relay: fix race conditions when multiple networks disconnect simultaneously

plugins/relay.py

@@ -199,22 +199,8 @@ def getPrefixModes(irc, remoteirc, channel, user, mlist=None):
                 modes += remoteirc.cmodes[pmode]
     return modes
 
-def getRemoteSid(irc, remoteirc):
+def spawnRelayServer(irc, remoteirc):
-    """Gets the remote server SID representing remoteirc on irc, spawning
+    irc.connected.wait()
-    it if it doesn't exist."""
-    # Don't spawn servers too early.
-    irc.connected.wait(2)
-
-    try:
-        spawnservers = irc.conf['relay']['spawn_servers']
-    except KeyError:
-        spawnservers = True
-    if not spawnservers:
-        return irc.sid
-    with spawnlocks_servers[irc.name]:
-        try:
-            sid = relayservers[irc.name][remoteirc.name]
-        except KeyError:
     try:
         # ENDBURST is delayed by 3 secs on supported IRCds to prevent
         # triggering join-flood protection and the like.
@@ -229,28 +215,49 @@ def getRemoteSid(irc, remoteirc):
         handle_disconnect(irc, None, 'PYLINK_DISCONNECT_RELAY_FORCED', {})
         irc.disconnect()
         raise
-            else:
+
+    # Mark the server as a relay server
     irc.servers[sid].remote = remoteirc.name
+
+    # Assign the newly spawned server as our relay server for the target net.
     relayservers[irc.name][remoteirc.name] = sid
+
     return sid
 
-def getRemoteUser(irc, remoteirc, user, spawnIfMissing=True):
+def getRemoteSid(irc, remoteirc):
-    """Gets the UID of the relay client for the given IRC network/user pair,
+    """Gets the remote server SID representing remoteirc on irc, spawning
-    spawning one if it doesn't exist and spawnIfMissing is True."""
+    it if it doesn't exist."""
-    # If the user (stored here as {('netname', 'UID'):
+
-    # {'network1': 'UID1', 'network2': 'UID2'}}) exists, don't spawn it
-    # again!
     try:
-        if user == irc.pseudoclient.uid:
+        spawnservers = irc.conf['relay']['spawn_servers']
-            return remoteirc.pseudoclient.uid
-    except AttributeError:  # Network hasn't been initialized yet?
-        pass
-    with spawnlocks[irc.name]:
-        try:
-            u = relayusers[(irc.name, user)][remoteirc.name]
     except KeyError:
+        spawnservers = True
+    if not spawnservers:
+        return irc.sid
+
+    with spawnlocks_servers[irc.name]:
+        try:
+            sid = relayservers[irc.name][remoteirc.name]
+        except KeyError:
+            log.debug('(%s) getRemoteSid: %s.relay doesn\'t have a known SID, spawning.', irc.name, remoteirc.name)
+            sid = spawnRelayServer(irc, remoteirc)
+
+        log.debug('(%s) getRemoteSid: got %s for %s.relay', irc.name, sid, remoteirc.name)
+        if sid not in irc.servers:
+            log.debug('(%s) getRemoteSid: SID %s for %s.relay doesn\'t exist, respawning', irc.name, sid, remoteirc.name)
+            # Our stored server doesn't exist anymore. This state is probably a holdover from a netsplit,
+            # so let's refresh it.
+            sid = spawnRelayServer(irc, remoteirc)
+        elif sid in irc.servers and irc.servers[sid].remote != remoteirc.name:
+            log.debug('(%s) getRemoteSid: %s.relay != %s.relay, respawning', irc.name, irc.servers[sid].remote, remoteirc.name)
+            sid = spawnRelayServer(irc, remoteirc)
+
+        log.debug('(%s) getRemoteSid: got %s for %s.relay (round 2)', irc.name, sid, remoteirc.name)
+        return sid
+
+def spawnRelayUser(irc, remoteirc, user):
     userobj = irc.users.get(user)
-            if userobj is None or (not spawnIfMissing) or (not remoteirc.connected.is_set()):
+    if userobj is None:
         # The query wasn't actually a valid user, or the network hasn't
         # been connected yet... Oh well!
         return
@@ -310,9 +317,39 @@ def getRemoteUser(irc, remoteirc, user, spawnIfMissing=True):
     away = userobj.away
     if away:
         remoteirc.proto.away(u, away)
+
     relayusers[(irc.name, user)][remoteirc.name] = u
     return u
 
+def getRemoteUser(irc, remoteirc, user, spawnIfMissing=True):
+    """Gets the UID of the relay client for the given IRC network/user pair,
+    spawning one if it doesn't exist and spawnIfMissing is True."""
+
+    log.debug('(%s) getRemoteUser: waiting for irc.connected', irc.name)
+    irc.connected.wait()
+    log.debug('(%s) getRemoteUser: waiting for %s.connected', irc.name, remoteirc.name)
+    remoteirc.connected.wait()
+
+    # If the user (stored here as {('netname', 'UID'):
+    # {'network1': 'UID1', 'network2': 'UID2'}}) exists, don't spawn it
+    # again!
+    try:
+        if user == irc.pseudoclient.uid:
+            return remoteirc.pseudoclient.uid
+    except AttributeError:  # Network hasn't been initialized yet?
+        pass
+    with spawnlocks[irc.name]:
+        u = None
+        try:
+            u = relayusers[(irc.name, user)][remoteirc.name]
+        except KeyError:
+            if spawnIfMissing:
+                u = spawnRelayUser(irc, remoteirc, user)
+
+        if u and ((u not in remoteirc.users) or remoteirc.users[u].remote != (irc.name, user)):
+            spawnRelayUser(irc, remoteirc, user)
+        return u
+
 def getOrigUser(irc, user, targetirc=None):
     """
     Given the UID of a relay client, returns a tuple of the home network name
@@ -757,10 +794,12 @@ def handle_join(irc, numeric, command, args):
 utils.add_hook(handle_join, 'JOIN')
 
 def handle_quit(irc, numeric, command, args):
+    with spawnlocks[irc.name]:
         for netname, user in relayusers[(irc.name, numeric)].copy().items():
             remoteirc = world.networkobjects[netname]
             remoteirc.proto.quit(user, args['text'])
         del relayusers[(irc.name, numeric)]
+
 utils.add_hook(handle_quit, 'QUIT')
 
 def handle_squit(irc, numeric, command, args):
@@ -1164,16 +1203,13 @@ def handle_disconnect(irc, numeric, command, args):
             if irc.name in v:
                 del relayusers[k][irc.name]
             if k[0] == irc.name:
-                try:
                 handle_quit(irc, k[1], 'PYLINK_DISCONNECT', {'text': 'Relay network lost connection.'})
-                    del relayusers[k]
+
-                except KeyError:
-                    pass
     # SQUIT all relay pseudoservers spawned for us, and remove them
     # from our relay subservers index.
     with spawnlocks_servers[irc.name]:
         for name, ircobj in world.networkobjects.copy().items():
-            if name != irc.name and ircobj.connected.is_set():
+            if name != irc.name:
                 try:
                     rsid = relayservers[name][irc.name]
                 except KeyError:
@@ -1181,15 +1217,10 @@ def handle_disconnect(irc, numeric, command, args):
                 else:
                     ircobj.proto.squit(ircobj.sid, rsid, text='Relay network lost connection.')
 
-            try:
+            if irc.name in relayservers[name]:
                 del relayservers[name][irc.name]
-            except KeyError:
-                pass
 
-        try:
         del relayservers[irc.name]
-        except KeyError:
-            pass
 
 utils.add_hook(handle_disconnect, "PYLINK_DISCONNECT")