Annotation of loncom/loncnew, revision 1.92

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: #
1.92    ! foxr        5: # $Id: loncnew,v 1.91 2010/12/20 11:31:52 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");
1.92    ! foxr      444:     Event->timer(interval => 1, cb => \&Tick,
        !           445: 	hard => 1);
1.1       foxr      446: }
1.3       albertel  447: 
1.1       foxr      448: =pod
1.3       albertel  449: 
1.1       foxr      450: =head2 ServerToIdle
1.3       albertel  451: 
                    452: This function is called when a connection to the server is
                    453: ready for more work.
                    454: 
                    455: If there is work in the Work queue the top element is dequeued
1.1       foxr      456: and the connection will start to work on it.  If the work queue is
                    457: empty, the connection is pushed on the idle connection stack where
                    458: it will either get another work unit, or alternatively, if it sits there
                    459: long enough, it will be shut down and released.
                    460: 
1.3       albertel  461: =cut
1.1       foxr      462: 
                    463: sub ServerToIdle {
                    464:     my $Socket   = shift;	# Get the socket.
1.49      foxr      465:     $KeyMode = $Socket->{AuthenticationMode};
1.7       foxr      466:     delete($ActiveTransactions{$Socket}); # Server has no transaction
1.1       foxr      467: 
1.29      foxr      468:     &Debug(5, "Server to idle");
1.1       foxr      469: 
                    470:     #  If there's work to do, start the transaction:
                    471: 
1.23      foxr      472:     my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
1.29      foxr      473:     if ($reqdata ne undef)  {
                    474: 	Debug(5, "Queue gave request data: ".$reqdata->getRequest());
1.7       foxr      475: 	&StartRequest($Socket,  $reqdata);
1.8       foxr      476: 
1.1       foxr      477:     } else {
                    478: 	
                    479:     #  There's no work waiting, so push the server to idle list.
1.29      foxr      480: 	&Debug(5, "No new work requests, server connection going idle");
1.1       foxr      481: 	$IdleConnections->push($Socket);
                    482:     }
                    483: }
1.3       albertel  484: 
1.1       foxr      485: =pod
1.3       albertel  486: 
1.1       foxr      487: =head2 ClientWritable
1.3       albertel  488: 
                    489: Event callback for when a client socket is writable.
                    490: 
                    491: This callback is established when a transaction reponse is
                    492: avaiable from lond.  The response is forwarded to the unix socket
                    493: as it becomes writable in this sub.
                    494: 
1.1       foxr      495: Parameters:
                    496: 
1.3       albertel  497: =item Event
                    498: 
                    499: The event that has been triggered. Event->w->data is
                    500: the data and Event->w->fd is the socket to write.
1.1       foxr      501: 
                    502: =cut
1.3       albertel  503: 
1.1       foxr      504: sub ClientWritable {
                    505:     my $Event    = shift;
                    506:     my $Watcher  = $Event->w;
1.84      albertel  507:     if (!defined($Watcher)) {
                    508: 	&child_exit(-1,'No watcher for event in ClientWritable');
                    509:     }
1.1       foxr      510:     my $Data     = $Watcher->data;
                    511:     my $Socket   = $Watcher->fd;
                    512: 
                    513:     # Try to send the data:
                    514: 
                    515:     &Debug(6, "ClientWritable writing".$Data);
                    516:     &Debug(9, "Socket is: ".$Socket);
                    517: 
1.6       foxr      518:     if($Socket->connected) {
                    519: 	my $result = $Socket->send($Data, 0);
                    520: 	
                    521: 	# $result undefined: the write failed.
                    522: 	# otherwise $result is the number of bytes written.
                    523: 	# Remove that preceding string from the data.
                    524: 	# If the resulting data is empty, destroy the watcher
                    525: 	# and set up a read event handler to accept the next
                    526: 	# request.
                    527: 	
                    528: 	&Debug(9,"Send result is ".$result." Defined: ".defined($result));
1.29      foxr      529: 	if($result ne undef) {
1.6       foxr      530: 	    &Debug(9, "send result was defined");
                    531: 	    if($result == length($Data)) { # Entire string sent.
