Fetching contributors…
Cannot retrieve contributors at this time
980 lines (726 sloc) 36.4 KB
.. index::
   single: security


:app:`Pyramid` provides an optional, declarative, security system. Security in :app:`Pyramid` is separated into authentication and authorization. The two systems communicate via :term:`principal` identifiers. Authentication is merely the mechanism by which credentials provided in the :term:`request` are resolved to one or more :term:`principal` identifiers. These identifiers represent the users and groups that are in effect during the request. Authorization then determines access based on the :term:`principal` identifiers, the requested :term:`permission`, and a :term:`context`.

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

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

.. index::
   single: authorization policy

Enabling an Authorization Policy

:app:`Pyramid` does not enable any authorization policy by default. 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

Use the :meth:`~pyramid.config.Configurator.set_authorization_policy` method of the :class:`~pyramid.config.Configurator` to enable an authorization policy.

You must also enable an :term:`authentication policy` in order to enable the authorization policy. This is because authorization, in general, depends upon authentication. Use the :meth:`~pyramid.config.Configurator.set_authentication_policy` method during application setup to specify the authentication policy.

For example:


The authentication_policy and authorization_policy arguments may also be passed to their respective methods mentioned above as :term:`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 :term:`userid`, and matches that userid's :term:`principals <principal>` against the principals present in any :term:`ACL` found in the resource tree when attempting to call some :term:`view`.

While it is possible to mix and match different authentication and authorization policies, it is an error to configure a Pyramid application with an authentication policy but without the authorization policy or vice versa. If you do this, you'll receive an error at application startup time.

.. seealso::

    See also the :mod:`pyramid.authorization` and :mod:`pyramid.authentication`
    modules for alternative implementations of authorization and authentication

.. index::
   single: permissions
   single: protecting views

Protecting Views with Permissions

To protect a :term:`view callable` from invocation based on a user's security settings when a particular type of resource becomes the :term:`context`, you must pass a :term:`permission` to :term:`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 :meth:`pyramid.config.Configurator.add_view` API:

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

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 :term:`context` resource in order to be able to invoke the blog_entry_add_view view. If they do not, the :term:`Forbidden view` will be invoked.

.. index::
   pair: permission; default

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", :app:`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.

The :meth:`pyramid.config.Configurator.set_default_permission` method supports configuring a default permission for an application.

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 :data:``, the default permission is ignored, and the view is registered without a permission (making it available to all callers regardless of their credentials).


When you register a default permission, all views (even :term:`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 :data:`` permission.

.. index::
   single: ACL
   single: access control list
   pair: resource; ACL

Assigning ACLs to Your Resource Objects

When the default :app:`Pyramid` :term:`authorization policy` determines whether a user possesses a particular permission with respect to a resource, it examines the :term:`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:

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

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 content management systems where fine-grained access is required on an object-by-object basis.

Dynamic ACLs are also possible by turning the ACL into a callable on the resource. This may allow the ACL to dynamically generate rules based on properties of the instance.


Writing __acl__ as properties is discouraged because an AttributeError occurring in fget or fset will be silently dismissed (this is consistent with Python getattr and hasattr behaviors). For dynamic ACLs, simply use callables, as documented above.

.. index::
   single: ACE
   single: access control entry

Elements of an ACL

Here's an example ACL:

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

Each element of an ACL is an :term:`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 :data:``, or :data:``, representing the action to take when the ACE matches. The second element is a :term:`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 :term:`authentication policy` policy is written to respect group information. See :ref:`extending_default_authentication_policies`.

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:

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:

The authorization policy will deny everyone the view permission, even though later in the ACL, there 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.

.. index::
   single: principal
   single: principal names

Special Principal Names

Special principal names exist in the :mod:`` module. They can be imported for use in your own code to populate ACLs, e.g., :data:``.


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.


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).
.. index::
   single: permission names
   single: special permission names

Special Permissions

Special permission names exist in the :mod:`` module. These can be imported for use in ACLs.


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.
.. index::
   single: special ACE
   single: ACE (special)

Special ACEs

A convenience :term:`ACE` is defined representing a deny to everyone of all permissions in :data:``. 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:

Under the hood, the :data:`` ACE equals the following:

.. index::
   single: ACL inheritance
   pair: location-aware; security

ACL Inheritance and Location-Awareness

While the default :term:`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.

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

.. seealso::

    See also :ref:`location_module` for documentations of functions which use

.. seealso::

    See also :ref:`location_aware`.

.. index::
   single: forbidden view

Changing the Forbidden View

When :app:`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 :ref:`changing_the_forbidden_view` within :ref:`hooks_chapter` for instructions on how to create a custom forbidden view and arrange for it to be called when view authorization is denied.

