Annotation of loncom/loncnew, revision 1.91

1.1       foxr        1: #!/usr/bin/perl
1.2       albertel    2: # The LearningOnline Network with CAPA
                      3: # lonc maintains the connections to remote computers
                      4: #
1.91    ! foxr        5: # $Id: loncnew,v 1.90 2010/12/20 11:15:51 foxr Exp $
1.2       albertel    6: #
                      7: # Copyright Michigan State University Board of Trustees
                      8: #
                      9: # This file is part of the LearningOnline Network with CAPA (LON-CAPA).
1.17      foxr       10: ## LON-CAPA is free software; you can redistribute it and/or modify
1.2       albertel   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: #
1.1       foxr       28: #
1.15      foxr       29: # new lonc handles n request out bver m connections to londs.
1.1       foxr       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).
1.7       foxr       36: #    - Add general logic for dispatching requests and timeouts. (done).
                     37: #    - Add support for the lonc/lond requests.          (done).
1.38      foxr       38: #    - Add logging/status monitoring.                    (done)
                     39: #    - Add Signal handling - HUP restarts. USR1 status report. (done)
1.7       foxr       40: #    - Add Configuration file I/O                       (done).
1.38      foxr       41: #    - Add management/status request interface.         (done)
1.8       foxr       42: #    - Add deferred request capability.                  (done)
1.38      foxr       43: #    - Detect transmission timeouts.                     (done)
1.7       foxr       44: #
                     45: 
1.23      foxr       46: use strict;
1.1       foxr       47: use lib "/home/httpd/lib/perl/";
                     48: use Event qw(:DEFAULT );
                     49: use POSIX qw(:signal_h);
1.12      foxr       50: use POSIX;
1.1       foxr       51: use IO::Socket;
                     52: use IO::Socket::INET;
                     53: use IO::Socket::UNIX;
1.9       foxr       54: use IO::File;
1.6       foxr       55: use IO::Handle;
1.1       foxr       56: use Socket;
                     57: use Crypt::IDEA;
                     58: use LONCAPA::Queue;
                     59: use LONCAPA::Stack;
                     60: use LONCAPA::LondConnection;
1.7       foxr       61: use LONCAPA::LondTransaction;
1.1       foxr       62: use LONCAPA::Configuration;
1.67      albertel   63: use Fcntl qw(:flock);
1.1       foxr       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: 
1.83      albertel   75: my %ChildPid;			# by pid -> host.
                     76: my %ChildHost;			# by host.
1.62      foxr       77: my %listening_to;		# Socket->host table for who the parent
                     78:                                 # is listening to.
                     79: my %parent_dispatchers;         # host-> listener watcher events. 
1.1       foxr       80: 
1.65      foxr       81: my %parent_handlers;		# Parent signal handlers...
                     82: 
1.9       foxr       83: my $MaxConnectionCount = 10;	# Will get from config later.
1.1       foxr       84: my $ClientConnection = 0;	# Uniquifier for client events.
                     85: 
1.9       foxr       86: my $DebugLevel = 0;
1.29      foxr       87: my $NextDebugLevel= 2;		# So Sigint can toggle this.
1.89      www        88: my $IdleTimeout= 7200;		# Wait two hours before pruning connections.
1.1       foxr       89: 
1.39      foxr       90: my $LogTransactions = 0;	# When True, all transactions/replies get logged.
1.65      foxr       91: my $executable      = $0;	# Get the full path to me.
1.39      foxr       92: 
1.1       foxr       93: #
                     94: #  The variables below are only used by the child processes.
                     95: #
                     96: my $RemoteHost;			# Name of host child is talking to.
1.81      albertel   97: my $RemoteHostId;		# default lonid of host child is talking to.
1.82      albertel   98: my @all_host_ids;
1.20      albertel   99: my $UnixSocketDir= $perlvar{'lonSockDir'};
1.1       foxr      100: my $IdleConnections = Stack->new(); # Set of idle connections
                    101: my %ActiveConnections;		# Connections to the remote lond.
1.7       foxr      102: my %ActiveTransactions;		# LondTransactions in flight.
1.1       foxr      103: my %ActiveClients;		# Serial numbers of active clients by socket.
                    104: my $WorkQueue       = Queue->new(); # Queue of pending transactions.
                    105: my $ConnectionCount = 0;
1.4       foxr      106: my $IdleSeconds     = 0;	# Number of seconds idle.
1.9       foxr      107: my $Status          = "";	# Current status string.
1.14      foxr      108: my $RecentLogEntry  = "";
1.72      albertel  109: my $ConnectionRetries=5;	# Number of connection retries allowed.
                    110: my $ConnectionRetriesLeft=5;	# Number of connection retries remaining.
1.40      foxr      111: my $LondVersion     = "unknown"; # Version of lond we talk with.
1.49      foxr      112: my $KeyMode         = "";       # e.g. ssl, local, insecure from last connect.
1.54      foxr      113: my $LondConnecting  = 0;       # True when a connection is being built.
1.1       foxr      114: 
1.60      foxr      115: 
                    116: 
1.62      foxr      117: my $I_am_child      = 0;	# True if this is the child process.
1.57      foxr      118: 
1.1       foxr      119: #
1.9       foxr      120: #   The hash below gives the HTML format for log messages
                    121: #   given a severity.
                    122: #    
                    123: my %LogFormats;
                    124: 
1.45      albertel  125: $LogFormats{"CRITICAL"} = "<font color='red'>CRITICAL: %s</font>";
                    126: $LogFormats{"SUCCESS"}  = "<font color='green'>SUCCESS: %s</font>";
                    127: $LogFormats{"INFO"}     = "<font color='yellow'>INFO: %s</font>";
                    128: $LogFormats{"WARNING"}  = "<font color='blue'>WARNING: %s</font>";
1.9       foxr      129: $LogFormats{"DEFAULT"}  = " %s ";
                    130: 
1.10      foxr      131: 
1.57      foxr      132: #  UpdateStatus;
                    133: #    Update the idle status display to show how many connections
                    134: #    are left, retries and other stuff.
                    135: #
                    136: sub UpdateStatus {
                    137:     if ($ConnectionRetriesLeft > 0) {
                    138: 	ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount
                    139: 		   ." Retries remaining: ".$ConnectionRetriesLeft
                    140: 		   ." ($KeyMode)");
                    141:     } else {
                    142: 	ShowStatus(GetServerHost()." >> DEAD <<");
                    143:     }
                    144: }
                    145: 
1.10      foxr      146: 
                    147: =pod
                    148: 
                    149: =head2 LogPerm
                    150: 
                    151: Makes an entry into the permanent log file.
                    152: 
                    153: =cut
1.69      matthew   154: 
1.10      foxr      155: sub LogPerm {
                    156:     my $message=shift;
                    157:     my $execdir=$perlvar{'lonDaemons'};
                    158:     my $now=time;
                    159:     my $local=localtime($now);
                    160:     my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
1.86      albertel  161:     chomp($message);
1.10      foxr      162:     print $fh "$now:$message:$local\n";
                    163: }
1.9       foxr      164: 
                    165: =pod
                    166: 
                    167: =head2 Log
                    168: 
                    169: Logs a message to the log file.
                    170: Parameters:
                    171: 
                    172: =item severity
                    173: 
                    174: One of CRITICAL, WARNING, INFO, SUCCESS used to select the
                    175: format string used to format the message.  if the severity is
                    176: not a defined severity the Default format string is used.
                    177: 
                    178: =item message
                    179: 
                    180: The base message.  In addtion to the format string, the message
                    181: will be appended to a string containing the name of our remote
                    182: host and the time will be formatted into the message.
                    183: 
                    184: =cut
                    185: 
                    186: sub Log {
1.47      foxr      187: 
                    188:     my ($severity, $message) = @_;
                    189: 
1.9       foxr      190:     if(!$LogFormats{$severity}) {
                    191: 	$severity = "DEFAULT";
                    192:     }
                    193: 
                    194:     my $format = $LogFormats{$severity};
                    195:     
                    196:     #  Put the window dressing in in front of the message format:
                    197: 
                    198:     my $now   = time;
                    199:     my $local = localtime($now);
                    200:     my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
1.76      albertel  201:     $finalformat = $finalformat.$format."\n";
1.9       foxr      202: 
                    203:     # open the file and put the result.
                    204: 
                    205:     my $execdir = $perlvar{'lonDaemons'};
                    206:     my $fh      = IO::File->new(">>$execdir/logs/lonc.log");
                    207:     my $msg = sprintf($finalformat, $message);
1.14      foxr      208:     $RecentLogEntry = $msg;
1.9       foxr      209:     print $fh $msg;
                    210:     
1.10      foxr      211:     
1.9       foxr      212: }
1.6       foxr      213: 
1.3       albertel  214: 
1.1       foxr      215: =pod
1.3       albertel  216: 
                    217: =head2 GetPeerName
                    218: 
                    219: Returns the name of the host that a socket object is connected to.
                    220: 
1.1       foxr      221: =cut
                    222: 
                    223: sub GetPeername {
1.47      foxr      224: 
                    225: 
                    226:     my ($connection, $AdrFamily) = @_;
                    227: 
1.1       foxr      228:     my $peer       = $connection->peername();
                    229:     my $peerport;
                    230:     my $peerip;
                    231:     if($AdrFamily == AF_INET) {
                    232: 	($peerport, $peerip) = sockaddr_in($peer);
1.23      foxr      233: 	my $peername    = gethostbyaddr($peerip, $AdrFamily);
1.1       foxr      234: 	return $peername;
                    235:     } elsif ($AdrFamily == AF_UNIX) {
                    236: 	my $peerfile;
                    237: 	($peerfile) = sockaddr_un($peer);
                    238: 	return $peerfile;
                    239:     }
                    240: }
                    241: =pod
1.3       albertel  242: 
1.1       foxr      243: =head2 Debug
1.3       albertel  244: 
                    245: Invoked to issue a debug message.
                    246: 
1.1       foxr      247: =cut
1.3       albertel  248: 
1.1       foxr      249: sub Debug {
1.47      foxr      250: 
                    251:     my ($level, $message) = @_;
                    252: 
1.1       foxr      253:     if ($level <= $DebugLevel) {
1.23      foxr      254: 	Log("INFO", "-Debug- $message host = $RemoteHost");
1.1       foxr      255:     }
                    256: }
                    257: 
                    258: sub SocketDump {
1.47      foxr      259: 
                    260:     my ($level, $socket) = @_;
                    261: 
1.1       foxr      262:     if($level <= $DebugLevel) {
1.48      foxr      263: 	$socket->Dump(-1);	# Ensure it will get dumped.
1.1       foxr      264:     }
                    265: }
1.3       albertel  266: 
1.1       foxr      267: =pod
1.3       albertel  268: 
1.5       foxr      269: =head2 ShowStatus
                    270: 
                    271:  Place some text as our pid status.
1.10      foxr      272:  and as what we return in a SIGUSR1
1.5       foxr      273: 
                    274: =cut
1.69      matthew   275: 
1.5       foxr      276: sub ShowStatus {
1.10      foxr      277:     my $state = shift;
                    278:     my $now = time;
                    279:     my $local = localtime($now);
                    280:     $Status   = $local.": ".$state;
                    281:     $0='lonc: '.$state.' '.$local;
1.5       foxr      282: }
                    283: 
                    284: =pod
                    285: 
1.69      matthew   286: =head2 SocketTimeout
1.15      foxr      287: 
                    288:     Called when an action on the socket times out.  The socket is 
                    289:    destroyed and any active transaction is failed.
                    290: 
                    291: 
                    292: =cut
1.69      matthew   293: 
1.15      foxr      294: sub SocketTimeout {
                    295:     my $Socket = shift;
1.38      foxr      296:     Log("WARNING", "A socket timeout was detected");
1.52      foxr      297:     Debug(5, " SocketTimeout called: ");
1.48      foxr      298:     $Socket->Dump(0);
1.42      foxr      299:     if(exists($ActiveTransactions{$Socket})) {
1.43      albertel  300: 	FailTransaction($ActiveTransactions{$Socket});
1.42      foxr      301:     }
1.22      foxr      302:     KillSocket($Socket);	# A transaction timeout also counts as
                    303:                                 # a connection failure:
                    304:     $ConnectionRetriesLeft--;
1.42      foxr      305:     if($ConnectionRetriesLeft <= 0) {
1.52      foxr      306: 	Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
1.56      foxr      307: 	$LondConnecting = 0;
1.42      foxr      308:     }
                    309: 
1.15      foxr      310: }
1.80      albertel  311: 
1.64      foxr      312: #
                    313: #   This function should be called by the child in all cases where it must
1.80      albertel  314: #   exit.  The child process must create a lock file for the AF_UNIX socket
                    315: #   in order to prevent connection requests from lonnet in the time between
                    316: #   process exit and the parent picking up the listen again.
                    317: #
1.64      foxr      318: # Parameters:
                    319: #     exit_code           - Exit status value, however see the next parameter.
                    320: #     message             - If this optional parameter is supplied, the exit
                    321: #                           is via a die with this message.
                    322: #
                    323: sub child_exit {
                    324:     my ($exit_code, $message) = @_;
                    325: 
                    326:     # Regardless of how we exit, we may need to do the lock thing:
                    327: 
1.80      albertel  328:     #
                    329:     #  Create a lock file since there will be a time window
                    330:     #  between our exit and the parent's picking up the listen
                    331:     #  during which no listens will be done on the
                    332:     #  lonnet client socket.
