diff options
Diffstat (limited to 'libmproxy')
| -rw-r--r-- | libmproxy/contentviews.py | 4 | ||||
| -rw-r--r-- | libmproxy/controller.py | 2 | ||||
| -rw-r--r-- | libmproxy/main.py | 6 | ||||
| -rw-r--r-- | libmproxy/models/http.py | 11 | ||||
| -rw-r--r-- | libmproxy/protocol/__init__.py | 8 | ||||
| -rw-r--r-- | libmproxy/protocol/base.py | 2 | ||||
| -rw-r--r-- | libmproxy/protocol/http.py | 234 | ||||
| -rw-r--r-- | libmproxy/protocol/http1.py | 70 | ||||
| -rw-r--r-- | libmproxy/protocol/http2.py | 414 | ||||
| -rw-r--r-- | libmproxy/protocol/tls.py | 2 | ||||
| -rw-r--r-- | libmproxy/proxy/server.py | 4 |
11 files changed, 539 insertions, 218 deletions
diff --git a/libmproxy/contentviews.py b/libmproxy/contentviews.py index 80955b0f..c0652c18 100644 --- a/libmproxy/contentviews.py +++ b/libmproxy/contentviews.py @@ -35,12 +35,12 @@ from .contrib.wbxml.ASCommandResponse import ASCommandResponse try: import pyamf from pyamf import remoting, flex -except ImportError: # pragma nocover +except ImportError: # pragma no cover pyamf = None try: import cssutils -except ImportError: # pragma nocover +except ImportError: # pragma no cover cssutils = None else: cssutils.log.setLevel(logging.CRITICAL) diff --git a/libmproxy/controller.py b/libmproxy/controller.py index 712ab1d2..9a059856 100644 --- a/libmproxy/controller.py +++ b/libmproxy/controller.py @@ -56,7 +56,7 @@ class Channel: try: # The timeout is here so we can handle a should_exit event. g = m.reply.q.get(timeout=0.5) - except Queue.Empty: # pragma: nocover + except Queue.Empty: # pragma: no cover continue return g diff --git a/libmproxy/main.py b/libmproxy/main.py index 655d573d..1c3cbf78 100644 --- a/libmproxy/main.py +++ b/libmproxy/main.py @@ -37,7 +37,7 @@ def get_server(dummy_server, options): sys.exit(1) -def mitmproxy(args=None): # pragma: nocover +def mitmproxy(args=None): # pragma: no cover from . import console check_pyopenssl_version() @@ -68,7 +68,7 @@ def mitmproxy(args=None): # pragma: nocover pass -def mitmdump(args=None): # pragma: nocover +def mitmdump(args=None): # pragma: no cover from . import dump check_pyopenssl_version() @@ -103,7 +103,7 @@ def mitmdump(args=None): # pragma: nocover pass -def mitmweb(args=None): # pragma: nocover +def mitmweb(args=None): # pragma: no cover from . import web check_pyopenssl_version() diff --git a/libmproxy/models/http.py b/libmproxy/models/http.py index e07dff69..8a0b226d 100644 --- a/libmproxy/models/http.py +++ b/libmproxy/models/http.py @@ -11,9 +11,14 @@ from netlib.tcp import Address from .. import version, stateobject from .flow import Flow +from collections import OrderedDict class MessageMixin(stateobject.StateObject): - _stateobject_attributes = dict( + # The restoration order is important currently, e.g. because + # of .content setting .headers["content-length"] automatically. + # Using OrderedDict is the short term fix, restoring state should + # be implemented without side-effects again. + _stateobject_attributes = OrderedDict( http_version=bytes, headers=Headers, timestamp_start=float, @@ -241,8 +246,6 @@ class HTTPRequest(MessageMixin, Request): timestamp_end=request.timestamp_end, form_out=(request.form_out if hasattr(request, 'form_out') else None), ) - if hasattr(request, 'stream_id'): - req.stream_id = request.stream_id return req def __hash__(self): @@ -347,8 +350,6 @@ class HTTPResponse(MessageMixin, Response): timestamp_start=response.timestamp_start, timestamp_end=response.timestamp_end, ) - if hasattr(response, 'stream_id'): - resp.stream_id = response.stream_id return resp def _refresh_cookie(self, c, delta): diff --git a/libmproxy/protocol/__init__.py b/libmproxy/protocol/__init__.py index d46f16f5..ea958d06 100644 --- a/libmproxy/protocol/__init__.py +++ b/libmproxy/protocol/__init__.py @@ -27,15 +27,19 @@ as late as possible; this makes server replay without any outgoing connections p from __future__ import (absolute_import, print_function, division) from .base import Layer, ServerConnectionMixin, Kill -from .http import Http1Layer, UpstreamConnectLayer, Http2Layer from .tls import TlsLayer from .tls import is_tls_record_magic from .tls import TlsClientHello +from .http import UpstreamConnectLayer +from .http1 import Http1Layer +from .http2 import Http2Layer from .rawtcp import RawTCPLayer __all__ = [ "Layer", "ServerConnectionMixin", "Kill", - "Http1Layer", "UpstreamConnectLayer", "Http2Layer", "TlsLayer", "is_tls_record_magic", "TlsClientHello", + "UpstreamConnectLayer", + "Http1Layer", + "Http2Layer", "RawTCPLayer", ] diff --git a/libmproxy/protocol/base.py b/libmproxy/protocol/base.py index 4eb034c0..40fcaf65 100644 --- a/libmproxy/protocol/base.py +++ b/libmproxy/protocol/base.py @@ -14,7 +14,7 @@ class _LayerCodeCompletion(object): Dummy class that provides type hinting in PyCharm, which simplifies development a lot. """ - def __init__(self, **mixin_args): # pragma: nocover + def __init__(self, **mixin_args): # pragma: no cover super(_LayerCodeCompletion, self).__init__(**mixin_args) if True: return diff --git a/libmproxy/protocol/http.py b/libmproxy/protocol/http.py index 12d09e71..f3240b85 100644 --- a/libmproxy/protocol/http.py +++ b/libmproxy/protocol/http.py @@ -1,26 +1,30 @@ from __future__ import (absolute_import, print_function, division) + import sys import traceback - import six from netlib import tcp from netlib.exceptions import HttpException, HttpReadDisconnect, NetlibException -from netlib.http import http1, Headers -from netlib.http import CONTENT_MISSING -from netlib.tcp import Address -from netlib.http.http2.connections import HTTP2Protocol -from netlib.http.http2.frame import GoAwayFrame, PriorityFrame, WindowUpdateFrame +from netlib.http import Headers, CONTENT_MISSING + +from h2.exceptions import H2Error + from .. import utils from ..exceptions import HttpProtocolException, ProtocolException from ..models import ( - HTTPFlow, HTTPRequest, HTTPResponse, make_error_response, make_connect_response, Error, expect_continue_response + HTTPFlow, + HTTPResponse, + make_error_response, + make_connect_response, + Error, + expect_continue_response ) + from .base import Layer, Kill -class _HttpLayer(Layer): - supports_streaming = False +class _HttpTransmissionLayer(Layer): def read_request(self): raise NotImplementedError() @@ -32,37 +36,18 @@ class _HttpLayer(Layer): raise NotImplementedError() def read_response(self, request): - raise NotImplementedError() - - def send_response(self, response): - raise NotImplementedError() - - def check_close_connection(self, flow): - raise NotImplementedError() - - -class _StreamingHttpLayer(_HttpLayer): - supports_streaming = True - - def read_response_headers(self): - raise NotImplementedError - - def read_response_body(self, request, response): - raise NotImplementedError() - yield "this is a generator" # pragma: no cover - - def read_response(self, request): response = self.read_response_headers() response.data.content = b"".join( self.read_response_body(request, response) ) return response - def send_response_headers(self, response): - raise NotImplementedError + def read_response_headers(self): + raise NotImplementedError() - def send_response_body(self, response, chunks): + def read_response_body(self, request, response): raise NotImplementedError() + yield "this is a generator" # pragma: no cover def send_response(self, response): if response.content == CONTENT_MISSING: @@ -70,164 +55,14 @@ class _StreamingHttpLayer(_HttpLayer): self.send_response_headers(response) self.send_response_body(response, [response.content]) - -class Http1Layer(_StreamingHttpLayer): - - def __init__(self, ctx, mode): - super(Http1Layer, self).__init__(ctx) - self.mode = mode - - def read_request(self): - req = http1.read_request(self.client_conn.rfile, body_size_limit=self.config.body_size_limit) - return HTTPRequest.wrap(req) - - def read_request_body(self, request): - expected_size = http1.expected_http_body_size(request) - return http1.read_body(self.client_conn.rfile, expected_size, self.config.body_size_limit) - - def send_request(self, request): - self.server_conn.wfile.write(http1.assemble_request(request)) - self.server_conn.wfile.flush() - - def read_response_headers(self): - resp = http1.read_response_head(self.server_conn.rfile) - return HTTPResponse.wrap(resp) - - def read_response_body(self, request, response): - expected_size = http1.expected_http_body_size(request, response) - return http1.read_body(self.server_conn.rfile, expected_size, self.config.body_size_limit) - def send_response_headers(self, response): - raw = http1.assemble_response_head(response) - self.client_conn.wfile.write(raw) - self.client_conn.wfile.flush() + raise NotImplementedError() def send_response_body(self, response, chunks): - for chunk in http1.assemble_body(response.headers, chunks): - self.client_conn.wfile.write(chunk) - self.client_conn.wfile.flush() - - def check_close_connection(self, flow): - request_close = http1.connection_close( - flow.request.http_version, - flow.request.headers - ) - response_close = http1.connection_close( - flow.response.http_version, - flow.response.headers - ) - read_until_eof = http1.expected_http_body_size(flow.request, flow.response) == -1 - close_connection = request_close or response_close or read_until_eof - if flow.request.form_in == "authority" and flow.response.status_code == 200: - # Workaround for https://github.com/mitmproxy/mitmproxy/issues/313: - # Charles Proxy sends a CONNECT response with HTTP/1.0 - # and no Content-Length header - - return False - return close_connection - - 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_from_client) - self.server_protocol = HTTP2Protocol(self.server_conn, is_server=False, - unhandled_frame_cb=self.handle_unexpected_frame_from_server) - - 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): - 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 to prevent client buffer filling up - # maintain a send buffer size, and read WindowUpdateFrames from client to increase the send buffer - self.client_conn.send(self.client_protocol.assemble(message)) + raise NotImplementedError() def check_close_connection(self, flow): - # TODO: add a timer to disconnect after a 10 second timeout - return False - - def connect(self): - self.ctx.connect() - self.server_protocol = HTTP2Protocol(self.server_conn, is_server=False, - unhandled_frame_cb=self.handle_unexpected_frame_from_server) - 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_from_server) - self.server_protocol.perform_connection_preface() - - def __call__(self): - self.server_protocol.perform_connection_preface() - layer = HttpLayer(self, self.mode) - layer() - - # terminate the connection - self.client_conn.send(GoAwayFrame().to_bytes()) - - def handle_unexpected_frame_from_client(self, frame): - if isinstance(frame, WindowUpdateFrame): - # Clients are sending WindowUpdate frames depending on their flow control algorithm. - # Since we cannot predict these frames, and we do not need to respond to them, - # simply accept them, and hide them from the log. - # Ideally we should keep track of our own flow control window and - # stall transmission if the outgoing flow control buffer is full. - return - if isinstance(frame, PriorityFrame): - # Clients are sending Priority frames depending on their implementation. - # The RFC does not clearly state when or which priority preferences should be set. - # Since we cannot predict these frames, and we do not need to respond to them, - # simply accept them, and hide them from the log. - # Ideally we should forward them to the server. - return - if isinstance(frame, GoAwayFrame): - # Client wants to terminate the connection, - # relay it to the server. - self.server_conn.send(frame.to_bytes()) - return - self.log("Unexpected HTTP2 frame from client: %s" % frame.human_readable(), "info") - - def handle_unexpected_frame_from_server(self, frame): - if isinstance(frame, WindowUpdateFrame): - # Servers are sending WindowUpdate frames depending on their flow control algorithm. - # Since we cannot predict these frames, and we do not need to respond to them, - # simply accept them, and hide them from the log. - # Ideally we should keep track of our own flow control window and - # stall transmission if the outgoing flow control buffer is full. - return - if isinstance(frame, GoAwayFrame): - # Server wants to terminate the connection, - # relay it to the client. - self.client_conn.send(frame.to_bytes()) - return - self.log("Unexpected HTTP2 frame from server: %s" % frame.human_readable(), "info") + raise NotImplementedError() class ConnectServerConnection(object): @@ -285,7 +120,7 @@ class UpstreamConnectLayer(Layer): def set_server(self, address, server_tls=None, sni=None): if self.ctx.server_conn: self.ctx.disconnect() - address = Address.wrap(address) + address = tcp.Address.wrap(address) self.connect_request.host = address.host self.connect_request.port = address.port self.server_conn.address = address @@ -400,7 +235,8 @@ class HttpLayer(Layer): try: response = make_error_response(code, message) self.send_response(response) - except NetlibException: + except (NetlibException, H2Error): + self.log(traceback.format_exc(), "debug") pass def change_upstream_proxy_server(self, address): @@ -420,7 +256,7 @@ class HttpLayer(Layer): layer() def send_response_to_client(self, flow): - if not (self.supports_streaming and flow.response.stream): + if not flow.response.stream: # no streaming: # we already received the full response from the server and can # send it to the client straight away. @@ -441,10 +277,7 @@ class HttpLayer(Layer): 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(flow.request) + flow.response = self.read_response_headers() try: get_response() @@ -474,15 +307,14 @@ class HttpLayer(Layer): if flow == Kill: raise Kill() - if self.supports_streaming: - if flow.response.stream: - flow.response.data.content = CONTENT_MISSING - else: - flow.response.data.content = b"".join(self.read_response_body( - flow.request, - flow.response - )) - flow.response.timestamp_end = utils.timestamp() + if flow.response.stream: + flow.response.data.content = CONTENT_MISSING + else: + flow.response.data.content = b"".join(self.read_response_body( + flow.request, + flow.response + )) + 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. diff --git a/libmproxy/protocol/http1.py b/libmproxy/protocol/http1.py new file mode 100644 index 00000000..fc2cf07a --- /dev/null +++ b/libmproxy/protocol/http1.py @@ -0,0 +1,70 @@ +from __future__ import (absolute_import, print_function, division) + +import six + +from netlib import tcp +from netlib.http import http1 + +from .http import _HttpTransmissionLayer, HttpLayer +from .. import utils +from ..models import HTTPRequest, HTTPResponse + + +class Http1Layer(_HttpTransmissionLayer): + + def __init__(self, ctx, mode): + super(Http1Layer, self).__init__(ctx) + self.mode = mode + + def read_request(self): + req = http1.read_request(self.client_conn.rfile, body_size_limit=self.config.body_size_limit) + return HTTPRequest.wrap(req) + + def read_request_body(self, request): + expected_size = http1.expected_http_body_size(request) + return http1.read_body(self.client_conn.rfile, expected_size, self.config.body_size_limit) + + def send_request(self, request): + self.server_conn.wfile.write(http1.assemble_request(request)) + self.server_conn.wfile.flush() + + def read_response_headers(self): + resp = http1.read_response_head(self.server_conn.rfile) + return HTTPResponse.wrap(resp) + + def read_response_body(self, request, response): + expected_size = http1.expected_http_body_size(request, response) + return http1.read_body(self.server_conn.rfile, expected_size, self.config.body_size_limit) + + def send_response_headers(self, response): + raw = http1.assemble_response_head(response) + self.client_conn.wfile.write(raw) + self.client_conn.wfile.flush() + + def send_response_body(self, response, chunks): + for chunk in http1.assemble_body(response.headers, chunks): + self.client_conn.wfile.write(chunk) + self.client_conn.wfile.flush() + + def check_close_connection(self, flow): + request_close = http1.connection_close( + flow.request.http_version, + flow.request.headers + ) + response_close = http1.connection_close( + flow.response.http_version, + flow.response.headers + ) + read_until_eof = http1.expected_http_body_size(flow.request, flow.response) == -1 + close_connection = request_close or response_close or read_until_eof + if flow.request.form_in == "authority" and flow.response.status_code == 200: + # Workaround for https://github.com/mitmproxy/mitmproxy/issues/313: + # Charles Proxy sends a CONNECT response with HTTP/1.0 + # and no Content-Length header + + return False + return close_connection + + def __call__(self): + layer = HttpLayer(self, self.mode) + layer() diff --git a/libmproxy/protocol/http2.py b/libmproxy/protocol/http2.py new file mode 100644 index 00000000..de068836 --- /dev/null +++ b/libmproxy/protocol/http2.py @@ -0,0 +1,414 @@ +from __future__ import (absolute_import, print_function, division) + +import threading +import time +import Queue + +from netlib.tcp import ssl_read_select +from netlib.exceptions import HttpException +from netlib.http import Headers +from netlib.utils import http2_read_raw_frame + +import h2 +from h2.connection import H2Connection +from h2.events import * + +from .base import Layer +from .http import _HttpTransmissionLayer, HttpLayer +from .. import utils +from ..models import HTTPRequest, HTTPResponse +from ..exceptions import HttpProtocolException +from ..exceptions import ProtocolException + + +class SafeH2Connection(H2Connection): + + def __init__(self, conn, *args, **kwargs): + super(SafeH2Connection, self).__init__(*args, **kwargs) + self.conn = conn + self.lock = threading.RLock() + + def safe_close_connection(self, error_code): + with self.lock: + self.close_connection(error_code) + self.conn.send(self.data_to_send()) + + def safe_increment_flow_control(self, stream_id, length): + if length == 0: + return + + with self.lock: + self.increment_flow_control_window(length) + self.conn.send(self.data_to_send()) + with self.lock: + if stream_id in self.streams and not self.streams[stream_id].closed: + self.increment_flow_control_window(length, stream_id=stream_id) + self.conn.send(self.data_to_send()) + + def safe_reset_stream(self, stream_id, error_code): + with self.lock: + try: + self.reset_stream(stream_id, error_code) + except h2.exceptions.StreamClosedError: + # stream is already closed - good + pass + self.conn.send(self.data_to_send()) + + def safe_update_settings(self, new_settings): + with self.lock: + self.update_settings(new_settings) + self.conn.send(self.data_to_send()) + + def safe_send_headers(self, is_zombie, stream_id, headers): + with self.lock: + if is_zombie(): + return + self.send_headers(stream_id, headers) + self.conn.send(self.data_to_send()) + + def safe_send_body(self, is_zombie, stream_id, chunks): + for chunk in chunks: + position = 0 + while position < len(chunk): + self.lock.acquire() + if is_zombie(): + self.lock.release() + return + max_outbound_frame_size = self.max_outbound_frame_size + frame_chunk = chunk[position:position + max_outbound_frame_size] + if self.local_flow_control_window(stream_id) < len(frame_chunk): + self.lock.release() + time.sleep(0) + continue + self.send_data(stream_id, frame_chunk) + self.conn.send(self.data_to_send()) + self.lock.release() + position += max_outbound_frame_size + with self.lock: + if is_zombie(): + return + self.end_stream(stream_id) + self.conn.send(self.data_to_send()) + + +class Http2Layer(Layer): + + def __init__(self, ctx, mode): + super(Http2Layer, self).__init__(ctx) + self.mode = mode + self.streams = dict() + self.server_to_client_stream_ids = dict([(0, 0)]) + self.client_conn.h2 = SafeH2Connection(self.client_conn, client_side=False) + + # make sure that we only pass actual SSL.Connection objects in here, + # because otherwise ssl_read_select fails! + self.active_conns = [self.client_conn.connection] + + def _initiate_server_conn(self): + self.server_conn.h2 = SafeH2Connection(self.server_conn, client_side=True) + self.server_conn.h2.initiate_connection() + self.server_conn.send(self.server_conn.h2.data_to_send()) + self.active_conns.append(self.server_conn.connection) + + def connect(self): # pragma: no cover + raise ValueError("CONNECT inside an HTTP2 stream is not supported.") + # self.ctx.connect() + # self.server_conn.connect() + # self._initiate_server_conn() + + def set_server(self): # pragma: no cover + raise NotImplementedError("Cannot change server for HTTP2 connections.") + + def disconnect(self): # pragma: no cover + raise NotImplementedError("Cannot dis- or reconnect in HTTP2 connections.") + + def next_layer(self): # pragma: no cover + # WebSockets over HTTP/2? + # CONNECT for proxying? + raise NotImplementedError() + + def _handle_event(self, event, source_conn, other_conn, is_server): + if hasattr(event, 'stream_id'): + if is_server and event.stream_id % 2 == 1: + eid = self.server_to_client_stream_ids[event.stream_id] + else: + eid = event.stream_id + + if isinstance(event, RequestReceived): + headers = Headers([[str(k), str(v)] for k, v in event.headers]) + self.streams[eid] = Http2SingleStreamLayer(self, eid, headers) + self.streams[eid].timestamp_start = time.time() + self.streams[eid].start() + elif isinstance(event, ResponseReceived): + headers = Headers([[str(k), str(v)] for k, v in event.headers]) + self.streams[eid].queued_data_length = 0 + self.streams[eid].timestamp_start = time.time() + self.streams[eid].response_headers = headers + self.streams[eid].response_arrived.set() + elif isinstance(event, DataReceived): + if self.config.body_size_limit and self.streams[eid].queued_data_length > self.config.body_size_limit: + raise HttpException("HTTP body too large. Limit is {}.".format(self.config.body_size_limit)) + self.streams[eid].data_queue.put(event.data) + self.streams[eid].queued_data_length += len(event.data) + source_conn.h2.safe_increment_flow_control(event.stream_id, event.flow_controlled_length) + elif isinstance(event, StreamEnded): + self.streams[eid].timestamp_end = time.time() + self.streams[eid].data_finished.set() + elif isinstance(event, StreamReset): + self.streams[eid].zombie = time.time() + if eid in self.streams and event.error_code == 0x8: + if is_server: + other_stream_id = self.streams[eid].client_stream_id + else: + other_stream_id = self.streams[eid].server_stream_id + if other_stream_id is not None: + other_conn.h2.safe_reset_stream(other_stream_id, event.error_code) + elif isinstance(event, RemoteSettingsChanged): + new_settings = dict([(id, cs.new_value) for (id, cs) in event.changed_settings.iteritems()]) + other_conn.h2.safe_update_settings(new_settings) + elif isinstance(event, ConnectionTerminated): + # Do not immediately terminate the other connection. + # Some streams might be still sending data to the client. + return False + elif isinstance(event, PushedStreamReceived): + # pushed stream ids should be uniq and not dependent on race conditions + # only the parent stream id must be looked up first + parent_eid = self.server_to_client_stream_ids[event.parent_stream_id] + with self.client_conn.h2.lock: + self.client_conn.h2.push_stream(parent_eid, event.pushed_stream_id, event.headers) + + headers = Headers([[str(k), str(v)] for k, v in event.headers]) + headers['x-mitmproxy-pushed'] = 'true' + self.streams[event.pushed_stream_id] = Http2SingleStreamLayer(self, event.pushed_stream_id, headers) + self.streams[event.pushed_stream_id].timestamp_start = time.time() + self.streams[event.pushed_stream_id].pushed = True + self.streams[event.pushed_stream_id].parent_stream_id = parent_eid + self.streams[event.pushed_stream_id].timestamp_end = time.time() + self.streams[event.pushed_stream_id].request_data_finished.set() + self.streams[event.pushed_stream_id].start() + elif isinstance(event, TrailersReceived): + raise NotImplementedError() + + return True + + def _cleanup_streams(self): + death_time = time.time() - 10 + for stream_id in self.streams.keys(): + zombie = self.streams[stream_id].zombie + if zombie and zombie <= death_time: + self.streams.pop(stream_id, None) + + def __call__(self): + if self.server_conn: + self._initiate_server_conn() + + preamble = self.client_conn.rfile.read(24) + self.client_conn.h2.initiate_connection() + self.client_conn.h2.receive_data(preamble) + self.client_conn.send(self.client_conn.h2.data_to_send()) + + while True: + r = ssl_read_select(self.active_conns, 1) + for conn in r: + source_conn = self.client_conn if conn == self.client_conn.connection else self.server_conn + other_conn = self.server_conn if conn == self.client_conn.connection else self.client_conn + is_server = (conn == self.server_conn.connection) + + with source_conn.h2.lock: + try: + raw_frame = b''.join(http2_read_raw_frame(source_conn.rfile)) + except: + for stream in self.streams.values(): + stream.zombie = time.time() + return + + events = source_conn.h2.receive_data(raw_frame) + source_conn.send(source_conn.h2.data_to_send()) + + for event in events: + if not self._handle_event(event, source_conn, other_conn, is_server): + return + + self._cleanup_streams() + + +class Http2SingleStreamLayer(_HttpTransmissionLayer, threading.Thread): + + def __init__(self, ctx, stream_id, request_headers): + super(Http2SingleStreamLayer, self).__init__(ctx) + self.zombie = None + self.client_stream_id = stream_id + self.server_stream_id = None + self.request_headers = request_headers + self.response_headers = None + self.pushed = False + + self.request_data_queue = Queue.Queue() + self.request_queued_data_length = 0 + self.request_data_finished = threading.Event() + + self.response_arrived = threading.Event() + self.response_data_queue = Queue.Queue() + self.response_queued_data_length = 0 + self.response_data_finished = threading.Event() + + @property + def data_queue(self): + if self.response_arrived.is_set(): + return self.response_data_queue + else: + return self.request_data_queue + + @property + def queued_data_length(self): + if self.response_arrived.is_set(): + return self.response_queued_data_length + else: + return self.request_queued_data_length + + @property + def data_finished(self): + if self.response_arrived.is_set(): + return self.response_data_finished + else: + return self.request_data_finished + + @queued_data_length.setter + def queued_data_length(self, v): + if self.response_arrived.is_set(): + return self.response_queued_data_length + else: + return self.request_queued_data_length + + def is_zombie(self): + return self.zombie is not None + + def read_request(self): + self.request_data_finished.wait() + + authority = self.request_headers.get(':authority', '') + method = self.request_headers.get(':method', 'GET') + scheme = self.request_headers.get(':scheme', 'https') + path = self.request_headers.get(':path', '/') + host = None + port = None + + if path == '*' or path.startswith("/"): + form_in = "relative" + elif method == 'CONNECT': # pragma: no cover + raise NotImplementedError("CONNECT over HTTP/2 is not implemented.") + else: # pragma: no cover + form_in = "absolute" + # FIXME: verify if path or :host contains what we need + scheme, host, port, _ = utils.parse_url(path) + + if authority: + host, _, port = authority.partition(':') + + if not host: + host = 'localhost' + if not port: + port = 443 if scheme == 'https' else 80 + port = int(port) + + data = [] + while self.request_data_queue.qsize() > 0: + data.append(self.request_data_queue.get()) + data = b"".join(data) + + return HTTPRequest( + form_in, + method, + scheme, + host, + port, + path, + (2, 0), + self.request_headers, + data, + timestamp_start=self.timestamp_start, + timestamp_end=self.timestamp_end, + ) + + def send_request(self, message): + if self.pushed: + # nothing to do here + return + + with self.server_conn.h2.lock: + # We must not assign a stream id if we are already a zombie. + if self.zombie: + return + + self.server_stream_id = self.server_conn.h2.get_next_available_stream_id() + self.server_to_client_stream_ids[self.server_stream_id] = self.client_stream_id + + self.server_conn.h2.safe_send_headers( + self.is_zombie, + self.server_stream_id, + message.headers + ) + self.server_conn.h2.safe_send_body( + self.is_zombie, + self.server_stream_id, + message.body + ) + + def read_response_headers(self): + self.response_arrived.wait() + + status_code = int(self.response_headers.get(':status', 502)) + + return HTTPResponse( + http_version=(2, 0), + status_code=status_code, + reason='', + headers=self.response_headers, + content=None, + timestamp_start=self.timestamp_start, + timestamp_end=self.timestamp_end, + ) + + def read_response_body(self, request, response): + while True: + try: + yield self.response_data_queue.get(timeout=1) + except Queue.Empty: + pass + if self.response_data_finished.is_set(): + while self.response_data_queue.qsize() > 0: + yield self.response_data_queue.get() + return + if self.zombie: + return + + def send_response_headers(self, response): + self.client_conn.h2.safe_send_headers( + self.is_zombie, + self.client_stream_id, + response.headers + ) + + def send_response_body(self, _response, chunks): + self.client_conn.h2.safe_send_body( + self.is_zombie, + self.client_stream_id, + chunks + ) + + def check_close_connection(self, flow): + # This layer only handles a single stream. + # RFC 7540 8.1: An HTTP request/response exchange fully consumes a single stream. + return True + + def connect(self): # pragma: no cover + raise ValueError("CONNECT inside an HTTP2 stream is not supported.") + + def set_server(self, *args, **kwargs): # pragma: no cover + # do not mess with the server connection - all streams share it. + pass + + def run(self): + layer = HttpLayer(self, self.mode) + layer() + self.zombie = time.time() diff --git a/libmproxy/protocol/tls.py b/libmproxy/protocol/tls.py index af1a6055..ccae1661 100644 --- a/libmproxy/protocol/tls.py +++ b/libmproxy/protocol/tls.py @@ -349,7 +349,7 @@ class TlsLayer(Layer): layer = self.ctx.next_layer(self) layer() - def __repr__(self): + def __repr__(self): # pragma: no cover if self._client_tls and self._server_tls: return "TlsLayer(client and server)" elif self._client_tls: diff --git a/libmproxy/proxy/server.py b/libmproxy/proxy/server.py index 750cb1a4..d208cff5 100644 --- a/libmproxy/proxy/server.py +++ b/libmproxy/proxy/server.py @@ -103,9 +103,9 @@ class ConnectionHandler(object): return Socks5Proxy(root_context) elif mode == "regular": return HttpProxy(root_context) - elif callable(mode): # pragma: nocover + elif callable(mode): # pragma: no cover return mode(root_context) - else: # pragma: nocover + else: # pragma: no cover raise ValueError("Unknown proxy mode: %s" % mode) def handle(self): |