#!/usr/bin/env python
"""
The flashproxy client transport plugin.
"""

import argparse
import BaseHTTPServer
import array
import base64
import cStringIO
import flashproxy
import os
import os.path
import select
import socket
import struct
import subprocess
import sys
import threading
import time
import traceback

from flashproxy.util import parse_addr_spec, addr_family, format_addr, safe_str, safe_format_addr
from flashproxy.reg import DEFAULT_TRANSPORT

from hashlib import sha1

try:
    import numpy
except ImportError:
    numpy = None

# Default local port in managed mode (choose one arbitrarily).
DEFAULT_LOCAL_PORT_MANAGED = 0
# Default local port in external mode.
DEFAULT_LOCAL_PORT_EXTERNAL = 9001
DEFAULT_REMOTE_PORT = 9000
DEFAULT_REGISTER_METHODS = ["appspot", "email", "http"]
DEFAULT_PORT_FORWARDING_HELPER = "tor-fw-helper"

# We will re-register if we have fewer than this many waiting proxies. The
# facilitator may choose to ignore our requests.
DESIRED_NUMBER_OF_PROXIES = 3

# We accept up to this many bytes from a socket not yet matched with a partner
# before disconnecting it.
UNCONNECTED_BUFFER_LIMIT = 10240

LOG_DATE_FORMAT = "%Y-%m-%d %H:%M:%S"

class options(object):
    local_addrs = []
    remote_addrs = []
    register_addr = None

    managed = True

    daemonize = False
    log_filename = None
    log_file = sys.stdout
    pid_filename = None
    port_forwarding = False
    port_forwarding_helper = DEFAULT_PORT_FORWARDING_HELPER
    port_forwarding_external = None
    register = False
    register_commands = []

    # registration options
    address_family = socket.AF_UNSPEC
    transport = DEFAULT_TRANSPORT
    safe_logging = True
    facilitator_url = None
    facilitator_pubkey_filename = None

log_lock = threading.Lock()
def log(msg):
    with log_lock:
        print >> options.log_file, (u"%s %s" % (time.strftime(LOG_DATE_FORMAT), msg)).encode("UTF-8")
        options.log_file.flush()

def format_sockaddr(sockaddr):
    host, port = socket.getnameinfo(sockaddr, socket.NI_NUMERICHOST | socket.NI_NUMERICSERV)
    port = int(port)
    return format_addr((host, port))

def safe_format_sockaddr(sockaddr):
    return safe_str(format_sockaddr(sockaddr))

def safe_format_peername(s):
    try:
        return safe_format_sockaddr(s.getpeername())
    except socket.error, e:
        return "<unconnected>"


def apply_mask_numpy(payload, mask_key):
    if len(payload) == 0:
        return ""
    payload_a = numpy.frombuffer(payload, dtype="|u4", count=len(payload)//4)
    m, = numpy.frombuffer(mask_key, dtype="|u4", count=1)
    result = numpy.bitwise_xor(payload_a, m).tostring()
    i = len(payload) // 4 * 4
    if i < len(payload):
        remainder = []
        while i < len(payload):
            remainder.append(chr(ord(payload[i]) ^ ord(mask_key[i % 4])))
            i += 1
        result = result + "".join(remainder)
    return result

def apply_mask_py(payload, mask_key):
    result = array.array("B", payload)
    m = array.array("B", mask_key)
    i = 0
    while i < len(result) - 7:
        result[i] ^= m[0]
        result[i+1] ^= m[1]
        result[i+2] ^= m[2]
        result[i+3] ^= m[3]
        result[i+4] ^= m[0]
        result[i+5] ^= m[1]
        result[i+6] ^= m[2]
        result[i+7] ^= m[3]
        i += 8
    while i < len(result):
        result[i] ^= m[i%4]
        i += 1
    return result.tostring()

if numpy is not None:
    apply_mask = apply_mask_numpy
else:
    apply_mask = apply_mask_py

class WebSocketFrame(object):
    def __init__(self):
        self.fin = False
        self.opcode = None
        self.payload = None

    def is_control(self):
        return (self.opcode & 0x08) != 0

class WebSocketMessage(object):
    def __init__(self):
        self.opcode = None
        self.payload = None

    def is_control(self):
        return (self.opcode & 0x08) != 0

class WebSocketDecoder(object):
    """RFC 6455 section 5 is about the WebSocket framing format."""
    # Raise an exception rather than buffer anything larger than this.
    MAX_MESSAGE_LENGTH = 1024 * 1024

    class MaskingError(ValueError):
        pass

    def __init__(self, use_mask = False):
        """use_mask should be True for server-to-client sockets, and False for
        client-to-server sockets."""
        self.use_mask = use_mask

        # Per-frame state.
        self.buf = ""

        # Per-message state.
        self.message_buf = ""
        self.message_opcode = None

    def feed(self, data):
        self.buf += data

    def read_frame(self):
        """Read a frame from the internal buffer, if one is available. Returns a
        WebSocketFrame object, or None if there are no complete frames to
        read."""
        # RFC 6255 section 5.2.
        if len(self.buf) < 2:
            return None
        offset = 0
        b0, b1 = struct.unpack_from(">BB", self.buf, offset)
        offset += 2
        fin = (b0 & 0x80) != 0
        opcode = b0 & 0x0f
        frame_masked = (b1 & 0x80) != 0
        payload_len = b1 & 0x7f

        if payload_len == 126:
            if len(self.buf) < offset + 2:
                return None
            payload_len, = struct.unpack_from(">H", self.buf, offset)
            offset += 2
        elif payload_len == 127:
            if len(self.buf) < offset + 8:
                return None
            payload_len, = struct.unpack_from(">Q", self.buf, offset)
            offset += 8

        if frame_masked:
            if not self.use_mask:
                # "A client MUST close a connection if it detects a masked
                # frame."
                raise self.MaskingError("Got masked payload from server")
            if len(self.buf) < offset + 4:
                return None
            mask_key = self.buf[offset:offset+4]
            offset += 4
        else:
            if self.use_mask:
                # "The server MUST close the connection upon receiving a frame
                # that is not masked."
                raise self.MaskingError("Got unmasked payload from client")
            mask_key = None

        if payload_len > self.MAX_MESSAGE_LENGTH:
            raise ValueError("Refusing to buffer payload of %d bytes" % payload_len)

        if len(self.buf) < offset + payload_len:
            return None
        if mask_key:
            payload = apply_mask(self.buf[offset:offset+payload_len], mask_key)
        else:
            payload = self.buf[offset:offset+payload_len]
        self.buf = self.buf[offset+payload_len:]

        frame = WebSocketFrame()
        frame.fin = fin
        frame.opcode = opcode
        frame.payload = payload

        return frame

    def read_message(self):
        """Read a complete message. If the opcode is 1, the payload is decoded
        from a UTF-8 binary string to a unicode string. If a control frame is
        read while another fragmented message is in progress, the control frame
        is returned as a new message immediately. Returns None if there is no
        complete frame to be read."""
        # RFC 6455 section 5.4 is about fragmentation.
        while True:
            frame = self.read_frame()
            if frame is None:
                return None
            # "Control frames (see Section 5.5) MAY be injected in the middle of
            # a fragmented message. Control frames themselves MUST NOT be
            # fragmented."
            if frame.is_control():
                if not frame.fin:
                    raise ValueError("Control frame (opcode %d) has FIN bit clear" % frame.opcode)
                message = WebSocketMessage()
                message.opcode = frame.opcode
                message.payload = frame.payload
                return message

            if self.message_opcode is None:
                if frame.opcode == 0:
                    raise ValueError("First frame has opcode 0")
                self.message_opcode = frame.opcode
            else:
                if frame.opcode != 0:
                    raise ValueError("Non-first frame has nonzero opcode %d" % frame.opcode)
            if len(self.message_buf) + len(frame.payload) > self.MAX_MESSAGE_LENGTH:
                raise ValueError("Refusing to buffer payload of %d bytes" % (len(self.message_buf) + len(frame.payload)))
            self.message_buf += frame.payload

            if frame.fin:
                break
        message = WebSocketMessage()
        message.opcode = self.message_opcode
        message.payload = self.message_buf
        self.postprocess_message(message)
        self.message_opcode = None
        self.message_buf = ""

        return message

    def postprocess_message(self, message):
        if message.opcode == 1:
            message.payload = message.payload.decode("utf-8")
        return message

class WebSocketEncoder(object):
    def __init__(self, use_mask = False):
        self.use_mask = use_mask

    def encode_frame(self, opcode, payload):
        if opcode >= 16:
            raise ValueError("Opcode of %d is >= 16" % opcode)
        length = len(payload)

        if self.use_mask:
            mask_key = os.urandom(4)
            payload = apply_mask(payload, mask_key)
            mask_bit = 0x80
        else:
            mask_key = ""
            mask_bit = 0x00

        if length < 126:
            len_b, len_ext = length, ""
        elif length < 0x10000:
            len_b, len_ext = 126, struct.pack(">H", length)
        elif length < 0x10000000000000000:
            len_b, len_ext = 127, struct.pack(">Q", length)
        else:
            raise ValueError("payload length of %d is too long" % length)

        return chr(0x80 | opcode) + chr(mask_bit | len_b) + len_ext + mask_key + payload

    def encode_message(self, opcode, payload):
        if opcode == 1:
            payload = payload.encode("utf-8")
        return self.encode_frame(opcode, payload)

# WebSocket implementations generally support text (opcode 1) messages, which
# are UTF-8-encoded text. Not all support binary (opcode 2) messages. During the
# WebSocket handshake, we use the "base64" value of the Sec-WebSocket-Protocol
# header field to indicate that text frames should encoded UTF-8-encoded
# base64-encoded binary data. Binary messages are always interpreted verbatim,
# but text messages are rejected if "base64" was not negotiated.
#
# The idea here is that browsers that know they don't support binary messages
# can negotiate "base64" with both endpoints and still reliably transport binary
# data. Those that know they can support binary messages can just use binary
# messages in the straightforward way.

class WebSocketBinaryDecoder(object):
    def __init__(self, protocols, use_mask = False):
        self.dec = WebSocketDecoder(use_mask)
        self.base64 = "base64" in protocols

    def feed(self, data):
        self.dec.feed(data)

    def read(self):
        """Returns None when there are currently no data to be read. Returns ""
        when a close message is received."""
        while True:
            message = self.dec.read_message()
            if message is None:
                return None
            elif message.opcode == 1:
                if not self.base64:
                    raise ValueError("Received text message on decoder incapable of base64")
                payload = base64.b64decode(message.payload)
                if payload:
                    return payload
            elif message.opcode == 2:
                if message.payload:
                    return message.payload
            elif message.opcode == 8:
                return ""
            # Ignore all other opcodes.
        return None

class WebSocketBinaryEncoder(object):
    def __init__(self, protocols, use_mask = False):
        self.enc = WebSocketEncoder(use_mask)
        self.base64 = "base64" in protocols

    def encode(self, data):
        if self.base64:
            return self.enc.encode_message(1, base64.b64encode(data))
        else:
            return self.enc.encode_message(2, data)


def listen_socket(addr):
    """Return a socket listening on the given address."""
    addrinfo = socket.getaddrinfo(addr[0], addr[1], 0, socket.SOCK_STREAM, socket.IPPROTO_TCP)[0]
    s = socket.socket(addrinfo[0], addrinfo[1], addrinfo[2])
    s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
    if addrinfo[0] == socket.AF_INET6 and socket.has_ipv6:
        # Set the IPV6_V6ONLY socket option, otherwise some operating systems
        # will listen on an IPv4 address as well as IPv6 by default. For
        # example, "::" will listen on both "::" and "0.0.0.0", and "::1" will
        # listen on both "::1" and "127.0.0.1". See
        # https://trac.torproject.org/projects/tor/ticket/4760.
        try:
            s.setsockopt(socket.IPPROTO_IPV6, socket.IPV6_V6ONLY, 1)
        except AttributeError:
            # Python 2.7.3 on Windows does not define IPPROTO_IPV6; see
            # http://bugs.python.org/issue6926. IPV6_V6ONLY is the default
            # behavior on Windows anyway, so we can skip the setsockopt.
            pass
        except socket.error:
            # Seen on Windows XP:
            # socket.error: [Errno 109] Protocol not available
            pass
    s.bind(addr)
    s.listen(10)
    return s

# How long to wait for a WebSocket request on the remote socket. It is limited
# to avoid Slowloris-like attacks.
WEBSOCKET_REQUEST_TIMEOUT = 2.0

# This subclass of BaseHTTPRequestHandler is essentially a means of parsing an
# HTTP request.
class WebSocketRequestHandler(BaseHTTPServer.BaseHTTPRequestHandler):
    def __init__(self, request_text, fd):
        self.rfile = cStringIO.StringIO(request_text)
        self.wfile = fd.makefile()
        self.error = False
        self.raw_requestline = self.rfile.readline()
        self.parse_request()

    def log_message(self, *args):
        pass

    def send_error(self, code, message = None):
        BaseHTTPServer.BaseHTTPRequestHandler.send_error(self, code, message)
        self.error = True

MAGIC_GUID = "258EAFA5-E914-47DA-95CA-C5AB0DC85B11"

def handle_websocket_request(fd):
    try:
        request_text = fd.recv(10 * 1024)
    except socket.error, e:
        log(u"Socket error while receiving WebSocket request: %s" % repr(str(e)))
        return None
    handler = WebSocketRequestHandler(request_text, fd)
    if handler.error or not hasattr(handler, "path"):
        return None
    method = handler.command
    path = handler.path
    headers = handler.headers

    # See RFC 6455 section 4.2.1 for this sequence of checks.
    #
    # 1. An HTTP/1.1 or higher GET request, including a "Request-URI"...
    if method != "GET":
        handler.send_error(405)
        return None
    if path != "/":
        handler.send_error(404)
        return None

    # 2. A |Host| header field containing the server's authority.
    # We deliberately skip this test.

    # 3. An |Upgrade| header field containing the value "websocket", treated as
    # an ASCII case-insensitive value.
    upgrade = headers.get("upgrade")
    if upgrade is None:
        handler.send_error(400)
        return None
    if "websocket" not in [x.strip().lower() for x in upgrade.split(",")]:
        handler.send_error(400)
        return None

    # 4. A |Connection| header field that includes the token "Upgrade", treated
    # as an ASCII case-insensitive value.
    connection = headers.get("connection")
    if connection is None:
        handler.send_error(400)
        return None
    if "upgrade" not in [x.strip().lower() for x in connection.split(",")]:
        handler.send_error(400)
        return None

    # 5. A |Sec-WebSocket-Key| header field with a base64-encoded value that,
    # when decoded, is 16 bytes in length.
    try:
        key = headers.get("sec-websocket-key")
        if len(base64.b64decode(key)) != 16:
            raise TypeError("Sec-WebSocket-Key must be 16 bytes")
    except TypeError:
        handler.send_error(400)
        return None

    # 6. A |Sec-WebSocket-Version| header field, with a value of 13. We also
    # allow 8 from draft-ietf-hybi-thewebsocketprotocol-10.
    version = headers.get("sec-websocket-version")
    KNOWN_VERSIONS = ["8", "13"]
    if version not in KNOWN_VERSIONS:
        # "If this version does not match a version understood by the server,
        # the server MUST abort the WebSocket handshake described in this
        # section and instead send an appropriate HTTP error code (such as 426
        # Upgrade Required) and a |Sec-WebSocket-Version| header field
        # indicating the version(s) the server is capable of understanding."
        handler.send_response(426)
        handler.send_header("Sec-WebSocket-Version", ", ".join(KNOWN_VERSIONS))
        handler.end_headers()
        return None

    # 7. Optionally, an |Origin| header field.

    # 8. Optionally, a |Sec-WebSocket-Protocol| header field, with a list of
    # values indicating which protocols the client would like to speak, ordered
    # by preference.
    protocols_str = headers.get("sec-websocket-protocol")
    if protocols_str is None:
        protocols = []
    else:
        protocols = [x.strip().lower() for x in protocols_str.split(",")]

    # 9. Optionally, a |Sec-WebSocket-Extensions| header field...

    # 10. Optionally, other header fields...

    # See RFC 6455 section 4.2.2, item 5 for these steps.

    # 1. A Status-Line with a 101 response code as per RFC 2616.
    handler.send_response(101)
    # 2. An |Upgrade| header field with value "websocket" as per RFC 2616.
    handler.send_header("Upgrade", "websocket")
    # 3. A |Connection| header field with value "Upgrade".
    handler.send_header("Connection", "Upgrade")
    # 4. A |Sec-WebSocket-Accept| header field.  The value of this header field
    # is constructed by concatenating /key/, defined above in step 4 in Section
    # 4.2.2, with the string "258EAFA5-E914-47DA-95CA-C5AB0DC85B11", taking the
    # SHA-1 hash of this concatenated value to obtain a 20-byte value and
    # base64-encoding (see Section 4 of [RFC4648]) this 20-byte hash.
    accept_key = base64.b64encode(sha1(key + MAGIC_GUID).digest())
    handler.send_header("Sec-WebSocket-Accept", accept_key)
    # 5.  Optionally, a |Sec-WebSocket-Protocol| header field, with a value
    # /subprotocol/ as defined in step 4 in Section 4.2.2.
    if "base64" in protocols:
        handler.send_header("Sec-WebSocket-Protocol", "base64")
    # 6.  Optionally, a |Sec-WebSocket-Extensions| header field...

    handler.end_headers()

    return protocols

def grab_string(s, pos):
    """Grab a NUL-terminated string from the given string, starting at the given
    offset. Return (pos, str) tuple, or (pos, None) on error."""
    i = pos
    while i < len(s):
        if s[i] == '\0':
            return (i + 1, s[pos:i])
        i += 1
    return pos, None

# http://ftp.icm.edu.pl/packages/socks/socks4/SOCKS4.protocol
# https://en.wikipedia.org/wiki/SOCKS#SOCKS4a
def parse_socks_request(data):
    """Parse the 8-byte SOCKS header at the beginning of data. Returns a
    (dest, port) tuple. Raises ValueError on error."""
    try:
        ver, cmd, dport, o1, o2, o3, o4 = struct.unpack(">BBHBBBB", data[:8])
    except struct.error:
        raise ValueError("Couldn't unpack SOCKS4 header")
    if ver != 4:
        raise ValueError("Wrong SOCKS version (%d)" % ver)
    if cmd != 1:
        raise ValueError("Wrong SOCKS command (%d)" % cmd)
    pos, userid = grab_string(data, 8)
    if userid is None:
        raise ValueError("Couldn't read userid")
    if o1 == 0 and o2 == 0 and o3 == 0 and o4 != 0:
        pos, dest = grab_string(data, pos)
        if dest is None:
            raise ValueError("Couldn't read destination")
    else:
        dest = "%d.%d.%d.%d" % (o1, o2, o3, o4)
    return dest, dport

def handle_socks_request(fd):
    try:
        addr = fd.getpeername()
        data = fd.recv(100)
    except socket.error, e:
        log(u"Socket error from SOCKS-pending: %s" % repr(str(e)))
        return False
    try:
        dest_addr = parse_socks_request(data)
    except ValueError, e:
        log(u"Error parsing SOCKS request: %s." % str(e))
        # Error reply.
        fd.sendall(struct.pack(">BBHBBBB", 0, 91, 0, 0, 0, 0, 0))
        return False
    log(u"Got SOCKS request for %s." % safe_format_addr(dest_addr))
    fd.sendall(struct.pack(">BBHBBBB", 0, 90, dest_addr[1], 127, 0, 0, 1))
    # Note we throw away the requested address and port.
    return True

def report_pending():
    log(u"locals  (%d): %s" % (len(locals), [safe_format_peername(x) for x in locals]))
    log(u"remotes (%d): %s" % (len(remotes), [safe_format_peername(x) for x in remotes]))

def forward_ports(pairs):
    """Attempt to forward all given pairs (external, internal) pairs of ports
    using port_forwarding_helper."""
    command = [options.port_forwarding_helper]
    basename = os.path.basename(command[0])
    for external, internal in pairs:
        command += ["-p", "%d:%d" % (external, internal)]
    try:
        log(u"Running port forwarding command: %s" % " ".join(command))
        p = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
        stdout, stderr = p.communicate()
    except OSError, e:
        log(u"Error running %s: %s" % (basename, str(e)))
        return False
    for line in stdout.splitlines():
        log(u"%s: %s" % (basename, line))
    for line in stderr.splitlines():
        log(u"%s: %s" % (basename, line))
    if p.returncode != 0:
        log("%s exited with status %d." % (basename, p.returncode))
        return False
    return True

def forward_listeners(listeners):
    """Attempt to forward the ports belonging to the given listening sockets.
    Non-IPv4 addresses are ignored. If options.port_forwarding_external is not
    None, only the first IPv4 address in the list will be forwarded."""
    forward_list = []
    for listener in remote_listen:
        host, port = socket.getnameinfo(listener.getsockname(), socket.NI_NUMERICHOST | socket.NI_NUMERICSERV)
        port = int(port)
        af = addr_family(host)
        if af != socket.AF_INET:
            # I guess tor-fw-helper can only handle IPv4.
            log(u"Not forwarding to %s because it is not an IPv4 address." % format_addr((host, port)))
            continue
        if options.port_forwarding_external is not None:
            forward_list.append((options.port_forwarding_external, port))
            # A fixed external address means we can forward only one port.
            break
        else:
            forward_list.append((port, port))
    forward_ports(forward_list)

register_condvar = threading.Condition()
# register_flag true means registration_thread_func should register at its next
# opportunity.
register_flag = False
def register():
    global register_flag
    if not options.register:
        return
    with register_condvar:
        register_flag = True
        register_condvar.notify()

def register_using_command(command):
    basename = os.path.basename(command[0])
    try:
        log(u"Running command: %s" % " ".join(command))
        p = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
        stdout, stderr = p.communicate()
    except OSError, e:
        log(u"Error running %s: %s" % (basename, str(e)))
        return False
    for line in stdout.splitlines():
        log(u"%s: %s" % (basename, line))
    for line in stderr.splitlines():
        log(u"%s: %s" % (basename, line))
    if p.returncode != 0:
        log("%s exited with status %d." % (basename, p.returncode))
        return False
    return True

def register_one():
    spec = format_addr(options.register_addr)
    log(u"Trying to register \"%s\"." % spec)
    for command in options.register_commands:
        if register_using_command(command + [spec]):
            break
    else:
        log(u"All registration commands failed.")

def registration_thread_func():
    global register_flag
    while True:
        with register_condvar:
            while not register_flag:
                register_condvar.wait()
            register_flag = False
        if len(unlinked_remotes) < DESIRED_NUMBER_OF_PROXIES:
            register_one()

def proxy_chunk_local_to_remote(local, remote, data = None):
    if data is None:
        try:
            data = local.recv(65536)
        except socket.error, e: # Can be "Connection reset by peer".
            log(u"Socket error from local: %s" % repr(str(e)))
            remote.close()
            return False
    if not data:
        log(u"EOF from local %s." % safe_format_peername(local))
        local.close()
        remote.close()
        return False
    else:
        try:
            remote.send_chunk(data)
        except socket.error, e:
            log(u"Socket error writing to remote: %s" % repr(str(e)))
            local.close()
            return False
        return True

def proxy_chunk_remote_to_local(remote, local, data = None):
    if data is None:
        try:
            data = remote.recv(65536)
        except socket.error, e: # Can be "Connection reset by peer".
            log(u"Socket error from remote: %s" % repr(str(e)))
            local.close()
            return False
    if not data:
        log(u"EOF from remote %s." % safe_format_peername(remote))
        remote.close()
        local.close()
        return False
    else:
        remote.dec.feed(data)
        while True:
            try:
                data = remote.dec.read()
            except (WebSocketDecoder.MaskingError, ValueError), e:
                log(u"WebSocket decode error from remote: %s" % repr(str(e)))
                remote.close()
                local.close()
                return False
            if data is None:
                break
            elif not data:
                log(u"WebSocket close from remote %s." % safe_format_peername(remote))
                remote.close()
                local.close()
                return False
            try:
                local.send_chunk(data)
            except socket.error, e:
                log(u"Socket error writing to local: %s" % repr(str(e)))
                remote.close()
                return False
        return True

def receive_unlinked(fd, label):
    """Receive and buffer data on a socket that has not been linked yet. Returns
    True iff there was no error and the socket may still be used; otherwise, the
    socket will be closed before returning."""

    try:
        data = fd.recv(1024)
    except socket.error, e:
        log(u"Socket error from %s: %s" % (label, repr(str(e))))
        fd.close()
        return False
    if not data:
        log(u"EOF from unlinked %s %s with %d bytes buffered." % (label, safe_format_peername(fd), len(fd.buf)))
        fd.close()
        return False
    else:
        log(u"Data from unlinked %s %s (%d bytes)." % (label, safe_format_peername(fd), len(data)))
        fd.buf += data
        if len(fd.buf) >= UNCONNECTED_BUFFER_LIMIT:
            log(u"Refusing to buffer more than %d bytes from %s %s." % (UNCONNECTED_BUFFER_LIMIT, label, safe_format_peername(fd)))
            fd.close()
            return False
        return True

def match_proxies():
    while unlinked_remotes and unlinked_locals:
        remote = unlinked_remotes.pop(0)
        local = unlinked_locals.pop(0)
        log(u"Linking %s and %s." % (safe_format_peername(local), safe_format_peername(remote)))
        remote.partner = local
        local.partner = remote
        if remote.buf:
            if not proxy_chunk_remote_to_local(remote, local, remote.buf):
                remotes.remove(remote)
                locals.remove(local)
                register()
                return
        if local.buf:
            if not proxy_chunk_local_to_remote(local, remote, local.buf):
                remotes.remove(remote)
                locals.remove(local)
                return

class TimeoutSocket(object):
    def __init__(self, fd):
        self.fd = fd
        self.birthday = time.time()

