mirror of
https://github.com/Grasscutters/mitmproxy.git
synced 2024-11-24 09:15:55 +00:00
590 lines
19 KiB
Python
590 lines
19 KiB
Python
from __future__ import (absolute_import, print_function, division)
|
|
import string
|
|
import sys
|
|
import time
|
|
|
|
from netlib import odict, utils, tcp, http
|
|
from netlib.http import semantics
|
|
from ..exceptions import *
|
|
|
|
|
|
class TCPHandler(object):
|
|
|
|
def __init__(self, rfile, wfile=None):
|
|
self.rfile = rfile
|
|
self.wfile = wfile
|
|
|
|
|
|
class HTTP1Protocol(semantics.ProtocolMixin):
|
|
|
|
ALPN_PROTO_HTTP1 = 'http/1.1'
|
|
|
|
def __init__(self, tcp_handler=None, rfile=None, wfile=None):
|
|
self.tcp_handler = tcp_handler or TCPHandler(rfile, wfile)
|
|
|
|
def read_request(
|
|
self,
|
|
include_body=True,
|
|
body_size_limit=None,
|
|
allow_empty=False,
|
|
):
|
|
"""
|
|
Parse an HTTP request from a file stream
|
|
|
|
Args:
|
|
include_body (bool): Read response body as well
|
|
body_size_limit (bool): Maximum body size
|
|
wfile (file): If specified, HTTP Expect headers are handled
|
|
automatically, by writing a HTTP 100 CONTINUE response to the stream.
|
|
|
|
Returns:
|
|
Request: The HTTP request
|
|
|
|
Raises:
|
|
HttpError: If the input is invalid.
|
|
"""
|
|
timestamp_start = time.time()
|
|
if hasattr(self.tcp_handler.rfile, "reset_timestamps"):
|
|
self.tcp_handler.rfile.reset_timestamps()
|
|
|
|
httpversion, host, port, scheme, method, path, headers, body = (
|
|
None, None, None, None, None, None, None, None)
|
|
|
|
request_line = self._get_request_line()
|
|
if not request_line:
|
|
if allow_empty:
|
|
return http.EmptyRequest()
|
|
else:
|
|
raise tcp.NetLibDisconnect()
|
|
|
|
request_line_parts = self._parse_init(request_line)
|
|
if not request_line_parts:
|
|
raise HttpError(
|
|
400,
|
|
"Bad HTTP request line: %s" % repr(request_line)
|
|
)
|
|
method, path, httpversion = request_line_parts
|
|
|
|
if path == '*' or path.startswith("/"):
|
|
form_in = "relative"
|
|
if not utils.isascii(path):
|
|
raise HttpError(
|
|
400,
|
|
"Bad HTTP request line: %s" % repr(request_line)
|
|
)
|
|
elif method == 'CONNECT':
|
|
form_in = "authority"
|
|
r = self._parse_init_connect(request_line)
|
|
if not r:
|
|
raise HttpError(
|
|
400,
|
|
"Bad HTTP request line: %s" % repr(request_line)
|
|
)
|
|
host, port, httpversion = r
|
|
path = None
|
|
else:
|
|
form_in = "absolute"
|
|
r = self._parse_init_proxy(request_line)
|
|
if not r:
|
|
raise HttpError(
|
|
400,
|
|
"Bad HTTP request line: %s" % repr(request_line)
|
|
)
|
|
_, scheme, host, port, path, _ = r
|
|
|
|
headers = self.read_headers()
|
|
if headers is None:
|
|
raise HttpError(400, "Invalid headers")
|
|
|
|
expect_header = headers.get_first("expect", "").lower()
|
|
if expect_header == "100-continue" and httpversion == (1, 1):
|
|
self.tcp_handler.wfile.write(
|
|
'HTTP/1.1 100 Continue\r\n'
|
|
'\r\n'
|
|
)
|
|
self.tcp_handler.wfile.flush()
|
|
del headers['expect']
|
|
|
|
if include_body:
|
|
body = self.read_http_body(
|
|
headers,
|
|
body_size_limit,
|
|
method,
|
|
None,
|
|
True
|
|
)
|
|
|
|
if hasattr(self.tcp_handler.rfile, "first_byte_timestamp"):
|
|
# more accurate timestamp_start
|
|
timestamp_start = self.tcp_handler.rfile.first_byte_timestamp
|
|
|
|
timestamp_end = time.time()
|
|
|
|
return http.Request(
|
|
form_in,
|
|
method,
|
|
scheme,
|
|
host,
|
|
port,
|
|
path,
|
|
httpversion,
|
|
headers,
|
|
body,
|
|
timestamp_start,
|
|
timestamp_end,
|
|
)
|
|
|
|
def read_response(
|
|
self,
|
|
request_method,
|
|
body_size_limit,
|
|
include_body=True,
|
|
):
|
|
"""
|
|
Returns an http.Response
|
|
|
|
By default, both response header and body are read.
|
|
If include_body=False is specified, body may be one of the
|
|
following:
|
|
- None, if the response is technically allowed to have a response body
|
|
- "", if the response must not have a response body (e.g. it's a
|
|
response to a HEAD request)
|
|
"""
|
|
timestamp_start = time.time()
|
|
if hasattr(self.tcp_handler.rfile, "reset_timestamps"):
|
|
self.tcp_handler.rfile.reset_timestamps()
|
|
|
|
line = self.tcp_handler.rfile.readline()
|
|
# Possible leftover from previous message
|
|
if line == "\r\n" or line == "\n":
|
|
line = self.tcp_handler.rfile.readline()
|
|
if not line:
|
|
raise HttpErrorConnClosed(502, "Server disconnect.")
|
|
parts = self.parse_response_line(line)
|
|
if not parts:
|
|
raise HttpError(502, "Invalid server response: %s" % repr(line))
|
|
proto, code, msg = parts
|
|
httpversion = self._parse_http_protocol(proto)
|
|
if httpversion is None:
|
|
raise HttpError(502, "Invalid HTTP version in line: %s" % repr(proto))
|
|
headers = self.read_headers()
|
|
if headers is None:
|
|
raise HttpError(502, "Invalid headers.")
|
|
|
|
if include_body:
|
|
body = self.read_http_body(
|
|
headers,
|
|
body_size_limit,
|
|
request_method,
|
|
code,
|
|
False
|
|
)
|
|
else:
|
|
# if include_body==False then a None body means the body should be
|
|
# read separately
|
|
body = None
|
|
|
|
if hasattr(self.tcp_handler.rfile, "first_byte_timestamp"):
|
|
# more accurate timestamp_start
|
|
timestamp_start = self.tcp_handler.rfile.first_byte_timestamp
|
|
|
|
if include_body:
|
|
timestamp_end = time.time()
|
|
else:
|
|
timestamp_end = None
|
|
|
|
return http.Response(
|
|
httpversion,
|
|
code,
|
|
msg,
|
|
headers,
|
|
body,
|
|
timestamp_start=timestamp_start,
|
|
timestamp_end=timestamp_end,
|
|
)
|
|
|
|
def assemble_request(self, request):
|
|
assert isinstance(request, semantics.Request)
|
|
|
|
if request.body == semantics.CONTENT_MISSING:
|
|
raise http.HttpError(
|
|
502,
|
|
"Cannot assemble flow with CONTENT_MISSING"
|
|
)
|
|
first_line = self._assemble_request_first_line(request)
|
|
headers = self._assemble_request_headers(request)
|
|
return "%s\r\n%s\r\n%s" % (first_line, headers, request.body)
|
|
|
|
def assemble_response(self, response):
|
|
assert isinstance(response, semantics.Response)
|
|
|
|
if response.body == semantics.CONTENT_MISSING:
|
|
raise http.HttpError(
|
|
502,
|
|
"Cannot assemble flow with CONTENT_MISSING"
|
|
)
|
|
first_line = self._assemble_response_first_line(response)
|
|
headers = self._assemble_response_headers(response)
|
|
return "%s\r\n%s\r\n%s" % (first_line, headers, response.body)
|
|
|
|
def read_headers(self):
|
|
"""
|
|
Read a set of headers.