                    333:     #
                    334:     my $lock_file = &GetLoncSocketPath().".lock";
                    335:     open(LOCK,">$lock_file");
                    336:     print LOCK "Contents not important";
                    337:     close(LOCK);
1.81      albertel  338:     unlink(&GetLoncSocketPath());
1.64      foxr      339: 
1.80      albertel  340:     if ($message) {
                    341: 	die($message);
1.64      foxr      342:     } else {
                    343: 	exit($exit_code);
                    344:     }
                    345: }
1.35      foxr      346: #----------------------------- Timer management ------------------------
1.15      foxr      347: 
                    348: =pod
                    349: 
1.1       foxr      350: =head2 Tick
1.3       albertel  351: 
                    352: Invoked  each timer tick.
                    353: 
1.1       foxr      354: =cut
                    355: 
1.5       foxr      356: 
1.1       foxr      357: sub Tick {
1.52      foxr      358:     my ($Event)       = @_;
                    359:     my $clock_watcher = $Event->w;
                    360: 
1.1       foxr      361:     my $client;
1.57      foxr      362:     UpdateStatus();
                    363: 
1.4       foxr      364:     # Is it time to prune connection count:
                    365: 
                    366: 
                    367:     if($IdleConnections->Count()  && 
                    368:        ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
1.52      foxr      369: 	$IdleSeconds++;
1.4       foxr      370: 	if($IdleSeconds > $IdleTimeout) { # Prune a connection...
1.23      foxr      371: 	    my $Socket = $IdleConnections->pop();
1.6       foxr      372: 	    KillSocket($Socket);
1.54      foxr      373: 	    $IdleSeconds = 0;	# Otherwise all connections get trimmed to fast.
1.57      foxr      374: 	    UpdateStatus();
1.80      albertel  375: 	    if(($ConnectionCount == 0)) {
1.64      foxr      376: 		&child_exit(0);
                    377: 
1.57      foxr      378: 	    }
1.4       foxr      379: 	}
                    380:     } else {
                    381: 	$IdleSeconds = 0;	# Reset idle count if not idle.
                    382:     }
1.15      foxr      383:     #
                    384:     #  For each inflight transaction, tick down its timeout counter.
                    385:     #
1.35      foxr      386: 
1.34      albertel  387:     foreach my $item (keys %ActiveConnections) {
                    388: 	my $State = $ActiveConnections{$item}->data->GetState();
1.35      foxr      389: 	if ($State ne 'Idle') {
1.34      albertel  390: 	    Debug(5,"Ticking Socket $State $item");
                    391: 	    $ActiveConnections{$item}->data->Tick();
                    392: 	}
1.15      foxr      393:     }
1.5       foxr      394:     # Do we have work in the queue, but no connections to service them?
                    395:     # If so, try to make some new connections to get things going again.
                    396:     #
1.57      foxr      397:     #   Note this code is dead now...
                    398:     #
1.5       foxr      399:     my $Requests = $WorkQueue->Count();
1.56      foxr      400:     if (($ConnectionCount == 0)  && ($Requests > 0) && (!$LondConnecting)) { 
1.10      foxr      401: 	if ($ConnectionRetriesLeft > 0) {
1.56      foxr      402: 	    Debug(5,"Work but no connections, Make a new one");
                    403: 	    my $success;
                    404: 	    $success    = &MakeLondConnection;
                    405: 	    if($success == 0) { # All connections failed:
1.29      foxr      406: 		Debug(5,"Work in queue failed to make any connectiouns\n");
1.22      foxr      407: 		EmptyQueue();	# Fail pending transactions with con_lost.
1.42      foxr      408: 		CloseAllLondConnections(); # Should all be closed but....
1.10      foxr      409: 	    }
                    410: 	} else {
1.56      foxr      411: 	    $LondConnecting = 0;
1.22      foxr      412: 	    ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
1.29      foxr      413: 	    Debug(5,"Work in queue, but gave up on connections..flushing\n");
1.10      foxr      414: 	    EmptyQueue();	# Connections can't be established.
1.42      foxr      415: 	    CloseAllLondConnections(); # Should all already be closed but...
1.5       foxr      416: 	}
                    417:        
                    418:     }
1.49      foxr      419:     if ($ConnectionCount == 0) {
                    420: 	$KeyMode = ""; 
1.52      foxr      421: 	$clock_watcher->cancel();
1.49      foxr      422:     }
1.66      albertel  423:     &UpdateStatus();
1.1       foxr      424: }
                    425: 
                    426: =pod
1.3       albertel  427: 
1.1       foxr      428: =head2 SetupTimer
                    429: 
1.3       albertel  430: Sets up a 1 per sec recurring timer event.  The event handler is used to:
1.1       foxr      431: 
1.3       albertel  432: =item
                    433: 
                    434: Trigger timeouts on communications along active sockets.
                    435: 
                    436: =item
                    437: 
                    438: Trigger disconnections of idle sockets.
1.1       foxr      439: 
                    440: =cut
                    441: 
                    442: sub SetupTimer {
1.52      foxr      443:     Debug(6, "SetupTimer");
                    444:     Event->timer(interval => 1, cb => \&Tick );
1.1       foxr      445: }
1.3       albertel  446: 
1.1       foxr      447: =pod
1.3       albertel  448: 
1.1       foxr      449: =head2 ServerToIdle
1.3       albertel  450: 
                    451: This function is called when a connection to the server is
                    452: ready for more work.
                    453: 
                    454: If there is work in the Work queue the top element is dequeued
1.1       foxr      455: and the connection will start to work on it.  If the work queue is
                    456: empty, the connection is pushed on the idle connection stack where
                    457: it will either get another work unit, or alternatively, if it sits there
                    458: long enough, it will be shut down and released.
                    459: 
1.3       albertel  460: =cut
1.1       foxr      461: 
                    462: sub ServerToIdle {
                    463:     my $Socket   = shift;	# Get the socket.
1.49      foxr      464:     $KeyMode = $Socket->{AuthenticationMode};
1.7       foxr      465:     delete($ActiveTransactions{$Socket}); # Server has no transaction
1.1       foxr      466: 
1.29      foxr      467:     &Debug(5, "Server to idle");
1.1       foxr      468: 
                    469:     #  If there's work to do, start the transaction:
                    470: 
1.23      foxr      471:     my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
1.29      foxr      472:     if ($reqdata ne undef)  {
                    473: 	Debug(5, "Queue gave request data: ".$reqdata->getRequest());
1.7       foxr      474: 	&StartRequest($Socket,  $reqdata);
1.8       foxr      475: 
1.1       foxr      476:     } else {
                    477: 	
                    478:     #  There's no work waiting, so push the server to idle list.
1.29      foxr      479: 	&Debug(5, "No new work requests, server connection going idle");
1.1       foxr      480: 	$IdleConnections->push($Socket);
                    481:     }
                    482: }
1.3       albertel  483: 
1.1       foxr      484: =pod
1.3       albertel  485: 
1.1       foxr      486: =head2 ClientWritable
1.3       albertel  487: 
                    488: Event callback for when a client socket is writable.
                    489: 
                    490: This callback is established when a transaction reponse is
                    491: avaiable from lond.  The response is forwarded to the unix socket
                    492: as it becomes writable in this sub.
                    493: 
1.1       foxr      494: Parameters:
                    495: 
1.3       albertel  496: =item Event
                    497: 
                    498: The event that has been triggered. Event->w->data is
                    499: the data and Event->w->fd is the socket to write.
1.1       foxr      500: 
                    501: =cut
1.3       albertel  502: 
1.1       foxr      503: sub ClientWritable {
                    504:     my $Event    = shift;
                    505:     my $Watcher  = $Event->w;
1.84      albertel  506:     if (!defined($Watcher)) {
                    507: 	&child_exit(-1,'No watcher for event in ClientWritable');
                    508:     }
1.1       foxr      509:     my $Data     = $Watcher->data;
                    510:     my $Socket   = $Watcher->fd;
                    511: 
                    512:     # Try to send the data:
                    513: 
                    514:     &Debug(6, "ClientWritable writing".$Data);
                    515:     &Debug(9, "Socket is: ".$Socket);
                    516: 
1.6       foxr      517:     if($Socket->connected) {
                    518: 	my $result = $Socket->send($Data, 0);
                    519: 	
                    520: 	# $result undefined: the write failed.
                    521: 	# otherwise $result is the number of bytes written.
                    522: 	# Remove that preceding string from the data.
                    523: 	# If the resulting data is empty, destroy the watcher
                    524: 	# and set up a read event handler to accept the next
                    525: 	# request.
                    526: 	
                    527: 	&Debug(9,"Send result is ".$result." Defined: ".defined($result));
1.29      foxr      528: 	if($result ne undef) {
1.6       foxr      529: 	    &Debug(9, "send result was defined");
                    530: 	    if($result == length($Data)) { # Entire string sent.
                    531: 		&Debug(9, "ClientWritable data all written");
                    532: 		$Watcher->cancel();
                    533: 		#
                    534: 		#  Set up to read next request from socket:
                    535: 		
                    536: 		my $descr     = sprintf("Connection to lonc client %d",
                    537: 					$ActiveClients{$Socket});
                    538: 		Event->io(cb    => \&ClientRequest,
                    539: 			  poll  => 'r',
                    540: 			  desc  => $descr,
                    541: 			  data  => "",
                    542: 			  fd    => $Socket);
                    543: 		
                    544: 	    } else {		# Partial string sent.
                    545: 		$Watcher->data(substr($Data, $result));
1.15      foxr      546: 		if($result == 0) {    # client hung up on us!!
1.52      foxr      547: 		    # Log("INFO", "lonc pipe client hung up on us!");
1.15      foxr      548: 		    $Watcher->cancel;
                    549: 		    $Socket->shutdown(2);
                    550: 		    $Socket->close();
                    551: 		}
1.6       foxr      552: 	    }
                    553: 	    
                    554: 	} else {			# Error of some sort...
                    555: 	    
                    556: 	    # Some errnos are possible:
                    557: 	    my $errno = $!;
                    558: 	    if($errno == POSIX::EWOULDBLOCK   ||
                    559: 	       $errno == POSIX::EAGAIN        ||
                    560: 	       $errno == POSIX::EINTR) {
                    561: 		# No action taken?
                    562: 	    } else {		# Unanticipated errno.
                    563: 		&Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
                    564: 		$Watcher->cancel;	# Stop the watcher.
                    565: 		$Socket->shutdown(2); # Kill connection
                    566: 		$Socket->close();	# Close the socket.
                    567: 	    }
1.1       foxr      568: 	    
                    569: 	}
1.6       foxr      570:     } else {
                    571: 	$Watcher->cancel();	# A delayed request...just cancel.
1.84      albertel  572: 	return;
1.1       foxr      573:     }
                    574: }
                    575: 
                    576: =pod
1.3       albertel  577: 
1.1       foxr      578: =head2 CompleteTransaction
1.3       albertel  579: 
                    580: Called when the reply data has been received for a lond 
1.1       foxr      581: transaction.   The reply data must now be sent to the
                    582: ultimate client on the other end of the Unix socket.  This is
                    583: done by setting up a writable event for the socket with the
                    584: data the reply data.
1.3       albertel  585: 
1.1       foxr      586: Parameters:
1.3       albertel  587: 
                    588: =item Socket
                    589: 
                    590: Socket on which the lond transaction occured.  This is a
                    591: LondConnection. The data received is in the TransactionReply member.
                    592: 
1.7       foxr      593: =item Transaction
1.3       albertel  594: 
1.7       foxr      595: The transaction that is being completed.
1.1       foxr      596: 
                    597: =cut
1.3       albertel  598: 
1.1       foxr      599: sub CompleteTransaction {
1.29      foxr      600:     &Debug(5,"Complete transaction");
1.47      foxr      601: 
                    602:     my ($Socket, $Transaction) = @_;
1.1       foxr      603: 
1.7       foxr      604:     if (!$Transaction->isDeferred()) { # Normal transaction
                    605: 	my $data   = $Socket->GetReply(); # Data to send.
1.39      foxr      606: 	if($LogTransactions) {
                    607: 	    Log("SUCCESS", "Reply from lond: '$data'");
                    608: 	}
1.7       foxr      609: 	StartClientReply($Transaction, $data);
                    610:     } else {			# Delete deferred transaction file.
1.9       foxr      611: 	Log("SUCCESS", "A delayed transaction was completed");
1.86      albertel  612: 	LogPerm("S:".$Transaction->getClient().":".$Transaction->getRequest());
                    613: 	unlink($Transaction->getFile());
1.7       foxr      614:     }
1.6       foxr      615: }
1.42      foxr      616: 
1.6       foxr      617: =pod
1.42      foxr      618: 
1.6       foxr      619: =head1 StartClientReply
                    620: 
                    621:    Initiates a reply to a client where the reply data is a parameter.
                    622: 
1.7       foxr      623: =head2  parameters:
                    624: 
                    625: =item Transaction
                    626: 
                    627:     The transaction for which we are responding to the client.
                    628: 
                    629: =item data
                    630: 
                    631:     The data to send to apached client.
