1: #!/usr/bin/perl
2: # The LearningOnline Network with CAPA
3: # lonc maintains the connections to remote computers
4: #
5: # $Id: loncnew,v 1.65 2004/10/05 11:16:57 foxr Exp $
6: #
7: # Copyright Michigan State University Board of Trustees
8: #
9: # This file is part of the LearningOnline Network with CAPA (LON-CAPA).
10: ## LON-CAPA is free software; you can redistribute it and/or modify
11: # it under the terms of the GNU General Public License as published by
12: # the Free Software Foundation; either version 2 of the License, or
13: # (at your option) any later version.
14: #
15: # LON-CAPA is distributed in the hope that it will be useful,
16: # but WITHOUT ANY WARRANTY; without even the implied warranty of
17: # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18: # GNU General Public License for more details.
19: #
20: # You should have received a copy of the GNU General Public License
21: # along with LON-CAPA; if not, write to the Free Software
22: # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23: #
24: # /home/httpd/html/adm/gpl.txt
25: #
26: # http://www.lon-capa.org/
27: #
28: #
29: # new lonc handles n request out bver m connections to londs.
30: # This module is based on the Event class.
31: # Development iterations:
32: # - Setup basic event loop. (done)
33: # - Add timer dispatch. (done)
34: # - Add ability to accept lonc UNIX domain sockets. (done)
35: # - Add ability to create/negotiate lond connections (done).
36: # - Add general logic for dispatching requests and timeouts. (done).
37: # - Add support for the lonc/lond requests. (done).
38: # - Add logging/status monitoring. (done)
39: # - Add Signal handling - HUP restarts. USR1 status report. (done)
40: # - Add Configuration file I/O (done).
41: # - Add management/status request interface. (done)
42: # - Add deferred request capability. (done)
43: # - Detect transmission timeouts. (done)
44: #
45:
46: use strict;
47: use lib "/home/httpd/lib/perl/";
48: use Event qw(:DEFAULT );
49: use POSIX qw(:signal_h);
50: use POSIX;
51: use IO::Socket;
52: use IO::Socket::INET;
53: use IO::Socket::UNIX;
54: use IO::File;
55: use IO::Handle;
56: use Socket;
57: use Crypt::IDEA;
58: use LONCAPA::Queue;
59: use LONCAPA::Stack;
60: use LONCAPA::LondConnection;
61: use LONCAPA::LondTransaction;
62: use LONCAPA::Configuration;
63: use LONCAPA::HashIterator;
64:
65:
66: # Read the httpd configuration file to get perl variables
67: # normally set in apache modules:
68:
69: my $perlvarref = LONCAPA::Configuration::read_conf('loncapa.conf');
70: my %perlvar = %{$perlvarref};
71:
72: #
73: # parent and shared variables.
74:
75: my %ChildHash; # by pid -> host.
76: my %HostToPid; # By host -> pid.
77: my %HostHash; # by loncapaname -> IP.
78: my %listening_to; # Socket->host table for who the parent
79: # is listening to.
80: my %parent_dispatchers; # host-> listener watcher events.
81:
82: my %parent_handlers; # Parent signal handlers...
83:
84: my $MaxConnectionCount = 10; # Will get from config later.
85: my $ClientConnection = 0; # Uniquifier for client events.
86:
87: my $DebugLevel = 0;
88: my $NextDebugLevel= 2; # So Sigint can toggle this.
89: my $IdleTimeout= 600; # Wait 10 minutes before pruning connections.
90:
91: my $LogTransactions = 0; # When True, all transactions/replies get logged.
92: my $executable = $0; # Get the full path to me.
93:
94: #
95: # The variables below are only used by the child processes.
96: #
97: my $RemoteHost; # Name of host child is talking to.
98: my $UnixSocketDir= $perlvar{'lonSockDir'};
99: my $IdleConnections = Stack->new(); # Set of idle connections
100: my %ActiveConnections; # Connections to the remote lond.
101: my %ActiveTransactions; # LondTransactions in flight.
102: my %ActiveClients; # Serial numbers of active clients by socket.
103: my $WorkQueue = Queue->new(); # Queue of pending transactions.
104: my $ConnectionCount = 0;
105: my $IdleSeconds = 0; # Number of seconds idle.
106: my $Status = ""; # Current status string.
107: my $RecentLogEntry = "";
108: my $ConnectionRetries=2; # Number of connection retries allowed.
109: my $ConnectionRetriesLeft=2; # Number of connection retries remaining.
110: my $LondVersion = "unknown"; # Version of lond we talk with.
111: my $KeyMode = ""; # e.g. ssl, local, insecure from last connect.
112: my $LondConnecting = 0; # True when a connection is being built.
113:
114:
115:
116: my $DieWhenIdle = 1; # When true children die when trimmed -> 0.
117: my $I_am_child = 0; # True if this is the child process.
118:
119: #
120: # The hash below gives the HTML format for log messages
121: # given a severity.
122: #
123: my %LogFormats;
124:
125: $LogFormats{"CRITICAL"} = "<font color='red'>CRITICAL: %s</font>";
126: $LogFormats{"SUCCESS"} = "<font color='green'>SUCCESS: %s</font>";
127: $LogFormats{"INFO"} = "<font color='yellow'>INFO: %s</font>";
128: $LogFormats{"WARNING"} = "<font color='blue'>WARNING: %s</font>";
129: $LogFormats{"DEFAULT"} = " %s ";
130:
131:
132: # UpdateStatus;
133: # Update the idle status display to show how many connections
134: # are left, retries and other stuff.
135: #
136: sub UpdateStatus {
137: if ($ConnectionRetriesLeft > 0) {
138: ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount
139: ." Retries remaining: ".$ConnectionRetriesLeft
140: ." ($KeyMode)");
141: } else {
142: ShowStatus(GetServerHost()." >> DEAD <<");
143: }
144: }
145:
146:
147: =pod
148:
149: =head2 LogPerm
150:
151: Makes an entry into the permanent log file.
152:
153: =cut
154: sub LogPerm {
155: my $message=shift;
156: my $execdir=$perlvar{'lonDaemons'};
157: my $now=time;
158: my $local=localtime($now);
159: my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
160: print $fh "$now:$message:$local\n";
161: }
162:
163: =pod
164:
165: =head2 Log
166:
167: Logs a message to the log file.
168: Parameters:
169:
170: =item severity
171:
172: One of CRITICAL, WARNING, INFO, SUCCESS used to select the
173: format string used to format the message. if the severity is
174: not a defined severity the Default format string is used.
175:
176: =item message
177:
178: The base message. In addtion to the format string, the message
179: will be appended to a string containing the name of our remote
180: host and the time will be formatted into the message.
181:
182: =cut
183:
184: sub Log {
185:
186: my ($severity, $message) = @_;
187:
188: if(!$LogFormats{$severity}) {
189: $severity = "DEFAULT";
190: }
191:
192: my $format = $LogFormats{$severity};
193:
194: # Put the window dressing in in front of the message format:
195:
196: my $now = time;
197: my $local = localtime($now);
198: my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
199: my $finalformat = $finalformat.$format."\n";
200:
201: # open the file and put the result.
202:
203: my $execdir = $perlvar{'lonDaemons'};
204: my $fh = IO::File->new(">>$execdir/logs/lonc.log");
205: my $msg = sprintf($finalformat, $message);
206: $RecentLogEntry = $msg;
207: print $fh $msg;
208:
209:
210: }
211:
212:
213: =pod
214:
215: =head2 GetPeerName
216:
217: Returns the name of the host that a socket object is connected to.
218:
219: =cut
220:
221: sub GetPeername {
222:
223:
224: my ($connection, $AdrFamily) = @_;
225:
226: my $peer = $connection->peername();
227: my $peerport;
228: my $peerip;
229: if($AdrFamily == AF_INET) {
230: ($peerport, $peerip) = sockaddr_in($peer);
231: my $peername = gethostbyaddr($peerip, $AdrFamily);
232: return $peername;
233: } elsif ($AdrFamily == AF_UNIX) {
234: my $peerfile;
235: ($peerfile) = sockaddr_un($peer);
236: return $peerfile;
237: }
238: }
239: =pod
240:
241: =head2 Debug
242:
243: Invoked to issue a debug message.
244:
245: =cut
246:
247: sub Debug {
248:
249: my ($level, $message) = @_;
250:
251: if ($level <= $DebugLevel) {
252: Log("INFO", "-Debug- $message host = $RemoteHost");
253: }
254: }
255:
256: sub SocketDump {
257:
258: my ($level, $socket) = @_;
259:
260: if($level <= $DebugLevel) {
261: $socket->Dump(-1); # Ensure it will get dumped.
262: }
263: }
264:
265: =pod
266:
267: =head2 ShowStatus
268:
269: Place some text as our pid status.
