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). -l sets a logfile to capture message traffic from
17 channels. -n sets the nick and -p the nickserv password. The -V
18 option prints the program version and exits.
20 Design and code by Eric S. Raymond <esr@thyrsus.com>. See the project
21 resource page at <http://www.catb.org/~esr/irker/>.
23 Requires Python 2.6 or 2.5 with the simplejson library installed, and
24 the irc client library at version >= 3.4 which requires 2.6: see
26 http://pypi.python.org/pypi/irc/
28 # These things might need tuning
33 XMIT_TTL = (3 * 60 * 60) # Time to live, seconds from last transmit
34 PING_TTL = (15 * 60) # Time to live, seconds from last PING
35 HANDSHAKE_TTL = 60 # Time to live, seconds from nick transmit
36 CHANNEL_TTL = (3 * 60 * 60) # Time to live, seconds from last transmit
37 DISCONNECT_TTL = (24 * 60 * 60) # Time to live, seconds from last connect
38 UNSEEN_TTL = 60 # Time to live, seconds since first request
39 CHANNEL_MAX = 18 # Max channels open per socket (default)
40 ANTI_FLOOD_DELAY = 0.5 # Anti-flood delay after transmissions, seconds
41 ANTI_BUZZ_DELAY = 0.09 # Anti-buzz delay after queue-empty check
42 CONNECTION_MAX = 200 # To avoid hitting a thread limit
44 # No user-serviceable parts below this line
48 import sys, getopt, urlparse, time, random, socket, signal, re
49 import threading, Queue, SocketServer
50 import irc.client, logging
52 import simplejson as json # Faster, also makes us Python-2.4-compatible
56 # Sketch of implementation:
58 # One Irker object manages multiple IRC sessions. It holds a map of
59 # Dispatcher objects, one per (server, port) combination, which are
60 # responsible for routing messages to one of any number of Connection
61 # objects that do the actual socket conversations. The reason for the
62 # Dispatcher layer is that IRC daemons limit the number of channels a
63 # client (that is, from the daemon's point of view, a socket) can be
64 # joined to, so each session to a server needs a flock of Connection
65 # instances each with its own socket.
67 # Connections are timed out and removed when either they haven't seen a
68 # PING for a while (indicating that the server may be stalled or down)
69 # or there has been no message traffic to them for a while, or
70 # even if the queue is nonempty but efforts to connect have failed for
73 # There are multiple threads. One accepts incoming traffic from all
74 # servers. Each Connection also has a consumer thread and a
75 # thread-safe message queue. The program main appends messages to
76 # queues as JSON requests are received; the consumer threads try to
77 # ship them to servers. When a socket write stalls, it only blocks an
78 # individual consumer thread; if it stalls long enough, the session
79 # will be timed out. This solves the biggest problem with a
80 # single-threaded implementation, which is that you can't count on a
81 # single stalled write not hanging all other traffic - you're at the
82 # mercy of the length of the buffers in the TCP/IP layer.
84 # Message delivery is thus not reliable in the face of network stalls,
85 # but this was considered acceptable because IRC (notoriously) has the
86 # same problem - there is little point in reliable delivery to a relay
87 # that is down or unreliable.
89 # This code uses only NICK, JOIN, PART, MODE, and PRIVMSG. It is strictly
90 # compliant to RFC1459, except for the interpretation and use of the
91 # DEAF and CHANLIMIT and (obsolete) MAXCHANNELS features. CHANLIMIT
92 # is as described in the Internet RFC draft
93 # draft-brocklesby-irc-isupport-03 at <http://www.mirc.com/isupport.html>.
94 # The ",isnick" feature is as described in
95 # <http://ftp.ics.uci.edu/pub/ietf/uri/draft-mirashi-url-irc-01.txt>.
98 def __init__(self, irkerd, servername, port):
100 self.servername = servername
102 self.nick_trial = None
103 self.connection = None
105 self.last_xmit = time.time()
106 self.last_ping = time.time()
107 self.channels_joined = {}
108 self.channel_limits = {}
109 # The consumer thread
110 self.queue = Queue.Queue()
112 def nickname(self, n=None):
113 "Return a name for the nth server connection."
117 return (namestyle % n)
120 def handle_ping(self):
121 "Register the fact that the server has pinged this connection."
