3 irkerd - a simple IRC multiplexer daemon
5 Listens for JSON objects of the form {'to':<irc-url>, 'privmsg':<text>}
6 and relays messages to IRC channels. Each request must be followed by
9 The <text> must be a string. The value of the 'to' attribute can be a
10 string containing an IRC URL (e.g. 'irc://chat.freenet.net/botwar') or
11 a list of such strings; in the latter case the message is broadcast to
12 all listed channels. Note that the channel portion of the URL need
13 *not* have a leading '#' unless the channel name itself does.
15 Options: -d sets the debug-message level (probably only of interest to
16 developers). The -V option prints the program version and exits.
18 Design and code by Eric S. Raymond <esr@thyrsus.com>. See the project
19 resource page at <http://www.catb.org/~esr/irker/>.
21 Requires Python 2.6 or 2.5 with the simplejson library installed, and
22 the irc client library at version >= 2.0.2: see
24 http://pypi.python.org/pypi/irc/
26 # These things might need tuning
31 NAMESTYLE = "irker%03d" # IRC nick template - must contain '%d'
32 XMIT_TTL = (3 * 60 * 60) # Time to live, seconds from last transmit
33 PING_TTL = (15 * 60) # Time to live, seconds from last PING
34 HANDSHAKE_TTL = 60 # Time to live, seconds from nick transmit
35 CHANNEL_TTL = (3 * 60 * 60) # Time to live, seconds from last transmit
36 DISCONNECT_TTL = (24 * 60 * 60) # Time to live, seconds from last connect
37 UNSEEN_TTL = 60 # Time to live, seconds since first request
38 CHANNEL_MAX = 18 # Max channels open per socket (default)
39 ANTI_FLOOD_DELAY = 0.5 # Anti-flood delay after transmissions, seconds
40 ANTI_BUZZ_DELAY = 0.09 # Anti-buzz delay after queue-empty check
42 # No user-serviceable parts below this line
46 # This black magic imports support for green threads (coroutines),
47 # then has kinky sex with the import library internals, replacing
48 # "threading" with a coroutine-using imposter. Threads then become
49 # ultra-light-weight and cooperatively scheduled.
52 eventlet.monkey_patch()
54 # With greenlets we don't worry about thread exhaustion, only the
55 # file descriptor limit (typically 1024 on modern Unixes). Thus we
56 # can handle a lot more concurrent sessions and generate less
57 # join/leave spam under heavy load.
60 # Threads are more expensive if we have to use OS-level ones
61 # rather than greenlets. We need to avoid pushing thread limits
62 # as well as fd limits. See security.txt for discussion.
66 import sys, getopt, urlparse, time, random, socket
67 import threading, Queue, SocketServer
68 import irc.client, logging
70 import simplejson as json # Faster, also makes us Python-2.4-compatible
74 # Sketch of implementation:
76 # One Irker object manages multiple IRC sessions. It holds a map of
77 # Dispatcher objects, one per (server, port) combination, which are
78 # responsible for routing messages to one of any number of Connection
79 # objects that do the actual socket conversations. The reason for the
80 # Dispatcher layer is that IRC daemons limit the number of channels a
81 # client (that is, from the daemon's point of view, a socket) can be
82 # joined to, so each session to a server needs a flock of Connection
83 # instances each with its own socket.
85 # Connections are timed out and removed when either they haven't seen a
86 # PING for a while (indicating that the server may be stalled or down)
87 # or there has been no message traffic to them for a while, or
88 # even if the queue is nonempty but efforts to connect have failed for
91 # There are multiple threads. One accepts incoming traffic from all servers.
92 # Each Connection also has a consumer thread and a thread-safe message queue.
93 # The program main appends messages to queues as JSON requests are received;
94 # the consumer threads try to ship them to servers. When a socket write
95 # stalls, it only blocks an individual consumer thread; if it stalls long
96 # enough, the session will be timed out.
98 # Message delivery is thus not reliable in the face of network stalls,
99 # but this was considered acceptable because IRC (notoriously) has the
100 # same problem - there is little point in reliable delivery to a relay
101 # that is down or unreliable.