270: and as what we return in a SIGUSR1
271:
272: =cut
273: sub ShowStatus {
274: my $state = shift;
275: my $now = time;
276: my $local = localtime($now);
277: $Status = $local.": ".$state;
278: $0='lonc: '.$state.' '.$local;
279: }
280:
281: =pod
282:
283: =head 2 SocketTimeout
284:
285: Called when an action on the socket times out. The socket is
286: destroyed and any active transaction is failed.
287:
288:
289: =cut
290: sub SocketTimeout {
291: my $Socket = shift;
292: Log("WARNING", "A socket timeout was detected");
293: Debug(5, " SocketTimeout called: ");
294: $Socket->Dump(0);
295: if(exists($ActiveTransactions{$Socket})) {
296: FailTransaction($ActiveTransactions{$Socket});
297: }
298: KillSocket($Socket); # A transaction timeout also counts as
299: # a connection failure:
300: $ConnectionRetriesLeft--;
301: if($ConnectionRetriesLeft <= 0) {
302: Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
303: $LondConnecting = 0;
304: }
305:
306: }
307: #
308: # This function should be called by the child in all cases where it must
309: # exit. If the child process is running with the DieWhenIdle turned on
310: # it must create a lock file for the AF_UNIX socket in order to prevent
311: # connection requests from lonnet in the time between process exit
312: # and the parent picking up the listen again.
313: # Parameters:
314: # exit_code - Exit status value, however see the next parameter.
315: # message - If this optional parameter is supplied, the exit
316: # is via a die with this message.
317: #
318: sub child_exit {
319: my ($exit_code, $message) = @_;
320:
321: # Regardless of how we exit, we may need to do the lock thing:
322:
323: if($DieWhenIdle) {
324: #
325: # Create a lock file since there will be a time window
326: # between our exit and the parent's picking up the listen
327: # during which no listens will be done on the
328: # lonnet client socket.
329: #
330: my $lock_file = GetLoncSocketPath().".lock";
331: open(LOCK,">$lock_file");
332: print LOCK "Contents not important";
333: close(LOCK);
334:
335: exit(0);
336: }
337: # Now figure out how we exit:
338:
339: if($message) {
340: die $message;
341: } else {
342: exit($exit_code);
343: }
344: }
345: #----------------------------- Timer management ------------------------
346:
347: =pod
348:
349: =head2 Tick
350:
351: Invoked each timer tick.
352:
353: =cut
354:
355:
356: sub Tick {
357: my ($Event) = @_;
358: my $clock_watcher = $Event->w;
359:
360: my $client;
361: UpdateStatus();
362:
363: # Is it time to prune connection count:
364:
365:
366: if($IdleConnections->Count() &&
367: ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
368: $IdleSeconds++;
369: if($IdleSeconds > $IdleTimeout) { # Prune a connection...
370: my $Socket = $IdleConnections->pop();
371: KillSocket($Socket);
372: $IdleSeconds = 0; # Otherwise all connections get trimmed to fast.
373: UpdateStatus();
374: if(($ConnectionCount == 0) && $DieWhenIdle) {
375: &child_exit(0);
376:
377: }
378: }
379: } else {
380: $IdleSeconds = 0; # Reset idle count if not idle.
381: }
382: #
383: # For each inflight transaction, tick down its timeout counter.
384: #
385:
386: foreach my $item (keys %ActiveConnections) {
387: my $State = $ActiveConnections{$item}->data->GetState();
388: if ($State ne 'Idle') {
389: Debug(5,"Ticking Socket $State $item");
390: $ActiveConnections{$item}->data->Tick();
391: }
392: }
393: # Do we have work in the queue, but no connections to service them?
394: # If so, try to make some new connections to get things going again.
395: #
396: # Note this code is dead now...
397: #
398: my $Requests = $WorkQueue->Count();
399: if (($ConnectionCount == 0) && ($Requests > 0) && (!$LondConnecting)) {
400: if ($ConnectionRetriesLeft > 0) {
401: Debug(5,"Work but no connections, Make a new one");
402: my $success;
403: $success = &MakeLondConnection;
404: if($success == 0) { # All connections failed:
405: Debug(5,"Work in queue failed to make any connectiouns\n");
406: EmptyQueue(); # Fail pending transactions with con_lost.
407: CloseAllLondConnections(); # Should all be closed but....
408: }
409: } else {
410: $LondConnecting = 0;
411: ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
412: Debug(5,"Work in queue, but gave up on connections..flushing\n");
413: EmptyQueue(); # Connections can't be established.
414: CloseAllLondConnections(); # Should all already be closed but...
415: }
416:
417: }
418: if ($ConnectionCount == 0) {
419: $KeyMode = "";
420: $clock_watcher->cancel();
421: }
422: }
423:
424: =pod
425:
426: =head2 SetupTimer
427:
428: Sets up a 1 per sec recurring timer event. The event handler is used to:
429:
430: =item
431:
432: Trigger timeouts on communications along active sockets.
433:
434: =item
435:
436: Trigger disconnections of idle sockets.
437:
438: =cut
439:
440: sub SetupTimer {
441: Debug(6, "SetupTimer");
442: Event->timer(interval => 1, cb => \&Tick );
443: }
444:
445: =pod
446:
447: =head2 ServerToIdle
448:
449: This function is called when a connection to the server is
450: ready for more work.
451:
452: If there is work in the Work queue the top element is dequeued
453: and the connection will start to work on it. If the work queue is
454: empty, the connection is pushed on the idle connection stack where
455: it will either get another work unit, or alternatively, if it sits there
456: long enough, it will be shut down and released.
457:
458: =cut
459:
460: sub ServerToIdle {
461: my $Socket = shift; # Get the socket.
462: $KeyMode = $Socket->{AuthenticationMode};
463: delete($ActiveTransactions{$Socket}); # Server has no transaction
464:
465: &Debug(5, "Server to idle");
466:
467: # If there's work to do, start the transaction:
468:
469: my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
470: if ($reqdata ne undef) {
471: Debug(5, "Queue gave request data: ".$reqdata->getRequest());
472: &StartRequest($Socket, $reqdata);
473:
474: } else {
475:
476: # There's no work waiting, so push the server to idle list.
477: &Debug(5, "No new work requests, server connection going idle");
478: $IdleConnections->push($Socket);
479: }
480: }
481:
482: =pod
483:
484: =head2 ClientWritable
485:
486: Event callback for when a client socket is writable.
487:
488: This callback is established when a transaction reponse is
489: avaiable from lond. The response is forwarded to the unix socket
490: as it becomes writable in this sub.
491:
492: Parameters:
493:
494: =item Event
495:
496: The event that has been triggered. Event->w->data is
497: the data and Event->w->fd is the socket to write.
498:
499: =cut
500:
501: sub ClientWritable {
502: my $Event = shift;
503: my $Watcher = $Event->w;
504: my $Data = $Watcher->data;
505: my $Socket = $Watcher->fd;
506:
507: # Try to send the data:
508:
509: &Debug(6, "ClientWritable writing".$Data);
510: &Debug(9, "Socket is: ".$Socket);
511:
512: if($Socket->connected) {
513: my $result = $Socket->send($Data, 0);
514:
515: # $result undefined: the write failed.
516: # otherwise $result is the number of bytes written.
517: # Remove that preceding string from the data.
518: # If the resulting data is empty, destroy the watcher
519: # and set up a read event handler to accept the next
520: # request.
521:
522: &Debug(9,"Send result is ".$result." Defined: ".defined($result));
523: if($result ne undef) {
524: &Debug(9, "send result was defined");
525: if($result == length($Data)) { # Entire string sent.
526: &Debug(9, "ClientWritable data all written");
527: $Watcher->cancel();
528: #
529: # Set up to read next request from socket:
530:
531: my $descr = sprintf("Connection to lonc client %d",
532: $ActiveClients{$Socket});
533: Event->io(cb => \&ClientRequest,
534: poll => 'r',
535: desc => $descr,
536: data => "",
537: fd => $Socket);
538:
539: } else { # Partial string sent.
540: $Watcher->data(substr($Data, $result));
541: if($result == 0) { # client hung up on us!!
542: # Log("INFO", "lonc pipe client hung up on us!");
543: $Watcher->cancel;
544: $Socket->shutdown(2);
545: $Socket->close();
546: }
547: }
548:
549: } else { # Error of some sort...
550:
551: # Some errnos are possible:
552: my $errno = $!;
553: if($errno == POSIX::EWOULDBLOCK ||
554: $errno == POSIX::EAGAIN ||
555: $errno == POSIX::EINTR) {
556: # No action taken?
557: } else { # Unanticipated errno.
558: &Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
559: $Watcher->cancel; # Stop the watcher.
