diff options
Diffstat (limited to 'netlib')
| -rw-r--r-- | netlib/tcp.py | 32 |
1 files changed, 20 insertions, 12 deletions
diff --git a/netlib/tcp.py b/netlib/tcp.py index 10269aa4..84008e2c 100644 --- a/netlib/tcp.py +++ b/netlib/tcp.py @@ -247,24 +247,32 @@ def close_socket(sock): """ try: # We already indicate that we close our end. - # If we close RD, any further received bytes would result in a RST being set, which we want to avoid - # for our purposes sock.shutdown(socket.SHUT_WR) # may raise "Transport endpoint is not connected" on Linux # Section 4.2.2.13 of RFC 1122 tells us that a close() with any # pending readable data could lead to an immediate RST being sent (which is the case on Windows). # http://ia600609.us.archive.org/22/items/TheUltimateSo_lingerPageOrWhyIsMyTcpNotReliable/the-ultimate-so_linger-page-or-why-is-my-tcp-not-reliable.html # - # However, we cannot rely on the shutdown()-followed-by-read()-eof technique proposed by the page above: - # Some remote machines just don't send a TCP FIN, which would leave us in the unfortunate situation that - # recv() would block infinitely. - # As a workaround, we set a timeout here even if we are in blocking mode. - # Please let us know if you have a better solution to this problem. - - sock.settimeout(sock.gettimeout() or 20) - # may raise a timeout/disconnect exception. - while sock.recv(4096): # pragma: no cover - pass + # This in turn results in the following issue: If we send an error page to the client and then close the socket, + # the RST may be received by the client before the error page and the users sees a connection error rather than + # the error page. Thus, we try to empty the read buffer on Windows first. + # (see https://github.com/mitmproxy/mitmproxy/issues/527#issuecomment-93782988) + # + if os.name == "nt": # pragma: no cover + # We cannot rely on the shutdown()-followed-by-read()-eof technique proposed by the page above: + # Some remote machines just don't send a TCP FIN, which would leave us in the unfortunate situation that + # recv() would block infinitely. + # As a workaround, we set a timeout here even if we are in blocking mode. + sock.settimeout(sock.gettimeout() or 20) + + # limit at a megabyte so that we don't read infinitely + for _ in xrange(1024 ** 3 // 4096): + # may raise a timeout/disconnect exception. + if not sock.recv(4096): + break + + # Now we can close the other half as well. + sock.shutdown(socket.SHUT_RD) except socket.error: pass |