    def age(self):
        return time.time() - self.birthday

    def __getattr__(self, name):
        return getattr(self.fd, name)

class RemoteSocket(object):
    def __init__(self, fd, protocols):
        self.fd = fd
        self.buf = ""
        self.partner = None
        self.dec = WebSocketBinaryDecoder(protocols, use_mask = True)
        self.enc = WebSocketBinaryEncoder(protocols, use_mask = False)

    def send_chunk(self, data):
        self.sendall(self.enc.encode(data))

    def __getattr__(self, name):
        return getattr(self.fd, name)

class LocalSocket(object):
    def __init__(self, fd):
        self.fd = fd
        self.buf = ""
        self.partner = None

    def send_chunk(self, data):
        self.sendall(data)

    def __getattr__(self, name):
        return getattr(self.fd, name)

def proxy_loop():
    while True:
        rset = remote_listen + local_listen + websocket_pending + socks_pending + locals + remotes
        rset, _, _ = select.select(rset, [], [], WEBSOCKET_REQUEST_TIMEOUT)
        for fd in rset:
            if fd in remote_listen:
                remote_c, addr = fd.accept()
                log(u"Remote connection from %s." % safe_format_sockaddr(addr))
                websocket_pending.append(TimeoutSocket(remote_c))
            elif fd in local_listen:
                local_c, addr = fd.accept()
                log(u"Local connection from %s." % safe_format_sockaddr(addr))
                socks_pending.append(local_c)
                register()
            elif fd in websocket_pending:
                log(u"Data from WebSocket-pending %s." % safe_format_peername(fd))
                protocols = handle_websocket_request(fd)
                if protocols is not None:
                    wrapped = RemoteSocket(fd, protocols)
                    remotes.append(wrapped)
                    unlinked_remotes.append(wrapped)
                else:
                    fd.close()
                    register()
                websocket_pending.remove(fd)
                report_pending()
            elif fd in socks_pending:
                log(u"SOCKS request from %s." % safe_format_peername(fd))
                if handle_socks_request(fd):
                    wrapped = LocalSocket(fd)
                    locals.append(wrapped)
                    unlinked_locals.append(wrapped)
                else:
                    fd.close()
                socks_pending.remove(fd)
                report_pending()
            elif fd in remotes:
                local = fd.partner
                if local:
                    if not proxy_chunk_remote_to_local(fd, local):
                        remotes.remove(fd)
                        locals.remove(local)
                        register()
                else:
                    if not receive_unlinked(fd, "remote"):
                        remotes.remove(fd)
                        unlinked_remotes.remove(fd)
                        register()
                    report_pending()
            elif fd in locals:
                remote = fd.partner
                if remote:
                    if not proxy_chunk_local_to_remote(fd, remote):
                        remotes.remove(remote)
                        locals.remove(fd)
                else:
                    if not receive_unlinked(fd, "local"):
                        locals.remove(fd)
                        unlinked_locals.remove(fd)
                    report_pending()
            match_proxies()
        while websocket_pending:
            pending = websocket_pending[0]
            if pending.age() < WEBSOCKET_REQUEST_TIMEOUT:
                break
            log(u"Expired remote connection from %s." % safe_format_peername(pending))
            pending.close()
            websocket_pending.pop(0)
            report_pending()

def build_register_command(method):
    # sys.path[0] usually contains the directory the script is located in.
    # py2exe overwrites this for bundled executables.
    if getattr(sys, "frozen", False):
        script_dir = os.path.dirname(sys.executable)
    else:
        script_dir = sys.path[0]
    if not script_dir:
        # Maybe the script was read from stdin; in any case don't guess at the directory.
        raise ValueError("Can't find executable directory for registration helpers")

    if method not in ("http", "appspot", "email"):
        raise ValueError("Unknown registration method \"%s\"" % method)

    args = []

    # facilitator selection
    if method == "http":
        if options.facilitator_url is not None:
            args += ["-f", options.facilitator_url]
    else:
        if options.facilitator_pubkey_filename is not None:
            args += ["--facilitator-pubkey", options.facilitator_pubkey_filename]

    # options shared by every registration helper.
    if options.address_family == socket.AF_INET:
        args += ["-4"]
    elif options.address_family == socket.AF_INET6:
        args += ["-6"]
    if options.transport is not None:
        args += ["--transport", options.transport]
    if not options.safe_logging:
        args += ["--unsafe-logging"]

    prog = os.path.join(script_dir, "flashproxy-reg-%s" % method)
    return [prog] + args

def pt_escape(s):
    result = []
    for c in s:
        if c == "\n":
            result.append("\\n")
        elif c == "\\":
            result.append("\\\\")
        elif 0 < ord(c) < 128:
            result.append(c)
        else:
            result.append("\\x%02x" % ord(c))
    return "".join(result)

def pt_line(keyword, *args):
    log(keyword + " " +  " ".join(pt_escape(x) for x in args))
    print keyword, " ".join(pt_escape(x) for x in args)
    sys.stdout.flush()

def pt_cmethoderror(msg):
    pt_line("CMETHOD-ERROR", msg)
    sys.exit(1)

def pt_get_client_transports(known, wildcard = None):
    result = []
    if os.environ.get("TOR_PT_CLIENT_TRANSPORTS") == "*":
        if wildcard is None:
            wildcard = known
        return wildcard
    for method in os.environ.get("TOR_PT_CLIENT_TRANSPORTS", "").split(","):
        if method in known:
            result.append(method)
    return result

def pt_setup_managed():
    TOR_PT_MANAGED_TRANSPORT_VER = os.environ.get("TOR_PT_MANAGED_TRANSPORT_VER")
    if TOR_PT_MANAGED_TRANSPORT_VER is None:
        pt_line("VERSION-ERROR", "no-version")
        print >> sys.stderr, """\