560: $Socket->shutdown(2); # Kill connection
561: $Socket->close(); # Close the socket.
562: }
563:
564: }
565: } else {
566: $Watcher->cancel(); # A delayed request...just cancel.
567: }
568: }
569:
570: =pod
571:
572: =head2 CompleteTransaction
573:
574: Called when the reply data has been received for a lond
575: transaction. The reply data must now be sent to the
576: ultimate client on the other end of the Unix socket. This is
577: done by setting up a writable event for the socket with the
578: data the reply data.
579:
580: Parameters:
581:
582: =item Socket
583:
584: Socket on which the lond transaction occured. This is a
585: LondConnection. The data received is in the TransactionReply member.
586:
587: =item Transaction
588:
589: The transaction that is being completed.
590:
591: =cut
592:
593: sub CompleteTransaction {
594: &Debug(5,"Complete transaction");
595:
596: my ($Socket, $Transaction) = @_;
597:
598: if (!$Transaction->isDeferred()) { # Normal transaction
599: my $data = $Socket->GetReply(); # Data to send.
600: if($LogTransactions) {
601: Log("SUCCESS", "Reply from lond: '$data'");
602: }
603: StartClientReply($Transaction, $data);
604: } else { # Delete deferred transaction file.
605: Log("SUCCESS", "A delayed transaction was completed");
606: LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest());
607: unlink $Transaction->getFile();
608: }
609: }
610:
611: =pod
612:
613: =head1 StartClientReply
614:
615: Initiates a reply to a client where the reply data is a parameter.
616:
617: =head2 parameters:
618:
619: =item Transaction
620:
621: The transaction for which we are responding to the client.
622:
623: =item data
624:
625: The data to send to apached client.
626:
627: =cut
628:
629: sub StartClientReply {
630:
631: my ($Transaction, $data) = @_;
632:
633: my $Client = $Transaction->getClient();
634:
635: &Debug(8," Reply was: ".$data);
636: my $Serial = $ActiveClients{$Client};
637: my $desc = sprintf("Connection to lonc client %d",
638: $Serial);
639: Event->io(fd => $Client,
640: poll => "w",
641: desc => $desc,
642: cb => \&ClientWritable,
643: data => $data);
644: }
645:
646: =pod
647:
648: =head2 FailTransaction
649:
650: Finishes a transaction with failure because the associated lond socket
651: disconnected. There are two possibilities:
652: - The transaction is deferred: in which case we just quietly
653: delete the transaction since there is no client connection.
654: - The transaction is 'live' in which case we initiate the sending
655: of "con_lost" to the client.
656:
657: Deleting the transaction means killing it from the %ActiveTransactions hash.
658:
659: Parameters:
660:
661: =item client
662:
663: The LondTransaction we are failing.
664:
665:
666: =cut
667:
668: sub FailTransaction {
669: my $transaction = shift;
670:
671: # If the socket is dead, that's already logged.
672:
673: if ($ConnectionRetriesLeft > 0) {
674: Log("WARNING", "Failing transaction "
675: .$transaction->getRequest());
676: }
677: Debug(1, "Failing transaction: ".$transaction->getRequest());
678: if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
679: my $client = $transaction->getClient();
680: Debug(1," Replying con_lost to ".$transaction->getRequest());
681: StartClientReply($transaction, "con_lost\n");
682: }
683:
684: }
685:
686: =pod
687: =head1 EmptyQueue
688:
689: Fails all items in the work queue with con_lost.
690: Note that each item in the work queue is a transaction.
691:
692: =cut
693: sub EmptyQueue {
694: $ConnectionRetriesLeft--; # Counts as connection failure too.
695: while($WorkQueue->Count()) {
696: my $request = $WorkQueue->dequeue(); # This is a transaction
697: FailTransaction($request);
698: }
699: }
700:
701: =pod
702:
703: =head2 CloseAllLondConnections
704:
705: Close all connections open on lond prior to exit e.g.
706:
707: =cut
708: sub CloseAllLondConnections {
709: foreach my $Socket (keys %ActiveConnections) {
710: if(exists($ActiveTransactions{$Socket})) {
711: FailTransaction($ActiveTransactions{$Socket});
712: }
713: KillSocket($Socket);
714: }
715: }
716: =cut
717:
718: =pod
719:
720: =head2 KillSocket
721:
722: Destroys a socket. This function can be called either when a socket
723: has died of 'natural' causes or because a socket needs to be pruned due to
724: idleness. If the socket has died naturally, if there are no longer any
725: live connections a new connection is created (in case there are transactions
726: in the queue). If the socket has been pruned, it is never re-created.
727:
728: Parameters:
729:
730: =item Socket
731:
732: The socket to kill off.
733:
734: =item Restart
735:
736: nonzero if we are allowed to create a new connection.
737:
738:
739: =cut
740: sub KillSocket {
741: my $Socket = shift;
742:
743: Log("WARNING", "Shutting down a socket");
744: $Socket->Shutdown();
745:
746: # If the socket came from the active connection set,
747: # delete its transaction... note that FailTransaction should
748: # already have been called!!!
749: # otherwise it came from the idle set.
750: #
751:
752: if(exists($ActiveTransactions{$Socket})) {
753: delete ($ActiveTransactions{$Socket});
754: }
755: if(exists($ActiveConnections{$Socket})) {
756: delete($ActiveConnections{$Socket});
757: $ConnectionCount--;
758: if ($ConnectionCount < 0) { $ConnectionCount = 0; }
759: }
760: # If the connection count has gone to zero and there is work in the
761: # work queue, the work all gets failed with con_lost.
762: #
763: if($ConnectionCount == 0) {
764: EmptyQueue();
765: CloseAllLondConnections; # Should all already be closed but...
766: }
767: }
768:
769: =pod
770:
771: =head2 LondReadable
772:
773: This function is called whenever a lond connection
774: is readable. The action is state dependent:
775:
776: =head3 State=Initialized
777:
778: We''re waiting for the challenge, this is a no-op until the
779: state changes.
780:
781: =head3 State=Challenged
782:
783: The challenge has arrived we need to transition to Writable.
784: The connection must echo the challenge back.
785:
786: =head3 State=ChallengeReplied
787:
788: The challenge has been replied to. The we are receiveing the
789: 'ok' from the partner.
790:
791: =head3 State=ReadingVersionString
792:
793: We have requested the lond version and are reading the
794: version back. Upon completion, we'll store the version away
795: for future use(?).
796:
797: =head3 State=HostSet
798:
799: We have selected the domain name of our peer (multhomed hosts)
800: and are getting the reply (presumably ok) back.
801:
802: =head3 State=RequestingKey
803:
804: The ok has been received and we need to send the request for
805: an encryption key. Transition to writable for that.
806:
807: =head3 State=ReceivingKey
808:
809: The the key has been requested, now we are reading the new key.
810:
811: =head3 State=Idle
812:
813: The encryption key has been negotiated or we have finished
814: reading data from the a transaction. If the callback data has
815: a client as well as the socket iformation, then we are
816: doing a transaction and the data received is relayed to the client
817: before the socket is put on the idle list.
818:
819: =head3 State=SendingRequest
820:
821: I do not think this state can be received here, but if it is,
822: the appropriate thing to do is to transition to writable, and send
823: the request.
824:
825: =head3 State=ReceivingReply
826:
827: We finished sending the request to the server and now transition
828: to readable to receive the reply.
829:
830: The parameter to this function are:
831:
832: The event. Implicit in this is the watcher and its data. The data
833: contains at least the lond connection object and, if a
834: transaction is in progress, the socket attached to the local client.
835:
836: =cut
837:
838: sub LondReadable {
839:
840: my $Event = shift;
841: my $Watcher = $Event->w;
842: my $Socket = $Watcher->data;
843: my $client = undef;
844:
845: &Debug(6,"LondReadable called state = ".$Socket->GetState());
846:
847:
848: my $State = $Socket->GetState(); # All action depends on the state.
849:
850: SocketDump(6, $Socket);
851: my $status = $Socket->Readable();
852:
853: &Debug(2, "Socket->Readable returned: $status");
854:
855: if($status != 0) {
856: # bad return from socket read. Currently this means that
857: # The socket has become disconnected. We fail the transaction.
858:
859: Log("WARNING",
860: "Lond connection lost.");
861: if(exists($ActiveTransactions{$Socket})) {
862: FailTransaction($ActiveTransactions{$Socket});
863: } else {
864: # Socket is connecting and failed... need to mark
865: # no longer connecting.
866:
867: $LondConnecting = 0;
868: }
869: $Watcher->cancel();
870: KillSocket($Socket);
871: $ConnectionRetriesLeft--; # Counts as connection failure
872: return;
873: }
874: SocketDump(6,$Socket);
875:
876: $State = $Socket->GetState(); # Update in case of transition.
877: &Debug(6, "After read, state is ".$State);
878:
879: if($State eq "Initialized") {
880:
881:
882: } elsif ($State eq "ChallengeReceived") {
883: # The challenge must be echoed back; The state machine
884: # in the connection takes care of setting that up. Just
885: # need to transition to writable:
886:
887: $Watcher->cb(\&LondWritable);
888: $Watcher->poll("w");
889:
890: } elsif ($State eq "ChallengeReplied") {
891:
892: } elsif ($State eq "RequestingVersion") {
893: # Need to ask for the version... that is writiability:
894:
895: $Watcher->cb(\&LondWritable);
896: $Watcher->poll("w");
897:
898: } elsif ($State eq "ReadingVersionString") {
899: # Read the rest of the version string...
900: } elsif ($State eq "SetHost") {
901: # Need to request the actual domain get set...
902:
903: $Watcher->cb(\&LondWritable);
904: $Watcher->poll("w");
905: } elsif ($State eq "HostSet") {
906: # Reading the 'ok' from the peer.
907:
908: } elsif ($State eq "RequestingKey") {
909: # The ok was received. Now we need to request the key
910: # That requires us to be writable:
911:
912: $Watcher->cb(\&LondWritable);
913: $Watcher->poll("w");
914:
915: } elsif ($State eq "ReceivingKey") {
916:
917: } elsif ($State eq "Idle") {
918:
919: # This is as good a spot as any to get the peer version
920: # string:
921:
922: if($LondVersion eq "unknown") {
923: $LondVersion = $Socket->PeerVersion();
924: Log("INFO", "Connected to lond version: $LondVersion");
925: }
926: # If necessary, complete a transaction and then go into the
927: # idle queue.
928: # Note that a trasition to idle indicates a live lond
929: # on the other end so reset the connection retries.
930: #
931: $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
932: $Watcher->cancel();
933: if(exists($ActiveTransactions{$Socket})) {
934: Debug(5,"Completing transaction!!");
935: CompleteTransaction($Socket,
936: $ActiveTransactions{$Socket});
937: } else {
938: Log("SUCCESS", "Connection ".$ConnectionCount." to "
939: .$RemoteHost." now ready for action");
940: }
941: ServerToIdle($Socket); # Next work unit or idle.
942:
943: #
944: $LondConnecting = 0; # Best spot I can think of for this.
945: #
946:
947: } elsif ($State eq "SendingRequest") {
948: # We need to be writable for this and probably don't belong
949: # here inthe first place.
950:
951: Deubg(6, "SendingRequest state encountered in readable");
952: $Watcher->poll("w");
953: $Watcher->cb(\&LondWritable);
954:
955: } elsif ($State eq "ReceivingReply") {
956:
957:
958: } else {
959: # Invalid state.
960: Debug(4, "Invalid state in LondReadable");
961: }
962: }
963:
964: =pod
965:
966: =head2 LondWritable
967:
968: This function is called whenever a lond connection
969: becomes writable while there is a writeable monitoring
970: event. The action taken is very state dependent:
971:
972: =head3 State = Connected
973:
974: The connection is in the process of sending the 'init' hailing to the
975: lond on the remote end. The connection object''s Writable member is
976: called. On error, ConnectionError is called to destroy the connection
977: and remove it from the ActiveConnections hash
978:
979: =head3 Initialized
980:
981: 'init' has been sent, writability monitoring is removed and
982: readability monitoring is started with LondReadable as the callback.
983:
984: =head3 ChallengeReceived
985:
986: The connection has received the who are you challenge from the remote
987: system, and is in the process of sending the challenge
988: response. Writable is called.
989:
990: =head3 ChallengeReplied
991:
992: The connection has replied to the initial challenge The we switch to
993: monitoring readability looking for the server to reply with 'ok'.
994:
995: =head3 RequestingKey
996:
997: The connection is in the process of requesting its encryption key.
998: Writable is called.
999:
1000: =head3 ReceivingKey
1001:
1002: The connection has sent the request for a key. Switch to readability
1003: monitoring to accept the key
1004:
1005: =head3 SendingRequest
1006:
1007: The connection is in the process of sending a request to the server.
1008: This request is part of a client transaction. All the states until
1009: now represent the client setup protocol. Writable is called.
1010:
1011: =head3 ReceivingReply
1012:
1013: The connection has sent a request. Now it must receive a reply.
1014: Readability monitoring is requested.
1015:
1016: This function is an event handler and therefore receives as
1017: a parameter the event that has fired. The data for the watcher
1018: of this event is a reference to a list of one or two elements,
1019: depending on state. The first (and possibly only) element is the
1020: socket. The second (present only if a request is in progress)
1021: is the socket on which to return a reply to the caller.
1022:
1023: =cut
1024:
1025: sub LondWritable {
1026: my $Event = shift;
1027: my $Watcher = $Event->w;
1028: my $Socket = $Watcher->data;
1029: my $State = $Socket->GetState();
1030:
1031: Debug(6,"LondWritable State = ".$State."\n");
1032:
1033:
1034: # Figure out what to do depending on the state of the socket:
1035:
1036:
1037:
1038:
1039: SocketDump(6,$Socket);
1040:
1041: # If the socket is writable, we must always write.
1042: # Only by writing will we undergo state transitions.
1043: # Old logic wrote in state specific code below, however
1044: # That forces us at least through another invocation of
1045: # this function after writability is possible again.
1046: # This logic also factors out common code for handling
1047: # write failures... in all cases, write failures
1048: # Kill the socket.
1049: # This logic makes the branches of the >big< if below
1050: # so that the writing states are actually NO-OPs.
1051:
1052: if ($Socket->Writable() != 0) {
1053: # The write resulted in an error.
1054: # We'll treat this as if the socket got disconnected:
1055: Log("WARNING", "Connection to ".$RemoteHost.
1056: " has been disconnected");
1057: if(exists($ActiveTransactions{$Socket})) {
1058: FailTransaction($ActiveTransactions{$Socket});
1059: } else {
1060: # In the process of conneting, so need to turn that off.
1061:
1062: $LondConnecting = 0;
1063: }
1064: $Watcher->cancel();
1065: KillSocket($Socket);
1066: return;
1067: }
1068:
1069:
1070:
1071: if ($State eq "Connected") {
1072:
1073: # "init" is being sent...
1074:
1075: } elsif ($State eq "Initialized") {
1076:
1077: # Now that init was sent, we switch
1078: # to watching for readability:
1079:
1080: $Watcher->cb(\&LondReadable);
1081: $Watcher->poll("r");
1082:
1083: } elsif ($State eq "ChallengeReceived") {
1084: # We received the challenge, now we
1085: # are echoing it back. This is a no-op,
1086: # we're waiting for the state to change
1087:
1088: } elsif ($State eq "ChallengeReplied") {
1089: # The echo was sent back, so we switch
1090: # to watching readability.
1091:
1092: $Watcher->cb(\&LondReadable);
1093: $Watcher->poll("r");
1094: } elsif ($State eq "RequestingVersion") {
1095: # Sending the peer a version request...
1096:
1097: } elsif ($State eq "ReadingVersionString") {
1098: # Transition to read since we have sent the
1099: # version command and now just need to read the
1100: # version string from the peer:
1101:
1102: $Watcher->cb(\&LondReadable);
1103: $Watcher->poll("r");
1104:
1105: } elsif ($State eq "SetHost") {
1106: # Setting the remote domain...
1107:
1108: } elsif ($State eq "HostSet") {
1109: # Back to readable to get the ok.
1110:
1111: $Watcher->cb(\&LondReadable);
1112: $Watcher->poll("r");
1113:
1114:
1115: } elsif ($State eq "RequestingKey") {
1116: # At this time we're requesting the key.
1117: # again, this is essentially a no-op.
1118:
1119: } elsif ($State eq "ReceivingKey") {
1120: # Now we need to wait for the key
1121: # to come back from the peer:
1122:
1123: $Watcher->cb(\&LondReadable);
1124: $Watcher->poll("r");
1125:
1126: } elsif ($State eq "SendingRequest") {
1127:
1128: # At this time we are sending a request to the
1129: # peer... write the next chunk:
1130:
1131:
1132: } elsif ($State eq "ReceivingReply") {
1133: # The send has completed. Wait for the
1134: # data to come in for a reply.
1135: Debug(8,"Writable sent request/receiving reply");
1136: $Watcher->cb(\&LondReadable);
1137: $Watcher->poll("r");
1138:
1139: } else {
1140: # Control only passes here on an error:
1141: # the socket state does not match any
1142: # of the known states... so an error
1143: # must be logged.
