FlaskSimpleAuth Documentation

This modules helps handling authentication, authorizations, parameters and provide other utils for Flask, controled from Flask configuration and the extended route decorator.

Contents

Motivation, Installation and Initialization.
Authentication
- Schemes (basic, param, token…)
- Password Management.
Authorizations
- Groups
- OAuth
- Objects
Parameters http, json and file parameters are managed with type hints.
Utils such as Reference wrapper, Error Handling and Miscellaneous Directives.
See Also Flask-Security, Flask-RESTful, Flask-AppBuilder, Flask-Login and others.
License, Versions, TODO.

Motivation

FlaskSimpleAuth is designed to help an application REST API back-end development. The features provided in this module point at fixing what I deem Flask lack of helpfulness and wrong style, such as:

not providing a simple and effective declarative security layer, which must address both authentication and authorization possibly in connection to application data, thus requires some kind of integration;
not providing a clean way to put authentication in the configuration only, where it belongs, and taking into account password management best practices.
not providing any real help with handling parameters, which is demonstrated by the fact that the user must find them in different dictionaries depending on where they come from, and having to access them through ugly global proxies such as request.

The resulting module allows clean application development which only has to focus on handling routes depending on cleanly typed parameters, with declarative security on each one, reducing both line counts and code complexity.

An emphasis is also put on performance by providing caching where it matters, so this is not only about style: Many hooks are provided to be able to take full control of various features, with reasonable defaults which make this less a necessity. Many key features rely on proven third-party packages such as passlib, PyJWT or flask-cors.

This module does not care much about web and application oriented features: it is a Flask framework extension which aims at better handling HTTP-related features. It could be used as a cleaner layer for other Flask application-oriented extensions such as Flask-Security.

Install

Use pip install FlaskSimpleAuth to install the module, or whatever other installation method you prefer. Depending on options, the following modules should be installed:

ProxyPatternPool for sharing.
cachetools and CacheToolsUtils for caching. Possibly pymemcache and redis for external back-ends.
passlib for password management, bcrypt for password hashing (default algorithm).
PyJWT for JSON Web Token (JWT), cryptography for pubkey-signed JWT.
Flask HTTPAuth for http-* authentication options.
Flask CORS for CORS handling.

These modules are installed with the corresponding options: password, jwt, memcached, redis, httpauth, cors: pip install FlaskSimpleAuth[jwt].

Sharing and cache-related modules are currently always installed, even if they are unused or may be desactivated.

Initialization

The module is simply initialize by calling its Flask constructor and providing a configuration through FSA_* directives (from a separate file or directly in the constructor).

import FlaskSimpleAuth as fsa
app = fsa.Flask("acme", FSA_MODE="debug")
app.config.from_envvar("ACME_CONFIG")

Once initialized, app behaves as a standard Flask object with many additions. The main change is the route decorator, an extended version of Flask’s own with an authorize parameter and transparent management of request parameters. Per-method shortcut decorators post, get, put, patch and delete which support the same extensions. The security first principle means that if the parameter is missing the route is closed with a 403.

@app.get("/store", authorize="OPEN")
def get_store(filter: str = None):
    # return store contents, possibly filtered
    ...

@app.post("/store", authorize="contributor")
def post_store(data: str):
    # append new data to store, return id
    ...

@app.get("/store/<id>", authorize="OPEN")
def get_store_id(id: int):
    # return data corresponding to id
    ...

Inside a request handling function, additional methods on app give access to authentication-dependent data, for instance:

get_user extracts the authenticated user or raise an exception, and current_user gets the authenticated user if any, or None. It can also be requested as a parameter with the CurrentUser type.
user_scope checks if the current token-authenticated user has some authorizations.
hash_password and check_password hash or check a password.
create_token computes a new authentication token for the current user.
check_user_password recheck a password for a user.

Various decorators/functions allow to register hooks, such as:

user_in_group, group_check, get_user_pass and object_perms functions/decorators to register authentication and authorization helper functions:
- a function to retrieve the password hash from the user name.
- a function which tells whether a user belong to some group.
- a function which tells whether a user is in a provided group or role.
- functions which define object ownership and access permissions.
password_quality a function/decorator to register a function to check for password quality.
password_check a function/decorator to register a new password checker, so as to handle recovery codes, for instance.
cast a function/decorator to register new str-to-some-type casts for function parameters.