                    532: 		&Debug(9, "ClientWritable data all written");
                    533: 		$Watcher->cancel();
                    534: 		#
                    535: 		#  Set up to read next request from socket:
                    536: 		
                    537: 		my $descr     = sprintf("Connection to lonc client %d",
                    538: 					$ActiveClients{$Socket});
                    539: 		Event->io(cb    => \&ClientRequest,
                    540: 			  poll  => 'r',
                    541: 			  desc  => $descr,
                    542: 			  data  => "",
                    543: 			  fd    => $Socket);
                    544: 		
                    545: 	    } else {		# Partial string sent.
                    546: 		$Watcher->data(substr($Data, $result));
1.15      foxr      547: 		if($result == 0) {    # client hung up on us!!
1.52      foxr      548: 		    # Log("INFO", "lonc pipe client hung up on us!");
1.15      foxr      549: 		    $Watcher->cancel;
                    550: 		    $Socket->shutdown(2);
                    551: 		    $Socket->close();
                    552: 		}
1.6       foxr      553: 	    }
                    554: 	    
                    555: 	} else {			# Error of some sort...
                    556: 	    
                    557: 	    # Some errnos are possible:
                    558: 	    my $errno = $!;
                    559: 	    if($errno == POSIX::EWOULDBLOCK   ||
                    560: 	       $errno == POSIX::EAGAIN        ||
                    561: 	       $errno == POSIX::EINTR) {
                    562: 		# No action taken?
                    563: 	    } else {		# Unanticipated errno.
                    564: 		&Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
                    565: 		$Watcher->cancel;	# Stop the watcher.
                    566: 		$Socket->shutdown(2); # Kill connection
                    567: 		$Socket->close();	# Close the socket.
                    568: 	    }
1.1       foxr      569: 	    
                    570: 	}
1.6       foxr      571:     } else {
                    572: 	$Watcher->cancel();	# A delayed request...just cancel.
1.84      albertel  573: 	return;
1.1       foxr      574:     }
                    575: }
                    576: 
                    577: =pod
1.3       albertel  578: 
1.1       foxr      579: =head2 CompleteTransaction
1.3       albertel  580: 
                    581: Called when the reply data has been received for a lond 
1.1       foxr      582: transaction.   The reply data must now be sent to the
                    583: ultimate client on the other end of the Unix socket.  This is
                    584: done by setting up a writable event for the socket with the
                    585: data the reply data.
1.3       albertel  586: 
1.1       foxr      587: Parameters:
1.3       albertel  588: 
                    589: =item Socket
                    590: 
                    591: Socket on which the lond transaction occured.  This is a
                    592: LondConnection. The data received is in the TransactionReply member.
                    593: 
1.7       foxr      594: =item Transaction
1.3       albertel  595: 
1.7       foxr      596: The transaction that is being completed.
1.1       foxr      597: 
                    598: =cut
1.3       albertel  599: 
1.1       foxr      600: sub CompleteTransaction {
1.29      foxr      601:     &Debug(5,"Complete transaction");
1.47      foxr      602: 
                    603:     my ($Socket, $Transaction) = @_;
1.1       foxr      604: 
1.7       foxr      605:     if (!$Transaction->isDeferred()) { # Normal transaction
                    606: 	my $data   = $Socket->GetReply(); # Data to send.
1.39      foxr      607: 	if($LogTransactions) {
                    608: 	    Log("SUCCESS", "Reply from lond: '$data'");
                    609: 	}
1.7       foxr      610: 	StartClientReply($Transaction, $data);
                    611:     } else {			# Delete deferred transaction file.
1.9       foxr      612: 	Log("SUCCESS", "A delayed transaction was completed");
1.86      albertel  613: 	LogPerm("S:".$Transaction->getClient().":".$Transaction->getRequest());
                    614: 	unlink($Transaction->getFile());
1.7       foxr      615:     }
1.6       foxr      616: }
1.42      foxr      617: 
1.6       foxr      618: =pod
1.42      foxr      619: 
1.6       foxr      620: =head1 StartClientReply
                    621: 
                    622:    Initiates a reply to a client where the reply data is a parameter.