1144:
1145: &Debug(4, "Invalid socket state ".$State."\n");
1146: }
1147:
1148: }
1149: =pod
1150:
1151: =cut
1152: sub QueueDelayed {
1153: Debug(3,"QueueDelayed called");
1154:
1155: my $path = "$perlvar{'lonSockDir'}/delayed";
1156:
1157: Debug(4, "Delayed path: ".$path);
1158: opendir(DIRHANDLE, $path);
1159:
1160: my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE;
1161: closedir(DIRHANDLE);
1162: my $dfname;
1163: my $reqfile;
1164: foreach $dfname (sort @alldelayed) {
1165: $reqfile = "$path/$dfname";
1166: Debug(4, "queueing ".$reqfile);
1167: my $Handle = IO::File->new($reqfile);
1168: my $cmd = <$Handle>;
1169: chomp $cmd; # There may or may not be a newline...
1170: $cmd = $cmd."\n"; # now for sure there's exactly one newline.
1171: my $Transaction = LondTransaction->new($cmd);
1172: $Transaction->SetDeferred($reqfile);
1173: QueueTransaction($Transaction);
1174: }
1175:
1176: }
1177:
1178: =pod
1179:
1180: =head2 MakeLondConnection
1181:
1182: Create a new lond connection object, and start it towards its initial
1183: idleness. Once idle, it becomes elligible to receive transactions
1184: from the work queue. If the work queue is not empty when the
1185: connection is completed and becomes idle, it will dequeue an entry and
1186: start off on it.
1187:
1188: =cut
1189:
1190: sub MakeLondConnection {
1191: Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
1192: .GetServerPort());
1193:
1194: my $Connection = LondConnection->new(&GetServerHost(),
1195: &GetServerPort());
1196:
1197: if($Connection eq undef) { # Needs to be more robust later.
1198: Log("CRITICAL","Failed to make a connection with lond.");
1199: $ConnectionRetriesLeft--;
1200: return 0; # Failure.
1201: } else {
1202:
1203: # The connection needs to have writability
1204: # monitored in order to send the init sequence
1205: # that starts the whole authentication/key
1206: # exchange underway.
1207: #
1208: my $Socket = $Connection->GetSocket();
1209: if($Socket eq undef) {
1210: &child_exit(-1, "did not get a socket from the connection");
1211: } else {
1212: &Debug(9,"MakeLondConnection got socket: ".$Socket);
1213: }
1214:
1215: $Connection->SetTimeoutCallback(\&SocketTimeout);
1216:
1217: my $event = Event->io(fd => $Socket,
1218: poll => 'w',
1219: cb => \&LondWritable,
1220: data => $Connection,
1221: desc => 'Connection to lond server');
1222: $ActiveConnections{$Connection} = $event;
1223: if ($ConnectionCount == 0) {
1224: &SetupTimer; # Need to handle timeouts with connections...
1225: }
1226: $ConnectionCount++;
1227: Debug(4, "Connection count = ".$ConnectionCount);
1228: if($ConnectionCount == 1) { # First Connection:
1229: QueueDelayed;
1230: }
1231: Log("SUCESS", "Created connection ".$ConnectionCount
1232: ." to host ".GetServerHost());
1233: $LondConnecting = 1; # Connection in progress.
1234: return 1; # Return success.
1235: }
1236:
1237: }
1238:
1239: =pod
1240:
1241: =head2 StartRequest
1242:
1243: Starts a lond request going on a specified lond connection.
1244: parameters are:
1245:
1246: =item $Lond
1247:
1248: Connection to the lond that will send the transaction and receive the
1249: reply.
1250:
1251: =item $Client
1252:
1253: Connection to the client that is making this request We got the
1254: request from this socket, and when the request has been relayed to
1255: lond and we get a reply back from lond it will get sent to this
1256: socket.
1257:
1258: =item $Request
1259:
1260: The text of the request to send.
1261:
1262: =cut
1263:
1264: sub StartRequest {
1265:
1266: my ($Lond, $Request) = @_;
1267:
1268: Debug(6, "StartRequest: ".$Request->getRequest());
1269:
1270: my $Socket = $Lond->GetSocket();
1271:
1272: $Request->Activate($Lond);
1273: $ActiveTransactions{$Lond} = $Request;
1274:
1275: $Lond->InitiateTransaction($Request->getRequest());
1276: my $event = Event->io(fd => $Socket,
1277: poll => "w",
1278: cb => \&LondWritable,
1279: data => $Lond,
1280: desc => "lond transaction connection");
1281: $ActiveConnections{$Lond} = $event;
1282: Debug(8," Start Request made watcher data with ".$event->data."\n");
1283: }
1284:
1285: =pod
1286:
1287: =head2 QueueTransaction
1288:
1289: If there is an idle lond connection, it is put to work doing this
1290: transaction. Otherwise, the transaction is placed in the work queue.
1291: If placed in the work queue and the maximum number of connections has
1292: not yet been created, a new connection will be started. Our goal is
1293: to eventually have a sufficient number of connections that the work
1294: queue will typically be empty. parameters are:
1295:
1296: =item Socket
1297:
1298: open on the lonc client.
1299:
1300: =item Request
1301:
1302: data to send to the lond.
1303:
1304: =cut
1305:
1306: sub QueueTransaction {
1307:
1308: my $requestData = shift; # This is a LondTransaction.
1309: my $cmd = $requestData->getRequest();
1310:
1311: Debug(6,"QueueTransaction: ".$cmd);
1312:
1313: my $LondSocket = $IdleConnections->pop();
1314: if(!defined $LondSocket) { # Need to queue request.
1315: Debug(5,"Must queue...");
1316: $WorkQueue->enqueue($requestData);
1317: Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
1318: if(($ConnectionCount < $MaxConnectionCount) && (! $LondConnecting)) {
1319:
1320: if($ConnectionRetriesLeft > 0) {
1321: Debug(5,"Starting additional lond connection");
1322: if(&MakeLondConnection() == 0) {
1323: EmptyQueue(); # Fail transactions, can't make connection.
1324: CloseAllLondConnections; # Should all be closed but...
1325: }
1326: } else {
1327: ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1328: $LondConnecting = 0;
1329: EmptyQueue(); # It's worse than that ... he's dead Jim.
1330: CloseAllLondConnections; # Should all be closed but..
1331: }
1332: }
1333: } else { # Can start the request:
1334: Debug(8,"Can start...");
1335: StartRequest($LondSocket, $requestData);
1336: }
1337: }
1338:
1339: #-------------------------- Lonc UNIX socket handling ---------------------
1340:
1341: =pod
1342:
1343: =head2 ClientRequest
1344: Callback that is called when data can be read from the UNIX domain
1345: socket connecting us with an apache server process.
1346:
1347: =cut
1348:
1349: sub ClientRequest {
1350: Debug(6, "ClientRequest");
1351: my $event = shift;
1352: my $watcher = $event->w;
1353: my $socket = $watcher->fd;
1354: my $data = $watcher->data;
1355: my $thisread;
1356:
1357: Debug(9, " Watcher named: ".$watcher->desc);
1358:
1359: my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
1360: Debug(8, "rcv: data length = ".length($thisread)
1361: ." read =".$thisread);
1362: unless (defined $rv && length($thisread)) {
1363: # Likely eof on socket.
1364: Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
1365: close($socket);
1366: $watcher->cancel();
1367: delete($ActiveClients{$socket});
1368: return;
1369: }
1370: Debug(8,"Data: ".$data." this read: ".$thisread);
1371: $data = $data.$thisread; # Append new data.
1372: $watcher->data($data);
1373: if($data =~ /\n$/) { # Request entirely read.
1374: if($data eq "close_connection_exit\n") {
1375: Log("CRITICAL",
1376: "Request Close Connection ... exiting");
1377: CloseAllLondConnections();
1378: exit;
1379: }
1380: Debug(8, "Complete transaction received: ".$data);
1381: if($LogTransactions) {
1382: Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
1383: }
1384: my $Transaction = LondTransaction->new($data);
1385: $Transaction->SetClient($socket);
1386: QueueTransaction($Transaction);
1387: $watcher->cancel(); # Done looking for input data.
1388: }
1389:
1390: }
1391:
1392: #
1393: # Accept a connection request for a client (lonc child) and
1394: # start up an event watcher to keep an eye on input from that
1395: # Event. This can be called both from NewClient and from
1396: # ChildProcess if we are started in DieWhenIdle mode.
1397: # Parameters:
1398: # $socket - The listener socket.
1399: # Returns:
1400: # NONE
1401: # Side Effects:
1402: # An event is made to watch the accepted connection.
1403: # Active clients hash is updated to reflect the new connection.
1404: # The client connection count is incremented.
1405: #
1406: sub accept_client {
1407: my ($socket) = @_;
1408:
1409: Debug(8, "Entering accept for lonc UNIX socket\n");
1410: my $connection = $socket->accept(); # Accept the client connection.
