File:
[LON-CAPA] /
loncom /
loncnew
Revision
1.56:
download - view:
text,
annotated -
select for diffs
Tue Sep 21 10:54:43 2004 UTC (20 years ago) by
foxr
Branches:
MAIN
CVS tags:
HEAD
Only allow new connections if there are no connection requests in progress.
This should prevent connection explosions due to short term load spikes.
I believe that connection explosion may also be responsible for transaction
failures due to resource issues.
1: #!/usr/bin/perl
2: # The LearningOnline Network with CAPA
3: # lonc maintains the connections to remote computers
4: #
5: # $Id: loncnew,v 1.56 2004/09/21 10:54:43 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: my $LondConnecting = 0; # True when a connection is being built.
108:
109: #
110: # The hash below gives the HTML format for log messages
111: # given a severity.
112: #
113: my %LogFormats;
114:
115: $LogFormats{"CRITICAL"} = "<font color='red'>CRITICAL: %s</font>";
116: $LogFormats{"SUCCESS"} = "<font color='green'>SUCCESS: %s</font>";
117: $LogFormats{"INFO"} = "<font color='yellow'>INFO: %s</font>";
118: $LogFormats{"WARNING"} = "<font color='blue'>WARNING: %s</font>";
119: $LogFormats{"DEFAULT"} = " %s ";
120:
121:
122:
123: =pod
124:
125: =head2 LogPerm
126:
127: Makes an entry into the permanent log file.
128:
129: =cut
130: sub LogPerm {
131: my $message=shift;
132: my $execdir=$perlvar{'lonDaemons'};
133: my $now=time;
134: my $local=localtime($now);
135: my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
136: print $fh "$now:$message:$local\n";
137: }
138:
139: =pod
140:
141: =head2 Log
142:
143: Logs a message to the log file.
144: Parameters:
145:
146: =item severity
147:
148: One of CRITICAL, WARNING, INFO, SUCCESS used to select the
149: format string used to format the message. if the severity is
150: not a defined severity the Default format string is used.
151:
152: =item message
153:
154: The base message. In addtion to the format string, the message
155: will be appended to a string containing the name of our remote
156: host and the time will be formatted into the message.
157:
158: =cut
159:
160: sub Log {
161:
162: my ($severity, $message) = @_;
163:
164: if(!$LogFormats{$severity}) {
165: $severity = "DEFAULT";
166: }
167:
168: my $format = $LogFormats{$severity};
169:
170: # Put the window dressing in in front of the message format:
171:
172: my $now = time;
173: my $local = localtime($now);
174: my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
175: my $finalformat = $finalformat.$format."\n";
176:
177: # open the file and put the result.
178:
179: my $execdir = $perlvar{'lonDaemons'};
180: my $fh = IO::File->new(">>$execdir/logs/lonc.log");
181: my $msg = sprintf($finalformat, $message);
182: $RecentLogEntry = $msg;
183: print $fh $msg;
184:
185:
186: }
187:
188:
189: =pod
190:
191: =head2 GetPeerName
192:
193: Returns the name of the host that a socket object is connected to.
194:
195: =cut
196:
197: sub GetPeername {
198:
199:
200: my ($connection, $AdrFamily) = @_;
201:
202: my $peer = $connection->peername();
203: my $peerport;
204: my $peerip;
205: if($AdrFamily == AF_INET) {
206: ($peerport, $peerip) = sockaddr_in($peer);
207: my $peername = gethostbyaddr($peerip, $AdrFamily);
208: return $peername;
209: } elsif ($AdrFamily == AF_UNIX) {
210: my $peerfile;
211: ($peerfile) = sockaddr_un($peer);
212: return $peerfile;
213: }
214: }
215: =pod
216:
217: =head2 Debug
218:
219: Invoked to issue a debug message.
220:
221: =cut
222:
223: sub Debug {
224:
225: my ($level, $message) = @_;
226:
227: if ($level <= $DebugLevel) {
228: Log("INFO", "-Debug- $message host = $RemoteHost");
229: }
230: }
231:
232: sub SocketDump {
233:
234: my ($level, $socket) = @_;
235:
236: if($level <= $DebugLevel) {
237: $socket->Dump(-1); # Ensure it will get dumped.
238: }
239: }
240:
241: =pod
242:
243: =head2 ShowStatus
244:
245: Place some text as our pid status.
246: and as what we return in a SIGUSR1
247:
248: =cut
249: sub ShowStatus {
250: my $state = shift;
251: my $now = time;
252: my $local = localtime($now);
253: $Status = $local.": ".$state;
254: $0='lonc: '.$state.' '.$local;
255: }
256:
257: =pod
258:
259: =head 2 SocketTimeout
260:
261: Called when an action on the socket times out. The socket is
262: destroyed and any active transaction is failed.
263:
264:
265: =cut
266: sub SocketTimeout {
267: my $Socket = shift;
268: Log("WARNING", "A socket timeout was detected");
269: Debug(5, " SocketTimeout called: ");
270: $Socket->Dump(0);
271: if(exists($ActiveTransactions{$Socket})) {
272: FailTransaction($ActiveTransactions{$Socket});
273: }
274: KillSocket($Socket); # A transaction timeout also counts as
275: # a connection failure:
276: $ConnectionRetriesLeft--;
277: if($ConnectionRetriesLeft <= 0) {
278: Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
279: $LondConnecting = 0;
280: }
281:
282: }
283: #----------------------------- Timer management ------------------------
284:
285: =pod
286:
287: =head2 Tick
288:
289: Invoked each timer tick.
290:
291: =cut
292:
293:
294: sub Tick {
295: my ($Event) = @_;
296: my $clock_watcher = $Event->w;
297:
298: my $client;
299: if($ConnectionRetriesLeft > 0) {
300: ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount
301: ." Retries remaining: ".$ConnectionRetriesLeft
302: ." ($KeyMode)");
303: } else {
304: ShowStatus(GetServerHost()." >> DEAD <<");
305: }
306: # Is it time to prune connection count:
307:
308:
309: if($IdleConnections->Count() &&
310: ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
311: $IdleSeconds++;
312: if($IdleSeconds > $IdleTimeout) { # Prune a connection...
313: my $Socket = $IdleConnections->pop();
314: KillSocket($Socket);
315: $IdleSeconds = 0; # Otherwise all connections get trimmed to fast.
316: }
317: } else {
318: $IdleSeconds = 0; # Reset idle count if not idle.
319: }
320: #
321: # For each inflight transaction, tick down its timeout counter.
322: #
323:
324: foreach my $item (keys %ActiveConnections) {
325: my $State = $ActiveConnections{$item}->data->GetState();
326: if ($State ne 'Idle') {
327: Debug(5,"Ticking Socket $State $item");
328: $ActiveConnections{$item}->data->Tick();
329: }
330: }
331: # Do we have work in the queue, but no connections to service them?
332: # If so, try to make some new connections to get things going again.
