1: #!/usr/bin/perl
2: # The LearningOnline Network with CAPA
3: # lonc maintains the connections to remote computers
4: #
5: # $Id: loncnew,v 1.67 2005/01/17 22:18:44 albertel 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: use Fcntl qw(:flock);
65:
66:
67: # Read the httpd configuration file to get perl variables
68: # normally set in apache modules:
69:
70: my $perlvarref = LONCAPA::Configuration::read_conf('loncapa.conf');
71: my %perlvar = %{$perlvarref};
72:
73: #
74: # parent and shared variables.
75:
76: my %ChildHash; # by pid -> host.
77: my %HostToPid; # By host -> pid.
78: my %HostHash; # by loncapaname -> IP.
79: my %listening_to; # Socket->host table for who the parent
80: # is listening to.
81: my %parent_dispatchers; # host-> listener watcher events.
82:
83: my %parent_handlers; # Parent signal handlers...
84:
85: my $MaxConnectionCount = 10; # Will get from config later.
86: my $ClientConnection = 0; # Uniquifier for client events.
87:
88: my $DebugLevel = 0;
89: my $NextDebugLevel= 2; # So Sigint can toggle this.
90: my $IdleTimeout= 600; # Wait 10 minutes before pruning connections.
91:
92: my $LogTransactions = 0; # When True, all transactions/replies get logged.
93: my $executable = $0; # Get the full path to me.
94:
95: #
96: # The variables below are only used by the child processes.
97: #
98: my $RemoteHost; # Name of host child is talking to.
99: my $UnixSocketDir= $perlvar{'lonSockDir'};
100: my $IdleConnections = Stack->new(); # Set of idle connections
101: my %ActiveConnections; # Connections to the remote lond.
102: my %ActiveTransactions; # LondTransactions in flight.
103: my %ActiveClients; # Serial numbers of active clients by socket.
104: my $WorkQueue = Queue->new(); # Queue of pending transactions.
105: my $ConnectionCount = 0;
106: my $IdleSeconds = 0; # Number of seconds idle.
107: my $Status = ""; # Current status string.
108: my $RecentLogEntry = "";
109: my $ConnectionRetries=2; # Number of connection retries allowed.
110: my $ConnectionRetriesLeft=2; # Number of connection retries remaining.
111: my $LondVersion = "unknown"; # Version of lond we talk with.
112: my $KeyMode = ""; # e.g. ssl, local, insecure from last connect.
113: my $LondConnecting = 0; # True when a connection is being built.
114:
115:
116:
117: my $DieWhenIdle = 1; # When true children die when trimmed -> 0.
118: my $I_am_child = 0; # True if this is the child process.
119:
120: #
121: # The hash below gives the HTML format for log messages
122: # given a severity.
123: #
124: my %LogFormats;
125:
126: $LogFormats{"CRITICAL"} = "<font color='red'>CRITICAL: %s</font>";
127: $LogFormats{"SUCCESS"} = "<font color='green'>SUCCESS: %s</font>";
128: $LogFormats{"INFO"} = "<font color='yellow'>INFO: %s</font>";
129: $LogFormats{"WARNING"} = "<font color='blue'>WARNING: %s</font>";
130: $LogFormats{"DEFAULT"} = " %s ";
131:
132:
133: # UpdateStatus;
134: # Update the idle status display to show how many connections
135: # are left, retries and other stuff.
136: #
137: sub UpdateStatus {
138: if ($ConnectionRetriesLeft > 0) {
139: ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount
140: ." Retries remaining: ".$ConnectionRetriesLeft
141: ." ($KeyMode)");
142: } else {
143: ShowStatus(GetServerHost()." >> DEAD <<");
144: }
145: }
146:
147:
148: =pod
149:
150: =head2 LogPerm
151:
152: Makes an entry into the permanent log file.
153:
154: =cut
155: sub LogPerm {
156: my $message=shift;
157: my $execdir=$perlvar{'lonDaemons'};
158: my $now=time;
159: my $local=localtime($now);
160: my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
161: print $fh "$now:$message:$local\n";
162: }
163:
164: =pod
165:
166: =head2 Log
167:
168: Logs a message to the log file.
169: Parameters:
170:
171: =item severity
172:
173: One of CRITICAL, WARNING, INFO, SUCCESS used to select the
174: format string used to format the message. if the severity is
175: not a defined severity the Default format string is used.
176:
177: =item message
178:
179: The base message. In addtion to the format string, the message
180: will be appended to a string containing the name of our remote
181: host and the time will be formatted into the message.
182:
183: =cut
184:
185: sub Log {
186:
187: my ($severity, $message) = @_;
188:
189: if(!$LogFormats{$severity}) {
190: $severity = "DEFAULT";
191: }
192:
193: my $format = $LogFormats{$severity};
194:
195: # Put the window dressing in in front of the message format:
196:
197: my $now = time;
198: my $local = localtime($now);
199: my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
200: my $finalformat = $finalformat.$format."\n";
201:
202: # open the file and put the result.
203:
204: my $execdir = $perlvar{'lonDaemons'};
205: my $fh = IO::File->new(">>$execdir/logs/lonc.log");
206: my $msg = sprintf($finalformat, $message);
207: $RecentLogEntry = $msg;
208: print $fh $msg;
209:
210:
211: }
212:
213:
214: =pod
215:
216: =head2 GetPeerName
217:
218: Returns the name of the host that a socket object is connected to.
219:
220: =cut
221:
222: sub GetPeername {
223:
224:
225: my ($connection, $AdrFamily) = @_;
226:
227: my $peer = $connection->peername();
228: my $peerport;
229: my $peerip;
230: if($AdrFamily == AF_INET) {
231: ($peerport, $peerip) = sockaddr_in($peer);
232: my $peername = gethostbyaddr($peerip, $AdrFamily);
233: return $peername;
234: } elsif ($AdrFamily == AF_UNIX) {
235: my $peerfile;
236: ($peerfile) = sockaddr_un($peer);
237: return $peerfile;
238: }
239: }
240: =pod
241:
242: =head2 Debug
243:
244: Invoked to issue a debug message.
245:
246: =cut
247:
248: sub Debug {
249:
250: my ($level, $message) = @_;
251:
252: if ($level <= $DebugLevel) {
253: Log("INFO", "-Debug- $message host = $RemoteHost");
254: }
255: }
256:
257: sub SocketDump {
258:
259: my ($level, $socket) = @_;
260:
261: if($level <= $DebugLevel) {
262: $socket->Dump(-1); # Ensure it will get dumped.
263: }
264: }
265:
266: =pod
267:
268: =head2 ShowStatus
269:
270: Place some text as our pid status.
271: and as what we return in a SIGUSR1
272:
273: =cut
274: sub ShowStatus {
275: my $state = shift;
276: my $now = time;
277: my $local = localtime($now);
278: $Status = $local.": ".$state;
279: $0='lonc: '.$state.' '.$local;
280: }
281:
282: =pod
283:
284: =head 2 SocketTimeout
285:
286: Called when an action on the socket times out. The socket is
287: destroyed and any active transaction is failed.
288:
289:
290: =cut
291: sub SocketTimeout {
292: my $Socket = shift;
293: Log("WARNING", "A socket timeout was detected");
294: Debug(5, " SocketTimeout called: ");
295: $Socket->Dump(0);
296: if(exists($ActiveTransactions{$Socket})) {
297: FailTransaction($ActiveTransactions{$Socket});
298: }
299: KillSocket($Socket); # A transaction timeout also counts as
300: # a connection failure:
301: $ConnectionRetriesLeft--;
302: if($ConnectionRetriesLeft <= 0) {
303: Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
304: $LondConnecting = 0;
305: }
306:
307: }
308: #
309: # This function should be called by the child in all cases where it must
310: # exit. If the child process is running with the DieWhenIdle turned on
311: # it must create a lock file for the AF_UNIX socket in order to prevent
312: # connection requests from lonnet in the time between process exit
313: # and the parent picking up the listen again.
314: # Parameters:
315: # exit_code - Exit status value, however see the next parameter.
316: # message - If this optional parameter is supplied, the exit
317: # is via a die with this message.
318: #
319: sub child_exit {
320: my ($exit_code, $message) = @_;
321:
322: # Regardless of how we exit, we may need to do the lock thing:
323:
324: if($DieWhenIdle) {
325: #
326: # Create a lock file since there will be a time window
327: # between our exit and the parent's picking up the listen
328: # during which no listens will be done on the
329: # lonnet client socket.
330: #
331: my $lock_file = GetLoncSocketPath().".lock";
332: open(LOCK,">$lock_file");
333: print LOCK "Contents not important";
334: close(LOCK);
335:
336: exit(0);
337: }
338: # Now figure out how we exit:
339:
340: if($message) {
341: die $message;
342: } else {
343: exit($exit_code);
344: }
345: }
346: #----------------------------- Timer management ------------------------
347:
348: =pod
349:
350: =head2 Tick
351:
352: Invoked each timer tick.
353:
354: =cut
355:
356:
357: sub Tick {
358: my ($Event) = @_;
359: my $clock_watcher = $Event->w;
360:
361: my $client;
362: UpdateStatus();
363:
364: # Is it time to prune connection count:
365:
366:
367: if($IdleConnections->Count() &&
368: ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
369: $IdleSeconds++;
370: if($IdleSeconds > $IdleTimeout) { # Prune a connection...
371: my $Socket = $IdleConnections->pop();
372: KillSocket($Socket);
373: $IdleSeconds = 0; # Otherwise all connections get trimmed to fast.
374: UpdateStatus();
375: if(($ConnectionCount == 0) && $DieWhenIdle) {
376: &child_exit(0);
377:
378: }
379: }
380: } else {
381: $IdleSeconds = 0; # Reset idle count if not idle.
382: }
383: #
384: # For each inflight transaction, tick down its timeout counter.
385: #
386:
387: foreach my $item (keys %ActiveConnections) {
388: my $State = $ActiveConnections{$item}->data->GetState();
389: if ($State ne 'Idle') {
390: Debug(5,"Ticking Socket $State $item");
391: $ActiveConnections{$item}->data->Tick();
392: }
393: }
394: # Do we have work in the queue, but no connections to service them?
395: # If so, try to make some new connections to get things going again.
396: #
397: # Note this code is dead now...
398: #
399: my $Requests = $WorkQueue->Count();
400: if (($ConnectionCount == 0) && ($Requests > 0) && (!$LondConnecting)) {
401: if ($ConnectionRetriesLeft > 0) {
402: Debug(5,"Work but no connections, Make a new one");
403: my $success;
404: $success = &MakeLondConnection;
405: if($success == 0) { # All connections failed:
406: Debug(5,"Work in queue failed to make any connectiouns\n");
407: EmptyQueue(); # Fail pending transactions with con_lost.
408: CloseAllLondConnections(); # Should all be closed but....
409: }
410: } else {
411: $LondConnecting = 0;
412: ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
413: Debug(5,"Work in queue, but gave up on connections..flushing\n");
414: EmptyQueue(); # Connections can't be established.
415: CloseAllLondConnections(); # Should all already be closed but...
416: }
417:
418: }
419: if ($ConnectionCount == 0) {
420: $KeyMode = "";
421: $clock_watcher->cancel();
422: }
423: &UpdateStatus();
424: }
425:
426: =pod
427:
428: =head2 SetupTimer
429:
430: Sets up a 1 per sec recurring timer event. The event handler is used to:
431:
432: =item
433:
434: Trigger timeouts on communications along active sockets.
435:
436: =item
437:
438: Trigger disconnections of idle sockets.
439:
440: =cut
441:
442: sub SetupTimer {
443: Debug(6, "SetupTimer");
444: Event->timer(interval => 1, cb => \&Tick );
445: }
446:
447: =pod
448:
449: =head2 ServerToIdle
450:
451: This function is called when a connection to the server is
452: ready for more work.
453:
454: If there is work in the Work queue the top element is dequeued
455: and the connection will start to work on it. If the work queue is
456: empty, the connection is pushed on the idle connection stack where
457: it will either get another work unit, or alternatively, if it sits there
458: long enough, it will be shut down and released.
459:
460: =cut
461:
462: sub ServerToIdle {
463: my $Socket = shift; # Get the socket.
464: $KeyMode = $Socket->{AuthenticationMode};
465: delete($ActiveTransactions{$Socket}); # Server has no transaction
466:
467: &Debug(5, "Server to idle");
468:
469: # If there's work to do, start the transaction:
470:
471: my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
472: if ($reqdata ne undef) {
473: Debug(5, "Queue gave request data: ".$reqdata->getRequest());
474: &StartRequest($Socket, $reqdata);
475:
476: } else {
477:
478: # There's no work waiting, so push the server to idle list.
479: &Debug(5, "No new work requests, server connection going idle");
480: $IdleConnections->push($Socket);
481: }
482: }
483:
484: =pod
485:
486: =head2 ClientWritable
487:
488: Event callback for when a client socket is writable.
489:
490: This callback is established when a transaction reponse is
491: avaiable from lond. The response is forwarded to the unix socket
492: as it becomes writable in this sub.
493:
494: Parameters:
495:
496: =item Event
497:
498: The event that has been triggered. Event->w->data is
499: the data and Event->w->fd is the socket to write.
500:
501: =cut
502:
503: sub ClientWritable {
504: my $Event = shift;
505: my $Watcher = $Event->w;
506: my $Data = $Watcher->data;
507: my $Socket = $Watcher->fd;
508:
509: # Try to send the data:
510:
511: &Debug(6, "ClientWritable writing".$Data);
512: &Debug(9, "Socket is: ".$Socket);
513:
514: if($Socket->connected) {
515: my $result = $Socket->send($Data, 0);
516:
517: # $result undefined: the write failed.
518: # otherwise $result is the number of bytes written.
519: # Remove that preceding string from the data.
520: # If the resulting data is empty, destroy the watcher
521: # and set up a read event handler to accept the next
522: # request.
523:
524: &Debug(9,"Send result is ".$result." Defined: ".defined($result));
525: if($result ne undef) {
526: &Debug(9, "send result was defined");
527: if($result == length($Data)) { # Entire string sent.
528: &Debug(9, "ClientWritable data all written");
529: $Watcher->cancel();
530: #
531: # Set up to read next request from socket:
532:
533: my $descr = sprintf("Connection to lonc client %d",
534: $ActiveClients{$Socket});
535: Event->io(cb => \&ClientRequest,
536: poll => 'r',
537: desc => $descr,
538: data => "",
539: fd => $Socket);
540:
541: } else { # Partial string sent.
542: $Watcher->data(substr($Data, $result));
543: if($result == 0) { # client hung up on us!!
544: # Log("INFO", "lonc pipe client hung up on us!");
545: $Watcher->cancel;
546: $Socket->shutdown(2);
547: $Socket->close();
548: }
549: }
550:
551: } else { # Error of some sort...
552:
553: # Some errnos are possible:
554: my $errno = $!;
555: if($errno == POSIX::EWOULDBLOCK ||
556: $errno == POSIX::EAGAIN ||
557: $errno == POSIX::EINTR) {
558: # No action taken?
559: } else { # Unanticipated errno.
560: &Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
561: $Watcher->cancel; # Stop the watcher.
562: $Socket->shutdown(2); # Kill connection
563: $Socket->close(); # Close the socket.
564: }
565:
566: }
567: } else {
568: $Watcher->cancel(); # A delayed request...just cancel.
569: }
570: }
571:
572: =pod
573:
574: =head2 CompleteTransaction
575:
576: Called when the reply data has been received for a lond
577: transaction. The reply data must now be sent to the
578: ultimate client on the other end of the Unix socket. This is
579: done by setting up a writable event for the socket with the
580: data the reply data.
581:
582: Parameters:
583:
584: =item Socket
585:
586: Socket on which the lond transaction occured. This is a
587: LondConnection. The data received is in the TransactionReply member.
588:
589: =item Transaction
590:
591: The transaction that is being completed.
592:
593: =cut
594:
595: sub CompleteTransaction {
596: &Debug(5,"Complete transaction");
597:
598: my ($Socket, $Transaction) = @_;
599:
600: if (!$Transaction->isDeferred()) { # Normal transaction
601: my $data = $Socket->GetReply(); # Data to send.
602: if($LogTransactions) {
603: Log("SUCCESS", "Reply from lond: '$data'");
604: }
605: StartClientReply($Transaction, $data);
606: } else { # Delete deferred transaction file.
607: Log("SUCCESS", "A delayed transaction was completed");
608: LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest());
609: unlink $Transaction->getFile();
610: }
611: }
612:
613: =pod
614:
615: =head1 StartClientReply
616:
617: Initiates a reply to a client where the reply data is a parameter.
618:
619: =head2 parameters:
620:
621: =item Transaction
622:
623: The transaction for which we are responding to the client.
624:
625: =item data
626:
627: The data to send to apached client.
628:
629: =cut
630:
631: sub StartClientReply {
632:
633: my ($Transaction, $data) = @_;
634:
635: my $Client = $Transaction->getClient();
636:
637: &Debug(8," Reply was: ".$data);
638: my $Serial = $ActiveClients{$Client};
639: my $desc = sprintf("Connection to lonc client %d",
640: $Serial);
641: Event->io(fd => $Client,
642: poll => "w",
643: desc => $desc,
644: cb => \&ClientWritable,
645: data => $data);
646: }
647:
648: =pod
649:
650: =head2 FailTransaction
651:
652: Finishes a transaction with failure because the associated lond socket
653: disconnected. There are two possibilities:
654: - The transaction is deferred: in which case we just quietly
655: delete the transaction since there is no client connection.
656: - The transaction is 'live' in which case we initiate the sending
657: of "con_lost" to the client.
658:
659: Deleting the transaction means killing it from the %ActiveTransactions hash.
660:
661: Parameters:
662:
663: =item client
664:
665: The LondTransaction we are failing.
666:
667:
668: =cut
669:
670: sub FailTransaction {
671: my $transaction = shift;
672:
673: # If the socket is dead, that's already logged.
674:
675: if ($ConnectionRetriesLeft > 0) {
676: Log("WARNING", "Failing transaction "
677: .$transaction->getRequest());
678: }
679: Debug(1, "Failing transaction: ".$transaction->getRequest());
680: if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
681: my $client = $transaction->getClient();
682: Debug(1," Replying con_lost to ".$transaction->getRequest());
683: StartClientReply($transaction, "con_lost\n");
684: }
685:
686: }
687:
688: =pod
689: =head1 EmptyQueue
690:
691: Fails all items in the work queue with con_lost.
692: Note that each item in the work queue is a transaction.
693:
694: =cut
695: sub EmptyQueue {
696: $ConnectionRetriesLeft--; # Counts as connection failure too.
697: while($WorkQueue->Count()) {
698: my $request = $WorkQueue->dequeue(); # This is a transaction
699: FailTransaction($request);
700: }
701: }
702:
703: =pod
704:
705: =head2 CloseAllLondConnections
706:
707: Close all connections open on lond prior to exit e.g.
708:
709: =cut
710: sub CloseAllLondConnections {
711: foreach my $Socket (keys %ActiveConnections) {
712: if(exists($ActiveTransactions{$Socket})) {
713: FailTransaction($ActiveTransactions{$Socket});
714: }
715: KillSocket($Socket);
716: }
717: }
718: =cut
719:
720: =pod
721:
722: =head2 KillSocket
723:
724: Destroys a socket. This function can be called either when a socket
725: has died of 'natural' causes or because a socket needs to be pruned due to
726: idleness. If the socket has died naturally, if there are no longer any
727: live connections a new connection is created (in case there are transactions
728: in the queue). If the socket has been pruned, it is never re-created.
729:
730: Parameters:
731:
732: =item Socket
733:
734: The socket to kill off.
735:
736: =item Restart
737:
738: nonzero if we are allowed to create a new connection.
739:
740:
741: =cut
742: sub KillSocket {
743: my $Socket = shift;
744:
745: Log("WARNING", "Shutting down a socket");
746: $Socket->Shutdown();
747:
748: # If the socket came from the active connection set,
749: # delete its transaction... note that FailTransaction should
750: # already have been called!!!
751: # otherwise it came from the idle set.
752: #
753:
754: if(exists($ActiveTransactions{$Socket})) {
755: delete ($ActiveTransactions{$Socket});
756: }
757: if(exists($ActiveConnections{$Socket})) {
758: delete($ActiveConnections{$Socket});
759: $ConnectionCount--;
760: if ($ConnectionCount < 0) { $ConnectionCount = 0; }
761: }
762: # If the connection count has gone to zero and there is work in the
763: # work queue, the work all gets failed with con_lost.
764: #
765: if($ConnectionCount == 0) {
766: EmptyQueue();
767: CloseAllLondConnections; # Should all already be closed but...
768: }
769: }
770:
771: =pod
772:
773: =head2 LondReadable
774:
775: This function is called whenever a lond connection
776: is readable. The action is state dependent:
777:
778: =head3 State=Initialized
779:
780: We''re waiting for the challenge, this is a no-op until the
781: state changes.
782:
783: =head3 State=Challenged
784:
785: The challenge has arrived we need to transition to Writable.
786: The connection must echo the challenge back.
787:
788: =head3 State=ChallengeReplied
789:
790: The challenge has been replied to. The we are receiveing the
791: 'ok' from the partner.
792:
793: =head3 State=ReadingVersionString
794:
795: We have requested the lond version and are reading the
796: version back. Upon completion, we'll store the version away
797: for future use(?).
798:
799: =head3 State=HostSet
800:
801: We have selected the domain name of our peer (multhomed hosts)
802: and are getting the reply (presumably ok) back.
803:
804: =head3 State=RequestingKey
805:
806: The ok has been received and we need to send the request for
807: an encryption key. Transition to writable for that.
808:
809: =head3 State=ReceivingKey
810:
811: The the key has been requested, now we are reading the new key.
812:
813: =head3 State=Idle
814:
815: The encryption key has been negotiated or we have finished
816: reading data from the a transaction. If the callback data has
817: a client as well as the socket iformation, then we are
818: doing a transaction and the data received is relayed to the client
819: before the socket is put on the idle list.
820:
821: =head3 State=SendingRequest
822:
823: I do not think this state can be received here, but if it is,
824: the appropriate thing to do is to transition to writable, and send
825: the request.
826:
827: =head3 State=ReceivingReply
828:
829: We finished sending the request to the server and now transition
830: to readable to receive the reply.
831:
832: The parameter to this function are:
833:
834: The event. Implicit in this is the watcher and its data. The data
835: contains at least the lond connection object and, if a
836: transaction is in progress, the socket attached to the local client.
837:
838: =cut
839:
840: sub LondReadable {
841:
842: my $Event = shift;
843: my $Watcher = $Event->w;
844: my $Socket = $Watcher->data;
845: my $client = undef;
846:
847: &Debug(6,"LondReadable called state = ".$Socket->GetState());
848:
849:
850: my $State = $Socket->GetState(); # All action depends on the state.
851:
852: SocketDump(6, $Socket);
853: my $status = $Socket->Readable();
854:
855: &Debug(2, "Socket->Readable returned: $status");
856:
857: if($status != 0) {
858: # bad return from socket read. Currently this means that
859: # The socket has become disconnected. We fail the transaction.
860:
861: Log("WARNING",
862: "Lond connection lost.");
863: if(exists($ActiveTransactions{$Socket})) {
864: FailTransaction($ActiveTransactions{$Socket});
865: } else {
866: # Socket is connecting and failed... need to mark
867: # no longer connecting.
868:
869: $LondConnecting = 0;
870: }
871: $Watcher->cancel();
872: KillSocket($Socket);
873: $ConnectionRetriesLeft--; # Counts as connection failure
874: return;
875: }
876: SocketDump(6,$Socket);
877:
878: $State = $Socket->GetState(); # Update in case of transition.
879: &Debug(6, "After read, state is ".$State);
880:
881: if($State eq "Initialized") {
882:
883:
884: } elsif ($State eq "ChallengeReceived") {
885: # The challenge must be echoed back; The state machine
886: # in the connection takes care of setting that up. Just
887: # need to transition to writable:
888:
889: $Watcher->cb(\&LondWritable);
890: $Watcher->poll("w");
891:
892: } elsif ($State eq "ChallengeReplied") {
893:
894: } elsif ($State eq "RequestingVersion") {
895: # Need to ask for the version... that is writiability:
896:
897: $Watcher->cb(\&LondWritable);
898: $Watcher->poll("w");
899:
900: } elsif ($State eq "ReadingVersionString") {
901: # Read the rest of the version string...
902: } elsif ($State eq "SetHost") {
903: # Need to request the actual domain get set...
904:
905: $Watcher->cb(\&LondWritable);
906: $Watcher->poll("w");
907: } elsif ($State eq "HostSet") {
908: # Reading the 'ok' from the peer.
909:
910: } elsif ($State eq "RequestingKey") {
911: # The ok was received. Now we need to request the key
912: # That requires us to be writable:
913:
914: $Watcher->cb(\&LondWritable);
915: $Watcher->poll("w");
916:
917: } elsif ($State eq "ReceivingKey") {
918:
919: } elsif ($State eq "Idle") {
920:
921: # This is as good a spot as any to get the peer version
922: # string:
923:
924: if($LondVersion eq "unknown") {
925: $LondVersion = $Socket->PeerVersion();
926: Log("INFO", "Connected to lond version: $LondVersion");
927: }
928: # If necessary, complete a transaction and then go into the
929: # idle queue.
930: # Note that a trasition to idle indicates a live lond
931: # on the other end so reset the connection retries.
932: #
933: $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
934: $Watcher->cancel();
935: if(exists($ActiveTransactions{$Socket})) {
936: Debug(5,"Completing transaction!!");
937: CompleteTransaction($Socket,
938: $ActiveTransactions{$Socket});
939: } else {
940: Log("SUCCESS", "Connection ".$ConnectionCount." to "
941: .$RemoteHost." now ready for action");
942: }
943: ServerToIdle($Socket); # Next work unit or idle.
944:
945: #
946: $LondConnecting = 0; # Best spot I can think of for this.
947: #
948:
949: } elsif ($State eq "SendingRequest") {
950: # We need to be writable for this and probably don't belong
951: # here inthe first place.
952:
953: Deubg(6, "SendingRequest state encountered in readable");
954: $Watcher->poll("w");
955: $Watcher->cb(\&LondWritable);
956:
957: } elsif ($State eq "ReceivingReply") {
958:
959:
960: } else {
961: # Invalid state.
962: Debug(4, "Invalid state in LondReadable");
963: }
964: }
965:
966: =pod
967:
968: =head2 LondWritable
969:
970: This function is called whenever a lond connection
971: becomes writable while there is a writeable monitoring
972: event. The action taken is very state dependent:
973:
974: =head3 State = Connected
975:
976: The connection is in the process of sending the 'init' hailing to the
977: lond on the remote end. The connection object''s Writable member is
978: called. On error, ConnectionError is called to destroy the connection
979: and remove it from the ActiveConnections hash
980:
981: =head3 Initialized
982:
983: 'init' has been sent, writability monitoring is removed and
984: readability monitoring is started with LondReadable as the callback.
985:
986: =head3 ChallengeReceived
987:
988: The connection has received the who are you challenge from the remote
989: system, and is in the process of sending the challenge
990: response. Writable is called.
991:
992: =head3 ChallengeReplied
993:
994: The connection has replied to the initial challenge The we switch to
995: monitoring readability looking for the server to reply with 'ok'.
996:
997: =head3 RequestingKey
998:
999: The connection is in the process of requesting its encryption key.
1000: Writable is called.
1001:
1002: =head3 ReceivingKey
1003:
1004: The connection has sent the request for a key. Switch to readability
1005: monitoring to accept the key
1006:
1007: =head3 SendingRequest
1008:
1009: The connection is in the process of sending a request to the server.
1010: This request is part of a client transaction. All the states until
1011: now represent the client setup protocol. Writable is called.
1012:
1013: =head3 ReceivingReply
1014:
1015: The connection has sent a request. Now it must receive a reply.
1016: Readability monitoring is requested.
1017:
1018: This function is an event handler and therefore receives as
1019: a parameter the event that has fired. The data for the watcher
1020: of this event is a reference to a list of one or two elements,
1021: depending on state. The first (and possibly only) element is the
1022: socket. The second (present only if a request is in progress)
1023: is the socket on which to return a reply to the caller.
1024:
1025: =cut
1026:
1027: sub LondWritable {
1028: my $Event = shift;
1029: my $Watcher = $Event->w;
1030: my $Socket = $Watcher->data;
1031: my $State = $Socket->GetState();
1032:
1033: Debug(6,"LondWritable State = ".$State."\n");
1034:
1035:
1036: # Figure out what to do depending on the state of the socket:
1037:
1038:
1039:
1040:
1041: SocketDump(6,$Socket);
1042:
1043: # If the socket is writable, we must always write.
1044: # Only by writing will we undergo state transitions.
1045: # Old logic wrote in state specific code below, however
1046: # That forces us at least through another invocation of
1047: # this function after writability is possible again.
1048: # This logic also factors out common code for handling
1049: # write failures... in all cases, write failures
1050: # Kill the socket.
1051: # This logic makes the branches of the >big< if below
1052: # so that the writing states are actually NO-OPs.
1053:
1054: if ($Socket->Writable() != 0) {
1055: # The write resulted in an error.
1056: # We'll treat this as if the socket got disconnected:
1057: Log("WARNING", "Connection to ".$RemoteHost.
1058: " has been disconnected");
1059: if(exists($ActiveTransactions{$Socket})) {
1060: FailTransaction($ActiveTransactions{$Socket});
1061: } else {
1062: # In the process of conneting, so need to turn that off.
1063:
1064: $LondConnecting = 0;
1065: }
1066: $Watcher->cancel();
1067: KillSocket($Socket);
1068: return;
1069: }
1070:
1071:
1072:
1073: if ($State eq "Connected") {
1074:
1075: # "init" is being sent...
1076:
1077: } elsif ($State eq "Initialized") {
1078:
1079: # Now that init was sent, we switch
1080: # to watching for readability:
1081:
1082: $Watcher->cb(\&LondReadable);
1083: $Watcher->poll("r");
1084:
1085: } elsif ($State eq "ChallengeReceived") {
1086: # We received the challenge, now we
1087: # are echoing it back. This is a no-op,
1088: # we're waiting for the state to change
1089:
1090: } elsif ($State eq "ChallengeReplied") {
1091: # The echo was sent back, so we switch
1092: # to watching readability.
1093:
1094: $Watcher->cb(\&LondReadable);
1095: $Watcher->poll("r");
1096: } elsif ($State eq "RequestingVersion") {
1097: # Sending the peer a version request...
1098:
1099: } elsif ($State eq "ReadingVersionString") {
1100: # Transition to read since we have sent the
1101: # version command and now just need to read the
1102: # version string from the peer:
1103:
1104: $Watcher->cb(\&LondReadable);
1105: $Watcher->poll("r");
1106:
1107: } elsif ($State eq "SetHost") {
1108: # Setting the remote domain...
1109:
1110: } elsif ($State eq "HostSet") {
1111: # Back to readable to get the ok.
1112:
1113: $Watcher->cb(\&LondReadable);
1114: $Watcher->poll("r");
1115:
1116:
1117: } elsif ($State eq "RequestingKey") {
1118: # At this time we're requesting the key.
1119: # again, this is essentially a no-op.
1120:
1121: } elsif ($State eq "ReceivingKey") {
1122: # Now we need to wait for the key
1123: # to come back from the peer:
1124:
1125: $Watcher->cb(\&LondReadable);
1126: $Watcher->poll("r");
1127:
1128: } elsif ($State eq "SendingRequest") {
1129:
1130: # At this time we are sending a request to the
1131: # peer... write the next chunk:
1132:
1133:
1134: } elsif ($State eq "ReceivingReply") {
1135: # The send has completed. Wait for the
1136: # data to come in for a reply.
1137: Debug(8,"Writable sent request/receiving reply");
1138: $Watcher->cb(\&LondReadable);
1139: $Watcher->poll("r");
1140:
1141: } else {
1142: # Control only passes here on an error:
1143: # the socket state does not match any
1144: # of the known states... so an error
1145: # must be logged.
1146:
1147: &Debug(4, "Invalid socket state ".$State."\n");
1148: }
1149:
1150: }
1151: =pod
1152:
1153: =cut
1154: sub QueueDelayed {
1155: Debug(3,"QueueDelayed called");
1156:
1157: my $path = "$perlvar{'lonSockDir'}/delayed";
1158:
1159: Debug(4, "Delayed path: ".$path);
1160: opendir(DIRHANDLE, $path);
1161:
1162: my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE;
1163: closedir(DIRHANDLE);
1164: my $dfname;
1165: my $reqfile;
1166: foreach $dfname (sort @alldelayed) {
1167: $reqfile = "$path/$dfname";
1168: Debug(4, "queueing ".$reqfile);
1169: my $Handle = IO::File->new($reqfile);
1170: my $cmd = <$Handle>;
1171: chomp $cmd; # There may or may not be a newline...
1172: $cmd = $cmd."\n"; # now for sure there's exactly one newline.
1173: my $Transaction = LondTransaction->new($cmd);
1174: $Transaction->SetDeferred($reqfile);
1175: QueueTransaction($Transaction);
1176: }
1177:
1178: }
1179:
1180: =pod
1181:
1182: =head2 MakeLondConnection
1183:
1184: Create a new lond connection object, and start it towards its initial
1185: idleness. Once idle, it becomes elligible to receive transactions
1186: from the work queue. If the work queue is not empty when the
1187: connection is completed and becomes idle, it will dequeue an entry and
1188: start off on it.