                    623: 
1.7       foxr      624: =head2  parameters:
                    625: 
                    626: =item Transaction
                    627: 
                    628:     The transaction for which we are responding to the client.
                    629: 
                    630: =item data
                    631: 
                    632:     The data to send to apached client.
                    633: 
1.6       foxr      634: =cut
1.42      foxr      635: 
1.6       foxr      636: sub StartClientReply {
1.1       foxr      637: 
1.47      foxr      638:     my ($Transaction, $data) = @_;
1.12      foxr      639: 
1.7       foxr      640:     my $Client   = $Transaction->getClient();
                    641: 
1.1       foxr      642:     &Debug(8," Reply was: ".$data);
                    643:     my $Serial         = $ActiveClients{$Client};
                    644:     my $desc           = sprintf("Connection to lonc client %d",
                    645: 				 $Serial);
                    646:     Event->io(fd       => $Client,
                    647: 	      poll     => "w",
                    648: 	      desc     => $desc,
                    649: 	      cb       => \&ClientWritable,
                    650: 	      data     => $data);
                    651: }
1.42      foxr      652: 
1.4       foxr      653: =pod
1.42      foxr      654: 
1.4       foxr      655: =head2 FailTransaction
                    656: 
                    657:   Finishes a transaction with failure because the associated lond socket
1.7       foxr      658:   disconnected.  There are two possibilities:
                    659:   - The transaction is deferred: in which case we just quietly
                    660:     delete the transaction since there is no client connection.
                    661:   - The transaction is 'live' in which case we initiate the sending
                    662:     of "con_lost" to the client.
                    663: 
1.42      foxr      664: Deleting the transaction means killing it from the %ActiveTransactions hash.
1.4       foxr      665: 
                    666: Parameters:
                    667: 
                    668: =item client  
                    669:  
1.7       foxr      670:    The LondTransaction we are failing.
                    671:  
1.42      foxr      672: 
1.4       foxr      673: =cut
                    674: 
                    675: sub FailTransaction {
1.7       foxr      676:     my $transaction = shift;
1.52      foxr      677:     
                    678:     #  If the socket is dead, that's already logged.
                    679: 
                    680:     if ($ConnectionRetriesLeft > 0) {
                    681: 	Log("WARNING", "Failing transaction "
1.71      albertel  682: 	    .$transaction->getLoggableRequest());
1.52      foxr      683:     }
1.71      albertel  684:     Debug(1, "Failing transaction: ".$transaction->getLoggableRequest());
1.10      foxr      685:     if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
1.11      foxr      686: 	my $client  = $transaction->getClient();
1.30      foxr      687: 	Debug(1," Replying con_lost to ".$transaction->getRequest());
1.11      foxr      688: 	StartClientReply($transaction, "con_lost\n");
1.7       foxr      689:     }
1.4       foxr      690: 
                    691: }
                    692: 
                    693: =pod
1.69      matthew   694: 
1.6       foxr      695: =head1  EmptyQueue
1.7       foxr      696: 
1.6       foxr      697:   Fails all items in the work queue with con_lost.
1.7       foxr      698:   Note that each item in the work queue is a transaction.
                    699: 
1.6       foxr      700: =cut
1.69      matthew   701: 
1.6       foxr      702: sub EmptyQueue {
1.22      foxr      703:     $ConnectionRetriesLeft--;	# Counts as connection failure too.
1.6       foxr      704:     while($WorkQueue->Count()) {
1.10      foxr      705: 	my $request = $WorkQueue->dequeue(); # This is a transaction
1.7       foxr      706: 	FailTransaction($request);
1.6       foxr      707:     }
                    708: }
                    709: 
                    710: =pod
1.4       foxr      711: 
1.9       foxr      712: =head2 CloseAllLondConnections
                    713: 
                    714: Close all connections open on lond prior to exit e.g.