# return password hash if any (see with FSA_GET_USER_PASS)
# None means that the user does not exists
@app.get_user_pass
def get_user_pass(user: str) -> str|None:
    return ...

# return whether user belong to some group (see with FSA_GROUP_CHECK)
@app.group_check("admin")
def user_is_admin(user: str) -> bool:
    return ...

# or alternative catch-all dynamic approach
# return whether user is in group (see with FSA_USER_IN_GROUP)
@app.user_in_group
def user_in_group(user: str, group: str) -> bool:
    return ...

# return whether user can access the `foo` object for an operation
# None will generates a 404
@app.object_perms("foo")
def allow_foo_access(user: str, fooid: int, mode: str) -> bool|None:
    return ...

These hooks allow taking over control of most internal processes, if needed.

Authentication

The main authentication configuration directive is FSA_AUTH which governs the authentication methods used by the get_user function, as described in the following sections. Defaut is httpd.

If a non-token scheme is provided, authentication will be token followed by the provided scheme, i.e. token is tried first anyway if enabled.

To take full control of authentication schemes, provide an ordered list. Note that it does not always make much sense to mix some schemes, e.g. basic and digest password storage assumptions are distinct and should not be merged. Also, only one HTTPAuth-based scheme can be active at a time.

Authentication is always performed on demand, either to check for a route authorization declared with authorize or when calling get_user.

The authentication scheme attempted on a route can be altered with the auth (or authn) parameter added to the route decorator. This may be used to restrict or change the authentication scheme on a route. Some combinations may or may not work depending on module internals, so this is only for special cases. A legitimate use for a REST API is to have FSA_AUTH defined to token and have only one password-authenticated route to obtain the token used by all other routes.

Authentication Schemes

The available authentication schemes are:

none

Deactivate authentication.
httpd

Inherit web server supplied authentication through request.remote_user. This is the default.

There are plenty authentication schemes available in a web server such as Apache or Nginx, including LDAP or other databases, all of which probably more efficiently implemented than python code, so it should be the preferred option. However, it could require significant configuration effort compared to the application-side approach.
basic

HTTP Basic password authentication, which rely on the Authorization HTTP header in the request.
- FSA_REALM provides the authentication realm.
See also Password Management below for how the password is retrieved and checked.
http-basic

Same as previous based on flask-HTTPAuth.
- FSA_REALM provides the authentication realm.
- FSA_HTTP_AUTH_OPTS allow to pass additional options to the HTTPAuth authentication class.
param

HTTP or JSON parameter for password authentication. User name and password are passed as request parameters.
- FSA_PARAM_USER is the parameter used for the user name. Default is USER.
- FSA_PARAM_PASS is the parameter used for the password. Default is PASS.
See also Password Management below for the password is retrieved and checked.
password

Tries basic then param authentication.
http-digest or digest

HTTP Digest authentication based on flask-HTTPAuth.

Note that the implementation relies on sessions, which may require the SECRET_KEY option to be set to something. The documentation states that server-side sessions are needed because otherwise the nonce and opaque parameters could be reused, which may be a security issue under some conditions. I’m unsure about that, but I agree that client-side cookie sessions are strange things best avoided if possible.
- FSA_REALM provides the authentication realm.
- FSA_HTTP_AUTH_OPTS allow to pass additional options to the HTTPAuth authentication class, such as use_ha1_pw, as a dictionary.
See also Password Management below for how the password is retrieved and checked. Note that password management is different for digest authentication because the simple hash of the password or the password itself is needed for the verification.
token

Only rely on signed tokens for authentication. A token certifies that a user is authenticated in a realm up to some time limit. The token is authenticated by a signature which is usually the hash of the payload (realm, user and limit) and a secret hold by the server.

There are two token types chosen with the FSA_TOKEN_TYPE configuration directive: fsa is a simple compact readable custom format, and jwt RFC 7519 standard based on PyJWT implementation.

The fsa token syntax is: <realm>:<user>:<limit>:<signature>, for instance: comics:calvin:20380119031407:4ee89cd4cc7afe0a86b26bdce6d11126. The time limit is a simple UTC timestamp YYYYMMDDHHmmSS that can be checked easily by the application client. Compared to jwt tokens, they are short and easy to interpret and compare manually, no decoding is involved. If an issuer is set (see FSA_TOKEN_ISSUER below), the name is appended to the realm after a /.