                    632: 
1.6       foxr      633: =cut
1.42      foxr      634: 
1.6       foxr      635: sub StartClientReply {
1.1       foxr      636: 
1.47      foxr      637:     my ($Transaction, $data) = @_;
1.12      foxr      638: 
1.7       foxr      639:     my $Client   = $Transaction->getClient();
                    640: 
1.1       foxr      641:     &Debug(8," Reply was: ".$data);
                    642:     my $Serial         = $ActiveClients{$Client};
                    643:     my $desc           = sprintf("Connection to lonc client %d",
                    644: 				 $Serial);
                    645:     Event->io(fd       => $Client,
                    646: 	      poll     => "w",
                    647: 	      desc     => $desc,
                    648: 	      cb       => \&ClientWritable,
                    649: 	      data     => $data);
                    650: }
1.42      foxr      651: 
1.4       foxr      652: =pod
1.42      foxr      653: 
1.4       foxr      654: =head2 FailTransaction
                    655: 
                    656:   Finishes a transaction with failure because the associated lond socket
1.7       foxr      657:   disconnected.  There are two possibilities:
                    658:   - The transaction is deferred: in which case we just quietly
                    659:     delete the transaction since there is no client connection.
                    660:   - The transaction is 'live' in which case we initiate the sending
                    661:     of "con_lost" to the client.
                    662: 
1.42      foxr      663: Deleting the transaction means killing it from the %ActiveTransactions hash.
1.4       foxr      664: 
                    665: Parameters:
                    666: 
                    667: =item client  
                    668:  
1.7       foxr      669:    The LondTransaction we are failing.
                    670:  
1.42      foxr      671: 
1.4       foxr      672: =cut
                    673: 
                    674: sub FailTransaction {
1.7       foxr      675:     my $transaction = shift;
1.52      foxr      676:     
                    677:     #  If the socket is dead, that's already logged.
                    678: 
                    679:     if ($ConnectionRetriesLeft > 0) {
                    680: 	Log("WARNING", "Failing transaction "
1.71      albertel  681: 	    .$transaction->getLoggableRequest());
1.52      foxr      682:     }
1.71      albertel  683:     Debug(1, "Failing transaction: ".$transaction->getLoggableRequest());
1.10      foxr      684:     if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
1.11      foxr      685: 	my $client  = $transaction->getClient();
1.30      foxr      686: 	Debug(1," Replying con_lost to ".$transaction->getRequest());
1.11      foxr      687: 	StartClientReply($transaction, "con_lost\n");
1.7       foxr      688:     }
1.4       foxr      689: 
                    690: }
                    691: 
                    692: =pod
1.69      matthew   693: 
1.6       foxr      694: =head1  EmptyQueue
1.7       foxr      695: 
1.6       foxr      696:   Fails all items in the work queue with con_lost.
1.7       foxr      697:   Note that each item in the work queue is a transaction.
                    698: 
1.6       foxr      699: =cut
1.69      matthew   700: 
1.6       foxr      701: sub EmptyQueue {
1.22      foxr      702:     $ConnectionRetriesLeft--;	# Counts as connection failure too.
1.6       foxr      703:     while($WorkQueue->Count()) {
1.10      foxr      704: 	my $request = $WorkQueue->dequeue(); # This is a transaction
1.7       foxr      705: 	FailTransaction($request);
1.6       foxr      706:     }
                    707: }
                    708: 
                    709: =pod
1.4       foxr      710: 
1.9       foxr      711: =head2 CloseAllLondConnections
                    712: 
                    713: Close all connections open on lond prior to exit e.g.
                    714: 
                    715: =cut
1.69      matthew   716: 
1.9       foxr      717: sub CloseAllLondConnections {
1.23      foxr      718:     foreach my $Socket (keys %ActiveConnections) {
1.42      foxr      719:       if(exists($ActiveTransactions{$Socket})) {
                    720: 	FailTransaction($ActiveTransactions{$Socket});
                    721:       }
                    722:       KillSocket($Socket);
1.9       foxr      723:     }
                    724: }
                    725: 
                    726: =pod
                    727: 
1.4       foxr      728: =head2 KillSocket
                    729:  
                    730: Destroys a socket.  This function can be called either when a socket
                    731: has died of 'natural' causes or because a socket needs to be pruned due to
                    732: idleness.  If the socket has died naturally, if there are no longer any 
                    733: live connections a new connection is created (in case there are transactions
                    734: in the queue).  If the socket has been pruned, it is never re-created.
                    735: 
                    736: Parameters:
1.1       foxr      737: 
1.4       foxr      738: =item Socket
                    739:  
                    740:   The socket to kill off.
                    741: 
                    742: =item Restart
                    743: 
                    744: nonzero if we are allowed to create a new connection.
                    745: 
1.69      matthew   746: =cut
1.4       foxr      747: 
                    748: sub KillSocket {
                    749:     my $Socket = shift;
                    750: 
1.17      foxr      751:     Log("WARNING", "Shutting down a socket");
1.9       foxr      752:     $Socket->Shutdown();
                    753: 
1.7       foxr      754:     #  If the socket came from the active connection set,
                    755:     #  delete its transaction... note that FailTransaction should
                    756:     #  already have been called!!!
                    757:     #  otherwise it came from the idle set.
                    758:     #  
1.4       foxr      759:     
                    760:     if(exists($ActiveTransactions{$Socket})) {
                    761: 	delete ($ActiveTransactions{$Socket});
                    762:     }
                    763:     if(exists($ActiveConnections{$Socket})) {
1.90      foxr      764: 	$ActiveConnections{$Socket}->cancel;
1.4       foxr      765: 	delete($ActiveConnections{$Socket});
1.37      albertel  766: 	$ConnectionCount--;
                    767: 	if ($ConnectionCount < 0) { $ConnectionCount = 0; }
1.4       foxr      768:     }
1.6       foxr      769:     #  If the connection count has gone to zero and there is work in the
                    770:     #  work queue, the work all gets failed with con_lost.
                    771:     #
                    772:     if($ConnectionCount == 0) {
1.22      foxr      773: 	EmptyQueue();
1.42      foxr      774: 	CloseAllLondConnections; # Should all already be closed but...
1.4       foxr      775:     }
1.90      foxr      776:     UpdateStatus();
1.4       foxr      777: }
1.1       foxr      778: 
                    779: =pod
1.3       albertel  780: 
1.1       foxr      781: =head2 LondReadable
1.3       albertel  782: 
1.1       foxr      783: This function is called whenever a lond connection
                    784: is readable.  The action is state dependent:
                    785: 
1.3       albertel  786: =head3 State=Initialized
                    787: 
                    788: We''re waiting for the challenge, this is a no-op until the
1.1       foxr      789: state changes.
1.3       albertel  790: 
1.1       foxr      791: =head3 State=Challenged 
1.3       albertel  792: 
                    793: The challenge has arrived we need to transition to Writable.
1.1       foxr      794: The connection must echo the challenge back.
1.3       albertel  795: 
1.1       foxr      796: =head3 State=ChallengeReplied
1.3       albertel  797: 
                    798: The challenge has been replied to.  The we are receiveing the 
1.1       foxr      799: 'ok' from the partner.
1.3       albertel  800: 
1.40      foxr      801: =head3  State=ReadingVersionString
                    802: 
                    803: We have requested the lond version and are reading the
                    804: version back.  Upon completion, we'll store the version away
                    805: for future use(?).
                    806: 
                    807: =head3 State=HostSet
                    808: 
                    809: We have selected the domain name of our peer (multhomed hosts)
                    810: and are getting the reply (presumably ok) back.
                    811: 
1.1       foxr      812: =head3 State=RequestingKey
1.3       albertel  813: 
                    814: The ok has been received and we need to send the request for
1.1       foxr      815: an encryption key.  Transition to writable for that.
1.3       albertel  816: 
1.1       foxr      817: =head3 State=ReceivingKey
1.3       albertel  818: 
                    819: The the key has been requested, now we are reading the new key.
                    820: 
1.1       foxr      821: =head3 State=Idle 
1.3       albertel  822: 
                    823: The encryption key has been negotiated or we have finished 
1.1       foxr      824: reading data from the a transaction.   If the callback data has
                    825: a client as well as the socket iformation, then we are 
                    826: doing a transaction and the data received is relayed to the client
                    827: before the socket is put on the idle list.
1.3       albertel  828: 
1.1       foxr      829: =head3 State=SendingRequest
1.3       albertel  830: 
                    831: I do not think this state can be received here, but if it is,
1.1       foxr      832: the appropriate thing to do is to transition to writable, and send
                    833: the request.
1.3       albertel  834: 
1.1       foxr      835: =head3 State=ReceivingReply
1.3       albertel  836: 
                    837: We finished sending the request to the server and now transition
1.1       foxr      838: to readable to receive the reply. 
                    839: 
                    840: The parameter to this function are:
1.3       albertel  841: 
1.1       foxr      842: The event. Implicit in this is the watcher and its data.  The data 
                    843: contains at least the lond connection object and, if a 
                    844: transaction is in progress, the socket attached to the local client.
                    845: 
1.3       albertel  846: =cut
1.1       foxr      847: 
                    848: sub LondReadable {
1.8       foxr      849: 
1.41      albertel  850:     my $Event      = shift;
                    851:     my $Watcher    = $Event->w;
                    852:     my $Socket     = $Watcher->data;
                    853:     my $client     = undef;
1.40      foxr      854: 
1.41      albertel  855:     &Debug(6,"LondReadable called state = ".$Socket->GetState());
1.40      foxr      856: 
                    857: 
1.41      albertel  858:     my $State = $Socket->GetState(); # All action depends on the state.
1.40      foxr      859: 
1.41      albertel  860:     SocketDump(6, $Socket);
                    861:     my $status = $Socket->Readable();
1.40      foxr      862: 
1.41      albertel  863:     &Debug(2, "Socket->Readable returned: $status");
1.40      foxr      864: 
1.41      albertel  865:     if($status != 0) {
                    866: 	# bad return from socket read. Currently this means that
                    867: 	# The socket has become disconnected. We fail the transaction.
1.40      foxr      868: 
1.41      albertel  869: 	Log("WARNING",
                    870: 	    "Lond connection lost.");
                    871: 	if(exists($ActiveTransactions{$Socket})) {
                    872: 	    FailTransaction($ActiveTransactions{$Socket});
1.56      foxr      873: 	} else {
                    874: 	    #  Socket is connecting and failed... need to mark
                    875: 	    #  no longer connecting.
                    876: 	   
                    877: 	    $LondConnecting = 0;
1.41      albertel  878: 	}
                    879: 	$Watcher->cancel();
                    880: 	KillSocket($Socket);
                    881: 	$ConnectionRetriesLeft--;       # Counts as connection failure
                    882: 	return;
                    883:     }
                    884:     SocketDump(6,$Socket);
1.17      foxr      885: 
1.41      albertel  886:     $State = $Socket->GetState(); # Update in case of transition.
                    887:     &Debug(6, "After read, state is ".$State);
1.1       foxr      888: 
1.41      albertel  889:     if($State eq "Initialized") {
1.1       foxr      890: 
                    891: 
1.41      albertel  892:     } elsif ($State eq "ChallengeReceived") {
1.1       foxr      893: 	#  The challenge must be echoed back;  The state machine
                    894: 	# in the connection takes care of setting that up.  Just
                    895: 	# need to transition to writable:
1.41      albertel  896: 	
                    897: 	$Watcher->cb(\&LondWritable);
                    898: 	$Watcher->poll("w");
1.1       foxr      899: 
1.41      albertel  900:     } elsif ($State eq "ChallengeReplied") {
1.1       foxr      901: 
1.41      albertel  902:     } elsif ($State eq "RequestingVersion") {
                    903: 	# Need to ask for the version... that is writiability:
1.1       foxr      904: 
1.41      albertel  905: 	$Watcher->cb(\&LondWritable);
                    906: 	$Watcher->poll("w");
                    907: 
                    908:     } elsif ($State eq "ReadingVersionString") {
                    909: 	# Read the rest of the version string... 
                    910:     } elsif ($State eq "SetHost") {
                    911: 	# Need to request the actual domain get set...
                    912: 
                    913: 	$Watcher->cb(\&LondWritable);
                    914: 	$Watcher->poll("w");
                    915:     } elsif ($State eq "HostSet") {
                    916: 	# Reading the 'ok' from the peer.
                    917: 
                    918:     } elsif ($State eq "RequestingKey") {
1.1       foxr      919: 	#  The ok was received.  Now we need to request the key
                    920: 	#  That requires us to be writable:
                    921: 
1.41      albertel  922: 	$Watcher->cb(\&LondWritable);
                    923: 	$Watcher->poll("w");
1.1       foxr      924: 
1.41      albertel  925:     } elsif ($State eq "ReceivingKey") {
1.1       foxr      926: 
1.41      albertel  927:     } elsif ($State eq "Idle") {
1.40      foxr      928:    
1.41      albertel  929: 	# This is as good a spot as any to get the peer version
                    930: 	# string:
1.40      foxr      931:    
1.41      albertel  932: 	if($LondVersion eq "unknown") {
                    933: 	    $LondVersion = $Socket->PeerVersion();
                    934: 	    Log("INFO", "Connected to lond version: $LondVersion");
                    935: 	}
1.1       foxr      936: 	# If necessary, complete a transaction and then go into the
                    937: 	# idle queue.
1.22      foxr      938: 	#  Note that a trasition to idle indicates a live lond
                    939: 	# on the other end so reset the connection retries.
                    940: 	#
1.41      albertel  941: 	$ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
                    942: 	$Watcher->cancel();
                    943: 	if(exists($ActiveTransactions{$Socket})) {
                    944: 	    Debug(5,"Completing transaction!!");
                    945: 	    CompleteTransaction($Socket, 
                    946: 				$ActiveTransactions{$Socket});
                    947: 	} else {
                    948: 	    Log("SUCCESS", "Connection ".$ConnectionCount." to "
                    949: 		.$RemoteHost." now ready for action");
                    950: 	}
                    951: 	ServerToIdle($Socket);	# Next work unit or idle.
