File:
[LON-CAPA] /
loncom /
loncnew
Revision
1.53:
download - view:
text,
annotated -
select for diffs
Mon Sep 20 09:34:31 2004 UTC (20 years ago) by
foxr
Branches:
MAIN
CVS tags:
HEAD
Fix the issue where loncnew would rapidly oscillate connections. The problem
was that once it started trimmning connections. they'd all get trimmed out
at the rate of one/sec, rather than at the rate of one /idle timeout
period... needed to reset the idle timeout each time a connection got
trimmed.
1: #!/usr/bin/perl
2: # The LearningOnline Network with CAPA
3: # lonc maintains the connections to remote computers
4: #
5: # $Id: loncnew,v 1.53 2004/09/20 09:34:31 foxr Exp $
6: #
7: # Copyright Michigan State University Board of Trustees
8: #
9: # This file is part of the LearningOnline Network with CAPA (LON-CAPA).
10: ## LON-CAPA is free software; you can redistribute it and/or modify
11: # it under the terms of the GNU General Public License as published by
12: # the Free Software Foundation; either version 2 of the License, or
13: # (at your option) any later version.
14: #
15: # LON-CAPA is distributed in the hope that it will be useful,
16: # but WITHOUT ANY WARRANTY; without even the implied warranty of
17: # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18: # GNU General Public License for more details.
19: #
20: # You should have received a copy of the GNU General Public License
21: # along with LON-CAPA; if not, write to the Free Software
22: # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23: #
24: # /home/httpd/html/adm/gpl.txt
25: #
26: # http://www.lon-capa.org/
27: #
28: #
29: # new lonc handles n request out bver m connections to londs.
30: # This module is based on the Event class.
31: # Development iterations:
32: # - Setup basic event loop. (done)
33: # - Add timer dispatch. (done)
34: # - Add ability to accept lonc UNIX domain sockets. (done)
35: # - Add ability to create/negotiate lond connections (done).
36: # - Add general logic for dispatching requests and timeouts. (done).
37: # - Add support for the lonc/lond requests. (done).
38: # - Add logging/status monitoring. (done)
39: # - Add Signal handling - HUP restarts. USR1 status report. (done)
40: # - Add Configuration file I/O (done).
41: # - Add management/status request interface. (done)
42: # - Add deferred request capability. (done)
43: # - Detect transmission timeouts. (done)
44: #
45:
46: use strict;
47: use lib "/home/httpd/lib/perl/";
48: use Event qw(:DEFAULT );
49: use POSIX qw(:signal_h);
50: use POSIX;
51: use IO::Socket;
52: use IO::Socket::INET;
53: use IO::Socket::UNIX;
54: use IO::File;
55: use IO::Handle;
56: use Socket;
57: use Crypt::IDEA;
58: use LONCAPA::Queue;
59: use LONCAPA::Stack;
60: use LONCAPA::LondConnection;
61: use LONCAPA::LondTransaction;
62: use LONCAPA::Configuration;
63: use LONCAPA::HashIterator;
64:
65:
66: # Read the httpd configuration file to get perl variables
67: # normally set in apache modules:
68:
69: my $perlvarref = LONCAPA::Configuration::read_conf('loncapa.conf');
70: my %perlvar = %{$perlvarref};
71:
72: #
73: # parent and shared variables.
74:
75: my %ChildHash; # by pid -> host.
76: my %HostToPid; # By host -> pid.
77: my %HostHash; # by loncapaname -> IP.
78:
79:
80: my $MaxConnectionCount = 10; # Will get from config later.
81: my $ClientConnection = 0; # Uniquifier for client events.
82:
83: my $DebugLevel = 0;
84: my $NextDebugLevel= 2; # So Sigint can toggle this.
85: my $IdleTimeout= 600; # Wait 10 minutes before pruning connections.
86:
87: my $LogTransactions = 0; # When True, all transactions/replies get logged.
88:
89: #
90: # The variables below are only used by the child processes.
91: #
92: my $RemoteHost; # Name of host child is talking to.
93: my $UnixSocketDir= $perlvar{'lonSockDir'};
94: my $IdleConnections = Stack->new(); # Set of idle connections
95: my %ActiveConnections; # Connections to the remote lond.
96: my %ActiveTransactions; # LondTransactions in flight.
97: my %ActiveClients; # Serial numbers of active clients by socket.
98: my $WorkQueue = Queue->new(); # Queue of pending transactions.
99: my $ConnectionCount = 0;
100: my $IdleSeconds = 0; # Number of seconds idle.
101: my $Status = ""; # Current status string.
102: my $RecentLogEntry = "";
103: my $ConnectionRetries=2; # Number of connection retries allowed.
104: my $ConnectionRetriesLeft=2; # Number of connection retries remaining.
105: my $LondVersion = "unknown"; # Version of lond we talk with.
106: my $KeyMode = ""; # e.g. ssl, local, insecure from last connect.
107:
108: #
109: # The hash below gives the HTML format for log messages
110: # given a severity.
111: #
112: my %LogFormats;
113:
114: $LogFormats{"CRITICAL"} = "<font color='red'>CRITICAL: %s</font>";
115: $LogFormats{"SUCCESS"} = "<font color='green'>SUCCESS: %s</font>";
116: $LogFormats{"INFO"} = "<font color='yellow'>INFO: %s</font>";
117: $LogFormats{"WARNING"} = "<font color='blue'>WARNING: %s</font>";
118: $LogFormats{"DEFAULT"} = " %s ";
119:
120:
121:
122: =pod
123:
124: =head2 LogPerm
125:
126: Makes an entry into the permanent log file.
127:
128: =cut
129: sub LogPerm {
130: my $message=shift;
131: my $execdir=$perlvar{'lonDaemons'};
132: my $now=time;
133: my $local=localtime($now);
134: my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
135: print $fh "$now:$message:$local\n";
136: }
137:
138: =pod
139:
140: =head2 Log
141:
142: Logs a message to the log file.
143: Parameters:
144:
145: =item severity
146:
147: One of CRITICAL, WARNING, INFO, SUCCESS used to select the
148: format string used to format the message. if the severity is
149: not a defined severity the Default format string is used.
150:
151: =item message
152:
153: The base message. In addtion to the format string, the message
154: will be appended to a string containing the name of our remote
155: host and the time will be formatted into the message.
156:
157: =cut
158:
159: sub Log {
160:
161: my ($severity, $message) = @_;
162:
163: if(!$LogFormats{$severity}) {
164: $severity = "DEFAULT";
165: }
166:
167: my $format = $LogFormats{$severity};
168:
169: # Put the window dressing in in front of the message format:
170:
171: my $now = time;
172: my $local = localtime($now);
173: my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
174: my $finalformat = $finalformat.$format."\n";
175:
176: # open the file and put the result.
177:
178: my $execdir = $perlvar{'lonDaemons'};
179: my $fh = IO::File->new(">>$execdir/logs/lonc.log");
180: my $msg = sprintf($finalformat, $message);
181: $RecentLogEntry = $msg;
182: print $fh $msg;
183:
184:
185: }
186:
187:
188: =pod
189:
190: =head2 GetPeerName
191:
192: Returns the name of the host that a socket object is connected to.
193:
194: =cut
195:
196: sub GetPeername {
197:
198:
199: my ($connection, $AdrFamily) = @_;
200:
201: my $peer = $connection->peername();
202: my $peerport;
203: my $peerip;
204: if($AdrFamily == AF_INET) {
205: ($peerport, $peerip) = sockaddr_in($peer);
206: my $peername = gethostbyaddr($peerip, $AdrFamily);
207: return $peername;
208: } elsif ($AdrFamily == AF_UNIX) {
209: my $peerfile;
210: ($peerfile) = sockaddr_un($peer);
211: return $peerfile;
212: }
213: }
214: =pod
215:
216: =head2 Debug
217:
218: Invoked to issue a debug message.