The following configuration directives are available:
- FSA_TOKEN_TYPE type of token, either fsa, jwt or None to disable. Default is fsa.
- FSA_TOKEN_CARRIER how to transport the token: bearer (Authorization HTTP header), param, cookie or header. Default is bearer.
- FSA_TOKEN_NAME name of parameter or cookie holding the token, or bearer scheme, or header name. Default is AUTH for param carrier, auth for cookie carrier, Bearer for HTTP Authorization header (bearer carrier), Auth for header carrier.
- FSA_REALM realm of authentication for token, basic or digest. Default is the simplified lower case application name. For jwt, this is translated as the audience. The application realm may be overriden on a route for creating MFA stages.
- FSA_TOKEN_ISSUER the issuer of the token. Default is None.
- FSA_TOKEN_SECRET secret string used for validating tokens. Default is a system-generated random string containing 256 bits. This default will only work with itself, as it is not shared across server instances or processes.
- FSA_TOKEN_SIGN secret string used for signing tokens, if different from previous secret. This is only relevant for public-key jwt schemes (R…, E…, P…). Default is to use the previous secret.
- FSA_TOKEN_DELAY number of minutes of token validity. Default is 60.0 minutes.
- FSA_TOKEN_GRACE number of minutes of grace time for token validity. Default is 0.0 minutes.
- FSA_TOKEN_ALGO algorithm used to sign the token. Default is blake2s for fsa and HS256 for jwt.
- FSA_TOKEN_LENGTH number of hash bytes kept for token signature. Default is 16 for fsa. The directive is ignored for jwt.
- FSA_TOKEN_RENEWAL for cookie tokens, the fraction of delay under which the cookie/token is renewed automatically. Default is 0.0, meaning no renewal.
Function create_token(user) creates a valid token for the user depending on the current scheme and detailed configuration. If user is not given, the current user is taken. Other parameters can be overriden, such as realm, delay, issuer, secret… to allow full control if necessary, eg for MFA.

Token authentication is always attempted unless the secret is empty. Setting FSA_AUTH to token results in only token authentication to be used.

Token authentication is usually much faster than password verification because password checks are designed to be slow so as to hinder password cracking, whereas token authentication relies on simple hashing for its security. Another benefit of token is that it avoids sending passwords over and over. The rational option is to use a password scheme to retrieve a token and then to use it till it expires. This can be enforced by setting FSA_AUTH to token and to only add auth="basic" on the login route used to retrieve a token.

Token expiration can be understood as a kind of automatic logout, which suggests to choose the delay with some care depending on the use case.

Internally token checks are cached so that even with slow JWT public-key schemes the performance impact should be low.

Note: it is possible to switch from an expiration model (the token tells when it expires) to a validity model (the token tells when it was generated) by setting FSA_TOKEN_DELAY to zero and FSA_TOKEN_GRACE to the amount of time a token should be considered valid.
oauth

Synonymous to token, but to be used on a route so as to trigger JWT scope authorizations on that route.

See also OAuth Authorizations below for how to use JWT token scopes.
http-token

Token scheme based on flask-HTTPAuth. Carrier is bearer or header.
- FSA_HTTP_AUTH_OPTS passes additional options to the HTTPAuth authentication class, such as header, as a dictionary.
fake

Trust a parameter for authentication claims. Only for local tests, obviously. This is enforced.
- FSA_FAKE_LOGIN is the parameter holding the user name. Default is LOGIN.

Other authentication schemes can be added by registering a new hook which:

is passed the application and request
returns the authenticated login as a str, or None
may raise an ErrorResponse if unhappy, eg to generate a 401

@app.authentication("code")
def code_authentication(app: Flask, req: Request) -> str|None:
    ...

@app.get("/code-authentication", authorize="AUTH", auth="code")
def get_code_authentication(user: CurrentUser):
    return f"Hello code-authenticated {user}!", 200

The FSA_AUTHENTICATION configuration directive is a dictionary which can be used for the same purpose.

Password Management

Password authentication is performed for the following authentication schemes: param, basic, http-basic, http-digest, digest, password.

The provided password management comprises handling password verification in the application, relying on standard password hashing schemes and a user-provided function to retrieve the password hash (get_user_pass), and/or delegating the whole verification process to a user-provided function (password_check).

For checking passwords internally, the password (salted hash) must be retrieved through get_user_pass(user). This function must be provided when the module is initialized. Because this function is cached by default, the cache expiration must be reached so that changes take effect, or the cache must be cleared manually, which may impair application performance.

The following configuration directives are available to configure passlib password checks:

FSA_PASSWORD_SCHEME password scheme to use for passwords. Default is bcrypt. See passlib documentation for available options, including the bad plaintext. Set to None to disable internal password checking.
FSA_PASSWORD_OPTS relevant options (for passlib.CryptContext). Default is ident 2y with 4 rounds.

Beware that modern password checking is often pretty expensive in order to thwart password cracking if the hashed passwords are leaked, so that you do not want to have to use that on every request in real life (eg hundreds milliseconds for passlib bcrypt 12 rounds). The above defaults result in manageable password checks of a few milliseconds. Consider using tokens to reduce the authentication load on each request.

For digest authentication, the password must be either in plaintext or a simple MD5 hash (RFC 2617). The authentication setup must be consistent (set use_ha1_pw as True for the later). As retrieving the stored information is enough to steal the password (plaintext) or at least impersonate a user (hash), consider avoiding digest altogether. HTTP Digest Authentication only makes sense for unencrypted connexions, which is bad practice anyway. It is just provided here for completeness.

Function hash_password(pass) computes the password salted digest compatible with the current configuration, and may be used by the application for setting or resetting passwords.

This function checks the password quality by relying on:

FSA_PASSWORD_LENGTH minimal password length, 0 to disable.
FSA_PASSWORD_RE list of regular expressions that a password must match.
FSA_PASSWORD_QUALITY hook function which returns whether the password is acceptable, possibly raising an exception to complain if not. This hook can also be filled with the password_quality method/decorator. It allows to plug a password strength estimator such as zxcvbn.

This application-managed standard password checking can be overridden by providing an alternate password checking function with a directive:

FSA_PASSWORD_CHECK hook function which returns whether user and password provided is acceptable. This allows to plug a LDAP server or a temporary password recovery scheme or other one-time or limited-time passwords sent by SMS or mail, for instance. This hook can also be filled with the password_check method/decorator. This alternate check is used if the primary check failed or is disactivated.

An opened route for user registration with mandatory parameters could look like that:

@app.post("/register", authorize="OPEN")
def post_register(user: str, password: str):
    if user_already_exists(user):
        return f"cannot create {user}", 409
    add_new_user_with_hashed_pass(user, app.hash_password(password))
    return "", 201

Because password checks are usually expensive, it is advisable to switch to token authentication. A token can be created on a path authenticated by a password method:

# token creation route for all registered users
@app.get("/login", authorize="AUTH")
def get_login():
    return jsonify(app.create_token()), 200

The client application will return the token as a parameter or in headers for authenticating later requests, till it expires.

Multi-factor authentication (MFA) is supported by generating intermediate tokens on distinct realms at different stages, as discussed in recipes and illustrated in the demo.

Note that route-dependent realms do not work with http-* authentications because the realm is frozen by the external implementation in this case. Also, The same shared secret is used to validate all realms.

Authorization

Authorizations are declared with the authorize (or authz) parameter to the route decorator or its per-method shortcuts. The modules supports three permission models:

a group-oriented model
a scope OAuth model
an object-oriented model

The parameter accepts a list of str and int for groups or scopes, and of tuple for object permissions. If a scalar is provided, it is assumed to be equivalent to a list of one element.

When multiple authorizations are required through a list, they are cumulative, that is all conditions must be met.

Group Authorizations

A group or role is identified as an integer or a string. There are three approaches to check for group membership:

use predefined group special values OPEN AUTH CLOSE.
associate a group membership function to each group name.
use the user_in_group(user, group) function to check whether the authenticated user belongs to a given group.

Because these functions are cached by default, the cache expiration must be reached so that changes take effect, or the cache must be cleared manually, which may impair application performance.

@app.get("/admin-only", authorize="ADMIN")
def get_admin_only():
    # only authenticated "ADMIN" users can get here!

There are three special values that can be passed to the authorize decorator:

OPEN declares that no authentication is needed on that route, i.e. anyone can get in, the route is open.
AUTH declares that all authenticated user can access this route, without group checks.
CLOSE returns a 403 on all access. It can be used to close a route temporarily. This is the default.

