diff options
Diffstat (limited to 'libmproxy/protocol2')
-rw-r--r-- | libmproxy/protocol2/__init__.py | 13 | ||||
-rw-r--r-- | libmproxy/protocol2/http.py | 588 | ||||
-rw-r--r-- | libmproxy/protocol2/http_proxy.py | 26 | ||||
-rw-r--r-- | libmproxy/protocol2/http_replay.py | 95 | ||||
-rw-r--r-- | libmproxy/protocol2/layer.py | 138 | ||||
-rw-r--r-- | libmproxy/protocol2/rawtcp.py | 66 | ||||
-rw-r--r-- | libmproxy/protocol2/reverse_proxy.py | 18 | ||||
-rw-r--r-- | libmproxy/protocol2/root_context.py | 94 | ||||
-rw-r--r-- | libmproxy/protocol2/socks_proxy.py | 59 | ||||
-rw-r--r-- | libmproxy/protocol2/tls.py | 288 | ||||
-rw-r--r-- | libmproxy/protocol2/transparent_proxy.py | 24 |
11 files changed, 1409 insertions, 0 deletions
diff --git a/libmproxy/protocol2/__init__.py b/libmproxy/protocol2/__init__.py new file mode 100644 index 00000000..61b9a77e --- /dev/null +++ b/libmproxy/protocol2/__init__.py @@ -0,0 +1,13 @@ +from __future__ import (absolute_import, print_function, division) +from .root_context import RootContext +from .socks_proxy import Socks5Proxy +from .reverse_proxy import ReverseProxy +from .http_proxy import HttpProxy, HttpUpstreamProxy +from .transparent_proxy import TransparentProxy +from .http import make_error_response + +__all__ = [ + "RootContext", + "Socks5Proxy", "ReverseProxy", "HttpProxy", "HttpUpstreamProxy", "TransparentProxy", + "make_error_response" +] diff --git a/libmproxy/protocol2/http.py b/libmproxy/protocol2/http.py new file mode 100644 index 00000000..a508ae8b --- /dev/null +++ b/libmproxy/protocol2/http.py @@ -0,0 +1,588 @@ +from __future__ import (absolute_import, print_function, division) + +from netlib import tcp +from netlib.http import status_codes, http1, HttpErrorConnClosed, HttpError +from netlib.http.semantics import CONTENT_MISSING +from netlib import odict +from netlib.tcp import NetLibError, Address +from netlib.http.http1 import HTTP1Protocol +from netlib.http.http2 import HTTP2Protocol + +from .. import version, utils +from ..exceptions import InvalidCredentials, HttpException, ProtocolException +from .layer import Layer +from ..proxy import Kill +from libmproxy.protocol import KILL, Error +from libmproxy.protocol.http import HTTPFlow +from libmproxy.protocol.http_wrappers import HTTPResponse, HTTPRequest + + +class _HttpLayer(Layer): + supports_streaming = False + + def read_request(self): + raise NotImplementedError() + + def send_request(self, request): + raise NotImplementedError() + + def read_response(self, request_method): + raise NotImplementedError() + + def send_response(self, response): + raise NotImplementedError() + + +class _StreamingHttpLayer(_HttpLayer): + supports_streaming = True + + def read_response_headers(self): + raise NotImplementedError + + def read_response_body(self, headers, request_method, response_code, max_chunk_size=None): + raise NotImplementedError() + yield "this is a generator" + + def send_response_headers(self, response): + raise NotImplementedError + + def send_response_body(self, response, chunks): + raise NotImplementedError() + + +class Http1Layer(_StreamingHttpLayer): + def __init__(self, ctx, mode): + super(Http1Layer, self).__init__(ctx) + self.mode = mode + self.client_protocol = HTTP1Protocol(self.client_conn) + self.server_protocol = HTTP1Protocol(self.server_conn) + + def read_request(self): + return HTTPRequest.from_protocol( + self.client_protocol, + body_size_limit=self.config.body_size_limit + ) + + def send_request(self, request): + self.server_conn.send(self.server_protocol.assemble(request)) + + def read_response(self, request_method): + return HTTPResponse.from_protocol( + self.server_protocol, + request_method=request_method, + body_size_limit=self.config.body_size_limit, + include_body=True + ) + + def send_response(self, response): + self.client_conn.send(self.client_protocol.assemble(response)) + + def read_response_headers(self): + return HTTPResponse.from_protocol( + self.server_protocol, + request_method=None, # does not matter if we don't read the body. + body_size_limit=self.config.body_size_limit, + include_body=False + ) + + def read_response_body(self, headers, request_method, response_code, max_chunk_size=None): + return self.server_protocol.read_http_body_chunked( + headers, + self.config.body_size_limit, + request_method, + response_code, + False, + max_chunk_size + ) + + def send_response_headers(self, response): + h = self.client_protocol._assemble_response_first_line(response) + self.client_conn.wfile.write(h + "\r\n") + h = self.client_protocol._assemble_response_headers( + response, + preserve_transfer_encoding=True + ) + self.client_conn.send(h + "\r\n") + + def send_response_body(self, response, chunks): + if self.client_protocol.has_chunked_encoding(response.headers): + chunks = ( + "%d\r\n%s\r\n" % (len(chunk), chunk) + for chunk in chunks + ) + for chunk in chunks: + self.client_conn.send(chunk) + + def connect(self): + self.ctx.connect() + self.server_protocol = HTTP1Protocol(self.server_conn) + + def reconnect(self): + self.ctx.reconnect() + self.server_protocol = HTTP1Protocol(self.server_conn) + + def set_server(self, *args, **kwargs): + self.ctx.set_server(*args, **kwargs) + self.server_protocol = HTTP1Protocol(self.server_conn) + + def __call__(self): + layer = HttpLayer(self, self.mode) + layer() + + +# TODO: The HTTP2 layer is missing multiplexing, which requires a major rewrite. +class Http2Layer(_HttpLayer): + def __init__(self, ctx, mode): + super(Http2Layer, self).__init__(ctx) + self.mode = mode + self.client_protocol = HTTP2Protocol(self.client_conn, is_server=True, + unhandled_frame_cb=self.handle_unexpected_frame) + self.server_protocol = HTTP2Protocol(self.server_conn, is_server=False, + unhandled_frame_cb=self.handle_unexpected_frame) + + def read_request(self): + request = HTTPRequest.from_protocol( + self.client_protocol, + body_size_limit=self.config.body_size_limit + ) + self._stream_id = request.stream_id + return request + + def send_request(self, message): + # TODO: implement flow control and WINDOW_UPDATE frames + self.server_conn.send(self.server_protocol.assemble(message)) + + def read_response(self, request_method): + return HTTPResponse.from_protocol( + self.server_protocol, + request_method=request_method, + body_size_limit=self.config.body_size_limit, + include_body=True, + stream_id=self._stream_id + ) + + def send_response(self, message): + # TODO: implement flow control and WINDOW_UPDATE frames + self.client_conn.send(self.client_protocol.assemble(message)) + + def connect(self): + self.ctx.connect() + self.server_protocol = HTTP2Protocol(self.server_conn, is_server=False, + unhandled_frame_cb=self.handle_unexpected_frame) + self.server_protocol.perform_connection_preface() + + def reconnect(self): + self.ctx.reconnect() + self.server_protocol = HTTP2Protocol(self.server_conn, is_server=False, + unhandled_frame_cb=self.handle_unexpected_frame) + self.server_protocol.perform_connection_preface() + + def set_server(self, *args, **kwargs): + self.ctx.set_server(*args, **kwargs) + self.server_protocol = HTTP2Protocol(self.server_conn, is_server=False, + unhandled_frame_cb=self.handle_unexpected_frame) + self.server_protocol.perform_connection_preface() + + def __call__(self): + self.server_protocol.perform_connection_preface() + layer = HttpLayer(self, self.mode) + layer() + + def handle_unexpected_frame(self, frm): + self.log("Unexpected HTTP2 Frame: %s" % frm.human_readable(), "info") + + +def make_error_response(status_code, message, headers=None): + response = status_codes.RESPONSES.get(status_code, "Unknown") + body = """ + <html> + <head> + <title>%d %s</title> + </head> + <body>%s</body> + </html> + """.strip() % (status_code, response, message) + + if not headers: + headers = odict.ODictCaseless() + headers["Server"] = [version.NAMEVERSION] + headers["Connection"] = ["close"] + headers["Content-Length"] = [len(body)] + headers["Content-Type"] = ["text/html"] + + return HTTPResponse( + (1, 1), # FIXME: Should be a string. + status_code, + response, + headers, + body, + ) + + +def make_connect_request(address): + address = Address.wrap(address) + return HTTPRequest( + "authority", "CONNECT", None, address.host, address.port, None, (1, 1), + odict.ODictCaseless(), "" + ) + + +def make_connect_response(httpversion): + headers = odict.ODictCaseless([ + ["Content-Length", "0"], + ["Proxy-Agent", version.NAMEVERSION] + ]) + return HTTPResponse( + httpversion, + 200, + "Connection established", + headers, + "", + ) + + +class ConnectServerConnection(object): + """ + "Fake" ServerConnection to represent state after a CONNECT request to an upstream proxy. + """ + + def __init__(self, address, ctx): + self.address = tcp.Address.wrap(address) + self._ctx = ctx + + @property + def via(self): + return self._ctx.server_conn + + def __getattr__(self, item): + return getattr(self.via, item) + + +class UpstreamConnectLayer(Layer): + def __init__(self, ctx, connect_request): + super(UpstreamConnectLayer, self).__init__(ctx) + self.connect_request = connect_request + self.server_conn = ConnectServerConnection( + (connect_request.host, connect_request.port), + self.ctx + ) + + def __call__(self): + layer = self.ctx.next_layer(self) + layer() + + def connect(self): + if not self.server_conn: + self.ctx.connect() + self.send_request(self.connect_request) + else: + pass # swallow the message + + def reconnect(self): + self.ctx.reconnect() + self.send_request(self.connect_request) + resp = self.read_response("CONNECT") + if resp.code != 200: + raise ProtocolException("Reconnect: Upstream server refuses CONNECT request") + + def set_server(self, address, server_tls=None, sni=None, depth=1): + if depth == 1: + if self.ctx.server_conn: + self.ctx.reconnect() + address = Address.wrap(address) + self.connect_request.host = address.host + self.connect_request.port = address.port + self.server_conn.address = address + else: + self.ctx.set_server(address, server_tls, sni, depth - 1) + + +class HttpLayer(Layer): + def __init__(self, ctx, mode): + super(HttpLayer, self).__init__(ctx) + self.mode = mode + self.__original_server_conn = None + "Contains the original destination in transparent mode, which needs to be restored" + "if an inline script modified the target server for a single http request" + + def __call__(self): + if self.mode == "transparent": + self.__original_server_conn = self.server_conn + while True: + try: + flow = HTTPFlow(self.client_conn, self.server_conn, live=self) + + try: + request = self.read_request() + except tcp.NetLibError: + # don't throw an error for disconnects that happen + # before/between requests. + return + + self.log("request", "debug", [repr(request)]) + + # Handle Proxy Authentication + self.authenticate(request) + + # Regular Proxy Mode: Handle CONNECT + if self.mode == "regular" and request.form_in == "authority": + self.handle_regular_mode_connect(request) + return + + # Make sure that the incoming request matches our expectations + self.validate_request(request) + + flow.request = request + self.process_request_hook(flow) + + if not flow.response: + self.establish_server_connection(flow) + self.get_response_from_server(flow) + + self.send_response_to_client(flow) + + if self.check_close_connection(flow): + return + + # TODO: Implement HTTP Upgrade + + # Upstream Proxy Mode: Handle CONNECT + if flow.request.form_in == "authority" and flow.response.code == 200: + self.handle_upstream_mode_connect(flow.request.copy()) + return + + except (HttpErrorConnClosed, NetLibError, HttpError, ProtocolException) as e: + if flow.request and not flow.response: + flow.error = Error(repr(e)) + self.channel.ask("error", flow) + try: + self.send_response(make_error_response( + getattr(e, "code", 502), + repr(e) + )) + except NetLibError: + pass + if isinstance(e, ProtocolException): + raise e + else: + raise ProtocolException("Error in HTTP connection: %s" % repr(e), e) + finally: + flow.live = False + + def handle_regular_mode_connect(self, request): + self.set_server((request.host, request.port)) + self.send_response(make_connect_response(request.httpversion)) + layer = self.ctx.next_layer(self) + layer() + + def handle_upstream_mode_connect(self, connect_request): + layer = UpstreamConnectLayer(self, connect_request) + layer() + + def check_close_connection(self, flow): + """ + Checks if the connection should be closed depending on the HTTP + semantics. Returns True, if so. + """ + + # TODO: add logic for HTTP/2 + + close_connection = ( + http1.HTTP1Protocol.connection_close( + flow.request.httpversion, + flow.request.headers + ) or http1.HTTP1Protocol.connection_close( + flow.response.httpversion, + flow.response.headers + ) or http1.HTTP1Protocol.expected_http_body_size( + flow.response.headers, + False, + flow.request.method, + flow.response.code) == -1 + ) + if flow.request.form_in == "authority" and flow.response.code == 200: + # Workaround for + # https://github.com/mitmproxy/mitmproxy/issues/313: Some + # proxies (e.g. Charles) send a CONNECT response with HTTP/1.0 + # and no Content-Length header + + return False + return close_connection + + def send_response_to_client(self, flow): + if not (self.supports_streaming and flow.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(flow.response) + else: + # streaming: + # First send the headers and then transfer the response incrementally + self.send_response_headers(flow.response) + chunks = self.read_response_body( + flow.response.headers, + flow.request.method, + flow.response.code, + max_chunk_size=4096 + ) + if callable(flow.response.stream): + chunks = flow.response.stream(chunks) + self.send_response_body(flow.response, chunks) + flow.response.timestamp_end = utils.timestamp() + + def get_response_from_server(self, flow): + def get_response(): + self.send_request(flow.request) + if self.supports_streaming: + flow.response = self.read_response_headers() + else: + flow.response = self.read_response() + + try: + get_response() + except (tcp.NetLibError, HttpErrorConnClosed) as v: + self.log( + "server communication error: %s" % repr(v), + 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 + self.reconnect() + get_response() + + # call the appropriate script hook - this is an opportunity for an + # inline script to set flow.stream = True + flow = self.channel.ask("responseheaders", flow) + if flow is None or flow == KILL: + raise Kill() + + if self.supports_streaming: + if flow.response.stream: + flow.response.content = CONTENT_MISSING + else: + flow.response.content = "".join(self.read_response_body( + flow.response.headers, + flow.request.method, + flow.response.code + )) + flow.response.timestamp_end = utils.timestamp() + + # no further manipulation of self.server_conn beyond this point + # we can safely set it as the final attribute value here. + flow.server_conn = self.server_conn + + self.log( + "response", + "debug", + [repr(flow.response)] + ) + response_reply = self.channel.ask("response", flow) + if response_reply is None or response_reply == KILL: + raise Kill() + + def process_request_hook(self, flow): + # 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 == "regular": + pass # only absolute-form at this point, nothing to do here. + elif self.mode == "upstream": + if flow.request.form_in == "authority": + flow.request.scheme = "http" # pseudo value + else: + flow.request.host = self.__original_server_conn.address.host + flow.request.port = self.__original_server_conn.address.port + flow.request.scheme = "https" if self.__original_server_conn.tls_established else "http" + + request_reply = self.channel.ask("request", flow) + if request_reply is None or request_reply == KILL: + raise Kill() + if isinstance(request_reply, HTTPResponse): + flow.response = request_reply + return + + def establish_server_connection(self, flow): + address = tcp.Address((flow.request.host, flow.request.port)) + tls = (flow.request.scheme == "https") + + if self.mode == "regular" or self.mode == "transparent": + # If there's an existing connection that doesn't match our expectations, kill it. + if address != self.server_conn.address or tls != self.server_conn.ssl_established: + self.set_server(address, tls, address.host) + # Establish connection is neccessary. + if not self.server_conn: + self.connect() + + # SetServer is not guaranteed to work with TLS: + # If there's not TlsLayer below which could catch the exception, + # TLS will not be established. + if tls and not self.server_conn.tls_established: + raise ProtocolException( + "Cannot upgrade to SSL, no TLS layer on the protocol stack.") + else: + if not self.server_conn: + self.connect() + if tls: + raise HttpException("Cannot change scheme in upstream proxy mode.") + """ + # This is a very ugly (untested) workaround to solve a very ugly problem. + if self.server_conn and self.server_conn.tls_established and not ssl: + self.reconnect() + elif ssl and not hasattr(self, "connected_to") or self.connected_to != address: + if self.server_conn.tls_established: + self.reconnect() + + self.send_request(make_connect_request(address)) + tls_layer = TlsLayer(self, False, True) + tls_layer._establish_tls_with_server() + """ + + def validate_request(self, request): + if request.form_in == "absolute" and request.scheme != "http": + self.send_response( + make_error_response(400, "Invalid request scheme: %s" % request.scheme)) + raise HttpException("Invalid request scheme: %s" % request.scheme) + + expected_request_forms = { + "regular": ("absolute",), # an authority request would already be handled. + "upstream": ("authority", "absolute"), + "transparent": ("relative",) + } + + allowed_request_forms = expected_request_forms[self.mode] + if request.form_in not in allowed_request_forms: + err_message = "Invalid HTTP request form (expected: %s, got: %s)" % ( + " or ".join(allowed_request_forms), request.form_in + ) + self.send_response(make_error_response(400, err_message)) + raise HttpException(err_message) + + if self.mode == "regular": + request.form_out = "relative" + + def authenticate(self, request): + if self.config.authenticator: + if self.config.authenticator.authenticate(request.headers): + self.config.authenticator.clean(request.headers) + else: + self.send_response(make_error_response( + 407, + "Proxy Authentication Required", + odict.ODictCaseless( + [ + [k, v] for k, v in + self.config.authenticator.auth_challenge_headers().items() + ]) + )) + raise InvalidCredentials("Proxy Authentication Required") diff --git a/libmproxy/protocol2/http_proxy.py b/libmproxy/protocol2/http_proxy.py new file mode 100644 index 00000000..2876c022 --- /dev/null +++ b/libmproxy/protocol2/http_proxy.py @@ -0,0 +1,26 @@ +from __future__ import (absolute_import, print_function, division) + +from .layer import Layer, ServerConnectionMixin + + +class HttpProxy(Layer, ServerConnectionMixin): + def __call__(self): + layer = self.ctx.next_layer(self) + try: + layer() + finally: + if self.server_conn: + self._disconnect() + + +class HttpUpstreamProxy(Layer, ServerConnectionMixin): + def __init__(self, ctx, server_address): + super(HttpUpstreamProxy, self).__init__(ctx, server_address=server_address) + + def __call__(self): + layer = self.ctx.next_layer(self) + try: + layer() + finally: + if self.server_conn: + self._disconnect() diff --git a/libmproxy/protocol2/http_replay.py b/libmproxy/protocol2/http_replay.py new file mode 100644 index 00000000..872ef9cd --- /dev/null +++ b/libmproxy/protocol2/http_replay.py @@ -0,0 +1,95 @@ +import threading +from netlib.http import HttpError +from netlib.http.http1 import HTTP1Protocol +from netlib.tcp import NetLibError + +from ..controller import Channel +from ..protocol import KILL, Error +from ..protocol.http_wrappers import HTTPResponse +from ..proxy import Log, Kill +from ..proxy.connection import ServerConnection +from .http import make_connect_request + + +class RequestReplayThread(threading.Thread): + name = "RequestReplayThread" + + def __init__(self, config, flow, masterq, should_exit): + """ + masterqueue can be a queue or None, if no scripthooks should be + processed. + """ + self.config, self.flow = config, flow + if masterq: + self.channel = Channel(masterq, should_exit) + else: + self.channel = None + super(RequestReplayThread, self).__init__() + + def run(self): + r = self.flow.request + form_out_backup = r.form_out + try: + self.flow.response = None + + # If we have a channel, run script hooks. + if self.channel: + request_reply = self.channel.ask("request", self.flow) + if request_reply is None or request_reply == KILL: + raise Kill() + elif isinstance(request_reply, HTTPResponse): + self.flow.response = request_reply + + if not self.flow.response: + # In all modes, we directly connect to the server displayed + if self.config.mode == "upstream": + server_address = self.config.upstream_server.address + server = ServerConnection(server_address) + server.connect() + protocol = HTTP1Protocol(server) + if r.scheme == "https": + connect_request = make_connect_request((r.host, r.port)) + server.send(protocol.assemble(connect_request)) + resp = protocol.read_response("CONNECT") + if resp.code != 200: + raise HttpError(502, "Upstream server refuses CONNECT request") + server.establish_ssl( + self.config.clientcerts, + sni=self.flow.server_conn.sni + ) + r.form_out = "relative" + else: + r.form_out = "absolute" + else: + server_address = (r.host, r.port) + server = ServerConnection(server_address) + server.connect() + protocol = HTTP1Protocol(server) + if r.scheme == "https": + server.establish_ssl( + self.config.clientcerts, + sni=self.flow.server_conn.sni + ) + r.form_out = "relative" + + server.send(protocol.assemble(r)) + self.flow.server_conn = server + self.flow.response = HTTPResponse.from_protocol( + protocol, + r.method, + body_size_limit=self.config.body_size_limit, + ) + if self.channel: + response_reply = self.channel.ask("response", self.flow) + if response_reply is None or response_reply == KILL: + raise Kill() + except (HttpError, NetLibError) as v: + self.flow.error = Error(repr(v)) + if self.channel: + self.channel.ask("error", self.flow) + except Kill: + # KillSignal should only be raised if there's a channel in the + # first place. + self.channel.tell("log", Log("Connection killed", "info")) + finally: + r.form_out = form_out_backup diff --git a/libmproxy/protocol2/layer.py b/libmproxy/protocol2/layer.py new file mode 100644 index 00000000..2b47cc26 --- /dev/null +++ b/libmproxy/protocol2/layer.py @@ -0,0 +1,138 @@ +""" +mitmproxy protocol architecture + +In mitmproxy, protocols are implemented as a set of layers, which are composed on top each other. +For example, the following scenarios depict possible scenarios (lowest layer first): + +Transparent HTTP proxy, no SSL: + TransparentModeLayer + HttpLayer + +Regular proxy, CONNECT request with WebSockets over SSL: + RegularModeLayer + HttpLayer + SslLayer + WebsocketLayer (or TcpLayer) + +Automated protocol detection by peeking into the buffer: + TransparentModeLayer + SslLayer + Http2Layer + +Communication between layers is done as follows: + - lower layers provide context information to higher layers + - higher layers can call functions provided by lower layers, + which are propagated until they reach a suitable layer. + +Further goals: + - Connections should always be peekable to make automatic protocol detection work. + - Upstream connections should be established as late as possible; + inline scripts shall have a chance to handle everything locally. +""" +from __future__ import (absolute_import, print_function, division) +from netlib import tcp +from ..proxy import Log +from ..proxy.connection import ServerConnection +from ..exceptions import ProtocolException + + +class _LayerCodeCompletion(object): + """ + Dummy class that provides type hinting in PyCharm, which simplifies development a lot. + """ + + def __init__(self, *args, **kwargs): + super(_LayerCodeCompletion, self).__init__(*args, **kwargs) + if True: + return + self.config = None + """@type: libmproxy.proxy.config.ProxyConfig""" + self.client_conn = None + """@type: libmproxy.proxy.connection.ClientConnection""" + self.channel = None + """@type: libmproxy.controller.Channel""" + + +class Layer(_LayerCodeCompletion): + def __init__(self, ctx, *args, **kwargs): + """ + Args: + ctx: The (read-only) higher layer. + """ + super(Layer, self).__init__(*args, **kwargs) + self.ctx = ctx + + def __call__(self): + """ + Logic of the layer. + Raises: + ProtocolException in case of protocol exceptions. + """ + raise NotImplementedError + + def __getattr__(self, name): + """ + Attributes not present on the current layer may exist on a higher layer. + """ + return getattr(self.ctx, name) + + def log(self, msg, level, subs=()): + full_msg = [ + "{}: {}".format(repr(self.client_conn.address), msg) + ] + for i in subs: + full_msg.append(" -> " + i) + full_msg = "\n".join(full_msg) + self.channel.tell("log", Log(full_msg, level)) + + @property + def layers(self): + return [self] + self.ctx.layers + + def __repr__(self): + return type(self).__name__ + + +class ServerConnectionMixin(object): + """ + Mixin that provides a layer with the capabilities to manage a server connection. + """ + + def __init__(self, server_address=None): + super(ServerConnectionMixin, self).__init__() + self.server_conn = ServerConnection(server_address) + + def reconnect(self): + address = self.server_conn.address + self._disconnect() + self.server_conn.address = address + self.connect() + + def set_server(self, address, server_tls=None, sni=None, depth=1): + if depth == 1: + if self.server_conn: + self._disconnect() + self.log("Set new server address: " + repr(address), "debug") + self.server_conn.address = address + else: + self.ctx.set_server(address, server_tls, sni, depth - 1) + + def _disconnect(self): + """ + Deletes (and closes) an existing server connection. + """ + self.log("serverdisconnect", "debug", [repr(self.server_conn.address)]) + self.server_conn.finish() + self.server_conn.close() + # self.channel.tell("serverdisconnect", self) + self.server_conn = ServerConnection(None) + + def connect(self): + if not self.server_conn.address: + raise ProtocolException("Cannot connect to server, no server address given.") + self.log("serverconnect", "debug", [repr(self.server_conn.address)]) + try: + self.server_conn.connect() + except tcp.NetLibError as e: + raise ProtocolException( + "Server connection to '%s' failed: %s" % (self.server_conn.address, e), e) diff --git a/libmproxy/protocol2/rawtcp.py b/libmproxy/protocol2/rawtcp.py new file mode 100644 index 00000000..b10217f1 --- /dev/null +++ b/libmproxy/protocol2/rawtcp.py @@ -0,0 +1,66 @@ +from __future__ import (absolute_import, print_function, division) +import socket +import select + +from OpenSSL import SSL + +from netlib.tcp import NetLibError +from netlib.utils import cleanBin +from ..exceptions import ProtocolException +from .layer import Layer + + +class RawTcpLayer(Layer): + chunk_size = 4096 + + def __init__(self, ctx, logging=True): + self.logging = logging + super(RawTcpLayer, self).__init__(ctx) + + def __call__(self): + self.connect() + + buf = memoryview(bytearray(self.chunk_size)) + + client = self.client_conn.connection + server = self.server_conn.connection + conns = [client, server] + + try: + while True: + r, _, _ = select.select(conns, [], [], 10) + for conn in r: + dst = server if conn == client else client + + size = conn.recv_into(buf, self.chunk_size) + if not size: + conns.remove(conn) + # Shutdown connection to the other peer + if isinstance(conn, SSL.Connection): + # We can't half-close a connection, so we just close everything here. + # Sockets will be cleaned up on a higher level. + return + else: + dst.shutdown(socket.SHUT_WR) + + if len(conns) == 0: + return + continue + + dst.sendall(buf[:size]) + + if self.logging: + # log messages are prepended with the client address, + # hence the "weird" direction string. + if dst == server: + direction = "-> tcp -> {}".format(repr(self.server_conn.address)) + else: + direction = "<- tcp <- {}".format(repr(self.server_conn.address)) + data = cleanBin(buf[:size].tobytes()) + self.log( + "{}\r\n{}".format(direction, data), + "info" + ) + + except (socket.error, NetLibError, SSL.Error) as e: + raise ProtocolException("TCP connection closed unexpectedly: {}".format(repr(e)), e) diff --git a/libmproxy/protocol2/reverse_proxy.py b/libmproxy/protocol2/reverse_proxy.py new file mode 100644 index 00000000..c4cabccc --- /dev/null +++ b/libmproxy/protocol2/reverse_proxy.py @@ -0,0 +1,18 @@ +from __future__ import (absolute_import, print_function, division) + +from .layer import Layer, ServerConnectionMixin +from .tls import TlsLayer + + +class ReverseProxy(Layer, ServerConnectionMixin): + def __init__(self, ctx, server_address, server_tls): + super(ReverseProxy, self).__init__(ctx, server_address=server_address) + self.server_tls = server_tls + + def __call__(self): + layer = self.ctx.next_layer(self) + try: + layer() + finally: + if self.server_conn: + self._disconnect() diff --git a/libmproxy/protocol2/root_context.py b/libmproxy/protocol2/root_context.py new file mode 100644 index 00000000..210ba6ab --- /dev/null +++ b/libmproxy/protocol2/root_context.py @@ -0,0 +1,94 @@ +from __future__ import (absolute_import, print_function, division) + +from netlib.http.http1 import HTTP1Protocol +from netlib.http.http2 import HTTP2Protocol + +from .rawtcp import RawTcpLayer +from .tls import TlsLayer, is_tls_record_magic +from .http import Http1Layer, Http2Layer +from .layer import ServerConnectionMixin +from .http_proxy import HttpProxy, HttpUpstreamProxy +from .reverse_proxy import ReverseProxy + +class RootContext(object): + """ + The outmost context provided to the root layer. + As a consequence, every layer has .client_conn, .channel, .next_layer() and .config. + """ + + def __init__(self, client_conn, config, channel): + self.client_conn = client_conn # Client Connection + self.channel = channel # provides .ask() method to communicate with FlowMaster + self.config = config # Proxy Configuration + + def next_layer(self, top_layer): + """ + This function determines the next layer in the protocol stack. + + Arguments: + top_layer: the current top layer. + + Returns: + The next layer + """ + + # 1. Check for --ignore. + if self.config.check_ignore(top_layer.server_conn.address): + return RawTcpLayer(top_layer, logging=False) + + d = top_layer.client_conn.rfile.peek(3) + client_tls = is_tls_record_magic(d) + + # 2. Always insert a TLS layer, even if there's neither client nor server tls. + # An inline script may upgrade from http to https, + # in which case we need some form of TLS layer. + if isinstance(top_layer, ReverseProxy): + return TlsLayer(top_layer, client_tls, top_layer.server_tls) + if isinstance(top_layer, ServerConnectionMixin): + return TlsLayer(top_layer, client_tls, client_tls) + + # 3. In Http Proxy mode and Upstream Proxy mode, the next layer is fixed. + if isinstance(top_layer, TlsLayer): + if isinstance(top_layer.ctx, HttpProxy): + return Http1Layer(top_layer, "regular") + if isinstance(top_layer.ctx, HttpUpstreamProxy): + return Http1Layer(top_layer, "upstream") + + # 4. Check for other TLS cases (e.g. after CONNECT). + if client_tls: + return TlsLayer(top_layer, True, True) + + # 4. Check for --tcp + if self.config.check_tcp(top_layer.server_conn.address): + return RawTcpLayer(top_layer) + + # 5. Check for TLS ALPN (HTTP1/HTTP2) + if isinstance(top_layer, TlsLayer): + alpn = top_layer.client_conn.get_alpn_proto_negotiated() + if alpn == HTTP2Protocol.ALPN_PROTO_H2: + return Http2Layer(top_layer, 'transparent') + if alpn == HTTP1Protocol.ALPN_PROTO_HTTP1: + return Http1Layer(top_layer, 'transparent') + + # 6. Assume HTTP1 by default + return Http1Layer(top_layer, 'transparent') + + # In a future version, we want to implement TCP passthrough as the last fallback, + # but we don't have the UI part ready for that. + # + # d = top_layer.client_conn.rfile.peek(3) + # is_ascii = ( + # len(d) == 3 and + # # better be safe here and don't expect uppercase... + # all(x in string.ascii_letters for x in d) + # ) + # # TODO: This could block if there are not enough bytes available? + # d = top_layer.client_conn.rfile.peek(len(HTTP2Protocol.CLIENT_CONNECTION_PREFACE)) + # is_http2_magic = (d == HTTP2Protocol.CLIENT_CONNECTION_PREFACE) + + @property + def layers(self): + return [] + + def __repr__(self): + return "RootContext" diff --git a/libmproxy/protocol2/socks_proxy.py b/libmproxy/protocol2/socks_proxy.py new file mode 100644 index 00000000..525520e8 --- /dev/null +++ b/libmproxy/protocol2/socks_proxy.py @@ -0,0 +1,59 @@ +from __future__ import (absolute_import, print_function, division) + +from netlib import socks +from netlib.tcp import NetLibError +from ..exceptions import Socks5Exception +from .layer import Layer, ServerConnectionMixin + + +class Socks5Proxy(Layer, ServerConnectionMixin): + def __call__(self): + try: + # Parse Client Greeting + client_greet = socks.ClientGreeting.from_file(self.client_conn.rfile, fail_early=True) + client_greet.assert_socks5() + if socks.METHOD.NO_AUTHENTICATION_REQUIRED not in client_greet.methods: + raise socks.SocksError( + socks.METHOD.NO_ACCEPTABLE_METHODS, + "mitmproxy only supports SOCKS without authentication" + ) + + # Send Server Greeting + server_greet = socks.ServerGreeting( + socks.VERSION.SOCKS5, + socks.METHOD.NO_AUTHENTICATION_REQUIRED + ) + server_greet.to_file(self.client_conn.wfile) + self.client_conn.wfile.flush() + + # Parse Connect Request + connect_request = socks.Message.from_file(self.client_conn.rfile) + connect_request.assert_socks5() + if connect_request.msg != socks.CMD.CONNECT: + raise socks.SocksError( + socks.REP.COMMAND_NOT_SUPPORTED, + "mitmproxy only supports SOCKS5 CONNECT." + ) + + # We always connect lazily, but we need to pretend to the client that we connected. + connect_reply = socks.Message( + socks.VERSION.SOCKS5, + socks.REP.SUCCEEDED, + connect_request.atyp, + # dummy