219:
220: =cut
221:
222: sub Debug {
223:
224: my ($level, $message) = @_;
225:
226: if ($level <= $DebugLevel) {
227: Log("INFO", "-Debug- $message host = $RemoteHost");
228: }
229: }
230:
231: sub SocketDump {
232:
233: my ($level, $socket) = @_;
234:
235: if($level <= $DebugLevel) {
236: $socket->Dump(-1); # Ensure it will get dumped.
237: }
238: }
239:
240: =pod
241:
242: =head2 ShowStatus
243:
244: Place some text as our pid status.
245: and as what we return in a SIGUSR1
246:
247: =cut
248: sub ShowStatus {
249: my $state = shift;
250: my $now = time;
251: my $local = localtime($now);
252: $Status = $local.": ".$state;
253: $0='lonc: '.$state.' '.$local;
254: }
255:
256: =pod
257:
258: =head 2 SocketTimeout
259:
260: Called when an action on the socket times out. The socket is
261: destroyed and any active transaction is failed.
262:
263:
264: =cut
265: sub SocketTimeout {
266: my $Socket = shift;
267: Log("WARNING", "A socket timeout was detected");
268: Debug(5, " SocketTimeout called: ");
269: $Socket->Dump(0);
270: if(exists($ActiveTransactions{$Socket})) {
271: FailTransaction($ActiveTransactions{$Socket});
272: }
273: KillSocket($Socket); # A transaction timeout also counts as
274: # a connection failure:
275: $ConnectionRetriesLeft--;
276: if($ConnectionRetriesLeft <= 0) {
277: Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
278: }
279:
280: }
281: #----------------------------- Timer management ------------------------
282:
283: =pod
284:
285: =head2 Tick
286:
287: Invoked each timer tick.
288:
289: =cut
290:
291:
292: sub Tick {
293: my ($Event) = @_;
294: my $clock_watcher = $Event->w;
295:
296: my $client;
297: if($ConnectionRetriesLeft > 0) {
298: ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount
299: ." Retries remaining: ".$ConnectionRetriesLeft
300: ." ($KeyMode)");
301: } else {
302: ShowStatus(GetServerHost()." >> DEAD <<");
303: }
304: # Is it time to prune connection count:
305:
306:
307: if($IdleConnections->Count() &&
308: ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
309: $IdleSeconds++;
310: if($IdleSeconds > $IdleTimeout) { # Prune a connection...
311: my $Socket = $IdleConnections->pop();
312: KillSocket($Socket);
313: IdleSeconds = 0; # Otherwise all connections get trimmed to fast.
314: }
315: } else {
316: $IdleSeconds = 0; # Reset idle count if not idle.
317: }
318: #
319: # For each inflight transaction, tick down its timeout counter.
320: #
321:
322: foreach my $item (keys %ActiveConnections) {
323: my $State = $ActiveConnections{$item}->data->GetState();
324: if ($State ne 'Idle') {
325: Debug(5,"Ticking Socket $State $item");
326: $ActiveConnections{$item}->data->Tick();
327: }
328: }
329: # Do we have work in the queue, but no connections to service them?
330: # If so, try to make some new connections to get things going again.
331: #
332:
333: my $Requests = $WorkQueue->Count();
334: if (($ConnectionCount == 0) && ($Requests > 0)) {
335: if ($ConnectionRetriesLeft > 0) {
336: my $Connections = ($Requests <= $MaxConnectionCount) ?
337: $Requests : $MaxConnectionCount;
338: Debug(5,"Work but no connections, start ".$Connections." of them");
339: my $successCount = 0;
340: for (my $i =0; $i < $Connections; $i++) {
341: $successCount += MakeLondConnection();
342: }
343: if($successCount == 0) { # All connections failed:
344: Debug(5,"Work in queue failed to make any connectiouns\n");
345: EmptyQueue(); # Fail pending transactions with con_lost.
346: CloseAllLondConnections(); # Should all be closed but....
347: }
348: } else {
349: ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
350: Debug(5,"Work in queue, but gave up on connections..flushing\n");
351: EmptyQueue(); # Connections can't be established.
352: CloseAllLondConnections(); # Should all already be closed but...
353: }
354:
355: }
356: if ($ConnectionCount == 0) {
357: $KeyMode = "";
358: $clock_watcher->cancel();
359: }
360: }
361:
362: =pod
363:
364: =head2 SetupTimer
365:
366: Sets up a 1 per sec recurring timer event. The event handler is used to:
367:
368: =item
369:
370: Trigger timeouts on communications along active sockets.
371:
372: =item
373:
374: Trigger disconnections of idle sockets.
375:
376: =cut
377:
378: sub SetupTimer {
379: Debug(6, "SetupTimer");
380: Event->timer(interval => 1, cb => \&Tick );
381: }
382:
383: =pod
384:
385: =head2 ServerToIdle
386:
387: This function is called when a connection to the server is
388: ready for more work.
389:
390: If there is work in the Work queue the top element is dequeued
391: and the connection will start to work on it. If the work queue is
392: empty, the connection is pushed on the idle connection stack where
393: it will either get another work unit, or alternatively, if it sits there
394: long enough, it will be shut down and released.
395:
396: =cut
397:
398: sub ServerToIdle {
399: my $Socket = shift; # Get the socket.
400: $KeyMode = $Socket->{AuthenticationMode};
401: delete($ActiveTransactions{$Socket}); # Server has no transaction
402:
403: &Debug(5, "Server to idle");
404:
405: # If there's work to do, start the transaction:
406:
407: my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
408: if ($reqdata ne undef) {
409: Debug(5, "Queue gave request data: ".$reqdata->getRequest());
410: &StartRequest($Socket, $reqdata);
411:
412: } else {
413:
414: # There's no work waiting, so push the server to idle list.
415: &Debug(5, "No new work requests, server connection going idle");
416: $IdleConnections->push($Socket);
417: }
418: }
419:
420: =pod
421:
422: =head2 ClientWritable
423:
424: Event callback for when a client socket is writable.
425:
426: This callback is established when a transaction reponse is
427: avaiable from lond. The response is forwarded to the unix socket
428: as it becomes writable in this sub.
429:
430: Parameters:
431:
432: =item Event
433:
434: The event that has been triggered. Event->w->data is
435: the data and Event->w->fd is the socket to write.
436:
437: =cut
438:
439: sub ClientWritable {
440: my $Event = shift;
441: my $Watcher = $Event->w;
442: my $Data = $Watcher->data;
443: my $Socket = $Watcher->fd;
444:
445: # Try to send the data:
446:
447: &Debug(6, "ClientWritable writing".$Data);
448: &Debug(9, "Socket is: ".$Socket);
449:
450: if($Socket->connected) {
451: my $result = $Socket->send($Data, 0);
452:
453: # $result undefined: the write failed.
454: # otherwise $result is the number of bytes written.
455: # Remove that preceding string from the data.
456: # If the resulting data is empty, destroy the watcher
457: # and set up a read event handler to accept the next
458: # request.
459:
460: &Debug(9,"Send result is ".$result." Defined: ".defined($result));
461: if($result ne undef) {
462: &Debug(9, "send result was defined");
463: if($result == length($Data)) { # Entire string sent.
