auth: Add detailed comments for auth subsystem.

Now that we've more or less stabilized our authentication/registration
subsystem how we want it, it seems worth adding proper documentation
for this.

Fixes #7619.
This commit is contained in:
Tim Abbott 2019-03-09 17:43:29 -08:00
parent 0b02ea4e54
commit 873aca4a82
2 changed files with 229 additions and 39 deletions

View File

@ -71,6 +71,13 @@ def create_preregistration_user(email: str, request: HttpRequest, realm_creation
def maybe_send_to_registration(request: HttpRequest, email: str, full_name: str='',
is_signup: bool=False, password_required: bool=True,
multiuse_object_key: str='') -> HttpResponse:
"""Given a successful authentication for an email address (i.e. we've
confirmed the user controls the email address) that does not
currently have a Zulip account in the target realm, send them to
the registration flow or the "continue to registration" flow,
depending on is_signup, whether the email address can join the
organization (checked in HomepageForm), and similar details.
"""
realm = get_realm(get_subdomain(request))
from_multiuse_invite = False
multiuse_obj = None
@ -85,8 +92,11 @@ def maybe_send_to_registration(request: HttpRequest, email: str, full_name: str=
form = HomepageForm({'email': email}, realm=realm, from_multiuse_invite=from_multiuse_invite)
if form.is_valid():
# Construct a PreregistrationUser object and send the user over to
# the confirmation view.
# If the email address is allowed to sign up for an account in
# this organization, construct a PreregistrationUser and
# Confirmation objects, and then send the user to account
# creation or confirm-continue-registration depending on
# is_signup.
prereg_user = None
if settings.ONLY_SSO:
try:
@ -117,14 +127,15 @@ def maybe_send_to_registration(request: HttpRequest, email: str, full_name: str=
return render(request,
'zerver/confirm_continue_registration.html',
context=context)
else:
# This email address it not allowed to join this organization, so
# just send the user back to the registration page.
url = reverse('register')
return render(request,
'zerver/accounts_home.html',
context={'form': form, 'current_url': lambda: url,
'from_multiuse_invite': from_multiuse_invite,
'multiuse_object_key': multiuse_object_key},
)
'multiuse_object_key': multiuse_object_key})
def redirect_to_subdomain_login_url() -> HttpResponseRedirect:
login_url = reverse('django.contrib.auth.views.login')
@ -139,6 +150,24 @@ def login_or_register_remote_user(request: HttpRequest, remote_username: Optiona
invalid_subdomain: bool=False, mobile_flow_otp: Optional[str]=None,
is_signup: bool=False, redirect_to: str='',
multiuse_object_key: str='') -> HttpResponse:
"""Given a successful authentication showing the user controls given
email address (remote_username) and potentially a UserProfile
object (if the user already has a Zulip account), redirect the
browser to the appropriate place:
* The logged-in app if the user already has a Zulip account and is
trying to login, potentially to an initial narrow or page that had been
saved in the `redirect_to` parameter.
* The registration form if is_signup was set (i.e. the user is
trying to create a Zulip account)
* A special `confirm_continue_registration.html` "do you want to
register or try another account" if the user doesn't have a
Zulip account but is_signup is False (i.e. the user tried to login
and then did social authentication selecting an email address that does
not have a Zulip account in this organization).
* A zulip:// URL to send control back to the mobile apps if they
are doing authentication using the mobile_flow_otp flow.
"""
email = remote_user_to_email(remote_username)
if user_profile is None or user_profile.is_mirror_dummy:
# We have verified the user controls an email address, but
@ -162,11 +191,15 @@ def login_or_register_remote_user(request: HttpRequest, remote_username: Optiona
# We can't use HttpResponseRedirect, since it only allows HTTP(S) URLs
response = HttpResponse(status=302)
response['Location'] = 'zulip://login?' + urllib.parse.urlencode(params)
# Maybe sending 'user_logged_in' signal is the better approach:
# Since we are returning an API key instead of going through
# the Django login() function (which creates a browser
# session, etc.), the "new login" signal handler (which
# triggers an email notification new logins) will not run
# automatically. So we call it manually here.
#
# Arguably, sending a fake 'user_logged_in' signal would be a better approach:
# user_logged_in.send(sender=user_profile.__class__, request=request, user=user_profile)
# Not doing this only because over here we don't add the user information
# in the session. If the signal receiver assumes that we do then that
# would cause problems.
email_on_new_login(sender=user_profile.__class__, request=request, user=user_profile)
# Mark this request as having a logged-in user for our server logs.
@ -456,6 +489,12 @@ _subdomain_token_salt = 'zerver.views.auth.log_into_subdomain'
@log_view_func
def log_into_subdomain(request: HttpRequest, token: str) -> HttpResponse:
"""Given a valid signed authentication token (generated by
redirect_and_log_into_subdomain called on auth.zulip.example.com),
call login_or_register_remote_user, passing all the authentication
result data that had been encoded in the signed token.
