5320ecbaf2
md5 is not an approved algorithm in FIPS mode, and trying to instantiate a hashlib.md5() will fail when the system is running in FIPS mode. md5 is allowed when in a non-security context. There is a plan to add a keyword parameter (usedforsecurity) to hashlib.md5() to annotate whether or not the instance is being used in a security context. In the case where it is not, the instantiation of md5 will be allowed. See https://bugs.python.org/issue9216 for more details. Some downstream python versions already support this parameter. To support these versions, a new encapsulation of md5() is added to swift/common/utils.py. This encapsulation is identical to the one being added to oslo.utils, but is recreated here to avoid adding a dependency. This patch is to replace the instances of hashlib.md5() with this new encapsulation, adding an annotation indicating whether the usage is a security context or not. While this patch seems large, it is really just the same change over and again. Reviewers need to pay particular attention as to whether the keyword parameter (usedforsecurity) is set correctly. Right now, all of them appear to be not used in a security context. Now that all the instances have been converted, we can update the bandit run to look for these instances and ensure that new invocations do not creep in. With this latest patch, the functional and unit tests all pass on a FIPS enabled system. Co-Authored-By: Pete Zaitcev Change-Id: Ibb4917da4c083e1e094156d748708b87387f2d87
3046 lines
122 KiB
Python
3046 lines
122 KiB
Python
#!/usr/bin/python -u
|
|
# Copyright (c) 2010-2012 OpenStack Foundation
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
|
# implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
from datetime import datetime
|
|
import io
|
|
import locale
|
|
import random
|
|
import six
|
|
from six.moves import urllib
|
|
import time
|
|
import unittest
|
|
import uuid
|
|
from copy import deepcopy
|
|
import eventlet
|
|
from swift.common.http import is_success, is_client_error
|
|
from swift.common.swob import normalize_etag
|
|
from swift.common.utils import md5
|
|
from email.utils import parsedate
|
|
|
|
if six.PY2:
|
|
from email.parser import FeedParser
|
|
else:
|
|
from email.parser import BytesFeedParser as FeedParser
|
|
|
|
import mock
|
|
|
|
from test.functional import normalized_urls, load_constraint, cluster_info
|
|
from test.functional import check_response, retry
|
|
import test.functional as tf
|
|
from test.functional.swift_test_client import Account, Connection, File, \
|
|
ResponseError, SkipTest
|
|
|
|
|
|
def setUpModule():
|
|
tf.setup_package()
|
|
|
|
|
|
def tearDownModule():
|
|
tf.teardown_package()
|
|
|
|
|
|
class Utils(object):
|
|
@classmethod
|
|
def create_ascii_name(cls, length=None):
|
|
return uuid.uuid4().hex
|
|
|
|
@classmethod
|
|
def create_utf8_name(cls, length=None):
|
|
if length is None:
|
|
length = 15
|
|
else:
|
|
length = int(length)
|
|
|
|
utf8_chars = u'\uF10F\uD20D\uB30B\u9409\u8508\u5605\u3703\u1801'\
|
|
u'\u0900\uF110\uD20E\uB30C\u940A\u8509\u5606\u3704'\
|
|
u'\u1802\u0901\uF111\uD20F\uB30D\u940B\u850A\u5607'\
|
|
u'\u3705\u1803\u0902\uF112\uD210\uB30E\u940C\u850B'\
|
|
u'\u5608\u3706\u1804\u0903\u03A9\u2603'
|
|
ustr = u''.join([random.choice(utf8_chars)
|
|
for x in range(length)])
|
|
if six.PY2:
|
|
return ustr.encode('utf-8')
|
|
return ustr
|
|
|
|
create_name = create_ascii_name
|
|
|
|
|
|
class BaseEnv(object):
|
|
account = conn = None
|
|
|
|
@classmethod
|
|
def setUp(cls):
|
|
cls.conn = Connection(tf.config)
|
|
cls.conn.authenticate()
|
|
cls.account = Account(cls.conn, tf.config.get('account',
|
|
tf.config['username']))
|
|
cls.account.delete_containers()
|
|
|
|
@classmethod
|
|
def tearDown(cls):
|
|
pass
|
|
|
|
|
|
class Base(unittest.TestCase):
|
|
env = BaseEnv
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
cls.env.tearDown()
|
|
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.env.setUp()
|
|
|
|
def setUp(self):
|
|
if tf.in_process:
|
|
tf.skip_if_no_xattrs()
|
|
|
|
def assert_body(self, body):
|
|
if not isinstance(body, bytes):
|
|
body = body.encode('utf-8')
|
|
response_body = self.env.conn.response.read()
|
|
self.assertEqual(response_body, body,
|
|
'Body returned: %s' % (response_body))
|
|
|
|
def assert_status(self, status_or_statuses):
|
|
self.assertTrue(
|
|
self.env.conn.response.status == status_or_statuses or
|
|
(hasattr(status_or_statuses, '__iter__') and
|
|
self.env.conn.response.status in status_or_statuses),
|
|
'Status returned: %d Expected: %s' %
|
|
(self.env.conn.response.status, status_or_statuses))
|
|
|
|
def assert_header(self, header_name, expected_value):
|
|
try:
|
|
actual_value = self.env.conn.response.getheader(header_name)
|
|
except KeyError:
|
|
self.fail(
|
|
'Expected header name %r not found in response.' % header_name)
|
|
self.assertEqual(expected_value, actual_value)
|
|
|
|
def assert_etag(self, unquoted_value):
|
|
if tf.cluster_info.get('etag_quoter', {}).get('enable_by_default'):
|
|
expected = '"%s"' % unquoted_value
|
|
else:
|
|
expected = unquoted_value
|
|
self.assert_header('etag', expected)
|
|
|
|
|
|
class Base2(object):
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
Utils.create_name = Utils.create_utf8_name
|
|
super(Base2, cls).setUpClass()
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
Utils.create_name = Utils.create_ascii_name
|
|
|
|
|
|
class TestAccountEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestAccountEnv, cls).setUp()
|
|
cls.containers = []
|
|
for i in range(10):
|
|
cont = cls.account.container(Utils.create_name())
|
|
if not cont.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
cls.containers.append(cont)
|
|
|
|
|
|
class TestAccountDev(Base):
|
|
env = TestAccountEnv
|
|
|
|
|
|
class TestAccountDevUTF8(Base2, TestAccountDev):
|
|
pass
|
|
|
|
|
|
class TestAccount(Base):
|
|
env = TestAccountEnv
|
|
|
|
def testNoAuthToken(self):
|
|
self.assertRaises(ResponseError, self.env.account.info,
|
|
cfg={'no_auth_token': True})
|
|
self.assert_status([401, 412])
|
|
|
|
self.assertRaises(ResponseError, self.env.account.containers,
|
|
cfg={'no_auth_token': True})
|
|
self.assert_status([401, 412])
|
|
|
|
def testInvalidUTF8Path(self):
|
|
valid_utf8 = Utils.create_utf8_name()
|
|
if six.PY2:
|
|
invalid_utf8 = valid_utf8[::-1]
|
|
else:
|
|
invalid_utf8 = (valid_utf8.encode('utf8')[::-1]).decode(
|
|
'utf-8', 'surrogateescape')
|
|
container = self.env.account.container(invalid_utf8)
|
|
self.assertFalse(container.create(cfg={'no_path_quote': True}))
|
|
self.assert_status(412)
|
|
self.assert_body('Invalid UTF8 or contains NULL')
|
|
|
|
def testVersionOnlyPath(self):
|
|
self.env.account.conn.make_request('PUT',
|
|
cfg={'version_only_path': True})
|
|
self.assert_status(412)
|
|
self.assert_body('Bad URL')
|
|
|
|
def testInvalidPath(self):
|
|
was_path = self.env.account.conn.storage_path
|
|
if (normalized_urls):
|
|
self.env.account.conn.storage_path = '/'
|
|
else:
|
|
self.env.account.conn.storage_path = "/%s" % was_path
|
|
try:
|
|
self.env.account.conn.make_request('GET')
|
|
self.assert_status(404)
|
|
finally:
|
|
self.env.account.conn.storage_path = was_path
|
|
|
|
def testPUTError(self):
|
|
if load_constraint('allow_account_management'):
|
|
raise SkipTest("Allow account management is enabled")
|
|
self.env.account.conn.make_request('PUT')
|
|
self.assert_status([403, 405])
|
|
|
|
def testAccountHead(self):
|
|
try_count = 0
|
|
while try_count < 5:
|
|
try_count += 1
|
|
|
|
info = self.env.account.info()
|
|
for field in ['object_count', 'container_count', 'bytes_used']:
|
|
self.assertGreaterEqual(info[field], 0)
|
|
|
|
if info['container_count'] == len(self.env.containers):
|
|
break
|
|
|
|
if try_count < 5:
|
|
time.sleep(1)
|
|
|
|
self.assertEqual(info['container_count'], len(self.env.containers))
|
|
self.assert_status(204)
|
|
|
|
def testContainerSerializedInfo(self):
|
|
container_info = {}
|
|
for container in self.env.containers:
|
|
info = {'bytes': 0}
|
|
info['count'] = random.randint(10, 30)
|
|
for i in range(info['count']):
|
|
file_item = container.file(Utils.create_name())
|
|
bytes = random.randint(1, 32768)
|
|
file_item.write_random(bytes)
|
|
info['bytes'] += bytes
|
|
|
|
container_info[container.name] = info
|
|
|
|
for format_type in ['json', 'xml']:
|
|
for a in self.env.account.containers(
|
|
parms={'format': format_type}):
|
|
self.assertGreaterEqual(a['count'], 0)
|
|
self.assertGreaterEqual(a['bytes'], 0)
|
|
|
|
headers = dict((k.lower(), v)
|
|
for k, v in self.env.conn.response.getheaders())
|
|
if format_type == 'json':
|
|
self.assertEqual(headers['content-type'],
|
|
'application/json; charset=utf-8')
|
|
elif format_type == 'xml':
|
|
self.assertEqual(headers['content-type'],
|
|
'application/xml; charset=utf-8')
|
|
|
|
def testListingLimit(self):
|
|
limit = load_constraint('account_listing_limit')
|
|
for lim in (1, 100, limit / 2, limit - 1, limit, limit + 1, limit * 2):
|
|
p = {'limit': lim}
|
|
|
|
if lim <= limit:
|
|
self.assertLessEqual(len(self.env.account.containers(parms=p)),
|
|
lim)
|
|
self.assert_status(200)
|
|
else:
|
|
self.assertRaises(ResponseError,
|
|
self.env.account.containers, parms=p)
|
|
self.assert_status(412)
|
|
|
|
def testContainerListing(self):
|
|
a = sorted([c.name for c in self.env.containers])
|
|
|
|
for format_type in [None, 'json', 'xml']:
|
|
b = self.env.account.containers(parms={'format': format_type})
|
|
|
|
if isinstance(b[0], dict):
|
|
b = [x['name'] for x in b]
|
|
|
|
self.assertEqual(a, b)
|
|
|
|
def testListDelimiter(self):
|
|
delimiter = '-'
|
|
containers = ['test', delimiter.join(['test', 'bar']),
|
|
delimiter.join(['test', 'foo'])]
|
|
for c in containers:
|
|
cont = self.env.account.container(c)
|
|
self.assertTrue(cont.create())
|
|
|
|
results = self.env.account.containers(parms={'delimiter': delimiter})
|
|
expected = ['test', 'test-']
|
|
results = [r for r in results if r in expected]
|
|
self.assertEqual(expected, results)
|
|
|
|
results = self.env.account.containers(parms={'delimiter': delimiter,
|
|
'reverse': 'yes'})
|
|
expected.reverse()
|
|
results = [r for r in results if r in expected]
|
|
self.assertEqual(expected, results)
|
|
|
|
def testListMultiCharDelimiter(self):
|
|
delimiter = '-&'
|
|
containers = ['test', delimiter.join(['test', 'bar']),
|
|
delimiter.join(['test', 'foo'])]
|
|
for c in containers:
|
|
cont = self.env.account.container(c)
|
|
self.assertTrue(cont.create())
|
|
|
|
results = self.env.account.containers(parms={'delimiter': delimiter})
|
|
expected = ['test', 'test-&']
|
|
results = [r for r in results if r in expected]
|
|
self.assertEqual(expected, results)
|
|
|
|
results = self.env.account.containers(parms={'delimiter': delimiter,
|
|
'reverse': 'yes'})
|
|
expected.reverse()
|
|
results = [r for r in results if r in expected]
|
|
self.assertEqual(expected, results)
|
|
|
|
def testListDelimiterAndPrefix(self):
|
|
delimiter = 'a'
|
|
containers = ['bar', 'bazar']
|
|
for c in containers:
|
|
cont = self.env.account.container(c)
|
|
self.assertTrue(cont.create())
|
|
|
|
results = self.env.account.containers(parms={'delimiter': delimiter,
|
|
'prefix': 'ba'})
|
|
expected = ['bar', 'baza']
|
|
results = [r for r in results if r in expected]
|
|
self.assertEqual(expected, results)
|
|
|
|
results = self.env.account.containers(parms={'delimiter': delimiter,
|
|
'prefix': 'ba',
|
|
'reverse': 'yes'})
|
|
expected.reverse()
|
|
results = [r for r in results if r in expected]
|
|
self.assertEqual(expected, results)
|
|
|
|
def testContainerListingLastModified(self):
|
|
expected = {}
|
|
for container in self.env.containers:
|
|
res = container.info()
|
|
expected[container.name] = time.mktime(
|
|
parsedate(res['last_modified']))
|
|
|
|
for format_type in ['json', 'xml']:
|
|
actual = {}
|
|
containers = self.env.account.containers(
|
|
parms={'format': format_type})
|
|
if isinstance(containers[0], dict):
|
|
for container in containers:
|
|
self.assertIn('name', container) # sanity
|
|
self.assertIn('last_modified', container) # sanity
|
|
# ceil by hand (wants easier way!)
|
|
datetime_str, micro_sec_str = \
|
|
container['last_modified'].split('.')
