Version bump for 1.18 release.
[irker.git] / irkerd
diff --git a/irkerd b/irkerd
index 1fe96a119e68d61b287fd8649fa0db7312fadfb9..2b2a9dffd63634f094bdeb3dc73078de6b933c0f 100755 (executable)
--- a/irkerd
+++ b/irkerd
@@ -1,4 +1,5 @@
 #!/usr/bin/env python
+from __future__ import with_statement
 """
 irkerd - a simple IRC multiplexer daemon
 
@@ -13,12 +14,15 @@ 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.
+developers). -l sets a logfile to capture message traffic from
+channels.  -n sets the nick and -p the nickserv password. 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
+Requires Python 2.6 or 2.5 with the simplejson library installed, and
+the irc client library at version >= 3.4 which requires 2.6: see
 
 http://pypi.python.org/pypi/irc/
 """
@@ -27,7 +31,6 @@ http://pypi.python.org/pypi/irc/
 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
 HANDSHAKE_TTL = 60             # Time to live, seconds from nick transmit
@@ -37,32 +40,13 @@ UNSEEN_TTL = 60                     # Time to live, seconds since first request
 CHANNEL_MAX = 18               # Max channels open per socket (default)
 ANTI_FLOOD_DELAY = 0.5         # Anti-flood delay after transmissions, seconds
 ANTI_BUZZ_DELAY = 0.09         # Anti-buzz delay after queue-empty check
+CONNECTION_MAX = 200           # To avoid hitting a thread limit
 
 # No user-serviceable parts below this line
 
-version = "1.8"
+version = "1.18"
 