@app.get("/closed", authorize="CLOSE")
def get_closed():
    # nobody can get here

@app.get("/authenticated", authorize="AUTH")
def get_authenticated():
    # AUTH-enticated users can get here

@app.get("/opened", authorize="OPEN")
def get_opened():
    # anyone can get here, no authentication is required

Note that this simplistic model does is not enough for non-trivial applications, where permissions on objects often depend on the object owner. For those, careful per-object and per-operation authorization are needed.

Groups can be registered with add_group or with FSA_AUTHZ_GROUPS. If done so, unregistered groups are rejected and result in a configuration error:

app.add_group("student", "professor")

@app.get("/students", authorize="admin")  # ERROR, unregistered group
def get_students():
    ...

OAuth Authorizations

OAuth authorizations are similar to group authorizations. They are attached to the current authentification performed through a token, on routes explicitely marked with auth="oauth". In that case, the authorize values are interpreted as scopes that must be provided by the token.

In order to simplify security implications, scopes and groups (user_in_group) authorizations cannot be mixed on a route: create distinct routes to handle these. Another current limitation is that only one issuer is allowed.

# /data is only accessible through a trusted JWT token with "read" scope
@app.get("/data", authorize="read", auth="oauth"):
def get_data(user: CurrentUser):
    return access_some_data(user), 200

Method user_scope allows to check whether the current user can perform some operation. It can be used with an object authorization rule.

Method add_scope and directive FSA_AUTHZ_SCOPES allow to register valid scopes that can be checked later. If not set, all scopes are considered valid.

The scope delegated authorization model suggests that the issuer is trusted to control accesses with any possible scope. This may or may not make sense from a security perspective depending on the use case. It makes perfect sense if the issuer is providing authorizations for accesses to itself, possibly from a third party.

Object Authorizations

Non trivial application have access permissions which depend on the data stored by the application. For instance, a user may alter a data because they own it, or access a data because they are friends of the owner.

In order to implement this model, the authorize decorator parameter can hold (domain, variable, mode) tuples which designate a permission domain (eg a table or object or concept name in the application), the name of a variable in the request (path or HTTP or JSON parameters) which identifies an object of the domain, and the operation or level of access necessary for this route:

@app.get("/message/<mid>", authorize=("msg", "mid", "read"))
def get_message_mid(mid: int):
    ...

The system will check whether the current user can access message mid in read mode by calling a per-domain user-supplied function:

@app.object_perms("msg")
def can_access_message(user: str, mid: int, mode: str) -> bool|None:
    # can user access message mid for operation mode?
    return ...

# also: app.object_perms("msg", can_access_message)

If the check function returns None, a 404 Not Found response is generated. If it returns False, a 403 Forbidden response is generated. If it returns True, the route function is called to generate the response.

If mode is not supplied, None is passed to the check function. If variable is not supplied, the first parameter of the route function is taken:

# same as authorize=("msg", "mid", None)
@app.patch("/message/<mid>", authorize=("msg",))
def patch_message_mid(mid: int):
    ...

The FSA_OBJECT_PERMS configuration directive can be set as a dictionary which maps domains to their access checking functions:

FSA_OBJECT_PERMS = { "msg": can_access_message, "blog": can_access_blog }

Because these functions are cached by default, the cache expiration must be reached so that changes take effect, or the cache must be cleared manually, which may impair application performance.

In the context of oauth authorizations, the per-domain object permission function can rely on user_scope to check whether some mode is allowed by the token.

Parameters

Request parameters (HTTP or JSON) are translated automatically to named function parameters by relying on function type annotations. Parameters are considered mandatory unless a default value is provided.

@app.get("/something/<id>", authorize=...)
def get_something_id(id: int, when: date, what: str = "nothing"):
    # `id` is an integer path-parameter
    # `when` is a mandatory date HTTP or JSON parameter
    # `what` is an optional string HTTP or JSON parameter
    ...

Request parameter string values are actually converted to the target type, and generate a 400 if the configuration fails. For int, base syntax is accepted for HTTP/JSON parameters, i.e. 0x11, 0o21, 0b10001 and 17 all mean decimal 17. For bool, False is an empty string, 0, False or F, otherwise the value is True. Type path is a special str type which allows to trigger accepting any path on a route. Type JsonData is a special type to convert, if necessary, a string value to JSON, expecting a list or a dictionary.