|
|
timestamp = time.mktime(
|
|
time.strptime(datetime_str,
|
|
"%Y-%m-%dT%H:%M:%S"))
|
|
if int(micro_sec_str):
|
|
timestamp += 1
|
|
actual[container['name']] = timestamp
|
|
|
|
self.assertEqual(expected, actual)
|
|
|
|
def testInvalidAuthToken(self):
|
|
hdrs = {'X-Auth-Token': 'bogus_auth_token'}
|
|
self.assertRaises(ResponseError, self.env.account.info, hdrs=hdrs)
|
|
self.assert_status(401)
|
|
|
|
def testLastContainerMarker(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
containers = self.env.account.containers(parms={
|
|
'format': format_type})
|
|
self.assertEqual(len(containers), len(self.env.containers))
|
|
self.assert_status(200)
|
|
|
|
marker = (containers[-1] if format_type is None
|
|
else containers[-1]['name'])
|
|
containers = self.env.account.containers(
|
|
parms={'format': format_type, 'marker': marker})
|
|
self.assertEqual(len(containers), 0)
|
|
if format_type is None:
|
|
self.assert_status(204)
|
|
else:
|
|
self.assert_status(200)
|
|
|
|
def testMarkerLimitContainerList(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
for marker in ['0', 'A', 'I', 'R', 'Z', 'a', 'i', 'r', 'z',
|
|
'abc123', 'mnop', 'xyz']:
|
|
|
|
limit = random.randint(2, 9)
|
|
containers = self.env.account.containers(
|
|
parms={'format': format_type,
|
|
'marker': marker,
|
|
'limit': limit})
|
|
self.assertLessEqual(len(containers), limit)
|
|
if containers:
|
|
if isinstance(containers[0], dict):
|
|
containers = [x['name'] for x in containers]
|
|
self.assertGreater(locale.strcoll(containers[0], marker),
|
|
0)
|
|
|
|
def testContainersOrderedByName(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
containers = self.env.account.containers(
|
|
parms={'format': format_type})
|
|
if isinstance(containers[0], dict):
|
|
containers = [x['name'] for x in containers]
|
|
self.assertEqual(sorted(containers, key=locale.strxfrm),
|
|
containers)
|
|
|
|
def testQuotedWWWAuthenticateHeader(self):
|
|
# check that the www-authenticate header value with the swift realm
|
|
# is correctly quoted.
|
|
conn = Connection(tf.config)
|
|
conn.authenticate()
|
|
inserted_html = '<b>Hello World'
|
|
hax = 'AUTH_haxx"\nContent-Length: %d\n\n%s' % (len(inserted_html),
|
|
inserted_html)
|
|
quoted_hax = urllib.parse.quote(hax)
|
|
conn.connection.request('GET', '/v1/' + quoted_hax, None, {})
|
|
resp = conn.connection.getresponse()
|
|
|
|
resp_headers = {}
|
|
for h, v in resp.getheaders():
|
|
h = h.lower()
|
|
if h in resp_headers:
|
|
# py2 would do this for us, but py3 apparently keeps them
|
|
# separate? Not sure which I like more...
|
|
resp_headers[h] += ',' + v
|
|
else:
|
|
resp_headers[h] = v
|
|
|
|
self.assertIn('www-authenticate', resp_headers)
|
|
actual = resp_headers['www-authenticate']
|
|
expected = 'Swift realm="%s"' % quoted_hax
|
|
# other middleware e.g. auth_token may also set www-authenticate
|
|
# headers in which case actual values will be a comma separated list.
|
|
# check that expected value is among the actual values
|
|
self.assertIn(expected, actual)
|
|
|
|
|
|
class TestAccountUTF8(Base2, TestAccount):
|
|
pass
|
|
|
|
|
|
class TestAccountNoContainers(Base):
|
|
def testGetRequest(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
self.assertFalse(self.env.account.containers(
|
|
parms={'format': format_type}))
|
|
|
|
if format_type is None:
|
|
self.assert_status(204)
|
|
else:
|
|
self.assert_status(200)
|
|
|
|
|
|
class TestAccountNoContainersUTF8(Base2, TestAccountNoContainers):
|
|
pass
|
|
|
|
|
|
class TestAccountSortingEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestAccountSortingEnv, cls).setUp()
|
|
postfix = Utils.create_name()
|
|
cls.cont_items = ('a1', 'a2', 'A3', 'b1', 'B2', 'a10', 'b10', 'zz')
|
|
cls.cont_items = ['%s%s' % (x, postfix) for x in cls.cont_items]
|
|
|
|
for container in cls.cont_items:
|
|
c = cls.account.container(container)
|
|
if not c.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
|
|
class TestAccountSorting(Base):
|
|
env = TestAccountSortingEnv
|
|
|
|
def testAccountContainerListSorting(self):
|
|
# name (byte order) sorting.
|
|
cont_list = sorted(self.env.cont_items)
|
|
for reverse in ('false', 'no', 'off', '', 'garbage'):
|
|
cont_listing = self.env.account.containers(
|
|
parms={'reverse': reverse})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing,
|
|
'Expected %s but got %s with reverse param %r'
|
|
% (cont_list, cont_listing, reverse))
|
|
|
|
def testAccountContainerListSortingReverse(self):
|
|
# name (byte order) sorting.
|
|
cont_list = sorted(self.env.cont_items)
|
|
cont_list.reverse()
|
|
for reverse in ('true', '1', 'yes', 'on', 't', 'y'):
|
|
cont_listing = self.env.account.containers(
|
|
parms={'reverse': reverse})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing,
|
|
'Expected %s but got %s with reverse param %r'
|
|
% (cont_list, cont_listing, reverse))
|
|
|
|
def testAccountContainerListSortingByPrefix(self):
|
|
cont_list = sorted(c for c in self.env.cont_items if c.startswith('a'))
|
|
cont_list.reverse()
|
|
cont_listing = self.env.account.containers(parms={
|
|
'reverse': 'on', 'prefix': 'a'})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing)
|
|
|
|
def testAccountContainerListSortingByMarkersExclusive(self):
|
|
first_item = self.env.cont_items[3] # 'b1' + postfix
|
|
last_item = self.env.cont_items[4] # 'B2' + postfix
|
|
|
|
cont_list = sorted(c for c in self.env.cont_items
|
|
if last_item < c < first_item)
|
|
cont_list.reverse()
|
|
cont_listing = self.env.account.containers(parms={
|
|
'reverse': 'on', 'marker': first_item, 'end_marker': last_item})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing)
|
|
|
|
def testAccountContainerListSortingByMarkersInclusive(self):
|
|
first_item = self.env.cont_items[3] # 'b1' + postfix
|
|
last_item = self.env.cont_items[4] # 'B2' + postfix
|
|
|
|
cont_list = sorted(c for c in self.env.cont_items
|
|
if last_item <= c <= first_item)
|
|
cont_list.reverse()
|
|
cont_listing = self.env.account.containers(parms={
|
|
'reverse': 'on', 'marker': first_item + '\x00',
|
|
'end_marker': last_item[:-1] + chr(ord(last_item[-1]) - 1)})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing)
|
|
|
|
def testAccountContainerListSortingByReversedMarkers(self):
|
|
cont_listing = self.env.account.containers(parms={
|
|
'reverse': 'on', 'marker': 'B', 'end_marker': 'b1'})
|
|
self.assert_status(204)
|
|
self.assertEqual([], cont_listing)
|
|
|
|
|
|
class TestContainerEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestContainerEnv, cls).setUp()
|
|
cls.container = cls.account.container(Utils.create_name())
|
|
if not cls.container.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
cls.file_count = 10
|
|
cls.file_size = 128
|
|
cls.files = list()
|
|
for x in range(cls.file_count):
|
|
file_item = cls.container.file(Utils.create_name())
|
|
file_item.write_random(cls.file_size)
|
|
cls.files.append(file_item.name)
|
|
|
|
|
|
class TestContainerDev(Base):
|
|
env = TestContainerEnv
|
|
|
|
|
|
class TestContainerDevUTF8(Base2, TestContainerDev):
|
|
pass
|
|
|
|
|
|
class TestContainer(Base):
|
|
env = TestContainerEnv
|
|
|
|
def testContainerNameLimit(self):
|
|
limit = load_constraint('max_container_name_length')
|
|
|
|
for lim in (limit - 100, limit - 10, limit - 1, limit,
|
|
limit + 1, limit + 10, limit + 100):
|
|
cont = self.env.account.container('a' * lim)
|
|
if lim <= limit:
|
|
self.assertTrue(cont.create())
|
|
self.assert_status((201, 202))
|
|
else:
|
|
self.assertFalse(cont.create())
|
|
self.assert_status(400)
|
|
|
|
def testFileThenContainerDelete(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
file_item = cont.file(Utils.create_name())
|
|
self.assertTrue(file_item.write_random())
|
|
|
|
self.assertTrue(file_item.delete())
|
|
self.assert_status(204)
|
|
self.assertNotIn(file_item.name, cont.files())
|
|
|
|
self.assertTrue(cont.delete())
|
|
self.assert_status(204)
|
|
self.assertNotIn(cont.name, self.env.account.containers())
|
|
|
|
def testFileListingLimitMarkerPrefix(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
|
|
files = sorted([Utils.create_name() for x in range(10)])
|
|
for f in files:
|
|
file_item = cont.file(f)
|
|
self.assertTrue(file_item.write_random())
|
|
|
|
for i in range(len(files)):
|
|
f = files[i]
|
|
for j in range(1, len(files) - i):
|
|
self.assertEqual(cont.files(parms={'limit': j, 'marker': f}),
|
|
files[i + 1: i + j + 1])
|
|
self.assertEqual(cont.files(parms={'marker': f}), files[i + 1:])
|
|
self.assertEqual(cont.files(parms={'marker': f, 'prefix': f}), [])
|
|
self.assertEqual(cont.files(parms={'prefix': f}), [f])
|
|
|
|
def testPrefixAndLimit(self):
|
|
load_constraint('container_listing_limit')
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
|
|
prefix_file_count = 10
|
|
limit_count = 2
|
|
prefixs = ['alpha/', 'beta/', 'kappa/']
|
|
prefix_files = {}
|
|
|
|
for prefix in prefixs:
|
|
prefix_files[prefix] = []
|
|
|
|
for i in range(prefix_file_count):
|
|
file_item = cont.file(prefix + Utils.create_name())
|
|
file_item.write()
|
|
prefix_files[prefix].append(file_item.name)
|
|
|
|
for format_type in [None, 'json', 'xml']:
|
|
for prefix in prefixs:
|
|
files = cont.files(parms={'prefix': prefix,
|
|
'format': format_type})
|
|
if isinstance(files[0], dict):
|
|
files = [x.get('name', x.get('subdir')) for x in files]
|
|
self.assertEqual(files, sorted(prefix_files[prefix]))
|
|
|
|
for format_type in [None, 'json', 'xml']:
|
|
for prefix in prefixs:
|
|
files = cont.files(parms={'limit': limit_count,
|
|
'prefix': prefix,
|
|
'format': format_type})
|
|
if isinstance(files[0], dict):
|
|
files = [x.get('name', x.get('subdir')) for x in files]
|
|
self.assertEqual(len(files), limit_count)
|
|
|
|
for file_item in files:
|
|
self.assertTrue(file_item.startswith(prefix))
|
|
|
|
def testListDelimiter(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
|
|
delimiter = '-'
|
|
files = ['test', delimiter.join(['test', 'bar']),
|
|
delimiter.join(['test', 'foo'])]
|
|
for f in files:
|
|
file_item = cont.file(f)
|
|
self.assertTrue(file_item.write_random())
|
|
|
|
for format_type in [None, 'json', 'xml']:
|
|
results = cont.files(parms={'format': format_type})
|
|
if isinstance(results[0], dict):
|
|
results = [x.get('name', x.get('subdir')) for x in results]
|
|
self.assertEqual(results, ['test', 'test-bar', 'test-foo'])
|
|
|
|
results = cont.files(parms={'delimiter': delimiter,
|
|
'format': format_type})
|
|
if isinstance(results[0], dict):
|
|
results = [x.get('name', x.get('subdir')) for x in results]
|
|
self.assertEqual(results, ['test', 'test-'])
|
|
|
|
results = cont.files(parms={'delimiter': delimiter,
|
|
'format': format_type,
|
|
'reverse': 'yes'})
|
|
if isinstance(results[0], dict):
|
|
results = [x.get('name', x.get('subdir')) for x in results]
|
|
self.assertEqual(results, ['test-', 'test'])
|
|
|
|
def testListMultiCharDelimiter(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
|
|
delimiter = '-&'
|
|
files = ['test', delimiter.join(['test', 'bar']),
|
|
delimiter.join(['test', 'foo']), "test-'baz"]
|
|
for f in files:
|
|
file_item = cont.file(f)
|
|
self.assertTrue(file_item.write_random())
|
|
|
|
for format_type in [None, 'json', 'xml']:
|
|
results = cont.files(parms={'format': format_type})
|
|
if isinstance(results[0], dict):
|
|
results = [x.get('name', x.get('subdir')) for x in results]
|
|
self.assertEqual(results, ['test', 'test-&bar', 'test-&foo',
|
|
"test-'baz"])
|
|
|
|
results = cont.files(parms={'delimiter': delimiter,
|
|
'format': format_type})
|
|
if isinstance(results[0], dict):
|
|
results = [x.get('name', x.get('subdir')) for x in results]
|
|
self.assertEqual(results, ['test', 'test-&', "test-'baz"])
|
|
|
|
results = cont.files(parms={'delimiter': delimiter,
|
|
'format': format_type,
|
|
'reverse': 'yes'})
|
|
if isinstance(results[0], dict):
|
|
results = [x.get('name', x.get('subdir')) for x in results]
|
|
self.assertEqual(results, ["test-'baz", 'test-&', 'test'])
|
|
|
|
def testListDelimiterAndPrefix(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
|
|
delimiter = 'a'
|
|
files = ['bar', 'bazar']
|
|
for f in files:
|
|
file_item = cont.file(f)
|
|
self.assertTrue(file_item.write_random())
|
|
|
|
results = cont.files(parms={'delimiter': delimiter, 'prefix': 'ba'})
|
|
self.assertEqual(results, ['bar', 'baza'])
|
|
|
|
results = cont.files(parms={'delimiter': delimiter,
|
|
'prefix': 'ba',
|
|
'reverse': 'yes'})
|
|
self.assertEqual(results, ['baza', 'bar'])
|
|
|
|
def testLeadingDelimiter(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
|
|
delimiter = '/'
|
|
files = ['test', delimiter.join(['', 'test', 'bar']),
|
|
delimiter.join(['', 'test', 'bar', 'foo'])]
|
|
