Enforce image_size_cap on v2 upload

image_size_cap should be checked and enforced on upload

Enforcement is in two places:
- on image metadata save
- during image save to backend store

(cherry picked from commit 92ab00fca6)
Conflicts:
	glance/location.py
	glance/tests/functional/v2/test_images.py
	glance/tests/unit/test_store_image.py

Closes-Bug: 1315321
Change-Id: I45bfb360703617bc394e9e27fe17adf43b09c0e1
Co-Author: Manuel Desbonnet <manuel.desbonnet@hp.com>

glance/db/__init__.py

@@ -32,6 +32,7 @@ db_opt = cfg.BoolOpt('use_tpool',
                      'all DB API calls')
 
 CONF = cfg.CONF
+CONF.import_opt('image_size_cap', 'glance.common.config')
 CONF.import_opt('metadata_encryption_key', 'glance.common.config')
 CONF.register_opt(db_opt)
 
@@ -148,6 +149,8 @@ class ImageRepo(object):
 
     def add(self, image):
         image_values = self._format_image_to_db(image)
+        if image_values['size'] > CONF.image_size_cap:
+            raise exception.ImageSizeLimitExceeded
         # the updated_at value is not set in the _format_image_to_db
         # function since it is specific to image create
         image_values['updated_at'] = image.updated_at
@@ -159,6 +162,8 @@ class ImageRepo(object):
 
     def save(self, image):
         image_values = self._format_image_to_db(image)
+        if image_values['size'] > CONF.image_size_cap:
+            raise exception.ImageSizeLimitExceeded
         try:
             new_values = self.db_api.image_update(self.context,
                                                   image.image_id,

glance/store/__init__.py

@@ -646,7 +646,10 @@ class ImageProxy(glance.domain.proxy.Image):
             size = 0  # NOTE(markwash): zero -> unknown size
         location, size, checksum, loc_meta = self.store_api.add_to_backend(
                 self.context, CONF.default_store,
-                self.image.image_id, utils.CooperativeReader(data), size)
+                self.image.image_id,
+                utils.LimitingReader(utils.CooperativeReader(data),
+                                     CONF.image_size_cap),
+                size)
         self.image.locations = [{'url': location, 'metadata': loc_meta}]
         self.image.size = size
         self.image.checksum = checksum

glance/tests/functional/__init__.py

@@ -279,6 +279,7 @@ class ApiServer(Server):
         self.pid_file = pid_file or os.path.join(self.test_dir, "api.pid")
         self.scrubber_datadir = os.path.join(self.test_dir, "scrubber")
         self.log_file = os.path.join(self.test_dir, "api.log")
+        self.image_size_cap = 1099511627776
         self.s3_store_host = "s3.amazonaws.com"
         self.s3_store_access_key = ""
         self.s3_store_secret_key = ""
@@ -332,6 +333,7 @@ metadata_encryption_key = %(metadata_encryption_key)s
 registry_host = 127.0.0.1
 registry_port = %(registry_port)s
 log_file = %(log_file)s
+image_size_cap = %(image_size_cap)d
 s3_store_host = %(s3_store_host)s
 s3_store_access_key = %(s3_store_access_key)s
 s3_store_secret_key = %(s3_store_secret_key)s

glance/tests/functional/v2/test_images.py

@@ -259,6 +259,48 @@ class TestImages(functional.FunctionalTest):
 
         self.stop_servers()
 
+    def test_image_size_cap(self):
+        self.api_server.image_size_cap = 128
+        self.start_servers(**self.__dict__.copy())
+        # create an image
+        path = self._url('/v2/images')
+        headers = self._headers({'content-type': 'application/json'})
+        data = json.dumps({'name': 'image-size-cap-test-image',
+                           'type': 'kernel', 'disk_format': 'aki',
+                           'container_format': 'aki'})
+        response = requests.post(path, headers=headers, data=data)
+        self.assertEqual(201, response.status_code)
+
+        image = json.loads(response.text)
+        image_id = image['id']
+
+        #try to populate it with oversized data
+        path = self._url('/v2/images/%s/file' % image_id)
+        headers = self._headers({'Content-Type': 'application/octet-stream'})
+
+        class StreamSim(object):
+            # Using a one-shot iterator to force chunked transfer in the PUT
+            # request
+            def __init__(self, size):
+                self.size = size
+
+            def __iter__(self):
+                yield 'Z' * self.size
+
+        response = requests.put(path, headers=headers, data=StreamSim(
+                                self.api_server.image_size_cap + 1))
+        self.assertEqual(413, response.status_code)
+
+        # hashlib.md5('Z'*129).hexdigest()
+        #     == '76522d28cb4418f12704dfa7acd6e7ee'
+        # If the image has this checksum, it means that the whole stream was
+        # accepted and written to the store, which should not be the case.
+        path = self._url('/v2/images/{0}'.format(image_id))
+        headers = self._headers({'content-type': 'application/json'})
+        response = requests.get(path, headers=headers)
+        image_checksum = json.loads(response.text).get('checksum')
+        self.assertNotEqual(image_checksum, '76522d28cb4418f12704dfa7acd6e7ee')
+
     def test_permissions(self):
         # Create an image that belongs to TENANT1
         path = self._url('/v2/images')

glance/tests/unit/test_store_image.py

@@ -126,8 +126,10 @@ class TestStoreImage(utils.BaseTestCase):
 
         self.stubs.Set(unit_test_utils.FakeStoreAPI, 'get_from_backend',
                        fake_get_from_backend)
-
-        self.assertEquals(image1.get_data().fd, 'ZZZ')
+        # This time, image1.get_data() returns the data wrapped in a
+        # LimitingReader|CooperativeReader pipeline, so peeking under
+        # the hood of those objects to get at the underlying string.
+        self.assertEquals(image1.get_data().data.fd, 'ZZZ')
         image1.locations.pop(0)
         self.assertEquals(len(image1.locations), 1)
         image2.delete()

glance/tests/unit/utils.py

@@ -149,7 +149,10 @@ class FakeStoreAPI(object):
             if image_id in location:
                 raise exception.Duplicate()
         if not size:
-            size = len(data.fd)
+            # 'data' is a string wrapped in a LimitingReader|CooperativeReader
+            # pipeline, so peek under the hood of those objects to get at the
+            # string itself.
+            size = len(data.data.fd)
         if (current_store_size + size) > store_max_size:
             raise exception.StorageFull()
         if context.user == USER2: