-
Eric S. Raymond authoredEric S. Raymond authored
Code owners
Assign users and groups as approvers for specific file changes. Learn more.
irkerd 19.71 KiB
#!/usr/bin/env python
"""
irkerd - a simple IRC multiplexer daemon
Listens for JSON objects of the form {'to':<irc-url>, 'privmsg':<text>}
and relays messages to IRC channels. Each request must be followed by
a newline.
The <text> must be a string. The value of the 'to' attribute can be a
string containing an IRC URL (e.g. 'irc://chat.freenet.net/botwar') or
a list of such strings; in the latter case the message is broadcast to
all listed channels. Note that the channel portion of the URL need
*not* have a leading '#' unless the channel name itself does.
Options: -d sets the debug-message level (probably only of interest to
developers). The -V option prints the program version and exits.
Design and code by Eric S. Raymond <esr@thyrsus.com>. See the project
resource page at <http://www.catb.org/~esr/irker/>.
Requires Python 2.6 and the irc client library at version >= 2.0.2: see
http://pypi.python.org/pypi/irc/
"""
# These things might need tuning
HOST = "localhost"
PORT = 6659
NAMESTYLE = "irker%03d" # IRC nick template - must contain '%d'
XMIT_TTL = (3 * 60 * 60) # Time to live, seconds from last transmit
PING_TTL = (15 * 60) # Time to live, seconds from last PING
DISCONNECT_TTL = (24 * 60 * 60) # Time to live, seconds from last connect
UNSEEN_TTL = 60 # Time to live, seconds since first request
CHANNEL_MAX = 18 # Max channels open per socket (default)
ANTI_FLOOD_DELAY = 0.125 # Anti-flood delay after transmissions, seconds
ANTI_BUZZ_DELAY = 0.09 # Anti-buzz delay after queue-empty check
# No user-serviceable parts below this line
# This black magic imports support for green threads (coroutines),
# then has kinky sex with the import library internals, replacing
# "threading" with a coroutine-using impostor. Threads then become
# ultra-light-weight and cooperatively scheduled.
try:
import eventlet; eventlet.monkey_patch()
green_threads = True
# With greenlets we don't worry about thread exhaustion, only the
# file descriptor limit (typically 1024 on modern Unixes). Thus we
# can handle a lot more concurrent sessions and generate less
# join/leave spam under heavy load.
CONNECTION_MAX = 1000
except ImportError:
# Threads are more expensive if we have to use OS-level ones
# rather than greenlets. We need to avoid pushing thread limits
# as well as fd limits. See security.txt for discussion.
CONNECTION_MAX = 200
green_threads = False
import sys, json, getopt, urlparse, time, random
import threading, Queue, SocketServer
import irc.client, logging
version = "1.2"