333: #
334:
335: my $Requests = $WorkQueue->Count();
336: if (($ConnectionCount == 0) && ($Requests > 0) && (!$LondConnecting)) {
337: if ($ConnectionRetriesLeft > 0) {
338: Debug(5,"Work but no connections, Make a new one");
339: my $success;
340: $success = &MakeLondConnection;
341: if($success == 0) { # All connections failed:
342: Debug(5,"Work in queue failed to make any connectiouns\n");
343: EmptyQueue(); # Fail pending transactions with con_lost.
344: CloseAllLondConnections(); # Should all be closed but....
345: }
346: } else {
347: $LondConnecting = 0;
348: ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
349: Debug(5,"Work in queue, but gave up on connections..flushing\n");
350: EmptyQueue(); # Connections can't be established.
351: CloseAllLondConnections(); # Should all already be closed but...
352: }
353:
354: }
355: if ($ConnectionCount == 0) {
356: $KeyMode = "";
357: $clock_watcher->cancel();
358: }
359: }
360:
361: =pod
362:
363: =head2 SetupTimer
364:
365: Sets up a 1 per sec recurring timer event. The event handler is used to:
366:
367: =item
368:
369: Trigger timeouts on communications along active sockets.
370:
371: =item
372:
373: Trigger disconnections of idle sockets.
374:
375: =cut
376:
377: sub SetupTimer {
378: Debug(6, "SetupTimer");
379: Event->timer(interval => 1, cb => \&Tick );
380: }
381:
382: =pod
383:
384: =head2 ServerToIdle
385:
386: This function is called when a connection to the server is
387: ready for more work.
388:
389: If there is work in the Work queue the top element is dequeued
390: and the connection will start to work on it. If the work queue is
391: empty, the connection is pushed on the idle connection stack where
392: it will either get another work unit, or alternatively, if it sits there
393: long enough, it will be shut down and released.
394:
395: =cut
396:
397: sub ServerToIdle {
398: my $Socket = shift; # Get the socket.
399: $KeyMode = $Socket->{AuthenticationMode};
400: delete($ActiveTransactions{$Socket}); # Server has no transaction
401:
402: &Debug(5, "Server to idle");
403:
404: # If there's work to do, start the transaction:
405:
406: my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
407: if ($reqdata ne undef) {
408: Debug(5, "Queue gave request data: ".$reqdata->getRequest());
409: &StartRequest($Socket, $reqdata);
410:
411: } else {
412:
413: # There's no work waiting, so push the server to idle list.
414: &Debug(5, "No new work requests, server connection going idle");
415: $IdleConnections->push($Socket);
416: }
417: }
418:
419: =pod
420:
421: =head2 ClientWritable
422:
423: Event callback for when a client socket is writable.
424:
425: This callback is established when a transaction reponse is
426: avaiable from lond. The response is forwarded to the unix socket
427: as it becomes writable in this sub.
428:
429: Parameters:
430:
431: =item Event
432:
433: The event that has been triggered. Event->w->data is
434: the data and Event->w->fd is the socket to write.
435:
436: =cut
437:
438: sub ClientWritable {
439: my $Event = shift;
440: my $Watcher = $Event->w;
441: my $Data = $Watcher->data;
442: my $Socket = $Watcher->fd;
443:
444: # Try to send the data:
445:
446: &Debug(6, "ClientWritable writing".$Data);
447: &Debug(9, "Socket is: ".$Socket);
448:
449: if($Socket->connected) {
450: my $result = $Socket->send($Data, 0);
451:
452: # $result undefined: the write failed.
453: # otherwise $result is the number of bytes written.
454: # Remove that preceding string from the data.
455: # If the resulting data is empty, destroy the watcher
456: # and set up a read event handler to accept the next
457: # request.
458:
459: &Debug(9,"Send result is ".$result." Defined: ".defined($result));
460: if($result ne undef) {
461: &Debug(9, "send result was defined");
462: if($result == length($Data)) { # Entire string sent.
463: &Debug(9, "ClientWritable data all written");
464: $Watcher->cancel();
465: #
466: # Set up to read next request from socket:
467:
468: my $descr = sprintf("Connection to lonc client %d",
469: $ActiveClients{$Socket});
470: Event->io(cb => \&ClientRequest,
471: poll => 'r',
472: desc => $descr,
473: data => "",
474: fd => $Socket);
475:
476: } else { # Partial string sent.
477: $Watcher->data(substr($Data, $result));
478: if($result == 0) { # client hung up on us!!
479: # Log("INFO", "lonc pipe client hung up on us!");
480: $Watcher->cancel;
481: $Socket->shutdown(2);
482: $Socket->close();
483: }
484: }
485:
486: } else { # Error of some sort...
487:
488: # Some errnos are possible:
489: my $errno = $!;
490: if($errno == POSIX::EWOULDBLOCK ||
491: $errno == POSIX::EAGAIN ||
492: $errno == POSIX::EINTR) {
493: # No action taken?
494: } else { # Unanticipated errno.
495: &Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
496: $Watcher->cancel; # Stop the watcher.
497: $Socket->shutdown(2); # Kill connection
498: $Socket->close(); # Close the socket.
499: }
500:
501: }
502: } else {
503: $Watcher->cancel(); # A delayed request...just cancel.
504: }
505: }
506:
507: =pod
508:
509: =head2 CompleteTransaction
510:
511: Called when the reply data has been received for a lond
512: transaction. The reply data must now be sent to the
513: ultimate client on the other end of the Unix socket. This is
514: done by setting up a writable event for the socket with the
515: data the reply data.
516:
517: Parameters:
518:
519: =item Socket
520:
521: Socket on which the lond transaction occured. This is a
522: LondConnection. The data received is in the TransactionReply member.
523:
524: =item Transaction
525:
526: The transaction that is being completed.
527:
528: =cut
529:
530: sub CompleteTransaction {
531: &Debug(5,"Complete transaction");
532:
533: my ($Socket, $Transaction) = @_;
534:
535: if (!$Transaction->isDeferred()) { # Normal transaction
536: my $data = $Socket->GetReply(); # Data to send.
537: if($LogTransactions) {
538: Log("SUCCESS", "Reply from lond: '$data'");
539: }
540: StartClientReply($Transaction, $data);
541: } else { # Delete deferred transaction file.
542: Log("SUCCESS", "A delayed transaction was completed");
543: LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest());
544: unlink $Transaction->getFile();
545: }
546: }
547:
548: =pod
549:
550: =head1 StartClientReply
551:
552: Initiates a reply to a client where the reply data is a parameter.
553:
554: =head2 parameters:
555:
556: =item Transaction
557:
558: The transaction for which we are responding to the client.
559:
560: =item data
561:
562: The data to send to apached client.
563:
564: =cut
565:
566: sub StartClientReply {
567:
568: my ($Transaction, $data) = @_;
569:
570: my $Client = $Transaction->getClient();
571:
572: &Debug(8," Reply was: ".$data);
573: my $Serial = $ActiveClients{$Client};
574: my $desc = sprintf("Connection to lonc client %d",
575: $Serial);
576: Event->io(fd => $Client,
577: poll => "w",
578: desc => $desc,
579: cb => \&ClientWritable,
580: data => $data);
581: }
582:
583: =pod
584:
585: =head2 FailTransaction
586:
587: Finishes a transaction with failure because the associated lond socket
588: disconnected. There are two possibilities:
589: - The transaction is deferred: in which case we just quietly
590: delete the transaction since there is no client connection.