464: &Debug(9, "ClientWritable data all written");
465: $Watcher->cancel();
466: #
467: # Set up to read next request from socket:
468:
469: my $descr = sprintf("Connection to lonc client %d",
470: $ActiveClients{$Socket});
471: Event->io(cb => \&ClientRequest,
472: poll => 'r',
473: desc => $descr,
474: data => "",
475: fd => $Socket);
476:
477: } else { # Partial string sent.
478: $Watcher->data(substr($Data, $result));
479: if($result == 0) { # client hung up on us!!
480: # Log("INFO", "lonc pipe client hung up on us!");
481: $Watcher->cancel;
482: $Socket->shutdown(2);
483: $Socket->close();
484: }
485: }
486:
487: } else { # Error of some sort...
488:
489: # Some errnos are possible:
490: my $errno = $!;
491: if($errno == POSIX::EWOULDBLOCK ||
492: $errno == POSIX::EAGAIN ||
493: $errno == POSIX::EINTR) {
494: # No action taken?
495: } else { # Unanticipated errno.
496: &Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
497: $Watcher->cancel; # Stop the watcher.
498: $Socket->shutdown(2); # Kill connection
499: $Socket->close(); # Close the socket.
500: }
501:
502: }
503: } else {
504: $Watcher->cancel(); # A delayed request...just cancel.
505: }
506: }
507:
508: =pod
509:
510: =head2 CompleteTransaction
511:
512: Called when the reply data has been received for a lond
513: transaction. The reply data must now be sent to the
514: ultimate client on the other end of the Unix socket. This is
515: done by setting up a writable event for the socket with the
516: data the reply data.
517:
518: Parameters:
519:
520: =item Socket
521:
522: Socket on which the lond transaction occured. This is a
523: LondConnection. The data received is in the TransactionReply member.
524:
525: =item Transaction
526:
527: The transaction that is being completed.
528:
529: =cut
530:
531: sub CompleteTransaction {
532: &Debug(5,"Complete transaction");
533:
534: my ($Socket, $Transaction) = @_;
535:
536: if (!$Transaction->isDeferred()) { # Normal transaction
537: my $data = $Socket->GetReply(); # Data to send.
538: if($LogTransactions) {
539: Log("SUCCESS", "Reply from lond: '$data'");
540: }
541: StartClientReply($Transaction, $data);
542: } else { # Delete deferred transaction file.
543: Log("SUCCESS", "A delayed transaction was completed");
544: LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest());
545: unlink $Transaction->getFile();
546: }
547: }
548:
549: =pod
550:
551: =head1 StartClientReply
552:
553: Initiates a reply to a client where the reply data is a parameter.
554:
555: =head2 parameters:
556:
557: =item Transaction
558:
559: The transaction for which we are responding to the client.
560:
561: =item data
562:
563: The data to send to apached client.
564:
565: =cut
566:
567: sub StartClientReply {
568:
569: my ($Transaction, $data) = @_;
570:
571: my $Client = $Transaction->getClient();
572:
573: &Debug(8," Reply was: ".$data);
574: my $Serial = $ActiveClients{$Client};
575: my $desc = sprintf("Connection to lonc client %d",
576: $Serial);
577: Event->io(fd => $Client,
578: poll => "w",
579: desc => $desc,
580: cb => \&ClientWritable,
581: data => $data);
582: }
583:
584: =pod
585:
586: =head2 FailTransaction
587:
588: Finishes a transaction with failure because the associated lond socket
589: disconnected. There are two possibilities:
590: - The transaction is deferred: in which case we just quietly
591: delete the transaction since there is no client connection.
592: - The transaction is 'live' in which case we initiate the sending
593: of "con_lost" to the client.
594:
595: Deleting the transaction means killing it from the %ActiveTransactions hash.
596:
597: Parameters:
598:
599: =item client
600:
601: The LondTransaction we are failing.
602:
603:
604: =cut
605:
606: sub FailTransaction {
607: my $transaction = shift;
608:
609: # If the socket is dead, that's already logged.
610:
611: if ($ConnectionRetriesLeft > 0) {
612: Log("WARNING", "Failing transaction "
613: .$transaction->getRequest());
614: }
615: Debug(1, "Failing transaction: ".$transaction->getRequest());
616: if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
617: my $client = $transaction->getClient();
618: Debug(1," Replying con_lost to ".$transaction->getRequest());
619: StartClientReply($transaction, "con_lost\n");
620: }
621:
622: }
623:
624: =pod
625: =head1 EmptyQueue
626:
627: Fails all items in the work queue with con_lost.
628: Note that each item in the work queue is a transaction.
629:
630: =cut
631: sub EmptyQueue {
632: $ConnectionRetriesLeft--; # Counts as connection failure too.
633: while($WorkQueue->Count()) {
634: my $request = $WorkQueue->dequeue(); # This is a transaction
635: FailTransaction($request);
636: }
637: }
638:
639: =pod
640:
641: =head2 CloseAllLondConnections
642:
643: Close all connections open on lond prior to exit e.g.
644:
645: =cut
646: sub CloseAllLondConnections {
647: foreach my $Socket (keys %ActiveConnections) {
648: if(exists($ActiveTransactions{$Socket})) {
649: FailTransaction($ActiveTransactions{$Socket});
650: }
651: KillSocket($Socket);
652: }
653: }
654: =cut
655:
656: =pod
657:
658: =head2 KillSocket
659:
660: Destroys a socket. This function can be called either when a socket
661: has died of 'natural' causes or because a socket needs to be pruned due to
662: idleness. If the socket has died naturally, if there are no longer any
663: live connections a new connection is created (in case there are transactions
664: in the queue). If the socket has been pruned, it is never re-created.
665:
666: Parameters:
667:
668: =item Socket
669:
670: The socket to kill off.
671:
672: =item Restart
673:
674: nonzero if we are allowed to create a new connection.
675:
676:
677: =cut
678: sub KillSocket {
679: my $Socket = shift;
680:
681: Log("WARNING", "Shutting down a socket");
682: $Socket->Shutdown();
683:
684: # If the socket came from the active connection set,
685: # delete its transaction... note that FailTransaction should
686: # already have been called!!!
687: # otherwise it came from the idle set.
688: #
689:
690: if(exists($ActiveTransactions{$Socket})) {
691: delete ($ActiveTransactions{$Socket});
692: }
693: if(exists($ActiveConnections{$Socket})) {
694: delete($ActiveConnections{$Socket});
695: $ConnectionCount--;
696: if ($ConnectionCount < 0) { $ConnectionCount = 0; }
697: }
698: # If the connection count has gone to zero and there is work in the
699: # work queue, the work all gets failed with con_lost.
700: #
701: if($ConnectionCount == 0) {
702: EmptyQueue();
703: CloseAllLondConnections; # Should all already be closed but...
704: }
705: }
706:
707: =pod
708:
709: =head2 LondReadable
710:
711: This function is called whenever a lond connection
712: is readable. The action is state dependent:
713:
714: =head3 State=Initialized
715:
716: We''re waiting for the challenge, this is a no-op until the
717: state changes.
718:
719: =head3 State=Challenged
720:
721: The challenge has arrived we need to transition to Writable.
722: The connection must echo the challenge back.
723:
724: =head3 State=ChallengeReplied
725:
726: The challenge has been replied to. The we are receiveing the
727: 'ok' from the partner.
728:
729: =head3 State=ReadingVersionString
730:
731: We have requested the lond version and are reading the
732: version back. Upon completion, we'll store the version away
733: for future use(?).