# Sketch of implementation:
#
# One Irker object manages multiple IRC sessions. It holds a map of
# Dispatcher objects, one per (server, port) combination, which are
# responsible for routing messages to one of any number of Connection
# objects that do the actual socket conversations. The reason for the
# Dispatcher layer is that IRC daemons limit the number of channels a
# client (that is, from the daemon's point of view, a socket) can be
# joined to, so each session to a server needs a flock of Connection
# instances each with its own socket.
#
# Connections are timed out and removed when either they haven't seen a
# PING for a while (indicating that the server may be stalled or down)
# or there has been no message traffic to them for a while, or
# even if the queue is nonempty but efforts to connect have failed for
# a long time.
#
# There are multiple threads. One accepts incoming traffic from all servers.
# Each Connection also has a consumer thread and a thread-safe message queue.
# The program main appends messages to queues as JSON requests are received;
# the consumer threads try to ship them to servers. When a socket write
# stalls, it only blocks an individual consumer thread; if it stalls long
# enough, the session will be timed out.
#
# Message delivery is thus not reliable in the face of network stalls,
# but this was considered acceptable because IRC (notoriously) has the
# same problem - there is little point in reliable delivery to a relay
# that is down or unreliable.
#
# This code uses only NICK, JOIN, MODE, and PRIVMSG. It is strictly
# compliant to RFC1459, except for the interpretation and use of the
# DEAF and CHANLIMIT and (obsolete) MAXCHANNELS features. CHANLIMIT
# is as described in the Internet RFC draft
# draft-brocklesby-irc-isupport-03 at <http://www.mirc.com/isupport.html>.
class Connection:
def __init__(self, irkerd, servername, port, nick_base=1):
self.irker = irkerd
self.servername = servername
self.port = port
self.nick_trial = nick_base
self.connection = None
self.status = "unseen"
self.last_xmit = time.time()
self.last_ping = time.time()
self.channels_joined = []
self.channel_limits = {}
# The consumer thread
self.queue = Queue.Queue()
self.thread = threading.Thread(target=self.dequeue)
self.thread.start()
def nickname(self, n=None):
"Return a name for the nth server connection."
if n is None:
n = self.nick_trial
return (NAMESTYLE % n)
def handle_ping(self):
"Register the fact that the server has pinged this connection."
self.last_ping = time.time()
def handle_welcome(self):
"The server says we're OK, with a non-conflicting nick."
self.status = "ready"
self.irker.debug(1, "nick %s accepted" % self.nickname())
def handle_badnick(self):
"The server says our nick has a conflict."
self.irker.debug(1, "nick %s rejected" % self.nickname())
# Randomness prevents a malicious user or bot from antcipating the
# next trial name in order to block us from completing the handshake.
self.nick_trial += random.randint(1, 3)
self.connection.nick(self.nickname())
def enqueue(self, channel, message):
"Enque a message for transmission."
self.queue.put((channel, message))
def dequeue(self):
"Try to ship pending messages from the queue."
while True:
# We want to be kind to the IRC servers and not hold unused
# sockets open forever, so they have a time-to-live. The
# loop is coded this particular way so that we can drop
# the actual server connection when its time-to-live
# expires, then reconnect and resume transmission if the
# queue fills up again.
if not self.connection:
self.connection = self.irker.irc.server()
self.connection.context = self
self.nick_trial = 1
self.channels_joined = []
# This will throw irc.client.ServerConnectionError on failure
try:
self.connection.connect(self.servername,
self.port,
nickname=self.nickname(),
username="irker",
ircname="irker relaying client")
self.status = "handshaking"
self.irker.debug(1, "XMIT_TTL bump (%s connection) at %s" % (self.servername, time.asctime()))
self.last_xmit = time.time()
except irc.client.ServerConnectionError:
self.status = "disconnected"
elif self.status == "handshaking":
# Don't buzz on the empty-queue test while we're handshaking
time.sleep(ANTI_BUZZ_DELAY)
elif self.queue.empty():
# Queue is empty, at some point we want to time out
# the connection rather than holding a socket open in
# the server forever.
now = time.time()
if now > self.last_xmit + XMIT_TTL \
or now > self.last_ping + PING_TTL:
self.irker.debug(1, "timing out inactive connection to %s at %s" % (self.servername, time.asctime()))
self.connection.context = None
self.connection.close()
self.connection = None
self.status = "disconnected"
else:
# Prevent this thread from hogging the CPU by pausing
# for just a little bit after the queue-empty check.
# As long as this is less that the duration of a human
# reflex arc it is highly unlikely any human will ever
# notice.
time.sleep(ANTI_BUZZ_DELAY)
elif self.status == "disconnected" \
and time.time() > self.last_xmit + DISCONNECT_TTL:
# Queue is nonempty, but the IRC server might be down. Letting
# failed connections retain queue space forever would be a
# memory leak.
self.status = "expired"
break
elif self.status == "unseen" \
and time.time() > self.last_xmit + UNSEEN_TTL:
