Merge pull request #413 from MuckRock/jwt-get-create

Author: allanlasser

config/settings/base.py

@@ -418,7 +418,7 @@
     "HTML_SELECT_CUTOFF": 20,
     "DEFAULT_AUTHENTICATION_CLASSES": (
         "rest_framework.authentication.SessionAuthentication",
-        "rest_framework_simplejwt.authentication.JWTAuthentication",
+        "documentcloud.core.authentication.SquareletJWTAuthentication",
         "documentcloud.core.authentication.ProcessingTokenAuthentication",
     ),
     "DEFAULT_VERSIONING_CLASS": "documentcloud.core.versioning.QueryParameterVersioning",

documentcloud/core/authentication.py

@@ -7,6 +7,17 @@
 
 # Standard Library
 import hmac
+import logging
+
+# Third Party
+import requests
+from rest_framework_simplejwt.authentication import JWTAuthentication
+from rest_framework_simplejwt.settings import api_settings
+from squarelet_auth import settings as squarelet_settings
+from squarelet_auth.users.utils import squarelet_update_or_create
+from squarelet_auth.utils import squarelet_get
+
+logger = logging.getLogger(__name__)
 
 
 class ProcessingTokenAuthentication(BaseAuthentication):
@@ -44,3 +55,59 @@ def authenticate_credentials(self, key):
 
     def authenticate_header(self, request):
         return "processing-token"
+
+
+class SquareletJWTAuthentication(JWTAuthentication):
+    """JWT authentication that lazily provisions users from Squarelet.
+
+    Squarelet issues JWTs for users who may not yet have a mirrored ``User``
+    row in DocumentCloud's database. That row is normally created on first
+    interactive login, or via the asynchronous cache-invalidation webhook.
+    Either of these can lose a race against an immediate API call, like when
+    we're trying to fetch information about a user's add-ons in Klaxon.
+
+    When the user is missing locally we fetch their data from Squarelet
+    synchronously, create the row inline, and retry, so the very first
+    authenticated request is self-healing and the timing race is eliminated.
+
+    Provisioning is gated by ``SQUARELET_DISABLE_CREATE`` (via
+    ``squarelet_auth.settings.DISABLE_CREATE``), matching the webhook's
+    ``pull_data`` task: where creating users from Squarelet is disabled, an
+    unknown user still 401s rather than being provisioned here.
+    """
+
+    def get_user(self, validated_token):
+        try:
+            return super().get_user(validated_token)
+        except exceptions.AuthenticationFailed as exc:
+            # Only provision when the token is valid but the user simply does
+            # not exist locally yet. Genuinely invalid tokens (and any other
+            # failures) must still surface as a 401. simplejwt wraps its detail
+            # in a dict (``{"detail": ..., "code": ...}``) while plain DRF uses
+            # an ``ErrorDetail`` string, so handle both shapes.
+            if isinstance(exc.detail, dict):
+                code = exc.detail.get("code")
+            else:
+                code = getattr(exc.detail, "code", None)
+            if code != "user_not_found":
+                raise
+
+            # Respect the same gate as the webhook's pull_data task: when
+            # creating users from Squarelet is disabled, don't provision them
+            # here either -- let the request 401.
+            if squarelet_settings.DISABLE_CREATE:
+                raise
+
+            uuid = validated_token[api_settings.USER_ID_CLAIM]
+            logger.info("[JWT] Lazily provisioning user from Squarelet: %s", uuid)
+            try:
+                resp = squarelet_get(f"/api/users/{uuid}/")
+                resp.raise_for_status()
+                squarelet_update_or_create(uuid, resp.json())
+            except requests.exceptions.RequestException:
+                logger.exception("[JWT] Failed to fetch user from Squarelet: %s", uuid)
+                # Re-raise the original auth failure so the request 401s
+                raise exc
+
+            # Retry now that the user should exist locally
+            return super().get_user(validated_token)

documentcloud/core/tests.py

@@ -4,6 +4,7 @@
 from django.db import transaction
 from django.test import TestCase
 from django.urls import reverse
+from rest_framework.exceptions import AuthenticationFailed
 
 # Standard Library
 import hashlib
@@ -14,8 +15,11 @@
 
 # Third Party
 import pytest
+import requests
+from rest_framework_simplejwt.settings import api_settings
 
 # DocumentCloud
+from documentcloud.core.authentication import SquareletJWTAuthentication
 from documentcloud.users.tests.factories import UserFactory
 