103 # This code uses only NICK, JOIN, PART, MODE, and PRIVMSG. It is strictly
104 # compliant to RFC1459, except for the interpretation and use of the
105 # DEAF and CHANLIMIT and (obsolete) MAXCHANNELS features. CHANLIMIT
106 # is as described in the Internet RFC draft
107 # draft-brocklesby-irc-isupport-03 at <http://www.mirc.com/isupport.html>.
108 # The ",isnick" feature is as described in
109 # <http://ftp.ics.uci.edu/pub/ietf/uri/draft-mirashi-url-irc-01.txt>.
112 def __init__(self, irkerd, servername, port):
114 self.servername = servername
116 self.nick_trial = None
117 self.connection = None
119 self.last_xmit = time.time()
120 self.last_ping = time.time()
121 self.channels_joined = {}
122 self.channel_limits = {}
123 # The consumer thread
124 self.queue = Queue.Queue()
126 def nickname(self, n=None):
127 "Return a name for the nth server connection."
130 return (NAMESTYLE % n)
131 def handle_ping(self):
132 "Register the fact that the server has pinged this connection."
133 self.last_ping = time.time()
134 def handle_welcome(self):
135 "The server says we're OK, with a non-conflicting nick."
136 self.status = "ready"
137 self.irker.debug(1, "nick %s accepted" % self.nickname())
138 def handle_badnick(self):
139 "The server says our nick has a conflict."
140 self.irker.debug(1, "nick %s rejected" % self.nickname())
141 # Randomness prevents a malicious user or bot from antcipating the
142 # next trial name in order to block us from completing the handshake.
143 self.nick_trial += random.randint(1, 3)
144 self.last_xmit = time.time()
145 self.connection.nick(self.nickname())
146 def handle_disconnect(self):
147 "Server disconnected us for flooding or some other reason."
148 self.connection = None
149 self.status = "disconnected"
150 def handle_kick(self, outof):
152 self.status = "handshaking"
154 del self.channels_joined[outof]
156 self.irker.logerr("kicked by %s from %s that's not joined"
157 % (self.servername, outof))
159 while not self.queue.empty():
160 (channel, message) = self.queue.get()
162 qcopy.append((channel, message))
163 for (channel, message) in qcopy:
164 self.queue.put((channel, message))
165 self.status = "ready"
166 def enqueue(self, channel, message):
167 "Enque a message for transmission."
168 if self.thread is None or not self.thread.is_alive():
169 self.status = "unseen"
170 self.thread = threading.Thread(target=self.dequeue)
171 self.thread.setDaemon(True)
173 self.queue.put((channel, message))
175 "Try to ship pending messages from the queue."
178 # We want to be kind to the IRC servers and not hold unused
179 # sockets open forever, so they have a time-to-live. The
180 # loop is coded this particular way so that we can drop
181 # the actual server connection when its time-to-live
182 # expires, then reconnect and resume transmission if the
183 # queue fills up again.
184 if self.queue.empty():
185 # Queue is empty, at some point we want to time out
186 # the connection rather than holding a socket open in
187 # the server forever.
189 xmit_timeout = now > self.last_xmit + XMIT_TTL
190 ping_timeout = now > self.last_ping + PING_TTL
191 if self.status == "disconnected":
192 # If the queue is empty, we can drop this connection.
193 self.status = "expired"
195 elif xmit_timeout or ping_timeout:
196 self.irker.debug(1, "timing out connection to %s at %s (ping_timeout=%s, xmit_timeout=%s)" % (self.servername, time.asctime(), ping_timeout, xmit_timeout))
197 with self.irker.library_lock:
198 self.connection.context = None
199 self.connection.quit("transmission timeout")
200 self.connection = None
201 self.status = "disconnected"
203 # Prevent this thread from hogging the CPU by pausing
204 # for just a little bit after the queue-empty check.
205 # As long as this is less that the duration of a human
206 # reflex arc it is highly unlikely any human will ever
208 time.sleep(ANTI_BUZZ_DELAY)
209 elif self.status == "disconnected" \
210 and time.time() > self.last_xmit + DISCONNECT_TTL:
211 # Queue is nonempty, but the IRC server might be
212 # down. Letting failed connections retain queue
213 # space forever would be a memory leak.
214 self.status = "expired"
216 elif not self.connection:
217 # Queue is nonempty but server isn't connected.
218 with self.irker.library_lock:
219 self.connection = self.irker.irc.server()
220 self.connection.context = self
221 # Try to avoid colliding with other instances
222 self.nick_trial = random.randint(1, 990)
223 self.channels_joined = {}
226 # irc.client.ServerConnectionError on failure
227 self.connection.connect(self.servername,
229 nickname=self.nickname(),
231 ircname="irker relaying client")
232 self.status = "handshaking"
233 self.irker.debug(1, "XMIT_TTL bump (%s connection) at %s" % (self.servername, time.asctime()))
234 self.last_xmit = time.time()
235 except irc.client.ServerConnectionError:
236 self.status = "disconnected"
237 elif self.status == "handshaking":
238 if time.time() > self.last_xmit + HANDSHAKE_TTL:
239 self.status = "expired"
242 # Don't buzz on the empty-queue test while we're
244 time.sleep(ANTI_BUZZ_DELAY)
245 elif self.status == "unseen" \
246 and time.time() > self.last_xmit + UNSEEN_TTL:
247 # Nasty people could attempt a denial-of-service
248 # attack by flooding us with requests with invalid
249 # servernames. We guard against this by rapidly
250 # expiring connections that have a nonempty queue but
251 # have never had a successful open.
252 self.status = "expired"
254 elif self.status == "ready":
255 with self.irker.library_lock:
256 (channel, message) = self.queue.get()
257 if channel not in self.channels_joined:
258 self.connection.join(channel)
259 self.irker.debug(1, "joining %s on %s." % (channel, self.servername))
260 for segment in message.split("\n"):
261 self.connection.privmsg(channel, segment)
262 time.sleep(ANTI_FLOOD_DELAY)
263 self.last_xmit = self.channels_joined[channel] = time.time()
264 self.irker.debug(1, "XMIT_TTL bump (%s transmission) at %s" % (self.servername, time.asctime()))
265 self.queue.task_done()
267 (exc_type, _exc_value, exc_traceback) = sys.exc_info()
268 self.irker.logerr("exception %s in thread for %s" % \
269 (exc_type, self.servername))
270 # This is so we can see tracebacks for errors inside the thread
271 # when we need to be able to for debugging purposes.
273 raise exc_type, _exc_value, exc_traceback
275 "Should this connection not be scavenged?"
276 return self.status != "expired"
277 def joined_to(self, channel):
278 "Is this connection joined to the specified channel?"
279 return channel in self.channels_joined
280 def accepting(self, channel):
281 "Can this connection accept a join of this channel?"
282 if self.channel_limits:
284 for already in self.channels_joined:
285 # This obscure code is because the RFCs allow separate limits
286 # by channel type (indicated by the first character of the name)
287 # a feature that is almost never actually used.
288 if already[0] == channel[0]:
290 return match_count < self.channel_limits.get(channel[0], CHANNEL_MAX)
292 return len(self.channels_joined) < CHANNEL_MAX
295 "Represent a transmission target."
296 def __init__(self, url):
297 parsed = urlparse.urlparse(url)
298 irchost, _, ircport = parsed.netloc.partition(':')
301 self.servername = irchost
302 # IRC channel names are case-insensitive. If we don't smash
303 # case here we may run into problems later. There was a bug
304 # observed on irc.rizon.net where an irkerd user specified #Channel,
305 # got kicked, and irkerd crashed because the server returned
306 # "#channel" in the notification that our kick handler saw.
307 self.channel = parsed.path.lstrip('/').lower()
308 isnick = self.channel.endswith(",isnick")
310 self.channel = self.channel[:-7]
311 if self.channel and not isnick and self.channel[0] not in "#&+":
312 self.channel = "#" + self.channel
313 self.port = int(ircport)
315 "Both components must be present for a valid target."