1411: Debug(8,"Connection request accepted from "
1412: .GetPeername($connection, AF_UNIX));
1413:
1414:
1415: my $description = sprintf("Connection to lonc client %d",
1416: $ClientConnection);
1417: Debug(9, "Creating event named: ".$description);
1418: Event->io(cb => \&ClientRequest,
1419: poll => 'r',
1420: desc => $description,
1421: data => "",
1422: fd => $connection);
1423: $ActiveClients{$connection} = $ClientConnection;
1424: $ClientConnection++;
1425: }
1426:
1427: =pod
1428:
1429: =head2 NewClient
1430:
1431: Callback that is called when a connection is received on the unix
1432: socket for a new client of lonc. The callback is parameterized by the
1433: event.. which is a-priori assumed to be an io event, and therefore has
1434: an fd member that is the Listener socket. We Accept the connection
1435: and register a new event on the readability of that socket:
1436:
1437: =cut
1438:
1439: sub NewClient {
1440: Debug(6, "NewClient");
1441: my $event = shift; # Get the event parameters.
1442: my $watcher = $event->w;
1443: my $socket = $watcher->fd; # Get the event' socket.
1444:
1445: &accept_client($socket);
1446: }
1447:
1448: =pod
1449:
1450: =head2 GetLoncSocketPath
1451:
1452: Returns the name of the UNIX socket on which to listen for client
1453: connections.
1454:
1455: =head2 Parameters:
1456:
1457: host (optional) - Name of the host socket to return.. defaults to
1458: the return from GetServerHost().
1459:
1460: =cut
1461:
1462: sub GetLoncSocketPath {
1463:
1464: my $host = GetServerHost(); # Default host.
1465: if (@_) {
1466: ($host) = @_; # Override if supplied.
1467: }
1468: return $UnixSocketDir."/".$host;
1469: }
1470:
1471: =pod
1472:
1473: =head2 GetServerHost
1474:
1475: Returns the host whose lond we talk with.
1476:
1477: =cut
1478:
1479: sub GetServerHost {
1480: return $RemoteHost; # Setup by the fork.
1481: }
1482:
1483: =pod
1484:
1485: =head2 GetServerPort
1486:
1487: Returns the lond port number.
1488:
1489: =cut
1490:
1491: sub GetServerPort {
1492: return $perlvar{londPort};
1493: }
1494:
1495: =pod
1496:
1497: =head2 SetupLoncListener
1498:
1499: Setup a lonc listener event. The event is called when the socket
1500: becomes readable.. that corresponds to the receipt of a new
1501: connection. The event handler established will accept the connection
1502: (creating a communcations channel), that int turn will establish
1503: another event handler to subess requests.
1504:
1505: =head2 Parameters:
1506:
1507: host (optional) Name of the host to set up a unix socket to.
1508:
1509: =cut
1510:
1511: sub SetupLoncListener {
1512:
1513: my $host = GetServerHost(); # Default host.
1514: if (@_) {
1515: ($host) = @_ # Override host with parameter.
1516: }
1517:
1518: my $socket;
1519: my $SocketName = GetLoncSocketPath($host);
1520: unlink($SocketName);
1521: unless ($socket =IO::Socket::UNIX->new(Local => $SocketName,
1522: Listen => 250,
1523: Type => SOCK_STREAM)) {
1524: if($I_am_child) {
1525: &child_exit(-1, "Failed to create a lonc listener socket");
1526: } else {
1527: die "Failed to create a lonc listner socket";
1528: }
1529: }
1530: return $socket;
1531: }
1532:
1533: #
1534: # Toggle transaction logging.
1535: # Implicit inputs:
1536: # LogTransactions
1537: # Implicit Outputs:
1538: # LogTransactions
1539: sub ToggleTransactionLogging {
1540: print STDERR "Toggle transaction logging...\n";
1541: if(!$LogTransactions) {
1542: $LogTransactions = 1;
1543: } else {
1544: $LogTransactions = 0;
1545: }
1546:
1547:
1548: Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
1549: }
1550:
1551: =pod
1552:
1553: =head2 ChildStatus
1554:
1555: Child USR1 signal handler to report the most recent status
1556: into the status file.
1557:
1558: We also use this to reset the retries count in order to allow the
1559: client to retry connections with a previously dead server.
1560: =cut
1561:
1562: sub ChildStatus {
1563: my $event = shift;
1564: my $watcher = $event->w;
1565:
1566: Debug(2, "Reporting child status because : ".$watcher->data);
1567: my $docdir = $perlvar{'lonDocRoot'};
1568: my $fh = IO::File->new(">>$docdir/lon-status/loncstatus.txt");
1569: print $fh $$."\t".$RemoteHost."\t".$Status."\t".
1570: $RecentLogEntry."\n";
1571: #
1572: # Write out information about each of the connections:
1573: #
1574: if ($DebugLevel > 2) {
1575: print $fh "Active connection statuses: \n";
1576: my $i = 1;
1577: print STDERR "================================= Socket Status Dump:\n";
1578: foreach my $item (keys %ActiveConnections) {
1579: my $Socket = $ActiveConnections{$item}->data;
1580: my $state = $Socket->GetState();
1581: print $fh "Connection $i State: $state\n";
1582: print STDERR "---------------------- Connection $i \n";
1583: $Socket->Dump(-1); # Ensure it gets dumped..
1584: $i++;
1585: }
1586: }
1587: $ConnectionRetriesLeft = $ConnectionRetries;
1588: }
1589:
1590: =pod
1591:
1592: =head2 SignalledToDeath
1593:
1594: Called in response to a signal that causes a chid process to die.
1595:
1596: =cut
1597:
1598:
1599: sub SignalledToDeath {
1600: my $event = shift;
1601: my $watcher= $event->w;
1602:
1603: Debug(2,"Signalled to death! via ".$watcher->data);
1604: my ($signal) = $watcher->data;
1605: chomp($signal);
1606: Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost "
1607: ."died through "."\"$signal\"");
1608: LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
1609: ."\"$signal\"");
1610: exit 0;
1611:
1612: }
1613:
1614: =head2 ToggleDebug
1615:
1616: This sub toggles trace debugging on and off.
1617:
1618: =cut
1619:
1620: sub ToggleDebug {
1621: my $Current = $DebugLevel;
1622: $DebugLevel = $NextDebugLevel;
1623: $NextDebugLevel = $Current;
1624:
1625: Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
1626:
1627: }
1628:
1629: =head2 ChildProcess
1630:
1631: This sub implements a child process for a single lonc daemon.
1632: Optional parameter:
1633: $socket - if provided, this is a socket already open for listen
1634: on the client socket. Otherwise, a new listen is set up.
1635:
1636: =cut
1637:
1638: sub ChildProcess {
1639: # If we are in DieWhenIdle mode, we've inherited all the
1640: # events of our parent and those have to be cancelled or else
1641: # all holy bloody chaos will result.. trust me, I already made
1642: # >that< mistake.
1643:
1644: my $host = GetServerHost();
1645: foreach my $listener (keys %parent_dispatchers) {
1646: my $watcher = $parent_dispatchers{$listener};
1647: my $s = $watcher->fd;
1648: if ($listener ne $host) { # Close everyone but me.
1649: Debug(5, "Closing listen socket for $listener");
1650: $s->close();
1651: }
1652: Debug(5, "Killing watcher for $listener");
1653:
1654: $watcher->cancel();
1655: delete($parent_dispatchers{$listener});
1656:
1657: }
1658:
1659: # kill off the parent's signal handlers too!
1660: #
1661:
1662: for my $handler (keys %parent_handlers) {
1663: my $watcher = $parent_handlers{$handler};
1664: $watcher->cancel();
1665: delete($parent_handlers{$handler});
1666: }
1667:
1668: $I_am_child = 1; # Seems like in spite of it all I may still getting
1669: # parent event dispatches.. flag I'm a child.
1670:
1671:
1672: #
1673: # Signals must be handled by the Event framework...
1674: #
1675:
1676: Event->signal(signal => "QUIT",
1677: cb => \&SignalledToDeath,
1678: data => "QUIT");
1679: Event->signal(signal => "HUP",
1680: cb => \&ChildStatus,
1681: data => "HUP");
1682: Event->signal(signal => "USR1",
1683: cb => \&ChildStatus,
1684: data => "USR1");
1685: Event->signal(signal => "USR2",
1686: cb => \&ToggleTransactionLogging);
1687: Event->signal(signal => "INT",
1688: cb => \&ToggleDebug,
1689: data => "INT");
1690:
1691: # Figure out if we got passed a socket or need to open one to listen for
1692: # client requests.
