Security

Pyramid provides an optional declarative authorization system that can prevent a view from being invoked based on an authorization policy. Before a view is invoked, the authorization system can use the credentials in the request along with the context resource to determine if access will be allowed. Here’s how it works at a high level:

  • A request is generated when a user visits the application.
  • Based on the request, a context resource is located through resource location. A context is located differently depending on whether the application uses traversal or URL dispatch, but a context is ultimately found in either case. See the URL Dispatch chapter for more information.
  • A view callable is located by view lookup using the context as well as other attributes of the request.
  • If an authentication policy is in effect, it is passed the request; it returns some number of principal identifiers.
  • If an authorization policy is in effect and the view configuration associated with the view callable that was found has a permission associated with it, the authorization policy is passed the context, some number of principal identifiers returned by the authentication policy, and the permission associated with the view; it will allow or deny access.
  • If the authorization policy allows access, the view callable is invoked.
  • If the authorization policy denies access, the view callable is not invoked; instead the forbidden view is invoked.

Security in Pyramid, unlike many systems, cleanly and explicitly separates authentication and authorization. Authentication is merely the mechanism by which credentials provided in the request are resolved to one or more principal identifiers. These identifiers represent the users and groups in effect during the request. Authorization then determines access based on the principal identifiers, the view callable being invoked, and the context resource.

Authorization is enabled by modifying your application to include an authentication policy and authorization policy. Pyramid comes with a variety of implementations of these policies. To provide maximal flexibility, Pyramid also allows you to create custom authentication policies and authorization policies.

Enabling an Authorization Policy

By default, Pyramid enables no authorization policy. All views are accessible by completely anonymous users. In order to begin protecting views from execution based on security settings, you need to enable an authorization policy.

Enabling an Authorization Policy Imperatively

Passing an authorization_policy argument to the constructor of the Configurator class enables an authorization policy.

You must also enable an authentication policy in order to enable the authorization policy. This is because authorization, in general, depends upon authentication. Use the authentication_policy argument to the Configurator class during application setup to specify an authentication policy.

For example:

1
2
3
4
5
6
7
from pyramid.config import Configurator
from pyramid.authentication import AuthTktAuthenticationPolicy
from pyramid.authorization import ACLAuthorizationPolicy
authentication_policy = AuthTktAuthenticationPolicy('seekrit')
authorization_policy = ACLAuthorizationPolicy()
config = Configurator(authentication_policy=authentication_policy,
                      authorization_policy=authorization_policy)

Note

the authentication_policy and authorization_policy arguments may also be passed to the Configurator as dotted Python name values, each representing the dotted name path to a suitable implementation global defined at Python module scope.

The above configuration enables a policy which compares the value of an “auth ticket” cookie passed in the request’s environment which contains a reference to a single principal against the principals present in any ACL found in the resource tree when attempting to call some view.

While it is possible to mix and match different authentication and authorization policies, it is an error to pass an authentication policy without the authorization policy or vice versa to a Configurator constructor.

See also the pyramid.authorization and pyramid.authentication modules for alternate implementations of authorization and authentication policies.

Protecting Views with Permissions

To protect a view callable from invocation based on a user’s security settings when a particular type of resource becomes the context, you must pass a permission to view configuration. Permissions are usually just strings, and they have no required composition: you can name permissions whatever you like.

For example, the following view declaration protects the view named add_entry.html when the context resource is of type Blog with the add permission using the pyramid.config.Configurator.add_view() API:

1
2
3
4
5
6
# config is an instance of pyramid.config.Configurator

config.add_view('mypackage.views.blog_entry_add_view',
                name='add_entry.html',
                context='mypackage.resources.Blog',
                permission='add')

The equivalent view registration including the add permission name may be performed via the @view_config decorator:

1
2
3
4
5
6
7
from pyramid.view import view_config
from resources import Blog

@view_config(context=Blog, name='add_entry.html', permission='add')
def blog_entry_add_view(request):
    """ Add blog entry code goes here """
    pass

As a result of any of these various view configuration statements, if an authorization policy is in place when the view callable is found during normal application operations, the requesting user will need to possess the add permission against the context resource in order to be able to invoke the blog_entry_add_view view. If he does not, the Forbidden view will be invoked.

