1: #!/usr/bin/perl
2: # The LearningOnline Network with CAPA
3: # lonc maintains the connections to remote computers
4: #
5: # $Id: loncnew,v 1.72 2005/08/29 18:22:48 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=5; # Number of connection retries allowed.
110: my $ConnectionRetriesLeft=5; # 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:
156: sub LogPerm {
157: my $message=shift;
158: my $execdir=$perlvar{'lonDaemons'};
159: my $now=time;
160: my $local=localtime($now);
161: my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
162: print $fh "$now:$message:$local\n";
163: }
164:
165: =pod
166:
167: =head2 Log
168:
169: Logs a message to the log file.
170: Parameters:
171:
172: =item severity
173:
174: One of CRITICAL, WARNING, INFO, SUCCESS used to select the
175: format string used to format the message. if the severity is
176: not a defined severity the Default format string is used.
177:
178: =item message
179:
180: The base message. In addtion to the format string, the message
181: will be appended to a string containing the name of our remote
182: host and the time will be formatted into the message.
183:
184: =cut
185:
186: sub Log {
187:
188: my ($severity, $message) = @_;
189:
190: if(!$LogFormats{$severity}) {
191: $severity = "DEFAULT";
192: }
193:
194: my $format = $LogFormats{$severity};
195:
196: # Put the window dressing in in front of the message format:
197:
198: my $now = time;
199: my $local = localtime($now);
200: my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
201: my $finalformat = $finalformat.$format."\n";
202:
203: # open the file and put the result.
204:
205: my $execdir = $perlvar{'lonDaemons'};
206: my $fh = IO::File->new(">>$execdir/logs/lonc.log");
207: my $msg = sprintf($finalformat, $message);
208: $RecentLogEntry = $msg;
209: print $fh $msg;
210:
211:
212: }
213:
214:
215: =pod
216:
217: =head2 GetPeerName
218:
219: Returns the name of the host that a socket object is connected to.
220:
221: =cut
222:
223: sub GetPeername {
224:
225:
226: my ($connection, $AdrFamily) = @_;
227:
228: my $peer = $connection->peername();
229: my $peerport;
230: my $peerip;
231: if($AdrFamily == AF_INET) {
232: ($peerport, $peerip) = sockaddr_in($peer);
233: my $peername = gethostbyaddr($peerip, $AdrFamily);
234: return $peername;
235: } elsif ($AdrFamily == AF_UNIX) {
236: my $peerfile;
237: ($peerfile) = sockaddr_un($peer);
238: return $peerfile;
239: }
240: }
241: =pod
242:
243: =head2 Debug
244:
245: Invoked to issue a debug message.
246:
247: =cut
248:
249: sub Debug {
250:
251: my ($level, $message) = @_;
252:
253: if ($level <= $DebugLevel) {
254: Log("INFO", "-Debug- $message host = $RemoteHost");
255: }
256: }
257:
258: sub SocketDump {
259:
260: my ($level, $socket) = @_;
261:
262: if($level <= $DebugLevel) {
263: $socket->Dump(-1); # Ensure it will get dumped.
264: }
265: }
266:
267: =pod
268:
269: =head2 ShowStatus
270:
271: Place some text as our pid status.
272: and as what we return in a SIGUSR1
273:
274: =cut
275:
276: sub ShowStatus {
277: my $state = shift;
278: my $now = time;
279: my $local = localtime($now);
280: $Status = $local.": ".$state;
281: $0='lonc: '.$state.' '.$local;
282: }
283:
284: =pod
285:
286: =head2 SocketTimeout
287:
288: Called when an action on the socket times out. The socket is
289: destroyed and any active transaction is failed.
290:
291:
292: =cut
293:
294: sub SocketTimeout {
295: my $Socket = shift;
296: Log("WARNING", "A socket timeout was detected");
297: Debug(5, " SocketTimeout called: ");
298: $Socket->Dump(0);
299: if(exists($ActiveTransactions{$Socket})) {
300: FailTransaction($ActiveTransactions{$Socket});
301: }
302: KillSocket($Socket); # A transaction timeout also counts as
303: # a connection failure:
304: $ConnectionRetriesLeft--;
305: if($ConnectionRetriesLeft <= 0) {
306: Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
307: $LondConnecting = 0;
308: }
309:
310: }
311: #
312: # This function should be called by the child in all cases where it must
313: # exit. If the child process is running with the DieWhenIdle turned on
314: # it must create a lock file for the AF_UNIX socket in order to prevent
315: # connection requests from lonnet in the time between process exit
316: # and the parent picking up the listen again.
317: # Parameters:
318: # exit_code - Exit status value, however see the next parameter.
319: # message - If this optional parameter is supplied, the exit
320: # is via a die with this message.
321: #
322: sub child_exit {
323: my ($exit_code, $message) = @_;
324:
325: # Regardless of how we exit, we may need to do the lock thing:
326:
327: if($DieWhenIdle) {
328: #
329: # Create a lock file since there will be a time window
330: # between our exit and the parent's picking up the listen
331: # during which no listens will be done on the
332: # lonnet client socket.
333: #
334: my $lock_file = GetLoncSocketPath().".lock";
335: open(LOCK,">$lock_file");
336: print LOCK "Contents not important";
337: close(LOCK);
338:
339: exit(0);
340: }
341: # Now figure out how we exit:
342:
343: if($message) {
344: die $message;
345: } else {
346: exit($exit_code);
347: }
348: }
349: #----------------------------- Timer management ------------------------
350:
351: =pod
352:
353: =head2 Tick
354:
355: Invoked each timer tick.
356:
357: =cut
358:
359:
360: sub Tick {
361: my ($Event) = @_;
362: my $clock_watcher = $Event->w;
363:
364: my $client;
365: UpdateStatus();
366:
367: # Is it time to prune connection count:
368:
369:
370: if($IdleConnections->Count() &&
371: ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
372: $IdleSeconds++;
373: if($IdleSeconds > $IdleTimeout) { # Prune a connection...
374: my $Socket = $IdleConnections->pop();
375: KillSocket($Socket);
376: $IdleSeconds = 0; # Otherwise all connections get trimmed to fast.
377: UpdateStatus();
378: if(($ConnectionCount == 0) && $DieWhenIdle) {
379: &child_exit(0);
380:
381: }
382: }
383: } else {
384: $IdleSeconds = 0; # Reset idle count if not idle.
385: }
386: #
387: # For each inflight transaction, tick down its timeout counter.
388: #
389:
390: foreach my $item (keys %ActiveConnections) {
391: my $State = $ActiveConnections{$item}->data->GetState();
392: if ($State ne 'Idle') {
393: Debug(5,"Ticking Socket $State $item");
394: $ActiveConnections{$item}->data->Tick();
395: }
396: }
397: # Do we have work in the queue, but no connections to service them?
398: # If so, try to make some new connections to get things going again.
399: #
400: # Note this code is dead now...
401: #
402: my $Requests = $WorkQueue->Count();
403: if (($ConnectionCount == 0) && ($Requests > 0) && (!$LondConnecting)) {
404: if ($ConnectionRetriesLeft > 0) {
405: Debug(5,"Work but no connections, Make a new one");
406: my $success;
407: $success = &MakeLondConnection;
408: if($success == 0) { # All connections failed:
409: Debug(5,"Work in queue failed to make any connectiouns\n");
410: EmptyQueue(); # Fail pending transactions with con_lost.
411: CloseAllLondConnections(); # Should all be closed but....
412: }
413: } else {
414: $LondConnecting = 0;
415: ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
416: Debug(5,"Work in queue, but gave up on connections..flushing\n");
417: EmptyQueue(); # Connections can't be established.
418: CloseAllLondConnections(); # Should all already be closed but...
419: }
420:
421: }
422: if ($ConnectionCount == 0) {
423: $KeyMode = "";
424: $clock_watcher->cancel();
425: }
426: &UpdateStatus();
427: }
428:
429: =pod
430:
431: =head2 SetupTimer
432:
433: Sets up a 1 per sec recurring timer event. The event handler is used to:
434:
435: =item
436:
437: Trigger timeouts on communications along active sockets.
438:
439: =item
440:
441: Trigger disconnections of idle sockets.
442:
443: =cut
444:
445: sub SetupTimer {
446: Debug(6, "SetupTimer");
447: Event->timer(interval => 1, cb => \&Tick );
448: }
449:
450: =pod
451:
452: =head2 ServerToIdle
453:
454: This function is called when a connection to the server is
455: ready for more work.
456:
457: If there is work in the Work queue the top element is dequeued
458: and the connection will start to work on it. If the work queue is
459: empty, the connection is pushed on the idle connection stack where
460: it will either get another work unit, or alternatively, if it sits there
461: long enough, it will be shut down and released.
462:
463: =cut
464:
465: sub ServerToIdle {
466: my $Socket = shift; # Get the socket.
467: $KeyMode = $Socket->{AuthenticationMode};
468: delete($ActiveTransactions{$Socket}); # Server has no transaction
469:
470: &Debug(5, "Server to idle");
471:
472: # If there's work to do, start the transaction:
473:
474: my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
475: if ($reqdata ne undef) {
476: Debug(5, "Queue gave request data: ".$reqdata->getRequest());
477: &StartRequest($Socket, $reqdata);
478:
479: } else {
480:
481: # There's no work waiting, so push the server to idle list.
482: &Debug(5, "No new work requests, server connection going idle");
483: $IdleConnections->push($Socket);
484: }
485: }
486:
487: =pod
488:
489: =head2 ClientWritable
490:
491: Event callback for when a client socket is writable.
492:
493: This callback is established when a transaction reponse is
494: avaiable from lond. The response is forwarded to the unix socket
495: as it becomes writable in this sub.
496:
497: Parameters:
498:
499: =item Event
500:
501: The event that has been triggered. Event->w->data is
502: the data and Event->w->fd is the socket to write.
503:
504: =cut
505:
506: sub ClientWritable {
507: my $Event = shift;
508: my $Watcher = $Event->w;
509: my $Data = $Watcher->data;
510: my $Socket = $Watcher->fd;
511:
512: # Try to send the data:
513:
514: &Debug(6, "ClientWritable writing".$Data);
515: &Debug(9, "Socket is: ".$Socket);
516:
517: if($Socket->connected) {
518: my $result = $Socket->send($Data, 0);
519:
520: # $result undefined: the write failed.
521: # otherwise $result is the number of bytes written.
522: # Remove that preceding string from the data.
523: # If the resulting data is empty, destroy the watcher
524: # and set up a read event handler to accept the next
525: # request.
526:
527: &Debug(9,"Send result is ".$result." Defined: ".defined($result));
528: if($result ne undef) {
529: &Debug(9, "send result was defined");
530: if($result == length($Data)) { # Entire string sent.
531: &Debug(9, "ClientWritable data all written");
532: $Watcher->cancel();
533: #
534: # Set up to read next request from socket:
535:
536: my $descr = sprintf("Connection to lonc client %d",
537: $ActiveClients{$Socket});
538: Event->io(cb => \&ClientRequest,
539: poll => 'r',
540: desc => $descr,
541: data => "",
542: fd => $Socket);
543:
544: } else { # Partial string sent.
545: $Watcher->data(substr($Data, $result));
546: if($result == 0) { # client hung up on us!!
547: # Log("INFO", "lonc pipe client hung up on us!");
548: $Watcher->cancel;
549: $Socket->shutdown(2);
550: $Socket->close();
551: }
552: }
553:
554: } else { # Error of some sort...
555:
556: # Some errnos are possible:
557: my $errno = $!;
558: if($errno == POSIX::EWOULDBLOCK ||
559: $errno == POSIX::EAGAIN ||
560: $errno == POSIX::EINTR) {
561: # No action taken?
562: } else { # Unanticipated errno.
563: &Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
564: $Watcher->cancel; # Stop the watcher.
565: $Socket->shutdown(2); # Kill connection
566: $Socket->close(); # Close the socket.
567: }
568:
569: }
570: } else {
571: $Watcher->cancel(); # A delayed request...just cancel.
572: }
573: }
574:
575: =pod
576:
577: =head2 CompleteTransaction
578:
579: Called when the reply data has been received for a lond
580: transaction. The reply data must now be sent to the
581: ultimate client on the other end of the Unix socket. This is
582: done by setting up a writable event for the socket with the
583: data the reply data.
584:
585: Parameters:
586:
587: =item Socket
588:
589: Socket on which the lond transaction occured. This is a
590: LondConnection. The data received is in the TransactionReply member.
591:
592: =item Transaction
593:
594: The transaction that is being completed.
595:
596: =cut
597:
598: sub CompleteTransaction {
599: &Debug(5,"Complete transaction");
600:
601: my ($Socket, $Transaction) = @_;
602:
603: if (!$Transaction->isDeferred()) { # Normal transaction
604: my $data = $Socket->GetReply(); # Data to send.
605: if($LogTransactions) {
606: Log("SUCCESS", "Reply from lond: '$data'");
607: }
608: StartClientReply($Transaction, $data);
609: } else { # Delete deferred transaction file.
610: Log("SUCCESS", "A delayed transaction was completed");
611: LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest());
612: unlink $Transaction->getFile();
613: }
614: }
615:
616: =pod
617:
618: =head1 StartClientReply
619:
620: Initiates a reply to a client where the reply data is a parameter.
621:
622: =head2 parameters:
623:
624: =item Transaction
625:
626: The transaction for which we are responding to the client.
627:
628: =item data
629:
630: The data to send to apached client.
631:
632: =cut
633:
634: sub StartClientReply {
635:
636: my ($Transaction, $data) = @_;
637:
638: my $Client = $Transaction->getClient();
639:
640: &Debug(8," Reply was: ".$data);
641: my $Serial = $ActiveClients{$Client};
642: my $desc = sprintf("Connection to lonc client %d",
643: $Serial);
644: Event->io(fd => $Client,
645: poll => "w",
646: desc => $desc,
647: cb => \&ClientWritable,
648: data => $data);
649: }
650:
651: =pod
652:
653: =head2 FailTransaction
654:
655: Finishes a transaction with failure because the associated lond socket
656: disconnected. There are two possibilities:
657: - The transaction is deferred: in which case we just quietly
658: delete the transaction since there is no client connection.
659: - The transaction is 'live' in which case we initiate the sending
660: of "con_lost" to the client.
661:
662: Deleting the transaction means killing it from the %ActiveTransactions hash.
663:
664: Parameters:
665:
666: =item client
667:
668: The LondTransaction we are failing.
669:
670:
671: =cut
672:
673: sub FailTransaction {
674: my $transaction = shift;
675:
676: # If the socket is dead, that's already logged.
677:
678: if ($ConnectionRetriesLeft > 0) {
679: Log("WARNING", "Failing transaction "
680: .$transaction->getLoggableRequest());
681: }
682: Debug(1, "Failing transaction: ".$transaction->getLoggableRequest());
683: if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
684: my $client = $transaction->getClient();
685: Debug(1," Replying con_lost to ".$transaction->getRequest());
686: StartClientReply($transaction, "con_lost\n");
687: }
688:
689: }
690:
691: =pod
692:
693: =head1 EmptyQueue
694:
695: Fails all items in the work queue with con_lost.
696: Note that each item in the work queue is a transaction.
697:
698: =cut
699:
700: sub EmptyQueue {
701: $ConnectionRetriesLeft--; # Counts as connection failure too.
702: while($WorkQueue->Count()) {
703: my $request = $WorkQueue->dequeue(); # This is a transaction
704: FailTransaction($request);
705: }
706: }
707:
708: =pod
709:
710: =head2 CloseAllLondConnections
711:
712: Close all connections open on lond prior to exit e.g.
713:
714: =cut
715:
716: sub CloseAllLondConnections {
717: foreach my $Socket (keys %ActiveConnections) {
718: if(exists($ActiveTransactions{$Socket})) {
719: FailTransaction($ActiveTransactions{$Socket});
720: }
721: KillSocket($Socket);
722: }
723: }
724:
725: =pod
726:
727: =head2 KillSocket
728:
729: Destroys a socket. This function can be called either when a socket
730: has died of 'natural' causes or because a socket needs to be pruned due to
731: idleness. If the socket has died naturally, if there are no longer any
732: live connections a new connection is created (in case there are transactions
733: in the queue). If the socket has been pruned, it is never re-created.
734:
735: Parameters:
736:
737: =item Socket
738:
739: The socket to kill off.
740:
741: =item Restart
742:
743: nonzero if we are allowed to create a new connection.
744:
745: =cut
746:
747: sub KillSocket {
748: my $Socket = shift;
749:
750: Log("WARNING", "Shutting down a socket");
751: $Socket->Shutdown();
752:
753: # If the socket came from the active connection set,
754: # delete its transaction... note that FailTransaction should
755: # already have been called!!!
756: # otherwise it came from the idle set.
757: #
758:
759: if(exists($ActiveTransactions{$Socket})) {
760: delete ($ActiveTransactions{$Socket});
761: }
762: if(exists($ActiveConnections{$Socket})) {
763: delete($ActiveConnections{$Socket});
764: $ConnectionCount--;
765: if ($ConnectionCount < 0) { $ConnectionCount = 0; }
766: }
767: # If the connection count has gone to zero and there is work in the
768: # work queue, the work all gets failed with con_lost.
769: #
770: if($ConnectionCount == 0) {
771: EmptyQueue();
772: CloseAllLondConnections; # Should all already be closed but...
773: }
774: }
775:
776: =pod
777:
778: =head2 LondReadable
779:
780: This function is called whenever a lond connection
781: is readable. The action is state dependent:
782:
783: =head3 State=Initialized
784:
785: We''re waiting for the challenge, this is a no-op until the
786: state changes.
787:
788: =head3 State=Challenged
789:
790: The challenge has arrived we need to transition to Writable.
791: The connection must echo the challenge back.
792:
793: =head3 State=ChallengeReplied
794:
795: The challenge has been replied to. The we are receiveing the
796: 'ok' from the partner.
797:
798: =head3 State=ReadingVersionString
799:
800: We have requested the lond version and are reading the
801: version back. Upon completion, we'll store the version away
802: for future use(?).
803:
804: =head3 State=HostSet
805:
806: We have selected the domain name of our peer (multhomed hosts)
807: and are getting the reply (presumably ok) back.
808:
809: =head3 State=RequestingKey
810:
811: The ok has been received and we need to send the request for
812: an encryption key. Transition to writable for that.
813:
814: =head3 State=ReceivingKey
815:
816: The the key has been requested, now we are reading the new key.
817:
818: =head3 State=Idle
819:
820: The encryption key has been negotiated or we have finished
821: reading data from the a transaction. If the callback data has
822: a client as well as the socket iformation, then we are
823: doing a transaction and the data received is relayed to the client
824: before the socket is put on the idle list.
825:
826: =head3 State=SendingRequest
827:
828: I do not think this state can be received here, but if it is,
829: the appropriate thing to do is to transition to writable, and send
830: the request.
831:
832: =head3 State=ReceivingReply
833:
834: We finished sending the request to the server and now transition
835: to readable to receive the reply.
836:
837: The parameter to this function are:
838:
839: The event. Implicit in this is the watcher and its data. The data
840: contains at least the lond connection object and, if a
841: transaction is in progress, the socket attached to the local client.
842:
843: =cut
844:
845: sub LondReadable {
846:
847: my $Event = shift;
848: my $Watcher = $Event->w;
849: my $Socket = $Watcher->data;
850: my $client = undef;
851:
852: &Debug(6,"LondReadable called state = ".$Socket->GetState());
853:
854:
855: my $State = $Socket->GetState(); # All action depends on the state.
856:
857: SocketDump(6, $Socket);
858: my $status = $Socket->Readable();
859:
860: &Debug(2, "Socket->Readable returned: $status");
861:
862: if($status != 0) {
863: # bad return from socket read. Currently this means that
864: # The socket has become disconnected. We fail the transaction.
865:
866: Log("WARNING",
867: "Lond connection lost.");
868: if(exists($ActiveTransactions{$Socket})) {
869: FailTransaction($ActiveTransactions{$Socket});
870: } else {
871: # Socket is connecting and failed... need to mark
872: # no longer connecting.
873:
874: $LondConnecting = 0;
875: }
876: $Watcher->cancel();
877: KillSocket($Socket);
878: $ConnectionRetriesLeft--; # Counts as connection failure
879: return;
880: }
881: SocketDump(6,$Socket);
882:
883: $State = $Socket->GetState(); # Update in case of transition.
884: &Debug(6, "After read, state is ".$State);
885:
886: if($State eq "Initialized") {
887:
888:
889: } elsif ($State eq "ChallengeReceived") {
890: # The challenge must be echoed back; The state machine
891: # in the connection takes care of setting that up. Just
892: # need to transition to writable:
893:
894: $Watcher->cb(\&LondWritable);
895: $Watcher->poll("w");
896:
897: } elsif ($State eq "ChallengeReplied") {
898:
899: } elsif ($State eq "RequestingVersion") {
900: # Need to ask for the version... that is writiability:
901:
902: $Watcher->cb(\&LondWritable);
903: $Watcher->poll("w");
904:
905: } elsif ($State eq "ReadingVersionString") {
906: # Read the rest of the version string...
907: } elsif ($State eq "SetHost") {
908: # Need to request the actual domain get set...
909:
910: $Watcher->cb(\&LondWritable);
911: $Watcher->poll("w");
912: } elsif ($State eq "HostSet") {
913: # Reading the 'ok' from the peer.
914:
915: } elsif ($State eq "RequestingKey") {
916: # The ok was received. Now we need to request the key
917: # That requires us to be writable:
918:
919: $Watcher->cb(\&LondWritable);
920: $Watcher->poll("w");
921:
922: } elsif ($State eq "ReceivingKey") {
923:
924: } elsif ($State eq "Idle") {
925:
926: # This is as good a spot as any to get the peer version
927: # string:
928:
929: if($LondVersion eq "unknown") {
930: $LondVersion = $Socket->PeerVersion();
931: Log("INFO", "Connected to lond version: $LondVersion");
932: }
933: # If necessary, complete a transaction and then go into the
934: # idle queue.
935: # Note that a trasition to idle indicates a live lond
936: # on the other end so reset the connection retries.
937: #
938: $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
939: $Watcher->cancel();
940: if(exists($ActiveTransactions{$Socket})) {
941: Debug(5,"Completing transaction!!");
942: CompleteTransaction($Socket,
943: $ActiveTransactions{$Socket});
944: } else {
945: Log("SUCCESS", "Connection ".$ConnectionCount." to "
946: .$RemoteHost." now ready for action");
947: }
948: ServerToIdle($Socket); # Next work unit or idle.
949:
950: #
951: $LondConnecting = 0; # Best spot I can think of for this.
952: #
953:
954: } elsif ($State eq "SendingRequest") {
955: # We need to be writable for this and probably don't belong
956: # here inthe first place.
957:
958: Deubg(6, "SendingRequest state encountered in readable");
959: $Watcher->poll("w");
960: $Watcher->cb(\&LondWritable);
961:
962: } elsif ($State eq "ReceivingReply") {
963:
964:
965: } else {
966: # Invalid state.
967: Debug(4, "Invalid state in LondReadable");
968: }
969: }
970:
971: =pod
972:
973: =head2 LondWritable
974:
975: This function is called whenever a lond connection
976: becomes writable while there is a writeable monitoring
977: event. The action taken is very state dependent:
978:
979: =head3 State = Connected
980:
981: The connection is in the process of sending the 'init' hailing to the
982: lond on the remote end. The connection object''s Writable member is
983: called. On error, ConnectionError is called to destroy the connection
984: and remove it from the ActiveConnections hash
985:
986: =head3 Initialized
987:
988: 'init' has been sent, writability monitoring is removed and
989: readability monitoring is started with LondReadable as the callback.
990:
991: =head3 ChallengeReceived
992:
993: The connection has received the who are you challenge from the remote
994: system, and is in the process of sending the challenge
995: response. Writable is called.
