def _create_request(self, flow, response=None):
request = Request(
method=flow.request.method,
url=flow.request.url,
headers=[(k, v) for k, v in flow.request.headers.items()],
body=flow.request.raw_content
)
# For websocket requests, the scheme of the request is overwritten with https
# in the initial CONNECT request so this hack explicitly sets the scheme back to wss.
if websockets.check_handshake(request.headers) and websockets.check_client_version(request.headers):
request.url = request.url.replace('https', 'wss', 1)
request.response = response
return request
def _process_flow(self, f):
try:
try:
request: http.HTTPRequest = self.read_request_headers(f)
except exceptions.HttpReadDisconnect:
# don't throw an error for disconnects that happen
# before/between requests.
return False
f.request = request
if request.first_line_format == "authority":
# The standards are silent on what we should do with a CONNECT
# request body, so although it's not common, it's allowed.
f.request.data.content = b"".join(
self.read_request_body(f.request))
f.request.data.trailers = self.read_request_trailers(f.request)
f.request.timestamp_end = time.time()
self.channel.ask("http_connect", f)
if self.mode is HTTPMode.regular:
return self.handle_regular_connect(f)
elif self.mode is HTTPMode.upstream:
return self.handle_upstream_connect(f)
else:
msg = "Unexpected CONNECT request."
self.send_error_response(400, msg)
return False
if not self.config.options.relax_http_form_validation:
validate_request_form(self.mode, request)
self.channel.ask("requestheaders", f)
# Re-validate request form in case the user has changed something.
if not self.config.options.relax_http_form_validation:
validate_request_form(self.mode, request)
if request.headers.get("expect", "").lower() == "100-continue":
# TODO: We may have to use send_response_headers for HTTP2
# here.
self.send_response(http.make_expect_continue_response())
request.headers.pop("expect")
if f.request.stream:
f.request.data.content = None
else:
f.request.data.content = b"".join(
self.read_request_body(request))
f.request.data.trailers = self.read_request_trailers(f.request)
request.timestamp_end = time.time()
except exceptions.HttpException as e:
# We optimistically guess there might be an HTTP client on the
# other end
self.send_error_response(400, repr(e))
# Request may be malformed at this point, so we unset it.
f.request = None
f.error = flow.Error(str(e))
self.channel.ask("error", f)
self.log("request", "warn",
["HTTP protocol error in client request: {}".format(e)])
return False
self.log("request", "debug", [repr(request)])
# set first line format to relative in regular mode,
# see https://github.com/mitmproxy/mitmproxy/issues/1759
if self.mode is HTTPMode.regular and request.first_line_format == "absolute":
request.authority = ""
# update host header in reverse mitmproxy mode
if self.config.options.mode.startswith(
"reverse:") and not self.config.options.keep_host_header:
f.request.host_header = self.config.upstream_server.address[0]
# Determine .scheme, .host and .port attributes for inline scripts. For
# absolute-form requests, they are directly given in the request. For
# authority-form requests, we only need to determine the request
# scheme. For relative-form requests, we need to determine host and
# port as well.
if self.mode is HTTPMode.transparent:
# Setting request.host also updates the host header, which we want
# to preserve
f.request.data.host = self.__initial_server_address[0]
f.request.data.port = self.__initial_server_address[1]
f.request.data.scheme = b"https" if self.__initial_server_tls else b"http"
self.channel.ask("request", f)
try:
if websockets.check_handshake(
request.headers) and websockets.check_client_version(
request.headers):
f.metadata['websocket'] = True
# We only support RFC6455 with WebSocket version 13
# allow inline scripts to manipulate the client handshake
self.channel.ask("websocket_handshake", f)
if not f.response:
self.establish_server_connection(f.request.host,
f.request.port,
f.request.scheme)
def get_response():
self.send_request_headers(f.request)
if f.request.stream:
chunks = self.read_request_body(f.request)
if callable(f.request.stream):
chunks = f.request.stream(chunks)
self.send_request_body(f.request, chunks)
else:
self.send_request_body(f.request,
[f.request.data.content])
self.send_request_trailers(f.request)
f.response = self.read_response_headers()
try:
get_response()
except exceptions.NetlibException as e:
self.log("server communication error: %s" % repr(e),
level="debug")
# In any case, we try to reconnect at least once. This is
# necessary because it might be possible that we already
# initiated an upstream connection after clientconnect that
# has already been expired, e.g consider the following event
# log:
# > clientconnect (transparent mode destination known)
# > serverconnect (required for client tls handshake)
# > read n% of large request
# > server detects timeout, disconnects
# > read (100-n)% of large request
# > send large request upstream
if isinstance(e, exceptions.Http2ProtocolException):
# do not try to reconnect for HTTP2
raise exceptions.ProtocolException(
"First and only attempt to get response via HTTP2 failed."
)
elif f.request.stream:
# We may have already consumed some request chunks already,
# so all we can do is signal downstream that upstream closed the connection.
self.send_error_response(408, "Request Timeout")
f.error = flow.Error(repr(e))
self.channel.ask("error", f)
return False
self.disconnect()
self.connect()
get_response()
# call the appropriate script hook - this is an opportunity for
# an inline script to set f.stream = True
self.channel.ask("responseheaders", f)
if f.response.stream:
f.response.data.content = None
else:
f.response.data.content = b"".join(
self.read_response_body(f.request, f.response))
f.response.timestamp_end = time.time()
# no further manipulation of self.server_conn beyond this point
# we can safely set it as the final attribute value here.
f.server_conn = self.server_conn
else:
# response was set by an inline script.
# we now need to emulate the responseheaders hook.
self.channel.ask("responseheaders", f)
f.response.data.trailers = self.read_response_trailers(
f.request, f.response)
self.log("response", "debug", [repr(f.response)])
self.channel.ask("response", f)
if not f.response.stream:
# no streaming:
# we already received the full response from the server and can
# send it to the client straight away.
self.send_response(f.response)
else:
# streaming:
# First send the headers and then transfer the response incrementally
self.send_response_headers(f.response)
chunks = self.read_response_body(f.request, f.response)
if callable(f.response.stream):
chunks = f.response.stream(chunks)
self.send_response_body(f.response, chunks)
f.response.timestamp_end = time.time()
if self.check_close_connection(f):
return False
# Handle 101 Switching Protocols
if f.response.status_code == 101:
# Handle a successful HTTP 101 Switching Protocols Response,
# received after e.g. a WebSocket upgrade request.
# Check for WebSocket handshake
is_websocket = (websockets.check_handshake(f.request.headers)
and websockets.check_handshake(
f.response.headers))
if is_websocket and not self.config.options.websocket:
self.log(
"Client requested WebSocket connection, but the protocol is disabled.",
"info")
if is_websocket and self.config.options.websocket:
layer = WebSocketLayer(self, f)
else:
layer = self.ctx.next_layer(self)
layer()
return False # should never be reached
except (exceptions.ProtocolException, exceptions.NetlibException) as e:
if not f.response:
self.send_error_response(502, repr(e))
f.error = flow.Error(str(e))
self.channel.ask("error", f)
return False
else:
raise exceptions.ProtocolException(
"Error in HTTP connection: %s" % repr(e))
finally:
if f:
f.live = False
return True
def __call__(self, f):
m = ((isinstance(f, http.HTTPFlow) and f.request
and net_websockets.check_handshake(f.request.headers))
or isinstance(f, websocket.WebSocketFlow))
return m