Awesome
Row Level Permissions for FastAPI
While trying out the excellent FastApi framework there was one peace missing for me: an easy, declarative way to define permissions of users (and roles/groups) on resources. Since I reall love the way Pyramid handles this, I re-implemented and adapted the system for FastApi (well, you might call it a blatant rip-off).
An extremely simple and incomplete example:
from fastapi import Depends, FastAPI
from fastapi.security import OAuth2PasswordBearer
from fastapi_permissions import configure_permissions, Allow, Deny
from pydantic import BaseModel
app = FastAPI()
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="/token")
class Item(BaseModel):
name: str
owner: str
def __acl__(self):
return [
(Allow, Authenticated, "view"),
(Allow, "role:admin", "edit"),
(Allow, f"user:{self.owner}", "delete"),
]
class User(BaseModel):
name: str
def principals(self):
return [f"user:{self.name}"]
def get_current_user(token: str = Depends(oauth2_scheme)):
...
def get_active_user_principals(user:User = Depends(get_current_user)):
...
def get_item(item_identifier):
...
# Permission is already wrapped in Depends()
Permission = configure_permissions(get_active_user_principals)
@app.get("/item/{item_identifier}")
async def show_item(item: Item=Permission("view", get_item)):
return [{"item": item}]
For a better example install fastapi_permissions
source in an virtual environment (see further below), and start a test server:
(permissions) $ uvicorn fastapi_permissions.example:app --reload
Visit http://127.0.0.1:8000/docs to try it out. There are two users available: "bob" and "alice", both have the password "secret".
The example is derived from the FastApi examples, so it should be familiar. New / added stuff is marked with comments in the source file fastapi_permissions/example.py
Why not use Scopes?
For most applications the use of scopes to determine the rights of a user is sufficient enough. So if scopes fit your application, please use them - they are already a part of the FastAPI framework.
While scopes are tied only to the state of the user, fastapi_permissions
also
take the state of the requested resource into account.
Let's take an scientific paper as an example: depending on the state of the submission process (like "draft", "submitted", "peer review" or "published") different users should have different permissions on viewing, editing or retracting. This could be acomplished with custom code in the path definition functions, but fastapi_permissions
offers a method to define these constraints in a single place.
There is a second case, where fastapi_permissions
might be the right addition to your app: If your brain is wired / preconditioned like mine to such a permission model - e.g. exposed for a long time to Pyramid...
Long Story Short: Use scopes until you need something different.
Concepts
Since fastapi_permissions
heavely derived from the Pyramid framework, I strongly suggest to take a look at its security documentation if anything is unclear to you.
The system depends on a couple of concepts not found in FastAPI:
- resources: objects that provide an access controll list
- access controll lists: a list of rules defining which principal has what permission
- principal: an identifier of a user or his/her associated groups/roles
- permission: an identifier (string) for an action on an object
resources & access controll lists
A resource provides an access controll list via it's __acl__
attribute. It can either be an property of an object or a callable. Each entry in the list is a tuple containing three values:
- an action:
fastapi_permissions.Allow
orfastapi_permissions.Deny
- a principal: e.g. "role:admin" or "user:bob"
- a permission or a tuple thereof: e.g. "edit" or ("view", "delete")
Examples:
from fastapi_permissions import Allow, Deny, Authenticated, Everyone
class StaticAclResource:
__acl__ = [
(Allow, Everyone, "view"),
(Allow, "role:user", "share")
]
class DynamicAclResource:
def __acl__(self):
return [
(Allow, Authenticated, "view"),
(Allow, "role:user", "share"),
(Allow, f"user:{self.owner}", "edit"),
]
# in contrast to pyramid, resources might be access conroll list themselves
# this can save some typing:
AclResourceAsList = [(Allow, Everyone, "view"), (Deny, "role:troll", "edit")]
You don't need to add any "deny-all-clause" at the end of the access controll list, this is automagically implied. All entries in a ACL are checked in the order provided in the list. This makes some complex configurations simple, but can sometimes be a pain in the lower back…
The two principals Everyone
and Authenticated
will be discussed in short time.
users & principal identifiers
You must provide a function that returns the principals of the current active user. The principals is just a list of strings, identifying the user and groups/roles the user belongs to:
Example:
def get_active_principals(user: User = Depends(get_current_user)):
if user:
# user is logged in
principals = [Everyone, Authenticated]
principals.extend(getattr(user, "principals", []))
else:
# user is not logged in
principals = [Everyone]
return principals
special principals
There are two special principals that also help providing access controll lists: Everyone
and Authenticated
.
The Everyone
principal should be added regardless of any other defined principals or login status, Authenticated
should only be added for a user that is logged in.
permissions
A permission is just a string that represents an action to be performed on a resource. Just make something up.
As with the special principals, there is a special permission that is usable as a wildcard: fastapi_permisssions.All
.