                    715: 
                    716: =cut
1.69      matthew   717: 
1.9       foxr      718: sub CloseAllLondConnections {
1.23      foxr      719:     foreach my $Socket (keys %ActiveConnections) {
1.42      foxr      720:       if(exists($ActiveTransactions{$Socket})) {
                    721: 	FailTransaction($ActiveTransactions{$Socket});
                    722:       }
                    723:       KillSocket($Socket);
1.9       foxr      724:     }
                    725: }
                    726: 
                    727: =pod
                    728: 
1.4       foxr      729: =head2 KillSocket
                    730:  
                    731: Destroys a socket.  This function can be called either when a socket
                    732: has died of 'natural' causes or because a socket needs to be pruned due to
                    733: idleness.  If the socket has died naturally, if there are no longer any 
                    734: live connections a new connection is created (in case there are transactions
                    735: in the queue).  If the socket has been pruned, it is never re-created.
                    736: 
                    737: Parameters:
1.1       foxr      738: 
1.4       foxr      739: =item Socket
                    740:  
                    741:   The socket to kill off.
                    742: 
                    743: =item Restart
                    744: 
                    745: nonzero if we are allowed to create a new connection.
                    746: 
1.69      matthew   747: =cut
1.4       foxr      748: 
                    749: sub KillSocket {
                    750:     my $Socket = shift;
                    751: 
1.17      foxr      752:     Log("WARNING", "Shutting down a socket");
1.9       foxr      753:     $Socket->Shutdown();
                    754: 
1.7       foxr      755:     #  If the socket came from the active connection set,
                    756:     #  delete its transaction... note that FailTransaction should
                    757:     #  already have been called!!!
                    758:     #  otherwise it came from the idle set.
                    759:     #  
1.4       foxr      760:     
                    761:     if(exists($ActiveTransactions{$Socket})) {
                    762: 	delete ($ActiveTransactions{$Socket});
                    763:     }
                    764:     if(exists($ActiveConnections{$Socket})) {
1.90      foxr      765: 	$ActiveConnections{$Socket}->cancel;
1.4       foxr      766: 	delete($ActiveConnections{$Socket});
1.37      albertel  767: 	$ConnectionCount--;
                    768: 	if ($ConnectionCount < 0) { $ConnectionCount = 0; }
1.4       foxr      769:     }
1.6       foxr      770:     #  If the connection count has gone to zero and there is work in the
                    771:     #  work queue, the work all gets failed with con_lost.
                    772:     #
                    773:     if($ConnectionCount == 0) {
1.22      foxr      774: 	EmptyQueue();
1.42      foxr      775: 	CloseAllLondConnections; # Should all already be closed but...
1.4       foxr      776:     }
1.90      foxr      777:     UpdateStatus();
1.4       foxr      778: }
1.1       foxr      779: 
                    780: =pod
1.3       albertel  781: 
1.1       foxr      782: =head2 LondReadable
1.3       albertel  783: 
1.1       foxr      784: This function is called whenever a lond connection
                    785: is readable.  The action is state dependent:
                    786: 
1.3       albertel  787: =head3 State=Initialized
                    788: 
                    789: We''re waiting for the challenge, this is a no-op until the
1.1       foxr      790: state changes.
1.3       albertel  791: 
1.1       foxr      792: =head3 State=Challenged 
1.3       albertel  793: 
                    794: The challenge has arrived we need to transition to Writable.
1.1       foxr      795: The connection must echo the challenge back.
1.3       albertel  796: 
1.1       foxr      797: =head3 State=ChallengeReplied
1.3       albertel  798: 
                    799: The challenge has been replied to.  The we are receiveing the 
1.1       foxr      800: 'ok' from the partner.
1.3       albertel  801: 
1.40      foxr      802: =head3  State=ReadingVersionString
                    803: 
                    804: We have requested the lond version and are reading the
                    805: version back.  Upon completion, we'll store the version away
                    806: for future use(?).
                    807: 
                    808: =head3 State=HostSet
                    809: 
                    810: We have selected the domain name of our peer (multhomed hosts)
                    811: and are getting the reply (presumably ok) back.