Setting a Default Permission

If a permission is not supplied to a view configuration, the registered view will always be executable by entirely anonymous users: any authorization policy in effect is ignored.

In support of making it easier to configure applications which are “secure by default”, Pyramid allows you to configure a default permission. If supplied, the default permission is used as the permission string to all view registrations which don’t otherwise name a permission argument.

These APIs are in support of configuring a default permission for an application:

  • The default_permission constructor argument to the Configurator constructor.
  • The pyramid.config.Configurator.set_default_permission() method.

When a default permission is registered:

  • If a view configuration names an explicit permission, the default permission is ignored for that view registration, and the view-configuration-named permission is used.
  • If a view configuration names the permission pyramid.security.NO_PERMISSION_REQUIRED, the default permission is ignored, and the view is registered without a permission (making it available to all callers regardless of their credentials).

Warning

When you register a default permission, all views (even exception view views) are protected by a permission. For all views which are truly meant to be anonymously accessible, you will need to associate the view’s configuration with the pyramid.security.NO_PERMISSION_REQUIRED permission.

Assigning ACLs to your Resource Objects

When the default Pyramid authorization policy determines whether a user possesses a particular permission with respect to a resource, it examines the ACL associated with the resource. An ACL is associated with a resource by adding an __acl__ attribute to the resource object. This attribute can be defined on the resource instance if you need instance-level security, or it can be defined on the resource class if you just need type-level security.

For example, an ACL might be attached to the resource for a blog via its class:

1
2
3
4
5
6
7
8
9
from pyramid.security import Everyone
from pyramid.security import Allow

class Blog(object):
    __acl__ = [
        (Allow, Everyone, 'view'),
        (Allow, 'group:editors', 'add'),
        (Allow, 'group:editors', 'edit'),
        ]

Or, if your resources are persistent, an ACL might be specified via the __acl__ attribute of an instance of a resource:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
from pyramid.security import Everyone
from pyramid.security import Allow

class Blog(object):
    pass

blog = Blog()

blog.__acl__ = [
        (Allow, Everyone, 'view'),
        (Allow, 'group:editors', 'add'),
        (Allow, 'group:editors', 'edit'),
        ]

Whether an ACL is attached to a resource’s class or an instance of the resource itself, the effect is the same. It is useful to decorate individual resource instances with an ACL (as opposed to just decorating their class) in applications such as “CMS” systems where fine-grained access is required on an object-by-object basis.

Elements of an ACL

Here’s an example ACL:

1
2
3
4
5
6
7
8
from pyramid.security import Everyone
from pyramid.security import Allow

__acl__ = [
        (Allow, Everyone, 'view'),
        (Allow, 'group:editors', 'add'),
        (Allow, 'group:editors', 'edit'),
        ]

The example ACL indicates that the pyramid.security.Everyone principal – a special system-defined principal indicating, literally, everyone – is allowed to view the blog, the group:editors principal is allowed to add to and edit the blog.

Each element of an ACL is an ACE or access control entry. For example, in the above code block, there are three ACEs: (Allow, Everyone, 'view'), (Allow, 'group:editors', 'add'), and (Allow, 'group:editors', 'edit').

The first element of any ACE is either pyramid.security.Allow, or pyramid.security.Deny, representing the action to take when the ACE matches. The second element is a principal. The third argument is a permission or sequence of permission names.

A principal is usually a user id, however it also may be a group id if your authentication system provides group information and the effective authentication policy policy is written to respect group information. For example, the pyramid.authentication.RepozeWho1AuthenicationPolicy respects group information if you configure it with a callback.