122 self.last_ping = time.time()
123 def handle_welcome(self):
124 "The server says we're OK, with a non-conflicting nick."
125 self.status = "ready"
126 self.irker.debug(1, "nick %s accepted" % self.nickname())
128 self.connection.privmsg("nickserv", "identify %s" % password)
129 def handle_badnick(self):
130 "The server says our nick is ill-formed or has a conflict."
131 self.irker.debug(1, "nick %s rejected" % self.nickname())
133 # Randomness prevents a malicious user or bot from
134 # anticipating the next trial name in order to block us
135 # from completing the handshake.
136 self.nick_trial += random.randint(1, 3)
137 self.last_xmit = time.time()
138 self.connection.nick(self.nickname())
139 # Otherwise fall through, it might be possible to
141 def handle_disconnect(self):
142 "Server disconnected us for flooding or some other reason."
143 self.connection = None
144 self.status = "disconnected"
145 def handle_kick(self, outof):
147 self.status = "handshaking"
149 del self.channels_joined[outof]
151 self.irker.logerr("kicked by %s from %s that's not joined"
152 % (self.servername, outof))
154 while not self.queue.empty():
155 (channel, message) = self.queue.get()
157 qcopy.append((channel, message))
158 for (channel, message) in qcopy:
159 self.queue.put((channel, message))
160 self.status = "ready"
161 def enqueue(self, channel, message):
162 "Enque a message for transmission."
163 if self.thread is None or not self.thread.is_alive():
164 self.status = "unseen"
165 self.thread = threading.Thread(target=self.dequeue)
166 self.thread.setDaemon(True)
168 self.queue.put((channel, message))
170 "Try to ship pending messages from the queue."
173 # We want to be kind to the IRC servers and not hold unused
174 # sockets open forever, so they have a time-to-live. The
175 # loop is coded this particular way so that we can drop
176 # the actual server connection when its time-to-live
177 # expires, then reconnect and resume transmission if the
178 # queue fills up again.
179 if self.queue.empty():
180 # Queue is empty, at some point we want to time out
181 # the connection rather than holding a socket open in
182 # the server forever.
184 xmit_timeout = now > self.last_xmit + XMIT_TTL
185 ping_timeout = now > self.last_ping + PING_TTL
186 if self.status == "disconnected":
187 # If the queue is empty, we can drop this connection.
188 self.status = "expired"
190 elif xmit_timeout or ping_timeout:
191 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))
192 with self.irker.irc.mutex:
193 self.connection.context = None
194 self.connection.quit("transmission timeout")
195 self.connection = None
196 self.status = "disconnected"
198 # Prevent this thread from hogging the CPU by pausing
199 # for just a little bit after the queue-empty check.
200 # As long as this is less that the duration of a human
201 # reflex arc it is highly unlikely any human will ever
203 time.sleep(ANTI_BUZZ_DELAY)
204 elif self.status == "disconnected" \
205 and time.time() > self.last_xmit + DISCONNECT_TTL:
206 # Queue is nonempty, but the IRC server might be
207 # down. Letting failed connections retain queue
208 # space forever would be a memory leak.
209 self.status = "expired"
211 elif not self.connection:
212 # Queue is nonempty but server isn't connected.
213 with self.irker.irc.mutex:
214 self.connection = self.irker.irc.server()
215 self.connection.context = self
216 # Try to avoid colliding with other instances
217 self.nick_trial = random.randint(1, 990)
218 self.channels_joined = {}
221 # irc.client.ServerConnectionError on failure
222 self.connection.connect(self.servername,
224 nickname=self.nickname(),
226 ircname="irker relaying client")
227 if hasattr(self.connection, "buffer"):
228 self.connection.buffer.errors = 'replace'
229 self.status = "handshaking"
230 self.irker.debug(1, "XMIT_TTL bump (%s connection) at %s" % (self.servername, time.asctime()))
231 self.last_xmit = time.time()
232 self.last_ping = time.time()
233 except irc.client.ServerConnectionError:
234 self.status = "disconnected"
235 elif self.status == "handshaking":
236 if time.time() > self.last_xmit + HANDSHAKE_TTL:
237 self.status = "expired"
240 # Don't buzz on the empty-queue test while we're
242 time.sleep(ANTI_BUZZ_DELAY)
243 elif self.status == "unseen" \
244 and time.time() > self.last_xmit + UNSEEN_TTL:
245 # Nasty people could attempt a denial-of-service
246 # attack by flooding us with requests with invalid
247 # servernames. We guard against this by rapidly
248 # expiring connections that have a nonempty queue but
249 # have never had a successful open.