734:
735: =head3 State=HostSet
736:
737: We have selected the domain name of our peer (multhomed hosts)
738: and are getting the reply (presumably ok) back.
739:
740: =head3 State=RequestingKey
741:
742: The ok has been received and we need to send the request for
743: an encryption key. Transition to writable for that.
744:
745: =head3 State=ReceivingKey
746:
747: The the key has been requested, now we are reading the new key.
748:
749: =head3 State=Idle
750:
751: The encryption key has been negotiated or we have finished
752: reading data from the a transaction. If the callback data has
753: a client as well as the socket iformation, then we are
754: doing a transaction and the data received is relayed to the client
755: before the socket is put on the idle list.
756:
757: =head3 State=SendingRequest
758:
759: I do not think this state can be received here, but if it is,
760: the appropriate thing to do is to transition to writable, and send
761: the request.
762:
763: =head3 State=ReceivingReply
764:
765: We finished sending the request to the server and now transition
766: to readable to receive the reply.
767:
768: The parameter to this function are:
769:
770: The event. Implicit in this is the watcher and its data. The data
771: contains at least the lond connection object and, if a
772: transaction is in progress, the socket attached to the local client.
773:
774: =cut
775:
776: sub LondReadable {
777:
778: my $Event = shift;
779: my $Watcher = $Event->w;
780: my $Socket = $Watcher->data;
781: my $client = undef;
782:
783: &Debug(6,"LondReadable called state = ".$Socket->GetState());
784:
785:
786: my $State = $Socket->GetState(); # All action depends on the state.
787:
788: SocketDump(6, $Socket);
789: my $status = $Socket->Readable();
790:
791: &Debug(2, "Socket->Readable returned: $status");
792:
793: if($status != 0) {
794: # bad return from socket read. Currently this means that
795: # The socket has become disconnected. We fail the transaction.
796:
797: Log("WARNING",
798: "Lond connection lost.");
799: if(exists($ActiveTransactions{$Socket})) {
800: FailTransaction($ActiveTransactions{$Socket});
801: }
802: $Watcher->cancel();
803: KillSocket($Socket);
804: $ConnectionRetriesLeft--; # Counts as connection failure
805: return;
806: }
807: SocketDump(6,$Socket);
808:
809: $State = $Socket->GetState(); # Update in case of transition.
810: &Debug(6, "After read, state is ".$State);
811:
812: if($State eq "Initialized") {
813:
814:
815: } elsif ($State eq "ChallengeReceived") {
816: # The challenge must be echoed back; The state machine
817: # in the connection takes care of setting that up. Just
818: # need to transition to writable:
819:
820: $Watcher->cb(\&LondWritable);
821: $Watcher->poll("w");
822:
823: } elsif ($State eq "ChallengeReplied") {
824:
825: } elsif ($State eq "RequestingVersion") {
826: # Need to ask for the version... that is writiability:
827:
828: $Watcher->cb(\&LondWritable);
829: $Watcher->poll("w");
830:
831: } elsif ($State eq "ReadingVersionString") {
832: # Read the rest of the version string...
833: } elsif ($State eq "SetHost") {
834: # Need to request the actual domain get set...
835:
836: $Watcher->cb(\&LondWritable);
837: $Watcher->poll("w");
838: } elsif ($State eq "HostSet") {
839: # Reading the 'ok' from the peer.
840:
841: } elsif ($State eq "RequestingKey") {
842: # The ok was received. Now we need to request the key
843: # That requires us to be writable:
844:
845: $Watcher->cb(\&LondWritable);
846: $Watcher->poll("w");
847:
848: } elsif ($State eq "ReceivingKey") {
849:
850: } elsif ($State eq "Idle") {
851:
852: # This is as good a spot as any to get the peer version
853: # string:
854:
855: if($LondVersion eq "unknown") {
856: $LondVersion = $Socket->PeerVersion();
857: Log("INFO", "Connected to lond version: $LondVersion");
858: }
859: # If necessary, complete a transaction and then go into the
860: # idle queue.
861: # Note that a trasition to idle indicates a live lond
862: # on the other end so reset the connection retries.
863: #
864: $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
865: $Watcher->cancel();
866: if(exists($ActiveTransactions{$Socket})) {
867: Debug(5,"Completing transaction!!");
868: CompleteTransaction($Socket,
869: $ActiveTransactions{$Socket});
870: } else {
871: Log("SUCCESS", "Connection ".$ConnectionCount." to "
872: .$RemoteHost." now ready for action");
873: }
874: ServerToIdle($Socket); # Next work unit or idle.
875:
876: } elsif ($State eq "SendingRequest") {
877: # We need to be writable for this and probably don't belong
878: # here inthe first place.
879:
880: Deubg(6, "SendingRequest state encountered in readable");
881: $Watcher->poll("w");
882: $Watcher->cb(\&LondWritable);
883:
884: } elsif ($State eq "ReceivingReply") {
885:
886:
887: } else {
888: # Invalid state.
889: Debug(4, "Invalid state in LondReadable");
890: }
891: }
892:
893: =pod
894:
895: =head2 LondWritable
896:
897: This function is called whenever a lond connection
898: becomes writable while there is a writeable monitoring
899: event. The action taken is very state dependent:
900:
901: =head3 State = Connected
902:
903: The connection is in the process of sending the 'init' hailing to the
904: lond on the remote end. The connection object''s Writable member is
905: called. On error, ConnectionError is called to destroy the connection
906: and remove it from the ActiveConnections hash
907:
908: =head3 Initialized
909:
910: 'init' has been sent, writability monitoring is removed and
911: readability monitoring is started with LondReadable as the callback.
912:
913: =head3 ChallengeReceived
914:
915: The connection has received the who are you challenge from the remote
916: system, and is in the process of sending the challenge
917: response. Writable is called.
918:
919: =head3 ChallengeReplied
920:
921: The connection has replied to the initial challenge The we switch to
922: monitoring readability looking for the server to reply with 'ok'.
923:
924: =head3 RequestingKey
925:
926: The connection is in the process of requesting its encryption key.
927: Writable is called.
928:
929: =head3 ReceivingKey
930:
931: The connection has sent the request for a key. Switch to readability
932: monitoring to accept the key
933:
934: =head3 SendingRequest
935:
936: The connection is in the process of sending a request to the server.
937: This request is part of a client transaction. All the states until
938: now represent the client setup protocol. Writable is called.
939:
940: =head3 ReceivingReply
941:
942: The connection has sent a request. Now it must receive a reply.
943: Readability monitoring is requested.
944:
945: This function is an event handler and therefore receives as
946: a parameter the event that has fired. The data for the watcher
947: of this event is a reference to a list of one or two elements,
948: depending on state. The first (and possibly only) element is the
949: socket. The second (present only if a request is in progress)
950: is the socket on which to return a reply to the caller.
951:
952: =cut
953:
954: sub LondWritable {
955: my $Event = shift;
956: my $Watcher = $Event->w;
957: my $Socket = $Watcher->data;
958: my $State = $Socket->GetState();
959:
960: Debug(6,"LondWritable State = ".$State."\n");
961:
962:
963: # Figure out what to do depending on the state of the socket:
964:
965:
966:
967:
968: SocketDump(6,$Socket);
969:
970: # If the socket is writable, we must always write.
971: # Only by writing will we undergo state transitions.
972: # Old logic wrote in state specific code below, however
973: # That forces us at least through another invocation of
974: # this function after writability is possible again.
975: # This logic also factors out common code for handling
976: # write failures... in all cases, write failures
977: # Kill the socket.