1693:
1694: my ($socket) = @_;
1695: if (!$socket) {
1696:
1697: $socket = SetupLoncListener();
1698: }
1699: # Establish an event to listen for client connection requests.
1700:
1701:
1702: Event->io(cb => \&NewClient,
1703: poll => 'r',
1704: desc => 'Lonc Listener Unix Socket',
1705: fd => $socket);
1706:
1707: $Event::Debuglevel = $DebugLevel;
1708:
1709: Debug(9, "Making initial lond connection for ".$RemoteHost);
1710:
1711: # Setup the initial server connection:
1712:
1713: # &MakeLondConnection(); // let first work request do it.
1714:
1715: # If We are in diwhenidle, need to accept the connection since the
1716: # event may not fire.
1717:
1718: if ($DieWhenIdle) {
1719: &accept_client($socket);
1720: }
1721:
1722: Debug(9,"Entering event loop");
1723: my $ret = Event::loop(); # Start the main event loop.
1724:
1725:
1726: &child_exit (-1,"Main event loop exited!!!");
1727: }
1728:
1729: # Create a new child for host passed in:
1730:
1731: sub CreateChild {
1732: my ($host, $socket) = @_;
1733:
1734: my $sigset = POSIX::SigSet->new(SIGINT);
1735: sigprocmask(SIG_BLOCK, $sigset);
1736: $RemoteHost = $host;
1737: Log("CRITICAL", "Forking server for ".$host);
1738: my $pid = fork;
1739: if($pid) { # Parent
1740: $RemoteHost = "Parent";
1741: $ChildHash{$pid} = $host;
1742: $HostToPid{$host}= $pid;
1743: sigprocmask(SIG_UNBLOCK, $sigset);
1744:
1745: } else { # child.
1746: ShowStatus("Connected to ".$RemoteHost);
1747: $SIG{INT} = 'DEFAULT';
1748: sigprocmask(SIG_UNBLOCK, $sigset);
1749: if(defined $socket) {
1750: &ChildProcess($socket);
1751: } else {
1752: ChildProcess; # Does not return.
1753: }
1754: }
1755: }
1756:
1757: # parent_client_connection:
1758: # Event handler that processes client connections for the parent process.
1759: # This sub is called when the parent is listening on a socket and
1760: # a connection request arrives. We must:
1761: # Start a child process to accept the connection request.
1762: # Kill our listen on the socket.
1763: # Parameter:
1764: # event - The event object that was created to monitor this socket.
1765: # event->w->fd is the socket.
1766: # Returns:
1767: # NONE
1768: #
1769: sub parent_client_connection {
1770: if ($I_am_child) {
1771: # Should not get here, but seem to anyway:
1772: &Debug(5," Child caught parent client connection event!!");
1773: my ($event) = @_;
1774: my $watcher = $event->w;
1775: $watcher->cancel(); # Try to kill it off again!!
1776: } else {
1777: &Debug(9, "parent_client_connection");
1778: my ($event) = @_;
1779: my $watcher = $event->w;
1780: my $socket = $watcher->fd;
1781:
1782: # Lookup the host associated with this socket:
1783:
1784: my $host = $listening_to{$socket};
1785:
1786: # Start the child:
1787:
1788:
1789:
1790: &Debug(9,"Creating child for $host (parent_client_connection)");
1791: &CreateChild($host, $socket);
1792:
1793: # Clean up the listen since now the child takes over until it exits.
1794:
1795: $watcher->cancel(); # Nolonger listening to this event
1796: delete($listening_to{$socket});
1797: delete($parent_dispatchers{$host});
1798: $socket->close();
1799: }
1800: }
1801:
1802: # parent_listen:
1803: # Opens a socket and starts a listen for the parent process on a client UNIX
1804: # domain socket.
1805: #
1806: # This involves:
1807: # Creating a socket for listen.
1808: # Removing any socket lock file
1809: # Adding an event handler for this socket becoming readable
1810: # To the parent's event dispatcher.
1811: # Parameters:
1812: # loncapa_host - LonCAPA cluster name of the host represented by the client
1813: # socket.
1814: # Returns:
1815: # NONE
1816: #
1817: sub parent_listen {
1818: my ($loncapa_host) = @_;
1819: Debug(5, "parent_listen: $loncapa_host");
1820:
1821: my $socket = &SetupLoncListener($loncapa_host);
1822: $listening_to{$socket} = $loncapa_host;
1823: if (!$socket) {
1824: die "Unable to create a listen socket for $loncapa_host";
1825: }
1826:
1827: my $lock_file = &GetLoncSocketPath($loncapa_host).".lock";
1828: unlink($lock_file); # No problem if it doesn't exist yet [startup e.g.]
1829:
1830: my $watcher = Event->io(cb => \&parent_client_connection,
1831: poll => 'r',
1832: desc => "Parent listener unix socket ($loncapa_host)",
1833: fd => $socket);
1834: $parent_dispatchers{$loncapa_host} = $watcher;
1835:
1836: }
1837:
1838:
1839: # listen_on_all_unix_sockets:
1840: # This sub initiates a listen on all unix domain lonc client sockets.
1841: # This will be called in the case where we are trimming idle processes.
1842: # When idle processes are trimmed, loncnew starts up with no children,
1843: # and only spawns off children when a connection request occurs on the
1844: # client unix socket. The spawned child continues to run until it has
1845: # been idle a while at which point it eventually exits and once more
1846: # the parent picks up the listen.
1847: #
1848: # Parameters:
1849: # NONE
1850: # Implicit Inputs:
1851: # The configuration file that has been read in by LondConnection.
1852: # Returns:
1853: # NONE
1854: #
1855: sub listen_on_all_unix_sockets {
1856: Debug(5, "listen_on_all_unix_sockets");
1857: my $host_iterator = &LondConnection::GetHostIterator();
1858: while (!$host_iterator->end()) {
1859: my $host_entry_ref = $host_iterator->get();
1860: my $host_name = $host_entry_ref->[0];
1861: Debug(9, "Listen for $host_name");
1862: &parent_listen($host_name);
1863: $host_iterator->next();
1864: }
1865: }
1866:
1867: # server_died is called whenever a child process exits.
1868: # Since this is dispatched via a signal, we must process all
1869: # dead children until there are no more left. The action
1870: # is to:
1871: # - Remove the child from the bookeeping hashes
1872: # - Re-establish a listen on the unix domain socket associated
1873: # with that host.
1874: # Parameters:
1875: # The event, but we don't actually care about it.
1876: sub server_died {
1877: &Debug(9, "server_died called...");
1878:
1879: while(1) { # Loop until waitpid nowait fails.
1880: my $pid = waitpid(-1, WNOHANG);
1881: if($pid <= 0) {
1882: return; # Nothing left to wait for.
1883: }
1884: # need the host to restart:
1885:
1886: my $host = $ChildHash{$pid};
1887: if($host) { # It's for real...
1888: &Debug(9, "Caught sigchild for $host");
1889: delete($ChildHash{$pid});
1890: delete($HostToPid{$host});
1891: &parent_listen($host);
1892:
1893: } else {
1894: &Debug(5, "Caught sigchild for pid not in hosts hash: $pid");
1895: }
1896: }
1897:
1898: }
1899:
1900: #
1901: # Parent process logic pass 1:
1902: # For each entry in the hosts table, we will
1903: # fork off an instance of ChildProcess to service the transactions
1904: # to that host. Each pid will be entered in a global hash
1905: # with the value of the key, the host.
1906: # The parent will then enter a loop to wait for process exits.
1907: # Each exit gets logged and the child gets restarted.
1908: #
1909:
1910: #
1911: # Fork and start in new session so hang-up isn't going to
1912: # happen without intent.
1913: #
1914:
1915:
1916:
1917:
1918:
1919:
1920: ShowStatus("Forming new session");
1921: my $childpid = fork;
1922: if ($childpid != 0) {
1923: sleep 4; # Give child a chacne to break to
1924: exit 0; # a new sesion.
1925: }
1926: #
1927: # Write my pid into the pid file so I can be located
1928: #
1929:
1930: ShowStatus("Parent writing pid file:");
1931: my $execdir = $perlvar{'lonDaemons'};
1932: open (PIDSAVE, ">$execdir/logs/lonc.pid");
1933: print PIDSAVE "$$\n";
1934: close(PIDSAVE);
1935:
1936:
1937:
1938: if (POSIX::setsid() < 0) {
1939: print "Could not create new session\n";
1940: exit -1;
1941: }
1942:
1943: ShowStatus("Forking node servers");
1944:
1945: Log("CRITICAL", "--------------- Starting children ---------------");
1946:
1947: LondConnection::ReadConfig; # Read standard config files.
