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 +2,15 @@
 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
+
+from .utils import redact_user_social_auth_pii
+
+logger = logging.getLogger(__name__)
 
 # Signal to retire a user from LMS-initiated mailings (course mailings, etc)
 # providing_args=["user"]
@@ -16,3 +23,32 @@
 # 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
+    """
+    Signal handler to redact PII from UserSocialAuth records before deletion.
+
+    This ensures that when SSO records are deleted (either via user retirement, manual unlinking,
+    or any other method), PII is redacted first. This prevents downstream systems that maintain
+    soft-deleted records from retaining sensitive user information.
+
+    Note: We call redact_user_social_auth_pii which saves the redacted data before the actual
+    deletion happens. This is intentional - downstream systems will capture the redacted
+    state before marking the record as deleted.
+
+    If redaction fails, the exception is re-raised to prevent deletion from proceeding,
+    ensuring GDPR compliance and preventing PII leaks to downstream systems.
+    """
+    try:
+        redact_user_social_auth_pii(instance)
+    except Exception as e:  # pylint: disable=broad-except
+        logger.exception(
+            "Failed to redact PII for UserSocialAuth before deletion: user_id=%s, provider=%s, error=%s",
+            instance.user_id,
+            instance.provider,
+            str(e)
+        )
+        # Re-raise to prevent deletion from proceeding without redaction
+        raise

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

@@ -6,10 +6,14 @@
 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.utils import (
+    redact_user_social_auth_pii,
+    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 +137,79 @@ 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_redact_user_social_auth_pii(self):
+        """
+        Test that redact_user_social_auth_pii correctly redacts uid and extra_data fields.
+        """
+        social_auth = self.create_social_auth()
+
+        redact_user_social_auth_pii(social_auth)
+        social_auth.refresh_from_db()
+
+        assert social_auth.uid == f'redacted_{social_auth.pk}@retired.invalid'
+        assert social_auth.extra_data == {}
+
+    def test_redact_user_social_auth_pii_idempotent(self):
+        """
+        Test that calling redact_user_social_auth_pii multiple times is idempotent.
+        """
+        social_auth = self.create_social_auth()
+
+        redact_user_social_auth_pii(social_auth)
+        # Duplicate call to redact user method to validate idempotency
+        redact_user_social_auth_pii(social_auth)
+        social_auth.refresh_from_db()
+
+        assert social_auth.uid == f'redacted_{social_auth.pk}@retired.invalid'
+        assert social_auth.extra_data == {}
+
+    def test_redact_multiple_sso_providers(self):
+        """
+        Test that redaction works correctly for multiple SSO providers.
+        """
+        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'}
+            ),
+        ]
+        for auth in auths:
+            redact_user_social_auth_pii(auth)
+            auth.refresh_from_db()
+            assert auth.uid == f'redacted_{auth.pk}@retired.invalid'
+            assert auth.extra_data == {}

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

@@ -196,6 +196,37 @@ def is_secondary_email_feature_enabled():
     return waffle.switch_is_active(ENABLE_SECONDARY_EMAIL_FEATURE_SWITCH)
 
 
+def redact_user_social_auth_pii(user_social_auth):
+    """
+    Redact PII from a UserSocialAuth record before deletion.
+
+    Downstream systems can retain deleted source rows as soft-deleted records, so sensitive
+    fields should be overwritten before deletion.
+
+    Uses a unique redacted UID per record (redacted_{pk}@retired.invalid) to avoid
+    IntegrityError from the unique_together constraint on (provider, uid).
+    """
+    if not user_social_auth or not user_social_auth.pk:
+        return
+
+    update_fields = {}
+
+    redacted_uid = f'redacted_{user_social_auth.pk}@retired.invalid'
+    if user_social_auth.uid != redacted_uid:
+        update_fields['uid'] = redacted_uid
+    if user_social_auth.extra_data:
+        update_fields['extra_data'] = {}
+
+    if not update_fields:
+        return
+
+    UserSocialAuth.objects.filter(pk=user_social_auth.pk).update(**update_fields)
+
+    # Keep instance in sync in case the caller reuses it.
+    for field_name, value in update_fields.items():
+        setattr(user_social_auth, field_name, value)
+
+
 def create_retirement_request_and_deactivate_account(user):
     """
     Adds user to retirement queue, unlinks social auth accounts, changes user passwords
@@ -204,8 +235,10 @@ 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
+    for social_auth in UserSocialAuth.objects.filter(user_id=user.id):
+        redact_user_social_auth_pii(social_auth)
+        social_auth.delete()
 
