mirror of
https://github.com/Tygs/0bin.git
synced 2023-08-10 21:13:00 +03:00
340 lines
13 KiB
Python
340 lines
13 KiB
Python
import os
|
|
import sys
|
|
|
|
from cherrypy._cpcompat import HTTPConnection, HTTPSConnection, ntob
|
|
from cherrypy._cpcompat import BytesIO
|
|
|
|
curdir = os.path.join(os.getcwd(), os.path.dirname(__file__))
|
|
has_space_filepath = os.path.join(curdir, 'static', 'has space.html')
|
|
bigfile_filepath = os.path.join(curdir, "static", "bigfile.log")
|
|
|
|
# The file size needs to be big enough such that half the size of it
|
|
# won't be socket-buffered (or server-buffered) all in one go. See
|
|
# test_file_stream.
|
|
BIGFILE_SIZE = 1024 * 1024 * 4
|
|
|
|
import cherrypy
|
|
from cherrypy.lib import static
|
|
from cherrypy.test import helper
|
|
|
|
|
|
class StaticTest(helper.CPWebCase):
|
|
|
|
def setup_server():
|
|
if not os.path.exists(has_space_filepath):
|
|
open(has_space_filepath, 'wb').write(ntob('Hello, world\r\n'))
|
|
if not os.path.exists(bigfile_filepath) or \
|
|
os.path.getsize(bigfile_filepath) != BIGFILE_SIZE:
|
|
open(bigfile_filepath, 'wb').write(ntob("x" * BIGFILE_SIZE))
|
|
|
|
class Root:
|
|
|
|
def bigfile(self):
|
|
from cherrypy.lib import static
|
|
self.f = static.serve_file(bigfile_filepath)
|
|
return self.f
|
|
bigfile.exposed = True
|
|
bigfile._cp_config = {'response.stream': True}
|
|
|
|
def tell(self):
|
|
if self.f.input.closed:
|
|
return ''
|
|
return repr(self.f.input.tell()).rstrip('L')
|
|
tell.exposed = True
|
|
|
|
def fileobj(self):
|
|
f = open(os.path.join(curdir, 'style.css'), 'rb')
|
|
return static.serve_fileobj(f, content_type='text/css')
|
|
fileobj.exposed = True
|
|
|
|
def bytesio(self):
|
|
f = BytesIO(ntob('Fee\nfie\nfo\nfum'))
|
|
return static.serve_fileobj(f, content_type='text/plain')
|
|
bytesio.exposed = True
|
|
|
|
class Static:
|
|
|
|
def index(self):
|
|
return 'You want the Baron? You can have the Baron!'
|
|
index.exposed = True
|
|
|
|
def dynamic(self):
|
|
return "This is a DYNAMIC page"
|
|
dynamic.exposed = True
|
|
|
|
root = Root()
|
|
root.static = Static()
|
|
|
|
rootconf = {
|
|
'/static': {
|
|
'tools.staticdir.on': True,
|
|
'tools.staticdir.dir': 'static',
|
|
'tools.staticdir.root': curdir,
|
|
},
|
|
'/style.css': {
|
|
'tools.staticfile.on': True,
|
|
'tools.staticfile.filename': os.path.join(curdir, 'style.css'),
|
|
},
|
|
'/docroot': {
|
|
'tools.staticdir.on': True,
|
|
'tools.staticdir.root': curdir,
|
|
'tools.staticdir.dir': 'static',
|
|
'tools.staticdir.index': 'index.html',
|
|
},
|
|
'/error': {
|
|
'tools.staticdir.on': True,
|
|
'request.show_tracebacks': True,
|
|
},
|
|
'/404test': {
|
|
'tools.staticdir.on': True,
|
|
'tools.staticdir.root': curdir,
|
|
'tools.staticdir.dir': 'static',
|
|
'error_page.404': error_page_404,
|
|
}
|
|
}
|
|
rootApp = cherrypy.Application(root)
|
|
rootApp.merge(rootconf)
|
|
|
|
test_app_conf = {
|
|
'/test': {
|
|
'tools.staticdir.index': 'index.html',
|
|
'tools.staticdir.on': True,
|
|
'tools.staticdir.root': curdir,
|
|
'tools.staticdir.dir': 'static',
|
|
},
|
|
}
|
|
testApp = cherrypy.Application(Static())
|
|
testApp.merge(test_app_conf)
|
|
|
|
vhost = cherrypy._cpwsgi.VirtualHost(rootApp, {'virt.net': testApp})
|
|
cherrypy.tree.graft(vhost)
|
|
setup_server = staticmethod(setup_server)
|
|
|
|
def teardown_server():
|
|
for f in (has_space_filepath, bigfile_filepath):
|
|
if os.path.exists(f):
|
|
try:
|
|
os.unlink(f)
|
|
except:
|
|
pass
|
|
teardown_server = staticmethod(teardown_server)
|
|
|
|
def testStatic(self):
|
|
self.getPage("/static/index.html")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/html')
|
|
self.assertBody('Hello, world\r\n')