1.54      foxr      952: 
                    953: 	#
                    954: 	$LondConnecting = 0;	# Best spot I can think of for this.
                    955: 	# 
1.6       foxr      956: 	
1.41      albertel  957:     } elsif ($State eq "SendingRequest") {
1.1       foxr      958: 	#  We need to be writable for this and probably don't belong
                    959: 	#  here inthe first place.
                    960: 
1.73      albertel  961: 	Debug(6, "SendingRequest state encountered in readable");
1.41      albertel  962: 	$Watcher->poll("w");
                    963: 	$Watcher->cb(\&LondWritable);
1.1       foxr      964: 
1.41      albertel  965:     } elsif ($State eq "ReceivingReply") {
1.1       foxr      966: 
                    967: 
1.41      albertel  968:     } else {
                    969: 	# Invalid state.
                    970: 	Debug(4, "Invalid state in LondReadable");
                    971:     }
1.1       foxr      972: }
1.3       albertel  973: 
1.1       foxr      974: =pod
1.3       albertel  975: 
1.1       foxr      976: =head2 LondWritable
1.3       albertel  977: 
1.1       foxr      978: This function is called whenever a lond connection
                    979: becomes writable while there is a writeable monitoring
                    980: event.  The action taken is very state dependent:
1.3       albertel  981: 
1.1       foxr      982: =head3 State = Connected 
1.3       albertel  983: 
                    984: The connection is in the process of sending the 'init' hailing to the
                    985: lond on the remote end.  The connection object''s Writable member is
                    986: called.  On error, ConnectionError is called to destroy the connection
                    987: and remove it from the ActiveConnections hash
                    988: 
1.1       foxr      989: =head3 Initialized
1.3       albertel  990: 
                    991: 'init' has been sent, writability monitoring is removed and
                    992: readability monitoring is started with LondReadable as the callback.
                    993: 
1.1       foxr      994: =head3 ChallengeReceived
1.3       albertel  995: 
                    996: The connection has received the who are you challenge from the remote
                    997: system, and is in the process of sending the challenge
                    998: response. Writable is called.
                    999: 
1.1       foxr     1000: =head3 ChallengeReplied
1.3       albertel 1001: 
                   1002: The connection has replied to the initial challenge The we switch to
                   1003: monitoring readability looking for the server to reply with 'ok'.
                   1004: 
1.1       foxr     1005: =head3 RequestingKey
1.3       albertel 1006: 
                   1007: The connection is in the process of requesting its encryption key.
                   1008: Writable is called.
                   1009: 
1.1       foxr     1010: =head3 ReceivingKey
1.3       albertel 1011: 
                   1012: The connection has sent the request for a key.  Switch to readability
                   1013: monitoring to accept the key
                   1014: 
1.1       foxr     1015: =head3 SendingRequest
1.3       albertel 1016: 
                   1017: The connection is in the process of sending a request to the server.
                   1018: This request is part of a client transaction.  All the states until
                   1019: now represent the client setup protocol. Writable is called.
                   1020: 
1.1       foxr     1021: =head3 ReceivingReply
                   1022: 
1.3       albertel 1023: The connection has sent a request.  Now it must receive a reply.
                   1024: Readability monitoring is requested.
                   1025: 
                   1026: This function is an event handler and therefore receives as
1.1       foxr     1027: a parameter the event that has fired.  The data for the watcher
                   1028: of this event is a reference to a list of one or two elements,
                   1029: depending on state. The first (and possibly only) element is the
                   1030: socket.  The second (present only if a request is in progress)
                   1031: is the socket on which to return a reply to the caller.
                   1032: 
                   1033: =cut
1.3       albertel 1034: 
1.1       foxr     1035: sub LondWritable {
                   1036:     my $Event   = shift;
                   1037:     my $Watcher = $Event->w;
1.8       foxr     1038:     my $Socket  = $Watcher->data;
                   1039:     my $State   = $Socket->GetState();
1.1       foxr     1040: 
1.8       foxr     1041:     Debug(6,"LondWritable State = ".$State."\n");
1.1       foxr     1042: 
1.8       foxr     1043:  
1.1       foxr     1044:     #  Figure out what to do depending on the state of the socket:
                   1045:     
                   1046: 
                   1047: 
                   1048: 
                   1049:     SocketDump(6,$Socket);
                   1050: 
1.42      foxr     1051:     #  If the socket is writable, we must always write.
                   1052:     # Only by writing will we undergo state transitions.
                   1053:     # Old logic wrote in state specific code below, however
                   1054:     # That forces us at least through another invocation of
                   1055:     # this function after writability is possible again.
                   1056:     # This logic also factors out common code for handling
                   1057:     # write failures... in all cases, write failures 
                   1058:     # Kill the socket.
                   1059:     #  This logic makes the branches of the >big< if below
                   1060:     # so that the writing states are actually NO-OPs.
                   1061: 
                   1062:     if ($Socket->Writable() != 0) {
1.43      albertel 1063: 	#  The write resulted in an error.
                   1064: 	# We'll treat this as if the socket got disconnected:
                   1065: 	Log("WARNING", "Connection to ".$RemoteHost.
                   1066: 	    " has been disconnected");
                   1067: 	if(exists($ActiveTransactions{$Socket})) {
                   1068: 	    FailTransaction($ActiveTransactions{$Socket});
1.56      foxr     1069: 	} else {
                   1070: 	    #  In the process of conneting, so need to turn that off.
                   1071: 	    
                   1072: 	    $LondConnecting = 0;
1.43      albertel 1073: 	}
                   1074: 	$Watcher->cancel();
                   1075: 	KillSocket($Socket);
                   1076: 	return;
1.42      foxr     1077:     }
                   1078: 
                   1079: 
                   1080: 
1.41      albertel 1081:     if      ($State eq "Connected")         {
1.1       foxr     1082: 
1.41      albertel 1083: 	#  "init" is being sent...
1.42      foxr     1084:  
1.41      albertel 1085:     } elsif ($State eq "Initialized")       {
1.4       foxr     1086: 
1.41      albertel 1087: 	# Now that init was sent, we switch 
                   1088: 	# to watching for readability:
1.1       foxr     1089: 
1.41      albertel 1090: 	$Watcher->cb(\&LondReadable);
                   1091: 	$Watcher->poll("r");
                   1092: 	
                   1093:     } elsif ($State eq "ChallengeReceived") {
                   1094: 	# We received the challenge, now we 
                   1095: 	# are echoing it back. This is a no-op,
                   1096: 	# we're waiting for the state to change
1.1       foxr     1097: 	
1.41      albertel 1098:     } elsif ($State eq "ChallengeReplied")  {
                   1099: 	# The echo was sent back, so we switch
                   1100: 	# to watching readability.
                   1101: 
                   1102: 	$Watcher->cb(\&LondReadable);
                   1103: 	$Watcher->poll("r");
                   1104:     } elsif ($State eq "RequestingVersion") {
                   1105: 	# Sending the peer a version request...
1.42      foxr     1106: 
1.41      albertel 1107:     } elsif ($State eq "ReadingVersionString") {
                   1108: 	# Transition to read since we have sent the
                   1109: 	# version command and now just need to read the
                   1110: 	# version string from the peer:
1.40      foxr     1111:       
1.41      albertel 1112: 	$Watcher->cb(\&LondReadable);
                   1113: 	$Watcher->poll("r");
1.40      foxr     1114:       
1.41      albertel 1115:     } elsif ($State eq "SetHost") {
                   1116: 	#  Setting the remote domain...
1.42      foxr     1117: 
1.41      albertel 1118:     } elsif ($State eq "HostSet") {
                   1119: 	# Back to readable to get the ok.
1.40      foxr     1120:       
1.41      albertel 1121: 	$Watcher->cb(\&LondReadable);
                   1122: 	$Watcher->poll("r");
1.40      foxr     1123:       
                   1124: 
1.41      albertel 1125:     } elsif ($State eq "RequestingKey")     {
                   1126: 	# At this time we're requesting the key.
                   1127: 	# again, this is essentially a no-op.
                   1128: 
                   1129:     } elsif ($State eq "ReceivingKey")      {
                   1130: 	# Now we need to wait for the key
                   1131: 	# to come back from the peer:
                   1132: 
                   1133: 	$Watcher->cb(\&LondReadable);
                   1134: 	$Watcher->poll("r");
                   1135: 
                   1136:     } elsif ($State eq "SendingRequest")    {
1.40      foxr     1137:  
1.41      albertel 1138: 	# At this time we are sending a request to the
1.1       foxr     1139: 	# peer... write the next chunk:
                   1140: 
1.41      albertel 1141: 
                   1142:     } elsif ($State eq "ReceivingReply")    {
                   1143: 	# The send has completed.  Wait for the
                   1144: 	# data to come in for a reply.
                   1145: 	Debug(8,"Writable sent request/receiving reply");
                   1146: 	$Watcher->cb(\&LondReadable);
                   1147: 	$Watcher->poll("r");
1.1       foxr     1148: 
1.41      albertel 1149:     } else {
                   1150: 	#  Control only passes here on an error: 
                   1151: 	#  the socket state does not match any
                   1152: 	#  of the known states... so an error
                   1153: 	#  must be logged.
1.1       foxr     1154: 
1.41      albertel 1155: 	&Debug(4, "Invalid socket state ".$State."\n");
                   1156:     }
1.1       foxr     1157:     
                   1158: }
1.81      albertel 1159: 
1.6       foxr     1160: =pod
                   1161:     
                   1162: =cut
1.69      matthew  1163: 
1.81      albertel 1164: 
1.6       foxr     1165: sub QueueDelayed {
1.8       foxr     1166:     Debug(3,"QueueDelayed called");
                   1167: 
1.6       foxr     1168:     my $path = "$perlvar{'lonSockDir'}/delayed";
1.8       foxr     1169: 
                   1170:     Debug(4, "Delayed path: ".$path);
1.6       foxr     1171:     opendir(DIRHANDLE, $path);
1.75      albertel 1172: 
1.82      albertel 1173:     my $host_id_re = '(?:'.join('|',map {quotemeta($_)} (@all_host_ids)).')';
1.75      albertel 1174:     my @alldelayed = grep(/\.$host_id_re$/, readdir(DIRHANDLE));
1.6       foxr     1175:     closedir(DIRHANDLE);
1.75      albertel 1176:     foreach my $dfname (sort(@alldelayed)) {
                   1177: 	my $reqfile = "$path/$dfname";
                   1178: 	my ($host_id) = ($dfname =~ /\.([^.]*)$/);
                   1179: 	Debug(4, "queueing ".$reqfile." for $host_id");
1.6       foxr     1180: 	my $Handle = IO::File->new($reqfile);
                   1181: 	my $cmd    = <$Handle>;
1.8       foxr     1182: 	chomp $cmd;		# There may or may not be a newline...
1.12      foxr     1183: 	$cmd = $cmd."\n";	# now for sure there's exactly one newline.
1.75      albertel 1184: 	my $Transaction = LondTransaction->new("sethost:$host_id:$cmd");
1.7       foxr     1185: 	$Transaction->SetDeferred($reqfile);
                   1186: 	QueueTransaction($Transaction);
1.6       foxr     1187:     }
                   1188:     
                   1189: }
1.1       foxr     1190: 
                   1191: =pod
1.3       albertel 1192: 
1.1       foxr     1193: =head2 MakeLondConnection
1.3       albertel 1194: 
                   1195: Create a new lond connection object, and start it towards its initial
                   1196: idleness.  Once idle, it becomes elligible to receive transactions
                   1197: from the work queue.  If the work queue is not empty when the
                   1198: connection is completed and becomes idle, it will dequeue an entry and
                   1199: start off on it.
                   1200: 
1.1       foxr     1201: =cut
1.3       albertel 1202: 
1.1       foxr     1203: sub MakeLondConnection {     
                   1204:     Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
                   1205: 	  .GetServerPort());
                   1206: 
                   1207:     my $Connection = LondConnection->new(&GetServerHost(),
1.81      albertel 1208: 					 &GetServerPort(),
                   1209: 					 &GetHostId());
1.1       foxr     1210: 
1.90      foxr     1211:     if($Connection eq undef) {	
1.9       foxr     1212: 	Log("CRITICAL","Failed to make a connection with lond.");
1.10      foxr     1213: 	$ConnectionRetriesLeft--;
                   1214: 	return 0;		# Failure.
1.5       foxr     1215:     }  else {
1.22      foxr     1216: 
1.82      albertel 1217: 	$LondConnecting = 1;	# Connection in progress.
1.5       foxr     1218: 	# The connection needs to have writability 
                   1219: 	# monitored in order to send the init sequence
                   1220: 	# that starts the whole authentication/key
                   1221: 	# exchange underway.
                   1222: 	#
                   1223: 	my $Socket = $Connection->GetSocket();
1.30      foxr     1224: 	if($Socket eq undef) {
1.64      foxr     1225: 	    &child_exit(-1, "did not get a socket from the connection");
1.5       foxr     1226: 	} else {
                   1227: 	    &Debug(9,"MakeLondConnection got socket: ".$Socket);
                   1228: 	}
1.1       foxr     1229: 	
1.21      foxr     1230: 	$Connection->SetTimeoutCallback(\&SocketTimeout);
                   1231: 
1.23      foxr     1232: 	my $event = Event->io(fd       => $Socket,
1.5       foxr     1233: 			   poll     => 'w',
                   1234: 			   cb       => \&LondWritable,
1.8       foxr     1235: 			   data     => $Connection,
1.5       foxr     1236: 			   desc => 'Connection to lond server');
                   1237: 	$ActiveConnections{$Connection} = $event;
1.52      foxr     1238: 	if ($ConnectionCount == 0) {
                   1239: 	    &SetupTimer;	# Need to handle timeouts with connections...