-# This black magic imports support for green threads (coroutines),
-# then has kinky sex with the import library internals, replacing
-# "threading" with a coroutine-using imposter.  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, getopt, urlparse, time, random, socket
+import sys, getopt, urlparse, time, random, socket, signal, re
 import threading, Queue, SocketServer
 import irc.client, logging
 try:
@@ -87,19 +71,23 @@ except ImportError:
 # 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.
+# 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. This solves the biggest problem with a
+# single-threaded implementation, which is that you can't count on a
+# single stalled write not hanging all other traffic - you're at the
+# mercy of the length of the buffers in the TCP/IP layer.
 #
 # 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
+# This code uses only NICK, JOIN, PART, 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
@@ -126,7 +114,10 @@ class Connection:
         "Return a name for the nth server connection."
         if n is None:
             n = self.nick_trial
-        return (NAMESTYLE % n)
+        if fallback:
+            return (namestyle % n)
+        else:
+            return namestyle
     def handle_ping(self):
         "Register the fact that the server has pinged this connection."
         self.last_ping = time.time()
@@ -134,14 +125,20 @@ class Connection:
         "The server says we're OK, with a non-conflicting nick."
         self.status = "ready"
         self.irker.debug(1, "nick %s accepted" % self.nickname())
+        if password:
+            self.connection.privmsg("nickserv", "identify %s" % password)
     def handle_badnick(self):
-        "The server says our nick has a conflict."
+        "The server says our nick is ill-formed or 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.last_xmit = time.time()
-        self.connection.nick(self.nickname())
+        if fallback:
+            # Randomness prevents a malicious user or bot from
+            # anticipating the next trial name in order to block us
+            # from completing the handshake.
+            self.nick_trial += random.randint(1, 3)
+            self.last_xmit = time.time()
+            self.connection.nick(self.nickname())
+        # Otherwise fall through, it might be possible to
+        # recover manually.
     def handle_disconnect(self):
         "Server disconnected us for flooding or some other reason."
         self.connection = None
@@ -187,12 +184,16 @@ class Connection:
                     now = time.time()
                     xmit_timeout = now > self.last_xmit + XMIT_TTL
                     ping_timeout = now > self.last_ping + PING_TTL
-                    if (xmit_timeout or ping_timeout) and self.status != "disconnected":
+                    if self.status == "disconnected":
+                        # If the queue is empty, we can drop this connection.
+                        self.status = "expired"
+                        break
+                    elif xmit_timeout or ping_timeout:
                         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))
-                        self.connection.context = None
-                        self.connection.quit("transmission timeout")
-                        self.connection.close()
-                        self.connection = None
+                        with self.irker.irc.mutex:
+                            self.connection.context = None
+                            self.connection.quit("transmission timeout")
+                            self.connection = None
                         self.status = "disconnected"
                     else:
                         # Prevent this thread from hogging the CPU by pausing
@@ -201,25 +202,37 @@ class Connection:
                         # 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 not self.connection:
                     # Queue is nonempty but server isn't connected.
-                    self.connection = self.irker.irc.server()
-                    self.connection.context = self
-                    # Try to avoid colliding with other instances
-                    self.nick_trial = random.randint(1, 990)
-                    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"
+                    with self.irker.irc.mutex:
+                        self.connection = self.irker.irc.server()
+                        self.connection.context = self
+                        # Try to avoid colliding with other instances
+                        self.nick_trial = random.randint(1, 990)
+                        self.channels_joined = {}
+                        try:
+                            # This will throw
+                            # irc.client.ServerConnectionError on failure
+                            self.connection.connect(self.servername,
+                                                self.port,
+                                                nickname=self.nickname(),
+                                                username="irker",
+                                                ircname="irker relaying client")
+                            if hasattr(self.connection, "buffer"):
+                                self.connection.buffer.errors = 'replace'
+                            self.status = "handshaking"
+                            self.irker.debug(1, "XMIT_TTL bump (%s connection) at %s" % (self.servername, time.asctime()))
+                            self.last_xmit = time.time()
+                            self.last_ping = time.time()
+                        except irc.client.ServerConnectionError:
+                            self.status = "disconnected"
                 elif self.status == "handshaking":
                     if time.time() > self.last_xmit + HANDSHAKE_TTL:
                         self.status = "expired"
@@ -228,13 +241,6 @@ class Connection:
                         # Don't buzz on the empty-queue test while we're
                         # handshaking
                         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
@@ -249,9 +255,23 @@ class Connection:
                     if channel not in self.channels_joined:
                         self.connection.join(channel)
                         self.irker.debug(1, "joining %s on %s." % (channel, self.servername))
-                    for segment in message.split("\n"):
-                        self.connection.privmsg(channel, segment)
-                        time.sleep(ANTI_FLOOD_DELAY)
+                    # An empty message might be used as a keepalive or
+                    # to join a channel for logging, so suppress the
+                    # privmsg send unless there is actual traffic.
+                    if message:
+                        for segment in message.split("\n"):
+                            # Truncate the message if it's too long,
+                            # but we're working with characters here,
+                            # not bytes, so we could be off.
+                            # 500 = 512 - CRLF - 'PRIVMSG ' - ' :'
+                            maxlength = 500 - len(channel)
+                            if len(segment) > maxlength:
+                                segment = segment[:maxlength]
+                            try:
+                                self.connection.privmsg(channel, segment)
+                            except ValueError as err:
+                                self.irker.debug(1, "irclib rejected a message to %s on %s because: %s" % (channel, self.servername, str(err)))
+                            time.sleep(ANTI_FLOOD_DELAY)
                     self.last_xmit = self.channels_joined[channel] = time.time()
                     self.irker.debug(1, "XMIT_TTL bump (%s transmission) at %s" % (self.servername, time.asctime()))
                     self.queue.task_done()
@@ -259,10 +279,21 @@ class Connection:
             (exc_type, _exc_value, exc_traceback) = sys.exc_info()
             self.irker.logerr("exception %s in thread for %s" % \
                               (exc_type, self.servername))
+
+            # Maybe this should have its own status?
+            self.status = "expired"
+
             # This is so we can see tracebacks for errors inside the thread
             # when we need to be able to for debugging purposes.
             if debuglvl > 0:
                 raise exc_type, _exc_value, exc_traceback
+        finally:
+            try:
+                # Make sure we don't leave any zombies behind
+                self.connection.close()
+            except:
+                # Irclib has a habit of throwing fresh exceptions here. Ignore that
+                pass
     def live(self):
         "Should this connection not be scavenged?"
         return self.status != "expired"
@@ -274,6 +305,9 @@ class Connection:
         if self.channel_limits:
             match_count = 0
             for already in self.channels_joined:
+                # This obscure code is because the RFCs allow separate limits
+                # by channel type (indicated by the first character of the name)
+                # a feature that is almost never actually used.
                 if already[0] == channel[0]:
                     match_count += 1
             return match_count < self.channel_limits.get(channel[0], CHANNEL_MAX)
@@ -283,6 +317,8 @@ class Connection:
 class Target():
     "Represent a transmission target."
     def __init__(self, url):
+        # Pre-2.6 Pythons don't recognize irc: as a valid URL prefix.
+        url = url.replace("irc://", "http://")
         parsed = urlparse.urlparse(url)
         irchost, _, ircport = parsed.netloc.partition(':')
         if not ircport:
@@ -294,6 +330,9 @@ class Target():
         # got kicked, and irkerd crashed because the server returned
         # "#channel" in the notification that our kick handler saw.
         self.channel = parsed.path.lstrip('/').lower()
+        # This deals with a tweak in recent versions of urlparse.
+        if parsed.fragment:
+            self.channel += "#" + parsed.fragment
         isnick = self.channel.endswith(",isnick")
         if isnick:
             self.channel = self.channel[:-7]
@@ -351,7 +390,7 @@ class Dispatcher:
         return len(self.connections) > 0
     def last_xmit(self):
         "Return the time of the most recent transmission."
-        return max([x.last_xmit for x in self.connections])
+        return max(x.last_xmit for x in self.connections)
 