978: # This logic makes the branches of the >big< if below
979: # so that the writing states are actually NO-OPs.
980:
981: if ($Socket->Writable() != 0) {
982: # The write resulted in an error.
983: # We'll treat this as if the socket got disconnected:
984: Log("WARNING", "Connection to ".$RemoteHost.
985: " has been disconnected");
986: if(exists($ActiveTransactions{$Socket})) {
987: FailTransaction($ActiveTransactions{$Socket});
988: }
989: $Watcher->cancel();
990: KillSocket($Socket);
991: return;
992: }
993:
994:
995:
996: if ($State eq "Connected") {
997:
998: # "init" is being sent...
999:
1000: } elsif ($State eq "Initialized") {
1001:
1002: # Now that init was sent, we switch
1003: # to watching for readability:
1004:
1005: $Watcher->cb(\&LondReadable);
1006: $Watcher->poll("r");
1007:
1008: } elsif ($State eq "ChallengeReceived") {
1009: # We received the challenge, now we
1010: # are echoing it back. This is a no-op,
1011: # we're waiting for the state to change
1012:
1013: } elsif ($State eq "ChallengeReplied") {
1014: # The echo was sent back, so we switch
1015: # to watching readability.
1016:
1017: $Watcher->cb(\&LondReadable);
1018: $Watcher->poll("r");
1019: } elsif ($State eq "RequestingVersion") {
1020: # Sending the peer a version request...
1021:
1022: } elsif ($State eq "ReadingVersionString") {
1023: # Transition to read since we have sent the
1024: # version command and now just need to read the
1025: # version string from the peer:
1026:
1027: $Watcher->cb(\&LondReadable);
1028: $Watcher->poll("r");
1029:
1030: } elsif ($State eq "SetHost") {
1031: # Setting the remote domain...
1032:
1033: } elsif ($State eq "HostSet") {
1034: # Back to readable to get the ok.
1035:
1036: $Watcher->cb(\&LondReadable);
1037: $Watcher->poll("r");
1038:
1039:
1040: } elsif ($State eq "RequestingKey") {
1041: # At this time we're requesting the key.
1042: # again, this is essentially a no-op.
1043:
1044: } elsif ($State eq "ReceivingKey") {
1045: # Now we need to wait for the key
1046: # to come back from the peer:
1047:
1048: $Watcher->cb(\&LondReadable);
1049: $Watcher->poll("r");
1050:
1051: } elsif ($State eq "SendingRequest") {
1052:
1053: # At this time we are sending a request to the
1054: # peer... write the next chunk:
1055:
1056:
1057: } elsif ($State eq "ReceivingReply") {
1058: # The send has completed. Wait for the
1059: # data to come in for a reply.
1060: Debug(8,"Writable sent request/receiving reply");
1061: $Watcher->cb(\&LondReadable);
1062: $Watcher->poll("r");
1063:
1064: } else {
1065: # Control only passes here on an error:
1066: # the socket state does not match any
1067: # of the known states... so an error
1068: # must be logged.
1069:
1070: &Debug(4, "Invalid socket state ".$State."\n");
1071: }
1072:
1073: }
1074: =pod
1075:
1076: =cut
1077: sub QueueDelayed {
1078: Debug(3,"QueueDelayed called");
1079:
1080: my $path = "$perlvar{'lonSockDir'}/delayed";
1081:
1082: Debug(4, "Delayed path: ".$path);
1083: opendir(DIRHANDLE, $path);
1084:
1085: my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE;
1086: closedir(DIRHANDLE);
1087: my $dfname;
1088: my $reqfile;
1089: foreach $dfname (sort @alldelayed) {
1090: $reqfile = "$path/$dfname";
1091: Debug(4, "queueing ".$reqfile);
1092: my $Handle = IO::File->new($reqfile);
1093: my $cmd = <$Handle>;
1094: chomp $cmd; # There may or may not be a newline...
1095: $cmd = $cmd."\n"; # now for sure there's exactly one newline.
1096: my $Transaction = LondTransaction->new($cmd);
1097: $Transaction->SetDeferred($reqfile);
1098: QueueTransaction($Transaction);
1099: }
1100:
1101: }
1102:
1103: =pod
1104:
1105: =head2 MakeLondConnection
1106:
1107: Create a new lond connection object, and start it towards its initial
1108: idleness. Once idle, it becomes elligible to receive transactions
1109: from the work queue. If the work queue is not empty when the
1110: connection is completed and becomes idle, it will dequeue an entry and
1111: start off on it.
1112:
1113: =cut
1114:
1115: sub MakeLondConnection {
1116: Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
1117: .GetServerPort());
1118:
1119: my $Connection = LondConnection->new(&GetServerHost(),
1120: &GetServerPort());
1121:
1122: if($Connection eq undef) { # Needs to be more robust later.
1123: Log("CRITICAL","Failed to make a connection with lond.");
1124: $ConnectionRetriesLeft--;
1125: return 0; # Failure.
1126: } else {
1127:
1128: # The connection needs to have writability
1129: # monitored in order to send the init sequence
1130: # that starts the whole authentication/key
1131: # exchange underway.
1132: #
1133: my $Socket = $Connection->GetSocket();
1134: if($Socket eq undef) {
1135: die "did not get a socket from the connection";
1136: } else {
1137: &Debug(9,"MakeLondConnection got socket: ".$Socket);
1138: }
1139:
1140: $Connection->SetTimeoutCallback(\&SocketTimeout);
1141:
1142: my $event = Event->io(fd => $Socket,
1143: poll => 'w',
1144: cb => \&LondWritable,
1145: data => $Connection,
1146: desc => 'Connection to lond server');
1147: $ActiveConnections{$Connection} = $event;
1148: if ($ConnectionCount == 0) {
1149: &SetupTimer; # Need to handle timeouts with connections...
1150: }
1151: $ConnectionCount++;
1152: Debug(4, "Connection count = ".$ConnectionCount);
1153: if($ConnectionCount == 1) { # First Connection:
1154: QueueDelayed;
1155: }
1156: Log("SUCESS", "Created connection ".$ConnectionCount
1157: ." to host ".GetServerHost());
1158: return 1; # Return success.
1159: }
1160:
1161: }
1162:
1163: =pod
1164:
1165: =head2 StartRequest
1166:
1167: Starts a lond request going on a specified lond connection.
1168: parameters are:
1169:
1170: =item $Lond
1171:
1172: Connection to the lond that will send the transaction and receive the
1173: reply.
1174:
1175: =item $Client
1176:
1177: Connection to the client that is making this request We got the
1178: request from this socket, and when the request has been relayed to
1179: lond and we get a reply back from lond it will get sent to this
1180: socket.
1181:
1182: =item $Request
1183:
1184: The text of the request to send.
1185:
1186: =cut
1187:
1188: sub StartRequest {
1189:
1190: my ($Lond, $Request) = @_;
1191:
1192: Debug(6, "StartRequest: ".$Request->getRequest());
1193:
1194: my $Socket = $Lond->GetSocket();
1195:
1196: $Request->Activate($Lond);
1197: $ActiveTransactions{$Lond} = $Request;
1198:
1199: $Lond->InitiateTransaction($Request->getRequest());
1200: my $event = Event->io(fd => $Socket,
1201: poll => "w",
1202: cb => \&LondWritable,
1203: data => $Lond,
1204: desc => "lond transaction connection");
1205: $ActiveConnections{$Lond} = $event;
1206: Debug(8," Start Request made watcher data with ".$event->data."\n");
1207: }
1208:
1209: =pod
1210:
1211: =head2 QueueTransaction
1212:
1213: If there is an idle lond connection, it is put to work doing this
1214: transaction. Otherwise, the transaction is placed in the work queue.