996:
997: =head3 ChallengeReplied
998:
999: The connection has replied to the initial challenge The we switch to
1000: monitoring readability looking for the server to reply with 'ok'.
1001:
1002: =head3 RequestingKey
1003:
1004: The connection is in the process of requesting its encryption key.
1005: Writable is called.
1006:
1007: =head3 ReceivingKey
1008:
1009: The connection has sent the request for a key. Switch to readability
1010: monitoring to accept the key
1011:
1012: =head3 SendingRequest
1013:
1014: The connection is in the process of sending a request to the server.
1015: This request is part of a client transaction. All the states until
1016: now represent the client setup protocol. Writable is called.
1017:
1018: =head3 ReceivingReply
1019:
1020: The connection has sent a request. Now it must receive a reply.
1021: Readability monitoring is requested.
1022:
1023: This function is an event handler and therefore receives as
1024: a parameter the event that has fired. The data for the watcher
1025: of this event is a reference to a list of one or two elements,
1026: depending on state. The first (and possibly only) element is the
1027: socket. The second (present only if a request is in progress)
1028: is the socket on which to return a reply to the caller.
1029:
1030: =cut
1031:
1032: sub LondWritable {
1033: my $Event = shift;
1034: my $Watcher = $Event->w;
1035: my $Socket = $Watcher->data;
1036: my $State = $Socket->GetState();
1037:
1038: Debug(6,"LondWritable State = ".$State."\n");
1039:
1040:
1041: # Figure out what to do depending on the state of the socket:
1042:
1043:
1044:
1045:
1046: SocketDump(6,$Socket);
1047:
1048: # If the socket is writable, we must always write.
1049: # Only by writing will we undergo state transitions.
1050: # Old logic wrote in state specific code below, however
1051: # That forces us at least through another invocation of
1052: # this function after writability is possible again.
1053: # This logic also factors out common code for handling
1054: # write failures... in all cases, write failures
1055: # Kill the socket.
1056: # This logic makes the branches of the >big< if below
1057: # so that the writing states are actually NO-OPs.
1058:
1059: if ($Socket->Writable() != 0) {
1060: # The write resulted in an error.
1061: # We'll treat this as if the socket got disconnected:
1062: Log("WARNING", "Connection to ".$RemoteHost.
1063: " has been disconnected");
1064: if(exists($ActiveTransactions{$Socket})) {
1065: FailTransaction($ActiveTransactions{$Socket});
1066: } else {
1067: # In the process of conneting, so need to turn that off.
1068:
1069: $LondConnecting = 0;
1070: }
1071: $Watcher->cancel();
1072: KillSocket($Socket);
1073: return;
1074: }
1075:
1076:
1077:
1078: if ($State eq "Connected") {
1079:
1080: # "init" is being sent...
1081:
1082: } elsif ($State eq "Initialized") {
1083:
1084: # Now that init was sent, we switch
1085: # to watching for readability:
1086:
1087: $Watcher->cb(\&LondReadable);
1088: $Watcher->poll("r");
1089:
1090: } elsif ($State eq "ChallengeReceived") {
1091: # We received the challenge, now we
1092: # are echoing it back. This is a no-op,
1093: # we're waiting for the state to change
1094:
1095: } elsif ($State eq "ChallengeReplied") {
1096: # The echo was sent back, so we switch
1097: # to watching readability.
1098:
1099: $Watcher->cb(\&LondReadable);
1100: $Watcher->poll("r");
1101: } elsif ($State eq "RequestingVersion") {
1102: # Sending the peer a version request...
1103:
1104: } elsif ($State eq "ReadingVersionString") {
1105: # Transition to read since we have sent the
1106: # version command and now just need to read the
1107: # version string from the peer:
1108:
1109: $Watcher->cb(\&LondReadable);
1110: $Watcher->poll("r");
1111:
1112: } elsif ($State eq "SetHost") {
1113: # Setting the remote domain...
1114:
1115: } elsif ($State eq "HostSet") {
1116: # Back to readable to get the ok.
1117:
1118: $Watcher->cb(\&LondReadable);
1119: $Watcher->poll("r");
1120:
1121:
1122: } elsif ($State eq "RequestingKey") {
1123: # At this time we're requesting the key.
1124: # again, this is essentially a no-op.
1125:
1126: } elsif ($State eq "ReceivingKey") {
1127: # Now we need to wait for the key
1128: # to come back from the peer:
1129:
1130: $Watcher->cb(\&LondReadable);
1131: $Watcher->poll("r");
1132:
1133: } elsif ($State eq "SendingRequest") {
1134:
1135: # At this time we are sending a request to the
1136: # peer... write the next chunk:
1137:
1138:
1139: } elsif ($State eq "ReceivingReply") {
1140: # The send has completed. Wait for the
1141: # data to come in for a reply.
1142: Debug(8,"Writable sent request/receiving reply");
1143: $Watcher->cb(\&LondReadable);
1144: $Watcher->poll("r");
1145:
1146: } else {
1147: # Control only passes here on an error:
1148: # the socket state does not match any
1149: # of the known states... so an error
1150: # must be logged.
1151:
1152: &Debug(4, "Invalid socket state ".$State."\n");
1153: }
1154:
1155: }
1156: =pod
1157:
1158: =cut
1159:
1160: sub QueueDelayed {
1161: Debug(3,"QueueDelayed called");
1162:
1163: my $path = "$perlvar{'lonSockDir'}/delayed";
1164:
1165: Debug(4, "Delayed path: ".$path);
1166: opendir(DIRHANDLE, $path);
1167:
1168: my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE;
1169: closedir(DIRHANDLE);
1170: my $dfname;
1171: my $reqfile;
1172: foreach $dfname (sort @alldelayed) {
1173: $reqfile = "$path/$dfname";
1174: Debug(4, "queueing ".$reqfile);
1175: my $Handle = IO::File->new($reqfile);
1176: my $cmd = <$Handle>;
1177: chomp $cmd; # There may or may not be a newline...
1178: $cmd = $cmd."\n"; # now for sure there's exactly one newline.
1179: my $Transaction = LondTransaction->new($cmd);
1180: $Transaction->SetDeferred($reqfile);
1181: QueueTransaction($Transaction);
1182: }
1183:
1184: }
1185:
1186: =pod
1187:
1188: =head2 MakeLondConnection
1189:
1190: Create a new lond connection object, and start it towards its initial
1191: idleness. Once idle, it becomes elligible to receive transactions
1192: from the work queue. If the work queue is not empty when the
1193: connection is completed and becomes idle, it will dequeue an entry and
1194: start off on it.
1195:
1196: =cut
1197:
1198: sub MakeLondConnection {
1199: Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
1200: .GetServerPort());
1201:
1202: my $Connection = LondConnection->new(&GetServerHost(),
1203: &GetServerPort());
1204:
1205: if($Connection eq undef) { # Needs to be more robust later.
1206: Log("CRITICAL","Failed to make a connection with lond.");
1207: $ConnectionRetriesLeft--;
1208: return 0; # Failure.
1209: } else {
1210:
1211: # The connection needs to have writability
1212: # monitored in order to send the init sequence
1213: # that starts the whole authentication/key
1214: # exchange underway.
1215: #
1216: my $Socket = $Connection->GetSocket();
1217: if($Socket eq undef) {
1218: &child_exit(-1, "did not get a socket from the connection");
1219: } else {
1220: &Debug(9,"MakeLondConnection got socket: ".$Socket);
1221: }
1222:
1223: $Connection->SetTimeoutCallback(\&SocketTimeout);
1224:
1225: my $event = Event->io(fd => $Socket,
1226: poll => 'w',
1227: cb => \&LondWritable,
1228: data => $Connection,
1229: desc => 'Connection to lond server');
1230: $ActiveConnections{$Connection} = $event;
1231: if ($ConnectionCount == 0) {
1232: &SetupTimer; # Need to handle timeouts with connections...