No TOR_PT_MANAGED_TRANSPORT_VER found in environment.
If you are running flashproxy-client from the command line and not from
a ClientTransportPlugin configuration line, you must use the --external
option.\
"""
        sys.exit(1)

    for ver in TOR_PT_MANAGED_TRANSPORT_VER.split(","):
        if ver == "1":
            pt_line("VERSION", ver)
            break
    else:
        pt_line("VERSION-ERROR", "no-version")
        sys.exit(1)

    client_transports = pt_get_client_transports(["flashproxy", "websocket"], ["flashproxy"])
    if not client_transports:
        pt_cmethods_done()
        sys.exit(1)
    return client_transports

def pt_cmethod(method_name, addr):
    pt_line("CMETHOD", method_name, "socks4", format_sockaddr(addr))

def pt_cmethods_done():
    pt_line("CMETHODS", "DONE")

def main():
    global remote_listen, local_listen
    global locals, remotes
    global websocket_pending, socks_pending
    global unlinked_locals, unlinked_remotes

    parser = argparse.ArgumentParser(
        usage="%(prog)s --register [OPTIONS] [LOCAL][:PORT] [REMOTE][:PORT]",
        formatter_class=argparse.RawDescriptionHelpFormatter,
        description="""\
Wait for connections on a local and a remote port. When any pair of connections
exists, data is ferried between them until one side is closed.

The local connection acts as a SOCKS4a proxy, but the host and port in the SOCKS
request are ignored and the local connection is always linked to a remote
connection.

By default, runs as a managed proxy: informs a parent Tor process of support for
the "flashproxy" or "websocket" pluggable transport. In managed mode, the LOCAL
port is chosen arbitrarily instead of the default; this can be overridden by
including a LOCAL port in the command. This is the way the program should be
invoked in a torrc ClientTransportPlugin "exec" line. Use the --external option
to run as an external proxy that does not interact with Tor.

If any of the --register, --register-addr, or --register-methods options are
used, then your IP address will be sent to the facilitator so that proxies can
connect to you. You need to register in some way in order to get any service.
The --facilitator option allows controlling which facilitator is used; if
omitted, it uses a public default.""",
        epilog="""\
The -4, -6, --unsafe-logging, --transport and --facilitator-pubkey options are
propagated to the child registration helpers. For backwards compatilibility,
the --facilitator option is also propagated to the http registration helper.
If you need to pass more options, use TODO #9976.""")
    flashproxy.util.add_module_opts(parser)
    flashproxy.reg.add_module_opts(parser)
    parser.add_argument("-f", "--facilitator", metavar="URL",
        help="register with the facilitator at this URL, default %(default)s. "
        "This is passed to the http registration ONLY.")
    # specific opts and args
    parser.add_argument("--daemon",
        help="daemonize (Unix only).", action="store_true")
    parser.add_argument("--external",
        help="be an external (non-managed) proxy - don't interact with Tor "
        "using environment variables and stdout.", action="store_true")
    parser.add_argument("-l", "--log", metavar="FILENAME",
        help="write log to FILENAME (default stderr).")
    parser.add_argument("--pidfile", metavar="FILENAME",
        help="write PID to FILENAME after daemonizing.")
    parser.add_argument("--port-forwarding",
        help="attempt to forward REMOTE port.", action="store_true")
    parser.add_argument("--port-forwarding-helper", metavar="PROGRAM",
        help="use the given PROGRAM to forward ports, default %s. Implies "
        "--port-forwarding." % DEFAULT_PORT_FORWARDING_HELPER)
    parser.add_argument("--port-forwarding-external", metavar="PORT",
        help="forward the external PORT to REMOTE on the local host, default "
        "same as the REMOTE. Implies --port-forwarding.", type=int)
    parser.add_argument("-r", "--register",
        help="register with the facilitator.", action="store_true")
    parser.add_argument("--register-addr", metavar="ADDR",
        help="register the given address (in case it differs from REMOTE). "
        "Implies --register.")
    parser.add_argument("--register-methods", metavar="METHOD[,METHOD...]",
        help="register using the given comma-separated list of methods. "
        "Implies --register. Possible methods are appspot,email,http. Default "
        "is %s." % ",".join(DEFAULT_REGISTER_METHODS),
        type=lambda x: None if x is None else x.split(",") if x else [])
    parser.add_argument("local_addr", metavar="LOCAL:PORT",
        help="local addr+port to listen on, default all localhost addresses on "
        "port %s. In managed mode, the port is chosen arbitrarily if not given."
        % DEFAULT_LOCAL_PORT_EXTERNAL, default="", nargs="?")
    parser.add_argument("remote_addr", metavar="REMOTE:PORT",
        help="remote addr+port to listen on, default all addresses on port %s"
        % DEFAULT_REMOTE_PORT, default="", nargs="?")

    ns = parser.parse_args(sys.argv[1:])
    # set registration options
    options.address_family = ns.address_family
    options.transport = ns.transport
    options.safe_logging = not ns.unsafe_logging
    options.facilitator_url = ns.facilitator
    options.facilitator_pubkey_filename = ns.facilitator_pubkey

    options.managed = not ns.external

    # do registration if any of the register options were set
    do_register = (ns.register or
                   ns.register_addr is not None or
                   ns.register_methods is not None)

    # do port forwarding if any of the port-forwarding options were set
    do_port_forwarding = (ns.port_forwarding or
                          ns.port_forwarding_helper is not None or
                          ns.port_forwarding_external is not None)

    options.log_filename = ns.log
    options.daemonize = ns.daemon
    options.pid_filename = ns.pidfile

    if options.log_filename:
        options.log_file = open(options.log_filename, "a")
        # Send error tracebacks to the log.
        sys.stderr = options.log_file
    else:
        options.log_file = sys.stderr

    if options.managed:
        method_names = pt_setup_managed()
    else:
        method_names = ["flashproxy"]

    if options.managed:
        default_local_port = DEFAULT_LOCAL_PORT_MANAGED
    else:
        default_local_port = DEFAULT_LOCAL_PORT_EXTERNAL
    default_remote_port = DEFAULT_REMOTE_PORT

    local_addr = parse_addr_spec(ns.local_addr, defhost="", defport=default_local_port)
    remote_addr = parse_addr_spec(ns.remote_addr, defhost="", defport=default_remote_port)

    if local_addr[0]:
        options.local_addrs.append(local_addr)
    else:
        options.local_addrs.append(("127.0.0.1", local_addr[1]))
        # Listen on both IPv4 and IPv6 if no host is given, unless we are in
        # managed mode.
        if not options.managed and socket.has_ipv6:
            options.local_addrs.append(("::1", local_addr[1]))
    if remote_addr[0]:
        options.remote_addrs.append(remote_addr)
    else:
        options.remote_addrs.append(("0.0.0.0", remote_addr[1]))
        if socket.has_ipv6:
            options.remote_addrs.append(("::", remote_addr[1]))

    # Determine registration info if requested.
    options.register = do_register
    register_addr_spec = ns.register_addr
    register_methods = ns.register_methods

    if not register_methods:
        register_methods = DEFAULT_REGISTER_METHODS
    for method in register_methods:
        options.register_commands.append(build_register_command(method))

    options.port_forwarding = do_port_forwarding
    options.port_forwarding_helper = ns.port_forwarding_helper or DEFAULT_PORT_FORWARDING_HELPER
    options.port_forwarding_external = ns.port_forwarding_external

    # Remote sockets, accepting remote WebSocket connections from proxies.
    remote_listen = []
    for addr in options.remote_addrs:
        try:
            listen = listen_socket(addr)
        except socket.error, e:
            log(u"Failed to listen remote on %s: %s." % (addr, str(e)))
            continue
        remote_listen.append(listen)
        log(u"Listening remote on %s." % format_sockaddr(listen.getsockname()))
        if options.register_addr is None:
            host, port = socket.getnameinfo(listen.getsockname(), socket.NI_NUMERICHOST | socket.NI_NUMERICSERV)
            port = int(port)
            if not remote_addr[0]:
                # Make host part blank (not 0.0.0.0) if unspecified.
                host = ""
            options.register_addr = parse_addr_spec(register_addr_spec or ":", host, port)
    if not remote_listen:
        log(u"Failed to open any remote listeners, quitting.")
        pt_cmethoderror("Failed to open any remote listeners.")
    # Local sockets, accepting SOCKS requests from localhost
    local_listen = []
    for addr in options.local_addrs:
        for method_name in method_names:
            try:
                listen = listen_socket(addr)
            except socket.error, e:
                log(u"Failed to listen local on %s: %s." % (addr, str(e)))
                continue
            local_listen.append(listen)
            log(u"Listening local on %s." % format_sockaddr(listen.getsockname()))
            if options.managed:
                pt_cmethod(method_name, listen.getsockname())
    if not local_listen:
        log(u"Failed to open any local listeners, quitting.")
        pt_cmethoderror("Failed to open any local listeners.")
    if options.managed:
        pt_cmethods_done()

    # Attempt to forward ports if requested.
    if options.port_forwarding:
        forward_listeners(remote_listen)

    # New remote sockets waiting to finish their WebSocket negotiation.
    websocket_pending = []
    # Remote connection sockets.
    remotes = []
    # Remotes not yet linked with a local. This is a subset of remotes.
    unlinked_remotes = []
    # New local sockets waiting to finish their SOCKS negotiation.
    socks_pending = []
    # Local Tor sockets, after SOCKS negotiation.
    locals = []
    # Locals not yet linked with a remote. This is a subset of remotes.
    unlinked_locals = []

    if options.daemonize:
        log(u"Daemonizing.")
        pid = os.fork()
        if pid != 0:
            if options.pid_filename:
                f = open(options.pid_filename, "w")
                print >> f, pid
                f.close()
            sys.exit(0)

    if options.register:
        registration_thread = threading.Thread(target=registration_thread_func, name="register")
        registration_thread.daemon = True
        registration_thread.start()
        register()

    try:
        proxy_loop()
    except Exception:
        exc = traceback.format_exc()
        log("".join(exc))

if __name__ == "__main__":
    main()