1215: If placed in the work queue and the maximum number of connections has
1216: not yet been created, a new connection will be started. Our goal is
1217: to eventually have a sufficient number of connections that the work
1218: queue will typically be empty. parameters are:
1219:
1220: =item Socket
1221:
1222: open on the lonc client.
1223:
1224: =item Request
1225:
1226: data to send to the lond.
1227:
1228: =cut
1229:
1230: sub QueueTransaction {
1231:
1232: my $requestData = shift; # This is a LondTransaction.
1233: my $cmd = $requestData->getRequest();
1234:
1235: Debug(6,"QueueTransaction: ".$cmd);
1236:
1237: my $LondSocket = $IdleConnections->pop();
1238: if(!defined $LondSocket) { # Need to queue request.
1239: Debug(5,"Must queue...");
1240: $WorkQueue->enqueue($requestData);
1241: if($ConnectionCount < $MaxConnectionCount) {
1242: if($ConnectionRetriesLeft > 0) {
1243: Debug(5,"Starting additional lond connection");
1244: if(MakeLondConnection() == 0) {
1245: EmptyQueue(); # Fail transactions, can't make connection.
1246: CloseAllLondConnections; # Should all be closed but...
1247: }
1248: } else {
1249: ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1250: EmptyQueue(); # It's worse than that ... he's dead Jim.
1251: CloseAllLondConnections; # Should all be closed but..
1252: }
1253: }
1254: } else { # Can start the request:
1255: Debug(8,"Can start...");
1256: StartRequest($LondSocket, $requestData);
1257: }
1258: }
1259:
1260: #-------------------------- Lonc UNIX socket handling ---------------------
1261:
1262: =pod
1263:
1264: =head2 ClientRequest
1265: Callback that is called when data can be read from the UNIX domain
1266: socket connecting us with an apache server process.
1267:
1268: =cut
1269:
1270: sub ClientRequest {
1271: Debug(6, "ClientRequest");
1272: my $event = shift;
1273: my $watcher = $event->w;
1274: my $socket = $watcher->fd;
1275: my $data = $watcher->data;
1276: my $thisread;
1277:
1278: Debug(9, " Watcher named: ".$watcher->desc);
1279:
1280: my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
1281: Debug(8, "rcv: data length = ".length($thisread)
1282: ." read =".$thisread);
1283: unless (defined $rv && length($thisread)) {
1284: # Likely eof on socket.
1285: Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
1286: close($socket);
1287: $watcher->cancel();
1288: delete($ActiveClients{$socket});
1289: return;
1290: }
1291: Debug(8,"Data: ".$data." this read: ".$thisread);
1292: $data = $data.$thisread; # Append new data.
1293: $watcher->data($data);
1294: if($data =~ /\n$/) { # Request entirely read.
1295: if($data eq "close_connection_exit\n") {
1296: Log("CRITICAL",
1297: "Request Close Connection ... exiting");
1298: CloseAllLondConnections();
1299: exit;
1300: }
1301: Debug(8, "Complete transaction received: ".$data);
1302: if($LogTransactions) {
1303: Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
1304: }
1305: my $Transaction = LondTransaction->new($data);
1306: $Transaction->SetClient($socket);
1307: QueueTransaction($Transaction);
1308: $watcher->cancel(); # Done looking for input data.
1309: }
1310:
1311: }
1312:
1313:
1314: =pod
1315:
1316: =head2 NewClient
1317:
1318: Callback that is called when a connection is received on the unix
1319: socket for a new client of lonc. The callback is parameterized by the
1320: event.. which is a-priori assumed to be an io event, and therefore has
1321: an fd member that is the Listener socket. We Accept the connection
1322: and register a new event on the readability of that socket:
1323:
1324: =cut
1325:
1326: sub NewClient {
1327: Debug(6, "NewClient");
1328: my $event = shift; # Get the event parameters.
1329: my $watcher = $event->w;
1330: my $socket = $watcher->fd; # Get the event' socket.
1331: my $connection = $socket->accept(); # Accept the client connection.
1332: Debug(8,"Connection request accepted from "
1333: .GetPeername($connection, AF_UNIX));
1334:
1335:
1336: my $description = sprintf("Connection to lonc client %d",
1337: $ClientConnection);
1338: Debug(9, "Creating event named: ".$description);
1339: Event->io(cb => \&ClientRequest,
1340: poll => 'r',
1341: desc => $description,
1342: data => "",
1343: fd => $connection);
1344: $ActiveClients{$connection} = $ClientConnection;
1345: $ClientConnection++;
1346: }
1347:
1348: =pod
1349:
1350: =head2 GetLoncSocketPath
1351:
1352: Returns the name of the UNIX socket on which to listen for client
1353: connections.
1354:
1355: =cut
1356:
1357: sub GetLoncSocketPath {
1358: return $UnixSocketDir."/".GetServerHost();
1359: }
1360:
1361: =pod
1362:
1363: =head2 GetServerHost
1364:
1365: Returns the host whose lond we talk with.
1366:
1367: =cut
1368:
1369: sub GetServerHost {
1370: return $RemoteHost; # Setup by the fork.
1371: }
1372:
1373: =pod
1374:
1375: =head2 GetServerPort
1376:
1377: Returns the lond port number.
1378:
1379: =cut
1380:
1381: sub GetServerPort {
1382: return $perlvar{londPort};
1383: }
1384:
1385: =pod
1386:
1387: =head2 SetupLoncListener
1388:
1389: Setup a lonc listener event. The event is called when the socket
1390: becomes readable.. that corresponds to the receipt of a new
1391: connection. The event handler established will accept the connection
1392: (creating a communcations channel), that int turn will establish
1393: another event handler to subess requests.
1394:
1395: =cut
1396:
1397: sub SetupLoncListener {
1398:
1399: my $socket;
1400: my $SocketName = GetLoncSocketPath();
1401: unlink($SocketName);
1402: unless ($socket =IO::Socket::UNIX->new(Local => $SocketName,
1403: Listen => 10,
1404: Type => SOCK_STREAM)) {
1405: die "Failed to create a lonc listner socket";
1406: }
1407: Event->io(cb => \&NewClient,
1408: poll => 'r',
1409: desc => 'Lonc listener Unix Socket',
1410: fd => $socket);
1411: }
1412:
1413: #
1414: # Toggle transaction logging.
1415: # Implicit inputs:
1416: # LogTransactions
1417: # Implicit Outputs:
1418: # LogTransactions
1419: sub ToggleTransactionLogging {
1420: print STDERR "Toggle transaction logging...\n";
1421: if(!$LogTransactions) {
1422: $LogTransactions = 1;
1423: } else {
1424: $LogTransactions = 0;
1425: }
1426:
1427:
1428: Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
1429: }
1430:
1431: =pod
1432:
1433: =head2 ChildStatus
1434:
1435: Child USR1 signal handler to report the most recent status
1436: into the status file.
1437:
1438: We also use this to reset the retries count in order to allow the
1439: client to retry connections with a previously dead server.
1440: =cut
1441:
1442: sub ChildStatus {
1443: my $event = shift;
1444: my $watcher = $event->w;
1445:
1446: Debug(2, "Reporting child status because : ".$watcher->data);
1447: my $docdir = $perlvar{'lonDocRoot'};
1448: my $fh = IO::File->new(">>$docdir/lon-status/loncstatus.txt");
1449: print $fh $$."\t".$RemoteHost."\t".$Status."\t".