"""
try:
data = signing.loads(token, salt=_subdomain_token_salt, max_age=15)
except signing.SignatureExpired as e:
@ -480,19 +519,36 @@ def log_into_subdomain(request: HttpRequest, token: str) -> HttpResponse:
else:
multiuse_object_key = ''
# We cannot pass the actual authenticated user_profile object that
# was fetched by the original authentication backend and passed
# into redirect_and_log_into_subdomain through a signed URL token,
# so we need to re-fetch it from the database.
if is_signup:
# If we are signing up, user_profile should be None. In case
# email_address already exists, user will get an error message.
# If we are creating a new user account, user_profile will
# always have been None, so we set that here. In the event
# that a user account with this email was somehow created in a
# race, the eventual registration code will catch that and
# throw an error, so we don't need to check for that here.
user_profile = None
return_data = {} # type: Dict[str, Any]
else:
# We can be reasonably confident that this subdomain actually
# has a corresponding realm, since it was referenced in a
# signed cookie. But we probably should add some error
# handling for the case where the realm disappeared in the
# meantime.
# We're just trying to login. We can be reasonably confident
# that this subdomain actually has a corresponding active
# realm, since the signed cookie proves there was one very
# recently. But as part of fetching the UserProfile object
# for the target user, we use DummyAuthBackend, which
# conveniently re-validates that the realm and user account
# were not deactivated in the meantime.
# Note: Ideally, we'd have a nice user-facing error message
# for the case where this auth fails (because e.g. the realm
# or user was deactivated since the signed cookie was
# generated < 15 seconds ago), but the authentication result
# is correct in those cases and such a race would be very
# rare, so a nice error message is low priority.
realm = get_realm(subdomain)
user_profile, return_data = authenticate_remote_user(realm, email_address)
invalid_subdomain = bool(return_data.get('invalid_subdomain'))
return login_or_register_remote_user(request, email_address, user_profile,
full_name, invalid_subdomain=invalid_subdomain,

View File

@ -1,3 +1,17 @@
# Documentation for Zulip's authentication backends is split across a few places:
#
# * https://zulip.readthedocs.io/en/latest/production/authentication-methods.html and
# zproject/prod_settings_template.py have user-level configuration documentation.
# * https://zulip.readthedocs.io/en/latest/subsystems/auth.html has developer-level
# documentation, especially on testing authentication backends in the Zulip
# development environment.
#
# Django upstream's documentation for authentication backends is also
# helpful background. The most important detail to understand for
# reading this file is that the Django authenticate() function will
# call the authenticate methods of all backends registered in
# settings.AUTHENTICATION_BACKENDS that have a function signature
# matching the args/kwargs passed in the authenticate() call.
import logging
from typing import Any, Dict, List, Optional, Set, Tuple, Union
@ -25,6 +39,12 @@ from zerver.models import CustomProfileField, PreregistrationUser, UserProfile,
custom_profile_fields_for_realm, get_default_stream_groups, get_user_profile_by_id, \
remote_user_to_email, email_to_username, get_realm, get_user_by_delivery_email
# This first batch of methods is used by other code in Zulip to check
# whether a given authentication backend is enabled for a given realm.
# In each case, we both needs to check at the server level (via
# `settings.AUTHENTICATION_BACKENDS`, queried via
# `django.contrib.auth.get_backends`) and at the realm level (via the
# `Realm.authentication_methods` BitField).
def pad_method_dict(method_dict: Dict[str, bool]) -> Dict[str, bool]:
"""Pads an authentication methods dict to contain all auth backends
supported by the software, regardless of whether they are
@ -79,6 +99,11 @@ def require_email_format_usernames(realm: Optional[Realm]=None) -> bool:
def common_get_active_user(email: str, realm: Realm,
return_data: Optional[Dict[str, Any]]=None) -> Optional[UserProfile]:
"""This is the core common function used by essentially all
authentication backends to check if there's an active user account
with a given email address in the organization, handling both
user-level and realm-level deactivation correctly.
