fix: Redact SSO PII before deletion

openedx/core/djangoapps/user_api/accounts/__init__.py

@@ -6,6 +6,9 @@
 from django.utils.text import format_lazy
 from django.utils.translation import gettext_lazy as _
 
+# Import signals to ensure they are registered
+from . import signals  # noqa: F401, pylint: disable=unused-import
+
 # The maximum length for the bio ("about me") account field
 BIO_MAX_LENGTH = 300
 

openedx/core/djangoapps/user_api/accounts/signals.py

@@ -2,8 +2,18 @@
 Django Signal related functionality for user_api accounts
 """
 
+import logging
 
-from django.dispatch import Signal
+from django.db.models.signals import pre_delete
+from django.dispatch import Signal, receiver
+from social_django.models import UserSocialAuth
+
+logger = logging.getLogger(__name__)
+
+# Prefix and suffix used to build a per-record redacted uid for UserSocialAuth.
+# Both the signal handler and bulk retirement code use these to stay in sync.
+REDACTED_SOCIAL_AUTH_UID_PREFIX = 'redacted-before-delete-'
+REDACTED_SOCIAL_AUTH_UID_SUFFIX = '@safe.com'
 
 # Signal to retire a user from LMS-initiated mailings (course mailings, etc)
 # providing_args=["user"]
@@ -16,3 +26,40 @@
 # Signal to retire LMS misc information
 # providing_args=["user"]
 USER_RETIRE_LMS_MISC = Signal()
+
+
+@receiver(pre_delete, sender=UserSocialAuth)
+def redact_social_auth_pii_before_deletion(sender, instance, **kwargs):  # pylint: disable=unused-argument
+    """
+    Redacts PII fields (uid, extra_data) before UserSocialAuth deletion.
+
+    Replaces uid with REDACTED_SOCIAL_AUTH_UID_PREFIX + pk + REDACTED_SOCIAL_AUTH_UID_SUFFIX
+    and clears extra_data.
+    Blocks deletion if redaction fails to prevent PII leaks to downstream systems.
+    """
+    if not instance or not instance.pk:
+        return
+
+    try:
+        update_fields = {}
+        redacted_uid = f'{REDACTED_SOCIAL_AUTH_UID_PREFIX}{instance.pk}{REDACTED_SOCIAL_AUTH_UID_SUFFIX}'
+
+        # These fields may have already been redacted as part of a bulk retirement,
+        # so we skip the update if it is already done to reduce query count.
+        if instance.uid != redacted_uid:
+            update_fields['uid'] = redacted_uid
+        if instance.extra_data:
+            update_fields['extra_data'] = {}
+
+        if not update_fields:
+            return
+
+        UserSocialAuth.objects.filter(pk=instance.pk).update(**update_fields)
+    except Exception:  # pylint: disable=broad-except
+        logger.exception(
+            "Failed to redact PII for UserSocialAuth before deletion: user_id=%s, provider=%s",
+            instance.user_id,
+            instance.provider,
+        )
+        # Re-raise to prevent deletion from proceeding without redaction
+        raise

openedx/core/djangoapps/user_api/accounts/tests/test_utils.py

@@ -6,10 +6,17 @@
 from completion.test_utils import CompletionWaffleTestMixin
 from django.test import TestCase
 from django.test.utils import override_settings
+from social_django.models import UserSocialAuth
 
 from common.djangoapps.student.models import CourseEnrollment
 from common.djangoapps.student.tests.factories import UserFactory
-from openedx.core.djangoapps.user_api.accounts.utils import retrieve_last_sitewide_block_completed
+from openedx.core.djangoapps.user_api.accounts.signals import (
+    REDACTED_SOCIAL_AUTH_UID_PREFIX,
+    REDACTED_SOCIAL_AUTH_UID_SUFFIX,
+)
+from openedx.core.djangoapps.user_api.accounts.utils import (
+    retrieve_last_sitewide_block_completed,
+)
 from openedx.core.djangolib.testing.utils import skip_unless_lms
 from xmodule.modulestore.tests.django_utils import (
     SharedModuleStoreTestCase,  # lint-amnesty, pylint: disable=wrong-import-order
@@ -133,3 +140,131 @@ def test_retrieve_last_sitewide_block_completed(self):
                )
 
         assert empty_block_url is None
+
+
+@skip_unless_lms
+class RedactUserSocialAuthPIITest(TestCase):
+    """
+    Tests for SSO PII redaction before deletion.
+    """
+
+    def setUp(self):
+        super().setUp()
+        self.user = UserFactory.create(username='testuser', email='testuser@example.com')
+
+    def create_social_auth(self, provider='google-oauth2', uid='user@example.com', extra_data=None):
+        """
+        Helper method to create UserSocialAuth instances for testing.
+        """
+        if extra_data is None:
+            extra_data = {
+                'email': 'user@example.com',
+                'name': 'Test User',
+                'id': '123456789',
+            }
+        return UserSocialAuth.objects.create(
+            user=self.user,
+            provider=provider,
+            uid=uid,
+            extra_data=extra_data,
+        )
+
+    def test_delete_redacts_user_social_auth_pii(self):
+        """
+        Test that deleting social auth redacts uid and extra_data before removal.
+        """
+        social_auth = self.create_social_auth()
+        social_auth_id = social_auth.id
+
+        captured_states = []
+
+        def capture_state_before_delete(sender, instance, **kwargs):  # pylint: disable=unused-argument
+            instance.refresh_from_db()
+            captured_states.append({
+                'id': instance.id,
+                'uid': instance.uid,
+                'extra_data': dict(instance.extra_data) if instance.extra_data else {},
+            })
+
+        from django.db.models.signals import pre_delete
+
+        pre_delete.connect(capture_state_before_delete, sender=UserSocialAuth)
+        try:
+            social_auth.delete()
+        finally:
+            pre_delete.disconnect(capture_state_before_delete, sender=UserSocialAuth)
+
+        assert captured_states == [{
+            'id': social_auth_id,
+            'uid': f'{REDACTED_SOCIAL_AUTH_UID_PREFIX}{social_auth_id}{REDACTED_SOCIAL_AUTH_UID_SUFFIX}',
+            'extra_data': {},
+        }]
+        assert not UserSocialAuth.objects.filter(id=social_auth_id).exists()
+
+    def test_delete_already_redacted_user_social_auth_is_idempotent(self):
+        """
+        Test that deleting an already redacted social auth keeps the redacted state.
+        """
+        social_auth = self.create_social_auth()
+        social_auth.uid = f'{REDACTED_SOCIAL_AUTH_UID_PREFIX}{social_auth.pk}{REDACTED_SOCIAL_AUTH_UID_SUFFIX}'
+        social_auth.extra_data = {}
+        social_auth.save(update_fields=['uid', 'extra_data'])
+        social_auth_id = social_auth.id
+
+        captured_states = []
+
+        def capture_state_before_delete(sender, instance, **kwargs):  # pylint: disable=unused-argument
+            instance.refresh_from_db()
+            captured_states.append((instance.uid, instance.extra_data))
+
+        from django.db.models.signals import pre_delete
+
+        pre_delete.connect(capture_state_before_delete, sender=UserSocialAuth)
+        try:
+            social_auth.delete()
+        finally:
+            pre_delete.disconnect(capture_state_before_delete, sender=UserSocialAuth)
+
+        assert captured_states == [
+            (f'{REDACTED_SOCIAL_AUTH_UID_PREFIX}{social_auth_id}{REDACTED_SOCIAL_AUTH_UID_SUFFIX}', {}),
+        ]
+        assert not UserSocialAuth.objects.filter(id=social_auth_id).exists()
+
+    def test_delete_redacts_multiple_sso_providers(self):
+        """
+        Test that deletion redacts multiple SSO providers before removal.
+        """
+        auths = [
+            self.create_social_auth(
+                provider='google-oauth2',
+                uid='google@example.com',
+                extra_data={'email': 'google@example.com', 'name': 'Google User'}
+            ),
+            self.create_social_auth(
+                provider='tpa-saml',
+                uid='saml@example.com',
+                extra_data={'email': 'saml@example.com', 'name': 'SAML User', 'uid': 'saml-uid'}
+            ),
+        ]
+        # Save IDs before deletion (they become None after delete)
+        auth_ids = [auth.pk for auth in auths]
+
+        captured_states = []
+
+        def capture_state_before_delete(sender, instance, **kwargs):  # pylint: disable=unused-argument
+            instance.refresh_from_db()
+            captured_states.append((instance.provider, instance.uid, instance.extra_data))
+
+        from django.db.models.signals import pre_delete
+
+        pre_delete.connect(capture_state_before_delete, sender=UserSocialAuth)
+        try:
+            for auth in auths:
+                auth.delete()
+        finally:
+            pre_delete.disconnect(capture_state_before_delete, sender=UserSocialAuth)
+
+        assert sorted(captured_states) == sorted([
+            ('google-oauth2', f'{REDACTED_SOCIAL_AUTH_UID_PREFIX}{auth_ids[0]}{REDACTED_SOCIAL_AUTH_UID_SUFFIX}', {}),
+            ('tpa-saml', f'{REDACTED_SOCIAL_AUTH_UID_PREFIX}{auth_ids[1]}{REDACTED_SOCIAL_AUTH_UID_SUFFIX}', {}),
+        ])