If one parameter is a dict of keyword arguments, all remaining request parameters are added to it, as shown below:

@app.put("/awesome", authorize="AUTH")
def put_awesome(**kwargs):
    ...

Pydantic-generated classes and dataclasses work out of the box both with HTTP and JSON parameters.

class Search(pydantic.BaseModel):
    words: list[str]
    limit: int

# or
@dataclass
class Search:
    words: list[str]
    limit: int

@app.get("/search", authorize="OPEN")
def get_search(q: Search):
    ...

Generic types can be used, although with restrictions: only combinations of list and str-key dict of standard Python types are supported and checked. Using custom classes may or may not work, consider using data classes instead.

@app.get("/question", authorize="OPEN")
def get_question(q: list[str]):
    ...

Note that list[*] when using HTTP parameters are assumed to be repeated parameters, not one parameter with a list value.

Custom classes can be used as parameter types, provided that the constructor accepts a string (for HTTP parameters) or whatever value provided (for JSON) to build the expected type.

class EmailAddr:
    def __init__(self, addr: str):
        self._addr = addr

@app.get("/mail/<addr>", authorize="AUTH")
def get_mail_addr(addr: EmailAddr):
    ...

Defining new types can be used to factor out some parameter checks, for instance requiring a positive integer:

class nat(int):  # this demonstrate Python simplicity
    def __new__(cls, val):
        if val < 0:
            raise ValueError(f"nat value must be positive: {val}")
        return super().__new__(cls, val)

@app.get("/pos", authorize="OPEN")
def get_pos(i: nat, j: nat):
    # i and j are positive integers
    ...

If the constructor does not match, a custom function can be provided with the cast function/decorator and will be called automatically to convert parameters:

class House:
    ...

@app.cast(House)
def strToHouse(s: str) -> House:
    return ...

# or: app.cast(House, strToHouse)

@app.get("/house/<h>", authorize="OPEN")
def get_house_h(h: House)
    ...

The FSA_CAST directive can also be defined as a dictionary mapping types to their conversion functions:

FSA_CAST = { House: strToHouse, ... }

As a special case, the Request, Session, Globals, Environ, CurrentApp, CurrentUser, Cookie, Header and FileStorage types, when used for parameters, result in the request, session, g flask special objects, environ WSGI parameter, the current authenticated user, the current application, the cookie value, the header value or a special file parameter (for upload) to be passed as this parameter to the function, allowing to keep a functional programming style by hidding away these special proxies.

More special parameters can be added with the special_parameter app function/decorator, by providing a type and a function which is given the parameter name (usually useless, but not always) and returns the expected value. For instance, the Request and FileStorage definitions roughly correspond to:

app.special_parameter(Request, lambda _: request)
app.special_parameter(FileStorage, lambda p: request.files[p])

An example use-case is to make user-related data easily available through such a special type:

class User:
    def __init__(self, login: str):
        self.firstname, self.lastname = db.get_user_data(login=login)

app.special_parameter(User, lambda _: User(app.current_user()))

@app.get("/hello", authorize="AUTH")
def get_hello(user: User):
    return f"Hello {user.firstname} {user.lastname}!", 200

The FSA_SPECIAL_PARAMETER directive can also be defined as a dictionary mapping types to their parameter value function.

Python parameter names can be prepended with a _, which is ignored when translating HTTP parameters. This allows to use python keywords as parameter names, such as pass or def.

@app.put("/user/<pass>", authorize="AUTH")
def put_user_pass(_pass: str, _def: str, _import: str):
    ...

Finally, configuration directive FSA_REJECT_UNEXPECTED_PARAM tells whether to reject requests with unexpected parameters. Default is True.

Utils

Utilities include the Reference generic object wrapper class, error handling utilities and miscellaneous configuration directives which cover security, caching and CORS.

`Reference` Object Wrapper

This class provides a proxy object based on the Proxy class from ProxyPatternPool.

This class implements a generic share-able global variable which can be used by modules (eg app, blueprints…) with its initialization differed.

Under the hood, most methods calls are forwarded to a possibly sub-thread-local object stored inside the wrapper, so that the Reference object mostly behaves like the wrapped object itself.

See the module for a detailed documentation.