     # Change LMS password & email
     user.email = get_retired_email_by_email(user.email)

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

@@ -10,6 +10,7 @@
 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.utils import redact_user_social_auth_pii
 from ...models import BulkUserRetirementConfig, UserRetirementStatus
 
 logger = logging.getLogger(__name__)
@@ -144,8 +145,10 @@ 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
+                    for social_auth in UserSocialAuth.objects.filter(user_id=user.id):
+                        redact_user_social_auth_pii(social_auth)
+                        social_auth.delete()
                     # Change LMS password & email
                     user.email = get_retired_email_by_email(user.email)
                     user.set_unusable_password()

openedx/core/djangoapps/user_api/management/tests/test_retire_user.py

@@ -5,10 +5,12 @@
 
 import csv
 import os
+from unittest import mock
 
 import pytest
 from django.contrib.auth.models import User  # lint-amnesty, pylint: disable=imported-auth-user
 from django.core.management import CommandError, call_command
+from social_django.models import UserSocialAuth
 
 from common.djangoapps.student.tests.factories import UserFactory  # lint-amnesty, pylint: disable=wrong-import-order
 from openedx.core.djangoapps.user_api.accounts.tests.retirement_helpers import (  # lint-amnesty, pylint: disable=unused-import, wrong-import-order
@@ -107,3 +109,81 @@ def test_retire_with_username_email_userfile(setup_retirement_states):  # lint-a
     with pytest.raises(CommandError, match=r'You cannot use userfile option with username and user_email'):
         call_command('retire_user', user_file=user_file, username=username, user_email=user_email)
     remove_user_file()
+
+
+@skip_unless_lms
+def test_retire_user_redacts_sso_pii_before_deletion(setup_retirement_states):  # lint-amnesty, pylint: disable=redefined-outer-name, unused-argument  # noqa: F811
+    """
+    Test that SSO PII is redacted before UserSocialAuth records are deleted during retirement.
+
+    This test verifies the order of operations by capturing the record's state
+    at the moment of deletion to ensure it was already redacted.
+    """
+    user = UserFactory.create(username='sso-user', email='sso-user@example.com')
+    social_auth = UserSocialAuth.objects.create(
+        user=user,
+        provider='google-oauth2',
+        uid='sso-user@example.com',
+        extra_data={
+            'email': 'sso-user@example.com',
+            'name': 'SSO Test User',
+            'id': '123456789',
+        }
+    )
+    social_auth_id = social_auth.id
+
+    # Capture the state at the moment of deletion to verify redaction happened first
+    captured_state = {}
+    original_delete = UserSocialAuth.delete
+
+    def capture_state_and_delete(self):
+        """Wrapper to capture state before deletion."""
+        # Refresh from database to get the actual current state
+        self.refresh_from_db()
+        captured_state['uid'] = self.uid
+        captured_state['extra_data'] = dict(self.extra_data) if self.extra_data else {}
+        # Call original delete
+        return original_delete(self)
+
+    with mock.patch.object(UserSocialAuth, 'delete', capture_state_and_delete):
+        call_command('retire_user', username=user.username, user_email=user.email)
+
+    # Verify that at the moment of deletion, the record was already redacted
+    assert captured_state['uid'] == f'redacted_{social_auth_id}@retired.invalid', \
+        "UID should be redacted before deletion"
+    assert captured_state['extra_data'] == {}, \
+        "extra_data should be empty before deletion"
+
+    # Verify deletion completed
+    assert not UserSocialAuth.objects.filter(id=social_auth_id).exists()
+
+    retired_user_status = UserRetirementStatus.objects.filter(original_username=user.username).first()
+    assert retired_user_status is not None
+    assert retired_user_status.original_email == 'sso-user@example.com'
+
+
+@skip_unless_lms
+def test_retire_user_calls_redaction_for_each_social_auth(setup_retirement_states):  # lint-amnesty, pylint: disable=redefined-outer-name, unused-argument  # noqa: F811
+    """
+    Test that redact_user_social_auth_pii is called for each UserSocialAuth record during retirement.
+    """
+    user = UserFactory.create(username='multi-sso-user', email='multi-sso@example.com')
+    UserSocialAuth.objects.create(
+        user=user,
+        provider='google-oauth2',
+        uid='google-multi@example.com',
+        extra_data={'email': 'google-multi@example.com', 'name': 'Google User'}
+    )
+    UserSocialAuth.objects.create(
+        user=user,
+        provider='tpa-saml',
+        uid='saml-multi@example.com',
+        extra_data={'email': 'saml-multi@example.com', 'name': 'SAML User', 'uid': 'saml-123'}
+    )
+
+    with mock.patch(
+        'openedx.core.djangoapps.user_api.management.commands.retire_user.redact_user_social_auth_pii'
+    ) as mock_redact:
+        call_command('retire_user', username=user.username, user_email=user.email)
+
+    assert mock_redact.call_count == 2