Usage
There are some things you must provide before using the permissions system:
- a callable (FastApi dependency) that returns the principal of the logged in (active) user
- a resource with an access controll list
Configuring the permissions system
Simple configuration with some defaults:
from fastapi_permissions import configure_permissions
# must be provided
def get_active_principals(...):
""" returns the principals of the current logged in user"""
...
# Permission is already wrapped in Depends()
Permission = configure_permissions(get_active_principals)
One configuration option is available:
- permission_exception:
- this exception will be raised if a permission is denied
- defaults to fastapi_permissions.permission_exception
from fastapi_permissions import configure_permissions
# must be provided
def get_active_principals(...):
""" returns the principals of the current logged in user"""
...
# Permission is already wrapped in Depends()
Permission = configure_permissions(
get_active_principals,
permission_exception
)
using permissions in path operation
To use access controll in a path operation, you call the perviously configured function with a permission and the resource. If the permission is granted, the requested resource the permission is checked on will be returned, or in this case, the acl list
from fastapi_permissions import configure_permissions, Allow
# must be provided
def get_active_principals(...):
""" returns the principals of the current logged in user"""
...
example_acl = [(Allow, "role:user", "view")]
# Permission is already wrapped in Depends()
Permission = configure_permissions(get_active_principals)
@app.get("/")
async def root(acls:list=Permission("view", example_acl)):
return {"OK"}
Instead of using an access controll list directly, you can also provide a dependency function that might fetch a resource from a database, the resouce should provide its access controll list via the __acl__
attribute:
from fastapi_permissions import configure_permissions, Allow
# must be provided
def get_active_principals(...):
""" returns the principals of the current logged in user"""
...
# fetches a resource from the database
def get_item(item_id: int):
""" returns a resource from the database
The resource provides an access controll list via its "__acl__" attribute.
"""
...
# Permission is alredy wrapped in Depends()
Permission = configure_permissions(get_active_principals)
@app.get("/item/{item_id}")
async def show_item(item:Item=Permission("view", get_item)):
return {"item": item}
helper functions
Sometimes you might want to check permissions inside a function and not as the definition of a path operation:
With has_permission(user_principals, permission, resource)
you can preform the permission check programatically. The function signature can easily be remebered with something like "John eat apple?". The result will be either True
or False
, so no need for try/except blocks \o/.
from fastapi_permissions import (
has_permission, Allow, All, Everyone, Authenticated
)
user_principals == [Everyone, Authenticated, "role:owner", "user:bob"]
apple_acl == [(Allow, "role:owner", All)]
if has_permission(user_principals, "eat", apple_acl):
print "Yum!"
The other function provided is list_permissions(user_principals, resource)
this will return a dict of all available permissions and a boolean value if the permission is granted or denied:
from fastapi_permissions import list_permissions, Allow, All
user_principals == [Everyone, Authenticated, "role:owner", "user:bob"]
apple_acl == [(Allow, "role:owner", All)]
print(list_permissions(user_principals, apple_acl))
{"permissions:*": True}
Please note, that "permissions:*"
is the string representation of fastapi_permissions.All
.
How it works
The main work is done in the has_permissions()
function, but the most interesting ones (at least for me) are the configure_permissions()
and permission_dependency_factory()
functions.
Wait. I didn't tell you about the latter one?
The permission()
thingy used in the path operation definition before is actually the mentioned permission_dependency_factory()
. The configure_permissions()
function just provisiones it with some default values using functools.partial
. This reduces the function signature from permission_dependency_factory(permission, resource, active_principals_func, permission_exception)
down to partial_function(permission, resource)
.
The permission_dependency_factory
returns another function with the signature permission_dependency(Depends(resource), Depends(active_principals_func))
. This is the acutal signature, that Depends()
uses in the path operation definition to search and inject the dependencies. The rest is just some closure magic ;-).
Or in other words: to have a nice API, the Depends()
in the path operation function should only have a function signature for retrieving the active user and the resource. On the other side, when writing the code, I wanted to only specifiy the parts relevant to the path operation function: the resource and the permission. The rest is just on how to make it work.
(F.)A.Q.
Permission check on collection of resources
How to use the library with something like this: List[Item]=Permission("edit", get_items)
.
The question was actually issue #3 and I have written a longer answer in the issue, please have a look there.
Dev & Test virtual environment
There is an easy to use make command for setting up a virtual environment, installing the required packages and installing the project in an editable way.
$ git clone https://github.com/holgi/fastapi-permissions.git
$ cd fastapi-permissions
$ make devenv
$ source .venv/bin/activate
Then you can test any changes locally with make test
. This will stop
on the first error and not report coverage.
(permissions) $ make test
If you can also run all tests and get a coverage report with
(permissions) $ make coverage
And when ready to test everything as an installed package (bonus point if
using make clean
before)
(permissions) $ make tox
Thanks
- Sebastián Ramírez, for creating FastAPI
- William, for fixing my stupid bug