for f in files:
|
|
file_item = cont.file(f)
|
|
self.assertTrue(file_item.write_random())
|
|
|
|
results = cont.files(parms={'delimiter': delimiter})
|
|
self.assertEqual(results, [delimiter, 'test'])
|
|
|
|
def testCreate(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
self.assert_status(201)
|
|
self.assertIn(cont.name, self.env.account.containers())
|
|
|
|
def testContainerFileListOnContainerThatDoesNotExist(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
container = self.env.account.container(Utils.create_name())
|
|
self.assertRaises(ResponseError, container.files,
|
|
parms={'format': format_type})
|
|
self.assert_status(404)
|
|
|
|
def testUtf8Container(self):
|
|
valid_utf8 = Utils.create_utf8_name()
|
|
if six.PY2:
|
|
invalid_utf8 = valid_utf8[::-1]
|
|
else:
|
|
invalid_utf8 = (valid_utf8.encode('utf8')[::-1]).decode(
|
|
'utf-8', 'surrogateescape')
|
|
container = self.env.account.container(valid_utf8)
|
|
self.assertTrue(container.create(cfg={'no_path_quote': True}))
|
|
self.assertIn(container.name, self.env.account.containers())
|
|
self.assertEqual(container.files(), [])
|
|
self.assertTrue(container.delete())
|
|
|
|
container = self.env.account.container(invalid_utf8)
|
|
self.assertFalse(container.create(cfg={'no_path_quote': True}))
|
|
self.assert_status(412)
|
|
self.assertRaises(ResponseError, container.files,
|
|
cfg={'no_path_quote': True})
|
|
self.assert_status(412)
|
|
|
|
def testCreateOnExisting(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
self.assert_status(201)
|
|
self.assertTrue(cont.create())
|
|
self.assert_status(202)
|
|
|
|
def testSlashInName(self):
|
|
if six.PY2:
|
|
cont_name = list(Utils.create_name().decode('utf-8'))
|
|
else:
|
|
cont_name = list(Utils.create_name())
|
|
cont_name[random.randint(2, len(cont_name) - 2)] = '/'
|
|
cont_name = ''.join(cont_name)
|
|
if six.PY2:
|
|
cont_name = cont_name.encode('utf-8')
|
|
|
|
cont = self.env.account.container(cont_name)
|
|
self.assertFalse(cont.create(cfg={'no_path_quote': True}),
|
|
'created container with name %s' % (cont_name))
|
|
self.assert_status(404)
|
|
self.assertNotIn(cont.name, self.env.account.containers())
|
|
|
|
def testDelete(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
self.assert_status(201)
|
|
self.assertTrue(cont.delete())
|
|
self.assert_status(204)
|
|
self.assertNotIn(cont.name, self.env.account.containers())
|
|
|
|
def testDeleteOnContainerThatDoesNotExist(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertFalse(cont.delete())
|
|
self.assert_status(404)
|
|
|
|
def testDeleteOnContainerWithFiles(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(cont.create())
|
|
file_item = cont.file(Utils.create_name())
|
|
file_item.write_random(self.env.file_size)
|
|
self.assertIn(file_item.name, cont.files())
|
|
self.assertFalse(cont.delete())
|
|
self.assert_status(409)
|
|
|
|
def testFileCreateInContainerThatDoesNotExist(self):
|
|
file_item = File(self.env.conn, self.env.account, Utils.create_name(),
|
|
Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write)
|
|
self.assert_status(404)
|
|
|
|
def testLastFileMarker(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
files = self.env.container.files(parms={'format': format_type})
|
|
self.assertEqual(len(files), len(self.env.files))
|
|
self.assert_status(200)
|
|
|
|
marker = files[-1] if format_type is None else files[-1]['name']
|
|
files = self.env.container.files(
|
|
parms={'format': format_type, 'marker': marker})
|
|
self.assertEqual(len(files), 0)
|
|
|
|
if format_type is None:
|
|
self.assert_status(204)
|
|
else:
|
|
self.assert_status(200)
|
|
|
|
def testContainerFileList(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
files = self.env.container.files(parms={'format': format_type})
|
|
self.assert_status(200)
|
|
if isinstance(files[0], dict):
|
|
files = [x['name'] for x in files]
|
|
|
|
for file_item in self.env.files:
|
|
self.assertIn(file_item, files)
|
|
|
|
for file_item in files:
|
|
self.assertIn(file_item, self.env.files)
|
|
|
|
def _testContainerFormattedFileList(self, format_type):
|
|
expected = {}
|
|
for name in self.env.files:
|
|
expected[name] = self.env.container.file(name).info()
|
|
|
|
file_list = self.env.container.files(parms={'format': format_type})
|
|
self.assert_status(200)
|
|
for actual in file_list:
|
|
name = actual['name']
|
|
self.assertIn(name, expected)
|
|
if tf.cluster_info.get('etag_quoter', {}).get('enable_by_default'):
|
|
self.assertEqual(expected[name]['etag'],
|
|
'"%s"' % actual['hash'])
|
|
else:
|
|
self.assertEqual(expected[name]['etag'], actual['hash'])
|
|
self.assertEqual(
|
|
expected[name]['content_type'], actual['content_type'])
|
|
self.assertEqual(
|
|
expected[name]['content_length'], actual['bytes'])
|
|
expected.pop(name)
|
|
self.assertFalse(expected) # sanity check
|
|
|
|
def testContainerJsonFileList(self):
|
|
self._testContainerFormattedFileList('json')
|
|
|
|
def testContainerXmlFileList(self):
|
|
self._testContainerFormattedFileList('xml')
|
|
|
|
def testMarkerLimitFileList(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
for marker in ['0', 'A', 'I', 'R', 'Z', 'a', 'i', 'r', 'z',
|
|
'abc123', 'mnop', 'xyz']:
|
|
limit = random.randint(2, self.env.file_count - 1)
|
|
files = self.env.container.files(parms={'format': format_type,
|
|
'marker': marker,
|
|
'limit': limit})
|
|
|
|
if not files:
|
|
continue
|
|
|
|
if isinstance(files[0], dict):
|
|
files = [x['name'] for x in files]
|
|
|
|
self.assertLessEqual(len(files), limit)
|
|
if files:
|
|
if isinstance(files[0], dict):
|
|
files = [x['name'] for x in files]
|
|
self.assertGreater(locale.strcoll(files[0], marker), 0)
|
|
|
|
def testFileOrder(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
files = self.env.container.files(parms={'format': format_type})
|
|
if isinstance(files[0], dict):
|
|
files = [x['name'] for x in files]
|
|
self.assertEqual(sorted(files, key=locale.strxfrm), files)
|
|
|
|
def testContainerInfo(self):
|
|
info = self.env.container.info()
|
|
self.assert_status(204)
|
|
self.assertEqual(info['object_count'], self.env.file_count)
|
|
self.assertEqual(info['bytes_used'],
|
|
self.env.file_count * self.env.file_size)
|
|
|
|
def testContainerInfoOnContainerThatDoesNotExist(self):
|
|
container = self.env.account.container(Utils.create_name())
|
|
self.assertRaises(ResponseError, container.info)
|
|
self.assert_status(404)
|
|
|
|
def testContainerFileListWithLimit(self):
|
|
for format_type in [None, 'json', 'xml']:
|
|
files = self.env.container.files(parms={'format': format_type,
|
|
'limit': 2})
|
|
self.assertEqual(len(files), 2)
|
|
|
|
def testContainerExistenceCachingProblem(self):
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertRaises(ResponseError, cont.files)
|
|
self.assertTrue(cont.create())
|
|
self.assertEqual(cont.files(), [])
|
|
|
|
cont = self.env.account.container(Utils.create_name())
|
|
self.assertRaises(ResponseError, cont.files)
|
|
self.assertTrue(cont.create())
|
|
# NB: no GET! Make sure the PUT cleared the cached 404
|
|
file_item = cont.file(Utils.create_name())
|
|
file_item.write_random()
|
|
|
|
def testContainerLastModified(self):
|
|
container = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(container.create())
|
|
info = container.info()
|
|
t0 = info['last_modified']
|
|
# last modified header is in date format which supports in second
|
|
# so we need to wait to increment a sec in the header.
|
|
eventlet.sleep(1)
|
|
|
|
# POST container change last modified timestamp
|
|
self.assertTrue(
|
|
container.update_metadata({'x-container-meta-japan': 'mitaka'}))
|
|
info = container.info()
|
|
t1 = info['last_modified']
|
|
self.assertNotEqual(t0, t1)
|
|
eventlet.sleep(1)
|
|
|
|
# PUT container (overwrite) also change last modified
|
|
self.assertTrue(container.create())
|
|
info = container.info()
|
|
t2 = info['last_modified']
|
|
self.assertNotEqual(t1, t2)
|
|
eventlet.sleep(1)
|
|
|
|
# PUT object doesn't change container last modified timestamp
|
|
obj = container.file(Utils.create_name())
|
|
self.assertTrue(
|
|
obj.write(b"aaaaa", hdrs={'Content-Type': 'text/plain'}))
|
|
info = container.info()
|
|
t3 = info['last_modified']
|
|
self.assertEqual(t2, t3)
|
|
|
|
# POST object also doesn't change container last modified timestamp
|
|
self.assertTrue(
|
|
obj.sync_metadata({'us': 'austin'}))
|
|
info = container.info()
|
|
t4 = info['last_modified']
|
|
self.assertEqual(t2, t4)
|
|
|
|
|
|
class TestContainerUTF8(Base2, TestContainer):
|
|
pass
|
|
|
|
|
|
class TestContainerSortingEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestContainerSortingEnv, cls).setUp()
|
|
cls.container = cls.account.container(Utils.create_name())
|
|
if not cls.container.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
cls.file_items = ('a1', 'a2', 'A3', 'b1', 'B2', 'a10', 'b10', 'zz')
|
|
cls.files = list()
|
|
cls.file_size = 128
|
|
for name in cls.file_items:
|
|
file_item = cls.container.file(name)
|
|
file_item.write_random(cls.file_size)
|
|
cls.files.append(file_item.name)
|
|
|
|
|
|
class TestContainerSorting(Base):
|
|
env = TestContainerSortingEnv
|
|
|
|
def testContainerFileListSortingReversed(self):
|
|
file_list = list(sorted(self.env.file_items))
|
|
file_list.reverse()
|
|
for reverse in ('true', '1', 'yes', 'on', 't', 'y'):
|
|
cont_files = self.env.container.files(parms={'reverse': reverse})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files,
|
|
'Expected %s but got %s with reverse param %r'
|
|
% (file_list, cont_files, reverse))
|
|
|
|
def testContainerFileSortingByPrefixReversed(self):
|
|
cont_list = sorted(c for c in self.env.file_items if c.startswith('a'))
|
|
cont_list.reverse()
|
|
cont_listing = self.env.container.files(parms={
|
|
'reverse': 'on', 'prefix': 'a'})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing)
|
|
|
|
def testContainerFileSortingByMarkersExclusiveReversed(self):
|
|
first_item = self.env.file_items[3] # 'b1' + postfix
|
|
last_item = self.env.file_items[4] # 'B2' + postfix
|
|
|
|
cont_list = sorted(c for c in self.env.file_items
|
|
if last_item < c < first_item)
|
|
cont_list.reverse()
|
|
cont_listing = self.env.container.files(parms={
|
|
'reverse': 'on', 'marker': first_item, 'end_marker': last_item})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing)
|
|
|
|
def testContainerFileSortingByMarkersInclusiveReversed(self):
|
|
first_item = self.env.file_items[3] # 'b1' + postfix
|
|
last_item = self.env.file_items[4] # 'B2' + postfix
|
|
|
|
cont_list = sorted(c for c in self.env.file_items
|
|
if last_item <= c <= first_item)
|
|
cont_list.reverse()
|
|
cont_listing = self.env.container.files(parms={
|
|
'reverse': 'on', 'marker': first_item + '\x00',
|
|
'end_marker': last_item[:-1] + chr(ord(last_item[-1]) - 1)})
|
|
self.assert_status(200)
|
|
self.assertEqual(cont_list, cont_listing)
|
|
|
|
def testContainerFileSortingByReversedMarkersReversed(self):
|
|
cont_listing = self.env.container.files(parms={
|
|
'reverse': 'on', 'marker': 'B', 'end_marker': 'b1'})
|
|
self.assert_status(204)
|
|
self.assertEqual([], cont_listing)
|
|
|
|
def testContainerFileListSorting(self):
|
|
file_list = list(sorted(self.env.file_items))
|
|
cont_files = self.env.container.files()
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
# Lets try again but with reverse is specifically turned off
|
|
cont_files = self.env.container.files(parms={'reverse': 'off'})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
cont_files = self.env.container.files(parms={'reverse': 'false'})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
cont_files = self.env.container.files(parms={'reverse': 'no'})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
cont_files = self.env.container.files(parms={'reverse': ''})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
# Lets try again but with a incorrect reverse values
|
|
cont_files = self.env.container.files(parms={'reverse': 'foo'})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
cont_files = self.env.container.files(parms={'reverse': 'hai'})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
cont_files = self.env.container.files(parms={'reverse': 'o=[]::::>'})
|
|
self.assert_status(200)
|
|
self.assertEqual(file_list, cont_files)
|
|
|
|
|
|
class TestContainerPathsEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestContainerPathsEnv, cls).setUp()
|
|
cls.file_size = 8
|
|
|
|
cls.container = cls.account.container(Utils.create_name())
|
|
if not cls.container.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
cls.files = [
|
|
'/file1',
|
|
'/file A',
|
|
'/dir1/',
|
|
'/dir2/',
|
|
'/dir1/file2',
|
|
'/dir1/subdir1/',
|
|
'/dir1/subdir2/',
|
|