 
@@ -93,3 +97,91 @@ def test_invalid_signature(self):
             f"{user.mailkey}@uploads.documentcloud.org", sign=False
         )
         assert response.status_code == 403
+
+
+@pytest.mark.django_db()
+class TestSquareletJWTAuthentication:
+    """Tests for lazy user provisioning during JWT authentication"""
+
+    def token(self, user_uuid):
+        """Build a minimal validated token carrying the user's uuid claim"""
+        return {api_settings.USER_ID_CLAIM: str(user_uuid)}
+
+    @mock.patch("documentcloud.core.authentication.squarelet_update_or_create")
+    @mock.patch("documentcloud.core.authentication.squarelet_get")
+    def test_existing_user(self, mock_get, mock_update):
+        """A user that already exists locally is returned without a callback"""
+        user = UserFactory()
+        auth = SquareletJWTAuthentication()
+
+        result = auth.get_user(self.token(user.uuid))
+
+        assert result == user
+        mock_get.assert_not_called()
+        mock_update.assert_not_called()
+
+    @mock.patch(
+        "documentcloud.core.authentication.squarelet_settings.DISABLE_CREATE", False
+    )
+    @mock.patch("documentcloud.core.authentication.squarelet_update_or_create")
+    @mock.patch("documentcloud.core.authentication.squarelet_get")
+    def test_lazy_provision_missing_user(self, mock_get, mock_update):
+        """A missing user is fetched from Squarelet, created, and returned"""
+        missing_uuid = uuid.uuid4()
+        data = {"preferred_username": "newuser", "organizations": []}
+        mock_get.return_value.json.return_value = data
+        # Simulate squarelet_update_or_create creating the local mirror row
+        mock_update.side_effect = lambda _uuid, _data: UserFactory(uuid=missing_uuid)
+        auth = SquareletJWTAuthentication()
+
+        result = auth.get_user(self.token(missing_uuid))
+
+        assert result.uuid == missing_uuid
+        mock_get.assert_called_once_with(f"/api/users/{missing_uuid}/")
+        # The uuid comes off the JWT claim as a string, matching how the
+        # webhook's pull_data task calls squarelet_update_or_create
+        mock_update.assert_called_once_with(str(missing_uuid), data)
+
+    @mock.patch("documentcloud.core.authentication.squarelet_update_or_create")
+    @mock.patch("documentcloud.core.authentication.squarelet_get")
+    def test_invalid_token_not_provisioned(self, mock_get, mock_update):
+        """A token without a user claim must 401 without contacting Squarelet"""
+        auth = SquareletJWTAuthentication()
+
+        with pytest.raises(AuthenticationFailed):
+            auth.get_user({})
+
+        mock_get.assert_not_called()
+        mock_update.assert_not_called()
+
+    @mock.patch(
+        "documentcloud.core.authentication.squarelet_settings.DISABLE_CREATE", False
+    )
+    @mock.patch("documentcloud.core.authentication.squarelet_update_or_create")
+    @mock.patch("documentcloud.core.authentication.squarelet_get")
+    def test_squarelet_fetch_fails(self, mock_get, mock_update):
+        """If the Squarelet fetch fails, the request still 401s"""
+        missing_uuid = uuid.uuid4()
+        mock_get.side_effect = requests.exceptions.RequestException
+        auth = SquareletJWTAuthentication()
+
+        with pytest.raises(AuthenticationFailed):
+            auth.get_user(self.token(missing_uuid))
+
+        mock_update.assert_not_called()
+
+    @mock.patch(
+        "documentcloud.core.authentication.squarelet_settings.DISABLE_CREATE", True
+    )
+    @mock.patch("documentcloud.core.authentication.squarelet_update_or_create")
+    @mock.patch("documentcloud.core.authentication.squarelet_get")
+    def test_disable_create_skips_provisioning(self, mock_get, mock_update):
+        """When SQUARELET_DISABLE_CREATE is set, missing users still 401"""
+        missing_uuid = uuid.uuid4()
+        auth = SquareletJWTAuthentication()
+
+        with pytest.raises(AuthenticationFailed):
+            auth.get_user(self.token(missing_uuid))
+
+        mock_get.assert_not_called()
+        mock_update.assert_not_called()