Error Handling

Raising the ErrorResponse exception with a message, status, and possibly headers and content type from any user-defined function generates a Response of this status with the text message as contents.

The following directives provide convenient configuration about error handling:

FSA_ERROR_RESPONSE sets the handler for generating actual responses on errors. Text values plain or json generate simple text/plain or application/json responses. Using json:error generates a JSON dictionary with key error holding the error message. The response generation can be fully overriden by providing a callable which expects the error message, status code, headers and content type as parameters. This handler can be restricted to apply only to FSA-generated errors, see FSA_HANDLE_ALL_ERRORS below. Default is plain.
FSA_HANDLE_ALL_ERRORS whether to handle all 4xx and 5xx errors, i.e. take full responsability for generating error responses using FSA internal error handler (see FSA_ERROR_RESPONSE above). Default is True. When set to False, some errors may generate their own response in any format based on Flask default error response generator.
FSA_KEEP_USER_ERRORS whether to refrain from handling user errors and let them pass to the outer WSGI infrastructure instead. User errors are intercepted anyway, traced and raised again. They may occur from any user-provided functions such as various hooks and route functions. Default is False
FSA_SERVER_ERROR controls the status code returned on the module internal errors, to help distinguish these from other internal errors which may occur. Default is 500.
FSA_NOT_FOUND_ERROR controls the status code returned when a permission checks returns None. Default is 404.

A possible pattern is to put checks, say about parameters, in a function which raises the ErrorResponse exception to control client-facing results, and to call this function for checking these parameters:

def check_foo_param(foo):
    if db_has_no_foo(foo):
        raise ErrorResponse(f"no such foo: {foo}", 404)
    if this_foo_is_confidential(foo, app.current_user()):
        raise ErrorResponse(f"no way foo: {foo}", 403)

@app.get("/foo/<fid>", authorize="AUTH")
def get_foo_fid(fid: int):
    check_foo_param(fid)
    ...

Such checks can also be performed in an object constructor, with the inconvenience that the objects become specific to the HTTP context.

Miscellaneous Configuration Directives

Some directives govern various details for this extension internal working.

FSA_MODE set module mode, expecting prod, dev, debug to debug4. This changes the module verbosity. Under dev or debug, FSA-* headers is added to show informations about the request, the authentication and the elapsed time from the application code perspective. Default is prod.
FSA_LOGGING_LEVEL adjust module internal logging level. Default is None.
FSA_SECURE only allows secured requests on non-local connections. Default is True.
FSA_LOCAL sets the internal object isolation level. It must be consistent with the module WSGI usage. Possible values are process, thread (several threads can be used by the WSGI server), werkzeug (may work with sub-thread level request handling, eg greenlets), gevent and eventlet. Default is thread.
FSA_ADD_HEADERS allows to add headers to the generated response, as a dictionary. Keys are header names and values are either strings, which are used as is, or functions which are called with the response as a parameter to generate a value. None returned values are silently ignored. The corresponding add_headers method allows to add headers as keyword arguments. Default is empty.
FSA_DEFAULT_CONTENT_TYPE allows to replace the default text/html header added for string or byte responses. Default is None, meaning no replacement.
FSA_JSON_STREAMING allows to force jsonify to generate a string instead of a string generator when serializing a generator. Default True is to stream the JSON output, which may interact badly with the database transactions in some cases depending on the driver and WSGI server.
FSA_BEFORE_REQUEST and FSA_AFTER_REQUEST allow to add a list of before and after request hooks from the configuration instead of the actual application code. As a slight deviation from Flask before request hook, before request functions are passed the current request as an argument. They are executed first (just after some internal initializations and before application-provided before request hooks) and last, respectively. Defaults are empty.
FSA_BEFORE_EXEC allows to add a list of functions called after all preprocessing and just before the actual execution of the route function. The hooks are passed the request, the login and the authentication scheme. It may return a response to shortcut the route function. Such functions can also be register with the before_exec function/decorator. Defaults are empty.