1233: }
1234: $ConnectionCount++;
1235: Debug(4, "Connection count = ".$ConnectionCount);
1236: if($ConnectionCount == 1) { # First Connection:
1237: QueueDelayed;
1238: }
1239: Log("SUCESS", "Created connection ".$ConnectionCount
1240: ." to host ".GetServerHost());
1241: $LondConnecting = 1; # Connection in progress.
1242: return 1; # Return success.
1243: }
1244:
1245: }
1246:
1247: =pod
1248:
1249: =head2 StartRequest
1250:
1251: Starts a lond request going on a specified lond connection.
1252: parameters are:
1253:
1254: =item $Lond
1255:
1256: Connection to the lond that will send the transaction and receive the
1257: reply.
1258:
1259: =item $Client
1260:
1261: Connection to the client that is making this request We got the
1262: request from this socket, and when the request has been relayed to
1263: lond and we get a reply back from lond it will get sent to this
1264: socket.
1265:
1266: =item $Request
1267:
1268: The text of the request to send.
1269:
1270: =cut
1271:
1272: sub StartRequest {
1273:
1274: my ($Lond, $Request) = @_;
1275:
1276: Debug(6, "StartRequest: ".$Request->getRequest());
1277:
1278: my $Socket = $Lond->GetSocket();
1279:
1280: $Request->Activate($Lond);
1281: $ActiveTransactions{$Lond} = $Request;
1282:
1283: $Lond->InitiateTransaction($Request->getRequest());
1284: my $event = Event->io(fd => $Socket,
1285: poll => "w",
1286: cb => \&LondWritable,
1287: data => $Lond,
1288: desc => "lond transaction connection");
1289: $ActiveConnections{$Lond} = $event;
1290: Debug(8," Start Request made watcher data with ".$event->data."\n");
1291: }
1292:
1293: =pod
1294:
1295: =head2 QueueTransaction
1296:
1297: If there is an idle lond connection, it is put to work doing this
1298: transaction. Otherwise, the transaction is placed in the work queue.
1299: If placed in the work queue and the maximum number of connections has
1300: not yet been created, a new connection will be started. Our goal is
1301: to eventually have a sufficient number of connections that the work
1302: queue will typically be empty. parameters are:
1303:
1304: =item Socket
1305:
1306: open on the lonc client.
1307:
1308: =item Request
1309:
1310: data to send to the lond.
1311:
1312: =cut
1313:
1314: sub QueueTransaction {
1315:
1316: my $requestData = shift; # This is a LondTransaction.
1317: my $cmd = $requestData->getRequest();
1318:
1319: Debug(6,"QueueTransaction: ".$cmd);
1320:
1321: my $LondSocket = $IdleConnections->pop();
1322: if(!defined $LondSocket) { # Need to queue request.
1323: Debug(5,"Must queue...");
1324: $WorkQueue->enqueue($requestData);
1325: Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
1326: if(($ConnectionCount < $MaxConnectionCount) && (! $LondConnecting)) {
1327:
1328: if($ConnectionRetriesLeft > 0) {
1329: Debug(5,"Starting additional lond connection");
1330: if(&MakeLondConnection() == 0) {
1331: EmptyQueue(); # Fail transactions, can't make connection.
1332: CloseAllLondConnections; # Should all be closed but...
1333: }
1334: } else {
1335: ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1336: $LondConnecting = 0;
1337: EmptyQueue(); # It's worse than that ... he's dead Jim.
1338: CloseAllLondConnections; # Should all be closed but..
1339: }
1340: }
1341: } else { # Can start the request:
1342: Debug(8,"Can start...");
1343: StartRequest($LondSocket, $requestData);
1344: }
1345: }
1346:
1347: #-------------------------- Lonc UNIX socket handling ---------------------
1348:
1349: =pod
1350:
1351: =head2 ClientRequest
1352: Callback that is called when data can be read from the UNIX domain
1353: socket connecting us with an apache server process.
1354:
1355: =cut
1356:
1357: sub ClientRequest {
1358: Debug(6, "ClientRequest");
1359: my $event = shift;
1360: my $watcher = $event->w;
1361: my $socket = $watcher->fd;
1362: my $data = $watcher->data;
1363: my $thisread;
1364:
1365: Debug(9, " Watcher named: ".$watcher->desc);
1366:
1367: my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
1368: Debug(8, "rcv: data length = ".length($thisread)
1369: ." read =".$thisread);
1370: unless (defined $rv && length($thisread)) {
1371: # Likely eof on socket.
1372: Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
1373: close($socket);
1374: $watcher->cancel();
1375: delete($ActiveClients{$socket});
1376: return;
1377: }
1378: Debug(8,"Data: ".$data." this read: ".$thisread);
1379: $data = $data.$thisread; # Append new data.
1380: $watcher->data($data);
1381: if($data =~ /\n$/) { # Request entirely read.
1382: if($data eq "close_connection_exit\n") {
1383: Log("CRITICAL",
1384: "Request Close Connection ... exiting");
1385: CloseAllLondConnections();
1386: exit;
1387: }
1388: Debug(8, "Complete transaction received: ".$data);
1389: if($LogTransactions) {
1390: Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
1391: }
1392: my $Transaction = LondTransaction->new($data);
1393: $Transaction->SetClient($socket);
1394: QueueTransaction($Transaction);
1395: $watcher->cancel(); # Done looking for input data.
1396: }
1397:
1398: }
1399:
1400: #
1401: # Accept a connection request for a client (lonc child) and
1402: # start up an event watcher to keep an eye on input from that
1403: # Event. This can be called both from NewClient and from
1404: # ChildProcess if we are started in DieWhenIdle mode.
1405: # Parameters:
1406: # $socket - The listener socket.
1407: # Returns:
1408: # NONE
1409: # Side Effects:
1410: # An event is made to watch the accepted connection.
1411: # Active clients hash is updated to reflect the new connection.
1412: # The client connection count is incremented.
1413: #
1414: sub accept_client {
1415: my ($socket) = @_;
1416:
1417: Debug(8, "Entering accept for lonc UNIX socket\n");
1418: my $connection = $socket->accept(); # Accept the client connection.
1419: Debug(8,"Connection request accepted from "
1420: .GetPeername($connection, AF_UNIX));
1421:
1422:
1423: my $description = sprintf("Connection to lonc client %d",
1424: $ClientConnection);
1425: Debug(9, "Creating event named: ".$description);
1426: Event->io(cb => \&ClientRequest,
1427: poll => 'r',
1428: desc => $description,
1429: data => "",
1430: fd => $connection);
1431: $ActiveClients{$connection} = $ClientConnection;
1432: $ClientConnection++;
1433: }
1434:
1435: =pod
1436:
1437: =head2 NewClient
1438:
1439: Callback that is called when a connection is received on the unix
1440: socket for a new client of lonc. The callback is parameterized by the
1441: event.. which is a-priori assumed to be an io event, and therefore has
1442: an fd member that is the Listener socket. We Accept the connection
1443: and register a new event on the readability of that socket:
1444:
1445: =cut
1446:
1447: sub NewClient {
1448: Debug(6, "NewClient");
1449: my $event = shift; # Get the event parameters.
1450: my $watcher = $event->w;
1451: my $socket = $watcher->fd; # Get the event' socket.
1452:
1453: &accept_client($socket);
1454: }
1455:
1456: =pod
1457:
1458: =head2 GetLoncSocketPath
1459:
1460: Returns the name of the UNIX socket on which to listen for client
1461: connections.
1462:
1463: =head2 Parameters:
1464:
1465: host (optional) - Name of the host socket to return.. defaults to
1466: the return from GetServerHost().
1467:
1468: =cut
1469:
1470: sub GetLoncSocketPath {
1471:
1472: my $host = GetServerHost(); # Default host.
1473: if (@_) {
1474: ($host) = @_; # Override if supplied.
1475: }
1476: return $UnixSocketDir."/".$host;
1477: }
1478:
1479: =pod
1480:
1481: =head2 GetServerHost
1482:
1483: Returns the host whose lond we talk with.
1484:
1485: =cut
1486:
1487: sub GetServerHost {
1488: return $RemoteHost; # Setup by the fork.
1489: }
1490:
1491: =pod
1492:
1493: =head2 GetServerPort
1494:
1495: Returns the lond port number.
1496:
1497: =cut
1498:
1499: sub GetServerPort {
1500: return $perlvar{londPort};
1501: }
1502:
1503: =pod
1504:
1505: =head2 SetupLoncListener
1506:
1507: Setup a lonc listener event. The event is called when the socket
1508: becomes readable.. that corresponds to the receipt of a new
1509: connection. The event handler established will accept the connection
1510: (creating a communcations channel), that int turn will establish
1511: another event handler to subess requests.
1512:
1513: =head2 Parameters:
1514:
1515: host (optional) Name of the host to set up a unix socket to.
1516:
1517: =cut
1518:
1519: sub SetupLoncListener {
1520:
1521: my $host = GetServerHost(); # Default host.
1522: if (@_) {
1523: ($host) = @_ # Override host with parameter.
1524: }
1525:
1526: my $socket;
1527: my $SocketName = GetLoncSocketPath($host);
1528: unlink($SocketName);
1529: unless ($socket =IO::Socket::UNIX->new(Local => $SocketName,
1530: Listen => 250,
1531: Type => SOCK_STREAM)) {
1532: if($I_am_child) {
1533: &child_exit(-1, "Failed to create a lonc listener socket");
1534: } else {
1535: die "Failed to create a lonc listner socket";
1536: }
1537: }
1538: return $socket;
1539: }
1540:
1541: #
1542: # Toggle transaction logging.
1543: # Implicit inputs:
1544: # LogTransactions
1545: # Implicit Outputs:
1546: # LogTransactions
1547: sub ToggleTransactionLogging {
1548: print STDERR "Toggle transaction logging...\n";
1549: if(!$LogTransactions) {
1550: $LogTransactions = 1;
1551: } else {
1552: $LogTransactions = 0;
1553: }
1554:
1555:
1556: Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
1557: }
1558:
1559: =pod
1560:
1561: =head2 ChildStatus
1562:
1563: Child USR1 signal handler to report the most recent status
1564: into the status file.
1565:
1566: We also use this to reset the retries count in order to allow the
1567: client to retry connections with a previously dead server.
1568:
1569: =cut
1570:
1571: sub ChildStatus {
1572: my $event = shift;
1573: my $watcher = $event->w;
1574:
1575: Debug(2, "Reporting child status because : ".$watcher->data);
1576: my $docdir = $perlvar{'lonDocRoot'};
1577:
1578: open(LOG,">>$docdir/lon-status/loncstatus.txt");
1579: flock(LOG,LOCK_EX);
1580: print LOG $$."\t".$RemoteHost."\t".$Status."\t".
1581: $RecentLogEntry."\n";
1582: #
1583: # Write out information about each of the connections:
1584: #
1585: if ($DebugLevel > 2) {
1586: print LOG "Active connection statuses: \n";
1587: my $i = 1;
1588: print STDERR "================================= Socket Status Dump:\n";
1589: foreach my $item (keys %ActiveConnections) {
1590: my $Socket = $ActiveConnections{$item}->data;
1591: my $state = $Socket->GetState();
1592: print LOG "Connection $i State: $state\n";
1593: print STDERR "---------------------- Connection $i \n";
1594: $Socket->Dump(-1); # Ensure it gets dumped..
1595: $i++;
1596: }
1597: }
1598: flock(LOG,LOCK_UN);
1599: close(LOG);
1600: $ConnectionRetriesLeft = $ConnectionRetries;
1601: UpdateStatus();
1602: }
1603:
1604: =pod
1605:
1606: =head2 SignalledToDeath
1607:
1608: Called in response to a signal that causes a chid process to die.
1609:
1610: =cut
1611:
1612:
1613: sub SignalledToDeath {
1614: my $event = shift;
1615: my $watcher= $event->w;
1616:
1617: Debug(2,"Signalled to death! via ".$watcher->data);
1618: my ($signal) = $watcher->data;
1619: chomp($signal);
1620: Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost "
1621: ."died through "."\"$signal\"");
1622: #LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
1623: # ."\"$signal\"");
1624: exit 0;
1625:
1626: }
1627:
1628: =pod
1629:
1630: =head2 ToggleDebug
1631:
1632: This sub toggles trace debugging on and off.
1633:
1634: =cut
1635:
1636: sub ToggleDebug {
1637: my $Current = $DebugLevel;
1638: $DebugLevel = $NextDebugLevel;
1639: $NextDebugLevel = $Current;
1640:
1641: Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
1642:
1643: }
1644:
1645: =pod
1646:
1647: =head2 ChildProcess
1648:
1649: This sub implements a child process for a single lonc daemon.
1650: Optional parameter:
1651: $socket - if provided, this is a socket already open for listen
1652: on the client socket. Otherwise, a new listen is set up.
1653:
1654: =cut
1655:
1656: sub ChildProcess {
1657: # If we are in DieWhenIdle mode, we've inherited all the
1658: # events of our parent and those have to be cancelled or else
1659: # all holy bloody chaos will result.. trust me, I already made
1660: # >that< mistake.
1661:
1662: my $host = GetServerHost();
1663: foreach my $listener (keys %parent_dispatchers) {
1664: my $watcher = $parent_dispatchers{$listener};
1665: my $s = $watcher->fd;
1666: if ($listener ne $host) { # Close everyone but me.
1667: Debug(5, "Closing listen socket for $listener");
1668: $s->close();
1669: }
1670: Debug(5, "Killing watcher for $listener");
1671:
1672: $watcher->cancel();
1673: delete($parent_dispatchers{$listener});
1674:
1675: }
1676:
1677: # kill off the parent's signal handlers too!
1678: #
1679:
1680: for my $handler (keys %parent_handlers) {
1681: my $watcher = $parent_handlers{$handler};
1682: $watcher->cancel();
1683: delete($parent_handlers{$handler});
1684: }
1685:
1686: $I_am_child = 1; # Seems like in spite of it all I may still getting
1687: # parent event dispatches.. flag I'm a child.
1688:
1689:
1690: #
1691: # Signals must be handled by the Event framework...
1692: #
1693:
1694: Event->signal(signal => "QUIT",
1695: cb => \&SignalledToDeath,
1696: data => "QUIT");
1697: Event->signal(signal => "HUP",
1698: cb => \&ChildStatus,
1699: data => "HUP");
1700: Event->signal(signal => "USR1",
1701: cb => \&ChildStatus,
1702: data => "USR1");
1703: Event->signal(signal => "USR2",
1704: cb => \&ToggleTransactionLogging);
1705: Event->signal(signal => "INT",
1706: cb => \&ToggleDebug,
1707: data => "INT");
1708:
1709: # Figure out if we got passed a socket or need to open one to listen for
1710: # client requests.
1711:
1712: my ($socket) = @_;
1713: if (!$socket) {
1714:
1715: $socket = SetupLoncListener();
1716: }
1717: # Establish an event to listen for client connection requests.
1718:
1719:
1720: Event->io(cb => \&NewClient,
1721: poll => 'r',
1722: desc => 'Lonc Listener Unix Socket',
1723: fd => $socket);
1724:
1725: $Event::Debuglevel = $DebugLevel;
1726:
1727: Debug(9, "Making initial lond connection for ".$RemoteHost);
1728:
1729: # Setup the initial server connection:
1730:
1731: # &MakeLondConnection(); // let first work request do it.
1732:
1733: # If We are in diwhenidle, need to accept the connection since the
1734: # event may not fire.
1735:
1736: if ($DieWhenIdle) {
1737: &accept_client($socket);
1738: }
1739:
1740: Debug(9,"Entering event loop");
1741: my $ret = Event::loop(); # Start the main event loop.
1742:
1743:
1744: &child_exit (-1,"Main event loop exited!!!");
1745: }
1746:
1747: # Create a new child for host passed in:
1748:
1749: sub CreateChild {
1750: my ($host, $socket) = @_;
1751:
1752: my $sigset = POSIX::SigSet->new(SIGINT);
1753: sigprocmask(SIG_BLOCK, $sigset);
1754: $RemoteHost = $host;
1755: Log("CRITICAL", "Forking server for ".$host);
1756: my $pid = fork;
1757: if($pid) { # Parent
1758: $RemoteHost = "Parent";
1759: $ChildHash{$pid} = $host;
1760: $HostToPid{$host}= $pid;
1761: sigprocmask(SIG_UNBLOCK, $sigset);
1762:
1763: } else { # child.