                    812: 
1.1       foxr      813: =head3 State=RequestingKey
1.3       albertel  814: 
                    815: The ok has been received and we need to send the request for
1.1       foxr      816: an encryption key.  Transition to writable for that.
1.3       albertel  817: 
1.1       foxr      818: =head3 State=ReceivingKey
1.3       albertel  819: 
                    820: The the key has been requested, now we are reading the new key.
                    821: 
1.1       foxr      822: =head3 State=Idle 
1.3       albertel  823: 
                    824: The encryption key has been negotiated or we have finished 
1.1       foxr      825: reading data from the a transaction.   If the callback data has
                    826: a client as well as the socket iformation, then we are 
                    827: doing a transaction and the data received is relayed to the client
                    828: before the socket is put on the idle list.
1.3       albertel  829: 
1.1       foxr      830: =head3 State=SendingRequest
1.3       albertel  831: 
                    832: I do not think this state can be received here, but if it is,
1.1       foxr      833: the appropriate thing to do is to transition to writable, and send
                    834: the request.
1.3       albertel  835: 
1.1       foxr      836: =head3 State=ReceivingReply
1.3       albertel  837: 
                    838: We finished sending the request to the server and now transition
1.1       foxr      839: to readable to receive the reply. 
                    840: 
                    841: The parameter to this function are:
1.3       albertel  842: 
1.1       foxr      843: The event. Implicit in this is the watcher and its data.  The data 
                    844: contains at least the lond connection object and, if a 
                    845: transaction is in progress, the socket attached to the local client.
                    846: 
1.3       albertel  847: =cut
1.1       foxr      848: 
                    849: sub LondReadable {
1.8       foxr      850: 
1.41      albertel  851:     my $Event      = shift;
                    852:     my $Watcher    = $Event->w;
                    853:     my $Socket     = $Watcher->data;
                    854:     my $client     = undef;
1.40      foxr      855: 
1.41      albertel  856:     &Debug(6,"LondReadable called state = ".$Socket->GetState());
1.40      foxr      857: 
                    858: 
1.41      albertel  859:     my $State = $Socket->GetState(); # All action depends on the state.
1.40      foxr      860: 
1.41      albertel  861:     SocketDump(6, $Socket);
                    862:     my $status = $Socket->Readable();
1.40      foxr      863: 
1.41      albertel  864:     &Debug(2, "Socket->Readable returned: $status");
1.40      foxr      865: 
1.41      albertel  866:     if($status != 0) {
                    867: 	# bad return from socket read. Currently this means that
                    868: 	# The socket has become disconnected. We fail the transaction.
1.40      foxr      869: 
1.41      albertel  870: 	Log("WARNING",
                    871: 	    "Lond connection lost.");
                    872: 	if(exists($ActiveTransactions{$Socket})) {
                    873: 	    FailTransaction($ActiveTransactions{$Socket});
1.56      foxr      874: 	} else {
                    875: 	    #  Socket is connecting and failed... need to mark
                    876: 	    #  no longer connecting.
                    877: 	   
                    878: 	    $LondConnecting = 0;
1.41      albertel  879: 	}
                    880: 	$Watcher->cancel();
                    881: 	KillSocket($Socket);
                    882: 	$ConnectionRetriesLeft--;       # Counts as connection failure
                    883: 	return;
                    884:     }
                    885:     SocketDump(6,$Socket);
1.17      foxr      886: 
1.41      albertel  887:     $State = $Socket->GetState(); # Update in case of transition.