1189:
1190: =cut
1191:
1192: sub MakeLondConnection {
1193: Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
1194: .GetServerPort());
1195:
1196: my $Connection = LondConnection->new(&GetServerHost(),
1197: &GetServerPort());
1198:
1199: if($Connection eq undef) { # Needs to be more robust later.
1200: Log("CRITICAL","Failed to make a connection with lond.");
1201: $ConnectionRetriesLeft--;
1202: return 0; # Failure.
1203: } else {
1204:
1205: # The connection needs to have writability
1206: # monitored in order to send the init sequence
1207: # that starts the whole authentication/key
1208: # exchange underway.
1209: #
1210: my $Socket = $Connection->GetSocket();
1211: if($Socket eq undef) {
1212: &child_exit(-1, "did not get a socket from the connection");
1213: } else {
1214: &Debug(9,"MakeLondConnection got socket: ".$Socket);
1215: }
1216:
1217: $Connection->SetTimeoutCallback(\&SocketTimeout);
1218:
1219: my $event = Event->io(fd => $Socket,
1220: poll => 'w',
1221: cb => \&LondWritable,
1222: data => $Connection,
1223: desc => 'Connection to lond server');
1224: $ActiveConnections{$Connection} = $event;
1225: if ($ConnectionCount == 0) {
1226: &SetupTimer; # Need to handle timeouts with connections...
1227: }
1228: $ConnectionCount++;
1229: Debug(4, "Connection count = ".$ConnectionCount);
1230: if($ConnectionCount == 1) { # First Connection:
1231: QueueDelayed;
1232: }
1233: Log("SUCESS", "Created connection ".$ConnectionCount
1234: ." to host ".GetServerHost());
1235: $LondConnecting = 1; # Connection in progress.
1236: return 1; # Return success.
1237: }
1238:
1239: }
1240:
1241: =pod
1242:
1243: =head2 StartRequest
1244:
1245: Starts a lond request going on a specified lond connection.
1246: parameters are:
1247:
1248: =item $Lond
1249:
1250: Connection to the lond that will send the transaction and receive the
1251: reply.
1252:
1253: =item $Client
1254:
1255: Connection to the client that is making this request We got the
1256: request from this socket, and when the request has been relayed to
1257: lond and we get a reply back from lond it will get sent to this
1258: socket.
1259:
1260: =item $Request
1261:
1262: The text of the request to send.
1263:
1264: =cut
1265:
1266: sub StartRequest {
1267:
1268: my ($Lond, $Request) = @_;
1269:
1270: Debug(6, "StartRequest: ".$Request->getRequest());
1271:
1272: my $Socket = $Lond->GetSocket();
1273:
1274: $Request->Activate($Lond);
1275: $ActiveTransactions{$Lond} = $Request;
1276:
1277: $Lond->InitiateTransaction($Request->getRequest());
1278: my $event = Event->io(fd => $Socket,
1279: poll => "w",
1280: cb => \&LondWritable,
1281: data => $Lond,
1282: desc => "lond transaction connection");
1283: $ActiveConnections{$Lond} = $event;
1284: Debug(8," Start Request made watcher data with ".$event->data."\n");
1285: }
1286:
1287: =pod
1288:
1289: =head2 QueueTransaction
1290:
1291: If there is an idle lond connection, it is put to work doing this
1292: transaction. Otherwise, the transaction is placed in the work queue.
1293: If placed in the work queue and the maximum number of connections has
1294: not yet been created, a new connection will be started. Our goal is
1295: to eventually have a sufficient number of connections that the work
1296: queue will typically be empty. parameters are:
1297:
1298: =item Socket
1299:
1300: open on the lonc client.
1301:
1302: =item Request
1303:
1304: data to send to the lond.
1305:
1306: =cut
1307:
1308: sub QueueTransaction {
1309:
1310: my $requestData = shift; # This is a LondTransaction.
1311: my $cmd = $requestData->getRequest();
1312:
1313: Debug(6,"QueueTransaction: ".$cmd);
1314:
1315: my $LondSocket = $IdleConnections->pop();
1316: if(!defined $LondSocket) { # Need to queue request.
1317: Debug(5,"Must queue...");
1318: $WorkQueue->enqueue($requestData);
1319: Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
1320: if(($ConnectionCount < $MaxConnectionCount) && (! $LondConnecting)) {
1321:
1322: if($ConnectionRetriesLeft > 0) {
1323: Debug(5,"Starting additional lond connection");
1324: if(&MakeLondConnection() == 0) {
1325: EmptyQueue(); # Fail transactions, can't make connection.
1326: CloseAllLondConnections; # Should all be closed but...
1327: }
1328: } else {
1329: ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1330: $LondConnecting = 0;
1331: EmptyQueue(); # It's worse than that ... he's dead Jim.
1332: CloseAllLondConnections; # Should all be closed but..
1333: }
1334: }
1335: } else { # Can start the request:
1336: Debug(8,"Can start...");
1337: StartRequest($LondSocket, $requestData);
1338: }
1339: }
1340:
1341: #-------------------------- Lonc UNIX socket handling ---------------------
1342:
1343: =pod
1344:
1345: =head2 ClientRequest
1346: Callback that is called when data can be read from the UNIX domain
1347: socket connecting us with an apache server process.
1348:
1349: =cut
1350:
1351: sub ClientRequest {
1352: Debug(6, "ClientRequest");
1353: my $event = shift;
1354: my $watcher = $event->w;
1355: my $socket = $watcher->fd;
1356: my $data = $watcher->data;
1357: my $thisread;
1358:
1359: Debug(9, " Watcher named: ".$watcher->desc);
1360:
1361: my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
1362: Debug(8, "rcv: data length = ".length($thisread)
1363: ." read =".$thisread);
1364: unless (defined $rv && length($thisread)) {
1365: # Likely eof on socket.
1366: Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
1367: close($socket);
1368: $watcher->cancel();
1369: delete($ActiveClients{$socket});
1370: return;
1371: }
1372: Debug(8,"Data: ".$data." this read: ".$thisread);
1373: $data = $data.$thisread; # Append new data.
1374: $watcher->data($data);
1375: if($data =~ /\n$/) { # Request entirely read.
1376: if($data eq "close_connection_exit\n") {
1377: Log("CRITICAL",
1378: "Request Close Connection ... exiting");
1379: CloseAllLondConnections();
1380: exit;
1381: }
1382: Debug(8, "Complete transaction received: ".$data);
1383: if($LogTransactions) {
1384: Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
1385: }
1386: my $Transaction = LondTransaction->new($data);
1387: $Transaction->SetClient($socket);
1388: QueueTransaction($Transaction);
1389: $watcher->cancel(); # Done looking for input data.
1390: }
1391:
1392: }
1393:
1394: #
1395: # Accept a connection request for a client (lonc child) and
1396: # start up an event watcher to keep an eye on input from that
1397: # Event. This can be called both from NewClient and from
1398: # ChildProcess if we are started in DieWhenIdle mode.
1399: # Parameters:
1400: # $socket - The listener socket.
1401: # Returns:
1402: # NONE
1403: # Side Effects:
1404: # An event is made to watch the accepted connection.
1405: # Active clients hash is updated to reflect the new connection.
1406: # The client connection count is incremented.
1407: #
1408: sub accept_client {
1409: my ($socket) = @_;
1410:
1411: Debug(8, "Entering accept for lonc UNIX socket\n");
1412: my $connection = $socket->accept(); # Accept the client connection.
1413: Debug(8,"Connection request accepted from "
1414: .GetPeername($connection, AF_UNIX));
1415:
1416:
1417: my $description = sprintf("Connection to lonc client %d",
1418: $ClientConnection);
1419: Debug(9, "Creating event named: ".$description);
1420: Event->io(cb => \&ClientRequest,
1421: poll => 'r',
1422: desc => $description,
1423: data => "",
1424: fd => $connection);
1425: $ActiveClients{$connection} = $ClientConnection;
1426: $ClientConnection++;
1427: }
1428:
1429: =pod
1430:
1431: =head2 NewClient
1432:
1433: Callback that is called when a connection is received on the unix
1434: socket for a new client of lonc. The callback is parameterized by the
1435: event.. which is a-priori assumed to be an io event, and therefore has
1436: an fd member that is the Listener socket. We Accept the connection
1437: and register a new event on the readability of that socket:
1438:
1439: =cut
1440:
1441: sub NewClient {
1442: Debug(6, "NewClient");
1443: my $event = shift; # Get the event parameters.