value, we don't have an upstream connection yet. + connect_request.addr + ) + connect_reply.to_file(self.client_conn.wfile) + self.client_conn.wfile.flush() + + except (socks.SocksError, NetLibError) as e: + raise Socks5Exception("SOCKS5 mode failure: %s" % repr(e), e) + + self.server_conn.address = connect_request.addr + + layer = self.ctx.next_layer(self) + try: + layer() + finally: + if self.server_conn: + self._disconnect() diff --git a/libmproxy/protocol2/tls.py b/libmproxy/protocol2/tls.py new file mode 100644 index 00000000..73bb12f3 --- /dev/null +++ b/libmproxy/protocol2/tls.py @@ -0,0 +1,288 @@ +from __future__ import (absolute_import, print_function, division) + +import struct + +from construct import ConstructError + +from netlib.tcp import NetLibError, NetLibInvalidCertificateError +from netlib.http.http1 import HTTP1Protocol +from ..contrib.tls._constructs import ClientHello +from ..exceptions import ProtocolException +from .layer import Layer + + +def is_tls_record_magic(d): + """ + Returns: + True, if the passed bytes start with the TLS record magic bytes. + False, otherwise. + """ + d = d[:3] + + # TLS ClientHello magic, works for SSLv3, TLSv1.0, TLSv1.1, TLSv1.2 + # http://www.moserware.com/2009/06/first-few-milliseconds-of-https.html#client-hello + return ( + len(d) == 3 and + d[0] == '\x16' and + d[1] == '\x03' and + d[2] in ('\x00', '\x01', '\x02', '\x03') + ) + + +class TlsLayer(Layer): + def __init__(self, ctx, client_tls, server_tls): + self.client_sni = None + self.client_alpn_protocols = None + + super(TlsLayer, self).__init__(ctx) + self._client_tls = client_tls + self._server_tls = server_tls + + self._sni_from_server_change = None + + def __call__(self): + """ + The strategy for establishing SSL is as follows: + First, we determine whether we need the server cert to establish ssl with the client. + If so, we first connect to the server and then to the client. + If not, we only connect to the client and do the server_ssl lazily on a Connect message. + + An additional complexity is that establish ssl with the server may require a SNI value from the client. + In an ideal world, we'd do the following: + 1. Start the SSL handshake with the client + 2. Check if the client sends a SNI. + 3. Pause the client handshake, establish SSL with the server. + 4. Finish the client handshake with the certificate from the server. + There's just one issue: We cannot get a callback from OpenSSL if the client doesn't send a SNI. :( + Thus, we manually peek into the connection and parse the ClientHello message to obtain both SNI and ALPN values. + + Further notes: + - OpenSSL 1.0.2 introduces a callback that would help here: + https://www.openssl.org/docs/ssl/SSL_CTX_set_cert_cb.html + - The original mitmproxy issue is https://github.com/mitmproxy/mitmproxy/issues/427 + """ + + client_tls_requires_server_cert = ( + self._client_tls and self._server_tls and not self.config.no_upstream_cert + ) + + if self._client_tls: + self._parse_client_hello() + + if client_tls_requires_server_cert: + self._establish_tls_with_client_and_server() + elif self._client_tls: + self._establish_tls_with_client() + + layer = self.ctx.next_layer(self) + layer() + + def __repr__(self): + if self._client_tls and self._server_tls: + return "TlsLayer(client and server)" + elif self._client_tls: + return "TlsLayer(client)" + elif self._server_tls: + return "TlsLayer(server)" + else: + return "TlsLayer(inactive)" + + def _get_client_hello(self): + """ + Peek into the socket and read all records that contain the initial client hello message. + + Returns: + The raw handshake packet bytes, without TLS record header(s). + """ + client_hello = "" + client_hello_size = 1 + offset = 0 + while len(client_hello) < client_hello_size: + record_header = self.client_conn.rfile.peek(offset + 5)[offset:] + if not is_tls_record_magic(record_header) or len(record_header) != 5: + raise ProtocolException('Expected TLS record, got "%s" instead.' % record_header) + record_size = struct.unpack("!H", record_header[3:])[0] + 5 + record_body = self.client_conn.rfile.peek(offset + record_size)[offset + 5:] + if len(record_body) != record_size - 5: + raise ProtocolException("Unexpected EOF in TLS handshake: %s" % record_body) + client_hello += record_body + offset += record_size + client_hello_size = struct.unpack("!I", '\x00' + client_hello[1:4])[0] + 4 + return client_hello + + def _parse_client_hello(self): + """ + Peek into the connection, read the initial client hello and parse it to obtain ALPN values. + """ + try: + raw_client_hello = self._get_client_hello()[4:] # exclude handshake header. + except ProtocolException as e: + self.log("Cannot parse Client Hello: %s" % repr(e), "error") + return + + try: + client_hello = ClientHello.parse(raw_client_hello) + except ConstructError as e: + self.log("Cannot parse Client Hello: %s" % repr(e), "error") + self.log("Raw Client Hello:\r\n:%s" % raw_client_hello.encode("hex"), "debug") + return + + for extension in client_hello.extensions: + if extension.type == 0x00: + if len(extension.server_names) != 1 or extension.server_names[0].type != 0: + self.log("Unknown Server Name Indication: %s" % extension.server_names, "error") + self.client_sni = extension.server_names[0].name + elif extension.type == 0x10: + self.client_alpn_protocols = list(extension.alpn_protocols) + + self.log( + "Parsed Client Hello: sni=%s, alpn=%s" % (self.client_sni, self.client_alpn_protocols), + "debug" + ) + + def connect(self): + if not self.server_conn: + self.ctx.connect() + if self._server_tls and not self.server_conn.tls_established: + self._establish_tls_with_server() + + def