'/dir1/subdir1/file2',
|
|
'/dir1/subdir1/file3',
|
|
'/dir1/subdir1/file4',
|
|
'/dir1/subdir1/subsubdir1/',
|
|
'/dir1/subdir1/subsubdir1/file5',
|
|
'/dir1/subdir1/subsubdir1/file6',
|
|
'/dir1/subdir1/subsubdir1/file7',
|
|
'/dir1/subdir1/subsubdir1/file8',
|
|
'/dir1/subdir1/subsubdir2/',
|
|
'/dir1/subdir1/subsubdir2/file9',
|
|
'/dir1/subdir1/subsubdir2/file0',
|
|
'file1',
|
|
'dir1/',
|
|
'dir2/',
|
|
'dir1/file2',
|
|
'dir1/subdir1/',
|
|
'dir1/subdir2/',
|
|
'dir1/subdir1/file2',
|
|
'dir1/subdir1/file3',
|
|
'dir1/subdir1/file4',
|
|
'dir1/subdir1/subsubdir1/',
|
|
'dir1/subdir1/subsubdir1/file5',
|
|
'dir1/subdir1/subsubdir1/file6',
|
|
'dir1/subdir1/subsubdir1/file7',
|
|
'dir1/subdir1/subsubdir1/file8',
|
|
'dir1/subdir1/subsubdir2/',
|
|
'dir1/subdir1/subsubdir2/file9',
|
|
'dir1/subdir1/subsubdir2/file0',
|
|
'dir1/subdir with spaces/',
|
|
'dir1/subdir with spaces/file B',
|
|
'dir1/subdir+with{whatever/',
|
|
'dir1/subdir+with{whatever/file D',
|
|
]
|
|
|
|
stored_files = set()
|
|
for f in cls.files:
|
|
file_item = cls.container.file(f)
|
|
if f.endswith('/'):
|
|
file_item.write(hdrs={'Content-Type': 'application/directory'})
|
|
else:
|
|
file_item.write_random(cls.file_size,
|
|
hdrs={'Content-Type':
|
|
'application/directory'})
|
|
if (normalized_urls):
|
|
nfile = '/'.join(filter(None, f.split('/')))
|
|
if (f[-1] == '/'):
|
|
nfile += '/'
|
|
stored_files.add(nfile)
|
|
else:
|
|
stored_files.add(f)
|
|
cls.stored_files = sorted(stored_files)
|
|
|
|
|
|
class TestContainerPaths(Base):
|
|
env = TestContainerPathsEnv
|
|
|
|
def testTraverseContainer(self):
|
|
found_files = []
|
|
found_dirs = []
|
|
|
|
def recurse_path(path, count=0):
|
|
if count > 10:
|
|
raise ValueError('too deep recursion')
|
|
|
|
for file_item in self.env.container.files(parms={'path': path}):
|
|
self.assertTrue(file_item.startswith(path))
|
|
if file_item.endswith('/'):
|
|
recurse_path(file_item, count + 1)
|
|
found_dirs.append(file_item)
|
|
else:
|
|
found_files.append(file_item)
|
|
|
|
recurse_path('')
|
|
for file_item in self.env.stored_files:
|
|
if file_item.startswith('/'):
|
|
self.assertNotIn(file_item, found_dirs)
|
|
self.assertNotIn(file_item, found_files)
|
|
elif file_item.endswith('/'):
|
|
self.assertIn(file_item, found_dirs)
|
|
self.assertNotIn(file_item, found_files)
|
|
else:
|
|
self.assertIn(file_item, found_files)
|
|
self.assertNotIn(file_item, found_dirs)
|
|
|
|
found_files = []
|
|
found_dirs = []
|
|
recurse_path('/')
|
|
for file_item in self.env.stored_files:
|
|
if not file_item.startswith('/'):
|
|
self.assertNotIn(file_item, found_dirs)
|
|
self.assertNotIn(file_item, found_files)
|
|
elif file_item.endswith('/'):
|
|
self.assertIn(file_item, found_dirs)
|
|
self.assertNotIn(file_item, found_files)
|
|
else:
|
|
self.assertIn(file_item, found_files)
|
|
self.assertNotIn(file_item, found_dirs)
|
|
|
|
def testContainerListing(self):
|
|
for format_type in (None, 'json', 'xml'):
|
|
files = self.env.container.files(parms={'format': format_type})
|
|
|
|
if isinstance(files[0], dict):
|
|
files = [str(x['name']) for x in files]
|
|
|
|
self.assertEqual(files, self.env.stored_files)
|
|
|
|
for format_type in ('json', 'xml'):
|
|
for file_item in self.env.container.files(parms={'format':
|
|
format_type}):
|
|
self.assertGreaterEqual(int(file_item['bytes']), 0)
|
|
self.assertIn('last_modified', file_item)
|
|
if file_item['name'].endswith('/'):
|
|
self.assertEqual(file_item['content_type'],
|
|
'application/directory')
|
|
|
|
def testStructure(self):
|
|
def assert_listing(path, file_list):
|
|
files = self.env.container.files(parms={'path': path})
|
|
self.assertEqual(sorted(file_list, key=locale.strxfrm), files)
|
|
if not normalized_urls:
|
|
assert_listing('/', ['/dir1/', '/dir2/', '/file1', '/file A'])
|
|
assert_listing('/dir1',
|
|
['/dir1/file2', '/dir1/subdir1/', '/dir1/subdir2/'])
|
|
assert_listing('/dir1/',
|
|
['/dir1/file2', '/dir1/subdir1/', '/dir1/subdir2/'])
|
|
assert_listing('/dir1/subdir1',
|
|
['/dir1/subdir1/subsubdir2/', '/dir1/subdir1/file2',
|
|
'/dir1/subdir1/file3', '/dir1/subdir1/file4',
|
|
'/dir1/subdir1/subsubdir1/'])
|
|
assert_listing('/dir1/subdir2', [])
|
|
assert_listing('', ['file1', 'dir1/', 'dir2/'])
|
|
else:
|
|
assert_listing('', ['file1', 'dir1/', 'dir2/', 'file A'])
|
|
assert_listing('dir1', ['dir1/file2', 'dir1/subdir1/',
|
|
'dir1/subdir2/', 'dir1/subdir with spaces/',
|
|
'dir1/subdir+with{whatever/'])
|
|
assert_listing('dir1/subdir1',
|
|
['dir1/subdir1/file4', 'dir1/subdir1/subsubdir2/',
|
|
'dir1/subdir1/file2', 'dir1/subdir1/file3',
|
|
'dir1/subdir1/subsubdir1/'])
|
|
assert_listing('dir1/subdir1/subsubdir1',
|
|
['dir1/subdir1/subsubdir1/file7',
|
|
'dir1/subdir1/subsubdir1/file5',
|
|
'dir1/subdir1/subsubdir1/file8',
|
|
'dir1/subdir1/subsubdir1/file6'])
|
|
assert_listing('dir1/subdir1/subsubdir1/',
|
|
['dir1/subdir1/subsubdir1/file7',
|
|
'dir1/subdir1/subsubdir1/file5',
|
|
'dir1/subdir1/subsubdir1/file8',
|
|
'dir1/subdir1/subsubdir1/file6'])
|
|
assert_listing('dir1/subdir with spaces/',
|
|
['dir1/subdir with spaces/file B'])
|
|
|
|
|
|
class TestFileEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestFileEnv, cls).setUp()
|
|
if not tf.skip2:
|
|
# creating another account and connection
|
|
# for account to account copy tests
|
|
config2 = deepcopy(tf.config)
|
|
config2['account'] = tf.config['account2']
|
|
config2['username'] = tf.config['username2']
|
|
config2['password'] = tf.config['password2']
|
|
cls.conn2 = Connection(config2)
|
|
cls.conn2.authenticate()
|
|
|
|
cls.account2 = cls.conn2.get_account()
|
|
cls.account2.delete_containers()
|
|
|
|
cls.container = cls.account.container(Utils.create_name())
|
|
if not cls.container.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
cls.file_size = 128
|
|
|
|
# With keystoneauth we need the accounts to have had the project
|
|
# domain id persisted as sysmeta prior to testing ACLs. This may
|
|
# not be the case if, for example, the account was created using
|
|
# a request with reseller_admin role, when project domain id may
|
|
# not have been known. So we ensure that the project domain id is
|
|
# in sysmeta by making a POST to the accounts using an admin role.
|
|
cls.account.update_metadata()
|
|
if not tf.skip2:
|
|
cls.account2.update_metadata()
|
|
|
|
|
|
class TestFileDev(Base):
|
|
env = TestFileEnv
|
|
|
|
|
|
class TestFileDevUTF8(Base2, TestFileDev):
|
|
pass
|
|
|
|
|
|
class TestFile(Base):
|
|
env = TestFileEnv
|
|
|
|
def testGetResponseHeaders(self):
|
|
obj_data = b'test_body'
|
|
|
|
def do_test(put_hdrs, get_hdrs, expected_hdrs, unexpected_hdrs):
|
|
filename = Utils.create_name()
|
|
file_item = self.env.container.file(filename)
|
|
resp = file_item.write(
|
|
data=obj_data, hdrs=put_hdrs, return_resp=True)
|
|
|
|
# put then get an object
|
|
resp.read()
|
|
read_data = file_item.read(hdrs=get_hdrs)
|
|
self.assertEqual(obj_data, read_data) # sanity check
|
|
resp_headers = file_item.conn.response.getheaders()
|
|
|
|
# check the *list* of all header (name, value) pairs rather than
|
|
# constructing a dict in case of repeated names in the list
|
|
errors = []
|
|
for k, v in resp_headers:
|
|
if k.lower() in unexpected_hdrs:
|
|
errors.append('Found unexpected header %s: %s' % (k, v))
|
|
for k, v in expected_hdrs.items():
|
|
matches = [hdr for hdr in resp_headers if hdr[0].lower() == k]
|
|
if not matches:
|
|
errors.append('Missing expected header %s' % k)
|
|
for (got_k, got_v) in matches:
|
|
# The Connection: header is parsed by cluster's LB and may
|
|
# be returned in either original lowercase or camel-cased.
|
|
if k == 'connection':
|
|
got_v = got_v.lower()
|
|
if got_v != v:
|
|
errors.append('Expected %s but got %s for %s' %
|
|
(v, got_v, k))
|
|
if errors:
|
|
self.fail(
|
|
'Errors in response headers:\n %s' % '\n '.join(errors))
|
|
|
|
put_headers = {'X-Object-Meta-Fruit': 'Banana',
|
|
'X-Delete-After': '10000',
|
|
'Content-Type': 'application/test'}
|
|
expected_headers = {'content-length': str(len(obj_data)),
|
|
'x-object-meta-fruit': 'Banana',
|
|
'accept-ranges': 'bytes',
|
|
'content-type': 'application/test',
|
|
'etag': md5(
|
|
obj_data, usedforsecurity=False).hexdigest(),
|
|
'last-modified': mock.ANY,
|
|
'date': mock.ANY,
|
|
'x-delete-at': mock.ANY,
|
|
'x-trans-id': mock.ANY,
|
|
'x-openstack-request-id': mock.ANY}
|
|
if tf.cluster_info.get('etag_quoter', {}).get('enable_by_default'):
|
|
expected_headers['etag'] = '"%s"' % expected_headers['etag']
|
|
unexpected_headers = ['connection', 'x-delete-after']
|
|
do_test(put_headers, {}, expected_headers, unexpected_headers)
|
|
|
|
get_headers = {'Connection': 'keep-alive'}
|
|
expected_headers['connection'] = 'keep-alive'
|
|
unexpected_headers = ['x-delete-after']
|
|
do_test(put_headers, get_headers, expected_headers, unexpected_headers)
|
|
|
|
def testCopy(self):
|
|
# makes sure to test encoded characters
|
|
source_filename = 'dealde%2Fl04 011e%204c8df/flash.png'
|
|
file_item = self.env.container.file(source_filename)
|
|
|
|
metadata = {}
|
|
metadata[Utils.create_ascii_name()] = Utils.create_name()
|
|
put_headers = {'Content-Type': 'application/test',
|
|
'Content-Encoding': 'gzip',
|
|
'Content-Disposition': 'attachment; filename=myfile'}
|
|
file_item.metadata = metadata
|
|
data = file_item.write_random(hdrs=put_headers)
|
|
|
|
# the allowed headers are configurable in object server, so we cannot
|
|
# assert that content-encoding and content-disposition get *copied*
|
|
# unless they were successfully set on the original PUT, so populate
|
|
# expected_headers by making a HEAD on the original object
|
|
file_item.initialize()
|
|
self.assertEqual('application/test', file_item.content_type)
|
|
resp_headers = dict(file_item.conn.response.getheaders())
|
|
expected_headers = {}
|
|
for k, v in put_headers.items():
|
|
if k.lower() in resp_headers:
|
|
expected_headers[k] = v
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
# copy both from within and across containers
|
|
for cont in (self.env.container, dest_cont):
|
|
# copy both with and without initial slash
|
|
for prefix in ('', '/'):
|
|
dest_filename = Utils.create_name()
|
|
|
|
extra_hdrs = {'X-Object-Meta-Extra': 'fresh'}
|
|
self.assertTrue(file_item.copy(
|
|
'%s%s' % (prefix, cont), dest_filename, hdrs=extra_hdrs))
|
|
|
|
# verify container listing for copy
|
|
listing = cont.files(parms={'format': 'json'})
|
|
for obj in listing:
|
|
if obj['name'] == dest_filename:
|
|
break
|
|
else:
|
|
self.fail('Failed to find %s in listing' % dest_filename)
|
|
|
|
self.assertEqual(file_item.size, obj['bytes'])
|
|
self.assertEqual(normalize_etag(file_item.etag), obj['hash'])
|
|
self.assertEqual(file_item.content_type, obj['content_type'])
|
|
|
|
file_copy = cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_copy.read())
|
|
self.assertTrue(file_copy.initialize())
|
|
expected_metadata = dict(metadata)
|
|
# new metadata should be merged with existing
|
|
expected_metadata['extra'] = 'fresh'
|
|
self.assertDictEqual(expected_metadata, file_copy.metadata)
|
|
resp_headers = dict(file_copy.conn.response.getheaders())
|
|
for k, v in expected_headers.items():
|
|
self.assertIn(k.lower(), resp_headers)
|
|
self.assertEqual(v, resp_headers[k.lower()])
|
|
|
|
# repeat copy with updated content-type, content-encoding and
|
|
# content-disposition, which should get updated
|
|
extra_hdrs = {
|
|
'X-Object-Meta-Extra': 'fresher',
|
|
'Content-Type': 'application/test-changed',
|
|
'Content-Encoding': 'not_gzip',
|
|
'Content-Disposition': 'attachment; filename=notmyfile'}
|
|
self.assertTrue(file_item.copy(
|
|
'%s%s' % (prefix, cont), dest_filename, hdrs=extra_hdrs))
|
|
|
|
self.assertIn(dest_filename, cont.files())
|
|
|
|
file_copy = cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_copy.read())
|
|
self.assertTrue(file_copy.initialize())
|
|
expected_metadata['extra'] = 'fresher'
|
|
self.assertDictEqual(expected_metadata, file_copy.metadata)
|
|
resp_headers = dict(file_copy.conn.response.getheaders())
|
|
# if k is in expected_headers then we can assert its new value
|
|
for k, v in expected_headers.items():
|
|
v = extra_hdrs.get(k, v)
|
|
self.assertIn(k.lower(), resp_headers)
|
|
self.assertEqual(v, resp_headers[k.lower()])
|
|
|
|
# verify container listing for copy
|
|
listing = cont.files(parms={'format': 'json'})
|
|
for obj in listing:
|
|
if obj['name'] == dest_filename:
|
|
break
|
|
else:
|
|
self.fail('Failed to find %s in listing' % dest_filename)
|
|
|
|
self.assertEqual(file_item.size, obj['bytes'])
|
|
self.assertEqual(normalize_etag(file_item.etag), obj['hash'])
|
|
self.assertEqual(
|
|
'application/test-changed', obj['content_type'])
|
|
|
|
# repeat copy with X-Fresh-Metadata header - existing user
|
|
# metadata should not be copied, new completely replaces it.
|
|
extra_hdrs = {'Content-Type': 'application/test-updated',
|
|
'X-Object-Meta-Extra': 'fresher',
|
|
'X-Fresh-Metadata': 'true'}
|
|
self.assertTrue(file_item.copy(
|
|
'%s%s' % (prefix, cont), dest_filename, hdrs=extra_hdrs))
|
|
|
|
self.assertIn(dest_filename, cont.files())
|
|
|
|
file_copy = cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_copy.read())
|
|
self.assertTrue(file_copy.initialize())
|
|
self.assertEqual('application/test-updated',
|
|
file_copy.content_type)
|
|
expected_metadata = {'extra': 'fresher'}
|
|
self.assertDictEqual(expected_metadata, file_copy.metadata)
|
|
resp_headers = dict(file_copy.conn.response.getheaders())
|
|
for k in ('Content-Disposition', 'Content-Encoding'):
|
|
self.assertNotIn(k.lower(), resp_headers)
|
|
|
|
# verify container listing for copy
|
|
listing = cont.files(parms={'format': 'json'})
|
|
for obj in listing:
|
|
if obj['name'] == dest_filename:
|
|
break
|
|
else:
|
|
self.fail('Failed to find %s in listing' % dest_filename)
|
|
|
|
self.assertEqual(file_item.size, obj['bytes'])
|
|
self.assertEqual(normalize_etag(file_item.etag), obj['hash'])
|
|
self.assertEqual(
|
|
'application/test-updated', obj['content_type'])
|
|
|
|
def testCopyRange(self):
|
|
# makes sure to test encoded characters
|
|
source_filename = 'dealde%2Fl04 011e%204c8df/flash.png'
|
|
file_item = self.env.container.file(source_filename)
|
|
|
|
metadata = {Utils.create_ascii_name(): Utils.create_name()}
|
|
|
|
data = file_item.write_random(1024)
|
|
file_item.sync_metadata(metadata)
|
|
file_item.initialize()
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
expected_body = data[100:201]
|
|
expected_etag = md5(expected_body, usedforsecurity=False)
|
|
# copy both from within and across containers
|
|
for cont in (self.env.container, dest_cont):
|
|
# copy both with and without initial slash
|
|
for prefix in ('', '/'):
|
|
dest_filename = Utils.create_name()
|
|
|
|
file_item.copy('%s%s' % (prefix, cont), dest_filename,
|
|
hdrs={'Range': 'bytes=100-200'})
|
|
self.assertEqual(201, file_item.conn.response.status)
|
|
|
|
# verify container listing for copy
|
|
listing = cont.files(parms={'format': 'json'})
|
|
for obj in listing:
|
|
if obj['name'] == dest_filename:
|
|
break
|
|
else:
|
|
self.fail('Failed to find %s in listing' % dest_filename)
|
|
|
|
self.assertEqual(101, obj['bytes'])
|
|
self.assertEqual(expected_etag.hexdigest(), obj['hash'])
|
|
self.assertEqual(file_item.content_type, obj['content_type'])
|
|
|
|
# verify copy object
|
|
copy_file_item = cont.file(dest_filename)
|
|
self.assertEqual(expected_body, copy_file_item.read())
|
|
self.assertTrue(copy_file_item.initialize())
|
|
self.assertEqual(metadata, copy_file_item.metadata)
|
|
|
|
def testCopyAccount(self):
|
|
# makes sure to test encoded characters
|
|
source_filename = 'dealde%2Fl04 011e%204c8df/flash.png'
|
|
file_item = self.env.container.file(source_filename)
|
|
|
|
metadata = {Utils.create_ascii_name(): Utils.create_name()}
|
|
|
|
data = file_item.write_random()
|
|
file_item.sync_metadata(metadata)
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
acct = self.env.conn.account_name
|
|
# copy both from within and across containers
|
|
for cont in (self.env.container, dest_cont):
|
|
# copy both with and without initial slash
|
|
for prefix in ('', '/'):
|
|
dest_filename = Utils.create_name()
|
|
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.copy_account(acct,
|
|
'%s%s' % (prefix, cont),
|
|
dest_filename)
|
|
|
|
self.assertIn(dest_filename, cont.files())
|
|
|
|
file_item = cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_item.read())
|
|
self.assertTrue(file_item.initialize())
|
|
self.assertEqual(metadata, file_item.metadata)
|
|
|
|
if not tf.skip2:
|
|
dest_cont = self.env.account2.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create(hdrs={
|
|
'X-Container-Write': self.env.conn.user_acl
|
|
}))
|
|
|
|
acct = self.env.conn2.account_name
|
|
# copy both with and without initial slash
|
|
for prefix in ('', '/'):
|
|
dest_filename = Utils.create_name()
|
|
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.copy_account(acct,
|
|
'%s%s' % (prefix, dest_cont),
|
|
dest_filename)
|
|
|
|
self.assertIn(dest_filename, dest_cont.files())
|
|
|
|
file_item = dest_cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_item.read())
|
|
self.assertTrue(file_item.initialize())
|
|
self.assertEqual(metadata, file_item.metadata)
|
|
|
|
def testCopy404s(self):
|
|
source_filename = Utils.create_name()
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.write_random()
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
for prefix in ('', '/'):
|
|
# invalid source container
|
|
source_cont = self.env.account.container(Utils.create_name())
|
|
file_item = source_cont.file(source_filename)
|
|
self.assertRaises(ResponseError, file_item.copy,
|
|
'%s%s' % (prefix, self.env.container),
|
|
Utils.create_name())
|
|
self.assert_status(404)
|
|
|
|
self.assertRaises(ResponseError, file_item.copy,
|
|
'%s%s' % (prefix, dest_cont),
|
|
Utils.create_name())
|
|
self.assert_status(404)
|
|
|
|
# invalid source object
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.copy,
|
|
'%s%s' % (prefix, self.env.container),
|
|
Utils.create_name())
|
|
self.assert_status(404)
|
|
|
|
self.assertRaises(ResponseError, file_item.copy,
|
|
'%s%s' % (prefix, dest_cont),
|
|
Utils.create_name())
|
|
self.assert_status(404)
|
|
|
|
# invalid destination container
|
|
file_item = self.env.container.file(source_filename)
|
|
self.assertRaises(ResponseError, file_item.copy,
|
|
'%s%s' % (prefix, Utils.create_name()),
|
|
Utils.create_name())
|
|
|
|
def testCopyAccount404s(self):
|
|
if tf.skip2:
|
|
raise SkipTest('Account2 not set')
|
|
acct = self.env.conn.account_name
|
|
acct2 = self.env.conn2.account_name
|
|
source_filename = Utils.create_name()
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.write_random()
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create(hdrs={
|
|
'X-Container-Read': self.env.conn2.user_acl
|
|
}))
|
|
dest_cont2 = self.env.account2.container(Utils.create_name())
|
|
self.assertTrue(dest_cont2.create(hdrs={
|
|
'X-Container-Write': self.env.conn.user_acl,
|
|
'X-Container-Read': self.env.conn.user_acl
|
|
}))
|
|
|
|
for acct, cont in ((acct, dest_cont), (acct2, dest_cont2)):
|
|
for prefix in ('', '/'):
|
|
# invalid source container
|
|
source_cont = self.env.account.container(Utils.create_name())
|
|
file_item = source_cont.file(source_filename)
|
|
self.assertRaises(ResponseError, file_item.copy_account,
|
|
acct,
|
|
'%s%s' % (prefix, self.env.container),
|
|
Utils.create_name())
|
|
# there is no such source container but user has
|
|
# permissions to do a GET (done internally via COPY) for
|
|
# objects in his own account.
|
|
self.assert_status(404)
|
|
|
|
self.assertRaises(ResponseError, file_item.copy_account,
|
|
acct,
|
|
'%s%s' % (prefix, cont),
|
|
Utils.create_name())
|
|
self.assert_status(404)
|
|
|
|
# invalid source object
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.copy_account,
|
|
acct,
|
|
'%s%s' % (prefix, self.env.container),
|
|
Utils.create_name())
|
|
# there is no such source container but user has
|
|
# permissions to do a GET (done internally via COPY) for
|
|
# objects in his own account.
|
|
self.assert_status(404)
|
|
|
|
self.assertRaises(ResponseError, file_item.copy_account,
|
|
acct,
|
|
'%s%s' % (prefix, cont),
|
|
Utils.create_name())
|
|
self.assert_status(404)
|
|
|
|
# invalid destination container
|
|
file_item = self.env.container.file(source_filename)
|
|
self.assertRaises(ResponseError, file_item.copy_account,
|
|
acct,
|
|
'%s%s' % (prefix, Utils.create_name()),
|
|
Utils.create_name())
|
|
if acct == acct2:
|
|
# there is no such destination container
|
|
# and foreign user can have no permission to write there
|
|
self.assert_status(403)
|
|
else:
|
|
self.assert_status(404)
|
|
|
|
def testCopyNoDestinationHeader(self):
|
|
source_filename = Utils.create_name()
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.write_random()
|
|
|
|
file_item = self.env.container.file(source_filename)
|
|
self.assertRaises(ResponseError, file_item.copy, Utils.create_name(),
|
|
Utils.create_name(),
|
|
cfg={'no_destination': True})
|
|
self.assert_status(412)
|
|
|
|
def testCopyDestinationSlashProblems(self):
|
|
source_filename = Utils.create_name()
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.write_random()
|
|
|
|
# no slash
|
|
self.assertRaises(ResponseError, file_item.copy, Utils.create_name(),
|
|
Utils.create_name(),
|
|
cfg={'destination': Utils.create_name()})
|
|
self.assert_status(412)
|
|
|
|
# too many slashes
|
|
self.assertRaises(ResponseError, file_item.copy, Utils.create_name(),
|
|
Utils.create_name(),
|
|
cfg={'destination': '//%s' % Utils.create_name()})
|
|
self.assert_status(412)
|
|
|
|
def testCopyFromHeader(self):
|
|
source_filename = Utils.create_name()
|
|
file_item = self.env.container.file(source_filename)
|
|
|
|
metadata = {}
|
|
for i in range(1):
|
|
metadata[Utils.create_ascii_name()] = Utils.create_name()
|
|
file_item.metadata = metadata
|
|
|
|
data = file_item.write_random()
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
# copy both from within and across containers
|
|
for cont in (self.env.container, dest_cont):
|
|
# copy both with and without initial slash
|
|
for prefix in ('', '/'):
|
|
dest_filename = Utils.create_name()
|
|
|
|
file_item = cont.file(dest_filename)
|
|
file_item.write(hdrs={'X-Copy-From': '%s%s/%s' % (
|
|
prefix, self.env.container.name, source_filename)})
|
|
|
|
self.assertIn(dest_filename, cont.files())
|
|
|
|
file_item = cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_item.read())
|
|
self.assertTrue(file_item.initialize())
|
|
self.assertEqual(metadata, file_item.metadata)
|
|
|
|
def testCopyFromAccountHeader(self):
|
|
if tf.skip2:
|
|
raise SkipTest('Account2 not set')
|
|
acct = self.env.conn.account_name
|
|
src_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(src_cont.create(hdrs={
|
|
'X-Container-Read': self.env.conn2.user_acl
|
|
}))
|
|
source_filename = Utils.create_name()
|
|
file_item = src_cont.file(source_filename)
|
|
|
|
metadata = {}
|
|
for i in range(1):
|
|
metadata[Utils.create_ascii_name()] = Utils.create_name()
|
|
file_item.metadata = metadata
|
|
|
|
data = file_item.write_random()
|
|
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
dest_cont2 = self.env.account2.container(Utils.create_name())
|
|
self.assertTrue(dest_cont2.create(hdrs={
|
|
'X-Container-Write': self.env.conn.user_acl
|
|
}))
|
|
|
|
for cont in (src_cont, dest_cont, dest_cont2):
|
|
# copy both with and without initial slash
|
|
for prefix in ('', '/'):
|
|
dest_filename = Utils.create_name()
|
|
|
|
file_item = cont.file(dest_filename)
|
|
file_item.write(hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' % (
|
|
prefix,
|
|
src_cont.name,
|
|
source_filename)})
|
|
|
|
self.assertIn(dest_filename, cont.files())
|
|
|
|
file_item = cont.file(dest_filename)
|
|
|
|
self.assertEqual(data, file_item.read())
|
|
self.assertTrue(file_item.initialize())
|
|
self.assertEqual(metadata, file_item.metadata)
|
|
|
|
def testCopyFromHeader404s(self):
|
|
source_filename = Utils.create_name()
|
|
file_item = self.env.container.file(source_filename)
|
|
file_item.write_random()
|
|
|
|
for prefix in ('', '/'):
|
|
# invalid source container
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
copy_from = ('%s%s/%s'
|
|
% (prefix, Utils.create_name(), source_filename))
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From': copy_from})
|
|
self.assert_status(404)
|
|
|
|
# invalid source object
|
|
copy_from = ('%s%s/%s'
|
|
% (prefix, self.env.container.name,
|
|
Utils.create_name()))
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From': copy_from})
|
|
self.assert_status(404)
|
|
|
|
# invalid destination container
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
file_item = dest_cont.file(Utils.create_name())
|
|
copy_from = ('%s%s/%s'
|
|
% (prefix, self.env.container.name, source_filename))
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From': copy_from})
|
|
self.assert_status(404)
|
|
|
|
def testCopyFromAccountHeader404s(self):
|
|
if tf.skip2:
|
|
raise SkipTest('Account2 not set')
|
|
acct = self.env.conn2.account_name
|
|
src_cont = self.env.account2.container(Utils.create_name())
|
|
self.assertTrue(src_cont.create(hdrs={
|
|
'X-Container-Read': self.env.conn.user_acl
|
|
}))
|
|
source_filename = Utils.create_name()
|
|
file_item = src_cont.file(source_filename)
|
|
file_item.write_random()
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
for prefix in ('', '/'):
|
|
# invalid source container
|
|
file_item = dest_cont.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' %
|
|
(prefix,
|
|
Utils.create_name(),
|
|
source_filename)})
|
|
self.assert_status(403)
|
|
|
|
# invalid source object
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' %
|
|
(prefix,
|
|
src_cont,
|
|
Utils.create_name())})
|
|
self.assert_status(404)
|
|
|
|
# invalid destination container
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
file_item = dest_cont.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' %
|
|
(prefix,
|
|
src_cont,
|
|
source_filename)})
|
|
self.assert_status(404)
|
|
|
|
def testCopyFromAccountHeader403s(self):
|
|
if tf.skip2:
|
|
raise SkipTest('Account2 not set')
|
|
acct = self.env.conn2.account_name
|
|
src_cont = self.env.account2.container(Utils.create_name())
|
|
self.assertTrue(src_cont.create()) # Primary user has no access
|
|
source_filename = Utils.create_name()
|
|
file_item = src_cont.file(source_filename)
|
|
file_item.write_random()
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(dest_cont.create())
|
|
|
|
for prefix in ('', '/'):
|
|
# invalid source container
|
|
file_item = dest_cont.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' %
|
|
(prefix,
|
|
Utils.create_name(),
|
|
source_filename)})
|
|
self.assert_status(403)
|
|
|
|
# invalid source object
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' %
|
|
(prefix,
|
|
src_cont,
|
|
Utils.create_name())})
|
|
self.assert_status(403)
|
|
|
|
# invalid destination container
|
|
dest_cont = self.env.account.container(Utils.create_name())
|
|
file_item = dest_cont.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write,
|
|
hdrs={'X-Copy-From-Account': acct,
|
|
'X-Copy-From': '%s%s/%s' %
|
|
(prefix,
|
|
src_cont,
|
|
source_filename)})
|
|
self.assert_status(403)
|
|
|
|
def testNameLimit(self):
|
|
limit = load_constraint('max_object_name_length')
|
|
|
|
for lim in (1, 10, limit // 2, limit - 1, limit, limit + 1, limit * 2):
|
|
file_item = self.env.container.file('a' * lim)
|
|
|
|
if lim <= limit:
|
|
self.assertTrue(file_item.write())
|
|
self.assert_status(201)
|
|
else:
|
|
self.assertRaises(ResponseError, file_item.write)
|
|
self.assert_status(400)
|
|
|
|
def testQuestionMarkInName(self):
|
|
if Utils.create_name == Utils.create_ascii_name:
|
|
file_name = list(Utils.create_name())
|
|
file_name[random.randint(2, len(file_name) - 2)] = '?'
|
|
file_name = "".join(file_name)
|
|
else:
|
|
file_name = Utils.create_name(6) + '?' + Utils.create_name(6)
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
self.assertTrue(file_item.write(cfg={'no_path_quote': True}))
|
|
self.assertNotIn(file_name, self.env.container.files())
|
|
self.assertIn(file_name.split('?')[0], self.env.container.files())
|
|
|
|
def testDeleteThen404s(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertTrue(file_item.write_random())
|
|
self.assert_status(201)
|
|
|
|
self.assertTrue(file_item.delete())
|
|
self.assert_status(204)
|
|
|
|
file_item.metadata = {Utils.create_ascii_name(): Utils.create_name()}
|
|
|
|
for method in (file_item.info,
|
|
file_item.read,
|
|
file_item.sync_metadata,
|
|
file_item.delete):
|
|
self.assertRaises(ResponseError, method)
|
|
self.assert_status(404)
|
|
|
|
def testBlankMetadataName(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.metadata = {'': Utils.create_name()}
|
|
self.assertRaises(ResponseError, file_item.write_random)
|
|
self.assert_status(400)
|
|
|
|
def testMetadataNumberLimit(self):
|
|
number_limit = load_constraint('max_meta_count')
|
|
size_limit = load_constraint('max_meta_overall_size')
|
|
|
|
for i in (number_limit - 10, number_limit - 1, number_limit,
|
|
number_limit + 1, number_limit + 10, number_limit + 100):
|
|
|
|
j = size_limit // (i * 2)
|
|
|
|
metadata = {}
|
|
while len(metadata.keys()) < i:
|
|
key = Utils.create_ascii_name()
|
|
val = Utils.create_name()
|
|
|
|
if len(key) > j:
|
|
key = key[:j]
|
|
# NB: we'll likely write object metadata that's *not* UTF-8
|
|
if six.PY2:
|
|
val = val[:j]
|
|
else:
|
|
val = val.encode('utf8')[:j].decode(
|
|
'utf8', 'surrogateescape')
|
|
|
|
metadata[key] = val
|
|
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.metadata = metadata
|
|
|
|
if i <= number_limit:
|
|
self.assertTrue(file_item.write())
|
|
self.assert_status(201)
|
|
self.assertTrue(file_item.sync_metadata())
|
|
self.assert_status(202)
|
|
else:
|
|
self.assertRaises(ResponseError, file_item.write)
|
|
self.assert_status(400)
|
|
file_item.metadata = {}
|
|
self.assertTrue(file_item.write())
|
|
self.assert_status(201)
|
|
file_item.metadata = metadata
|
|
self.assertRaises(ResponseError, file_item.sync_metadata)
|
|
self.assert_status(400)
|
|
|
|
def testContentTypeGuessing(self):
|
|
file_types = {'wav': 'audio/x-wav', 'txt': 'text/plain',
|
|
'zip': 'application/zip'}
|
|
|
|
container = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(container.create())
|
|
|
|
for i in file_types.keys():
|
|
file_item = container.file(Utils.create_name() + '.' + i)
|
|
file_item.write(b'', cfg={'no_content_type': True})
|
|
|
|
file_types_read = {}
|
|
for i in container.files(parms={'format': 'json'}):
|
|
file_types_read[i['name'].split('.')[1]] = i['content_type']
|
|
|
|
self.assertEqual(file_types, file_types_read)
|
|
|
|
def testRangedGets(self):
|
|
# We set the file_length to a strange multiple here. This is to check
|
|
# that ranges still work in the EC case when the requested range
|
|
# spans EC segment boundaries. The 1 MiB base value is chosen because
|
|
# that's a common EC segment size. The 1.33 multiple is to ensure we
|
|
# aren't aligned on segment boundaries
|
|
file_length = int(1048576 * 1.33)
|
|
range_size = file_length // 10
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
data = file_item.write_random(file_length)
|
|
|
|
for i in range(0, file_length, range_size):
|
|
range_string = 'bytes=%d-%d' % (i, i + range_size - 1)
|
|
hdrs = {'Range': range_string}
|
|
self.assertEqual(
|
|
data[i: i + range_size], file_item.read(hdrs=hdrs),
|
|
range_string)
|
|
|
|
range_string = 'bytes=-%d' % (i)
|
|
hdrs = {'Range': range_string}
|
|
if i == 0:
|
|
# RFC 2616 14.35.1
|
|
# "If a syntactically valid byte-range-set includes ... at
|
|
# least one suffix-byte-range-spec with a NON-ZERO
|
|
# suffix-length, then the byte-range-set is satisfiable.
|
|
# Otherwise, the byte-range-set is unsatisfiable.
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(416)
|
|
self.assert_header('content-range', 'bytes */%d' % file_length)
|
|
else:
|
|
self.assertEqual(file_item.read(hdrs=hdrs), data[-i:])
|
|
self.assert_header('content-range', 'bytes %d-%d/%d' % (
|
|
file_length - i, file_length - 1, file_length))
|
|
self.assert_etag(file_item.md5)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
|
|
range_string = 'bytes=%d-' % (i)
|
|
hdrs = {'Range': range_string}
|
|
self.assertEqual(
|
|
file_item.read(hdrs=hdrs), data[i - file_length:],
|
|
range_string)
|
|
|
|
range_string = 'bytes=%d-%d' % (file_length + 1000, file_length + 2000)
|
|
hdrs = {'Range': range_string}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(416)
|
|
self.assert_header('content-range', 'bytes */%d' % file_length)
|
|
self.assert_etag(file_item.md5)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
|
|
range_string = 'bytes=%d-%d' % (file_length - 1000, file_length + 2000)
|
|
hdrs = {'Range': range_string}
|
|
self.assertEqual(file_item.read(hdrs=hdrs), data[-1000:], range_string)
|
|
|
|
hdrs = {'Range': '0-4'}
|
|
self.assertEqual(file_item.read(hdrs=hdrs), data, '0-4')
|
|
|
|
# RFC 2616 14.35.1
|
|
# "If the entity is shorter than the specified suffix-length, the
|
|
# entire entity-body is used."
|
|
range_string = 'bytes=-%d' % (file_length + 10)
|
|
hdrs = {'Range': range_string}
|
|
self.assertEqual(file_item.read(hdrs=hdrs), data, range_string)
|
|
|
|
def testMultiRangeGets(self):
|
|
file_length = 10000
|
|
range_size = file_length // 10
|
|
subrange_size = range_size // 10
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
data = file_item.write_random(
|
|
file_length, hdrs={"Content-Type":
|
|
"lovecraft/rugose; squamous=true"})
|
|
|
|
for i in range(0, file_length, range_size):
|
|
range_string = 'bytes=%d-%d,%d-%d,%d-%d' % (
|
|
i, i + subrange_size - 1,
|
|
i + 2 * subrange_size, i + 3 * subrange_size - 1,
|
|
i + 4 * subrange_size, i + 5 * subrange_size - 1)
|
|
hdrs = {'Range': range_string}
|
|
|
|
fetched = file_item.read(hdrs=hdrs)
|
|
self.assert_status(206)
|
|
content_type = file_item.content_type
|
|
self.assertTrue(content_type.startswith("multipart/byteranges"))
|
|
self.assertIsNone(file_item.content_range)
|
|
|
|
# email.parser.FeedParser wants a message with headers on the
|
|
# front, then two CRLFs, and then a body (like emails have but
|
|
# HTTP response bodies don't). We fake it out by constructing a
|
|
# one-header preamble containing just the Content-Type, then
|
|
# feeding in the response body.
|
|
parser = FeedParser()
|
|
parser.feed(b"Content-Type: %s\r\n\r\n" % content_type.encode())
|
|
parser.feed(fetched)
|
|
root_message = parser.close()
|
|
self.assertTrue(root_message.is_multipart())
|
|
|
|
byteranges = root_message.get_payload()
|
|
self.assertEqual(len(byteranges), 3)
|
|
|
|
self.assertEqual(byteranges[0]['Content-Type'],
|
|
"lovecraft/rugose; squamous=true")
|
|
self.assertEqual(
|
|
byteranges[0]['Content-Range'],
|
|
"bytes %d-%d/%d" % (i, i + subrange_size - 1, file_length))
|
|
self.assertEqual(
|
|
byteranges[0].get_payload(decode=True),
|
|
data[i:(i + subrange_size)])
|
|
|
|
self.assertEqual(byteranges[1]['Content-Type'],
|
|
"lovecraft/rugose; squamous=true")
|
|
self.assertEqual(
|
|
byteranges[1]['Content-Range'],
|
|
"bytes %d-%d/%d" % (i + 2 * subrange_size,
|
|
i + 3 * subrange_size - 1, file_length))
|
|
self.assertEqual(
|
|
byteranges[1].get_payload(decode=True),
|
|
data[(i + 2 * subrange_size):(i + 3 * subrange_size)])
|
|
|
|
self.assertEqual(byteranges[2]['Content-Type'],
|
|
"lovecraft/rugose; squamous=true")
|
|
self.assertEqual(
|
|
byteranges[2]['Content-Range'],
|
|
"bytes %d-%d/%d" % (i + 4 * subrange_size,
|
|
i + 5 * subrange_size - 1, file_length))
|
|
self.assertEqual(
|
|
byteranges[2].get_payload(decode=True),
|
|
data[(i + 4 * subrange_size):(i + 5 * subrange_size)])
|
|
|
|
# The first two ranges are satisfiable but the third is not; the
|
|
# result is a multipart/byteranges response containing only the two
|
|
# satisfiable byteranges.
|
|
range_string = 'bytes=%d-%d,%d-%d,%d-%d' % (
|
|
0, subrange_size - 1,
|
|
2 * subrange_size, 3 * subrange_size - 1,
|
|
file_length, file_length + subrange_size - 1)
|
|
hdrs = {'Range': range_string}
|
|
fetched = file_item.read(hdrs=hdrs)
|
|
self.assert_status(206)
|
|
content_type = file_item.content_type
|
|
self.assertTrue(content_type.startswith("multipart/byteranges"))
|
|
self.assertIsNone(file_item.content_range)
|
|
|
|
parser = FeedParser()
|
|
parser.feed(b"Content-Type: %s\r\n\r\n" % content_type.encode())
|
|
parser.feed(fetched)
|
|
root_message = parser.close()
|
|
|
|
self.assertTrue(root_message.is_multipart())
|
|
byteranges = root_message.get_payload()
|
|
self.assertEqual(len(byteranges), 2)
|
|
|
|
self.assertEqual(byteranges[0]['Content-Type'],
|
|
"lovecraft/rugose; squamous=true")
|
|
self.assertEqual(
|
|
byteranges[0]['Content-Range'],
|
|
"bytes %d-%d/%d" % (0, subrange_size - 1, file_length))
|
|
self.assertEqual(byteranges[0].get_payload(decode=True),
|
|
data[:subrange_size])
|
|
|
|
self.assertEqual(byteranges[1]['Content-Type'],
|
|
"lovecraft/rugose; squamous=true")
|
|
self.assertEqual(
|
|
byteranges[1]['Content-Range'],
|
|
"bytes %d-%d/%d" % (2 * subrange_size, 3 * subrange_size - 1,
|
|
file_length))
|
|
self.assertEqual(
|
|
byteranges[1].get_payload(decode=True),
|
|
data[(2 * subrange_size):(3 * subrange_size)])
|
|
|
|
# The first range is satisfiable but the second is not; the
|
|
# result is either a multipart/byteranges response containing one
|
|
# byterange or a normal, non-MIME 206 response.
|
|
range_string = 'bytes=%d-%d,%d-%d' % (
|
|
0, subrange_size - 1,
|
|
file_length, file_length + subrange_size - 1)
|
|
hdrs = {'Range': range_string}
|
|
fetched = file_item.read(hdrs=hdrs)
|
|
self.assert_status(206)
|
|
content_type = file_item.content_type
|
|
if content_type.startswith("multipart/byteranges"):
|
|
self.assertIsNone(file_item.content_range)
|
|
parser = FeedParser()
|
|
parser.feed(b"Content-Type: %s\r\n\r\n" % content_type.encode())
|
|
parser.feed(fetched)
|
|
root_message = parser.close()
|
|
|
|
self.assertTrue(root_message.is_multipart())
|
|
byteranges = root_message.get_payload()
|
|
self.assertEqual(len(byteranges), 1)
|
|
|
|
self.assertEqual(byteranges[0]['Content-Type'],
|
|
"lovecraft/rugose; squamous=true")
|
|
self.assertEqual(
|
|
byteranges[0]['Content-Range'],
|
|
"bytes %d-%d/%d" % (0, subrange_size - 1, file_length))
|
|
self.assertEqual(byteranges[0].get_payload(decode=True),
|
|
data[:subrange_size])
|
|
else:
|
|
self.assertEqual(
|
|
file_item.content_range,
|
|
"bytes %d-%d/%d" % (0, subrange_size - 1, file_length))
|
|
self.assertEqual(content_type, "lovecraft/rugose; squamous=true")
|
|
self.assertEqual(fetched, data[:subrange_size])