                    888:     &Debug(6, "After read, state is ".$State);
1.1       foxr      889: 
1.41      albertel  890:     if($State eq "Initialized") {
1.1       foxr      891: 
                    892: 
1.41      albertel  893:     } elsif ($State eq "ChallengeReceived") {
1.1       foxr      894: 	#  The challenge must be echoed back;  The state machine
                    895: 	# in the connection takes care of setting that up.  Just
                    896: 	# need to transition to writable:
1.41      albertel  897: 	
                    898: 	$Watcher->cb(\&LondWritable);
                    899: 	$Watcher->poll("w");
1.1       foxr      900: 
1.41      albertel  901:     } elsif ($State eq "ChallengeReplied") {
1.1       foxr      902: 
1.41      albertel  903:     } elsif ($State eq "RequestingVersion") {
                    904: 	# Need to ask for the version... that is writiability:
1.1       foxr      905: 
1.41      albertel  906: 	$Watcher->cb(\&LondWritable);
                    907: 	$Watcher->poll("w");
                    908: 
                    909:     } elsif ($State eq "ReadingVersionString") {
                    910: 	# Read the rest of the version string... 
                    911:     } elsif ($State eq "SetHost") {
                    912: 	# Need to request the actual domain get set...
                    913: 
                    914: 	$Watcher->cb(\&LondWritable);
                    915: 	$Watcher->poll("w");
                    916:     } elsif ($State eq "HostSet") {
                    917: 	# Reading the 'ok' from the peer.
                    918: 
                    919:     } elsif ($State eq "RequestingKey") {
1.1       foxr      920: 	#  The ok was received.  Now we need to request the key
                    921: 	#  That requires us to be writable:
                    922: 
1.41      albertel  923: 	$Watcher->cb(\&LondWritable);
                    924: 	$Watcher->poll("w");
1.1       foxr      925: 
1.41      albertel  926:     } elsif ($State eq "ReceivingKey") {
1.1       foxr      927: 
1.41      albertel  928:     } elsif ($State eq "Idle") {
1.40      foxr      929:    
1.41      albertel  930: 	# This is as good a spot as any to get the peer version
                    931: 	# string:
1.40      foxr      932:    
1.41      albertel  933: 	if($LondVersion eq "unknown") {
                    934: 	    $LondVersion = $Socket->PeerVersion();
                    935: 	    Log("INFO", "Connected to lond version: $LondVersion");
                    936: 	}
1.1       foxr      937: 	# If necessary, complete a transaction and then go into the
                    938: 	# idle queue.
1.22      foxr      939: 	#  Note that a trasition to idle indicates a live lond
                    940: 	# on the other end so reset the connection retries.
                    941: 	#
1.41      albertel  942: 	$ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
                    943: 	$Watcher->cancel();
                    944: 	if(exists($ActiveTransactions{$Socket})) {
                    945: 	    Debug(5,"Completing transaction!!");
                    946: 	    CompleteTransaction($Socket, 
                    947: 				$ActiveTransactions{$Socket});
                    948: 	} else {
                    949: 	    Log("SUCCESS", "Connection ".$ConnectionCount." to "
                    950: 		.$RemoteHost." now ready for action");
                    951: 	}
                    952: 	ServerToIdle($Socket);	# Next work unit or idle.
1.54      foxr      953: 
                    954: 	#
                    955: 	$LondConnecting = 0;	# Best spot I can think of for this.
                    956: 	# 
1.6       foxr      957: 	
1.41      albertel  958:     } elsif ($State eq "SendingRequest") {
1.1       foxr      959: 	#  We need to be writable for this and probably don't belong
                    960: 	#  here inthe first place.
                    961: 
1.73      albertel  962: 	Debug(6, "SendingRequest state encountered in readable");
1.41      albertel  963: 	$Watcher->poll("w");
                    964: 	$Watcher->cb(\&LondWritable);
1.1       foxr      965: 
1.41      albertel  966:     } elsif ($State eq "ReceivingReply") {
1.1       foxr      967: 
                    968: 
1.41      albertel  969:     } else {
                    970: 	# Invalid state.