reconnect(self): + self.ctx.reconnect() + if self._server_tls and not self.server_conn.tls_established: + self._establish_tls_with_server() + + def set_server(self, address, server_tls=None, sni=None, depth=1): + self.ctx.set_server(address, server_tls, sni, depth) + if depth == 1 and server_tls is not None: + self._sni_from_server_change = sni + self._server_tls = server_tls + + @property + def sni_for_server_connection(self): + if self._sni_from_server_change is False: + return None + else: + return self._sni_from_server_change or self.client_sni + + @property + def alpn_for_client_connection(self): + return self.server_conn.get_alpn_proto_negotiated() + + def __alpn_select_callback(self, conn_, options): + """ + Once the client signals the alternate protocols it supports, + we reconnect upstream with the same list and pass the server's choice down to the client. + """ + + # This gets triggered if we haven't established an upstream connection yet. + default_alpn = HTTP1Protocol.ALPN_PROTO_HTTP1 + # alpn_preference = netlib.http.http2.HTTP2Protocol.ALPN_PROTO_H2 + + if self.alpn_for_client_connection in options: + choice = bytes(self.alpn_for_client_connection) + elif default_alpn in options: + choice = bytes(default_alpn) + else: + choice = options[0] + self.log("ALPN for client: %s" % choice, "debug") + return choice + + def _establish_tls_with_client_and_server(self): + self.ctx.connect() + + # If establishing TLS with the server fails, we try to establish TLS with the client nonetheless + # to send an error message over TLS. + try: + self._establish_tls_with_server() + except Exception as e: + try: + self._establish_tls_with_client() + except: + pass + raise e + + self._establish_tls_with_client() + + def _establish_tls_with_client(self): + self.log("Establish TLS with client", "debug") + cert, key, chain_file = self._find_cert() + + try: + self.client_conn.convert_to_ssl( + cert, key, + method=self.config.openssl_method_client, + options=self.config.openssl_options_client, + cipher_list=self.config.ciphers_client, + dhparams=self.config.certstore.dhparams, + chain_file=chain_file, + alpn_select_callback=self.__alpn_select_callback, + ) + except NetLibError as e: + raise ProtocolException("Cannot establish TLS with client: %s" % repr(e), e) + + def _establish_tls_with_server(self): + self.log("Establish TLS with server", "debug") + try: + # We only support http/1.1 and h2. + # If the server only supports spdy (next to http/1.1), it may select that + # and mitmproxy would enter TCP passthrough mode, which we want to avoid. + deprecated_http2_variant = lambda x: x.startswith("h2-") or x.startswith("spdy") + if self.client_alpn_protocols: + alpn = filter(lambda x: not deprecated_http2_variant(x), self.client_alpn_protocols) + else: + alpn = None + + self.server_conn.establish_ssl( + self.config.clientcerts, + self.sni_for_server_connection, + method=self.config.openssl_method_server, + options=self.config.openssl_options_server, + verify_options=self.config.openssl_verification_mode_server, + ca_path=self.config.openssl_trusted_cadir_server, + ca_pemfile=self.config.openssl_trusted_ca_server, + cipher_list=self.config.ciphers_server, + alpn_protos=alpn, + ) + tls_cert_err = self.server_conn.ssl_verification_error + if tls_cert_err is not None: + self.log( + "TLS verification failed for upstream server at depth %s with error: %s" % + (tls_cert_err['depth'], tls_cert_err['errno']), + "error") + self.log("Ignoring server verification error, continuing with connection", "error") + except NetLibInvalidCertificateError as e: + tls_cert_err = self.server_conn.ssl_verification_error + self.log( + "TLS verification failed for upstream server at depth %s with error: %s" % + (tls_cert_err['depth'], tls_cert_err['errno']), + "error") + self.log("Aborting connection attempt", "error") + raise ProtocolException("Cannot establish TLS with server: %s" % repr(e), e) + except NetLibError as e: + raise ProtocolException("Cannot establish TLS with server: %s" % repr(e), e) + + self.log("ALPN selected by server: %s" % self.alpn_for_client_connection, "debug") + + def _find_cert(self): + host = self.server_conn.address.host + sans = set() + # Incorporate upstream certificate + use_upstream_cert = ( + self.server_conn and + self.server_conn.tls_established and + (not self.config.no_upstream_cert) + ) + if use_upstream_cert: + upstream_cert = self.server_conn.cert + sans.update(upstream_cert.altnames) + if upstream_cert.cn: + sans.add(host) + host = upstream_cert.cn.decode("utf8").encode("idna") + # Also add SNI values. + if self.client_sni: + sans.add(self.client_sni) + if self._sni_from_server_change: + sans.add(self._sni_from_server_change) + + sans.discard(host) + return self.config.certstore.get_cert(host, list(sans)) diff --git a/libmproxy/protocol2/transparent_proxy.py b/libmproxy/protocol2/transparent_proxy.py new file mode 100644 index 00000000..e6ebf115 --- /dev/null +++ b/libmproxy/protocol2/transparent_proxy.py @@ -0,0 +1,24 @@ +from __future__ import (absolute_import, print_function, division) + +from ..exceptions import ProtocolException +from .. import platform +from .layer import Layer, ServerConnectionMixin + + +class TransparentProxy(Layer, ServerConnectionMixin): + def __init__(self, ctx): + super(TransparentProxy, self).__init__(ctx) + self.resolver = platform.resolver() + + def __call__(self): + try: + self.server_conn.address = self.resolver.original_addr(self.client_conn.connection) + except Exception as e: + raise ProtocolException("Transparent mode failure: %s" % repr(e), e) + + layer = self.ctx.next_layer(self) + try: + layer() + finally: + if self.server_conn: + self._disconnect() |