# Nasty people could attempt a denial-of-service
# attack by flooding us with requests with invalid
# servernames. We guard against this by rapidly
# expiring connections that have a nonempty queue but
# have never had a successful open.
self.status = "expired"
break
elif self.status == "ready":
(channel, message) = self.queue.get()
if channel not in self.channels_joined:
self.channels_joined.append(channel)
if channel[0] not in "#&+":
channel = "#" + channel
self.connection.join(channel)
for segment in message.split("\n"):
self.connection.privmsg(channel, segment)
self.last_xmit = time.time()
self.irker.debug(1, "XMIT_TTL bump (%s transmission) at %s" % (self.servername, time.asctime()))
self.queue.task_done()
time.sleep(ANTI_FLOOD_DELAY)
def live(self):
"Should this connection not be scavenged?"
return self.status != "expired"
def joined_to(self, channel):
"Is this connection joined to the specified channel?"
return channel in self.channels_joined
def accepting(self, channel):
"Can this connection accept a join of this channel?"
if self.channel_limits:
match_count = 0
for already in self.channels_joined:
if already[0] == channel[0]:
match_count += 1
return match_count < self.channel_limits.get(channel[0], CHANNEL_MAX)
else:
return len(self.channels_joined) < CHANNEL_MAX
class Target():
"Represent a transmission target."
def __init__(self, url):
parsed = urlparse.urlparse(url)
irchost, _, ircport = parsed.netloc.partition(':')
if not ircport:
ircport = 6667
self.servername = irchost
self.channel = parsed.path.lstrip('/')
self.port = int(ircport)
def valid(self):
"Both components must be present for a valid target."
self.servername and self.channel
def server(self):
"Return a hashable tuple representing the destination server."
return (self.servername, self.port)
class Dispatcher:
"Manage connections to a particular server-port combination."
def __init__(self, irkerd, servername, port):
self.irker = irkerd
self.servername = servername
self.port = port
self.connections = []
def dispatch(self, channel, message):
"Dispatch messages for our server-port combination."
connections = [x for x in self.connections if x.live()]
eligibles = [x for x in connections if x.joined_to(channel)] \
or [x for x in connections if x.accepting(channel)]
if not eligibles:
newconn = Connection(self.irker,
self.servername,
self.port,
len(self.connections)+1)
self.connections.append(newconn)
eligibles = [newconn]
eligibles[0].enqueue(channel, message)
def live(self):
"Does this server-port combination have any live connections?"
self.connections = [x for x in self.connections if x.live()]
return len(self.connections) > 0
class Irker:
"Persistent IRC multiplexer."
def __init__(self, debuglevel=0):
self.debuglevel = debuglevel
self.irc = irc.client.IRC()
self.irc.add_global_handler("ping", self._handle_ping)
self.irc.add_global_handler("welcome", self._handle_welcome)
self.irc.add_global_handler("erroneusnickname", self._handle_badnick)
self.irc.add_global_handler("nicknameinuse", self._handle_badnick)
self.irc.add_global_handler("nickcollision", self._handle_badnick)
self.irc.add_global_handler("unavailresource", self._handle_badnick)
self.irc.add_global_handler("featurelist", self._handle_features)
thread = threading.Thread(target=self.irc.process_forever)
self.irc._thread = thread
thread.start()
self.servers = {}
def logerr(self, errmsg):
"Log a processing error."
sys.stderr.write("irkerd: " + errmsg + "\n")
def debug(self, level, errmsg):
"Debugging information."
if self.debuglevel >= level:
sys.stderr.write("irkerd: %s\n" % errmsg)
def _handle_ping(self, connection, _event):
"PING arrived, bump the last-received time for the connection."
if connection.context:
connection.context.handle_ping()
def _handle_welcome(self, connection, _event):
"Welcome arrived, nick accepted for this connection."
if connection.context:
connection.context.handle_welcome()
def _handle_badnick(self, connection, _event):
"Nick not accepted for this connection."
if connection.context:
connection.context.handle_badnick()
def _handle_features(self, connection, event):
"Determine if and how we can set deaf mode."
if connection.context:
cxt = connection.context
for lump in event.arguments():
if lump.startswith("DEAF="):
connection.mode(cxt.nickname(), "+"+lump[5:])
elif lump.startswith("MAXCHANNELS="):
m = int(lump[12:])
for pref in "#&+":
cxt.channel_limits[pref] = m
self.debug(1, "%s maxchannels is %d" \
% (connection.server, m))
elif lump.startswith("CHANLIMIT=#:"):
limits = lump[10:].split(",")
try:
for token in limits:
(prefixes, limit) = token.split(":")
limit = int(limit)
for c in prefixes:
cxt.channel_limits[c] = limit
self.debug(1, "%s channel limit map is %s" \
% (connection.server, cxt.channel_limits))
except ValueError:
self.logerr("ill-formed CHANLIMIT property")
def handle(self, line):
"Perform a JSON relay request."
try:
request = json.loads(line.strip())
if type(request) != type({}):
self.logerr("request in tot a JSON dictionary: %s" % repr(request))
elif "to" not in request or "privmsg" not in request:
self.logerr("malformed reqest - 'to' or 'privmsg' missing: %s" % repr(request))
else:
channels = request['to']
message = request['privmsg']
if type(channels) not in (type([]), type(u"")) \
or type(message) != type(u""):
self.logerr("malformed request - unexpected types: %s" % repr(request))
else:
if type(channels) == type(u""):
channels = [channels]
for url in channels:
if type(url) != type(u""):
self.logerr("malformed request - unexpected type: %s" % repr(request))
else:
target = Target(url)
if not target.valid():
return
if target.server() not in self.servers:
self.servers[target.server()] = Dispatcher(self, target.servername, target.port)
self.servers[target.server()].dispatch(target.channel, message)
# GC dispatchers with no active connections
servernames = self.servers.keys()
for servername in servernames:
if not self.servers[servername].live():
del self.servers[servername]
# If we might be pushing a resource limit
# even after garbage collection, remove a
# session. The goal here is to head off
# DoS attacks that aim at exhausting
# thread space or file descriptors. The
# cost is that attempts to DoS this
# service will cause lots of join/leave
# spam as we scavenge old channels after
# connecting to new ones. The particular
# method used for selecting a session to
# be terminated doesn't matter much; we
# choose the one longest idle on the
# assumption that message activity is likely
# to be clumpy.
oldest = None
if len(self.servers) >= CONNECTION_MAX:
for (name, server) in self.servers.items():
if not oldest or server.last_xmit < self.servers[oldest].last_xmit:
oldest = name
del self.servers[oldest]
except ValueError:
self.logerr("can't recognize JSON on input: %s" % repr(line))
class IrkerTCPHandler(SocketServer.StreamRequestHandler):
def handle(self):
while True:
line = self.rfile.readline()
if not line:
break
irker.handle(line.strip())
class IrkerUDPHandler(SocketServer.BaseRequestHandler):
def handle(self):
data = self.request[0].strip()
#socket = self.request[1]
irker.handle(data)
if __name__ == '__main__':
debuglvl = 0
(options, arguments) = getopt.getopt(sys.argv[1:], "d:V")
for (opt, val) in options:
if opt == '-d': # Enable debug/progress messages
debuglvl = int(val)
if debuglvl > 1:
logging.basicConfig(level=logging.DEBUG)
elif opt == '-V': # Emit version and exit
sys.stdout.write("irkerd version %s\n" % version)
sys.exit(0)
irker = Irker(debuglevel=debuglvl)
tcpserver = SocketServer.TCPServer((HOST, PORT), IrkerTCPHandler)
udpserver = SocketServer.UDPServer((HOST, PORT), IrkerUDPHandler)
threading.Thread(target=tcpserver.serve_forever).start()
threading.Thread(target=udpserver.serve_forever).start()
# Main thread has to stay alive forever for the cooperative
# scheduling of the green threads to work.
if green_threads:
while True:
time.sleep(10)
# end