"""
try:
user_profile = get_user_by_delivery_email(email, realm)
except UserProfile.DoesNotExist:
@ -105,18 +130,26 @@ def common_get_active_user(email: str, realm: Realm,
return user_profile
class ZulipAuthMixin:
"""This common mixin is used to override Django's default behavior for
looking up a logged-in user by ID to use a version that fetches
from memcached before checking the database (avoiding a database
query in most cases).
"""
def get_user(self, user_profile_id: int) -> Optional[UserProfile]:
""" Get a UserProfile object from the user_profile_id. """
"""Override the Django method for getting a UserProfile object from
the user_profile_id,."""
try:
return get_user_profile_by_id(user_profile_id)
except UserProfile.DoesNotExist:
return None
class ZulipDummyBackend(ZulipAuthMixin):
"""
Used when we want to log you in without checking any
"""Used when we want to log you in without checking any
authentication (i.e. new user registration or when otherwise
authentication has already been checked earlier in the process).
We ensure that this backend only ever successfully authenticates
when explicitly requested by including the use_dummy_backend kwarg.
"""
def authenticate(self, username: Optional[str]=None, realm: Optional[Realm]=None,
@ -131,10 +164,9 @@ class ZulipDummyBackend(ZulipAuthMixin):
class EmailAuthBackend(ZulipAuthMixin):
"""
Email Authentication Backend
Email+Password Authentication Backend (the default).
Allows a user to sign in using an email/password pair rather than
a username/password pair.
Allows a user to sign in using an email/password pair.
"""
def authenticate(self, username: Optional[str]=None, password: Optional[str]=None,
@ -206,6 +238,14 @@ class GoogleMobileOauth2Backend(ZulipAuthMixin):
return common_get_active_user(token_payload["email"], realm, return_data)
class ZulipRemoteUserBackend(RemoteUserBackend):
"""Authentication backend that reads the Apache REMOTE_USER variable.
Used primarily in enterprise environments with an SSO solution
that has an Apache REMOTE_USER integration. For manual testing, see
https://zulip.readthedocs.io/en/latest/production/authentication-methods.html
See also remote_user_sso in zerver/views/auth.py.
"""
create_unknown_user = False
def authenticate(self, remote_user: Optional[str], realm: Optional[Realm]=None,
@ -227,6 +267,13 @@ def is_valid_email(email: str) -> bool:
return True
def email_belongs_to_ldap(realm: Realm, email: str) -> bool:
"""Used to make determinations on whether a user's email address is
managed by LDAP. For environments using both LDAP and
Email+Password authentication, we do not allow EmailAuthBackend
authentication for email addresses managed by LDAP (to avoid a
security issue where one create separate credentials for an LDAP
user), and this function is used to enforce that rule.
"""
if not ldap_auth_enabled(realm):
return False
@ -253,27 +300,34 @@ class ZulipLDAPConfigurationError(Exception):
LDAP_USER_ACCOUNT_CONTROL_DISABLED_MASK = 2
class ZulipLDAPAuthBackendBase(ZulipAuthMixin, LDAPBackend):
"""Common code between LDAP authentication (ZulipLDAPAuthBackend) and
using LDAP just to sync user data (ZulipLDAPUserPopulator).
To fully understand our LDAP backend, you may want to skim
django_auth_ldap/backend.py from the upstream django-auth-ldap
library. It's not a lot of code, and searching around in that
file makes the flow for LDAP authentication clear.
"""
def __init__(self) -> None:
# Used to initialize a fake LDAP directly for both manual
# and automated testing in a development environment where
# there is no actual LDAP server.
if settings.DEVELOPMENT and settings.FAKE_LDAP_MODE: # nocoverage
init_fakeldap()
# Don't use Django LDAP's permissions functions
# Disable django-auth-ldap's permissions functions -- we don't use
# the standard Django user/group permissions system because they
# are prone to performance issues.
def has_perm(self, user: Optional[UserProfile], perm: Any, obj: Any=None) -> bool:
# Using Any type is safe because we are not doing anything with
# the arguments.
return False
def has_module_perms(self, user: Optional[UserProfile], app_label: Optional[str]) -> bool:
return False
def get_all_permissions(self, user: Optional[UserProfile], obj: Any=None) -> Set[Any]:
