This package implements an authentication policy for Pyramid that using JSON Web Tokens. This standard (RFC 7519) is often used to secure backend APIs. The excellent PyJWT library is used for the JWT encoding / decoding logic.
Enabling JWT support in a Pyramid application is very simple:
from pyramid.config import Configurator
from pyramid.authorization import ACLAuthorizationPolicy
def main():
config = Configurator()
# Pyramid requires an authorization policy to be active.
config.set_authorization_policy(ACLAuthorizationPolicy())
# Enable JWT authentication.
config.include('pyramid_jwt')
config.set_jwt_authentication_policy('secret')This will set a JWT authentication policy using the Authorization HTTP header with a JWT scheme to retrieve tokens. Using another HTTP header is trivial:
config.set_jwt_authentication_policy('secret', http_header='X-My-Header')If your application needs to decode tokens which contain an Audience claim you can extend this with:
config.set_jwt_authentication_policy('secret',
auth_type='Bearer',
callback=add_role_principals,
audience="example.org")To make creating valid tokens easier a new create_jwt_token method is
added to the request. You can use this in your view to create tokens. A simple
authentication view for a REST backend could look something like this:
@view_config('login', request_method='POST', renderer='json')
def login(request):
login = request.POST['login']
password = request.POST['password']
user_id = authenticate(login, password) # You will need to implement this.
if user_id:
return {
'result': 'ok',
'token': request.create_jwt_token(user_id)
}
else:
return {
'result': 'error'
}Unless you are using JWT cookies within cookies (see the next section), the
standard remember() and forget() functions from Pyramid are not useful.
Trying to use them while regular (header-based) JWT authentication is enabled
will result in a warning.
Optionally, you can use cookies as a transport for the JWT Cookies. This is an
useful technique to allow browser-based web apps to consume your REST APIs
without the hassle of managing token storage (where to store JWT cookies is a
known-issue), since http_only cookies cannot be handled by Javascript
running on the page
Using JWT within cookies have some added benefits, the first one being sliding
sessions: Tokens inside cookies will automatically be reissued whenever
reissue_time is past.
from pyramid.config import Configurator
from pyramid.authorization import ACLAuthorizationPolicy
def main():
config = Configurator()
# Pyramid requires an authorization policy to be active.
config.set_authorization_policy(ACLAuthorizationPolicy())
# Enable JWT authentication.
config.include('pyramid_jwt')
config.set_jwt_cookie_authentication_policy(
'secret', reissue_time=7200
)When working with JWT alone, there's no standard for manually invalidating a token: Either the token validity expires, or the application needs to handle a token blacklist (or even better, a whitelist)
On the other hand, when using cookies, this library allows the app to logout
a given user by erasing its cookie: This policy follows the standard cookie
deletion mechanism respected by most browsers, so a call to Pyramid's
forget() function will instruct the browser remove that cookie, effectively
throwing that JWT token away, even though it may still be valid.
See Creating a JWT within a cookie for examples.
Normally pyramid_jwt only makes a single JWT claim: the subject (or
sub claim) is set to the principal. You can also add extra claims to the
token by passing keyword parameters to the create_jwt_token method.
token = request.create_jwt_token(user.id,
name=user.name,
admin=(user.role == 'admin'))All claims found in a JWT token can be accessed through the jwt_claims
dictionary property on a request. For the above example you can retrieve the
name and admin-status for the user directly from the request:
print('User id: %d' % request.authenticated_userid)
print('Users name: %s', request.jwt_claims['name'])
if request.jwt_claims['admin']:
print('This user is an admin!')Keep in mind that data jwt_claims only reflects the claims from a JWT
token and do not check if the user is valid: the callback configured for the
authentication policy is not checked. For this reason you should always use
request.authenticated_userid instead of request.jwt_claims['sub'].
You can also use extra claims to manage extra principals for users. For example you could claims to represent add group membership or roles for a user. This requires two steps: first add the extra claims to the JWT token as shown above, and then use the authentication policy's callback hook to turn the extra claim into principals. Here is a quick example:
def add_role_principals(userid, request):
return ['role:%s' % role for role in request.jwt_claims.get('roles', [])]
config.set_jwt_authentication_policy(callback=add_role_principals)You can then use the role principals in an ACL:
class MyView:
__acl__ = [
(Allow, Everyone, ['read']),
(Allow, 'role:admin', ['create', 'update']),
]After creating and returning the token through your API with
create_jwt_token you can test by issuing an HTTP authorization header type
for JWT.