1444: my $watcher = $event->w;
1445: my $socket = $watcher->fd; # Get the event' socket.
1446:
1447: &accept_client($socket);
1448: }
1449:
1450: =pod
1451:
1452: =head2 GetLoncSocketPath
1453:
1454: Returns the name of the UNIX socket on which to listen for client
1455: connections.
1456:
1457: =head2 Parameters:
1458:
1459: host (optional) - Name of the host socket to return.. defaults to
1460: the return from GetServerHost().
1461:
1462: =cut
1463:
1464: sub GetLoncSocketPath {
1465:
1466: my $host = GetServerHost(); # Default host.
1467: if (@_) {
1468: ($host) = @_; # Override if supplied.
1469: }
1470: return $UnixSocketDir."/".$host;
1471: }
1472:
1473: =pod
1474:
1475: =head2 GetServerHost
1476:
1477: Returns the host whose lond we talk with.
1478:
1479: =cut
1480:
1481: sub GetServerHost {
1482: return $RemoteHost; # Setup by the fork.
1483: }
1484:
1485: =pod
1486:
1487: =head2 GetServerPort
1488:
1489: Returns the lond port number.
1490:
1491: =cut
1492:
1493: sub GetServerPort {
1494: return $perlvar{londPort};
1495: }
1496:
1497: =pod
1498:
1499: =head2 SetupLoncListener
1500:
1501: Setup a lonc listener event. The event is called when the socket
1502: becomes readable.. that corresponds to the receipt of a new
1503: connection. The event handler established will accept the connection
1504: (creating a communcations channel), that int turn will establish
1505: another event handler to subess requests.
1506:
1507: =head2 Parameters:
1508:
1509: host (optional) Name of the host to set up a unix socket to.
1510:
1511: =cut
1512:
1513: sub SetupLoncListener {
1514:
1515: my $host = GetServerHost(); # Default host.
1516: if (@_) {
1517: ($host) = @_ # Override host with parameter.
1518: }
1519:
1520: my $socket;
1521: my $SocketName = GetLoncSocketPath($host);
1522: unlink($SocketName);
1523: unless ($socket =IO::Socket::UNIX->new(Local => $SocketName,
1524: Listen => 250,
1525: Type => SOCK_STREAM)) {
1526: if($I_am_child) {
1527: &child_exit(-1, "Failed to create a lonc listener socket");
1528: } else {
1529: die "Failed to create a lonc listner socket";
1530: }
1531: }
1532: return $socket;
1533: }
1534:
1535: #
1536: # Toggle transaction logging.
1537: # Implicit inputs:
1538: # LogTransactions
1539: # Implicit Outputs:
1540: # LogTransactions
1541: sub ToggleTransactionLogging {
1542: print STDERR "Toggle transaction logging...\n";
1543: if(!$LogTransactions) {
1544: $LogTransactions = 1;
1545: } else {
1546: $LogTransactions = 0;
1547: }
1548:
1549:
1550: Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
1551: }
1552:
1553: =pod
1554:
1555: =head2 ChildStatus
1556:
1557: Child USR1 signal handler to report the most recent status
1558: into the status file.
1559:
1560: We also use this to reset the retries count in order to allow the
1561: client to retry connections with a previously dead server.
1562: =cut
1563:
1564: sub ChildStatus {
1565: my $event = shift;
1566: my $watcher = $event->w;
1567:
1568: Debug(2, "Reporting child status because : ".$watcher->data);
1569: my $docdir = $perlvar{'lonDocRoot'};
1570:
1571: open(LOG,">>$docdir/lon-status/loncstatus.txt");
1572: flock(LOG,LOCK_EX);
1573: print LOG $$."\t".$RemoteHost."\t".$Status."\t".
1574: $RecentLogEntry."\n";
1575: #
1576: # Write out information about each of the connections:
1577: #
1578: if ($DebugLevel > 2) {
1579: print LOG "Active connection statuses: \n";
1580: my $i = 1;
1581: print STDERR "================================= Socket Status Dump:\n";
1582: foreach my $item (keys %ActiveConnections) {
1583: my $Socket = $ActiveConnections{$item}->data;
1584: my $state = $Socket->GetState();
1585: print LOG "Connection $i State: $state\n";
1586: print STDERR "---------------------- Connection $i \n";
1587: $Socket->Dump(-1); # Ensure it gets dumped..
1588: $i++;
1589: }
1590: }
1591: flock(LOG,LOCK_UN);
1592: close(LOG);
1593: $ConnectionRetriesLeft = $ConnectionRetries;
1594: }
1595:
1596: =pod
1597:
1598: =head2 SignalledToDeath
1599:
1600: Called in response to a signal that causes a chid process to die.
1601:
1602: =cut
1603:
1604:
1605: sub SignalledToDeath {
1606: my $event = shift;
1607: my $watcher= $event->w;
1608:
1609: Debug(2,"Signalled to death! via ".$watcher->data);
1610: my ($signal) = $watcher->data;
1611: chomp($signal);
1612: Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost "
1613: ."died through "."\"$signal\"");
1614: LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
1615: ."\"$signal\"");
1616: exit 0;
1617:
1618: }
1619:
1620: =head2 ToggleDebug
1621:
1622: This sub toggles trace debugging on and off.
1623:
1624: =cut
1625:
1626: sub ToggleDebug {
1627: my $Current = $DebugLevel;
1628: $DebugLevel = $NextDebugLevel;
1629: $NextDebugLevel = $Current;
1630:
1631: Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
1632:
1633: }
1634:
1635: =head2 ChildProcess
1636:
1637: This sub implements a child process for a single lonc daemon.
1638: Optional parameter:
1639: $socket - if provided, this is a socket already open for listen
1640: on the client socket. Otherwise, a new listen is set up.
1641:
1642: =cut
1643:
1644: sub ChildProcess {
1645: # If we are in DieWhenIdle mode, we've inherited all the
1646: # events of our parent and those have to be cancelled or else
1647: # all holy bloody chaos will result.. trust me, I already made
1648: # >that< mistake.
1649:
1650: my $host = GetServerHost();
1651: foreach my $listener (keys %parent_dispatchers) {
1652: my $watcher = $parent_dispatchers{$listener};
1653: my $s = $watcher->fd;
1654: if ($listener ne $host) { # Close everyone but me.
1655: Debug(5, "Closing listen socket for $listener");
1656: $s->close();
1657: }
1658: Debug(5, "Killing watcher for $listener");
1659:
1660: $watcher->cancel();
1661: delete($parent_dispatchers{$listener});
1662:
1663: }
1664:
1665: # kill off the parent's signal handlers too!
1666: #
1667:
1668: for my $handler (keys %parent_handlers) {
1669: my $watcher = $parent_handlers{$handler};
1670: $watcher->cancel();
1671: delete($parent_handlers{$handler});
1672: }
1673:
1674: $I_am_child = 1; # Seems like in spite of it all I may still getting
1675: # parent event dispatches.. flag I'm a child.
1676:
1677:
1678: #
1679: # Signals must be handled by the Event framework...
1680: #
1681:
1682: Event->signal(signal => "QUIT",
1683: cb => \&SignalledToDeath,
1684: data => "QUIT");
1685: Event->signal(signal => "HUP",
1686: cb => \&ChildStatus,
1687: data => "HUP");
1688: Event->signal(signal => "USR1",
1689: cb => \&ChildStatus,
1690: data => "USR1");
1691: Event->signal(signal => "USR2",
1692: cb => \&ToggleTransactionLogging);
1693: Event->signal(signal => "INT",
1694: cb => \&ToggleDebug,
1695: data => "INT");
1696:
1697: # Figure out if we got passed a socket or need to open one to listen for
1698: # client requests.
1699:
1700: my ($socket) = @_;
1701: if (!$socket) {
1702:
1703: $socket = SetupLoncListener();
1704: }
1705: # Establish an event to listen for client connection requests.
1706:
1707:
1708: Event->io(cb => \&NewClient,
1709: poll => 'r',
1710: desc => 'Lonc Listener Unix Socket',
1711: fd => $socket);
1712:
1713: $Event::Debuglevel = $DebugLevel;
1714:
1715: Debug(9, "Making initial lond connection for ".$RemoteHost);
1716:
1717: # Setup the initial server connection:
1718:
1719: # &MakeLondConnection(); // let first work request do it.