250 self.status = "expired"
252 elif self.status == "ready":
253 (channel, message) = self.queue.get()
254 if channel not in self.channels_joined:
255 self.connection.join(channel)
256 self.irker.debug(1, "joining %s on %s." % (channel, self.servername))
257 # An empty message might be used as a keepalive or
258 # to join a channel for logging, so suppress the
259 # privmsg send unless there is actual traffic.
261 for segment in message.split("\n"):
262 # Truncate the message if it's too long,
263 # but we're working with characters here,
264 # not bytes, so we could be off.
265 # 500 = 512 - CRLF - 'PRIVMSG ' - ' :'
266 maxlength = 500 - len(channel)
267 if len(segment) > maxlength:
268 segment = segment[:maxlength]
270 self.connection.privmsg(channel, segment)
271 except ValueError as err:
272 self.irker.debug(1, "irclib rejected a message to %s on %s because: %s" % (channel, self.servername, str(err)))
273 time.sleep(ANTI_FLOOD_DELAY)
274 self.last_xmit = self.channels_joined[channel] = time.time()
275 self.irker.debug(1, "XMIT_TTL bump (%s transmission) at %s" % (self.servername, time.asctime()))
276 self.queue.task_done()
278 (exc_type, _exc_value, exc_traceback) = sys.exc_info()
279 self.irker.logerr("exception %s in thread for %s" % \
280 (exc_type, self.servername))
282 # Maybe this should have its own status?
283 self.status = "expired"
285 # This is so we can see tracebacks for errors inside the thread
286 # when we need to be able to for debugging purposes.
288 raise exc_type, _exc_value, exc_traceback
291 # Make sure we don't leave any zombies behind
292 self.connection.close()
294 # Irclib has a habit of throwing fresh exceptions here. Ignore that
297 "Should this connection not be scavenged?"
298 return self.status != "expired"
299 def joined_to(self, channel):
300 "Is this connection joined to the specified channel?"
301 return channel in self.channels_joined
302 def accepting(self, channel):
303 "Can this connection accept a join of this channel?"
304 if self.channel_limits:
306 for already in self.channels_joined:
307 # This obscure code is because the RFCs allow separate limits
308 # by channel type (indicated by the first character of the name)
309 # a feature that is almost never actually used.
310 if already[0] == channel[0]:
312 return match_count < self.channel_limits.get(channel[0], CHANNEL_MAX)
314 return len(self.channels_joined) < CHANNEL_MAX
317 "Represent a transmission target."
318 def __init__(self, url):
319 # Pre-2.6 Pythons don't recognize irc: as a valid URL prefix.
320 url = url.replace("irc://", "http://")
321 parsed = urlparse.urlparse(url)
322 irchost, _, ircport = parsed.netloc.partition(':')
325 self.servername = irchost
326 # IRC channel names are case-insensitive. If we don't smash
327 # case here we may run into problems later. There was a bug
328 # observed on irc.rizon.net where an irkerd user specified #Channel,
329 # got kicked, and irkerd crashed because the server returned
330 # "#channel" in the notification that our kick handler saw.
331 self.channel = parsed.path.lstrip('/').lower()
332 # This deals with a tweak in recent versions of urlparse.
334 self.channel += "#" + parsed.fragment
335 isnick = self.channel.endswith(",isnick")
337 self.channel = self.channel[:-7]
338 if self.channel and not isnick and self.channel[0] not in "#&+":
339 self.channel = "#" + self.channel
340 self.port = int(ircport)
342 "Both components must be present for a valid target."
343 return self.servername and self.channel
345 "Return a hashable tuple representing the destination server."
346 return (self.servername, self.port)
349 "Manage connections to a particular server-port combination."
350 def __init__(self, irkerd, servername, port):
352 self.servername = servername
354 self.connections = []
355 def dispatch(self, channel, message):
356 "Dispatch messages for our server-port combination."