# Using Any type is safe because we are not doing anything with
# the arguments and always return empty set.
return set()
def get_group_permissions(self, user: Optional[UserProfile], obj: Any=None) -> Set[Any]:
# Using Any type is safe because we are not doing anything with
# the arguments and always return empty set.
return set()
def django_to_ldap_username(self, username: str) -> str:
@ -306,12 +360,17 @@ class ZulipLDAPAuthBackendBase(ZulipAuthMixin, LDAPBackend):
do_change_avatar_fields(user, UserProfile.AVATAR_FROM_USER)
def is_account_control_disabled_user(self, ldap_user: _LDAPUser) -> bool:
"""Implements the userAccountControl check for whether a user has been
disabled in an Active Directory server being integrated with
Zulip via LDAP."""
account_control_value = ldap_user.attrs[settings.AUTH_LDAP_USER_ATTR_MAP['userAccountControl']][0]
ldap_disabled = bool(int(account_control_value) & LDAP_USER_ACCOUNT_CONTROL_DISABLED_MASK)
return ldap_disabled
@classmethod
def get_mapped_name(cls, ldap_user: _LDAPUser) -> Tuple[str, str]:
"""Constructs the user's Zulip full_name and short_name fields from
the LDAP data"""
if "full_name" in settings.AUTH_LDAP_USER_ATTR_MAP:
full_name_attr = settings.AUTH_LDAP_USER_ATTR_MAP["full_name"]
short_name = full_name = ldap_user.attrs[full_name_attr][0]
@ -386,6 +445,10 @@ class ZulipLDAPAuthBackendBase(ZulipAuthMixin, LDAPBackend):
def get_or_build_user(self, username: str,
ldap_user: _LDAPUser) -> Tuple[UserProfile, bool]:
"""This is used only in non-authentication contexts such as:
./manage.py sync_ldap_user_data
In authentication contexts, this is overriden in ZulipLDAPAuthBackend.
"""
(user, built) = super().get_or_build_user(username, ldap_user)
self.sync_avatar_from_ldap(user, ldap_user)
self.sync_full_name_from_ldap(user, ldap_user)
@ -423,12 +486,31 @@ class ZulipLDAPAuthBackend(ZulipLDAPAuthBackendBase):
return_data['outside_ldap_domain'] = True
return None
# Call into (ultimately) the django-auth-ldap authenticate
# function. This will check the username/password pair
# against the LDAP database, and assuming those are correct,
# end up calling `self.get_or_build_user` with the
# authenticated user's data from LDAP.
return ZulipLDAPAuthBackendBase.authenticate(self,
request=None,
username=username,
password=password)
def get_or_build_user(self, username: str, ldap_user: _LDAPUser) -> Tuple[UserProfile, bool]:
"""The main function of our authentication backend extension of
django-auth-ldap. When this is called (from `authenticate`),
django-auth-ldap will already have verified that the provided
username and password match those in the LDAP database.
This function's responsibility is to check (1) whether the
email address for this user obtained from LDAP has an active
account in this Zulip realm. If so, it will log them in.
Otherwise, to provide a seamless Single Sign-On experience
with LDAP, this function can automatically create a new Zulip
user account in the realm (assuming the realm is configured to
allow that email address to sign up).
"""
return_data = {} # type: Dict[str, Any]
if settings.LDAP_EMAIL_ATTR is not None:
@ -488,8 +570,13 @@ class ZulipLDAPAuthBackend(ZulipLDAPAuthBackendBase):
return user_profile, True
# Just like ZulipLDAPAuthBackend, but doesn't let you log in.
class ZulipLDAPUserPopulator(ZulipLDAPAuthBackendBase):
"""Just like ZulipLDAPAuthBackend, but doesn't let you log in. Used
for syncing data like names, avatars, and custom profile fields
from LDAP in `manage.py sync_ldap_user_data` as well as in
registration for organizations that use a different SSO solution
for managing login (often via RemoteUserBackend).
"""
def authenticate(self, username: str, password: str, realm: Optional[Realm]=None,
return_data: Optional[Dict[str, Any]]=None) -> None:
return None
@ -504,8 +591,8 @@ def sync_user_from_ldap(user_profile: UserProfile) -> bool:
return True
class DevAuthBackend(ZulipAuthMixin):
# Allow logging in as any user without a password.
# This is used for convenience when developing Zulip.
"""Allow logging in as any user without a password. This is used for
convenience when developing Zulip, and is disabled in production."""
def authenticate(self, dev_auth_username: Optional[str]=None, realm: Optional[Realm]=None,
return_data: Optional[Dict[str, Any]]=None) -> Optional[UserProfile]:
assert dev_auth_username is not None
@ -520,6 +607,8 @@ def social_associate_user_helper(backend: BaseAuth, return_data: Dict[str, Any],
"""Responsible for doing the Zulip-account lookup and validation parts
of the Zulip Social auth pipeline (similar to the authenticate()
methods in most other auth backends in this file).