1720:
1721: # If We are in diwhenidle, need to accept the connection since the
1722: # event may not fire.
1723:
1724: if ($DieWhenIdle) {
1725: &accept_client($socket);
1726: }
1727:
1728: Debug(9,"Entering event loop");
1729: my $ret = Event::loop(); # Start the main event loop.
1730:
1731:
1732: &child_exit (-1,"Main event loop exited!!!");
1733: }
1734:
1735: # Create a new child for host passed in:
1736:
1737: sub CreateChild {
1738: my ($host, $socket) = @_;
1739:
1740: my $sigset = POSIX::SigSet->new(SIGINT);
1741: sigprocmask(SIG_BLOCK, $sigset);
1742: $RemoteHost = $host;
1743: Log("CRITICAL", "Forking server for ".$host);
1744: my $pid = fork;
1745: if($pid) { # Parent
1746: $RemoteHost = "Parent";
1747: $ChildHash{$pid} = $host;
1748: $HostToPid{$host}= $pid;
1749: sigprocmask(SIG_UNBLOCK, $sigset);
1750:
1751: } else { # child.
1752: ShowStatus("Connected to ".$RemoteHost);
1753: $SIG{INT} = 'DEFAULT';
1754: sigprocmask(SIG_UNBLOCK, $sigset);
1755: if(defined $socket) {
1756: &ChildProcess($socket);
1757: } else {
1758: ChildProcess; # Does not return.
1759: }
1760: }
1761: }
1762:
1763: # parent_client_connection:
1764: # Event handler that processes client connections for the parent process.
1765: # This sub is called when the parent is listening on a socket and
1766: # a connection request arrives. We must:
1767: # Start a child process to accept the connection request.
1768: # Kill our listen on the socket.
1769: # Parameter:
1770: # event - The event object that was created to monitor this socket.
1771: # event->w->fd is the socket.
1772: # Returns:
1773: # NONE
1774: #
1775: sub parent_client_connection {
1776: if ($I_am_child) {
1777: # Should not get here, but seem to anyway:
1778: &Debug(5," Child caught parent client connection event!!");
1779: my ($event) = @_;
1780: my $watcher = $event->w;
1781: $watcher->cancel(); # Try to kill it off again!!
1782: } else {
1783: &Debug(9, "parent_client_connection");
1784: my ($event) = @_;
1785: my $watcher = $event->w;
1786: my $socket = $watcher->fd;
1787:
1788: # Lookup the host associated with this socket:
1789:
1790: my $host = $listening_to{$socket};
1791:
1792: # Start the child:
1793:
1794:
1795:
1796: &Debug(9,"Creating child for $host (parent_client_connection)");
1797: &CreateChild($host, $socket);
1798:
1799: # Clean up the listen since now the child takes over until it exits.
1800:
1801: $watcher->cancel(); # Nolonger listening to this event
1802: delete($listening_to{$socket});
1803: delete($parent_dispatchers{$host});
1804: $socket->close();
1805: }
1806: }
1807:
1808: # parent_listen:
1809: # Opens a socket and starts a listen for the parent process on a client UNIX
1810: # domain socket.
1811: #
1812: # This involves:
1813: # Creating a socket for listen.
1814: # Removing any socket lock file
1815: # Adding an event handler for this socket becoming readable
1816: # To the parent's event dispatcher.
1817: # Parameters:
1818: # loncapa_host - LonCAPA cluster name of the host represented by the client
1819: # socket.
1820: # Returns:
1821: # NONE
1822: #
1823: sub parent_listen {
1824: my ($loncapa_host) = @_;
1825: Debug(5, "parent_listen: $loncapa_host");
1826:
1827: my $socket = &SetupLoncListener($loncapa_host);
1828: $listening_to{$socket} = $loncapa_host;
1829: if (!$socket) {
1830: die "Unable to create a listen socket for $loncapa_host";
1831: }
1832:
1833: my $lock_file = &GetLoncSocketPath($loncapa_host).".lock";
1834: unlink($lock_file); # No problem if it doesn't exist yet [startup e.g.]
1835:
1836: my $watcher = Event->io(cb => \&parent_client_connection,
1837: poll => 'r',
1838: desc => "Parent listener unix socket ($loncapa_host)",
1839: fd => $socket);
1840: $parent_dispatchers{$loncapa_host} = $watcher;
1841:
1842: }
1843:
1844:
1845: # listen_on_all_unix_sockets:
1846: # This sub initiates a listen on all unix domain lonc client sockets.
1847: # This will be called in the case where we are trimming idle processes.
1848: # When idle processes are trimmed, loncnew starts up with no children,
1849: # and only spawns off children when a connection request occurs on the
1850: # client unix socket. The spawned child continues to run until it has
1851: # been idle a while at which point it eventually exits and once more
1852: # the parent picks up the listen.
1853: #
1854: # Parameters:
1855: # NONE
1856: # Implicit Inputs:
1857: # The configuration file that has been read in by LondConnection.
1858: # Returns:
1859: # NONE
1860: #
1861: sub listen_on_all_unix_sockets {
1862: Debug(5, "listen_on_all_unix_sockets");
1863: my $host_iterator = &LondConnection::GetHostIterator();
1864: while (!$host_iterator->end()) {
1865: my $host_entry_ref = $host_iterator->get();
1866: my $host_name = $host_entry_ref->[0];
1867: Debug(9, "Listen for $host_name");
1868: &parent_listen($host_name);
1869: $host_iterator->next();
1870: }
1871: }
1872:
1873: # server_died is called whenever a child process exits.
1874: # Since this is dispatched via a signal, we must process all
1875: # dead children until there are no more left. The action
1876: # is to:
1877: # - Remove the child from the bookeeping hashes
1878: # - Re-establish a listen on the unix domain socket associated
1879: # with that host.
1880: # Parameters:
1881: # The event, but we don't actually care about it.
1882: sub server_died {
1883: &Debug(9, "server_died called...");
1884:
1885: while(1) { # Loop until waitpid nowait fails.
1886: my $pid = waitpid(-1, WNOHANG);
1887: if($pid <= 0) {
1888: return; # Nothing left to wait for.
1889: }
1890: # need the host to restart:
1891:
1892: my $host = $ChildHash{$pid};
1893: if($host) { # It's for real...
1894: &Debug(9, "Caught sigchild for $host");
1895: delete($ChildHash{$pid});
1896: delete($HostToPid{$host});
1897: &parent_listen($host);
1898:
1899: } else {
1900: &Debug(5, "Caught sigchild for pid not in hosts hash: $pid");
1901: }
1902: }
1903:
1904: }
1905:
1906: #
1907: # Parent process logic pass 1:
1908: # For each entry in the hosts table, we will
1909: # fork off an instance of ChildProcess to service the transactions
1910: # to that host. Each pid will be entered in a global hash
1911: # with the value of the key, the host.
1912: # The parent will then enter a loop to wait for process exits.
1913: # Each exit gets logged and the child gets restarted.
1914: #
1915:
1916: #
1917: # Fork and start in new session so hang-up isn't going to
1918: # happen without intent.
1919: #
1920:
1921:
1922:
1923:
1924:
1925:
1926: ShowStatus("Forming new session");
1927: my $childpid = fork;
1928: if ($childpid != 0) {
1929: sleep 4; # Give child a chacne to break to
1930: exit 0; # a new sesion.
1931: }
1932: #
1933: # Write my pid into the pid file so I can be located
1934: #
1935:
1936: ShowStatus("Parent writing pid file:");
1937: my $execdir = $perlvar{'lonDaemons'};
1938: open (PIDSAVE, ">$execdir/logs/lonc.pid");
1939: print PIDSAVE "$$\n";
1940: close(PIDSAVE);
1941:
1942:
1943:
1944: if (POSIX::setsid() < 0) {
1945: print "Could not create new session\n";
1946: exit -1;
1947: }
1948:
1949: ShowStatus("Forking node servers");
1950:
1951: Log("CRITICAL", "--------------- Starting children ---------------");
1952:
1953: LondConnection::ReadConfig; # Read standard config files.
1954: my $HostIterator = LondConnection::GetHostIterator;
1955:
1956: if ($DieWhenIdle) {
1957: $RemoteHost = "[parent]";
1958: &listen_on_all_unix_sockets();
1959: } else {
1960:
1961: while (! $HostIterator->end()) {
1962:
1963: my $hostentryref = $HostIterator->get();
1964: CreateChild($hostentryref->[0]);
1965: $HostHash{$hostentryref->[0]} = $hostentryref->[4];
1966: $HostIterator->next();
1967: }
1968: }
1969:
1970: $RemoteHost = "Parent Server";
1971:
1972: # Maintain the population:
1973:
1974: ShowStatus("Parent keeping the flock");
1975:
1976:
1977: if ($DieWhenIdle) {
1978: # We need to setup a SIGChild event to handle the exit (natural or otherwise)
1979: # of the children.
