Fix #374: Strip EXIF metadata from uploaded avatars to avoid leaking private data

api/funkwhale_api/common/serializers.py

@@ -1,8 +1,12 @@
 import collections
+import io
+import PIL
+import os
 
 from rest_framework import serializers
 
 from django.core.exceptions import ObjectDoesNotExist
+from django.core.files.uploadedfile import SimpleUploadedFile
 from django.utils.encoding import smart_text
 from django.utils.translation import ugettext_lazy as _
 
@@ -190,3 +194,25 @@ def track_fields_for_update(*fields):
         return serializer_class
 
     return decorator
+
+
+class StripExifImageField(serializers.ImageField):
+    def to_internal_value(self, data):
+        file_obj = super().to_internal_value(data)
+
+        image = PIL.Image.open(file_obj)
+        data = list(image.getdata())
+        image_without_exif = PIL.Image.new(image.mode, image.size)
+        image_without_exif.putdata(data)
+
+        with io.BytesIO() as output:
+            image_without_exif.save(
+                output,
+                format=PIL.Image.EXTENSION[os.path.splitext(file_obj.name)[-1]],
+                quality=100,
+            )
+            content = output.getvalue()
+
+        return SimpleUploadedFile(
+            file_obj.name, content, content_type=file_obj.content_type
+        )

api/funkwhale_api/users/serializers.py

@@ -11,7 +11,7 @@ from rest_framework import serializers
 from versatileimagefield.serializers import VersatileImageFieldSerializer
 
 from funkwhale_api.activity import serializers as activity_serializers
-
+from funkwhale_api.common import serializers as common_serializers
 from . import models
 
 
@@ -66,7 +66,13 @@ class UserActivitySerializer(activity_serializers.ModelSerializer):
         return "Person"
 
 
-avatar_field = VersatileImageFieldSerializer(allow_null=True, sizes="square")
+class AvatarField(
+    common_serializers.StripExifImageField, VersatileImageFieldSerializer
+):
+    pass
+
+
+avatar_field = AvatarField(allow_null=True, sizes="square")
 
 
 class UserBasicSerializer(serializers.ModelSerializer):

api/tests/common/test_serializers.py

@@ -1,3 +1,8 @@
+import os
+import PIL
+
+from django.core.files.uploadedfile import SimpleUploadedFile
+
 import django_filters
 
 from funkwhale_api.common import serializers
@@ -163,3 +168,17 @@ def test_track_fields_for_update(mocker):
         {"field1": "value1", "field2": "value2"},
         {"field1": "newvalue1", "field2": "newvalue2"},
     )
+
+
+def test_strip_exif_field():
+    source_path = os.path.join(os.path.dirname(__file__), "exif.jpg")
+    source = PIL.Image.open(source_path)
+
+    assert bool(source._getexif())
+
+    with open(source_path, "rb") as f:
+        uploaded = SimpleUploadedFile("source.jpg", f.read(), content_type="image/jpeg")
+    field = serializers.StripExifImageField()
+
+    cleaned = PIL.Image.open(field.to_internal_value(uploaded))
+    assert cleaned._getexif() is None

changes/changelog.d/374.enhancement

@@ -0,0 +1 @@
+Strip EXIF metadata from uploaded avatars to avoid leaking private data (#374)