316 return self.servername and self.channel
318 "Return a hashable tuple representing the destination server."
319 return (self.servername, self.port)
322 "Manage connections to a particular server-port combination."
323 def __init__(self, irkerd, servername, port):
325 self.servername = servername
327 self.connections = []
328 def dispatch(self, channel, message):
329 "Dispatch messages for our server-port combination."
330 # First, check if there is room for another channel
331 # on any of our existing connections.
332 connections = [x for x in self.connections if x.live()]
333 eligibles = [x for x in connections if x.joined_to(channel)] \
334 or [x for x in connections if x.accepting(channel)]
336 eligibles[0].enqueue(channel, message)
338 # All connections are full up. Look for one old enough to be
341 for connection in connections:
342 for (chan, age) in connections.channels_joined.items():
343 if age < time.time() - CHANNEL_TTL:
344 ancients.append((connection, chan, age))
346 ancients.sort(key=lambda x: x[2])
347 (found_connection, drop_channel, _drop_age) = ancients[0]
348 found_connection.part(drop_channel, "scavenged by irkerd")
349 del found_connection.channels_joined[drop_channel]
350 #time.sleep(ANTI_FLOOD_DELAY)
351 found_connection.enqueue(channel, message)
353 # Didn't find any channels with no recent activity
354 newconn = Connection(self.irker,
357 self.connections.append(newconn)
358 newconn.enqueue(channel, message)
360 "Does this server-port combination have any live connections?"
361 self.connections = [x for x in self.connections if x.live()]
362 return len(self.connections) > 0
364 "Return the time of the most recent transmission."
365 return max(x.last_xmit for x in self.connections)
368 "Persistent IRC multiplexer."
369 def __init__(self, debuglevel=0):
370 self.debuglevel = debuglevel
371 self.irc = irc.client.IRC()
372 self.irc.add_global_handler("ping", self._handle_ping)
373 self.irc.add_global_handler("welcome", self._handle_welcome)
374 self.irc.add_global_handler("erroneusnickname", self._handle_badnick)
375 self.irc.add_global_handler("nicknameinuse", self._handle_badnick)
376 self.irc.add_global_handler("nickcollision", self._handle_badnick)
377 self.irc.add_global_handler("unavailresource", self._handle_badnick)
378 self.irc.add_global_handler("featurelist", self._handle_features)
379 self.irc.add_global_handler("disconnect", self._handle_disconnect)
380 self.irc.add_global_handler("kick", self._handle_kick)
381 self.library_lock = threading.Lock()
382 thread = threading.Thread(target=self._process_forever)
383 thread.setDaemon(True)
384 self.irc._thread = thread
387 def logerr(self, errmsg):
388 "Log a processing error."
389 sys.stderr.write("irkerd: " + errmsg + "\n")
390 def debug(self, level, errmsg):
391 "Debugging information."
392 if self.debuglevel >= level:
393 sys.stderr.write("irkerd: %s\n" % errmsg)
394 def _process_forever(self):
395 "IRC library process_forever with mutex."
396 self.debug(1, "process_forever()")
398 with self.library_lock:
399 self.irc.process_once(ANTI_BUZZ_DELAY)
400 def _handle_ping(self, connection, _event):
401 "PING arrived, bump the last-received time for the connection."
402 if connection.context:
403 connection.context.handle_ping()
404 def _handle_welcome(self, connection, _event):
405 "Welcome arrived, nick accepted for this connection."
406 if connection.context:
407 connection.context.handle_welcome()
408 def _handle_badnick(self, connection, _event):
409 "Nick not accepted for this connection."
410 if connection.context:
411 connection.context.handle_badnick()
412 def _handle_features(self, connection, event):
413 "Determine if and how we can set deaf mode."