GET /resource HTTP/1.1
Host: server.example.com
Authorization: JWT eyJhbGciOiJIUzI1NiIXVCJ9...TJVA95OrM7E20RMHrHDcEfxjoYZgeFONFh7HgQ
We can test using curl.
curl --header 'Authorization: JWT TOKEN' server.example.com/ROUTE_PATHconfig.add_route('example', '/ROUTE_PATH')
@view_config(route_name=example)
def some_action(request):
if request.authenticated_userid:
# Do somethingThere are a number of flags that specify how tokens are created and verified.
You can either set this in your .ini-file, or pass/override them directly to the
config.set_jwt_authentication_policy() function.
| Parameter | ini-file entry | Default | Description |
|---|---|---|---|
| private_key | jwt.private_key | Key used to hash or sign tokens. | |
| public_key | jwt.public_key | Key used to verify token signatures. Only used with asymmetric algorithms. | |
| algorithm | jwt.algorithm | HS512 | Hash or encryption algorithm |
| expiration | jwt.expiration | Number of seconds (or a datetime.timedelta instance) before a token expires. | |
| audience | jwt.audience | Proposed audience for the token | |
| leeway | jwt.leeway | 0 | Number of seconds a token is allowed to be expired before it is rejected. |
| http_header | jwt.http_header | Authorization | HTTP header used for tokens |
| auth_type | jwt.auth_type | JWT | Authentication type used in Authorization header. Unused for other HTTP headers. |
| json_encoder | None | A subclass of JSONEncoder to be used to encode principal and claims infos. |
The follow options applies to the cookie-based authentication policy:
| Parameter | ini-file entry | Default | Description |
|---|---|---|---|
| cookie_name | jwt.cookie_name | Authorization | Key used to identify the cookie. |
| cookie_path | jwt.cookie_path | None | Path for cookie. |
| https_only | jwt.https_only_cookie | True | Whether or not the token should only be sent through a secure HTTPS transport |
| reissue_time | jwt.cookie_reissue_time | None | Number of seconds (or a datetime.timedelta instance) before a cookie (and the token within it) is reissued |
This is a basic guide (that will assume for all following statements that you have followed the Readme for this project) that will explain how (and why) to use JWT to secure/restrict access to a pyramid REST style backend API, this guide will explain a basic overview on:
- Creating JWT's
- Decoding JWT's
- Restricting access to certain pyramid views via JWT's
First off, lets start with the first view in our pyramid project, this would normally be say a login view, this view has no permissions associated with it, any user can access and post login credentials to it, for example:
def authenticate_user(login, password):
# Note the below will not work, its just an example of returning a user
# object back to the JWT creation.
login_query = session.query(User).\
filter(User.login == login).\
filter(User.password == password).first()
if login_query:
user_dict = {
'userid': login_query.id,
'user_name': login_query.user_name,
'roles': login_query.roles
}
# An example of login_query.roles would be a list
# print(login_query.roles)
# ['admin', 'reports']
return user_dict
else:
# If we end up here, no logins have been found
return None
@view_config('login', request_method='POST', renderer='json')
def login(request):
'''Create a login view
'''
login = request.POST['login']
password = request.POST['password']
user = authenticate(login, password)
if user:
return {
'result': 'ok',
'token': request.create_jwt_token(
user['userid'],
roles=user['roles'],
userName=user['user_name']
)
}
else:
return {
'result': 'error',
'token': None
}Now what this does is return your JWT back to whatever front end application you may have, with the user details, along with their permissions, this will return a decoded token such as:
eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ1c2VyTmFtZSI6Imx1a2UiLCJyb2xlcyI6WyJhZG1pbiIsInJlcG9ydHMiXSwic3ViIjo0LCJpYXQiOjE1MTkwNDQyNzB9.__KjyW1U-tpAEvTbSJsasS-8CaFyXH784joUPONH6hQ
Now I would suggest heading over to JWT.io, copy this data into their page, and you will see the decoded token:
{
"userName": "luke",
"roles": [
"admin",
"reports"
],
"sub": 4,
"iat": 1519044270
}Note, at the bottom of jwt.io's webpage, that the signature shows verified, if you change the "secret" at the bottom, it will say "NOT Verified" this is because in order for any JWT process to be verified, the valid "secret" or "private key" must be used. It is important to note that any data sent in a JWT is accessible and readable by anyone.