1450: $RecentLogEntry."\n";
1451: #
1452: # Write out information about each of the connections:
1453: #
1454: if ($DebugLevel > 2) {
1455: print $fh "Active connection statuses: \n";
1456: my $i = 1;
1457: print STDERR "================================= Socket Status Dump:\n";
1458: foreach my $item (keys %ActiveConnections) {
1459: my $Socket = $ActiveConnections{$item}->data;
1460: my $state = $Socket->GetState();
1461: print $fh "Connection $i State: $state\n";
1462: print STDERR "---------------------- Connection $i \n";
1463: $Socket->Dump(-1); # Ensure it gets dumped..
1464: $i++;
1465: }
1466: }
1467: $ConnectionRetriesLeft = $ConnectionRetries;
1468: }
1469:
1470: =pod
1471:
1472: =head2 SignalledToDeath
1473:
1474: Called in response to a signal that causes a chid process to die.
1475:
1476: =cut
1477:
1478:
1479: sub SignalledToDeath {
1480: my $event = shift;
1481: my $watcher= $event->w;
1482:
1483: Debug(2,"Signalled to death! via ".$watcher->data);
1484: my ($signal) = $watcher->data;
1485: chomp($signal);
1486: Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost "
1487: ."died through "."\"$signal\"");
1488: LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
1489: ."\"$signal\"");
1490: exit 0;
1491:
1492: }
1493:
1494: =head2 ToggleDebug
1495:
1496: This sub toggles trace debugging on and off.
1497:
1498: =cut
1499:
1500: sub ToggleDebug {
1501: my $Current = $DebugLevel;
1502: $DebugLevel = $NextDebugLevel;
1503: $NextDebugLevel = $Current;
1504:
1505: Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
1506:
1507: }
1508:
1509: =head2 ChildProcess
1510:
1511: This sub implements a child process for a single lonc daemon.
1512:
1513: =cut
1514:
1515: sub ChildProcess {
1516:
1517:
1518: #
1519: # Signals must be handled by the Event framework...
1520: #
1521:
1522: Event->signal(signal => "QUIT",
1523: cb => \&SignalledToDeath,
1524: data => "QUIT");
1525: Event->signal(signal => "HUP",
1526: cb => \&ChildStatus,
1527: data => "HUP");
1528: Event->signal(signal => "USR1",
1529: cb => \&ChildStatus,
1530: data => "USR1");
1531: Event->signal(signal => "USR2",
1532: cb => \&ToggleTransactionLogging);
1533: Event->signal(signal => "INT",
1534: cb => \&ToggleDebug,
1535: data => "INT");
1536:
1537:
1538: SetupLoncListener();
1539:
1540: $Event::Debuglevel = $DebugLevel;
1541:
1542: Debug(9, "Making initial lond connection for ".$RemoteHost);
1543:
1544: # Setup the initial server connection:
1545:
1546: # &MakeLondConnection(); // let first work requirest do it.
1547:
1548:
1549: Debug(9,"Entering event loop");
1550: my $ret = Event::loop(); # Start the main event loop.
1551:
1552:
1553: die "Main event loop exited!!!";
1554: }
1555:
1556: # Create a new child for host passed in:
1557:
1558: sub CreateChild {
1559: my $host = shift;
1560:
1561: my $sigset = POSIX::SigSet->new(SIGINT);
1562: sigprocmask(SIG_BLOCK, $sigset);
1563: $RemoteHost = $host;
1564: Log("CRITICAL", "Forking server for ".$host);
1565: my $pid = fork;
1566: if($pid) { # Parent
1567: $RemoteHost = "Parent";
1568: $ChildHash{$pid} = $host;
1569: $HostToPid{$host}= $pid;
1570: sigprocmask(SIG_UNBLOCK, $sigset);
1571:
1572: } else { # child.
1573: ShowStatus("Connected to ".$RemoteHost);
1574: $SIG{INT} = 'DEFAULT';
1575: sigprocmask(SIG_UNBLOCK, $sigset);
1576: ChildProcess; # Does not return.
1577: }
1578:
1579: }
1580: #
1581: # Parent process logic pass 1:
1582: # For each entry in the hosts table, we will
1583: # fork off an instance of ChildProcess to service the transactions
1584: # to that host. Each pid will be entered in a global hash
1585: # with the value of the key, the host.
1586: # The parent will then enter a loop to wait for process exits.
1587: # Each exit gets logged and the child gets restarted.
1588: #
1589:
1590: #
1591: # Fork and start in new session so hang-up isn't going to
1592: # happen without intent.
1593: #
1594:
1595:
1596:
1597:
1598:
1599:
1600: ShowStatus("Forming new session");
1601: my $childpid = fork;
1602: if ($childpid != 0) {
1603: sleep 4; # Give child a chacne to break to
1604: exit 0; # a new sesion.
1605: }
1606: #
1607: # Write my pid into the pid file so I can be located
1608: #
1609:
1610: ShowStatus("Parent writing pid file:");
1611: my $execdir = $perlvar{'lonDaemons'};
1612: open (PIDSAVE, ">$execdir/logs/lonc.pid");
1613: print PIDSAVE "$$\n";
1614: close(PIDSAVE);
1615:
1616:
1617:
1618: if (POSIX::setsid() < 0) {
1619: print "Could not create new session\n";
1620: exit -1;
1621: }
1622:
1623: ShowStatus("Forking node servers");
1624:
1625: Log("CRITICAL", "--------------- Starting children ---------------");
1626:
1627: LondConnection::ReadConfig; # Read standard config files.
1628: my $HostIterator = LondConnection::GetHostIterator;
1629: while (! $HostIterator->end()) {
1630:
1631: my $hostentryref = $HostIterator->get();
1632: CreateChild($hostentryref->[0]);
1633: $HostHash{$hostentryref->[0]} = $hostentryref->[4];
1634: $HostIterator->next();
1635: }
1636: $RemoteHost = "Parent Server";
1637:
1638: # Maintain the population:
1639:
1640: ShowStatus("Parent keeping the flock");
1641:
1642: #
1643: # Set up parent signals:
1644: #
1645:
1646: $SIG{INT} = \&Terminate;
1647: $SIG{TERM} = \&Terminate;
1648: $SIG{HUP} = \&Restart;
1649: $SIG{USR1} = \&CheckKids;
1650: $SIG{USR2} = \&UpdateKids; # LonManage update request.
1651:
1652: while(1) {
1653: my $deadchild = wait();
1654: if(exists $ChildHash{$deadchild}) { # need to restart.
1655: my $deadhost = $ChildHash{$deadchild};
1656: delete($HostToPid{$deadhost});
1657: delete($ChildHash{$deadchild});
1658: Log("WARNING","Lost child pid= ".$deadchild.
1659: "Connected to host ".$deadhost);
1660: Log("INFO", "Restarting child procesing ".$deadhost);
1661: CreateChild($deadhost);
1662: }
1663: }
1664:
1665:
1666:
1667: =pod
1668:
1669: =head1 CheckKids
1670:
1671: Since kids do not die as easily in this implementation
1672: as the previous one, there is no need to restart the
1673: dead ones (all dead kids get restarted when they die!!)
1674: The only thing this function does is to pass USR1 to the
1675: kids so that they report their status.
1676:
1677: =cut
1678:
1679: sub CheckKids {
1680: Debug(2, "Checking status of children");
1681: my $docdir = $perlvar{'lonDocRoot'};
1682: my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
1683: my $now=time;
1684: my $local=localtime($now);
1685: print $fh "LONC status $local - parent $$ \n\n";
1686: foreach my $pid (keys %ChildHash) {
1687: Debug(2, "Sending USR1 -> $pid");
1688: kill 'USR1' => $pid; # Tell Child to report status.