Some control is available about internal caching features used for user authentication (user password access and token validations) and authorization (group and per-object permissions):

FSA_CACHE controls the type of cache to use, set to None to disallow caches. Values for standard cachetools cache classes are ttl, lru, tlru, lfu, mru, fifo, rr plus dict. MemCached is supported by setting it to memcached, and Redis with redis. Default is ttl. The directive can also be set to a MutableMapping instance to take direct control over the cache.
FSA_CACHE_OPTS sets internal cache options with a dictionary. This must contain the expected connection parameters for pymemcache.Client and for redis.Redis redis, for instance. For redis and ttl, an expiration ttl of 10 minutes is used and can be overwritten by providing the ttl parameter.
FSA_CACHE_SIZE controls size of internal cachetools caches. Default is 262144, which should use a few MiB. None means unbounded, more or less.
FSA_CACHE_PREFIX use this application-level prefix, useful for shared distributed caches. A good candidate could be app.name + ".". Default is None, meaning no prefix.
Method clear_caches allows to clear internal process caches. This is a mostly a bad idea, you should wait for the ttl.
Methods password_uncache, token_uncache, user_token_uncache, group_uncache, object_perms_uncache and auth_uncache allow to remove a particular entry from the shared cache, without waiting for its eventual expiration.

Note that uncaching features are on a best-effort basis, and may not work as expected especially with multi-process or multi-level cache settings. You should really wait for the ttl…

Web-application oriented features:

FSA_401_REDIRECT url to redirect to on 401. Default is None. This can be used for a web application login page.
FSA_URL_NAME name of parameter for the target URL after a successful login. Default is URL if redirect is activated, else None. Currently, the login page should use this parameter to redirect to when ok.
FSA_CORS and FSA_CORS_OPTS control CORS (Cross Origin Resource Sharing) settings.

CORS is a security feature implemented by web browsers to prevent JavaScript injection. It checks whether a server accepts requests from a given origin (i.e. from JavaScript code provided by some domain).

CORS request handling is enabled by setting FSA_CORS to True which allows requests from any origin. Default is False. Additional options are controled with FSA_CORS_OPTS. The implementation is simply delegated to the flask_cors Flask extension which must be available if the feature is enabled.

License

This software is public domain.

All software has bug, this is software, hence… Beware that you may lose your hairs or your friends because of it. If you like it, feel free to send a postcard to the author.

Versions

Sources, documentation and issues are hosted on GitHub. Install package from PyPI.

See all versions.

TODO

Todo or not todo…

Authn/Authz

demo LDAP auth? One class plus a new check_password?
add any token scheme?
use authlib?
test FSA_HTTP_AUTH_OPTS?
password re could use a dict for providing an explanation?
how to have several issuers and their signatures schemes?
add issuer route parameter? see realm.
declare scopes per domain?

Params

FSA_PARAM_STYLE any/http/json to restrict/force parameters? being lazy is not too bad?
check for bad char in parameter names
allow handling files in kwargs
add a filter on returned value? make_response?

Features

how to add a timeout? or manage an outside one?
logging default behavior is a pain and the maintainer is self satisfied. how to ensure that logging is initialized?
the doc and implementation should clarify exception handling, and possible overrides.
add ability to catch and process any user error. what about Flask?
json mode: generate json in more cases? automatically?
declare some exceptions to be turned into 400 instead of 500? currently this can be done below, eg anodb, maybe this is enough?

Software Engineering

reduce sloc?
check for more directive types (dynamically)?
add app.log?
take advantage of TypedDict?

Documentation

more recipes?
include demo? point to demo?
comparisons with other frameworks
use FlaskTester in tutorial?
add POST /token to tutorial?

FlaskSimpleAuth Documentation

Contents

Motivation

Install

Initialization

Authentication

Authentication Schemes

Password Management

Authorization

Group Authorizations

OAuth Authorizations

Object Authorizations

Parameters

Utils

`Reference` Object Wrapper

Error Handling

Miscellaneous Configuration Directives

See Also

Flask-Security

Flask-RESTful

Flask-AppBuilder

Others

License

Versions

TODO

Authn/Authz

Params

Features

Software Engineering

Documentation

FlaskSimpleAuth Documentation

Contents

Motivation

Install

Initialization

Authentication

Authentication Schemes

Password Management

Authorization

Group Authorizations

OAuth Authorizations

Object Authorizations

Parameters

Utils

Reference Object Wrapper

Error Handling

Miscellaneous Configuration Directives

See Also

Flask-Security

Flask-RESTful

Flask-AppBuilder

Flask-Login

Others

License

Versions

TODO

Authn/Authz

Params

Features

Software Engineering

Documentation

`Reference` Object Wrapper