Each ACE in an ACL is processed by an authorization policy in the order dictated by the ACL. So if you have an ACL like this:

1
2
3
4
5
6
7
8
from pyramid.security import Everyone
from pyramid.security import Allow
from pyramid.security import Deny

__acl__ = [
    (Allow, Everyone, 'view'),
    (Deny, Everyone, 'view'),
    ]

The default authorization policy will allow everyone the view permission, even though later in the ACL you have an ACE that denies everyone the view permission. On the other hand, if you have an ACL like this:

1
2
3
4
5
6
7
8
from pyramid.security import Everyone
from pyramid.security import Allow
from pyramid.security import Deny

__acl__ = [
    (Deny, Everyone, 'view'),
    (Allow, Everyone, 'view'),
    ]

The authorization policy will deny everyone the view permission, even though later in the ACL is an ACE that allows everyone.

The third argument in an ACE can also be a sequence of permission names instead of a single permission name. So instead of creating multiple ACEs representing a number of different permission grants to a single group:editors group, we can collapse this into a single ACE, as below.

1
2
3
4
5
6
7
from pyramid.security import Everyone
from pyramid.security import Allow

__acl__ = [
    (Allow, Everyone, 'view'),
    (Allow, 'group:editors', ('add', 'edit')),
    ]

Special Principal Names

Special principal names exist in the pyramid.security module. They can be imported for use in your own code to populate ACLs, e.g. pyramid.security.Everyone.

pyramid.security.Everyone

Literally, everyone, no matter what. This object is actually a string “under the hood” (system.Everyone). Every user “is” the principal named Everyone during every request, even if a security policy is not in use.

pyramid.security.Authenticated

Any user with credentials as determined by the current security policy. You might think of it as any user that is “logged in”. This object is actually a string “under the hood” (system.Authenticated).

Special Permissions

Special permission names exist in the pyramid.security module. These can be imported for use in ACLs.

pyramid.security.ALL_PERMISSIONS

An object representing, literally, all permissions. Useful in an ACL like so: (Allow, 'fred', ALL_PERMISSIONS). The ALL_PERMISSIONS object is actually a stand-in object that has a __contains__ method that always returns True, which, for all known authorization policies, has the effect of indicating that a given principal “has” any permission asked for by the system.

Special ACEs

A convenience ACE is defined representing a deny to everyone of all permissions in pyramid.security.DENY_ALL. This ACE is often used as the last ACE of an ACL to explicitly cause inheriting authorization policies to “stop looking up the traversal tree” (effectively breaking any inheritance). For example, an ACL which allows only fred the view permission for a particular resource despite what inherited ACLs may say when the default authorization policy is in effect might look like so:

1
2
3
4
from pyramid.security import Allow
from pyramid.security import DENY_ALL

__acl__ = [ (Allow, 'fred', 'view'), DENY_ALL ]

“Under the hood”, the pyramid.security.DENY_ALL ACE equals the following:

1
2
from pyramid.security import ALL_PERMISSIONS
__acl__ = [ (Deny, Everyone, ALL_PERMISSIONS) ]

ACL Inheritance and Location-Awareness

While the default authorization policy is in place, if a resource object does not have an ACL when it is the context, its parent is consulted for an ACL. If that object does not have an ACL, its parent is consulted for an ACL, ad infinitum, until we’ve reached the root and there are no more parents left.

In order to allow the security machinery to perform ACL inheritance, resource objects must provide location-awareness. Providing location-awareness means two things: the root object in the resource tree must have a __name__ attribute and a __parent__ attribute.

1
2
3
class Blog(object):
    __name__ = ''
    __parent__ = None

An object with a __parent__ attribute and a __name__ attribute is said to be location-aware. Location-aware objects define an __parent__ attribute which points at their parent object. The root object’s __parent__ is None.

See pyramid.location for documentations of functions which use location-awareness. See also Location-Aware Resources.

Changing the Forbidden View

When Pyramid denies a view invocation due to an authorization denial, the special forbidden view is invoked. “Out of the box”, this forbidden view is very plain. See Changing the Forbidden View within Using Hooks for instructions on how to create a custom forbidden view and arrange for it to be called when view authorization is denied.

Debugging View Authorization Failures

If your application in your judgment is allowing or denying view access inappropriately, start your application under a shell using the PYRAMID_DEBUG_AUTHORIZATION environment variable set to 1. For example:

$ PYRAMID_DEBUG_AUTHORIZATION=1 bin/paster serve myproject.ini

When any authorization takes place during a top-level view rendering, a message will be logged to the console (to stderr) about what ACE in which ACL permitted or denied the authorization based on authentication information.