1948: my $HostIterator = LondConnection::GetHostIterator;
1949:
1950: if ($DieWhenIdle) {
1951: $RemoteHost = "[parent]";
1952: &listen_on_all_unix_sockets();
1953: } else {
1954:
1955: while (! $HostIterator->end()) {
1956:
1957: my $hostentryref = $HostIterator->get();
1958: CreateChild($hostentryref->[0]);
1959: $HostHash{$hostentryref->[0]} = $hostentryref->[4];
1960: $HostIterator->next();
1961: }
1962: }
1963:
1964: $RemoteHost = "Parent Server";
1965:
1966: # Maintain the population:
1967:
1968: ShowStatus("Parent keeping the flock");
1969:
1970:
1971: if ($DieWhenIdle) {
1972: # We need to setup a SIGChild event to handle the exit (natural or otherwise)
1973: # of the children.
1974:
1975: Event->signal(cb => \&server_died,
1976: desc => "Child exit handler",
1977: signal => "CHLD");
1978:
1979:
1980: # Set up all the other signals we set up. We'll vector them off to the
1981: # same subs as we would for DieWhenIdle false and, if necessary, conditionalize
1982: # the code there.
1983:
1984: $parent_handlers{INT} = Event->signal(cb => \&Terminate,
1985: desc => "Parent INT handler",
1986: signal => "INT");
1987: $parent_handlers{TERM} = Event->signal(cb => \&Terminate,
1988: desc => "Parent TERM handler",
1989: signal => "TERM");
1990: $parent_handlers{HUP} = Event->signal(cb => \&Restart,
1991: desc => "Parent HUP handler.",
1992: signal => "HUP");
1993: $parent_handlers{USR1} = Event->signal(cb => \&CheckKids,
1994: desc => "Parent USR1 handler",
1995: signal => "USR1");
1996: $parent_handlers{USR2} = Event->signal(cb => \&UpdateKids,
1997: desc => "Parent USR2 handler.",
1998: signal => "USR2");
1999:
2000: # Start procdesing events.
2001:
2002: $Event::DebugLevel = $DebugLevel;
2003: Debug(9, "Parent entering event loop");
2004: my $ret = Event::loop();
2005: die "Main Event loop exited: $ret";
2006:
2007:
2008: } else {
2009: #
2010: # Set up parent signals:
2011: #
2012:
2013: $SIG{INT} = \&Terminate;
2014: $SIG{TERM} = \&Terminate;
2015: $SIG{HUP} = \&Restart;
2016: $SIG{USR1} = \&CheckKids;
2017: $SIG{USR2} = \&UpdateKids; # LonManage update request.
2018:
2019: while(1) {
2020: my $deadchild = wait();
2021: if(exists $ChildHash{$deadchild}) { # need to restart.
2022: my $deadhost = $ChildHash{$deadchild};
2023: delete($HostToPid{$deadhost});
2024: delete($ChildHash{$deadchild});
2025: Log("WARNING","Lost child pid= ".$deadchild.
2026: "Connected to host ".$deadhost);
2027: Log("INFO", "Restarting child procesing ".$deadhost);
2028: CreateChild($deadhost);
2029: }
2030: }
2031: }
2032:
2033:
2034: =pod
2035:
2036: =head1 CheckKids
2037:
2038: Since kids do not die as easily in this implementation
2039: as the previous one, there is no need to restart the
2040: dead ones (all dead kids get restarted when they die!!)
2041: The only thing this function does is to pass USR1 to the
2042: kids so that they report their status.
2043:
2044: =cut
2045:
2046: sub CheckKids {
2047: Debug(2, "Checking status of children");
2048: my $docdir = $perlvar{'lonDocRoot'};
2049: my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
2050: my $now=time;
2051: my $local=localtime($now);
2052: print $fh "LONC status $local - parent $$ \n\n";
2053: foreach my $host (keys %parent_dispatchers) {
2054: print $fh "LONC Parent process listening for $host\n";
2055: }
2056: foreach my $pid (keys %ChildHash) {
2057: Debug(2, "Sending USR1 -> $pid");
2058: kill 'USR1' => $pid; # Tell Child to report status.
2059: sleep 1; # Wait so file doesn't intermix.
2060: }
2061:
2062: }
2063:
2064: =pod
2065:
2066: =head1 UpdateKids
2067:
2068: parent's SIGUSR2 handler. This handler:
2069:
2070: =item
2071:
2072: Rereads the hosts file.
2073:
2074: =item
2075:
2076: Kills off (via sigint) children for hosts that have disappeared.
2077:
2078: =item
2079:
2080: QUITs children for hosts that already exist (this just forces a status display
2081: and resets the connection retry count for that host.
2082:
2083: =item
2084:
2085: Starts new children for hosts that have been added to the hosts.tab file since
2086: the start of the master program and maintains them.
2087:
2088: =cut
2089:
2090: sub UpdateKids {
2091:
2092: Log("INFO", "Updating connections via SIGUSR2");
2093:
2094: # I'm not sure what I was thinking in the first implementation.
2095: # someone will have to work hard to convince me the effect is any
2096: # different than Restart, especially now that we don't start up
2097: # per host servers automatically, may as well just restart.
2098: # The down side is transactions that are in flight will get timed out
2099: # (lost unless they are critical).
2100:
2101: &Restart();
2102:
2103: }
2104:
2105:
2106: =pod
2107:
2108: =head1 Restart
2109:
2110: Signal handler for HUP... all children are killed and
2111: we self restart. This is an el-cheapo way to re read
2112: the config file.
2113:
2114: =cut
2115:
2116: sub Restart {
2117: &KillThemAll; # First kill all the children.
2118: Log("CRITICAL", "Restarting");
2119: my $execdir = $perlvar{'lonDaemons'};
2120: unlink("$execdir/logs/lonc.pid");
2121: exec("$executable");
2122: }
2123:
2124: =pod
2125:
2126: =head1 KillThemAll
2127:
2128: Signal handler that kills all children by sending them a
2129: SIGHUP. Responds to sigint and sigterm.
2130:
2131: =cut
2132:
2133: sub KillThemAll {
2134: Debug(2, "Kill them all!!");
2135: local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die.
2136: foreach my $pid (keys %ChildHash) {
2137: my $serving = $ChildHash{$pid};
2138: ShowStatus("Nicely Killing lonc for $serving pid = $pid");
2139: Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
2140: kill 'QUIT' => $pid;
2141: }
2142:
2143:
2144: }
2145:
2146:
2147: #
2148: # Kill all children via KILL. Just in case the
2149: # first shot didn't get them.
2150:
2151: sub really_kill_them_all_dammit
2152: {
2153: Debug(2, "Kill them all Dammit");
2154: local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
2155: foreach my $pid (keys %ChildHash) {
2156: my $serving = $ChildHash{$pid};
2157: &ShowStatus("Nastily killing lonc for $serving pid = $pid");
2158: Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
2159: kill 'KILL' => $pid;
2160: delete($ChildHash{$pid});
2161: my $execdir = $perlvar{'lonDaemons'};
2162: unlink("$execdir/logs/lonc.pid");
2163: }
2164: }
2165: =pod
2166:
2167: =head1 Terminate
2168:
2169: Terminate the system.
2170:
2171: =cut
2172:
2173: sub Terminate {
2174: &Log("CRITICAL", "Asked to kill children.. first be nice...");
2175: &KillThemAll;
2176: #
2177: # By now they really should all be dead.. but just in case
2178: # send them all SIGKILL's after a bit of waiting:
2179:
2180: sleep(4);
2181: &Log("CRITICAL", "Now kill children nasty");
2182: &really_kill_them_all_dammit;
2183: Log("CRITICAL","Master process exiting");
2184: exit 0;
2185:
2186: }
2187: =pod
2188:
2189: =head1 Theory
2190:
2191: The event class is used to build this as a single process with an
2192: event driven model. The following events are handled:
2193:
2194: =item UNIX Socket connection Received
2195:
2196: =item Request data arrives on UNIX data transfer socket.
2197:
2198: =item lond connection becomes writable.
2199:
2200: =item timer fires at 1 second intervals.
2201:
2202: All sockets are run in non-blocking mode. Timeouts managed by the timer
2203: handler prevents hung connections.
2204:
2205: Key data structures:
2206:
2207: =item RequestQueue
2208:
2209: A queue of requests received from UNIX sockets that are
2210: waiting for a chance to be forwarded on a lond connection socket.
2211:
2212: =item ActiveConnections
2213:
2214: A hash of lond connections that have transactions in process that are
2215: available to be timed out.
2216:
2217: =item ActiveTransactions
2218:
2219: A hash indexed by lond connections that contain the client reply
2220: socket for each connection that has an active transaction on it.
2221:
2222: =item IdleConnections
2223:
2224: A hash of lond connections that have no work to do. These connections
2225: can be closed if they are idle for a long enough time.
2226:
2227: =cut
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>