|
|
|
|
# No byterange is satisfiable, so we get a 416 response.
|
|
range_string = 'bytes=%d-%d,%d-%d' % (
|
|
file_length, file_length + 2,
|
|
file_length + 100, file_length + 102)
|
|
hdrs = {'Range': range_string}
|
|
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(416)
|
|
self.assert_header('content-range', 'bytes */%d' % file_length)
|
|
|
|
def testRangedGetsWithLWSinHeader(self):
|
|
file_length = 10000
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
data = file_item.write_random(file_length)
|
|
|
|
for r in ('BYTES=0-999', 'bytes = 0-999', 'BYTES = 0 - 999',
|
|
'bytes = 0 - 999', 'bytes=0 - 999', 'bytes=0-999 '):
|
|
|
|
self.assertEqual(file_item.read(hdrs={'Range': r}), data[0:1000])
|
|
|
|
def testFileSizeLimit(self):
|
|
limit = load_constraint('max_file_size')
|
|
tsecs = 3
|
|
|
|
def timeout(seconds, method, *args, **kwargs):
|
|
try:
|
|
with eventlet.Timeout(seconds):
|
|
method(*args, **kwargs)
|
|
except eventlet.Timeout:
|
|
return True
|
|
else:
|
|
return False
|
|
|
|
# This loop will result in fallocate calls for 4x the limit
|
|
# (minus 111 bytes). With fallocate turned on in the object servers,
|
|
# this may fail if you don't have 4x the limit available on your
|
|
# data drives.
|
|
|
|
# Note that this test does not actually send any data to the system.
|
|
# All it does is ensure that a response (success or failure) comes
|
|
# back within 3 seconds. For the successful tests (size smaller
|
|
# than limit), the cluster will log a 499.
|
|
|
|
for i in (limit - 100, limit - 10, limit - 1, limit, limit + 1,
|
|
limit + 10, limit + 100):
|
|
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
|
|
if i <= limit:
|
|
self.assertTrue(timeout(tsecs, file_item.write,
|
|
cfg={'set_content_length': i}))
|
|
else:
|
|
self.assertRaises(ResponseError, timeout, tsecs,
|
|
file_item.write,
|
|
cfg={'set_content_length': i})
|
|
|
|
def testNoContentLengthForPut(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.write, b'testing',
|
|
cfg={'no_content_length': True})
|
|
self.assert_status(411)
|
|
|
|
def testDelete(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.write_random(self.env.file_size)
|
|
|
|
self.assertIn(file_item.name, self.env.container.files())
|
|
self.assertTrue(file_item.delete())
|
|
self.assertNotIn(file_item.name, self.env.container.files())
|
|
|
|
def testBadHeaders(self):
|
|
file_length = 100
|
|
|
|
# no content type on puts should be ok
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.write_random(file_length, cfg={'no_content_type': True})
|
|
self.assert_status(201)
|
|
|
|
# content length x
|
|
self.assertRaises(ResponseError, file_item.write_random, file_length,
|
|
hdrs={'Content-Length': 'X'},
|
|
cfg={'no_content_length': True})
|
|
self.assert_status(400)
|
|
|
|
# no content-length
|
|
self.assertRaises(ResponseError, file_item.write_random, file_length,
|
|
cfg={'no_content_length': True})
|
|
self.assert_status(411)
|
|
|
|
self.assertRaises(ResponseError, file_item.write_random, file_length,
|
|
hdrs={'transfer-encoding': 'gzip,chunked'},
|
|
cfg={'no_content_length': True})
|
|
self.assert_status(501)
|
|
|
|
# bad request types
|
|
# for req in ('LICK', 'GETorHEAD_base', 'container_info',
|
|
# 'best_response'):
|
|
for req in ('LICK', 'GETorHEAD_base'):
|
|
self.env.account.conn.make_request(req)
|
|
self.assert_status(405)
|
|
|
|
# bad range headers
|
|
self.assertEqual(
|
|
len(file_item.read(hdrs={'Range': 'parsecs=8-12'})),
|
|
file_length)
|
|
self.assert_status(200)
|
|
|
|
def testMetadataLengthLimits(self):
|
|
key_limit = load_constraint('max_meta_name_length')
|
|
value_limit = load_constraint('max_meta_value_length')
|
|
lengths = [[key_limit, value_limit], [key_limit, value_limit + 1],
|
|
[key_limit + 1, value_limit], [key_limit, 0],
|
|
[key_limit, value_limit * 10],
|
|
[key_limit * 10, value_limit]]
|
|
|
|
for l in lengths:
|
|
metadata = {'a' * l[0]: 'b' * l[1]}
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.metadata = metadata
|
|
|
|
if l[0] <= key_limit and l[1] <= value_limit:
|
|
self.assertTrue(file_item.write())
|
|
self.assert_status(201)
|
|
self.assertTrue(file_item.sync_metadata())
|
|
else:
|
|
self.assertRaises(ResponseError, file_item.write)
|
|
self.assert_status(400)
|
|
file_item.metadata = {}
|
|
self.assertTrue(file_item.write())
|
|
self.assert_status(201)
|
|
file_item.metadata = metadata
|
|
self.assertRaises(ResponseError, file_item.sync_metadata)
|
|
self.assert_status(400)
|
|
|
|
def testEtagWayoff(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
hdrs = {'etag': 'reallylonganddefinitelynotavalidetagvalue'}
|
|
self.assertRaises(ResponseError, file_item.write_random, hdrs=hdrs)
|
|
self.assert_status(422)
|
|
|
|
def testFileCreate(self):
|
|
for i in range(10):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
data = file_item.write_random()
|
|
self.assert_status(201)
|
|
self.assertEqual(data, file_item.read())
|
|
self.assert_status(200)
|
|
|
|
def testHead(self):
|
|
file_name = Utils.create_name()
|
|
content_type = Utils.create_name()
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.content_type = content_type
|
|
file_item.write_random(self.env.file_size)
|
|
|
|
expected_etag = file_item.md5
|
|
if tf.cluster_info.get('etag_quoter', {}).get('enable_by_default'):
|
|
expected_etag = '"%s"' % expected_etag
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
info = file_item.info()
|
|
|
|
self.assert_status(200)
|
|
self.assertEqual(info['content_length'], self.env.file_size)
|
|
self.assertEqual(info['etag'], expected_etag)
|
|
self.assertEqual(info['content_type'], content_type)
|
|
self.assertIn('last_modified', info)
|
|
|
|
def testDeleteOfFileThatDoesNotExist(self):
|
|
# in container that exists
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.delete)
|
|
self.assert_status(404)
|
|
|
|
# in container that does not exist
|
|
container = self.env.account.container(Utils.create_name())
|
|
file_item = container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.delete)
|
|
self.assert_status(404)
|
|
|
|
def testHeadOnFileThatDoesNotExist(self):
|
|
# in container that exists
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.info)
|
|
self.assert_status(404)
|
|
|
|
# in container that does not exist
|
|
container = self.env.account.container(Utils.create_name())
|
|
file_item = container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.info)
|
|
self.assert_status(404)
|
|
|
|
def testMetadataOnPost(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.write_random(self.env.file_size)
|
|
|
|
for i in range(10):
|
|
metadata = {}
|
|
for j in range(10):
|
|
metadata[Utils.create_ascii_name()] = Utils.create_name()
|
|
|
|
file_item.metadata = metadata
|
|
self.assertTrue(file_item.sync_metadata())
|
|
self.assert_status(202)
|
|
|
|
file_item = self.env.container.file(file_item.name)
|
|
self.assertTrue(file_item.initialize())
|
|
self.assert_status(200)
|
|
self.assertEqual(file_item.metadata, metadata)
|
|
|
|
def testGetContentType(self):
|
|
file_name = Utils.create_name()
|
|
content_type = Utils.create_name()
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.content_type = content_type
|
|
file_item.write_random()
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.read()
|
|
|
|
self.assertEqual(content_type, file_item.content_type)
|
|
|
|
def testGetOnFileThatDoesNotExist(self):
|
|
# in container that exists
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.read)
|
|
self.assert_status(404)
|
|
|
|
# in container that does not exist
|
|
container = self.env.account.container(Utils.create_name())
|
|
file_item = container.file(Utils.create_name())
|
|
self.assertRaises(ResponseError, file_item.read)
|
|
self.assert_status(404)
|
|
|
|
def testPostOnFileThatDoesNotExist(self):
|
|
# in container that exists
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.metadata['Field'] = 'Value'
|
|
self.assertRaises(ResponseError, file_item.sync_metadata)
|
|
self.assert_status(404)
|
|
|
|
# in container that does not exist
|
|
container = self.env.account.container(Utils.create_name())
|
|
file_item = container.file(Utils.create_name())
|
|
file_item.metadata['Field'] = 'Value'
|
|
self.assertRaises(ResponseError, file_item.sync_metadata)
|
|
self.assert_status(404)
|
|
|
|
def testMetadataOnPut(self):
|
|
for i in range(10):
|
|
metadata = {}
|
|
for j in range(10):
|
|
metadata[Utils.create_ascii_name()] = Utils.create_name()
|
|
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
file_item.metadata = metadata
|
|
file_item.write_random(self.env.file_size)
|
|
|
|
file_item = self.env.container.file(file_item.name)
|
|
self.assertTrue(file_item.initialize())
|
|
self.assert_status(200)
|
|
self.assertEqual(file_item.metadata, metadata)
|
|
|
|
def testSerialization(self):
|
|
container = self.env.account.container(Utils.create_name())
|
|
self.assertTrue(container.create())
|
|
|
|
files = []
|
|
for i in (0, 1, 10, 100, 1000, 10000):
|
|
files.append({'name': Utils.create_name(),
|
|
'content_type': Utils.create_name(), 'bytes': i})
|
|
|
|
write_time = time.time()
|
|
for f in files:
|
|
file_item = container.file(f['name'])
|
|
file_item.content_type = f['content_type']
|
|
file_item.write_random(f['bytes'])
|
|
|
|
f['hash'] = file_item.md5
|
|
f['json'] = False
|
|
f['xml'] = False
|
|
write_time = time.time() - write_time
|
|
|
|
for format_type in ['json', 'xml']:
|
|
for file_item in container.files(parms={'format': format_type}):
|
|
found = False
|
|
for f in files:
|
|
if f['name'] != file_item['name']:
|
|
continue
|
|
|
|
self.assertEqual(file_item['content_type'],
|
|
f['content_type'])
|
|
self.assertEqual(int(file_item['bytes']), f['bytes'])
|
|
|
|
d = datetime.strptime(
|
|
file_item['last_modified'].split('.')[0],
|
|
"%Y-%m-%dT%H:%M:%S")
|
|
lm = time.mktime(d.timetuple())
|
|
|
|
if 'last_modified' in f:
|
|
self.assertEqual(f['last_modified'], lm)
|
|
else:
|
|
f['last_modified'] = lm
|
|
|
|
f[format_type] = True
|
|
found = True
|
|
|
|
self.assertTrue(
|
|
found, 'Unexpected file %s found in '
|
|
'%s listing' % (file_item['name'], format_type))
|
|
|
|
headers = dict((h.lower(), v)
|
|
for h, v in self.env.conn.response.getheaders())
|
|
if format_type == 'json':
|
|
self.assertEqual(headers['content-type'],
|
|
'application/json; charset=utf-8')
|
|
elif format_type == 'xml':
|
|
self.assertEqual(headers['content-type'],
|
|
'application/xml; charset=utf-8')
|
|
|
|
lm_diff = max([f['last_modified'] for f in files]) -\
|
|
min([f['last_modified'] for f in files])
|
|
self.assertLess(lm_diff, write_time + 1,
|
|
'Diff in last modified times '
|
|
'should be less than time to write files')
|
|
|
|
for f in files:
|
|
for format_type in ['json', 'xml']:
|
|
self.assertTrue(
|
|
f[format_type], 'File %s not found in %s listing'
|
|
% (f['name'], format_type))
|
|
|
|
def testStackedOverwrite(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
|
|
for i in range(1, 11):
|
|
data = file_item.write_random(512)
|
|
file_item.write(data)
|
|
|
|
self.assertEqual(file_item.read(), data)
|
|
|
|
def testZeroByteFile(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
|
|
self.assertTrue(file_item.write(b''))
|
|
self.assertIn(file_item.name, self.env.container.files())
|
|
self.assertEqual(file_item.read(), b'')
|
|
|
|
def testEtagResponse(self):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
|
|
data = io.BytesIO(file_item.write_random(512))
|
|
self.assert_etag(File.compute_md5sum(data))
|
|
|
|
def testChunkedPut(self):
|
|
if (tf.web_front_end == 'apache2'):
|
|
raise SkipTest("Chunked PUT cannot be tested with apache2 web "
|
|
"front end")
|
|
|
|
def chunks(s, length=3):
|
|
i, j = 0, length
|
|
while i < len(s):
|
|
yield s[i:j]
|
|
i, j = j, j + length
|
|
|
|
data = File.random_data(10000)
|
|
etag = File.compute_md5sum(data)
|
|
|
|
for i in (1, 10, 100, 1000):
|
|
file_item = self.env.container.file(Utils.create_name())
|
|
|
|
for j in chunks(data, i):
|
|
file_item.chunked_write(j)
|
|
|
|
self.assertTrue(file_item.chunked_write())
|
|
self.assertEqual(data, file_item.read())
|
|
|
|
info = file_item.info()
|
|
self.assertEqual(normalize_etag(info['etag']), etag)
|
|
|
|
def test_POST(self):
|
|
# verify consistency between object and container listing metadata
|
|
file_name = Utils.create_name()
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.content_type = 'text/foobar'
|
|
file_item.write_random(1024)
|
|
|
|
# sanity check
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.initialize()
|
|
self.assertEqual('text/foobar', file_item.content_type)
|
|
self.assertEqual(1024, file_item.size)
|
|
etag = file_item.etag
|
|
|
|
# check container listing is consistent
|
|
listing = self.env.container.files(parms={'format': 'json'})
|
|
for f_dict in listing:
|
|
if f_dict['name'] == file_name:
|
|
break
|
|
else:
|
|
self.fail('Failed to find file %r in listing' % file_name)
|
|
self.assertEqual(1024, f_dict['bytes'])
|
|
self.assertEqual('text/foobar', f_dict['content_type'])
|
|