                   1240: 	}
1.5       foxr     1241: 	$ConnectionCount++;
1.8       foxr     1242: 	Debug(4, "Connection count = ".$ConnectionCount);
1.6       foxr     1243: 	if($ConnectionCount == 1) { # First Connection:
                   1244: 	    QueueDelayed;
                   1245: 	}
1.9       foxr     1246: 	Log("SUCESS", "Created connection ".$ConnectionCount
                   1247: 	    ." to host ".GetServerHost());
1.10      foxr     1248: 	return 1;		# Return success.
1.1       foxr     1249:     }
                   1250:     
                   1251: }
1.3       albertel 1252: 
1.1       foxr     1253: =pod
1.3       albertel 1254: 
1.1       foxr     1255: =head2 StartRequest
1.3       albertel 1256: 
                   1257: Starts a lond request going on a specified lond connection.
                   1258: parameters are:
                   1259: 
                   1260: =item $Lond
                   1261: 
                   1262: Connection to the lond that will send the transaction and receive the
                   1263: reply.
                   1264: 
                   1265: =item $Client
                   1266: 
                   1267: Connection to the client that is making this request We got the
                   1268: request from this socket, and when the request has been relayed to
                   1269: lond and we get a reply back from lond it will get sent to this
                   1270: socket.
                   1271: 
                   1272: =item $Request
                   1273: 
                   1274: The text of the request to send.
                   1275: 
1.1       foxr     1276: =cut
                   1277: 
                   1278: sub StartRequest {
1.47      foxr     1279: 
                   1280:     my ($Lond, $Request) = @_;
1.1       foxr     1281:     
1.7       foxr     1282:     Debug(6, "StartRequest: ".$Request->getRequest());
1.1       foxr     1283: 
                   1284:     my $Socket = $Lond->GetSocket();
                   1285:     
1.7       foxr     1286:     $Request->Activate($Lond);
                   1287:     $ActiveTransactions{$Lond} = $Request;
1.1       foxr     1288: 
1.7       foxr     1289:     $Lond->InitiateTransaction($Request->getRequest());
1.23      foxr     1290:     my $event = Event->io(fd      => $Socket,
1.1       foxr     1291: 		       poll    => "w",
                   1292: 		       cb      => \&LondWritable,
                   1293: 		       data    => $Lond,
                   1294: 		       desc    => "lond transaction connection");
                   1295:     $ActiveConnections{$Lond} = $event;
                   1296:     Debug(8," Start Request made watcher data with ".$event->data."\n");
                   1297: }
                   1298: 
                   1299: =pod
1.3       albertel 1300: 
1.1       foxr     1301: =head2 QueueTransaction
1.3       albertel 1302: 
                   1303: If there is an idle lond connection, it is put to work doing this
                   1304: transaction.  Otherwise, the transaction is placed in the work queue.
                   1305: If placed in the work queue and the maximum number of connections has
                   1306: not yet been created, a new connection will be started.  Our goal is
                   1307: to eventually have a sufficient number of connections that the work
                   1308: queue will typically be empty.  parameters are:
                   1309: 
                   1310: =item Socket
                   1311: 
                   1312: open on the lonc client.
                   1313: 
                   1314: =item Request
                   1315: 
                   1316: data to send to the lond.
1.1       foxr     1317: 
                   1318: =cut
1.3       albertel 1319: 
1.1       foxr     1320: sub QueueTransaction {
                   1321: 
1.7       foxr     1322:     my $requestData   = shift;	# This is a LondTransaction.
                   1323:     my $cmd           = $requestData->getRequest();
                   1324: 
                   1325:     Debug(6,"QueueTransaction: ".$cmd);
1.1       foxr     1326: 
                   1327:     my $LondSocket    = $IdleConnections->pop();
                   1328:     if(!defined $LondSocket) {	# Need to queue request.
1.29      foxr     1329: 	Debug(5,"Must queue...");
1.1       foxr     1330: 	$WorkQueue->enqueue($requestData);
1.56      foxr     1331: 	Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
                   1332: 	if(($ConnectionCount < $MaxConnectionCount)   && (! $LondConnecting)) {
                   1333: 
1.22      foxr     1334: 	    if($ConnectionRetriesLeft > 0) {
1.29      foxr     1335: 		Debug(5,"Starting additional lond connection");
1.56      foxr     1336: 		if(&MakeLondConnection() == 0) {
1.22      foxr     1337: 		    EmptyQueue();	# Fail transactions, can't make connection.
1.42      foxr     1338: 		    CloseAllLondConnections; # Should all be closed but...
1.22      foxr     1339: 		}
                   1340: 	    } else {
                   1341: 		ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1.56      foxr     1342: 		$LondConnecting = 0;
1.22      foxr     1343: 		EmptyQueue();	# It's worse than that ... he's dead Jim.
1.42      foxr     1344: 		CloseAllLondConnections; # Should all be closed but..
1.17      foxr     1345: 	    }
1.1       foxr     1346: 	}
                   1347:     } else {			# Can start the request:
                   1348: 	Debug(8,"Can start...");
1.7       foxr     1349: 	StartRequest($LondSocket,  $requestData);
1.1       foxr     1350:     }
                   1351: }
                   1352: 
                   1353: #-------------------------- Lonc UNIX socket handling ---------------------
1.3       albertel 1354: 
1.1       foxr     1355: =pod
1.3       albertel 1356: 
1.1       foxr     1357: =head2 ClientRequest
1.3       albertel 1358: Callback that is called when data can be read from the UNIX domain
                   1359: socket connecting us with an apache server process.
1.1       foxr     1360: 
                   1361: =cut
                   1362: 
                   1363: sub ClientRequest {
                   1364:     Debug(6, "ClientRequest");
                   1365:     my $event   = shift;
                   1366:     my $watcher = $event->w;
                   1367:     my $socket  = $watcher->fd;
                   1368:     my $data    = $watcher->data;
                   1369:     my $thisread;
                   1370: 
                   1371:     Debug(9, "  Watcher named: ".$watcher->desc);
                   1372: 
                   1373:     my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
                   1374:     Debug(8, "rcv:  data length = ".length($thisread)
                   1375: 	  ." read =".$thisread);
1.29      foxr     1376:     unless (defined $rv  && length($thisread)) {
1.1       foxr     1377: 	 # Likely eof on socket.
                   1378: 	Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
                   1379: 	close($socket);
                   1380: 	$watcher->cancel();
                   1381: 	delete($ActiveClients{$socket});
1.10      foxr     1382: 	return;
1.1       foxr     1383:     }
                   1384:     Debug(8,"Data: ".$data." this read: ".$thisread);
                   1385:     $data = $data.$thisread;	# Append new data.
                   1386:     $watcher->data($data);
1.44      albertel 1387:     if($data =~ /\n$/) {	# Request entirely read.
1.87      albertel 1388: 	if ($data eq "close_connection_exit\n") {
1.9       foxr     1389: 	    Log("CRITICAL",
                   1390: 		"Request Close Connection ... exiting");
                   1391: 	    CloseAllLondConnections();
                   1392: 	    exit;
1.87      albertel 1393: 	} elsif ($data eq "reset_retries\n") {
                   1394: 	    Log("INFO", "Resetting Connection Retries.");
                   1395: 	    $ConnectionRetriesLeft = $ConnectionRetries;
                   1396: 	    &UpdateStatus();
                   1397: 	    my $Transaction = LondTransaction->new($data);
                   1398: 	    $Transaction->SetClient($socket);
                   1399: 	    StartClientReply($Transaction, "ok\n");
                   1400: 	    $watcher->cancel();
                   1401: 	    return;
1.9       foxr     1402: 	}
1.1       foxr     1403: 	Debug(8, "Complete transaction received: ".$data);
1.87      albertel 1404: 	if ($LogTransactions) {
1.39      foxr     1405: 	    Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
                   1406: 	}
1.8       foxr     1407: 	my $Transaction = LondTransaction->new($data);
1.7       foxr     1408: 	$Transaction->SetClient($socket);
                   1409: 	QueueTransaction($Transaction);
1.1       foxr     1410: 	$watcher->cancel();	# Done looking for input data.
                   1411:     }
                   1412: 
                   1413: }
                   1414: 
1.62      foxr     1415: #
                   1416: #     Accept a connection request for a client (lonc child) and
                   1417: #    start up an event watcher to keep an eye on input from that 
                   1418: #    Event.  This can be called both from NewClient and from
1.80      albertel 1419: #    ChildProcess.
1.62      foxr     1420: # Parameters:
                   1421: #    $socket       - The listener socket.
                   1422: # Returns:
                   1423: #   NONE
                   1424: # Side Effects:
                   1425: #    An event is made to watch the accepted connection.
                   1426: #    Active clients hash is updated to reflect the new connection.
                   1427: #    The client connection count is incremented.
                   1428: #
                   1429: sub accept_client {
                   1430:     my ($socket) = @_;
                   1431: 
                   1432:     Debug(8, "Entering accept for lonc UNIX socket\n");
                   1433:     my $connection = $socket->accept();	# Accept the client connection.
                   1434:     Debug(8,"Connection request accepted from "
                   1435: 	  .GetPeername($connection, AF_UNIX));
                   1436: 
                   1437: 
                   1438:     my $description = sprintf("Connection to lonc client %d",
                   1439: 			      $ClientConnection);
                   1440:     Debug(9, "Creating event named: ".$description);
                   1441:     Event->io(cb      => \&ClientRequest,
                   1442: 	      poll    => 'r',
                   1443: 	      desc    => $description,
                   1444: 	      data    => "",
                   1445: 	      fd      => $connection);
                   1446:     $ActiveClients{$connection} = $ClientConnection;
                   1447:     $ClientConnection++;
                   1448: }
1.1       foxr     1449: 
                   1450: =pod
1.3       albertel 1451: 
1.1       foxr     1452: =head2  NewClient
1.3       albertel 1453: 
                   1454: Callback that is called when a connection is received on the unix
                   1455: socket for a new client of lonc.  The callback is parameterized by the
                   1456: event.. which is a-priori assumed to be an io event, and therefore has
                   1457: an fd member that is the Listener socket.  We Accept the connection
                   1458: and register a new event on the readability of that socket:
                   1459: 
1.1       foxr     1460: =cut
1.3       albertel 1461: 
1.1       foxr     1462: sub NewClient {
                   1463:     Debug(6, "NewClient");
                   1464:     my $event      = shift;		# Get the event parameters.
                   1465:     my $watcher    = $event->w; 
                   1466:     my $socket     = $watcher->fd;	# Get the event' socket.
                   1467: 
1.62      foxr     1468:     &accept_client($socket);
1.1       foxr     1469: }
1.3       albertel 1470: 
                   1471: =pod
                   1472: 
                   1473: =head2 GetLoncSocketPath
                   1474: 
                   1475: Returns the name of the UNIX socket on which to listen for client
                   1476: connections.
1.1       foxr     1477: 
1.58      foxr     1478: =head2 Parameters:
                   1479: 
                   1480:     host (optional)  - Name of the host socket to return.. defaults to
                   1481:                        the return from GetServerHost().
                   1482: 
1.1       foxr     1483: =cut
1.3       albertel 1484: 
1.1       foxr     1485: sub GetLoncSocketPath {
1.58      foxr     1486: 
                   1487:     my $host = GetServerHost();	# Default host.
                   1488:     if (@_) {
                   1489: 	($host)  = @_;		# Override if supplied.
                   1490:     }
                   1491:     return $UnixSocketDir."/".$host;
1.1       foxr     1492: }
                   1493: 
1.3       albertel 1494: =pod
                   1495: 
                   1496: =head2 GetServerHost
                   1497: 
                   1498: Returns the host whose lond we talk with.
                   1499: 
1.1       foxr     1500: =cut
1.3       albertel 1501: 
1.7       foxr     1502: sub GetServerHost {
1.1       foxr     1503:     return $RemoteHost;		# Setup by the fork.
                   1504: }
1.3       albertel 1505: 
                   1506: =pod
                   1507: 
1.81      albertel 1508: =head2 GetServerId
                   1509: 
                   1510: Returns the hostid whose lond we talk with.
                   1511: 
                   1512: =cut
                   1513: 
                   1514: sub GetHostId {
                   1515:     return $RemoteHostId;		# Setup by the fork.
                   1516: }
                   1517: 
                   1518: =pod
                   1519: 
1.3       albertel 1520: =head2 GetServerPort
                   1521: 
                   1522: Returns the lond port number.
                   1523: 
1.1       foxr     1524: =cut
1.3       albertel 1525: 
1.7       foxr     1526: sub GetServerPort {
1.1       foxr     1527:     return $perlvar{londPort};
                   1528: }
1.3       albertel 1529: 
                   1530: =pod
                   1531: 
                   1532: =head2 SetupLoncListener
                   1533: 
                   1534: Setup a lonc listener event.  The event is called when the socket
                   1535: becomes readable.. that corresponds to the receipt of a new
                   1536: connection.  The event handler established will accept the connection
                   1537: (creating a communcations channel), that int turn will establish
                   1538: another event handler to subess requests.
1.1       foxr     1539: 
1.58      foxr     1540: =head2  Parameters:
                   1541: 
                   1542:    host (optional)   Name of the host to set up a unix socket to.