.. index::
   single: debugging authorization failures

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 $VENV/bin/pserve 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.:

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 :meth:`pyramid.request.Request.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 :data:``, :data:``, :data:``, or :data:``, as documented in :ref:`security_module`. 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 :meth:`~pyramid.request.Request.has_permission` fails is often useful.

.. index::
   single: authentication policy (extending)

Extending Default Authentication Policies

Pyramid ships with some built in authentication policies for use in your applications. See :mod:`pyramid.authentication` for the available policies. They differ on their mechanisms for tracking authentication credentials between requests, however they all interface with your application in mostly the same way.

Above you learned about :ref:`assigning_acls`. Each :term:`principal` used in the :term:`ACL` is matched against the list returned from :meth:`pyramid.interfaces.IAuthenticationPolicy.effective_principals`. Similarly, :meth:`pyramid.request.Request.authenticated_userid` maps to :meth:`pyramid.interfaces.IAuthenticationPolicy.authenticated_userid`.

You may control these values by subclassing the default authentication policies. For example, below we subclass the :class:`pyramid.authentication.AuthTktAuthenticationPolicy` and define extra functionality to query our database before confirming that the :term:`userid` is valid in order to avoid blindly trusting the value in the cookie (what if the cookie is still valid, but the user has deleted their account?). We then use that :term:`userid` to augment the effective_principals with information about groups and other state for that user.

In most instances authenticated_userid and effective_principals are application-specific, whereas unauthenticated_userid, remember, and forget are generic and focused on transport and serialization of data between consecutive requests.

.. index::
   single: authentication policy (creating)

Creating Your Own Authentication Policy

:app:`Pyramid` ships with a number of useful out-of-the-box security policies (see :mod:`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:

After you do so, you can pass an instance of such a class into the :class:`~pyramid.config.Configurator.set_authentication_policy` method at configuration time to use it.

.. index::
   single: authorization policy (creating)

Creating Your Own Authorization Policy

An authorization policy is a policy that allows or denies access after a user has been authenticated. Most :app:`Pyramid` applications will use the default :class:`pyramid.authorization.ACLAuthorizationPolicy`.

However, in some cases, it's useful to be able to use a different authorization policy than the default :class:`~pyramid.authorization.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 :term:`ACL` objects.

:app:`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:

After you do so, you can pass an instance of such a class into the :class:`~pyramid.config.Configurator.set_authorization_policy` method at configuration time to use it.

Admonishment Against Secret-Sharing

A "secret" is required by various components of Pyramid. For example, the :term:`authentication policy` below uses a secret value seekrit:

authn_policy = AuthTktAuthenticationPolicy('seekrit', hashalg='sha512')

A :term:`session factory` also requires a secret:

my_session_factory = SignedCookieSessionFactory('itsaseekreet')

It is tempting to use the same secret for multiple Pyramid subsystems. For example, you might be tempted to use the value seekrit as the secret for both the authentication policy and the session factory defined above. This is a bad idea, because in both cases, these secrets are used to sign the payload of the data.

If you use the same secret for two different parts of your application for signing purposes, it may allow an attacker to get his chosen plaintext signed, which would allow the attacker to control the content of the payload. Re-using a secret across two different subsystems might drop the security of signing to zero. Keys should not be re-used across different contexts where an attacker has the possibility of providing a chosen plaintext.

.. index::
   single: preventing cross-site request forgery attacks
   single: cross-site request forgery attacks, prevention

Preventing Cross-Site Request Forgery Attacks

Cross-site request forgery attacks are a phenomenon whereby a user who is logged in to your website might inadvertantly load a URL because it is linked from, or embedded in, an attacker's website. If the URL is one that may modify or delete data, the consequences can be dire.

You can avoid most of these attacks by issuing a unique token to the browser and then requiring that it be present in all potentially unsafe requests. :app:`Pyramid` provides facilities to create and check CSRF tokens.

By default :app:`Pyramid` comes with a session-based CSRF implementation :class:`pyramid.csrf.SessionCSRFStoragePolicy`. To use it, you must first enable a :term:`session factory` as described in :ref:`using_the_default_session_factory` or :ref:`using_alternate_session_factories`. Alternatively, you can use a cookie-based implementation :class:`pyramid.csrf.CookieCSRFStoragePolicy` which gives some additional flexibility as it does not require a session for each user. You can also define your own implementation of :class:`pyramid.interfaces.ICSRFStoragePolicy` and register it with the :meth:`pyramid.config.Configurator.set_csrf_storage_policy` directive.