if tf.cluster_info.get('etag_quoter', {}).get('enable_by_default'):
|
|
self.assertEqual(etag, '"%s"' % f_dict['hash'])
|
|
else:
|
|
self.assertEqual(etag, f_dict['hash'])
|
|
put_last_modified = f_dict['last_modified']
|
|
|
|
# now POST updated content-type to each file
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.content_type = 'image/foobarbaz'
|
|
file_item.sync_metadata({'Test': 'blah'})
|
|
|
|
# sanity check object metadata
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.initialize()
|
|
|
|
self.assertEqual(1024, file_item.size)
|
|
self.assertEqual('image/foobarbaz', file_item.content_type)
|
|
self.assertEqual(etag, file_item.etag)
|
|
self.assertIn('test', file_item.metadata)
|
|
|
|
# check for consistency between object and container listing
|
|
listing = self.env.container.files(parms={'format': 'json'})
|
|
for f_dict in listing:
|
|
if f_dict['name'] == file_name:
|
|
break
|
|
else:
|
|
self.fail('Failed to find file %r in listing' % file_name)
|
|
self.assertEqual(1024, f_dict['bytes'])
|
|
self.assertEqual('image/foobarbaz', f_dict['content_type'])
|
|
self.assertLess(put_last_modified, f_dict['last_modified'])
|
|
if tf.cluster_info.get('etag_quoter', {}).get('enable_by_default'):
|
|
self.assertEqual(etag, '"%s"' % f_dict['hash'])
|
|
else:
|
|
self.assertEqual(etag, f_dict['hash'])
|
|
|
|
|
|
class TestFileUTF8(Base2, TestFile):
|
|
pass
|
|
|
|
|
|
class TestFileComparisonEnv(BaseEnv):
|
|
@classmethod
|
|
def setUp(cls):
|
|
super(TestFileComparisonEnv, cls).setUp()
|
|
cls.container = cls.account.container(Utils.create_name())
|
|
|
|
if not cls.container.create():
|
|
raise ResponseError(cls.conn.response)
|
|
|
|
cls.file_count = 20
|
|
cls.file_size = 128
|
|
cls.files = list()
|
|
for x in range(cls.file_count):
|
|
file_item = cls.container.file(Utils.create_name())
|
|
file_item.write_random(cls.file_size)
|
|
cls.files.append(file_item)
|
|
|
|
cls.time_old_f1 = time.strftime("%a, %d %b %Y %H:%M:%S GMT",
|
|
time.gmtime(time.time() - 86400))
|
|
cls.time_old_f2 = time.strftime("%A, %d-%b-%y %H:%M:%S GMT",
|
|
time.gmtime(time.time() - 86400))
|
|
cls.time_old_f3 = time.strftime("%a %b %d %H:%M:%S %Y",
|
|
time.gmtime(time.time() - 86400))
|
|
cls.time_new = time.strftime("%a, %d %b %Y %H:%M:%S GMT",
|
|
time.gmtime(time.time() + 86400))
|
|
|
|
|
|
class TestFileComparison(Base):
|
|
env = TestFileComparisonEnv
|
|
|
|
def testIfMatch(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-Match': file_item.md5}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
|
|
hdrs = {'If-Match': 'bogus'}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(412)
|
|
self.assert_etag(file_item.md5)
|
|
|
|
def testIfMatchMultipleEtags(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-Match': '"bogus1", "%s", "bogus2"' % file_item.md5}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
|
|
hdrs = {'If-Match': '"bogus1", "bogus2", "bogus3"'}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(412)
|
|
self.assert_etag(file_item.md5)
|
|
|
|
def testIfNoneMatch(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-None-Match': 'bogus'}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
|
|
hdrs = {'If-None-Match': file_item.md5}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(304)
|
|
self.assert_etag(file_item.md5)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
|
|
def testIfNoneMatchMultipleEtags(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-None-Match': '"bogus1", "bogus2", "bogus3"'}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
|
|
hdrs = {'If-None-Match':
|
|
'"bogus1", "bogus2", "%s"' % file_item.md5}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(304)
|
|
self.assert_etag(file_item.md5)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
|
|
def testIfModifiedSince(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-Modified-Since': self.env.time_old_f1}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
self.assertTrue(file_item.info(hdrs=hdrs))
|
|
|
|
hdrs = {'If-Modified-Since': self.env.time_new}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(304)
|
|
self.assert_etag(file_item.md5)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
self.assertRaises(ResponseError, file_item.info, hdrs=hdrs)
|
|
self.assert_status(304)
|
|
self.assert_etag(file_item.md5)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
|
|
def testIfUnmodifiedSince(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-Unmodified-Since': self.env.time_new}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
self.assertTrue(file_item.info(hdrs=hdrs))
|
|
|
|
hdrs = {'If-Unmodified-Since': self.env.time_old_f2}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(412)
|
|
self.assert_etag(file_item.md5)
|
|
self.assertRaises(ResponseError, file_item.info, hdrs=hdrs)
|
|
self.assert_status(412)
|
|
self.assert_etag(file_item.md5)
|
|
|
|
def testIfMatchAndUnmodified(self):
|
|
for file_item in self.env.files:
|
|
hdrs = {'If-Match': file_item.md5,
|
|
'If-Unmodified-Since': self.env.time_new}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
|
|
hdrs = {'If-Match': 'bogus',
|
|
'If-Unmodified-Since': self.env.time_new}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(412)
|
|
self.assert_etag(file_item.md5)
|
|
|
|
hdrs = {'If-Match': file_item.md5,
|
|
'If-Unmodified-Since': self.env.time_old_f3}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(412)
|
|
self.assert_etag(file_item.md5)
|
|
|
|
def testLastModified(self):
|
|
file_name = Utils.create_name()
|
|
content_type = Utils.create_name()
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
file_item.content_type = content_type
|
|
resp = file_item.write_random_return_resp(self.env.file_size)
|
|
put_last_modified = resp.getheader('last-modified')
|
|
etag = file_item.md5
|
|
|
|
file_item = self.env.container.file(file_name)
|
|
info = file_item.info()
|
|
self.assertIn('last_modified', info)
|
|
last_modified = info['last_modified']
|
|
self.assertEqual(put_last_modified, info['last_modified'])
|
|
|
|
hdrs = {'If-Modified-Since': last_modified}
|
|
self.assertRaises(ResponseError, file_item.read, hdrs=hdrs)
|
|
self.assert_status(304)
|
|
self.assert_etag(etag)
|
|
self.assert_header('accept-ranges', 'bytes')
|
|
|
|
hdrs = {'If-Unmodified-Since': last_modified}
|
|
self.assertTrue(file_item.read(hdrs=hdrs))
|
|
|
|
|
|
class TestFileComparisonUTF8(Base2, TestFileComparison):
|
|
pass
|
|
|
|
|
|
class TestServiceToken(unittest.TestCase):
|
|
|
|
def setUp(self):
|
|
if tf.skip_service_tokens:
|
|
raise SkipTest
|
|
|
|
if tf.in_process:
|
|
tf.skip_if_no_xattrs()
|
|
|
|
self.SET_TO_USERS_TOKEN = 1
|
|
self.SET_TO_SERVICE_TOKEN = 2
|
|
|
|
# keystoneauth and tempauth differ in allowing PUT account
|
|
# Even if keystoneauth allows it, the proxy-server uses
|
|
# allow_account_management to decide if accounts can be created
|
|
self.put_account_expect = is_client_error
|
|
if tf.swift_test_auth_version != '1':
|
|
if cluster_info.get('swift').get('allow_account_management'):
|
|
self.put_account_expect = is_success
|
|
|
|
def _scenario_generator(self):
|
|
paths = ((None, None), ('c', None), ('c', 'o'))
|
|
for path in paths:
|
|
for method in ('PUT', 'POST', 'HEAD', 'GET', 'OPTIONS'):
|
|
yield method, path[0], path[1]
|
|
for path in reversed(paths):
|
|
yield 'DELETE', path[0], path[1]
|
|
|
|
def _assert_is_authed_response(self, method, container, object, resp):
|
|
resp.read()
|
|
expect = is_success
|
|
if method == 'DELETE' and not container:
|
|
expect = is_client_error
|
|
if method == 'PUT' and not container:
|
|
expect = self.put_account_expect
|
|
self.assertTrue(expect(resp.status), 'Unexpected %s for %s %s %s'
|
|
% (resp.status, method, container, object))
|
|
|
|
def _assert_not_authed_response(self, method, container, object, resp):
|
|
resp.read()
|
|
expect = is_client_error
|
|
if method == 'OPTIONS':
|
|
expect = is_success
|
|
self.assertTrue(expect(resp.status), 'Unexpected %s for %s %s %s'
|
|
% (resp.status, method, container, object))
|
|
|
|
def prepare_request(self, method, use_service_account=False,
|
|
container=None, obj=None, body=None, headers=None,
|
|
x_auth_token=None,
|
|
x_service_token=None, dbg=False):
|
|
"""
|
|
Setup for making the request
|
|
|
|
When retry() calls the do_request() function, it calls it the
|
|
test user's token, the parsed path, a connection and (optionally)
|
|
a token from the test service user. We save options here so that
|
|
do_request() can make the appropriate request.
|
|
|
|
:param method: The operation (e.g. 'HEAD')
|
|
:param use_service_account: Optional. Set True to change the path to
|
|
be the service account
|
|
:param container: Optional. Adds a container name to the path
|
|
:param obj: Optional. Adds an object name to the path
|
|
:param body: Optional. Adds a body (string) in the request
|
|
:param headers: Optional. Adds additional headers.
|
|
:param x_auth_token: Optional. Default is SET_TO_USERS_TOKEN. One of:
|
|
SET_TO_USERS_TOKEN Put the test user's token in
|
|
X-Auth-Token
|
|
SET_TO_SERVICE_TOKEN Put the service token in X-Auth-Token
|
|
:param x_service_token: Optional. Default is to not set X-Service-Token
|
|
to any value. If specified, is one of following:
|
|
SET_TO_USERS_TOKEN Put the test user's token in
|
|
X-Service-Token
|
|
SET_TO_SERVICE_TOKEN Put the service token in
|
|
X-Service-Token
|
|
:param dbg: Optional. Set true to check request arguments
|
|
"""
|
|
self.method = method
|
|
self.use_service_account = use_service_account
|
|
self.container = container
|
|
self.obj = obj
|
|
self.body = body
|
|
self.headers = headers
|
|
if x_auth_token:
|
|
self.x_auth_token = x_auth_token
|
|
else:
|
|
self.x_auth_token = self.SET_TO_USERS_TOKEN
|
|
self.x_service_token = x_service_token
|
|
self.dbg = dbg
|
|
|
|
def do_request(self, url, token, parsed, conn, service_token=''):
|
|
if self.use_service_account:
|
|
path = self._service_account(parsed.path)
|
|
else:
|
|
path = parsed.path
|
|
if self.container:
|
|
path += '/%s' % self.container
|
|
if self.obj:
|
|
path += '/%s' % self.obj
|
|
headers = {}
|
|
if self.body:
|
|
headers.update({'Content-Length': len(self.body)})
|
|
if self.x_auth_token == self.SET_TO_USERS_TOKEN:
|
|
headers.update({'X-Auth-Token': token})
|
|
elif self.x_auth_token == self.SET_TO_SERVICE_TOKEN:
|
|
headers.update({'X-Auth-Token': service_token})
|
|
if self.x_service_token == self.SET_TO_USERS_TOKEN:
|
|
headers.update({'X-Service-Token': token})
|
|
elif self.x_service_token == self.SET_TO_SERVICE_TOKEN:
|
|
headers.update({'X-Service-Token': service_token})
|
|
if self.dbg:
|
|
print('DEBUG: conn.request: method:%s path:%s'
|
|
' body:%s headers:%s' % (self.method, path, self.body,
|
|
headers))
|
|
conn.request(self.method, path, self.body, headers=headers)
|
|
return check_response(conn)
|
|
|
|
def _service_account(self, path):
|
|
parts = path.split('/', 3)
|
|
account = parts[2]
|
|
try:
|
|
project_id = account[account.index('_') + 1:]
|
|
except ValueError:
|
|
project_id = account
|
|
parts[2] = '%s%s' % (tf.swift_test_service_prefix, project_id)
|
|
return '/'.join(parts)
|
|
|
|
def test_user_access_own_auth_account(self):
|
|
# This covers ground tested elsewhere (tests a user doing HEAD
|
|
# on own account). However, if this fails, none of the remaining
|
|
# tests will work
|
|
self.prepare_request('HEAD')
|
|
resp = retry(self.do_request)
|
|
resp.read()
|
|
self.assertIn(resp.status, (200, 204))
|
|
|
|
def test_user_cannot_access_service_account(self):
|
|
for method, container, obj in self._scenario_generator():
|
|
self.prepare_request(method, use_service_account=True,
|
|
container=container, obj=obj)
|
|
resp = retry(self.do_request)
|
|
self._assert_not_authed_response(method, container, obj, resp)
|
|
|
|
def test_service_user_denied_with_x_auth_token(self):
|
|
for method, container, obj in self._scenario_generator():
|
|
self.prepare_request(method, use_service_account=True,
|
|
container=container, obj=obj,
|
|
x_auth_token=self.SET_TO_SERVICE_TOKEN)
|
|
resp = retry(self.do_request, service_user=5)
|
|
self._assert_not_authed_response(method, container, obj, resp)
|
|
|
|
def test_service_user_denied_with_x_service_token(self):
|
|
for method, container, obj in self._scenario_generator():
|
|
self.prepare_request(method, use_service_account=True,
|
|
container=container, obj=obj,
|
|
x_auth_token=self.SET_TO_SERVICE_TOKEN,
|
|
x_service_token=self.SET_TO_SERVICE_TOKEN)
|
|
resp = retry(self.do_request, service_user=5)
|
|
self._assert_not_authed_response(method, container, obj, resp)
|
|
|
|
def test_user_plus_service_can_access_service_account(self):
|
|
for method, container, obj in self._scenario_generator():
|
|
self.prepare_request(method, use_service_account=True,
|
|
container=container, obj=obj,
|
|
x_auth_token=self.SET_TO_USERS_TOKEN,
|
|
x_service_token=self.SET_TO_SERVICE_TOKEN)
|
|
resp = retry(self.do_request, service_user=5)
|
|
self._assert_is_authed_response(method, container, obj, resp)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|