openedx/core/djangoapps/user_api/accounts/utils.py

@@ -11,6 +11,8 @@
 from completion.models import BlockCompletion
 from completion.waffle import ENABLE_COMPLETION_TRACKING_SWITCH
 from django.conf import settings
+from django.db.models import CharField, Value
+from django.db.models.functions import Cast, Concat
 from django.utils.translation import gettext as _
 from edx_django_utils.user import generate_password
 from social_django.models import UserSocialAuth
@@ -22,6 +24,7 @@
 from xmodule.modulestore.django import modulestore  # lint-amnesty, pylint: disable=wrong-import-order
 
 from ..models import UserRetirementStatus
+from .signals import REDACTED_SOCIAL_AUTH_UID_PREFIX, REDACTED_SOCIAL_AUTH_UID_SUFFIX
 
 ENABLE_SECONDARY_EMAIL_FEATURE_SWITCH = 'enable_secondary_email_feature'
 LOGGER = logging.getLogger(__name__)
@@ -204,8 +207,17 @@ def create_retirement_request_and_deactivate_account(user):
     # Add user to retirement queue.
     UserRetirementStatus.create_retirement(user)
 
-    # Unlink LMS social auth accounts
-    UserSocialAuth.objects.filter(user_id=user.id).delete()
+    # Redact and unlink LMS social auth accounts
+    social_auth_queryset = UserSocialAuth.objects.filter(user_id=user.id)
+    social_auth_queryset.update(
+        uid=Concat(
+            Value(REDACTED_SOCIAL_AUTH_UID_PREFIX),
+            Cast('id', output_field=CharField()),
+            Value(REDACTED_SOCIAL_AUTH_UID_SUFFIX),
+        ),
+        extra_data={},
+    )
+    social_auth_queryset.delete()
 
     # Change LMS password & email
     user.email = get_retired_email_by_email(user.email)

openedx/core/djangoapps/user_api/management/commands/retire_user.py

@@ -5,11 +5,14 @@
 from django.contrib.auth.models import User  # lint-amnesty, pylint: disable=imported-auth-user
 from django.core.management.base import BaseCommand, CommandError
 from django.db import transaction
+from django.db.models import CharField, Value
+from django.db.models.functions import Cast, Concat
 from social_django.models import UserSocialAuth
 
 from common.djangoapps.student.models import AccountRecovery, Registration, get_retired_email_by_email
 from openedx.core.djangolib.oauth2_retirement_utils import retire_dot_oauth2_models
 
+from ...accounts.signals import REDACTED_SOCIAL_AUTH_UID_PREFIX, REDACTED_SOCIAL_AUTH_UID_SUFFIX
 from ...models import BulkUserRetirementConfig, UserRetirementStatus
 
 logger = logging.getLogger(__name__)
@@ -144,8 +147,17 @@ def handle(self, *args, **options):
                 for user in users:
                     # Add user to retirement queue.
                     UserRetirementStatus.create_retirement(user)
-                    # Unlink LMS social auth accounts
-                    UserSocialAuth.objects.filter(user_id=user.id).delete()
+                    # Redact and unlink LMS social auth accounts
+                    social_auth_queryset = UserSocialAuth.objects.filter(user_id=user.id)
+                    social_auth_queryset.update(
+                        uid=Concat(
+                            Value(REDACTED_SOCIAL_AUTH_UID_PREFIX),
+                            Cast('id', output_field=CharField()),
+                            Value(REDACTED_SOCIAL_AUTH_UID_SUFFIX),
+                        ),
+                        extra_data={},
+                    )
+                    social_auth_queryset.delete()
                     # Change LMS password & email
                     user.email = get_retired_email_by_email(user.email)
                     user.set_unusable_password()