When building a request from a stream, try to get an accurate
start timestamp from the Reader. This was already in the code
and also used when building response objects, but was ommited
in commit ddf458b330
Without his logic and when the client is reusing a connection
to send requests, the timestamp_start of subsequent requests
is early and equal to when the connection started read blocking
- No output to stdout on load in examples - they muck up the test suite.
- Use the odict module directly, rather than aliasing it. The small convenience
this gives to scripters is not worth it.
- Move the cookie tests from the flow test module to the protocol_http test
module.
Found using fuzzing. Reproduction with pathoc, given "mitmproxy -s" and
pathod running on 9999:
get:'http://localhost:9999/p/':s'200:b\'foo\':h\'Content-Length\'=\'3\'':i58,'\x1a':r
return flow.FlowMaster.run(self)
File "/Users/aldo/mitmproxy/mitmproxy/libmproxy/controller.py", line 111, in run
self.tick(self.masterq, 0.01)
File "/Users/aldo/mitmproxy/mitmproxy/libmproxy/flow.py", line 613, in tick
return controller.Master.tick(self, q, timeout)
File "/Users/aldo/mitmproxy/mitmproxy/libmproxy/controller.py", line 101, in tick
self.handle(*msg)
File "/Users/aldo/mitmproxy/mitmproxy/libmproxy/controller.py", line 118, in handle
m(obj)
File "/Users/aldo/mitmproxy/mitmproxy/libmproxy/flow.py", line 738, in handle_responseheaders
self.stream_large_bodies.run(f, False)
File "/Users/aldo/mitmproxy/mitmproxy/libmproxy/flow.py", line 155, in run
r.headers, is_request, flow.request.method, code
File "/Users/aldo/mitmproxy/mitmproxy/netlib/http.py", line 401, in expected_http_body_size
raise HttpError(400 if is_request else 502, "Invalid content-length header: %s" % headers["content-length"])
netlib.http.HttpError: Invalid content-length header: ['\x1a3']