1980:
1981: Event->signal(cb => \&server_died,
1982: desc => "Child exit handler",
1983: signal => "CHLD");
1984:
1985:
1986: # Set up all the other signals we set up. We'll vector them off to the
1987: # same subs as we would for DieWhenIdle false and, if necessary, conditionalize
1988: # the code there.
1989:
1990: $parent_handlers{INT} = Event->signal(cb => \&Terminate,
1991: desc => "Parent INT handler",
1992: signal => "INT");
1993: $parent_handlers{TERM} = Event->signal(cb => \&Terminate,
1994: desc => "Parent TERM handler",
1995: signal => "TERM");
1996: $parent_handlers{HUP} = Event->signal(cb => \&Restart,
1997: desc => "Parent HUP handler.",
1998: signal => "HUP");
1999: $parent_handlers{USR1} = Event->signal(cb => \&CheckKids,
2000: desc => "Parent USR1 handler",
2001: signal => "USR1");
2002: $parent_handlers{USR2} = Event->signal(cb => \&UpdateKids,
2003: desc => "Parent USR2 handler.",
2004: signal => "USR2");
2005:
2006: # Start procdesing events.
2007:
2008: $Event::DebugLevel = $DebugLevel;
2009: Debug(9, "Parent entering event loop");
2010: my $ret = Event::loop();
2011: die "Main Event loop exited: $ret";
2012:
2013:
2014: } else {
2015: #
2016: # Set up parent signals:
2017: #
2018:
2019: $SIG{INT} = \&Terminate;
2020: $SIG{TERM} = \&Terminate;
2021: $SIG{HUP} = \&Restart;
2022: $SIG{USR1} = \&CheckKids;
2023: $SIG{USR2} = \&UpdateKids; # LonManage update request.
2024:
2025: while(1) {
2026: my $deadchild = wait();
2027: if(exists $ChildHash{$deadchild}) { # need to restart.
2028: my $deadhost = $ChildHash{$deadchild};
2029: delete($HostToPid{$deadhost});
2030: delete($ChildHash{$deadchild});
2031: Log("WARNING","Lost child pid= ".$deadchild.
2032: "Connected to host ".$deadhost);
2033: Log("INFO", "Restarting child procesing ".$deadhost);
2034: CreateChild($deadhost);
2035: }
2036: }
2037: }
2038:
2039:
2040: =pod
2041:
2042: =head1 CheckKids
2043:
2044: Since kids do not die as easily in this implementation
2045: as the previous one, there is no need to restart the
2046: dead ones (all dead kids get restarted when they die!!)
2047: The only thing this function does is to pass USR1 to the
2048: kids so that they report their status.
2049:
2050: =cut
2051:
2052: sub CheckKids {
2053: Debug(2, "Checking status of children");
2054: my $docdir = $perlvar{'lonDocRoot'};
2055: my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
2056: my $now=time;
2057: my $local=localtime($now);
2058: print $fh "LONC status $local - parent $$ \n\n";
2059: foreach my $host (keys %parent_dispatchers) {
2060: print $fh "LONC Parent process listening for $host\n";
2061: }
2062: foreach my $pid (keys %ChildHash) {
2063: Debug(2, "Sending USR1 -> $pid");
2064: kill 'USR1' => $pid; # Tell Child to report status.
2065: }
2066:
2067: }
2068:
2069: =pod
2070:
2071: =head1 UpdateKids
2072:
2073: parent's SIGUSR2 handler. This handler:
2074:
2075: =item
2076:
2077: Rereads the hosts file.
2078:
2079: =item
2080:
2081: Kills off (via sigint) children for hosts that have disappeared.
2082:
2083: =item
2084:
2085: QUITs children for hosts that already exist (this just forces a status display
2086: and resets the connection retry count for that host.
2087:
2088: =item
2089:
2090: Starts new children for hosts that have been added to the hosts.tab file since
2091: the start of the master program and maintains them.
2092:
2093: =cut
2094:
2095: sub UpdateKids {
2096:
2097: Log("INFO", "Updating connections via SIGUSR2");
2098:
2099: # I'm not sure what I was thinking in the first implementation.
2100: # someone will have to work hard to convince me the effect is any
2101: # different than Restart, especially now that we don't start up
2102: # per host servers automatically, may as well just restart.
2103: # The down side is transactions that are in flight will get timed out
2104: # (lost unless they are critical).
2105:
2106: &Restart();
2107:
2108: }
2109:
2110:
2111: =pod
2112:
2113: =head1 Restart
2114:
2115: Signal handler for HUP... all children are killed and
2116: we self restart. This is an el-cheapo way to re read
2117: the config file.
2118:
2119: =cut
2120:
2121: sub Restart {
2122: &KillThemAll; # First kill all the children.
2123: Log("CRITICAL", "Restarting");
2124: my $execdir = $perlvar{'lonDaemons'};
2125: unlink("$execdir/logs/lonc.pid");
2126: exec("$executable");
2127: }
2128:
2129: =pod
2130:
2131: =head1 KillThemAll
2132:
2133: Signal handler that kills all children by sending them a
2134: SIGHUP. Responds to sigint and sigterm.
2135:
2136: =cut
2137:
2138: sub KillThemAll {
2139: Debug(2, "Kill them all!!");
2140: local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die.
2141: foreach my $pid (keys %ChildHash) {
2142: my $serving = $ChildHash{$pid};
2143: ShowStatus("Nicely Killing lonc for $serving pid = $pid");
2144: Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
2145: kill 'QUIT' => $pid;
2146: }
2147:
2148:
2149: }
2150:
2151:
2152: #
2153: # Kill all children via KILL. Just in case the
2154: # first shot didn't get them.
2155:
2156: sub really_kill_them_all_dammit
2157: {
2158: Debug(2, "Kill them all Dammit");
2159: local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
2160: foreach my $pid (keys %ChildHash) {
2161: my $serving = $ChildHash{$pid};
2162: &ShowStatus("Nastily killing lonc for $serving pid = $pid");
2163: Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
2164: kill 'KILL' => $pid;
2165: delete($ChildHash{$pid});
2166: my $execdir = $perlvar{'lonDaemons'};
2167: unlink("$execdir/logs/lonc.pid");
2168: }
2169: }
2170: =pod
2171:
2172: =head1 Terminate
2173:
2174: Terminate the system.
2175:
2176: =cut
2177:
2178: sub Terminate {
2179: &Log("CRITICAL", "Asked to kill children.. first be nice...");
2180: &KillThemAll;
2181: #
2182: # By now they really should all be dead.. but just in case
2183: # send them all SIGKILL's after a bit of waiting:
2184:
2185: sleep(4);
2186: &Log("CRITICAL", "Now kill children nasty");
2187: &really_kill_them_all_dammit;
2188: Log("CRITICAL","Master process exiting");
2189: exit 0;
2190:
2191: }
2192: =pod
2193:
2194: =head1 Theory
2195:
2196: The event class is used to build this as a single process with an
2197: event driven model. The following events are handled:
2198:
2199: =item UNIX Socket connection Received
2200:
2201: =item Request data arrives on UNIX data transfer socket.
2202:
2203: =item lond connection becomes writable.
2204:
2205: =item timer fires at 1 second intervals.
2206:
2207: All sockets are run in non-blocking mode. Timeouts managed by the timer
2208: handler prevents hung connections.
2209:
2210: Key data structures:
2211:
2212: =item RequestQueue
2213:
2214: A queue of requests received from UNIX sockets that are
2215: waiting for a chance to be forwarded on a lond connection socket.
2216:
2217: =item ActiveConnections
2218:
2219: A hash of lond connections that have transactions in process that are
2220: available to be timed out.
2221:
2222: =item ActiveTransactions
2223:
2224: A hash indexed by lond connections that contain the client reply
2225: socket for each connection that has an active transaction on it.
2226:
2227: =item IdleConnections
2228:
2229: A hash of lond connections that have no work to do. These connections
2230: can be closed if they are idle for a long enough time.
2231:
2232: =cut
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/icons/back.gif){kind=icon}
Return to loncnew CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [LON-CAPA] / loncom