1764: ShowStatus("Connected to ".$RemoteHost);
1765: $SIG{INT} = 'DEFAULT';
1766: sigprocmask(SIG_UNBLOCK, $sigset);
1767: if(defined $socket) {
1768: &ChildProcess($socket);
1769: } else {
1770: ChildProcess; # Does not return.
1771: }
1772: }
1773: }
1774:
1775: # parent_client_connection:
1776: # Event handler that processes client connections for the parent process.
1777: # This sub is called when the parent is listening on a socket and
1778: # a connection request arrives. We must:
1779: # Start a child process to accept the connection request.
1780: # Kill our listen on the socket.
1781: # Parameter:
1782: # event - The event object that was created to monitor this socket.
1783: # event->w->fd is the socket.
1784: # Returns:
1785: # NONE
1786: #
1787: sub parent_client_connection {
1788: if ($I_am_child) {
1789: # Should not get here, but seem to anyway:
1790: &Debug(5," Child caught parent client connection event!!");
1791: my ($event) = @_;
1792: my $watcher = $event->w;
1793: $watcher->cancel(); # Try to kill it off again!!
1794: } else {
1795: &Debug(9, "parent_client_connection");
1796: my ($event) = @_;
1797: my $watcher = $event->w;
1798: my $socket = $watcher->fd;
1799:
1800: # Lookup the host associated with this socket:
1801:
1802: my $host = $listening_to{$socket};
1803:
1804: # Start the child:
1805:
1806:
1807:
1808: &Debug(9,"Creating child for $host (parent_client_connection)");
1809: &CreateChild($host, $socket);
1810:
1811: # Clean up the listen since now the child takes over until it exits.
1812:
1813: $watcher->cancel(); # Nolonger listening to this event
1814: delete($listening_to{$socket});
1815: delete($parent_dispatchers{$host});
1816: $socket->close();
1817: }
1818: }
1819:
1820: # parent_listen:
1821: # Opens a socket and starts a listen for the parent process on a client UNIX
1822: # domain socket.
1823: #
1824: # This involves:
1825: # Creating a socket for listen.
1826: # Removing any socket lock file
1827: # Adding an event handler for this socket becoming readable
1828: # To the parent's event dispatcher.
1829: # Parameters:
1830: # loncapa_host - LonCAPA cluster name of the host represented by the client
1831: # socket.
1832: # Returns:
1833: # NONE
1834: #
1835: sub parent_listen {
1836: my ($loncapa_host) = @_;
1837: Debug(5, "parent_listen: $loncapa_host");
1838:
1839: my $socket = &SetupLoncListener($loncapa_host);
1840: $listening_to{$socket} = $loncapa_host;
1841: if (!$socket) {
1842: die "Unable to create a listen socket for $loncapa_host";
1843: }
1844:
1845: my $lock_file = &GetLoncSocketPath($loncapa_host).".lock";
1846: unlink($lock_file); # No problem if it doesn't exist yet [startup e.g.]
1847:
1848: my $watcher = Event->io(cb => \&parent_client_connection,
1849: poll => 'r',
1850: desc => "Parent listener unix socket ($loncapa_host)",
1851: fd => $socket);
1852: $parent_dispatchers{$loncapa_host} = $watcher;
1853:
1854: }
1855:
1856:
1857: # listen_on_all_unix_sockets:
1858: # This sub initiates a listen on all unix domain lonc client sockets.
1859: # This will be called in the case where we are trimming idle processes.
1860: # When idle processes are trimmed, loncnew starts up with no children,
1861: # and only spawns off children when a connection request occurs on the
1862: # client unix socket. The spawned child continues to run until it has
1863: # been idle a while at which point it eventually exits and once more
1864: # the parent picks up the listen.
1865: #
1866: # Parameters:
1867: # NONE
1868: # Implicit Inputs:
1869: # The configuration file that has been read in by LondConnection.
1870: # Returns:
1871: # NONE
1872: #
1873: sub listen_on_all_unix_sockets {
1874: Debug(5, "listen_on_all_unix_sockets");
1875: my $host_iterator = &LondConnection::GetHostIterator();
1876: while (!$host_iterator->end()) {
1877: my $host_entry_ref = $host_iterator->get();
1878: my $host_name = $host_entry_ref->[0];
1879: Debug(9, "Listen for $host_name");
1880: &parent_listen($host_name);
1881: $host_iterator->next();
1882: }
1883: }
1884:
1885: # server_died is called whenever a child process exits.
1886: # Since this is dispatched via a signal, we must process all
1887: # dead children until there are no more left. The action
1888: # is to:
1889: # - Remove the child from the bookeeping hashes
1890: # - Re-establish a listen on the unix domain socket associated
1891: # with that host.
1892: # Parameters:
1893: # The event, but we don't actually care about it.
1894: sub server_died {
1895: &Debug(9, "server_died called...");
1896:
1897: while(1) { # Loop until waitpid nowait fails.
1898: my $pid = waitpid(-1, WNOHANG);
1899: if($pid <= 0) {
1900: return; # Nothing left to wait for.
1901: }
1902: # need the host to restart:
1903:
1904: my $host = $ChildHash{$pid};
1905: if($host) { # It's for real...
1906: &Debug(9, "Caught sigchild for $host");
1907: delete($ChildHash{$pid});
1908: delete($HostToPid{$host});
1909: &parent_listen($host);
1910:
1911: } else {
1912: &Debug(5, "Caught sigchild for pid not in hosts hash: $pid");
1913: }
1914: }
1915:
1916: }
1917:
1918: #
1919: # Parent process logic pass 1:
1920: # For each entry in the hosts table, we will
1921: # fork off an instance of ChildProcess to service the transactions
1922: # to that host. Each pid will be entered in a global hash
1923: # with the value of the key, the host.
1924: # The parent will then enter a loop to wait for process exits.
1925: # Each exit gets logged and the child gets restarted.
1926: #
1927:
1928: #
1929: # Fork and start in new session so hang-up isn't going to
1930: # happen without intent.
1931: #
1932:
1933:
1934:
1935:
1936:
1937:
1938: ShowStatus("Forming new session");
1939: my $childpid = fork;
1940: if ($childpid != 0) {
1941: sleep 4; # Give child a chacne to break to
1942: exit 0; # a new sesion.
1943: }
1944: #
1945: # Write my pid into the pid file so I can be located
1946: #
1947:
1948: ShowStatus("Parent writing pid file:");
1949: my $execdir = $perlvar{'lonDaemons'};
1950: open (PIDSAVE, ">$execdir/logs/lonc.pid");
1951: print PIDSAVE "$$\n";
1952: close(PIDSAVE);
1953:
1954:
1955:
1956: if (POSIX::setsid() < 0) {
1957: print "Could not create new session\n";
1958: exit -1;
1959: }
1960:
1961: ShowStatus("Forking node servers");
1962:
1963: Log("CRITICAL", "--------------- Starting children ---------------");
1964:
1965: LondConnection::ReadConfig; # Read standard config files.
1966: my $HostIterator = LondConnection::GetHostIterator;
1967:
1968: if ($DieWhenIdle) {
1969: $RemoteHost = "[parent]";
1970: &listen_on_all_unix_sockets();
1971: } else {
1972:
1973: while (! $HostIterator->end()) {
1974:
1975: my $hostentryref = $HostIterator->get();
1976: CreateChild($hostentryref->[0]);
1977: $HostHash{$hostentryref->[0]} = $hostentryref->[4];
1978: $HostIterator->next();
1979: }
1980: }
1981:
1982: $RemoteHost = "Parent Server";
1983:
1984: # Maintain the population:
1985:
1986: ShowStatus("Parent keeping the flock");
1987:
1988:
1989: if ($DieWhenIdle) {
1990: # We need to setup a SIGChild event to handle the exit (natural or otherwise)
1991: # of the children.
1992:
1993: Event->signal(cb => \&server_died,
1994: desc => "Child exit handler",
1995: signal => "CHLD");
1996:
1997:
1998: # Set up all the other signals we set up. We'll vector them off to the
1999: # same subs as we would for DieWhenIdle false and, if necessary, conditionalize
2000: # the code there.
2001:
2002: $parent_handlers{INT} = Event->signal(cb => \&Terminate,
2003: desc => "Parent INT handler",
2004: signal => "INT");
2005: $parent_handlers{TERM} = Event->signal(cb => \&Terminate,
2006: desc => "Parent TERM handler",
2007: signal => "TERM");
2008: $parent_handlers{HUP} = Event->signal(cb => \&Restart,
2009: desc => "Parent HUP handler.",
2010: signal => "HUP");
2011: $parent_handlers{USR1} = Event->signal(cb => \&CheckKids,
2012: desc => "Parent USR1 handler",
2013: signal => "USR1");
2014: $parent_handlers{USR2} = Event->signal(cb => \&UpdateKids,
2015: desc => "Parent USR2 handler.",
2016: signal => "USR2");
2017:
2018: # Start procdesing events.
2019:
2020: $Event::DebugLevel = $DebugLevel;
2021: Debug(9, "Parent entering event loop");
2022: my $ret = Event::loop();
2023: die "Main Event loop exited: $ret";
2024:
2025:
2026: } else {
2027: #
2028: # Set up parent signals:
2029: #
2030:
2031: $SIG{INT} = \&Terminate;
2032: $SIG{TERM} = \&Terminate;
2033: $SIG{HUP} = \&Restart;
2034: $SIG{USR1} = \&CheckKids;
2035: $SIG{USR2} = \&UpdateKids; # LonManage update request.
2036:
2037: while(1) {
2038: my $deadchild = wait();
2039: if(exists $ChildHash{$deadchild}) { # need to restart.
2040: my $deadhost = $ChildHash{$deadchild};
2041: delete($HostToPid{$deadhost});
2042: delete($ChildHash{$deadchild});
2043: Log("WARNING","Lost child pid= ".$deadchild.
2044: "Connected to host ".$deadhost);
2045: Log("INFO", "Restarting child procesing ".$deadhost);
2046: CreateChild($deadhost);
2047: }
2048: }
2049: }
2050:
2051:
2052: =pod
2053:
2054: =head1 CheckKids
2055:
2056: Since kids do not die as easily in this implementation
2057: as the previous one, there is no need to restart the
2058: dead ones (all dead kids get restarted when they die!!)
2059: The only thing this function does is to pass USR1 to the
2060: kids so that they report their status.
2061:
2062: =cut
2063:
2064: sub CheckKids {
2065: Debug(2, "Checking status of children");
2066: my $docdir = $perlvar{'lonDocRoot'};
2067: my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
2068: my $now=time;
2069: my $local=localtime($now);
2070: print $fh "LONC status $local - parent $$ \n\n";
2071: foreach my $host (keys %parent_dispatchers) {
2072: print $fh "LONC Parent process listening for $host\n";
2073: }
2074: foreach my $pid (keys %ChildHash) {
2075: Debug(2, "Sending USR1 -> $pid");
2076: kill 'USR1' => $pid; # Tell Child to report status.
2077: }
2078:
2079: }
2080:
2081: =pod
2082:
2083: =head1 UpdateKids
2084:
2085: parent's SIGUSR2 handler. This handler:
2086:
2087: =item
2088:
2089: Rereads the hosts file.
2090:
2091: =item
2092:
2093: Kills off (via sigint) children for hosts that have disappeared.
2094:
2095: =item
2096:
2097: QUITs children for hosts that already exist (this just forces a status display
2098: and resets the connection retry count for that host.
2099:
2100: =item
2101:
2102: Starts new children for hosts that have been added to the hosts.tab file since
2103: the start of the master program and maintains them.
2104:
2105: =cut
2106:
2107: sub UpdateKids {
2108:
2109: Log("INFO", "Updating connections via SIGUSR2");
2110:
2111: # I'm not sure what I was thinking in the first implementation.
2112: # someone will have to work hard to convince me the effect is any
2113: # different than Restart, especially now that we don't start up
2114: # per host servers automatically, may as well just restart.
2115: # The down side is transactions that are in flight will get timed out
2116: # (lost unless they are critical).
2117:
2118: &Restart();
2119:
2120: }
2121:
2122:
2123: =pod
2124:
2125: =head1 Restart
2126:
2127: Signal handler for HUP... all children are killed and
2128: we self restart. This is an el-cheapo way to re read
2129: the config file.
2130:
2131: =cut
2132:
2133: sub Restart {
2134: &KillThemAll; # First kill all the children.
2135: Log("CRITICAL", "Restarting");
2136: my $execdir = $perlvar{'lonDaemons'};
2137: unlink("$execdir/logs/lonc.pid");
2138: exec("$executable");
2139: }
2140:
2141: =pod
2142:
2143: =head1 KillThemAll
2144:
2145: Signal handler that kills all children by sending them a
2146: SIGHUP. Responds to sigint and sigterm.
2147:
2148: =cut
2149:
2150: sub KillThemAll {
2151: Debug(2, "Kill them all!!");
2152: local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die.
2153: foreach my $pid (keys %ChildHash) {
2154: my $serving = $ChildHash{$pid};
2155: ShowStatus("Nicely Killing lonc for $serving pid = $pid");
2156: Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
2157: kill 'QUIT' => $pid;
2158: }
2159:
2160:
2161: }
2162:
2163:
2164: #
2165: # Kill all children via KILL. Just in case the
2166: # first shot didn't get them.
2167:
2168: sub really_kill_them_all_dammit
2169: {
2170: Debug(2, "Kill them all Dammit");
2171: local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
2172: foreach my $pid (keys %ChildHash) {
2173: my $serving = $ChildHash{$pid};
2174: &ShowStatus("Nastily killing lonc for $serving pid = $pid");
2175: Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
2176: kill 'KILL' => $pid;
2177: delete($ChildHash{$pid});
2178: my $execdir = $perlvar{'lonDaemons'};
2179: unlink("$execdir/logs/lonc.pid");
2180: }
2181: }
2182:
2183: =pod
2184:
2185: =head1 Terminate
2186:
2187: Terminate the system.
2188:
2189: =cut
2190:
2191: sub Terminate {
2192: &Log("CRITICAL", "Asked to kill children.. first be nice...");
2193: &KillThemAll;
2194: #
2195: # By now they really should all be dead.. but just in case
2196: # send them all SIGKILL's after a bit of waiting:
2197:
2198: sleep(4);
2199: &Log("CRITICAL", "Now kill children nasty");
2200: &really_kill_them_all_dammit;
2201: Log("CRITICAL","Master process exiting");
2202: exit 0;
2203:
2204: }
2205: =pod
2206:
2207: =head1 Theory
2208:
2209: The event class is used to build this as a single process with an
2210: event driven model. The following events are handled:
2211:
2212: =item UNIX Socket connection Received
2213:
2214: =item Request data arrives on UNIX data transfer socket.
2215:
2216: =item lond connection becomes writable.
2217:
2218: =item timer fires at 1 second intervals.
2219:
2220: All sockets are run in non-blocking mode. Timeouts managed by the timer
2221: handler prevents hung connections.
2222:
2223: Key data structures:
2224:
2225: =item RequestQueue
2226:
2227: A queue of requests received from UNIX sockets that are
2228: waiting for a chance to be forwarded on a lond connection socket.
2229:
2230: =item ActiveConnections
2231:
2232: A hash of lond connections that have transactions in process that are
2233: available to be timed out.
2234:
2235: =item ActiveTransactions
2236:
2237: A hash indexed by lond connections that contain the client reply
2238: socket for each connection that has an active transaction on it.
2239:
2240: =item IdleConnections
2241:
2242: A hash of lond connections that have no work to do. These connections
2243: can be closed if they are idle for a long enough time.
2244:
2245: =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