This behavior can also be turned on in the application .ini file by setting the pyramid.debug_authorization key to true within the application’s configuration section, e.g.:

1
2
3
[app:main]
use = egg:MyProject
pyramid.debug_authorization = true

With this debug flag turned on, the response sent to the browser will also contain security debugging information in its body.

Debugging Imperative Authorization Failures

The pyramid.security.has_permission() API is used to check security within view functions imperatively. It returns instances of objects that are effectively booleans. But these objects are not raw True or False objects, and have information attached to them about why the permission was allowed or denied. The object will be one of pyramid.security.ACLAllowed, pyramid.security.ACLDenied, pyramid.security.Allowed, or pyramid.security.Denied, as documented in pyramid.security. At the very minimum these objects will have a msg attribute, which is a string indicating why the permission was denied or allowed. Introspecting this information in the debugger or via print statements when a call to has_permission() fails is often useful.

Creating Your Own Authentication Policy

Pyramid ships with a number of useful out-of-the-box security policies (see pyramid.authentication). However, creating your own authentication policy is often necessary when you want to control the “horizontal and vertical” of how your users authenticate. Doing so is a matter of creating an instance of something that implements the following interface:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
class IAuthenticationPolicy(object):
    """ An object representing a Pyramid authentication policy. """

    def authenticated_userid(self, request):
        """ Return the authenticated userid or ``None`` if no
        authenticated userid can be found. This method of the policy
        should ensure that a record exists in whatever persistent store is
        used related to the user (the user should not have been deleted);
        if a record associated with the current id does not exist in a
        persistent store, it should return ``None``."""

    def unauthenticated_userid(self, request):
        """ Return the *unauthenticated* userid.  This method performs the
        same duty as ``authenticated_userid`` but is permitted to return the
        userid based only on data present in the request; it needn't (and
        shouldn't) check any persistent store to ensure that the user record
        related to the request userid exists."""

    def effective_principals(self, request):
        """ Return a sequence representing the effective principals
        including the userid and any groups belonged to by the current
        user, including 'system' groups such as
        ``pyramid.security.Everyone`` and
        ``pyramid.security.Authenticated``. """

    def remember(self, request, principal, **kw):
        """ Return a set of headers suitable for 'remembering' the
        principal named ``principal`` when set in a response.  An
        individual authentication policy and its consumers can decide
        on the composition and meaning of **kw. """

    def forget(self, request):
        """ Return a set of headers suitable for 'forgetting' the
        current user on subsequent requests. """

After you do so, you can pass an instance of such a class into the Configurator class at configuration time as authentication_policy to use it.

Creating Your Own Authorization Policy

An authorization policy is a policy that allows or denies access after a user has been authenticated. By default, Pyramid will use the pyramid.authorization.ACLAuthorizationPolicy if an authentication policy is activated and an authorization policy isn’t otherwise specified.

In some cases, it’s useful to be able to use a different authorization policy than the default ACLAuthorizationPolicy. For example, it might be desirable to construct an alternate authorization policy which allows the application to use an authorization mechanism that does not involve ACL objects.

Pyramid ships with only a single default authorization policy, so you’ll need to create your own if you’d like to use a different one. Creating and using your own authorization policy is a matter of creating an instance of an object that implements the following interface:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
class IAuthorizationPolicy(object):
    """ An object representing a Pyramid authorization policy. """
    def permits(self, context, principals, permission):
        """ Return ``True`` if any of the ``principals`` is allowed the
        ``permission`` in the current ``context``, else return ``False``
        """

    def principals_allowed_by_permission(self, context, permission):
        """ Return a set of principal identifiers allowed by the
        ``permission`` in ``context``.  This behavior is optional; if you
        choose to not implement it you should define this method as
        something which raises a ``NotImplementedError``.  This method
        will only be called when the
        ``pyramid.security.principals_allowed_by_permission`` API is
        used."""

After you do so, you can pass an instance of such a class into the Configurator class at configuration time as authorization_policy to use it.