357 # First, check if there is room for another channel
358 # on any of our existing connections.
359 connections = [x for x in self.connections if x.live()]
360 eligibles = [x for x in connections if x.joined_to(channel)] \
361 or [x for x in connections if x.accepting(channel)]
363 eligibles[0].enqueue(channel, message)
365 # All connections are full up. Look for one old enough to be
368 for connection in connections:
369 for (chan, age) in connections.channels_joined.items():
370 if age < time.time() - CHANNEL_TTL:
371 ancients.append((connection, chan, age))
373 ancients.sort(key=lambda x: x[2])
374 (found_connection, drop_channel, _drop_age) = ancients[0]
375 found_connection.part(drop_channel, "scavenged by irkerd")
376 del found_connection.channels_joined[drop_channel]
377 #time.sleep(ANTI_FLOOD_DELAY)
378 found_connection.enqueue(channel, message)
380 # Didn't find any channels with no recent activity
381 newconn = Connection(self.irker,
384 self.connections.append(newconn)
385 newconn.enqueue(channel, message)
387 "Does this server-port combination have any live connections?"
388 self.connections = [x for x in self.connections if x.live()]
389 return len(self.connections) > 0
391 "Return the time of the most recent transmission."
392 return max(x.last_xmit for x in self.connections)
395 "Persistent IRC multiplexer."
396 def __init__(self, debuglevel=0):
397 self.debuglevel = debuglevel
398 self.irc = irc.client.IRC()
399 self.irc.add_global_handler("ping", self._handle_ping)
400 self.irc.add_global_handler("welcome", self._handle_welcome)
401 self.irc.add_global_handler("erroneusnickname", self._handle_badnick)
402 self.irc.add_global_handler("nicknameinuse", self._handle_badnick)
403 self.irc.add_global_handler("nickcollision", self._handle_badnick)
404 self.irc.add_global_handler("unavailresource", self._handle_badnick)
405 self.irc.add_global_handler("featurelist", self._handle_features)
406 self.irc.add_global_handler("disconnect", self._handle_disconnect)
407 self.irc.add_global_handler("kick", self._handle_kick)
408 self.irc.add_global_handler("all_raw_messages", self._handle_all_raw_messages)
409 thread = threading.Thread(target=self.irc.process_forever)
410 thread.setDaemon(True)
411 self.irc._thread = thread
414 def logerr(self, errmsg):
415 "Log a processing error."
416 sys.stderr.write("irkerd: " + errmsg + "\n")
417 def debug(self, level, errmsg):
418 "Debugging information."
419 if self.debuglevel >= level:
420 sys.stderr.write("irkerd: %s\n" % errmsg)
421 def _handle_ping(self, connection, _event):
422 "PING arrived, bump the last-received time for the connection."
423 if connection.context:
424 connection.context.handle_ping()
425 def _handle_welcome(self, connection, _event):
426 "Welcome arrived, nick accepted for this connection."
427 if connection.context:
428 connection.context.handle_welcome()
429 def _handle_badnick(self, connection, _event):
430 "Nick not accepted for this connection."
431 if connection.context:
432 connection.context.handle_badnick()
433 def _handle_features(self, connection, event):
434 "Determine if and how we can set deaf mode."
435 if connection.context:
436 cxt = connection.context
437 arguments = event.arguments
438 # irclib 5.0 compatibility, because the maintainer is a fool
439 if callable(arguments):
440 arguments = arguments()
441 for lump in arguments:
442 if lump.startswith("DEAF="):
444 connection.mode(cxt.nickname(), "+"+lump[5:])
445 elif lump.startswith("MAXCHANNELS="):
448 cxt.channel_limits[pref] = m
449 self.debug(1, "%s maxchannels is %d"
450 % (connection.server, m))
451 elif lump.startswith("CHANLIMIT=#:"):
452 limits = lump[10:].split(",")
455 (prefixes, limit) = token.split(":")
458 cxt.channel_limits[c] = limit
459 self.debug(1, "%s channel limit map is %s"
460 % (connection.server, cxt.channel_limits))
462 self.logerr("ill-formed CHANLIMIT property")
463 def _handle_disconnect(self, connection, _event):
464 "Server hung up the connection."