1689: sleep 1; # Wait so file doesn't intermix.
1690: }
1691: }
1692:
1693: =pod
1694:
1695: =head1 UpdateKids
1696:
1697: parent's SIGUSR2 handler. This handler:
1698:
1699: =item
1700:
1701: Rereads the hosts file.
1702:
1703: =item
1704:
1705: Kills off (via sigint) children for hosts that have disappeared.
1706:
1707: =item
1708:
1709: QUITs children for hosts that already exist (this just forces a status display
1710: and resets the connection retry count for that host.
1711:
1712: =item
1713:
1714: Starts new children for hosts that have been added to the hosts.tab file since
1715: the start of the master program and maintains them.
1716:
1717: =cut
1718:
1719: sub UpdateKids {
1720:
1721: Log("INFO", "Updating connections via SIGUSR2");
1722:
1723: # Just in case we need to kill our own lonc, we wait a few seconds to
1724: # give it a chance to receive and relay lond's response to the
1725: # re-init command.
1726: #
1727:
1728: sleep(2); # Wait a couple of seconds.
1729:
1730: my %hosts; # Indexed by loncapa hostname, value=ip.
1731:
1732: # Need to re-read the host table:
1733:
1734:
1735: LondConnection::ReadConfig();
1736: my $I = LondConnection::GetHostIterator;
1737: while (! $I->end()) {
1738: my $item = $I->get();
1739: $hosts{$item->[0]} = $item->[4];
1740: $I->next();
1741: }
1742:
1743: # The logic below is written for clarity not for efficiency.
1744: # Since I anticipate that this function is only rarely called, that's
1745: # appropriate. There are certainly ways to combine the loops below,
1746: # and anyone wishing to obscure the logic is welcome to go for it.
1747: # Note that we don't re-direct sigchild. Instead we do what's needed
1748: # to the data structures that keep track of children to ensure that
1749: # when sigchild is honored, no new child is born.
1750: #
1751:
1752: # For each existing child; if it's host doesn't exist, kill the child.
1753:
1754: foreach my $child (keys %ChildHash) {
1755: my $oldhost = $ChildHash{$child};
1756: if (!(exists $hosts{$oldhost})) {
1757: Log("CRITICAL", "Killing child for $oldhost host no longer exists");
1758: delete $ChildHash{$child};
1759: delete $HostToPid{$oldhost};
1760: kill 'QUIT' => $child;
1761: }
1762: }
1763: # For each remaining existing child; if it's host's ip has changed,
1764: # Restart the child on the new IP.
1765:
1766: foreach my $child (keys %ChildHash) {
1767: my $oldhost = $ChildHash{$child};
1768: my $oldip = $HostHash{$oldhost};
1769: if ($hosts{$oldhost} ne $oldip) {
1770:
1771: # kill the old child.
1772:
1773: Log("CRITICAL", "Killing child for $oldhost host ip has changed...");
1774: delete $ChildHash{$child};
1775: delete $HostToPid{$oldhost};
1776: kill 'QUIT' => $child;
1777:
1778: # Do the book-keeping needed to start a new child on the
1779: # new ip.
1780:
1781: $HostHash{$oldhost} = $hosts{$oldhost};
1782: CreateChild($oldhost);
1783: }
1784: }
1785: # Finally, for each new host, not in the host hash, create a
1786: # enter the host and create a new child.
1787: # Force a status display of any existing process.
1788:
1789: foreach my $host (keys %hosts) {
1790: if(!(exists $HostHash{$host})) {
1791: Log("INFO", "New host $host discovered in hosts.tab...");
1792: $HostHash{$host} = $hosts{$host};
1793: CreateChild($host);
1794: } else {
1795: kill 'HUP' => $HostToPid{$host}; # status display.
1796: }
1797: }
1798: }
1799:
1800:
1801: =pod
1802:
1803: =head1 Restart
1804:
1805: Signal handler for HUP... all children are killed and
1806: we self restart. This is an el-cheapo way to re read
1807: the config file.
1808:
1809: =cut
1810:
1811: sub Restart {
1812: &KillThemAll; # First kill all the children.
1813: Log("CRITICAL", "Restarting");
1814: my $execdir = $perlvar{'lonDaemons'};
1815: unlink("$execdir/logs/lonc.pid");
1816: exec("$execdir/loncnew");
1817: }
1818:
1819: =pod
1820:
1821: =head1 KillThemAll
1822:
1823: Signal handler that kills all children by sending them a
1824: SIGHUP. Responds to sigint and sigterm.
1825:
1826: =cut
1827:
1828: sub KillThemAll {
1829: Debug(2, "Kill them all!!");
1830: local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die.
1831: foreach my $pid (keys %ChildHash) {
1832: my $serving = $ChildHash{$pid};
1833: ShowStatus("Nicely Killing lonc for $serving pid = $pid");
1834: Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
1835: kill 'QUIT' => $pid;
1836: }
1837:
1838:
1839: }
1840:
1841:
1842: #
1843: # Kill all children via KILL. Just in case the
1844: # first shot didn't get them.
1845:
1846: sub really_kill_them_all_dammit
1847: {
1848: Debug(2, "Kill them all Dammit");
1849: local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
1850: foreach my $pid (keys %ChildHash) {
1851: my $serving = $ChildHash{$pid};
1852: &ShowStatus("Nastily killing lonc for $serving pid = $pid");
1853: Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
1854: kill 'KILL' => $pid;
1855: delete($ChildHash{$pid});
1856: my $execdir = $perlvar{'lonDaemons'};
1857: unlink("$execdir/logs/lonc.pid");
1858: }
1859: }
1860: =pod
1861:
1862: =head1 Terminate
1863:
1864: Terminate the system.
1865:
1866: =cut
1867:
1868: sub Terminate {
1869: &Log("CRITICAL", "Asked to kill children.. first be nice...");
1870: &KillThemAll;
1871: #
1872: # By now they really should all be dead.. but just in case
1873: # send them all SIGKILL's after a bit of waiting:
1874:
1875: sleep(4);
1876: &Log("CRITICAL", "Now kill children nasty");
1877: &really_kill_them_all_dammit;
1878: Log("CRITICAL","Master process exiting");
1879: exit 0;
1880:
1881: }
1882: =pod
1883:
1884: =head1 Theory
1885:
1886: The event class is used to build this as a single process with an
1887: event driven model. The following events are handled:
1888:
1889: =item UNIX Socket connection Received
1890:
1891: =item Request data arrives on UNIX data transfer socket.
1892:
1893: =item lond connection becomes writable.
1894:
1895: =item timer fires at 1 second intervals.
1896:
1897: All sockets are run in non-blocking mode. Timeouts managed by the timer
1898: handler prevents hung connections.
1899:
1900: Key data structures:
1901:
1902: =item RequestQueue
1903:
1904: A queue of requests received from UNIX sockets that are
1905: waiting for a chance to be forwarded on a lond connection socket.
1906:
1907: =item ActiveConnections
1908:
1909: A hash of lond connections that have transactions in process that are
1910: available to be timed out.
1911:
1912: =item ActiveTransactions
1913:
1914: A hash indexed by lond connections that contain the client reply
1915: socket for each connection that has an active transaction on it.
1916:
1917: =item IdleConnections
1918:
1919: A hash of lond connections that have no work to do. These connections
1920: can be closed if they are idle for a long enough time.
1921:
1922: =cut
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>