591: - The transaction is 'live' in which case we initiate the sending
592: of "con_lost" to the client.
593:
594: Deleting the transaction means killing it from the %ActiveTransactions hash.
595:
596: Parameters:
597:
598: =item client
599:
600: The LondTransaction we are failing.
601:
602:
603: =cut
604:
605: sub FailTransaction {
606: my $transaction = shift;
607:
608: # If the socket is dead, that's already logged.
609:
610: if ($ConnectionRetriesLeft > 0) {
611: Log("WARNING", "Failing transaction "
612: .$transaction->getRequest());
613: }
614: Debug(1, "Failing transaction: ".$transaction->getRequest());
615: if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
616: my $client = $transaction->getClient();
617: Debug(1," Replying con_lost to ".$transaction->getRequest());
618: StartClientReply($transaction, "con_lost\n");
619: }
620:
621: }
622:
623: =pod
624: =head1 EmptyQueue
625:
626: Fails all items in the work queue with con_lost.
627: Note that each item in the work queue is a transaction.
628:
629: =cut
630: sub EmptyQueue {
631: $ConnectionRetriesLeft--; # Counts as connection failure too.
632: while($WorkQueue->Count()) {
633: my $request = $WorkQueue->dequeue(); # This is a transaction
634: FailTransaction($request);
635: }
636: }
637:
638: =pod
639:
640: =head2 CloseAllLondConnections
641:
642: Close all connections open on lond prior to exit e.g.
643:
644: =cut
645: sub CloseAllLondConnections {
646: foreach my $Socket (keys %ActiveConnections) {
647: if(exists($ActiveTransactions{$Socket})) {
648: FailTransaction($ActiveTransactions{$Socket});
649: }
650: KillSocket($Socket);
651: }
652: }
653: =cut
654:
655: =pod
656:
657: =head2 KillSocket
658:
659: Destroys a socket. This function can be called either when a socket
660: has died of 'natural' causes or because a socket needs to be pruned due to
661: idleness. If the socket has died naturally, if there are no longer any
662: live connections a new connection is created (in case there are transactions
663: in the queue). If the socket has been pruned, it is never re-created.
664:
665: Parameters:
666:
667: =item Socket
668:
669: The socket to kill off.
670:
671: =item Restart
672:
673: nonzero if we are allowed to create a new connection.
674:
675:
676: =cut
677: sub KillSocket {
678: my $Socket = shift;
679:
680: Log("WARNING", "Shutting down a socket");
681: $Socket->Shutdown();
682:
683: # If the socket came from the active connection set,
684: # delete its transaction... note that FailTransaction should
685: # already have been called!!!
686: # otherwise it came from the idle set.
687: #
688:
689: if(exists($ActiveTransactions{$Socket})) {
690: delete ($ActiveTransactions{$Socket});
691: }
692: if(exists($ActiveConnections{$Socket})) {
693: delete($ActiveConnections{$Socket});
694: $ConnectionCount--;
695: if ($ConnectionCount < 0) { $ConnectionCount = 0; }
696: }
697: # If the connection count has gone to zero and there is work in the
698: # work queue, the work all gets failed with con_lost.
699: #
700: if($ConnectionCount == 0) {
701: EmptyQueue();
702: CloseAllLondConnections; # Should all already be closed but...
703: }
704: }
705:
706: =pod
707:
708: =head2 LondReadable
709:
710: This function is called whenever a lond connection
711: is readable. The action is state dependent:
712:
713: =head3 State=Initialized
714:
715: We''re waiting for the challenge, this is a no-op until the
716: state changes.
717:
718: =head3 State=Challenged
719:
720: The challenge has arrived we need to transition to Writable.
721: The connection must echo the challenge back.
722:
723: =head3 State=ChallengeReplied
724:
725: The challenge has been replied to. The we are receiveing the
726: 'ok' from the partner.
727:
728: =head3 State=ReadingVersionString
729:
730: We have requested the lond version and are reading the
731: version back. Upon completion, we'll store the version away
732: for future use(?).
733:
734: =head3 State=HostSet
735:
736: We have selected the domain name of our peer (multhomed hosts)
737: and are getting the reply (presumably ok) back.
738:
739: =head3 State=RequestingKey
740:
741: The ok has been received and we need to send the request for
742: an encryption key. Transition to writable for that.
743:
744: =head3 State=ReceivingKey
745:
746: The the key has been requested, now we are reading the new key.
747:
748: =head3 State=Idle
749:
750: The encryption key has been negotiated or we have finished
751: reading data from the a transaction. If the callback data has
752: a client as well as the socket iformation, then we are
753: doing a transaction and the data received is relayed to the client
754: before the socket is put on the idle list.
755:
756: =head3 State=SendingRequest
757:
758: I do not think this state can be received here, but if it is,
759: the appropriate thing to do is to transition to writable, and send
760: the request.
761:
762: =head3 State=ReceivingReply
763:
764: We finished sending the request to the server and now transition
765: to readable to receive the reply.
766:
767: The parameter to this function are:
768:
769: The event. Implicit in this is the watcher and its data. The data
770: contains at least the lond connection object and, if a
771: transaction is in progress, the socket attached to the local client.
772:
773: =cut
774:
775: sub LondReadable {
776:
777: my $Event = shift;
778: my $Watcher = $Event->w;
779: my $Socket = $Watcher->data;
780: my $client = undef;
781:
782: &Debug(6,"LondReadable called state = ".$Socket->GetState());
783:
784:
785: my $State = $Socket->GetState(); # All action depends on the state.
786:
787: SocketDump(6, $Socket);
788: my $status = $Socket->Readable();
789:
790: &Debug(2, "Socket->Readable returned: $status");
791:
792: if($status != 0) {
793: # bad return from socket read. Currently this means that
794: # The socket has become disconnected. We fail the transaction.
795:
796: Log("WARNING",
797: "Lond connection lost.");
798: if(exists($ActiveTransactions{$Socket})) {
799: FailTransaction($ActiveTransactions{$Socket});
800: } else {
801: # Socket is connecting and failed... need to mark
802: # no longer connecting.
803:
804: $LondConnecting = 0;
805: }
806: $Watcher->cancel();
807: KillSocket($Socket);
808: $ConnectionRetriesLeft--; # Counts as connection failure
809: return;
810: }
811: SocketDump(6,$Socket);
812:
813: $State = $Socket->GetState(); # Update in case of transition.
814: &Debug(6, "After read, state is ".$State);
815:
816: if($State eq "Initialized") {
817:
818:
819: } elsif ($State eq "ChallengeReceived") {
820: # The challenge must be echoed back; The state machine
821: # in the connection takes care of setting that up. Just
822: # need to transition to writable:
823:
824: $Watcher->cb(\&LondWritable);
825: $Watcher->poll("w");
826:
827: } elsif ($State eq "ChallengeReplied") {
828:
829: } elsif ($State eq "RequestingVersion") {
830: # Need to ask for the version... that is writiability:
831:
832: $Watcher->cb(\&LondWritable);
833: $Watcher->poll("w");
834:
835: } elsif ($State eq "ReadingVersionString") {
836: # Read the rest of the version string...
837: } elsif ($State eq "SetHost") {
838: # Need to request the actual domain get set...
839:
840: $Watcher->cb(\&LondWritable);
841: $Watcher->poll("w");
842: } elsif ($State eq "HostSet") {
843: # Reading the 'ok' from the peer.
844:
845: } elsif ($State eq "RequestingKey") {
846: # The ok was received. Now we need to request the key
847: # That requires us to be writable:
848:
849: $Watcher->cb(\&LondWritable);
850: $Watcher->poll("w");
851:
852: } elsif ($State eq "ReceivingKey") {
853:
854: } elsif ($State eq "Idle") {
855:
856: # This is as good a spot as any to get the peer version
857: # string:
858:
859: if($LondVersion eq "unknown") {
860: $LondVersion = $Socket->PeerVersion();
861: Log("INFO", "Connected to lond version: $LondVersion");
862: }
863: # If necessary, complete a transaction and then go into the
864: # idle queue.
865: # Note that a trasition to idle indicates a live lond
866: # on the other end so reset the connection retries.
867: #
868: $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
869: $Watcher->cancel();
870: if(exists($ActiveTransactions{$Socket})) {
871: Debug(5,"Completing transaction!!");
872: CompleteTransaction($Socket,
873: $ActiveTransactions{$Socket});
874: } else {
875: Log("SUCCESS", "Connection ".$ConnectionCount." to "
876: .$RemoteHost." now ready for action");
877: }
878: ServerToIdle($Socket); # Next work unit or idle.
879:
880: #
881: $LondConnecting = 0; # Best spot I can think of for this.
882: #
883:
884: } elsif ($State eq "SendingRequest") {
885: # We need to be writable for this and probably don't belong
886: # here inthe first place.
887:
888: Deubg(6, "SendingRequest state encountered in readable");
889: $Watcher->poll("w");
890: $Watcher->cb(\&LondWritable);
891:
892: } elsif ($State eq "ReceivingReply") {
893:
894:
895: } else {
896: # Invalid state.
897: Debug(4, "Invalid state in LondReadable");
898: }
899: }
900:
901: =pod
902:
903: =head2 LondWritable
904:
905: This function is called whenever a lond connection
906: becomes writable while there is a writeable monitoring
907: event. The action taken is very state dependent:
908:
909: =head3 State = Connected
910:
911: The connection is in the process of sending the 'init' hailing to the
912: lond on the remote end. The connection object''s Writable member is
913: called. On error, ConnectionError is called to destroy the connection
914: and remove it from the ActiveConnections hash
915:
916: =head3 Initialized
917:
918: 'init' has been sent, writability monitoring is removed and
919: readability monitoring is started with LondReadable as the callback.
920:
921: =head3 ChallengeReceived
922:
923: The connection has received the who are you challenge from the remote
924: system, and is in the process of sending the challenge
925: response. Writable is called.
926:
927: =head3 ChallengeReplied
928:
929: The connection has replied to the initial challenge The we switch to
930: monitoring readability looking for the server to reply with 'ok'.
931:
932: =head3 RequestingKey
933:
934: The connection is in the process of requesting its encryption key.
935: Writable is called.
936:
937: =head3 ReceivingKey
938:
939: The connection has sent the request for a key. Switch to readability
940: monitoring to accept the key
941:
942: =head3 SendingRequest
943:
944: The connection is in the process of sending a request to the server.
945: This request is part of a client transaction. All the states until
946: now represent the client setup protocol. Writable is called.
947:
948: =head3 ReceivingReply
949:
950: The connection has sent a request. Now it must receive a reply.
951: Readability monitoring is requested.
952:
953: This function is an event handler and therefore receives as
954: a parameter the event that has fired. The data for the watcher
955: of this event is a reference to a list of one or two elements,
956: depending on state. The first (and possibly only) element is the
957: socket. The second (present only if a request is in progress)
958: is the socket on which to return a reply to the caller.
959:
960: =cut
961:
962: sub LondWritable {
963: my $Event = shift;
964: my $Watcher = $Event->w;
965: my $Socket = $Watcher->data;
966: my $State = $Socket->GetState();
967:
968: Debug(6,"LondWritable State = ".$State."\n");
969:
970:
971: # Figure out what to do depending on the state of the socket:
972:
973:
974:
975:
976: SocketDump(6,$Socket);
977:
978: # If the socket is writable, we must always write.
979: # Only by writing will we undergo state transitions.
980: # Old logic wrote in state specific code below, however
981: # That forces us at least through another invocation of
982: # this function after writability is possible again.
983: # This logic also factors out common code for handling
984: # write failures... in all cases, write failures
985: # Kill the socket.
986: # This logic makes the branches of the >big< if below
987: # so that the writing states are actually NO-OPs.
988:
989: if ($Socket->Writable() != 0) {
990: # The write resulted in an error.
991: # We'll treat this as if the socket got disconnected:
992: Log("WARNING", "Connection to ".$RemoteHost.
993: " has been disconnected");
994: if(exists($ActiveTransactions{$Socket})) {
995: FailTransaction($ActiveTransactions{$Socket});
996: } else {
997: # In the process of conneting, so need to turn that off.
998:
999: $LondConnecting = 0;
1000: }
1001: $Watcher->cancel();
1002: KillSocket($Socket);
1003: return;
1004: }
1005:
1006:
1007:
1008: if ($State eq "Connected") {
1009:
1010: # "init" is being sent...
1011:
1012: } elsif ($State eq "Initialized") {
1013:
1014: # Now that init was sent, we switch
1015: # to watching for readability:
1016:
1017: $Watcher->cb(\&LondReadable);
1018: $Watcher->poll("r");
1019:
1020: } elsif ($State eq "ChallengeReceived") {
1021: # We received the challenge, now we
1022: # are echoing it back. This is a no-op,
1023: # we're waiting for the state to change
1024:
1025: } elsif ($State eq "ChallengeReplied") {
1026: # The echo was sent back, so we switch
1027: # to watching readability.
1028:
1029: $Watcher->cb(\&LondReadable);
1030: $Watcher->poll("r");
1031: } elsif ($State eq "RequestingVersion") {
1032: # Sending the peer a version request...
1033:
1034: } elsif ($State eq "ReadingVersionString") {
1035: # Transition to read since we have sent the
1036: # version command and now just need to read the
1037: # version string from the peer:
1038:
1039: $Watcher->cb(\&LondReadable);
1040: $Watcher->poll("r");
1041:
1042: } elsif ($State eq "SetHost") {
1043: # Setting the remote domain...
1044:
1045: } elsif ($State eq "HostSet") {
1046: # Back to readable to get the ok.
1047:
1048: $Watcher->cb(\&LondReadable);
1049: $Watcher->poll("r");
1050:
1051:
1052: } elsif ($State eq "RequestingKey") {
1053: # At this time we're requesting the key.