Returns a UserProfile object for successful authentication, and None otherwise.
"""
subdomain = backend.strategy.session_get('subdomain')
realm = get_realm(subdomain)
@ -536,8 +625,12 @@ def social_associate_user_helper(backend: BaseAuth, return_data: Dict[str, Any],
return_data["social_auth_failed_reason"] = kwargs['response']["auth_failed_reason"]
return None
elif hasattr(backend, 'get_verified_emails'):
# Some social backends, like GitHubAuthBackend, don't guarantee that
# the `details` data is validated.
# Some social backends, like GitHubAuthBackend, don't
# guarantee that the `details` data is validated (i.e., it's
# possible users can put any string they want in the "email"
# field of the `details` object). For those backends, we have
# custom per-backend code to properly fetch only verified
# email addresses from the appropriate third-party API.
verified_emails = backend.get_verified_emails(*args, **kwargs)
if len(verified_emails) == 0:
# TODO: Provide a nice error message screen to the user
@ -581,6 +674,12 @@ def social_auth_associate_user(
backend: BaseAuth,
*args: Any,
**kwargs: Any) -> Dict[str, Any]:
"""A simple wrapper function to reformat the return data from
social_associate_user_helper as a dictionary. The
python-social-auth infrastructure will then pass those values into
later stages of settings.SOCIAL_AUTH_PIPELINE, such as
social_auth_finish, as kwargs.
"""
return_data = {} # type: Dict[str, Any]
user_profile = social_associate_user_helper(
backend, return_data, *args, **kwargs)
@ -593,6 +692,13 @@ def social_auth_finish(backend: Any,
response: HttpResponse,
*args: Any,
**kwargs: Any) -> Optional[UserProfile]:
"""Given the determination in social_auth_associate_user for whether
the user should be authenticated, this takes care of actually
logging in the user (if appropriate) and redirecting the browser
to the appropriate next page depending on the situation. Read the
comments below as well as login_or_register_remote_user in
`zerver/views/auth.py` for the details on how that dispatch works.
"""
from zerver.views.auth import (login_or_register_remote_user,
redirect_and_log_into_subdomain)
@ -642,13 +748,41 @@ def social_auth_finish(backend: Any,
realm = Realm.objects.get(id=return_data["realm_id"])
multiuse_object_key = strategy.session_get('multiuse_object_key', '')
mobile_flow_otp = strategy.session_get('mobile_flow_otp')
# At this point, we have now confirmed that the user has
# demonstrated control over the target email address.
#
# The next step is to call login_or_register_remote_user, but
# there are two code paths here because of an optimization to save
# a redirect on mobile.
if mobile_flow_otp is not None:
# For mobile app authentication, login_or_register_remote_user
# will redirect to a special zulip:// URL that is handled by
# the app after a successful authentication; so we can
# redirect directly from here, saving a round trip over what
# we need to do to create session cookies on the right domain
# in the web login flow (below).
return login_or_register_remote_user(strategy.request, email_address,
user_profile, full_name,
invalid_subdomain=bool(invalid_subdomain),
mobile_flow_otp=mobile_flow_otp,
is_signup=is_signup,
redirect_to=redirect_to)
# If this authentication code were executing on
# subdomain.zulip.example.com, we would just call
# login_or_register_remote_user as in the mobile code path.
# However, because third-party SSO providers generally don't allow
# wildcard addresses in their redirect URLs, for multi-realm
# servers, we will have just completed authentication on e.g.
# auth.zulip.example.com (depending on
# settings.SOCIAL_AUTH_SUBDOMAIN), which cannot store cookies on
# the subdomain.zulip.example.com domain. So instead we serve a
# redirect (encoding the authentication result data in a
# cryptographically signed token) to a route on
# subdomain.zulip.example.com that will verify the signature and
# then call login_or_register_remote_user.
return redirect_and_log_into_subdomain(realm, full_name, email_address,
is_signup=is_signup,
redirect_to=redirect_to,