|
|
Stop once a blank line is reached.
|
|
|
|
Return a ODictCaseless object, or None if headers are invalid.
|
|
"""
|
|
ret = []
|
|
name = ''
|
|
while True:
|
|
line = self.tcp_handler.rfile.readline()
|
|
if not line or line == '\r\n' or line == '\n':
|
|
break
|
|
if line[0] in ' \t':
|
|
if not ret:
|
|
return None
|
|
# continued header
|
|
ret[-1][1] = ret[-1][1] + '\r\n ' + line.strip()
|
|
else:
|
|
i = line.find(':')
|
|
# We're being liberal in what we accept, here.
|
|
if i > 0:
|
|
name = line[:i]
|
|
value = line[i + 1:].strip()
|
|
ret.append([name, value])
|
|
else:
|
|
return None
|
|
return odict.ODictCaseless(ret)
|
|
|
|
|
|
def read_http_body(self, *args, **kwargs):
|
|
return "".join(
|
|
content for _, content, _ in self.read_http_body_chunked(*args, **kwargs)
|
|
)
|
|
|
|
|
|
def read_http_body_chunked(
|
|
self,
|
|
headers,
|
|
limit,
|
|
request_method,
|
|
response_code,
|
|
is_request,
|
|
max_chunk_size=None
|
|
):
|
|
"""
|
|
Read an HTTP message body:
|
|
headers: An ODictCaseless object
|
|
limit: Size limit.
|
|
is_request: True if the body to read belongs to a request, False
|
|
otherwise
|
|
"""
|
|
if max_chunk_size is None:
|
|
max_chunk_size = limit or sys.maxsize
|
|
|
|
expected_size = self.expected_http_body_size(
|
|
headers, is_request, request_method, response_code
|
|
)
|
|
|
|
if expected_size is None:
|
|
if self.has_chunked_encoding(headers):
|
|
# Python 3: yield from
|
|
for x in self._read_chunked(limit, is_request):
|
|
yield x
|
|
else: # pragma: nocover
|
|
raise HttpError(
|
|
400 if is_request else 502,
|
|
"Content-Length unknown but no chunked encoding"
|
|
)
|
|
elif expected_size >= 0:
|
|
if limit is not None and expected_size > limit:
|
|
raise HttpError(
|
|
400 if is_request else 509,
|
|
"HTTP Body too large. Limit is %s, content-length was %s" % (
|
|
limit, expected_size
|
|
)
|
|
)
|
|
bytes_left = expected_size
|
|
while bytes_left:
|
|
chunk_size = min(bytes_left, max_chunk_size)
|
|
content = self.tcp_handler.rfile.read(chunk_size)
|
|
yield "", content, ""
|
|
bytes_left -= chunk_size
|
|
else:
|
|
bytes_left = limit or -1
|
|
while bytes_left:
|
|
chunk_size = min(bytes_left, max_chunk_size)
|
|
content = self.tcp_handler.rfile.read(chunk_size)
|
|
if not content:
|
|
return
|
|
yield "", content, ""
|
|
bytes_left -= chunk_size
|
|
not_done = self.tcp_handler.rfile.read(1)
|
|
if not_done:
|
|
raise HttpError(
|
|
400 if is_request else 509,
|
|
"HTTP Body too large. Limit is %s," % limit
|
|
)
|
|
|
|
@classmethod
|
|
def expected_http_body_size(
|
|
self,
|
|
headers,
|
|
is_request,
|
|
request_method,
|
|
response_code,
|
|
):
|
|
"""
|
|
Returns the expected body length:
|
|
- a positive integer, if the size is known in advance
|
|
- None, if the size in unknown in advance (chunked encoding or invalid
|
|
data)
|
|
- -1, if all data should be read until end of stream.
|
|
|
|
May raise HttpError.
|
|
"""
|
|
# Determine response size according to
|
|
# http://tools.ietf.org/html/rfc7230#section-3.3
|
|
if request_method:
|
|
request_method = request_method.upper()
|
|
|
|
if (not is_request and (
|
|
request_method == "HEAD" or
|
|
(request_method == "CONNECT" and response_code == 200) or
|
|
response_code in [204, 304] or
|
|
100 <= response_code <= 199)):
|
|
return 0
|
|
if self.has_chunked_encoding(headers):
|
|
return None
|
|
if "content-length" in headers:
|
|
try:
|
|
size = int(headers["content-length"][0])
|
|
if size < 0:
|
|
raise ValueError()
|
|
return size
|
|
except ValueError:
|
|
return None
|
|
if is_request:
|
|
return 0
|
|
return -1
|
|
|
|
|
|
@classmethod
|
|
def has_chunked_encoding(self, headers):
|
|
return "chunked" in [
|
|
i.lower() for i in utils.get_header_tokens(headers, "transfer-encoding")
|
|
]
|
|
|
|
|
|
def _get_request_line(self):
|
|
"""
|
|
Get a line, possibly preceded by a blank.
|
|
"""
|
|
line = self.tcp_handler.rfile.readline()
|
|
if line == "\r\n" or line == "\n":
|
|
# Possible leftover from previous message
|
|
line = self.tcp_handler.rfile.readline()
|
|
return line
|
|
|
|
def _read_chunked(self, limit, is_request):
|
|
"""
|
|
Read a chunked HTTP body.
|
|
|
|
May raise HttpError.
|
|
"""
|
|
# FIXME: Should check if chunked is the final encoding in the headers
|
|
# http://tools.ietf.org/html/draft-ietf-httpbis-p1-messaging-16#section-3.3
|
|
# 3.3 2.
|
|
total = 0
|
|
code = 400 if is_request else 502
|
|
while True:
|
|
line = self.tcp_handler.rfile.readline(128)
|
|
if line == "":
|
|
raise HttpErrorConnClosed(code, "Connection closed prematurely")
|
|
if line != '\r\n' and line != '\n':
|
|
try:
|
|
length = int(line, 16)
|
|
except ValueError:
|
|
raise HttpError(
|
|
code,
|
|
"Invalid chunked encoding length: %s" % line
|
|
)
|
|
total += length
|
|
if limit is not None and total > limit:
|
|
msg = "HTTP Body too large. Limit is %s," \
|
|
" chunked content longer than %s" % (limit, total)
|
|
raise HttpError(code, msg)
|
|
chunk = self.tcp_handler.rfile.read(length)
|
|
suffix = self.tcp_handler.rfile.readline(5)
|
|
if suffix != '\r\n':
|
|
raise HttpError(code, "Malformed chunked body")
|
|
yield line, chunk, '\r\n'
|
|
if length == 0:
|
|
return
|
|
|
|
@classmethod
|
|
def _parse_http_protocol(self, line):
|
|
"""
|
|
Parse an HTTP protocol declaration.
|
|
Returns a (major, minor) tuple, or None.
|
|
"""
|
|
if not line.startswith("HTTP/"):
|
|
return None
|
|
_, version = line.split('/', 1)
|
|
if "." not in version:
|
|
return None
|
|
major, minor = version.split('.', 1)
|
|
try:
|
|
major = int(major)
|
|
minor = int(minor)
|
|
except ValueError:
|
|
return None
|
|
return major, minor
|
|
|
|
@classmethod
|
|
def _parse_init(self, line):
|
|
try:
|
|
method, url, protocol = string.split(line)
|
|
except ValueError:
|
|
return None
|
|
httpversion = self._parse_http_protocol(protocol)
|
|
if not httpversion:
|
|
return None
|
|
if not utils.isascii(method):
|
|
return None
|
|
return method, url, httpversion
|
|
|
|
@classmethod
|
|
def _parse_init_connect(self, line):
|
|
"""
|
|
Returns (host, port, httpversion) if line is a valid CONNECT line.