The following section would also work if pyramid did not create the JWT, all it needs to know to decode a JWT is the "secret" or "private key" used to create/sign the original JWT.By their nature JWT's aren't secure, but they can be used "to secure". In our example above, we returned the "roles" array in our JWT, this had two properties "admin" and "reports" so we could then in our pyramid application, setup an ACL to map JWT permissions to pyramid based security, for example in our projects __init__.py we could add:
from pyramid.security import ALL_PERMISSIONS
class RootACL(object):
__acl__ = [
(Allow, 'admin', ALL_PERMISSIONS),
(Allow, 'reports', ['reports'])
]
def __init__(self, request):
passWhat this ACL will do is allow anyone with the "admin" role in their JWT access to all views protected via a permission, where as users with "reports" in their JWT will only have access to views protected via the "reports" permission.
Now this ACL in itself is not enough to map the JWT permission to pyramids security backend, we need to also add the following to __init__.py:
from pyramid.authorization import ACLAuthorizationPolicy
def add_role_principals(userid, request):
return request.jwt_claims.get('roles', [])
def main(global_config, **settings):
""" This function returns a Pyramid WSGI application.
"""
config = Configurator(settings=settings)
...
# Enable JWT - JSON Web Token based authentication
config.set_root_factory(RootACL)
config.set_authorization_policy(ACLAuthorizationPolicy())
config.include('pyramid_jwt')
config.set_jwt_authentication_policy('myJWTsecretKeepThisSafe',
auth_type='Bearer',
callback=add_role_principals)This code will map any properties of the "roles" attribute of the JWT, run them through the ACL and then tie them into pyramids security framework.
Since cookie-based authentication is already standardized within Pyramid by the
remember() and forget() calls, you should simply use them:
from pyramid.response import Response
from pyramid.security import remember
@view_config('login', request_method='POST', renderer='json')
def login_with_cookies(request):
'''Create a login view
'''
login = request.POST['login']
password = request.POST['password']
user = authenticate(login, password) # From the previous snippet
if user:
headers = remember(
user['userid'],
roles=user['roles'],
userName=user['user_name']
)
return Response(headers=headers, body="OK") # Or maybe redirect somewhere else
return Response(status=403) # Or redirect back to loginPlease note that since the JWT cookies will be stored inside the cookies, there's no need for your app to explicitly include it on the response body. The browser (or whatever consuming this response) is responsible to keep that cookie for as long as it's valid, and re-send it on the following requests.
Also note that there's no need to decode the cookie manually. The Policy
handles that through the existing request.jwt_claims.
For example, a JWT could easily be manipulated, anyone could hijack the token, change the values of the "roles" array to gain access to a view they do not actually have access to. WRONG! pyramid_jwt checks the signature of all JWT tokens as part of the decode process, if it notices that the signature of the token is not as expected, it means either the application has been setup correctly with the wrong private key, OR an attacker has tried to manipulate the token.
The major security concern when working with JWT tokens is where to store the token itself: While pyramid_jwt is able to detect tampered tokens, nothing can be done if the actual valid token leaks. Any user with a valid token will be correctly authenticated within your app. Storing the token securely is outside the scope of this library.
When using JWT within a cookie, the browser (or tool consuming the cookie) is
responsible for storing it, but pyramid_jwt does set the http_only flag on
all cookies, so javascript running on the page cannot access these cookies,
which helps mitigate XSS attacks. It's still mentioning that the tokens are
still visible through the browser's debugging/inspection tools.
In the example posted above we creating an "admin" role that we gave ALL_PERMISSIONS access in our ACL, so any user with this role could access any view e.g.:
@view_config(route_name='view_a', request_method='GET',
permission="admin", renderer='json')
def view_a(request):
return
@view_config(route_name='view_b', request_method='GET',
permission="cpanel", renderer='json')
def view_b(request):
returnThis user would be able to access both of these views, however any user with the "reports" permission would not be able to access any of these views, they could only access permissions with "reports". Obviously in our use case, one user had both "admin" and "reports" permissions, so they would be able to access any view regardless.