465 self.debug(1, "server %s disconnected" % connection.server)
467 if connection.context:
468 connection.context.handle_disconnect()
469 def _handle_kick(self, connection, event):
470 "Server hung up the connection."
471 target = event.target
472 # irclib 5.0 compatibility, because the maintainer continues
476 self.debug(1, "irker has been kicked from %s on %s" % (target, connection.server))
477 if connection.context:
478 connection.context.handle_kick(target)
479 def _handle_all_raw_messages(self, _connection, event):
480 "Log all messages when in watcher mode."
481 with open(logfile, "a") as logfp:
482 logfp.write("%03f|%s|%s\n" % \
483 (time.time(), event.source, event.arguments[0]))
484 def handle(self, line):
485 "Perform a JSON relay request."
487 request = json.loads(line.strip())
488 if not isinstance(request, dict):
489 self.logerr("request is not a JSON dictionary: %r" % request)
490 elif "to" not in request or "privmsg" not in request:
491 self.logerr("malformed request - 'to' or 'privmsg' missing: %r" % request)
493 channels = request['to']
494 message = request['privmsg']
495 if not isinstance(channels, (list, basestring)):
496 self.logerr("malformed request - unexpected channel type: %r" % channels)
497 if not isinstance(message, basestring):
498 self.logerr("malformed request - unexpected message type: %r" % message)
500 if not isinstance(channels, list):
501 channels = [channels]
503 if not isinstance(url, basestring):
504 self.logerr("malformed request - URL has unexpected type: %r" % url)
507 if not target.valid():
509 if target.server() not in self.servers:
510 self.servers[target.server()] = Dispatcher(self, target.servername, target.port)
511 self.servers[target.server()].dispatch(target.channel, message)
512 # GC dispatchers with no active connections
513 servernames = self.servers.keys()
514 for servername in servernames:
515 if not self.servers[servername].live():
516 del self.servers[servername]
517 # If we might be pushing a resource limit
518 # even after garbage collection, remove a
519 # session. The goal here is to head off
520 # DoS attacks that aim at exhausting
521 # thread space or file descriptors. The
522 # cost is that attempts to DoS this
523 # service will cause lots of join/leave
524 # spam as we scavenge old channels after
525 # connecting to new ones. The particular
526 # method used for selecting a session to
527 # be terminated doesn't matter much; we
528 # choose the one longest idle on the
529 # assumption that message activity is likely
531 if len(self.servers) >= CONNECTION_MAX:
532 oldest = min(self.servers.keys(), key=lambda name: self.servers[name].last_xmit())
533 del self.servers[oldest]
535 self.logerr("can't recognize JSON on input: %r" % line)
537 self.logerr("wildly malformed JSON blew the parser stack.")
539 class IrkerTCPHandler(SocketServer.StreamRequestHandler):
542 line = self.rfile.readline()
545 irker.handle(line.strip())
547 class IrkerUDPHandler(SocketServer.BaseRequestHandler):
549 data = self.request[0].strip()
550 #socket = self.request[1]
553 if __name__ == '__main__':
555 namestyle = "irker%03d"
558 (options, arguments) = getopt.getopt(sys.argv[1:], "d:l:n:p:V:")
559 for (opt, val) in options:
560 if opt == '-d': # Enable debug/progress messages
563 logging.basicConfig(level=logging.DEBUG)
564 elif opt == '-l': # Logfile mode - report traffic read in
566 elif opt == '-n': # Force the nick
568 elif opt == '-p': # Set a nickserv password
570 elif opt == '-V': # Emit version and exit
571 sys.stdout.write("irkerd version %s\n" % version)
573 fallback = re.search("%.*d", namestyle)
574 irker = Irker(debuglevel=debuglvl)
575 irker.debug(1, "irkerd version %s" % version)
577 tcpserver = SocketServer.TCPServer((HOST, PORT), IrkerTCPHandler)
578 udpserver = SocketServer.UDPServer((HOST, PORT), IrkerUDPHandler)
579 for server in [tcpserver, udpserver]:
580 server = threading.Thread(target=server.serve_forever)
581 server.setDaemon(True)
585 except KeyboardInterrupt:
587 except socket.error, e:
588 sys.stderr.write("irkerd: server launch failed: %r\n" % e)