                   1543: 
1.1       foxr     1544: =cut
1.3       albertel 1545: 
1.1       foxr     1546: sub SetupLoncListener {
1.78      albertel 1547:     my ($host,$SocketName) = @_;
                   1548:     if (!$host) { $host = &GetServerHost(); }
                   1549:     if (!$SocketName) { $SocketName = &GetLoncSocketPath($host); }
1.1       foxr     1550: 
1.78      albertel 1551: 
                   1552:     unlink($SocketName);
1.58      foxr     1553: 
1.1       foxr     1554:     my $socket;
1.7       foxr     1555:     unless ($socket =IO::Socket::UNIX->new(Local  => $SocketName,
1.55      albertel 1556: 					    Listen => 250, 
1.1       foxr     1557: 					    Type   => SOCK_STREAM)) {
1.64      foxr     1558: 	if($I_am_child) {
                   1559: 	    &child_exit(-1, "Failed to create a lonc listener socket");
                   1560: 	} else {
                   1561: 	    die "Failed to create a lonc listner socket";
                   1562: 	}
1.1       foxr     1563:     }
1.59      foxr     1564:     return $socket;
1.1       foxr     1565: }
                   1566: 
1.39      foxr     1567: #
                   1568: #   Toggle transaction logging.
                   1569: #  Implicit inputs:  
                   1570: #     LogTransactions
                   1571: #  Implicit Outputs:
                   1572: #     LogTransactions
                   1573: sub ToggleTransactionLogging {
                   1574:     print STDERR "Toggle transaction logging...\n";
                   1575:     if(!$LogTransactions) {
                   1576: 	$LogTransactions = 1;
                   1577:     } else {
                   1578: 	$LogTransactions = 0;
                   1579:     }
                   1580: 
                   1581: 
                   1582:     Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
                   1583: }
                   1584: 
1.14      foxr     1585: =pod 
                   1586: 
                   1587: =head2 ChildStatus
                   1588:  
                   1589: Child USR1 signal handler to report the most recent status
                   1590: into the status file.
                   1591: 
1.22      foxr     1592: We also use this to reset the retries count in order to allow the
                   1593: client to retry connections with a previously dead server.
1.69      matthew  1594: 
1.14      foxr     1595: =cut
1.46      albertel 1596: 
1.14      foxr     1597: sub ChildStatus {
                   1598:     my $event = shift;
                   1599:     my $watcher = $event->w;
                   1600: 
                   1601:     Debug(2, "Reporting child status because : ".$watcher->data);
                   1602:     my $docdir = $perlvar{'lonDocRoot'};
1.67      albertel 1603:     
                   1604:     open(LOG,">>$docdir/lon-status/loncstatus.txt");
                   1605:     flock(LOG,LOCK_EX);
                   1606:     print LOG $$."\t".$RemoteHost."\t".$Status."\t".
1.14      foxr     1607: 	$RecentLogEntry."\n";
1.38      foxr     1608:     #
                   1609:     #  Write out information about each of the connections:
                   1610:     #
1.46      albertel 1611:     if ($DebugLevel > 2) {
1.67      albertel 1612: 	print LOG "Active connection statuses: \n";
1.46      albertel 1613: 	my $i = 1;
                   1614: 	print STDERR  "================================= Socket Status Dump:\n";
                   1615: 	foreach my $item (keys %ActiveConnections) {
                   1616: 	    my $Socket = $ActiveConnections{$item}->data;
                   1617: 	    my $state  = $Socket->GetState();
1.67      albertel 1618: 	    print LOG "Connection $i State: $state\n";
1.46      albertel 1619: 	    print STDERR "---------------------- Connection $i \n";
1.48      foxr     1620: 	    $Socket->Dump(-1);	# Ensure it gets dumped..
1.46      albertel 1621: 	    $i++;	
                   1622: 	}
1.38      foxr     1623:     }
1.67      albertel 1624:     flock(LOG,LOCK_UN);
                   1625:     close(LOG);
1.22      foxr     1626:     $ConnectionRetriesLeft = $ConnectionRetries;
1.70      albertel 1627:     UpdateStatus();
1.14      foxr     1628: }
                   1629: 
1.1       foxr     1630: =pod
1.3       albertel 1631: 
1.10      foxr     1632: =head2 SignalledToDeath
                   1633: 
                   1634: Called in response to a signal that causes a chid process to die.
                   1635: 
                   1636: =cut
                   1637: 
                   1638: 
                   1639: sub SignalledToDeath {
1.14      foxr     1640:     my $event  = shift;
                   1641:     my $watcher= $event->w;
                   1642: 
                   1643:     Debug(2,"Signalled to death! via ".$watcher->data);
1.17      foxr     1644:     my ($signal) = $watcher->data;
1.10      foxr     1645:     chomp($signal);
                   1646:     Log("CRITICAL", "Abnormal exit.  Child $$ for $RemoteHost "
                   1647: 	."died through "."\"$signal\"");
1.68      albertel 1648:     #LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
                   1649: #	    ."\"$signal\"");
1.12      foxr     1650:     exit 0;
1.10      foxr     1651: 
                   1652: }
1.16      foxr     1653: 
1.69      matthew  1654: =pod
                   1655: 
1.16      foxr     1656: =head2 ToggleDebug
                   1657: 
                   1658: This sub toggles trace debugging on and off.
                   1659: 
                   1660: =cut
                   1661: 
                   1662: sub ToggleDebug {
                   1663:     my $Current    = $DebugLevel;
                   1664:        $DebugLevel = $NextDebugLevel;
                   1665:        $NextDebugLevel = $Current;
                   1666: 
                   1667:     Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
                   1668: 
                   1669: }
                   1670: 
1.69      matthew  1671: =pod
                   1672: 
1.1       foxr     1673: =head2 ChildProcess
                   1674: 
                   1675: This sub implements a child process for a single lonc daemon.
1.61      foxr     1676: Optional parameter:
                   1677:    $socket  - if provided, this is a socket already open for listen
                   1678:               on the client socket. Otherwise, a new listen is set up.
1.1       foxr     1679: 
                   1680: =cut
                   1681: 
                   1682: sub ChildProcess {
1.80      albertel 1683:     #  We've inherited all the
1.62      foxr     1684:     #  events of our parent and those have to be cancelled or else
                   1685:     #  all holy bloody chaos will result.. trust me, I already made
                   1686:     #  >that< mistake.
                   1687: 
                   1688:     my $host = GetServerHost();
                   1689:     foreach my $listener (keys %parent_dispatchers) {
                   1690: 	my $watcher = $parent_dispatchers{$listener};
                   1691: 	my $s       = $watcher->fd;
                   1692: 	if ($listener ne $host) { # Close everyone but me.
                   1693: 	    Debug(5, "Closing listen socket for $listener");
                   1694: 	    $s->close();
                   1695: 	}
                   1696: 	Debug(5, "Killing watcher for $listener");
                   1697: 
                   1698: 	$watcher->cancel();
1.65      foxr     1699: 	delete($parent_dispatchers{$listener});
1.62      foxr     1700: 
                   1701:     }
1.65      foxr     1702: 
                   1703:     #  kill off the parent's signal handlers too!  
                   1704:     #
                   1705: 
                   1706:     for my $handler (keys %parent_handlers) {
                   1707: 	my $watcher = $parent_handlers{$handler};
                   1708: 	$watcher->cancel();
                   1709: 	delete($parent_handlers{$handler});
                   1710:     }
                   1711: 
1.64      foxr     1712:     $I_am_child    = 1;		# Seems like in spite of it all I may still getting
                   1713:                                 # parent event dispatches.. flag I'm a child.
1.1       foxr     1714: 
                   1715: 
1.14      foxr     1716:     #
                   1717:     #  Signals must be handled by the Event framework...
1.61      foxr     1718:     #
1.14      foxr     1719: 
                   1720:     Event->signal(signal   => "QUIT",
                   1721: 		  cb       => \&SignalledToDeath,
                   1722: 		  data     => "QUIT");
                   1723:     Event->signal(signal   => "HUP",
                   1724: 		  cb       => \&ChildStatus,
                   1725: 		  data     => "HUP");
                   1726:     Event->signal(signal   => "USR1",
                   1727: 		  cb       => \&ChildStatus,
                   1728: 		  data     => "USR1");
1.39      foxr     1729:     Event->signal(signal   => "USR2",
                   1730: 		  cb       => \&ToggleTransactionLogging);
1.16      foxr     1731:     Event->signal(signal   => "INT",
                   1732: 		  cb       => \&ToggleDebug,
                   1733: 		  data     => "INT");
1.1       foxr     1734: 
1.62      foxr     1735:     #  Figure out if we got passed a socket or need to open one to listen for
                   1736:     #  client requests.
                   1737: 
1.61      foxr     1738:     my ($socket) = @_;
                   1739:     if (!$socket) {
                   1740: 
                   1741: 	$socket =  SetupLoncListener();
                   1742:     }
1.62      foxr     1743:     #  Establish an event to listen for client connection requests.
                   1744: 
                   1745: 
1.59      foxr     1746:     Event->io(cb   => \&NewClient,
                   1747: 	      poll => 'r',
                   1748: 	      desc => 'Lonc Listener Unix Socket',
                   1749: 	      fd   => $socket);
1.1       foxr     1750:     
1.76      albertel 1751:     $Event::DebugLevel = $DebugLevel;
1.1       foxr     1752:     
                   1753:     Debug(9, "Making initial lond connection for ".$RemoteHost);
                   1754: 
                   1755: # Setup the initial server connection:
                   1756:     
1.62      foxr     1757:      # &MakeLondConnection(); // let first work request do it.
1.10      foxr     1758: 
1.80      albertel 1759:     #  need to accept the connection since the event may  not fire.
1.62      foxr     1760: 
1.80      albertel 1761:     &accept_client($socket);
1.5       foxr     1762: 
1.1       foxr     1763:     Debug(9,"Entering event loop");
                   1764:     my $ret = Event::loop();		#  Start the main event loop.
                   1765:     
                   1766:     
1.64      foxr     1767:     &child_exit (-1,"Main event loop exited!!!");
1.1       foxr     1768: }
                   1769: 
                   1770: #  Create a new child for host passed in:
                   1771: 
                   1772: sub CreateChild {
1.81      albertel 1773:     my ($host, $hostid) = @_;
1.52      foxr     1774: 
1.12      foxr     1775:     my $sigset = POSIX::SigSet->new(SIGINT);
                   1776:     sigprocmask(SIG_BLOCK, $sigset);
1.1       foxr     1777:     $RemoteHost = $host;
1.91    ! foxr     1778:     ShowStatus('Parent keeping the flock'); # Update time in status message.
1.9       foxr     1779:     Log("CRITICAL", "Forking server for ".$host);
1.23      foxr     1780:     my $pid          = fork;
1.1       foxr     1781:     if($pid) {			# Parent
1.17      foxr     1782: 	$RemoteHost = "Parent";
1.83      albertel 1783: 	$ChildPid{$pid} = $host;
1.12      foxr     1784: 	sigprocmask(SIG_UNBLOCK, $sigset);
1.82      albertel 1785: 	undef(@all_host_ids);
1.1       foxr     1786:     } else {			# child.
1.81      albertel 1787: 	$RemoteHostId = $hostid;
1.5       foxr     1788: 	ShowStatus("Connected to ".$RemoteHost);
1.23      foxr     1789: 	$SIG{INT} = 'DEFAULT';
1.12      foxr     1790: 	sigprocmask(SIG_UNBLOCK, $sigset);
1.81      albertel 1791: 	&ChildProcess();		# Does not return.
1.1       foxr     1792:     }
1.61      foxr     1793: }
1.1       foxr     1794: 
1.61      foxr     1795: # parent_client_connection:
                   1796: #    Event handler that processes client connections for the parent process.
                   1797: #    This sub is called when the parent is listening on a socket and
                   1798: #    a connection request arrives.  We must:
                   1799: #     Start a child process to accept the connection request.
                   1800: #     Kill our listen on the socket.
                   1801: # Parameter:
                   1802: #    event       - The event object that was created to monitor this socket.
                   1803: #                  event->w->fd is the socket.
                   1804: # Returns:
                   1805: #    NONE
                   1806: #
                   1807: sub parent_client_connection {
1.62      foxr     1808:     if ($I_am_child) {
                   1809: 	#  Should not get here, but seem to anyway:
                   1810: 	&Debug(5," Child caught parent client connection event!!");
                   1811: 	my ($event) = @_;
                   1812: 	my $watcher = $event->w;
                   1813: 	$watcher->cancel();	# Try to kill it off again!!
                   1814:     } else {
                   1815: 	&Debug(9, "parent_client_connection");
                   1816: 	my ($event)   = @_;
                   1817: 	my $watcher   = $event->w;
                   1818: 	my $socket    = $watcher->fd;
1.81      albertel 1819: 	my $connection = $socket->accept();	# Accept the client connection.
                   1820: 	Event->io(cb      => \&get_remote_hostname,
                   1821: 		  poll    => 'r',
                   1822: 		  data    => "",
                   1823: 		  fd      => $connection);
1.77      albertel 1824:     }
                   1825: }
                   1826: 
                   1827: sub get_remote_hostname {
1.82      albertel 1828:     my ($event)   = @_;
                   1829:     my $watcher   = $event->w;
                   1830:     my $socket    = $watcher->fd;
                   1831: 
                   1832:     my $thisread;
                   1833:     my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
                   1834:     Debug(8, "rcv:  data length = ".length($thisread)." read =".$thisread);
                   1835:     if (!defined($rv) || length($thisread) == 0) {
                   1836: 	# Likely eof on socket.