1054: # again, this is essentially a no-op.
1055:
1056: } elsif ($State eq "ReceivingKey") {
1057: # Now we need to wait for the key
1058: # to come back from the peer:
1059:
1060: $Watcher->cb(\&LondReadable);
1061: $Watcher->poll("r");
1062:
1063: } elsif ($State eq "SendingRequest") {
1064:
1065: # At this time we are sending a request to the
1066: # peer... write the next chunk:
1067:
1068:
1069: } elsif ($State eq "ReceivingReply") {
1070: # The send has completed. Wait for the
1071: # data to come in for a reply.
1072: Debug(8,"Writable sent request/receiving reply");
1073: $Watcher->cb(\&LondReadable);
1074: $Watcher->poll("r");
1075:
1076: } else {
1077: # Control only passes here on an error:
1078: # the socket state does not match any
1079: # of the known states... so an error
1080: # must be logged.
1081:
1082: &Debug(4, "Invalid socket state ".$State."\n");
1083: }
1084:
1085: }
1086: =pod
1087:
1088: =cut
1089: sub QueueDelayed {
1090: Debug(3,"QueueDelayed called");
1091:
1092: my $path = "$perlvar{'lonSockDir'}/delayed";
1093:
1094: Debug(4, "Delayed path: ".$path);
1095: opendir(DIRHANDLE, $path);
1096:
1097: my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE;
1098: closedir(DIRHANDLE);
1099: my $dfname;
1100: my $reqfile;
1101: foreach $dfname (sort @alldelayed) {
1102: $reqfile = "$path/$dfname";
1103: Debug(4, "queueing ".$reqfile);
1104: my $Handle = IO::File->new($reqfile);
1105: my $cmd = <$Handle>;
1106: chomp $cmd; # There may or may not be a newline...
1107: $cmd = $cmd."\n"; # now for sure there's exactly one newline.
1108: my $Transaction = LondTransaction->new($cmd);
1109: $Transaction->SetDeferred($reqfile);
1110: QueueTransaction($Transaction);
1111: }
1112:
1113: }
1114:
1115: =pod
1116:
1117: =head2 MakeLondConnection
1118:
1119: Create a new lond connection object, and start it towards its initial
1120: idleness. Once idle, it becomes elligible to receive transactions
1121: from the work queue. If the work queue is not empty when the
1122: connection is completed and becomes idle, it will dequeue an entry and
1123: start off on it.
1124:
1125: =cut
1126:
1127: sub MakeLondConnection {
1128: Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
1129: .GetServerPort());
1130:
1131: my $Connection = LondConnection->new(&GetServerHost(),
1132: &GetServerPort());
1133:
1134: if($Connection eq undef) { # Needs to be more robust later.
1135: Log("CRITICAL","Failed to make a connection with lond.");
1136: $ConnectionRetriesLeft--;
1137: return 0; # Failure.
1138: } else {
1139:
1140: # The connection needs to have writability
1141: # monitored in order to send the init sequence
1142: # that starts the whole authentication/key
1143: # exchange underway.
1144: #
1145: my $Socket = $Connection->GetSocket();
1146: if($Socket eq undef) {
1147: die "did not get a socket from the connection";
1148: } else {
1149: &Debug(9,"MakeLondConnection got socket: ".$Socket);
1150: }
1151:
1152: $Connection->SetTimeoutCallback(\&SocketTimeout);
1153:
1154: my $event = Event->io(fd => $Socket,
1155: poll => 'w',
1156: cb => \&LondWritable,
1157: data => $Connection,
1158: desc => 'Connection to lond server');
1159: $ActiveConnections{$Connection} = $event;
1160: if ($ConnectionCount == 0) {
1161: &SetupTimer; # Need to handle timeouts with connections...
1162: }
1163: $ConnectionCount++;
1164: Debug(4, "Connection count = ".$ConnectionCount);
1165: if($ConnectionCount == 1) { # First Connection:
1166: QueueDelayed;
1167: }
1168: Log("SUCESS", "Created connection ".$ConnectionCount
1169: ." to host ".GetServerHost());
1170: $LondConnecting = 1; # Connection in progress.
1171: return 1; # Return success.
1172: }
1173:
1174: }
1175:
1176: =pod
1177:
1178: =head2 StartRequest
1179:
1180: Starts a lond request going on a specified lond connection.
1181: parameters are:
1182:
1183: =item $Lond
1184:
1185: Connection to the lond that will send the transaction and receive the
1186: reply.
1187:
1188: =item $Client
1189:
1190: Connection to the client that is making this request We got the
1191: request from this socket, and when the request has been relayed to
1192: lond and we get a reply back from lond it will get sent to this
1193: socket.
1194:
1195: =item $Request
1196:
1197: The text of the request to send.
1198:
1199: =cut
1200:
1201: sub StartRequest {
1202:
1203: my ($Lond, $Request) = @_;
1204:
1205: Debug(6, "StartRequest: ".$Request->getRequest());
1206:
1207: my $Socket = $Lond->GetSocket();
1208:
1209: $Request->Activate($Lond);
1210: $ActiveTransactions{$Lond} = $Request;
1211:
1212: $Lond->InitiateTransaction($Request->getRequest());
1213: my $event = Event->io(fd => $Socket,
1214: poll => "w",
1215: cb => \&LondWritable,
1216: data => $Lond,
1217: desc => "lond transaction connection");
1218: $ActiveConnections{$Lond} = $event;
1219: Debug(8," Start Request made watcher data with ".$event->data."\n");
1220: }
1221:
1222: =pod
1223:
1224: =head2 QueueTransaction
1225:
1226: If there is an idle lond connection, it is put to work doing this
1227: transaction. Otherwise, the transaction is placed in the work queue.
1228: If placed in the work queue and the maximum number of connections has
1229: not yet been created, a new connection will be started. Our goal is
1230: to eventually have a sufficient number of connections that the work
1231: queue will typically be empty. parameters are:
1232:
1233: =item Socket
1234:
1235: open on the lonc client.
1236:
1237: =item Request
1238:
1239: data to send to the lond.
1240:
1241: =cut
1242:
1243: sub QueueTransaction {
1244:
1245: my $requestData = shift; # This is a LondTransaction.
1246: my $cmd = $requestData->getRequest();
1247:
1248: Debug(6,"QueueTransaction: ".$cmd);
1249:
1250: my $LondSocket = $IdleConnections->pop();
1251: if(!defined $LondSocket) { # Need to queue request.
1252: Debug(5,"Must queue...");
1253: $WorkQueue->enqueue($requestData);
1254: Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
1255: if(($ConnectionCount < $MaxConnectionCount) && (! $LondConnecting)) {
1256:
1257: if($ConnectionRetriesLeft > 0) {
1258: Debug(5,"Starting additional lond connection");
1259: if(&MakeLondConnection() == 0) {
1260: EmptyQueue(); # Fail transactions, can't make connection.
1261: CloseAllLondConnections; # Should all be closed but...
1262: }
1263: } else {
1264: ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1265: $LondConnecting = 0;
1266: EmptyQueue(); # It's worse than that ... he's dead Jim.