|
|
|
|
# Using a staticdir.root value in a subdir...
|
|
self.getPage("/docroot/index.html")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/html')
|
|
self.assertBody('Hello, world\r\n')
|
|
|
|
# Check a filename with spaces in it
|
|
self.getPage("/static/has%20space.html")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/html')
|
|
self.assertBody('Hello, world\r\n')
|
|
|
|
self.getPage("/style.css")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/css')
|
|
# Note: The body should be exactly 'Dummy stylesheet\n', but
|
|
# unfortunately some tools such as WinZip sometimes turn \n
|
|
# into \r\n on Windows when extracting the CherryPy tarball so
|
|
# we just check the content
|
|
self.assertMatchesBody('^Dummy stylesheet')
|
|
|
|
def test_fallthrough(self):
|
|
# Test that NotFound will then try dynamic handlers (see [878]).
|
|
self.getPage("/static/dynamic")
|
|
self.assertBody("This is a DYNAMIC page")
|
|
|
|
# Check a directory via fall-through to dynamic handler.
|
|
self.getPage("/static/")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/html;charset=utf-8')
|
|
self.assertBody('You want the Baron? You can have the Baron!')
|
|
|
|
def test_index(self):
|
|
# Check a directory via "staticdir.index".
|
|
self.getPage("/docroot/")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/html')
|
|
self.assertBody('Hello, world\r\n')
|
|
# The same page should be returned even if redirected.
|
|
self.getPage("/docroot")
|
|
self.assertStatus(301)
|
|
self.assertHeader('Location', '%s/docroot/' % self.base())
|
|
self.assertMatchesBody("This resource .* <a href=(['\"])%s/docroot/\\1>"
|
|
"%s/docroot/</a>." % (self.base(), self.base()))
|
|
|
|
def test_config_errors(self):
|
|
# Check that we get an error if no .file or .dir
|
|
self.getPage("/error/thing.html")
|
|
self.assertErrorPage(500)
|
|
if sys.version_info >= (3, 3):
|
|
errmsg = ntob("TypeError: staticdir\(\) missing 2 "
|
|
"required positional arguments")
|
|
else:
|
|
errmsg = ntob("TypeError: staticdir\(\) takes at least 2 "
|
|
"(positional )?arguments \(0 given\)")
|
|
self.assertMatchesBody(errmsg)
|
|
|
|
def test_security(self):
|
|
# Test up-level security
|
|
self.getPage("/static/../../test/style.css")
|
|
self.assertStatus((400, 403))
|
|
|
|
def test_modif(self):
|
|
# Test modified-since on a reasonably-large file
|
|
self.getPage("/static/dirback.jpg")
|
|
self.assertStatus("200 OK")
|
|
lastmod = ""
|
|
for k, v in self.headers:
|
|
if k == 'Last-Modified':
|
|
lastmod = v
|
|
ims = ("If-Modified-Since", lastmod)
|
|
self.getPage("/static/dirback.jpg", headers=[ims])
|
|
self.assertStatus(304)
|
|
self.assertNoHeader("Content-Type")
|
|
self.assertNoHeader("Content-Length")
|
|
self.assertNoHeader("Content-Disposition")
|
|
self.assertBody("")
|
|
|
|
def test_755_vhost(self):
|
|
self.getPage("/test/", [('Host', 'virt.net')])
|
|
self.assertStatus(200)
|
|
self.getPage("/test", [('Host', 'virt.net')])
|
|
self.assertStatus(301)
|
|
self.assertHeader('Location', self.scheme + '://virt.net/test/')
|
|
|
|
def test_serve_fileobj(self):
|
|
self.getPage("/fileobj")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/css;charset=utf-8')
|
|
self.assertMatchesBody('^Dummy stylesheet')
|
|
|
|
def test_serve_bytesio(self):
|
|
self.getPage("/bytesio")
|
|
self.assertStatus('200 OK')
|
|
self.assertHeader('Content-Type', 'text/plain;charset=utf-8')
|
|
self.assertHeader('Content-Length', 14)
|
|
self.assertMatchesBody('Fee\nfie\nfo\nfum')
|
|
|
|
def test_file_stream(self):
|
|
if cherrypy.server.protocol_version != "HTTP/1.1":
|
|
return self.skip()
|
|
|
|
self.PROTOCOL = "HTTP/1.1"
|
|
|
|
# Make an initial request
|
|
self.persistent = True
|
|
conn = self.HTTP_CONN
|
|
conn.putrequest("GET", "/bigfile", skip_host=True)
|
|
conn.putheader("Host", self.HOST)
|
|
conn.endheaders()
|
|
response = conn.response_class(conn.sock, method="GET")
|
|
response.begin()
|
|
self.assertEqual(response.status, 200)
|
|
|
|
body = ntob('')
|
|
remaining = BIGFILE_SIZE
|
|
while remaining > 0:
|
|
data = response.fp.read(65536)
|
|
if not data:
|
|
break
|
|
body += data
|
|
remaining -= len(data)
|
|
|
|
if self.scheme == "https":
|
|
newconn = HTTPSConnection
|
|
else:
|
|
newconn = HTTPConnection
|
|
s, h, b = helper.webtest.openURL(
|
|
ntob("/tell"), headers=[], host=self.HOST, port=self.PORT,
|
|
http_conn=newconn)
|
|
if not b:
|
|
# The file was closed on the server.
|
|
tell_position = BIGFILE_SIZE
|
|
else:
|
|
tell_position = int(b)
|
|
|
|
read_so_far = len(body)
|
|
|
|
# It is difficult for us to force the server to only read
|
|
# the bytes that we ask for - there are going to be buffers
|
|
# inbetween.
|
|
#
|
|
# CherryPy will attempt to write as much data as it can to
|
|
# the socket, and we don't have a way to determine what that
|
|
# size will be. So we make the following assumption - by
|
|
# the time we have read in the entire file on the server,
|
|
# we will have at least received half of it. If this is not
|
|
# the case, then this is an indicator that either:
|
|
# - machines that are running this test are using buffer
|
|
# sizes greater than half of BIGFILE_SIZE; or
|
|
# - streaming is broken.
|
|
#
|
|
# At the time of writing, we seem to have encountered
|
|
# buffer sizes bigger than 512K, so we've increased
|
|
# BIGFILE_SIZE to 4MB.
|
|
if tell_position >= BIGFILE_SIZE:
|
|
if read_so_far < (BIGFILE_SIZE / 2):
|
|
self.fail(
|
|
"The file should have advanced to position %r, but "
|
|
"has already advanced to the end of the file. It "
|
|
"may not be streamed as intended, or at the wrong "
|
|
"chunk size (64k)" % read_so_far)
|
|
elif tell_position < read_so_far:
|
|
self.fail(
|
|
"The file should have advanced to position %r, but has "
|
|
"only advanced to position %r. It may not be streamed "
|
|
"as intended, or at the wrong chunk size (64k)" %
|
|
(read_so_far, tell_position))
|
|
|
|
if body != ntob("x" * BIGFILE_SIZE):
|
|
self.fail("Body != 'x' * %d. Got %r instead (%d bytes)." %
|
|
(BIGFILE_SIZE, body[:50], len(body)))
|
|
conn.close()
|
|
|
|
def test_file_stream_deadlock(self):
|
|
if cherrypy.server.protocol_version != "HTTP/1.1":
|
|
return self.skip()
|
|
|
|
self.PROTOCOL = "HTTP/1.1"
|
|
|
|
# Make an initial request but abort early.
|
|
self.persistent = True
|
|
conn = self.HTTP_CONN
|
|
conn.putrequest("GET", "/bigfile", skip_host=True)
|
|
conn.putheader("Host", self.HOST)
|
|
conn.endheaders()
|
|
response = conn.response_class(conn.sock, method="GET")
|
|
response.begin()
|
|
self.assertEqual(response.status, 200)
|
|
body = response.fp.read(65536)
|
|
if body != ntob("x" * len(body)):
|
|
self.fail("Body != 'x' * %d. Got %r instead (%d bytes)." %
|
|
(65536, body[:50], len(body)))
|
|
response.close()
|
|
conn.close()
|
|
|
|
# Make a second request, which should fetch the whole file.
|
|
self.persistent = False
|
|
self.getPage("/bigfile")
|
|
if self.body != ntob("x" * BIGFILE_SIZE):
|
|
self.fail("Body != 'x' * %d. Got %r instead (%d bytes)." %
|
|
(BIGFILE_SIZE, self.body[:50], len(body)))
|
|
|
|
def test_error_page_with_serve_file(self):
|
|
self.getPage("/404test/yunyeen")
|
|
self.assertStatus(404)
|
|
self.assertInBody("I couldn't find that thing")
|
|
|
|
def error_page_404(status, message, traceback, version):
|
|
import os.path
|
|
return static.serve_file(os.path.join(curdir, 'static', '404.html'),
|
|
content_type='text/html')
|