                   1837: 	Debug(5,"Client Socket closed on lonc for p_c_c");
                   1838: 	close($socket);
                   1839: 	$watcher->cancel();
                   1840: 	return;
                   1841:     }
                   1842: 
                   1843:     my $data    = $watcher->data().$thisread;
                   1844:     $watcher->data($data);
                   1845:     if($data =~ /\n$/) {	# Request entirely read.
                   1846: 	chomp($data);
                   1847:     } else {
                   1848: 	return;
                   1849:     }
1.77      albertel 1850: 
1.82      albertel 1851:     &Debug(5,"Creating child for $data (parent_client_connection)");
                   1852:     (my $hostname,my $lonid,@all_host_ids) = split(':',$data);
1.83      albertel 1853:     $ChildHost{$hostname}++;
                   1854:     if ($ChildHost{$hostname} == 1) {
                   1855: 	&CreateChild($hostname,$lonid);
                   1856:     } else {
                   1857: 	&Log('WARNING',"Request for a second child on $hostname");
                   1858:     }
1.82      albertel 1859:     # Clean up the listen since now the child takes over until it exits.
                   1860:     $watcher->cancel();		# Nolonger listening to this event
                   1861:     $socket->send("done\n");
                   1862:     $socket->close();
1.61      foxr     1863: }
                   1864: 
                   1865: # parent_listen:
                   1866: #    Opens a socket and starts a listen for the parent process on a client UNIX
                   1867: #    domain socket.
                   1868: #
                   1869: #    This involves:
                   1870: #       Creating a socket for listen.
                   1871: #       Removing any socket lock file
                   1872: #       Adding an event handler for this socket becoming readable
                   1873: #         To the parent's event dispatcher.
                   1874: # Parameters:
                   1875: #    loncapa_host    - LonCAPA cluster name of the host represented by the client
                   1876: #                      socket.
                   1877: # Returns:
                   1878: #    NONE
                   1879: #
                   1880: sub parent_listen {
                   1881:     my ($loncapa_host) = @_;
                   1882:     Debug(5, "parent_listen: $loncapa_host");
                   1883: 
1.78      albertel 1884:     my ($socket,$file);
                   1885:     if (!$loncapa_host) {
                   1886: 	$loncapa_host = 'common_parent';
                   1887: 	$file         = $perlvar{'lonSockCreate'};
                   1888:     } else {
                   1889: 	$file         = &GetLoncSocketPath($loncapa_host);
                   1890:     }
                   1891:     $socket = &SetupLoncListener($loncapa_host,$file);
                   1892: 
1.62      foxr     1893:     $listening_to{$socket} = $loncapa_host;
1.61      foxr     1894:     if (!$socket) {
                   1895: 	die "Unable to create a listen socket for $loncapa_host";
                   1896:     }
                   1897:     
1.78      albertel 1898:     my $lock_file = $file.".lock";
1.61      foxr     1899:     unlink($lock_file);		# No problem if it doesn't exist yet [startup e.g.]
                   1900: 
1.77      albertel 1901:     my $watcher = 
                   1902: 	Event->io(cb    => \&parent_client_connection,
                   1903: 		  poll  => 'r',
                   1904: 		  desc  => "Parent listener unix socket ($loncapa_host)",
                   1905: 		  data => "",
                   1906: 		  fd    => $socket);
1.62      foxr     1907:     $parent_dispatchers{$loncapa_host} = $watcher;
1.61      foxr     1908: 
                   1909: }
                   1910: 
1.77      albertel 1911: sub parent_clean_up {
                   1912:     my ($loncapa_host) = @_;
1.87      albertel 1913:     Debug(1, "parent_clean_up: $loncapa_host");
1.77      albertel 1914: 
                   1915:     my $socket_file = &GetLoncSocketPath($loncapa_host);
                   1916:     unlink($socket_file);	# No problem if it doesn't exist yet [startup e.g.]
                   1917:     my $lock_file   = $socket_file.".lock";
                   1918:     unlink($lock_file);		# No problem if it doesn't exist yet [startup e.g.]
                   1919: }
                   1920: 
1.61      foxr     1921: 
1.83      albertel 1922: 
                   1923: #    This sub initiates a listen on the common unix domain lonc client socket.
                   1924: #    loncnew starts up with no children, and only spawns off children when a
                   1925: #    connection request occurs on the common client unix socket.  The spawned
                   1926: #    child continues to run until it has been idle a while at which point it
                   1927: #    eventually exits and once more the parent picks up the listen.
1.61      foxr     1928: #
                   1929: #  Parameters:
                   1930: #      NONE
                   1931: #  Implicit Inputs:
                   1932: #    The configuration file that has been read in by LondConnection.
                   1933: #  Returns:
                   1934: #     NONE
                   1935: #
1.77      albertel 1936: sub listen_on_common_socket {
                   1937:     Debug(5, "listen_on_common_socket");
1.78      albertel 1938:     &parent_listen();
1.77      albertel 1939: }
                   1940: 
1.63      foxr     1941: #   server_died is called whenever a child process exits.
                   1942: #   Since this is dispatched via a signal, we must process all
                   1943: #   dead children until there are no more left.  The action
                   1944: #   is to:
                   1945: #      - Remove the child from the bookeeping hashes
                   1946: #      - Re-establish a listen on the unix domain socket associated
                   1947: #        with that host.
                   1948: # Parameters:
                   1949: #    The event, but we don't actually care about it.
                   1950: sub server_died {
                   1951:     &Debug(9, "server_died called...");
                   1952:     
                   1953:     while(1) {			# Loop until waitpid nowait fails.
                   1954: 	my $pid = waitpid(-1, WNOHANG);
                   1955: 	if($pid <= 0) {
                   1956: 	    return;		# Nothing left to wait for.
                   1957: 	}
                   1958: 	# need the host to restart:
                   1959: 
1.83      albertel 1960: 	my $host = $ChildPid{$pid};
1.63      foxr     1961: 	if($host) {		# It's for real...
                   1962: 	    &Debug(9, "Caught sigchild for $host");
1.83      albertel 1963: 	    delete($ChildPid{$pid});
                   1964: 	    delete($ChildHost{$host});
1.81      albertel 1965: 	    &parent_clean_up($host);
                   1966: 
1.63      foxr     1967: 	} else {
                   1968: 	    &Debug(5, "Caught sigchild for pid not in hosts hash: $pid");
                   1969: 	}
                   1970:     }
                   1971: 
                   1972: }
                   1973: 
1.1       foxr     1974: #
                   1975: #  Parent process logic pass 1:
                   1976: #   For each entry in the hosts table, we will
                   1977: #  fork off an instance of ChildProcess to service the transactions
                   1978: #  to that host.  Each pid will be entered in a global hash
                   1979: #  with the value of the key, the host.
                   1980: #  The parent will then enter a loop to wait for process exits.
                   1981: #  Each exit gets logged and the child gets restarted.
                   1982: #
                   1983: 
1.5       foxr     1984: #
                   1985: #   Fork and start in new session so hang-up isn't going to 
                   1986: #   happen without intent.
                   1987: #
                   1988: 
                   1989: 
1.6       foxr     1990: 
                   1991: 
1.8       foxr     1992: 
1.6       foxr     1993: 
                   1994: ShowStatus("Forming new session");
                   1995: my $childpid = fork;
                   1996: if ($childpid != 0) {
                   1997:     sleep 4;			# Give child a chacne to break to
                   1998:     exit 0;			# a new sesion.
                   1999: }
1.8       foxr     2000: #
                   2001: #   Write my pid into the pid file so I can be located
                   2002: #
                   2003: 
                   2004: ShowStatus("Parent writing pid file:");
1.23      foxr     2005: my $execdir = $perlvar{'lonDaemons'};
1.8       foxr     2006: open (PIDSAVE, ">$execdir/logs/lonc.pid");
                   2007: print PIDSAVE "$$\n";
                   2008: close(PIDSAVE);
1.6       foxr     2009: 
1.17      foxr     2010: 
                   2011: 
1.6       foxr     2012: if (POSIX::setsid() < 0) {
                   2013:     print "Could not create new session\n";
                   2014:     exit -1;
                   2015: }
1.5       foxr     2016: 
                   2017: ShowStatus("Forking node servers");
                   2018: 
1.9       foxr     2019: Log("CRITICAL", "--------------- Starting children ---------------");
                   2020: 
1.31      foxr     2021: LondConnection::ReadConfig;               # Read standard config files.
1.1       foxr     2022: 
1.80      albertel 2023: $RemoteHost = "[parent]";
1.81      albertel 2024: &listen_on_common_socket();
1.60      foxr     2025: 
1.12      foxr     2026: $RemoteHost = "Parent Server";
1.1       foxr     2027: 
                   2028: # Maintain the population:
1.5       foxr     2029: 
                   2030: ShowStatus("Parent keeping the flock");
1.1       foxr     2031: 
1.12      foxr     2032: 
1.80      albertel 2033: # We need to setup a SIGChild event to handle the exit (natural or otherwise)
                   2034: # of the children.
1.61      foxr     2035: 
1.80      albertel 2036: Event->signal(cb       => \&server_died,
                   2037: 	      desc     => "Child exit handler",
                   2038: 	      signal   => "CHLD");
                   2039: 
                   2040: 
                   2041: # Set up all the other signals we set up.
                   2042: 
                   2043: $parent_handlers{INT} = Event->signal(cb       => \&Terminate,
                   2044: 				      desc     => "Parent INT handler",
                   2045: 				      signal   => "INT");
                   2046: $parent_handlers{TERM} = Event->signal(cb       => \&Terminate,
                   2047: 				       desc     => "Parent TERM handler",
                   2048: 				       signal   => "TERM");
1.81      albertel 2049: $parent_handlers{HUP}  = Event->signal(cb       => \&KillThemAll,
                   2050: 				       desc     => "Parent HUP handler.",
                   2051: 				       signal   => "HUP");
1.80      albertel 2052: $parent_handlers{USR1} = Event->signal(cb       => \&CheckKids,
                   2053: 				       desc     => "Parent USR1 handler",
                   2054: 				       signal   => "USR1");
                   2055: $parent_handlers{USR2} = Event->signal(cb       => \&UpdateKids,
                   2056: 				       desc     => "Parent USR2 handler.",
                   2057: 				       signal   => "USR2");
                   2058: 
                   2059: #  Start procdesing events.
                   2060: 
                   2061: $Event::DebugLevel = $DebugLevel;
                   2062: Debug(9, "Parent entering event loop");
                   2063: my $ret = Event::loop();
                   2064: die "Main Event loop exited: $ret";
1.14      foxr     2065: 
                   2066: =pod
                   2067: 
                   2068: =head1 CheckKids
                   2069: 
                   2070:   Since kids do not die as easily in this implementation
                   2071: as the previous one, there  is no need to restart the
                   2072: dead ones (all dead kids get restarted when they die!!)
                   2073: The only thing this function does is to pass USR1 to the
                   2074: kids so that they report their status.
                   2075: 
                   2076: =cut
                   2077: 
                   2078: sub CheckKids {
                   2079:     Debug(2, "Checking status of children");
                   2080:     my $docdir = $perlvar{'lonDocRoot'};
                   2081:     my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
                   2082:     my $now=time;
                   2083:     my $local=localtime($now);
                   2084:     print $fh "LONC status $local - parent $$ \n\n";
1.65      foxr     2085:     foreach my $host (keys %parent_dispatchers) {
                   2086: 	print $fh "LONC Parent process listening for $host\n";
                   2087:     }
1.83      albertel 2088:     foreach my $pid (keys %ChildPid) {
1.14      foxr     2089: 	Debug(2, "Sending USR1 -> $pid");
                   2090: 	kill 'USR1' => $pid;	# Tell Child to report status.
                   2091:     }
1.65      foxr     2092: 
1.14      foxr     2093: }
1.24      foxr     2094: 
                   2095: =pod
                   2096: 
                   2097: =head1  UpdateKids
                   2098: 
1.25      foxr     2099: parent's SIGUSR2 handler.  This handler:
1.24      foxr     2100: 
                   2101: =item
                   2102: 
                   2103: Rereads the hosts file.
                   2104: 
                   2105: =item
                   2106:  
                   2107: Kills off (via sigint) children for hosts that have disappeared.
                   2108: 
                   2109: =item
                   2110: 
1.27      foxr     2111: QUITs  children for hosts that already exist (this just forces a status display
1.24      foxr     2112: and resets the connection retry count for that host.
                   2113: 
                   2114: =item
                   2115: 
                   2116: Starts new children for hosts that have been added to the hosts.tab file since
                   2117: the start of the master program and maintains them.
                   2118: 
                   2119: =cut
                   2120: 
                   2121: sub UpdateKids {
1.27      foxr     2122: 
1.25      foxr     2123:     Log("INFO", "Updating connections via SIGUSR2");
1.27      foxr     2124: 
1.65      foxr     2125:     #  I'm not sure what I was thinking in the first implementation.
                   2126:     # someone will have to work hard to convince me the effect is any
                   2127:     # different than Restart, especially now that we don't start up 
                   2128:     # per host servers automatically, may as well just restart.
                   2129:     # The down side is transactions that are in flight will get timed out
                   2130:     # (lost unless they are critical).