1267: CloseAllLondConnections; # Should all be closed but..
1268: }
1269: }
1270: } else { # Can start the request:
1271: Debug(8,"Can start...");
1272: StartRequest($LondSocket, $requestData);
1273: }
1274: }
1275:
1276: #-------------------------- Lonc UNIX socket handling ---------------------
1277:
1278: =pod
1279:
1280: =head2 ClientRequest
1281: Callback that is called when data can be read from the UNIX domain
1282: socket connecting us with an apache server process.
1283:
1284: =cut
1285:
1286: sub ClientRequest {
1287: Debug(6, "ClientRequest");
1288: my $event = shift;
1289: my $watcher = $event->w;
1290: my $socket = $watcher->fd;
1291: my $data = $watcher->data;
1292: my $thisread;
1293:
1294: Debug(9, " Watcher named: ".$watcher->desc);
1295:
1296: my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
1297: Debug(8, "rcv: data length = ".length($thisread)
1298: ." read =".$thisread);
1299: unless (defined $rv && length($thisread)) {
1300: # Likely eof on socket.
1301: Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
1302: close($socket);
1303: $watcher->cancel();
1304: delete($ActiveClients{$socket});
1305: return;
1306: }
1307: Debug(8,"Data: ".$data." this read: ".$thisread);
1308: $data = $data.$thisread; # Append new data.
1309: $watcher->data($data);
1310: if($data =~ /\n$/) { # Request entirely read.
1311: if($data eq "close_connection_exit\n") {
1312: Log("CRITICAL",
1313: "Request Close Connection ... exiting");
1314: CloseAllLondConnections();
1315: exit;
1316: }
1317: Debug(8, "Complete transaction received: ".$data);
1318: if($LogTransactions) {
1319: Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
1320: }
1321: my $Transaction = LondTransaction->new($data);
1322: $Transaction->SetClient($socket);
1323: QueueTransaction($Transaction);
1324: $watcher->cancel(); # Done looking for input data.
1325: }
1326:
1327: }
1328:
1329:
1330: =pod
1331:
1332: =head2 NewClient
1333:
1334: Callback that is called when a connection is received on the unix
1335: socket for a new client of lonc. The callback is parameterized by the
1336: event.. which is a-priori assumed to be an io event, and therefore has
1337: an fd member that is the Listener socket. We Accept the connection
1338: and register a new event on the readability of that socket:
1339:
1340: =cut
1341:
1342: sub NewClient {
1343: Debug(6, "NewClient");
1344: my $event = shift; # Get the event parameters.
1345: my $watcher = $event->w;
1346: my $socket = $watcher->fd; # Get the event' socket.
1347: my $connection = $socket->accept(); # Accept the client connection.
1348: Debug(8,"Connection request accepted from "
1349: .GetPeername($connection, AF_UNIX));
1350:
1351:
1352: my $description = sprintf("Connection to lonc client %d",
1353: $ClientConnection);
1354: Debug(9, "Creating event named: ".$description);
1355: Event->io(cb => \&ClientRequest,
1356: poll => 'r',
1357: desc => $description,
1358: data => "",
1359: fd => $connection);
1360: $ActiveClients{$connection} = $ClientConnection;
1361: $ClientConnection++;
1362: }
1363:
1364: =pod
1365:
1366: =head2 GetLoncSocketPath
1367:
1368: Returns the name of the UNIX socket on which to listen for client
1369: connections.
1370:
1371: =cut
1372:
1373: sub GetLoncSocketPath {
1374: return $UnixSocketDir."/".GetServerHost();
1375: }
1376:
1377: =pod
1378:
1379: =head2 GetServerHost
1380:
1381: Returns the host whose lond we talk with.
1382:
1383: =cut
1384:
1385: sub GetServerHost {
1386: return $RemoteHost; # Setup by the fork.
1387: }
1388:
1389: =pod
1390:
1391: =head2 GetServerPort
1392:
1393: Returns the lond port number.
1394:
1395: =cut
1396:
1397: sub GetServerPort {
1398: return $perlvar{londPort};
1399: }
1400:
1401: =pod
1402:
1403: =head2 SetupLoncListener
1404:
1405: Setup a lonc listener event. The event is called when the socket
1406: becomes readable.. that corresponds to the receipt of a new
1407: connection. The event handler established will accept the connection
1408: (creating a communcations channel), that int turn will establish
1409: another event handler to subess requests.
1410:
1411: =cut
1412:
1413: sub SetupLoncListener {
1414:
1415: my $socket;
1416: my $SocketName = GetLoncSocketPath();
1417: unlink($SocketName);
1418: unless ($socket =IO::Socket::UNIX->new(Local => $SocketName,
1419: Listen => 250,
1420: Type => SOCK_STREAM)) {
1421: die "Failed to create a lonc listner socket";
1422: }
1423: Event->io(cb => \&NewClient,
1424: poll => 'r',
1425: desc => 'Lonc listener Unix Socket',
1426: fd => $socket);
1427: }
1428:
1429: #
1430: # Toggle transaction logging.
1431: # Implicit inputs:
1432: # LogTransactions
1433: # Implicit Outputs:
1434: # LogTransactions
1435: sub ToggleTransactionLogging {
1436: print STDERR "Toggle transaction logging...\n";
1437: if(!$LogTransactions) {
1438: $LogTransactions = 1;
1439: } else {
1440: $LogTransactions = 0;
1441: }
1442:
1443:
1444: Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
1445: }
1446:
1447: =pod
1448:
1449: =head2 ChildStatus
1450:
1451: Child USR1 signal handler to report the most recent status
1452: into the status file.
1453:
1454: We also use this to reset the retries count in order to allow the
1455: client to retry connections with a previously dead server.
1456: =cut
1457:
1458: sub ChildStatus {
1459: my $event = shift;
1460: my $watcher = $event->w;
1461:
1462: Debug(2, "Reporting child status because : ".$watcher->data);
1463: my $docdir = $perlvar{'lonDocRoot'};
1464: my $fh = IO::File->new(">>$docdir/lon-status/loncstatus.txt");
1465: print $fh $$."\t".$RemoteHost."\t".$Status."\t".
1466: $RecentLogEntry."\n";
1467: #
1468: # Write out information about each of the connections:
1469: #
1470: if ($DebugLevel > 2) {
1471: print $fh "Active connection statuses: \n";
1472: my $i = 1;
1473: print STDERR "================================= Socket Status Dump:\n";
1474: foreach my $item (keys %ActiveConnections) {
1475: my $Socket = $ActiveConnections{$item}->data;
1476: my $state = $Socket->GetState();
1477: print $fh "Connection $i State: $state\n";
1478: print STDERR "---------------------- Connection $i \n";
1479: $Socket->Dump(-1); # Ensure it gets dumped..
1480: $i++;
1481: }
1482: }
1483: $ConnectionRetriesLeft = $ConnectionRetries;
1484: }
1485:
1486: =pod
1487:
1488: =head2 SignalledToDeath
1489:
1490: Called in response to a signal that causes a chid process to die.
1491:
1492: =cut
1493:
1494:
1495: sub SignalledToDeath {
1496: my $event = shift;
1497: my $watcher= $event->w;
1498:
1499: Debug(2,"Signalled to death! via ".$watcher->data);
1500: my ($signal) = $watcher->data;
1501: chomp($signal);
1502: Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost "
1503: ."died through "."\"$signal\"");
1504: LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
1505: ."\"$signal\"");
1506: exit 0;
1507:
1508: }
1509:
1510: =head2 ToggleDebug
1511:
1512: This sub toggles trace debugging on and off.
1513:
1514: =cut
1515:
1516: sub ToggleDebug {
1517: my $Current = $DebugLevel;
1518: $DebugLevel = $NextDebugLevel;
1519: $NextDebugLevel = $Current;
1520:
1521: Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
1522:
1523: }
1524:
1525: =head2 ChildProcess
1526:
1527: This sub implements a child process for a single lonc daemon.
1528:
1529: =cut
1530:
1531: sub ChildProcess {
1532:
1533:
1534: #
1535: # Signals must be handled by the Event framework...
1536: #
1537:
1538: Event->signal(signal => "QUIT",
1539: cb => \&SignalledToDeath,
1540: data => "QUIT");
1541: Event->signal(signal => "HUP",
1542: cb => \&ChildStatus,
1543: data => "HUP");
1544: Event->signal(signal => "USR1",
1545: cb => \&ChildStatus,
1546: data => "USR1");
1547: Event->signal(signal => "USR2",
1548: cb => \&ToggleTransactionLogging);
1549: Event->signal(signal => "INT",
1550: cb => \&ToggleDebug,
1551: data => "INT");
1552:
1553:
1554: SetupLoncListener();
1555:
1556: $Event::Debuglevel = $DebugLevel;
1557:
1558: Debug(9, "Making initial lond connection for ".$RemoteHost);
1559:
1560: # Setup the initial server connection:
1561:
1562: # &MakeLondConnection(); // let first work requirest do it.
1563:
1564:
1565: Debug(9,"Entering event loop");
1566: my $ret = Event::loop(); # Start the main event loop.
1567:
1568:
1569: die "Main event loop exited!!!";
1570: }
1571:
1572: # Create a new child for host passed in:
1573:
1574: sub CreateChild {
1575: my $host = shift;
1576:
1577: my $sigset = POSIX::SigSet->new(SIGINT);
1578: sigprocmask(SIG_BLOCK, $sigset);
1579: $RemoteHost = $host;
1580: Log("CRITICAL", "Forking server for ".$host);
1581: my $pid = fork;
1582: if($pid) { # Parent
1583: $RemoteHost = "Parent";
1584: $ChildHash{$pid} = $host;
1585: $HostToPid{$host}= $pid;
1586: sigprocmask(SIG_UNBLOCK, $sigset);
1587:
1588: } else { # child.
1589: ShowStatus("Connected to ".$RemoteHost);
1590: $SIG{INT} = 'DEFAULT';
1591: sigprocmask(SIG_UNBLOCK, $sigset);
1592: ChildProcess; # Does not return.
1593: }
1594:
1595: }
1596: #
1597: # Parent process logic pass 1:
1598: # For each entry in the hosts table, we will
1599: # fork off an instance of ChildProcess to service the transactions
1600: # to that host. Each pid will be entered in a global hash
1601: # with the value of the key, the host.
1602: # The parent will then enter a loop to wait for process exits.
1603: # Each exit gets logged and the child gets restarted.
1604: #
1605:
1606: #
1607: # Fork and start in new session so hang-up isn't going to
1608: # happen without intent.
1609: #
1610:
1611:
1612:
1613:
1614:
1615:
1616: ShowStatus("Forming new session");
1617: my $childpid = fork;
1618: if ($childpid != 0) {
1619: sleep 4; # Give child a chacne to break to
1620: exit 0; # a new sesion.
1621: }
1622: #
1623: # Write my pid into the pid file so I can be located
1624: #
1625:
1626: ShowStatus("Parent writing pid file:");
1627: my $execdir = $perlvar{'lonDaemons'};
1628: open (PIDSAVE, ">$execdir/logs/lonc.pid");
1629: print PIDSAVE "$$\n";
1630: close(PIDSAVE);
1631:
1632:
1633:
1634: if (POSIX::setsid() < 0) {
1635: print "Could not create new session\n";
1636: exit -1;
1637: }
1638:
1639: ShowStatus("Forking node servers");
1640:
1641: Log("CRITICAL", "--------------- Starting children ---------------");
1642:
1643: LondConnection::ReadConfig; # Read standard config files.
1644: my $HostIterator = LondConnection::GetHostIterator;
1645: while (! $HostIterator->end()) {
1646:
1647: my $hostentryref = $HostIterator->get();
1648: CreateChild($hostentryref->[0]);
1649: $HostHash{$hostentryref->[0]} = $hostentryref->[4];
1650: $HostIterator->next();
1651: }
1652: $RemoteHost = "Parent Server";
1653:
1654: # Maintain the population:
1655:
1656: ShowStatus("Parent keeping the flock");
1657:
1658: #
1659: # Set up parent signals:
1660: #
1661:
1662: $SIG{INT} = \&Terminate;
1663: $SIG{TERM} = \&Terminate;
1664: $SIG{HUP} = \&Restart;
1665: $SIG{USR1} = \&CheckKids;
1666: $SIG{USR2} = \&UpdateKids; # LonManage update request.
1667:
1668: while(1) {
1669: my $deadchild = wait();
1670: if(exists $ChildHash{$deadchild}) { # need to restart.
1671: my $deadhost = $ChildHash{$deadchild};
1672: delete($HostToPid{$deadhost});
1673: delete($ChildHash{$deadchild});
1674: Log("WARNING","Lost child pid= ".$deadchild.
1675: "Connected to host ".$deadhost);
1676: Log("INFO", "Restarting child procesing ".$deadhost);
1677: CreateChild($deadhost);
1678: }
1679: }
1680:
1681:
1682:
1683: =pod
1684:
1685: =head1 CheckKids
1686:
1687: Since kids do not die as easily in this implementation
1688: as the previous one, there is no need to restart the
1689: dead ones (all dead kids get restarted when they die!!)
1690: The only thing this function does is to pass USR1 to the
1691: kids so that they report their status.
1692:
1693: =cut
1694:
1695: sub CheckKids {
1696: Debug(2, "Checking status of children");
1697: my $docdir = $perlvar{'lonDocRoot'};
1698: my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
1699: my $now=time;
1700: my $local=localtime($now);
1701: print $fh "LONC status $local - parent $$ \n\n";
1702: foreach my $pid (keys %ChildHash) {
1703: Debug(2, "Sending USR1 -> $pid");
1704: kill 'USR1' => $pid; # Tell Child to report status.
1705: sleep 1; # Wait so file doesn't intermix.
1706: }
1707: }
1708:
1709: =pod
1710:
1711: =head1 UpdateKids
1712:
1713: parent's SIGUSR2 handler. This handler:
1714:
1715: =item
1716:
1717: Rereads the hosts file.
1718:
1719: =item
1720:
1721: Kills off (via sigint) children for hosts that have disappeared.
1722:
1723: =item
1724:
1725: QUITs children for hosts that already exist (this just forces a status display
1726: and resets the connection retry count for that host.
1727:
1728: =item
1729:
1730: Starts new children for hosts that have been added to the hosts.tab file since
1731: the start of the master program and maintains them.
1732:
1733: =cut
1734:
1735: sub UpdateKids {
1736:
1737: Log("INFO", "Updating connections via SIGUSR2");
1738:
1739: # Just in case we need to kill our own lonc, we wait a few seconds to
1740: # give it a chance to receive and relay lond's response to the
1741: # re-init command.
1742: #
1743:
1744: sleep(2); # Wait a couple of seconds.
1745:
1746: my %hosts; # Indexed by loncapa hostname, value=ip.
1747:
1748: # Need to re-read the host table:
1749:
1750:
1751: LondConnection::ReadConfig();
1752: my $I = LondConnection::GetHostIterator;
1753: while (! $I->end()) {
1754: my $item = $I->get();
1755: $hosts{$item->[0]} = $item->[4];
1756: $I->next();
1757: }
1758:
1759: # The logic below is written for clarity not for efficiency.
1760: # Since I anticipate that this function is only rarely called, that's
1761: # appropriate. There are certainly ways to combine the loops below,
1762: # and anyone wishing to obscure the logic is welcome to go for it.
1763: # Note that we don't re-direct sigchild. Instead we do what's needed
1764: # to the data structures that keep track of children to ensure that
1765: # when sigchild is honored, no new child is born.
1766: #
1767:
1768: # For each existing child; if it's host doesn't exist, kill the child.
1769:
1770: foreach my $child (keys %ChildHash) {
1771: my $oldhost = $ChildHash{$child};
1772: if (!(exists $hosts{$oldhost})) {
1773: Log("CRITICAL", "Killing child for $oldhost host no longer exists");
1774: delete $ChildHash{$child};
1775: delete $HostToPid{$oldhost};
1776: kill 'QUIT' => $child;
1777: }
1778: }
1779: # For each remaining existing child; if it's host's ip has changed,
1780: # Restart the child on the new IP.
1781:
1782: foreach my $child (keys %ChildHash) {
1783: my $oldhost = $ChildHash{$child};
1784: my $oldip = $HostHash{$oldhost};
1785: if ($hosts{$oldhost} ne $oldip) {
1786:
1787: # kill the old child.
1788:
1789: Log("CRITICAL", "Killing child for $oldhost host ip has changed...");
1790: delete $ChildHash{$child};
1791: delete $HostToPid{$oldhost};
1792: kill 'QUIT' => $child;
1793:
1794: # Do the book-keeping needed to start a new child on the
1795: # new ip.
1796:
1797: $HostHash{$oldhost} = $hosts{$oldhost};
1798: CreateChild($oldhost);
1799: }
1800: }
1801: # Finally, for each new host, not in the host hash, create a
1802: # enter the host and create a new child.
1803: # Force a status display of any existing process.
1804:
1805: foreach my $host (keys %hosts) {
1806: if(!(exists $HostHash{$host})) {
1807: Log("INFO", "New host $host discovered in hosts.tab...");
1808: $HostHash{$host} = $hosts{$host};
1809: CreateChild($host);
1810: } else {
1811: kill 'HUP' => $HostToPid{$host}; # status display.
1812: }
1813: }
1814: }
1815:
1816:
1817: =pod
1818:
1819: =head1 Restart
1820:
1821: Signal handler for HUP... all children are killed and
1822: we self restart. This is an el-cheapo way to re read
1823: the config file.
1824:
1825: =cut
1826:
1827: sub Restart {
1828: &KillThemAll; # First kill all the children.
1829: Log("CRITICAL", "Restarting");
1830: my $execdir = $perlvar{'lonDaemons'};
1831: unlink("$execdir/logs/lonc.pid");
1832: exec("$execdir/loncnew");
1833: }
1834:
1835: =pod
1836:
1837: =head1 KillThemAll
1838:
1839: Signal handler that kills all children by sending them a
1840: SIGHUP. Responds to sigint and sigterm.
1841:
1842: =cut
1843:
1844: sub KillThemAll {
1845: Debug(2, "Kill them all!!");
1846: local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die.
1847: foreach my $pid (keys %ChildHash) {
1848: my $serving = $ChildHash{$pid};
1849: ShowStatus("Nicely Killing lonc for $serving pid = $pid");
1850: Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
1851: kill 'QUIT' => $pid;
1852: }
1853:
1854:
1855: }
1856:
1857:
1858: #
1859: # Kill all children via KILL. Just in case the
1860: # first shot didn't get them.
1861:
1862: sub really_kill_them_all_dammit
1863: {
1864: Debug(2, "Kill them all Dammit");
1865: local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
1866: foreach my $pid (keys %ChildHash) {
1867: my $serving = $ChildHash{$pid};
1868: &ShowStatus("Nastily killing lonc for $serving pid = $pid");
1869: Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
1870: kill 'KILL' => $pid;
1871: delete($ChildHash{$pid});
1872: my $execdir = $perlvar{'lonDaemons'};
1873: unlink("$execdir/logs/lonc.pid");
1874: }
1875: }
1876: =pod
1877:
1878: =head1 Terminate
1879:
1880: Terminate the system.
1881:
1882: =cut
1883:
1884: sub Terminate {
1885: &Log("CRITICAL", "Asked to kill children.. first be nice...");
1886: &KillThemAll;
1887: #
1888: # By now they really should all be dead.. but just in case
1889: # send them all SIGKILL's after a bit of waiting:
1890:
1891: sleep(4);
1892: &Log("CRITICAL", "Now kill children nasty");
1893: &really_kill_them_all_dammit;
1894: Log("CRITICAL","Master process exiting");
1895: exit 0;
1896:
1897: }
1898: =pod
1899:
1900: =head1 Theory
1901:
1902: The event class is used to build this as a single process with an
1903: event driven model. The following events are handled:
1904:
1905: =item UNIX Socket connection Received
1906:
1907: =item Request data arrives on UNIX data transfer socket.
1908:
1909: =item lond connection becomes writable.
1910:
1911: =item timer fires at 1 second intervals.
1912:
1913: All sockets are run in non-blocking mode. Timeouts managed by the timer
1914: handler prevents hung connections.
1915:
1916: Key data structures:
1917:
1918: =item RequestQueue
1919:
1920: A queue of requests received from UNIX sockets that are
1921: waiting for a chance to be forwarded on a lond connection socket.
1922:
1923: =item ActiveConnections
1924:
1925: A hash of lond connections that have transactions in process that are
1926: available to be timed out.
1927:
1928: =item ActiveTransactions
1929:
1930: A hash indexed by lond connections that contain the client reply
1931: socket for each connection that has an active transaction on it.
1932:
1933: =item IdleConnections
1934:
1935: A hash of lond connections that have no work to do. These connections
1936: can be closed if they are idle for a long enough time.
1937:
1938: =cut
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>