|
|
http://tools.ietf.org/html/draft-luotonen-web-proxy-tunneling-01 section 3.1
|
|
"""
|
|
v = self._parse_init(line)
|
|
if not v:
|
|
return None
|
|
method, url, httpversion = v
|
|
|
|
if method.upper() != 'CONNECT':
|
|
return None
|
|
try:
|
|
host, port = url.split(":")
|
|
except ValueError:
|
|
return None
|
|
try:
|
|
port = int(port)
|
|
except ValueError:
|
|
return None
|
|
if not utils.is_valid_port(port):
|
|
return None
|
|
if not utils.is_valid_host(host):
|
|
return None
|
|
return host, port, httpversion
|
|
|
|
@classmethod
|
|
def _parse_init_proxy(self, line):
|
|
v = self._parse_init(line)
|
|
if not v:
|
|
return None
|
|
method, url, httpversion = v
|
|
|
|
parts = utils.parse_url(url)
|
|
if not parts:
|
|
return None
|
|
scheme, host, port, path = parts
|
|
return method, scheme, host, port, path, httpversion
|
|
|
|
@classmethod
|
|
def _parse_init_http(self, line):
|
|
"""
|
|
Returns (method, url, httpversion)
|
|
"""
|
|
v = self._parse_init(line)
|
|
if not v:
|
|
return None
|
|
method, url, httpversion = v
|
|
if not utils.isascii(url):
|
|
return None
|
|
if not (url.startswith("/") or url == "*"):
|
|
return None
|
|
return method, url, httpversion
|
|
|
|
@classmethod
|
|
def connection_close(self, httpversion, headers):
|
|
"""
|
|
Checks the message to see if the client connection should be closed
|
|
according to RFC 2616 Section 8.1 Note that a connection should be
|
|
closed as well if the response has been read until end of the stream.
|
|
"""
|
|
# At first, check if we have an explicit Connection header.
|
|
if "connection" in headers:
|
|
toks = utils.get_header_tokens(headers, "connection")
|
|
if "close" in toks:
|
|
return True
|
|
elif "keep-alive" in toks:
|
|
return False
|
|
|
|
# If we don't have a Connection header, HTTP 1.1 connections are assumed to
|
|
# be persistent
|
|
return httpversion != (1, 1)
|
|
|
|
@classmethod
|
|
def parse_response_line(self, line):
|
|
parts = line.strip().split(" ", 2)
|
|
if len(parts) == 2: # handle missing message gracefully
|
|
parts.append("")
|
|
if len(parts) != 3:
|
|
return None
|
|
proto, code, msg = parts
|
|
try:
|
|
code = int(code)
|
|
except ValueError:
|
|
return None
|
|
return (proto, code, msg)
|
|
|
|
@classmethod
|
|
def _assemble_request_first_line(self, request):
|
|
return request.legacy_first_line()
|
|
|
|
def _assemble_request_headers(self, request):
|
|
headers = request.headers.copy()
|
|
for k in request._headers_to_strip_off:
|
|
del headers[k]
|
|
if 'host' not in headers and request.scheme and request.host and request.port:
|
|
headers["Host"] = [utils.hostport(request.scheme,
|
|
request.host,
|
|
request.port)]
|
|
|
|
# If content is defined (i.e. not None or CONTENT_MISSING), we always
|
|
# add a content-length header.
|
|
if request.body or request.body == "":
|
|
headers["Content-Length"] = [str(len(request.body))]
|
|
|
|
return headers.format()
|
|
|
|
def _assemble_response_first_line(self, response):
|
|
return 'HTTP/%s.%s %s %s' % (
|
|
response.httpversion[0],
|
|
response.httpversion[1],
|
|
response.status_code,
|
|
response.msg,
|
|
)
|
|
|
|
def _assemble_response_headers(
|
|
self,
|
|
response,
|
|
preserve_transfer_encoding=False,
|
|
):
|
|
headers = response.headers.copy()
|
|
for k in response._headers_to_strip_off:
|
|
del headers[k]
|
|
if not preserve_transfer_encoding:
|
|
del headers['Transfer-Encoding']
|
|
|
|
# If body is defined (i.e. not None or CONTENT_MISSING), we always
|
|
# add a content-length header.
|
|
if response.body or response.body == "":
|
|
headers["Content-Length"] = [str(len(response.body))]
|
|
|
|
return headers.format()
|