 class Irker:
     "Persistent IRC multiplexer."
@@ -367,6 +406,7 @@ class Irker:
         self.irc.add_global_handler("featurelist", self._handle_features)
         self.irc.add_global_handler("disconnect", self._handle_disconnect)
         self.irc.add_global_handler("kick", self._handle_kick)
+        self.irc.add_global_handler("all_raw_messages", self._handle_all_raw_messages)
         thread = threading.Thread(target=self.irc.process_forever)
         thread.setDaemon(True)
         self.irc._thread = thread
@@ -395,9 +435,14 @@ class Irker:
         "Determine if and how we can set deaf mode."
         if connection.context:
             cxt = connection.context
-            for lump in event.arguments():
+            arguments = event.arguments
+            # irclib 5.0 compatibility, because the maintainer is a fool
+            if callable(arguments):
+                arguments = arguments()
+            for lump in arguments:
                 if lump.startswith("DEAF="):
-                    connection.mode(cxt.nickname(), "+"+lump[5:])
+                    if not logfile:
+                        connection.mode(cxt.nickname(), "+"+lump[5:])
                 elif lump.startswith("MAXCHANNELS="):
                     m = int(lump[12:])
                     for pref in "#&+":
@@ -419,13 +464,25 @@ class Irker:
     def _handle_disconnect(self, connection, _event):
         "Server hung up the connection."
         self.debug(1, "server %s disconnected" % connection.server)
+        connection.close()
         if connection.context:
             connection.context.handle_disconnect()
     def _handle_kick(self, connection, event):
         "Server hung up the connection."
-        self.debug(1, "irker has been kicked from %s on %s" % (event.target(), connection.server))
+        target = event.target
+        # irclib 5.0 compatibility, because the maintainer continues
+        # to be a fool.
+        if callable(target):
+            target = target()
+        self.debug(1, "irker has been kicked from %s on %s" % (target, connection.server))
         if connection.context:
-            connection.context.handle_kick(event.target())
+            connection.context.handle_kick(target)
+    def _handle_all_raw_messages(self, _connection, event):
+        "Log all messages when in watcher mode."
+        if logfile:
+            with open(logfile, "a") as logfp:
+                logfp.write("%03f|%s|%s\n" % \
+                             (time.time(), event.source, event.arguments[0]))
     def handle(self, line):
         "Perform a JSON relay request."
         try:
@@ -473,13 +530,8 @@ class Irker:
                             # choose the one longest idle on the
                             # assumption that message activity is likely
                             # to be clumpy.
-                            oldest = None
-                            oldtime = float("inf")
                             if len(self.servers) >= CONNECTION_MAX:
-                                for (name, server) in self.servers.items():
-                                    if server.last_xmit() < oldtime:
-                                        oldest = name
-                                        oldtime = server.last_xmit()
+                                oldest = min(self.servers.keys(), key=lambda name: self.servers[name].last_xmit())
                                 del self.servers[oldest]
         except ValueError:
             self.logerr("can't recognize JSON on input: %r" % line)
@@ -502,15 +554,25 @@ class IrkerUDPHandler(SocketServer.BaseRequestHandler):
 
 if __name__ == '__main__':
     debuglvl = 0
-    (options, arguments) = getopt.getopt(sys.argv[1:], "d:V")
+    namestyle = "irker%03d"
+    password = None
+    logfile = None
+    (options, arguments) = getopt.getopt(sys.argv[1:], "d:l:n:p: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 == '-l':      # Logfile mode - report traffic read in
+            logfile = val
+        elif opt == '-n':      # Force the nick
+            namestyle = val
+        elif opt == '-p':      # Set a nickserv password
+            password = val
         elif opt == '-V':      # Emit version and exit
             sys.stdout.write("irkerd version %s\n" % version)
             sys.exit(0)
+    fallback = re.search("%.*d", namestyle)
     irker = Irker(debuglevel=debuglvl)
     irker.debug(1, "irkerd version %s" % version)
     try:
@@ -521,8 +583,7 @@ if __name__ == '__main__':
             server.setDaemon(True)
             server.start()
         try:
-            while True:
-                time.sleep(10)
+            signal.pause()
         except KeyboardInterrupt:
             raise SystemExit(1)
     except socket.error, e: