods focused and testable, as well as parametrize these conditions with arguments.

class AccountAccessPolicy(AccessPolicy):
    statements = [
        ## ... other statements ...
        {
            "action": ["withdraw"],
            "principal": ["*"],
            "effect": "allow",
            "condition": ["balance_is_positive", "user_must_be:owner"]     
        },
        {
            "action": ["upgrade_to_gold_status"],
            "principal": ["*"],
            "effect": "allow",
            "condition": ["user_must_be:account_advisor"]
        }
        ## ... other statements ...
    ]

    def balance_is_positive(self, request, view, action) -> bool:
        account = view.get_object()
        return account.balance > 0

    def user_must_be(self, request, view, action, field: str) -> bool:
        account = view.get_object()
        return getattr(account, field) == request.user

Notice how we're re-using the user_must_be method by parameterizing it with the model field that should be equal fo the user of the request: the statement will only be effective if this condition passes.

Re-Usable Conditions/Permissions

If you'd like to re-use custom conditions across policies, you can define them globally in a module and point to it via the setttings.

# in your project settings.py

DRF_ACCESS_POLICY = {"reusable_conditions": "myproject.global_access_conditions"}
# in myproject.global_access_conditions.py

def is_the_weather_nice(request, view, action: str) -> bool:
    data = weather_api.load_today()
    return data["temperature"] > 68

def user_must_be(self, request, view, action, field: str) -> bool:
    account = view.get_object()
    return getattr(account, field) == request.user

The policy class will first check its own methods for what's been defined in the condition property. If nothing is found, it will check the module defined in the reusable_conditions setting.

Multitenancy Data / Restricting QuerySets

You can define a class method on your policy class that takes a QuerySet and the current request and returns a securely scoped QuerySet representing only the database rows that the current user should have access to. This is helpful for multitenant situations or more generally when users should not have full visibility to model instances. Of course you could do this elsewhere in your code, but putting this method on the policy class keeps all access logic in a single place.

    class PhotoAlbumAccessPolicy(AccessPolicy):
        # ... statements, etc ...

        # Users can only access albums they have created
        @classmethod
        def scope_queryset(cls, request, qs):
            return qs.filter(creator=request.user)


    class TodoListAccessPolicy(AccessPolicy):
        # ... statements, etc ...

        # Users can only access todo lists owned by their organization
        @classmethod
        def scope_queryset(cls, request, qs):
            user_orgs = request.user.organizations.all()
            return qs.filter(org__id__in=user_orgs)

Attaching to ViewSets and Function-Based Views

You attach access policies the same way you do with regular DRF permissions.

For ViewSets, add it to permissions property:

class ArticleViewSet(ModelViewSet):
    permission_classes = (ArticleAccessPolicy, )

For function-based views, add it to permissions_classes decorator:

@api_view(["GET"])
@permission_classes((ArticleAccessPolicy,))
def create_article(request):
    ## you logic here...
    pass

Loading Statements from External Source

If you don't want your policy statements hardcoded into the classes, you can load them from an external data source: a great step to take because you can then change access rules without redeploying code.

Just define a method on your policy class called get_policy_statements, which has the following signature: get_policy_statements(self, request, view) -> List[dict]

Example:

class UserAccessPolicy(AccessPolicy):
    id = 'user-policy'

    def get_policy_statements(self, request, view) -> List[dict]:
        statements = data_api.load_json(self.id)
        return json.loads(statements)

You probably want to only define this method once on your own custom subclass of AccessPolicy, from which all your other access policies inherit.

Customizing User Group/Role Values

If you aren't using Django's built-in auth app, you may need to define a custom way to retrieve the role/group names to which the user belongs. Just define a method called get_user_group_values on your policy class. It is passed a single argument: the user of the current request. In the example below, the user model has a to-many relationship with a "roles", which have their "name" value in a field called "title".

class UserAccessPolicy(AccessPolicy):
    # ... other properties and methods ...

    def get_user_group_values(self, user) -> List[str]:
        return list(user.roles.values_list("title", flat=True))

Customizing Principal Prefixes

By default, the prefixes to identify the type of principle (user or group) are "id:" and "group:", respectively. You can customize this by setting these properties on your policy class:

class FriendRequestPolicy(permissions.BasePermission):
    group_prefix = "role:"
    id_prefix = "staff_id:"

    # .. the rest of you policy definition ..

Changelog

0.5.0 (September 2019)

  • Add option to define re-usable custom conditions/permissions in a module that can be referenced by multiple policies.

0.4.2 (June 2019)

  • Fixes readme format for Pypy display.

0.4.0 (June 2019)

  • Allow passing arguments to condition methods, via condition values formatted as {method_name}:{arg_value}.

0.3.0 (May 2019)

  • Adds special <safe_methods> action key that matches when the current request is an HTTP read-only method: HEAD, GET, OPTIONS.

0.2.0 (May 2019)

  • Adds special authenticated and anonymous principal keys to match any authenticated user and any non-authenticated user, respectively. Thanks @bogdandm for discussion/advice!

0.1.0 (May 2019)

  • Initial release

Testing

Tests are found in a simplified Django project in the /tests folder. Install the project requirements and do ./manage.py test to run them.

License

See License.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for drf-access-policy, version 0.5.0
Filename, size File type Python version Upload date Hashes
Filename, size drf-access-policy-0.5.0.tar.gz (11.2 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page