1.27      foxr     2131: 
1.81      albertel 2132:     &KillThemAll();
1.24      foxr     2133: }
                   2134: 
1.14      foxr     2135: 
1.13      foxr     2136: =pod
                   2137: 
                   2138: =head1 Restart
                   2139: 
                   2140: Signal handler for HUP... all children are killed and
                   2141: we self restart.  This is an el-cheapo way to re read
                   2142: the config file.
                   2143: 
                   2144: =cut
                   2145: 
                   2146: sub Restart {
1.23      foxr     2147:     &KillThemAll;		# First kill all the children.
1.13      foxr     2148:     Log("CRITICAL", "Restarting");
                   2149:     my $execdir = $perlvar{'lonDaemons'};
                   2150:     unlink("$execdir/logs/lonc.pid");
1.65      foxr     2151:     exec("$executable");
1.10      foxr     2152: }
1.12      foxr     2153: 
                   2154: =pod
                   2155: 
                   2156: =head1 KillThemAll
                   2157: 
                   2158: Signal handler that kills all children by sending them a 
1.17      foxr     2159: SIGHUP.  Responds to sigint and sigterm.
1.12      foxr     2160: 
                   2161: =cut
                   2162: 
1.10      foxr     2163: sub KillThemAll {
1.12      foxr     2164:     Debug(2, "Kill them all!!");
1.85      albertel 2165:     
                   2166:     #local($SIG{CHLD}) = 'IGNORE';
                   2167:     # Our children >will< die.
                   2168:     # but we need to catch their death and cleanup after them in case this is 
                   2169:     # a restart set of kills
                   2170:     my @allpids = keys(%ChildPid);
                   2171:     foreach my $pid (@allpids) {
1.83      albertel 2172: 	my $serving = $ChildPid{$pid};
1.52      foxr     2173: 	ShowStatus("Nicely Killing lonc for $serving pid = $pid");
                   2174: 	Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
1.17      foxr     2175: 	kill 'QUIT' => $pid;
1.12      foxr     2176:     }
1.85      albertel 2177:     ShowStatus("Finished killing child processes off.");
1.1       foxr     2178: }
1.12      foxr     2179: 
1.52      foxr     2180: 
                   2181: #
                   2182: #  Kill all children via KILL.  Just in case the
                   2183: #  first shot didn't get them.
                   2184: 
                   2185: sub really_kill_them_all_dammit
                   2186: {
                   2187:     Debug(2, "Kill them all Dammit");
                   2188:     local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
1.83      albertel 2189:     foreach my $pid (keys %ChildPid) {
                   2190: 	my $serving = $ChildPid{$pid};
1.52      foxr     2191: 	&ShowStatus("Nastily killing lonc for $serving pid = $pid");
                   2192: 	Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
                   2193: 	kill 'KILL' => $pid;
1.83      albertel 2194: 	delete($ChildPid{$pid});
1.52      foxr     2195: 	my $execdir = $perlvar{'lonDaemons'};
                   2196: 	unlink("$execdir/logs/lonc.pid");
                   2197:     }
                   2198: }
1.69      matthew  2199: 
1.14      foxr     2200: =pod
                   2201: 
                   2202: =head1 Terminate
                   2203:  
                   2204: Terminate the system.
                   2205: 
                   2206: =cut
                   2207: 
                   2208: sub Terminate {
1.52      foxr     2209:     &Log("CRITICAL", "Asked to kill children.. first be nice...");
                   2210:     &KillThemAll;
                   2211:     #
                   2212:     #  By now they really should all be dead.. but just in case 
                   2213:     #  send them all SIGKILL's after a bit of waiting:
                   2214: 
                   2215:     sleep(4);
                   2216:     &Log("CRITICAL", "Now kill children nasty");
                   2217:     &really_kill_them_all_dammit;
1.17      foxr     2218:     Log("CRITICAL","Master process exiting");
                   2219:     exit 0;
1.14      foxr     2220: 
                   2221: }
1.81      albertel 2222: 
                   2223: sub my_hostname {
                   2224:     use Sys::Hostname;
                   2225:     my $name = &hostname();
                   2226:     &Debug(9,"Name is $name");
                   2227:     return $name;
                   2228: }
                   2229: 
1.12      foxr     2230: =pod
1.1       foxr     2231: 
                   2232: =head1 Theory
1.3       albertel 2233: 
                   2234: The event class is used to build this as a single process with an
                   2235: event driven model.  The following events are handled:
1.1       foxr     2236: 
                   2237: =item UNIX Socket connection Received
                   2238: 
                   2239: =item Request data arrives on UNIX data transfer socket.
                   2240: 
                   2241: =item lond connection becomes writable.
                   2242: 
                   2243: =item timer fires at 1 second intervals.
                   2244: 
                   2245: All sockets are run in non-blocking mode.  Timeouts managed by the timer
                   2246: handler prevents hung connections.
                   2247: 
                   2248: Key data structures:
                   2249: 
1.3       albertel 2250: =item RequestQueue
                   2251: 
                   2252: A queue of requests received from UNIX sockets that are
                   2253: waiting for a chance to be forwarded on a lond connection socket.
                   2254: 
                   2255: =item ActiveConnections
                   2256: 
                   2257: A hash of lond connections that have transactions in process that are
                   2258: available to be timed out.
                   2259: 
                   2260: =item ActiveTransactions
                   2261: 
                   2262: A hash indexed by lond connections that contain the client reply
                   2263: socket for each connection that has an active transaction on it.
                   2264: 
                   2265: =item IdleConnections
                   2266: 
                   2267: A hash of lond connections that have no work to do.  These connections
                   2268: can be closed if they are idle for a long enough time.
1.1       foxr     2269: 
                   2270: =cut
1.88      foxr     2271: 
                   2272: =pod
                   2273: 
                   2274: =head1 Log messages
                   2275: 
                   2276: The following is a list of log messages that can appear in the 
                   2277: lonc.log file.  Each log file has a severity and a message.
                   2278: 
                   2279: =over 2
                   2280: 
                   2281: =item Warning  A socket timeout was detected
                   2282: 
                   2283: If there are pending transactions in the socket's queue,
                   2284: they are failed (saved if critical).  If the connection
                   2285: retry count gets exceeded by this, the
                   2286: remote host is marked as dead.
                   2287: Called when timeouts occured during the connection and
                   2288: connection dialog with a remote host.
                   2289: 
                   2290: =item Critical Host makred DEAD <hostname>   
                   2291: 
                   2292: The numer of retry counts for contacting a host was
                   2293: exceeded. The host is marked dead an no 
                   2294: further attempts will be made by that child.
                   2295: 
                   2296: =item Info lonc pipe client hung up on us     
                   2297: 
                   2298: Write to the client pipe indicated no data transferred
                   2299: Socket to remote host is shut down.  Reply to the client 
                   2300: is discarded.  Note: This is commented out in &ClientWriteable
                   2301: 
                   2302: =item Success  Reply from lond: <data>   
                   2303: 
                   2304: Can be enabled for debugging by setting LogTransactions to nonzero.
                   2305: Indicates a successful transaction with lond, <data> is the data received
                   2306: from the remote lond.
                   2307: 
                   2308: =item Success A delayed transaction was completed  
                   2309: 
                   2310: A transaction that must be reliable was executed and completed
                   2311: as lonc restarted.  This is followed by a mesage of the form
                   2312: 
                   2313:   S: client-name : request
                   2314: 
                   2315: =item WARNING  Failing transaction <cmd>:<subcmd>  
                   2316: 
                   2317: Transaction failed on a socket, but the failure retry count for the remote
                   2318: node has not yet been exhausted (the node is not yet marked dead).
                   2319: cmd is the command, subcmd is the subcommand.  This results from a con_lost
                   2320: when communicating with lond.
                   2321: 
                   2322: =item WARNING Shutting down a socket     
                   2323: 
                   2324: Called when a socket is being closed to lond.  This is emitted both when 
                   2325: idle pruning is being done and when the socket has been disconnected by the remote.
                   2326: 
                   2327: =item WARNING Lond connection lost.
                   2328: 
                   2329: Called when a read from lond's socket failed indicating lond has closed the 
                   2330: connection or died.  This should be followed by one or more
                   2331: 
                   2332:  "WARNING Failing transaction..." msgs for each in-flight or queued transaction.
                   2333: 
                   2334: =item INFO Connected to lond version:  <version> 
                   2335: 
                   2336: When connection negotiation is complete, the lond version is requested and logged here.
                   2337: 
                   2338: =item SUCCESS Connection n to host now ready for action
                   2339: 
                   2340: Emitted when connection has been completed with lond. n is then number of 
                   2341: concurrent connections and host, the host to which the connection has just
                   2342: been established.
                   2343: 
                   2344: =item WARNING Connection to host has been disconnected
                   2345: 
                   2346: Write to a lond resulted in failure status.  Connection to lond is dropped.
                   2347: 
                   2348: =item SUCCESS Created connection n to host host 
                   2349: 
                   2350: Initial connection request to host..(before negotiation).
                   2351: 
                   2352: =item CRITICAL Request Close Connection ... exiting
                   2353: 
                   2354: Client has sent "close_connection_exit"   The loncnew server is exiting.
                   2355: 
                   2356: =item INFO Resetting Connection Retries 
                   2357: 
                   2358: Client has sent "reset_retries" The lond connection retries are reset to zero for the
                   2359: corresponding lond.
                   2360: 
                   2361: =item SUCCESS Transaction <data>
                   2362: 
                   2363: Only emitted if the global variable $LogTransactions was set to true.
                   2364: A client has requested a lond transaction <data> is the contents of the request.
                   2365: 
                   2366: =item SUCCESS Toggled transaction logging <LogTransactions>
                   2367:                                     
                   2368: The state of the $LogTransactions global has been toggled, and its current value
                   2369: (after being toggled) is displayed.  When non zero additional logging of transactions
                   2370: is enabled for debugging purposes.  Transaction logging is toggled on receipt of a USR2
                   2371: signal.
                   2372: 
                   2373: =item CRITICAL Abnormal exit. Child <pid> for <host> died thorugh signal.
                   2374: 
                   2375: QUIT signal received.  lonc child process is exiting.
                   2376: 
                   2377: =item SUCCESS New debugging level for <RemoteHost> now <DebugLevel>
                   2378:                                     
                   2379: Debugging toggled for the host loncnew is talking with.
                   2380: Currently debugging is a level based scheme with higher number 
                   2381: conveying more information.  The daemon starts out at
                   2382: DebugLevel 0 and can toggle back and forth between that and
                   2383: DebugLevel 2  These are controlled by
                   2384: the global variables $DebugLevel and $NextDebugLevel
                   2385: The debug level can go up to 9.
                   2386: SIGINT toggles the debug level.  The higher the debug level the 
                   2387: more debugging information is spewed.  See the Debug
                   2388: sub in loncnew.
                   2389: 
                   2390: =item CRITICAL Forking server for host  
                   2391: 
                   2392: A child is being created to service requests for the specified host.
                   2393: 
                   2394: 
                   2395: =item WARNING Request for a second child on hostname
                   2396:                                     
                   2397: Somehow loncnew was asked to start a second child on a host that already had a child
                   2398: servicing it.  This request is not honored, but themessage is emitted.  This could happen
                   2399: due to a race condition.  When a client attempts to contact loncnew for a new host, a child
                   2400: is forked off to handle the requests for that server.  The parent then backs off the Unix
                   2401: domain socket leaving it for the child to service all requests.  If in the time between
                   2402: creating the child, and backing off, a new connection request comes in to the unix domain
                   2403: socket, this could trigger (unlikely but remotely possible),.
                   2404: 
                   2405: =item CRITICAL ------ Starting Children ----
                   2406: 
                   2407: This message should probably be changed to "Entering event loop"  as the loncnew only starts
                   2408: children as needed.  This message is emitted as new events are established and
                   2409: the event processing loop is entered.
                   2410: 
                   2411: =item INFO Updating connections via SIGUSR2
                   2412:                                     
                   2413: SIGUSR2 received. The original code would kill all clients, re-read the host file,
                   2414: then restart children for each host.  Now that childrean aree started on demand, this
                   2415: just kills all child processes and lets requests start them as needed again.
                   2416: 
                   2417: 
                   2418: =item CRITICAL Restarting
                   2419: 
                   2420: SigHUP received.  all the children are killed and the script exec's itself to start again.
                   2421: 
                   2422: =item CRITICAL Nicely killing lonc for host pid = <pid>
                   2423: 
                   2424: Attempting to kill the child that is serving the specified host (pid given) cleanly via
                   2425: SIGQUIT  The child should handle that, clean up nicely and exit.
                   2426: 
                   2427: =item CRITICAL Nastily killing lonc for host pid = <pid>
                   2428: 
                   2429: The child specified did not die when requested via SIGQUIT.  Therefore it is killed
                   2430: via SIGKILL.
                   2431: 
                   2432: =item CRITICAL Asked to kill children.. first be nice..
                   2433: 
                   2434: In the parent's INT handler.  INT kills the child processes.  This inidicate loncnew
                   2435: is about to attempt to kill all known children via SIGQUIT.  This message should be followed 
                   2436: by one "Nicely killing" message for each extant child.
                   2437: 
                   2438: =item CRITICAL Now kill children nasty 
                   2439: 
                   2440: In the parent's INT handler. remaining children are about to be killed via
                   2441: SIGKILL. Should be followed by a Nastily killing... for each lonc child that 
                   2442: refused to die.
                   2443: 
                   2444: =item CRITICAL Master process exiting
                   2445: 
                   2446: In the parent's INT handler. just prior to the exit 0 call.
                   2447: 
                   2448: =back
                   2449: 
                   2450: =cut

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>