For example:

from pyramid.config import Configurator

config = Configurator()
.. index::
   single: csrf.get_csrf_token

Using the csrf.get_csrf_token Method

To get the current CSRF token, use the :data:`pyramid.csrf.get_csrf_token` method.

from pyramid.csrf import get_csrf_token
token = get_csrf_token(request)

The get_csrf_token() method accepts a single argument: the request. It returns a CSRF token string. If get_csrf_token() or new_csrf_token() was invoked previously for this user, then the existing token will be returned. If no CSRF token previously existed for this user, then a new token will be set into the session and returned. The newly created token will be opaque and randomized.

Using the get_csrf_token global in templates

Templates have a get_csrf_token() method inserted into their globals, which allows you to get the current token without modifying the view code. This method takes no arguments and returns a CSRF token string. You can use the returned token as the value of a hidden field in a form that posts to a method that requires elevated privileges, or supply it as a request header in AJAX requests.

For example, include the CSRF token as a hidden field:

<form method="post" action="/myview">
  <input type="hidden" name="csrf_token" value="${get_csrf_token()}">
  <input type="submit" value="Delete Everything">

Or include it as a header in a jQuery AJAX request:

var csrfToken = "${get_csrf_token()}";
  type: "POST",
  url: "/myview",
  headers: { 'X-CSRF-Token': csrfToken }
}).done(function() {

The handler for the URL that receives the request should then require that the correct CSRF token is supplied.

.. index::
   single: csrf.new_csrf_token

Using the csrf.new_csrf_token Method

To explicitly create a new CSRF token, use the csrf.new_csrf_token() method. This differs only from csrf.get_csrf_token() inasmuch as it clears any existing CSRF token, creates a new CSRF token, sets the token into the user, and returns the token.

from pyramid.csrf import new_csrf_token
token = new_csrf_token(request)


It is not possible to force a new CSRF token from a template. If you want to regenerate your CSRF token then do it in the view code and return the new token as part of the context.

Checking CSRF Tokens Manually

In request handling code, you can check the presence and validity of a CSRF token with :func:`pyramid.csrf.check_csrf_token`. If the token is valid, it will return True, otherwise it will raise HTTPBadRequest. Optionally, you can specify raises=False to have the check return False instead of raising an exception.

By default, it checks for a POST parameter named csrf_token or a header named X-CSRF-Token.

from pyramid.csrf import check_csrf_token

def myview(request):
    # Require CSRF Token

    # ...

Checking CSRF Tokens Automatically

.. versionadded:: 1.7

:app:`Pyramid` supports automatically checking CSRF tokens on requests with an unsafe method as defined by RFC2616. Any other request may be checked manually. This feature can be turned on globally for an application using the :meth:`pyramid.config.Configurator.set_default_csrf_options` directive. For example:

from pyramid.config import Configurator

config = Configurator()

CSRF checking may be explicitly enabled or disabled on a per-view basis using the require_csrf view option. A value of True or False will override the default set by set_default_csrf_options. For example:

@view_config(route_name='hello', require_csrf=False)
def myview(request):
    # ...

When CSRF checking is active, the token and header used to find the supplied CSRF token will be csrf_token and X-CSRF-Token, respectively, unless otherwise overridden by set_default_csrf_options. The token is checked against the value in request.POST which is the submitted form body. If this value is not present, then the header will be checked.

In addition to token based CSRF checks, if the request is using HTTPS then the automatic CSRF checking will also check the referrer of the request to ensure that it matches one of the trusted origins. By default the only trusted origin is the current host, however additional origins may be configured by setting pyramid.csrf_trusted_origins to a list of domain names (and ports if they are non-standard). If a host in the list of domains starts with a . then that will allow all subdomains as well as the domain without the ..

If CSRF checks fail then a :class:`pyramid.exceptions.BadCSRFToken` or :class:`pyramid.exceptions.BadCSRFOrigin` exception will be raised. This exception may be caught and handled by an :term:`exception view` but, by default, will result in a 400 Bad Request response being sent to the client.

Checking CSRF Tokens with a View Predicate

.. deprecated:: 1.7
   Use the ``require_csrf`` option or read :ref:`auto_csrf_checking` instead
   to have :class:`pyramid.exceptions.BadCSRFToken` exceptions raised.

A convenient way to require a valid CSRF token for a particular view is to include check_csrf=True as a view predicate. See :meth:`pyramid.config.Configurator.add_view`.

@view_config(request_method='POST', check_csrf=True, ...)
def myview(request):


A mismatch of a CSRF token is treated like any other predicate miss, and the predicate system, when it doesn't find a view, raises HTTPNotFound instead of HTTPBadRequest, so check_csrf=True behavior is different from calling :func:`pyramid.csrf.check_csrf_token`.