414 if connection.context:
415 cxt = connection.context
416 for lump in event.arguments():
417 if lump.startswith("DEAF="):
418 connection.mode(cxt.nickname(), "+"+lump[5:])
419 elif lump.startswith("MAXCHANNELS="):
422 cxt.channel_limits[pref] = m
423 self.debug(1, "%s maxchannels is %d"
424 % (connection.server, m))
425 elif lump.startswith("CHANLIMIT=#:"):
426 limits = lump[10:].split(",")
429 (prefixes, limit) = token.split(":")
432 cxt.channel_limits[c] = limit
433 self.debug(1, "%s channel limit map is %s"
434 % (connection.server, cxt.channel_limits))
436 self.logerr("ill-formed CHANLIMIT property")
437 def _handle_disconnect(self, connection, _event):
438 "Server hung up the connection."
439 self.debug(1, "server %s disconnected" % connection.server)
441 if connection.context:
442 connection.context.handle_disconnect()
443 def _handle_kick(self, connection, event):
444 "Server hung up the connection."
445 self.debug(1, "irker has been kicked from %s on %s" % (event.target(), connection.server))
446 if connection.context:
447 connection.context.handle_kick(event.target())
448 def handle(self, line):
449 "Perform a JSON relay request."
451 request = json.loads(line.strip())
452 if not isinstance(request, dict):
453 self.logerr("request is not a JSON dictionary: %r" % request)
454 elif "to" not in request or "privmsg" not in request:
455 self.logerr("malformed request - 'to' or 'privmsg' missing: %r" % request)
457 channels = request['to']
458 message = request['privmsg']
459 if not isinstance(channels, (list, basestring)):
460 self.logerr("malformed request - unexpected channel type: %r" % channels)
461 if not isinstance(message, basestring):
462 self.logerr("malformed request - unexpected message type: %r" % message)
464 if not isinstance(channels, list):
465 channels = [channels]
467 if not isinstance(url, basestring):
468 self.logerr("malformed request - URL has unexpected type: %r" % url)
471 if not target.valid():
473 if target.server() not in self.servers:
474 self.servers[target.server()] = Dispatcher(self, target.servername, target.port)
475 self.servers[target.server()].dispatch(target.channel, message)
476 # GC dispatchers with no active connections
477 servernames = self.servers.keys()
478 for servername in servernames:
479 if not self.servers[servername].live():
480 del self.servers[servername]
481 # If we might be pushing a resource limit
482 # even after garbage collection, remove a
483 # session. The goal here is to head off
484 # DoS attacks that aim at exhausting
485 # thread space or file descriptors. The
486 # cost is that attempts to DoS this
487 # service will cause lots of join/leave
488 # spam as we scavenge old channels after
489 # connecting to new ones. The particular
490 # method used for selecting a session to
491 # be terminated doesn't matter much; we
492 # choose the one longest idle on the
493 # assumption that message activity is likely
495 if len(self.servers) >= CONNECTION_MAX:
496 oldest = min(self.servers.keys(), key=lambda name: self.servers[name].last_xmit())
497 del self.servers[oldest]
499 self.logerr("can't recognize JSON on input: %r" % line)
501 self.logerr("wildly malformed JSON blew the parser stack.")
503 class IrkerTCPHandler(SocketServer.StreamRequestHandler):
506 line = self.rfile.readline()
509 irker.handle(line.strip())
511 class IrkerUDPHandler(SocketServer.BaseRequestHandler):
513 data = self.request[0].strip()
514 #socket = self.request[1]
517 if __name__ == '__main__':
519 (options, arguments) = getopt.getopt(sys.argv[1:], "d:V")
520 for (opt, val) in options:
521 if opt == '-d': # Enable debug/progress messages
524 logging.basicConfig(level=logging.DEBUG)
525 elif opt == '-V': # Emit version and exit
526 sys.stdout.write("irkerd version %s\n" % version)
528 irker = Irker(debuglevel=debuglvl)
529 irker.debug(1, "irkerd version %s" % version)
531 tcpserver = SocketServer.TCPServer((HOST, PORT), IrkerTCPHandler)
532 udpserver = SocketServer.UDPServer((HOST, PORT), IrkerUDPHandler)
533 for server in [tcpserver, udpserver]:
534 server = threading.Thread(target=server.serve_forever)
535 server.setDaemon(True)
540 except KeyboardInterrupt:
542 except socket.error, e:
543 sys.stderr.write("irkerd: server launch failed: %r\n" % e)