                    971: 	Debug(4, "Invalid state in LondReadable");
                    972:     }
1.1       foxr      973: }
1.3       albertel  974: 
1.1       foxr      975: =pod
1.3       albertel  976: 
1.1       foxr      977: =head2 LondWritable
1.3       albertel  978: 
1.1       foxr      979: This function is called whenever a lond connection
                    980: becomes writable while there is a writeable monitoring
                    981: event.  The action taken is very state dependent:
1.3       albertel  982: 
1.1       foxr      983: =head3 State = Connected 
1.3       albertel  984: 
                    985: The connection is in the process of sending the 'init' hailing to the
                    986: lond on the remote end.  The connection object''s Writable member is
                    987: called.  On error, ConnectionError is called to destroy the connection
                    988: and remove it from the ActiveConnections hash
                    989: 
1.1       foxr      990: =head3 Initialized
1.3       albertel  991: 
                    992: 'init' has been sent, writability monitoring is removed and
                    993: readability monitoring is started with LondReadable as the callback.
                    994: 
1.1       foxr      995: =head3 ChallengeReceived
1.3       albertel  996: 
                    997: The connection has received the who are you challenge from the remote
                    998: system, and is in the process of sending the challenge
                    999: response. Writable is called.
                   1000: 
1.1       foxr     1001: =head3 ChallengeReplied
1.3       albertel 1002: 
                   1003: The connection has replied to the initial challenge The we switch to
                   1004: monitoring readability looking for the server to reply with 'ok'.
                   1005: 
1.1       foxr     1006: =head3 RequestingKey
1.3       albertel 1007: 
                   1008: The connection is in the process of requesting its encryption key.
                   1009: Writable is called.
                   1010: 
1.1       foxr     1011: =head3 ReceivingKey
1.3       albertel 1012: 
                   1013: The connection has sent the request for a key.  Switch to readability
                   1014: monitoring to accept the key
                   1015: 
1.1       foxr     1016: =head3 SendingRequest
1.3       albertel 1017: 
                   1018: The connection is in the process of sending a request to the server.
                   1019: This request is part of a client transaction.  All the states until
                   1020: now represent the client setup protocol. Writable is called.
                   1021: 
1.1       foxr     1022: =head3 ReceivingReply
                   1023: 
1.3       albertel 1024: The connection has sent a request.  Now it must receive a reply.
                   1025: Readability monitoring is requested.
                   1026: 
                   1027: This function is an event handler and therefore receives as
1.1       foxr     1028: a parameter the event that has fired.  The data for the watcher
                   1029: of this event is a reference to a list of one or two elements,
                   1030: depending on state. The first (and possibly only) element is the
                   1031: socket.  The second (present only if a request is in progress)
                   1032: is the socket on which to return a reply to the caller.
                   1033: 
                   1034: =cut
1.3       albertel 1035: 
1.1       foxr     1036: sub LondWritable {
                   1037:     my $Event   = shift;
                   1038:     my $Watcher = $Event->w;
1.8       foxr     1039:     my $Socket  = $Watcher->data;
                   1040:     my $State   = $Socket->GetState();
1.1       foxr     1041: 
1.8       foxr     1042:     Debug(6,"LondWritable State = ".$State."\n");
1.1       foxr     1043: 
1.8       foxr     1044:  
1.1       foxr     1045:     #  Figure out what to do depending on the state of the socket:
                   1046:     
                   1047: 
                   1048: 
                   1049: 
                   1050:     SocketDump(6,$Socket);
                   1051: 
1.42      foxr     1052:     #  If the socket is writable, we must always write.
                   1053:     # Only by writing will we undergo state transitions.
                   1054:     # Old logic wrote in state specific code below, however
                   1055:     # That forces us at least through another invocation of
                   1056:     # this function after writability is possible again.
                   1057:     # This logic also factors out common code for handling
                   1058:     # write failures... in all cases, write failures 
                   1059:     # Kill the socket.
                   1060:     #  This logic makes the branches of the >big< if below
                   1061:     # so that the writing states are actually NO-OPs.
                   1062: 
                   1063:     if ($Socket->Writable() != 0) {
1.43      albertel 1064: 	#  The write resulted in an error.
                   1065: 	# We'll treat this as if the socket got disconnected:
                   1066: 	Log("WARNING", "Connection to ".$RemoteHost.
                   1067: 	    " has been disconnected");
                   1068: 	if(exists($ActiveTransactions{$Socket})) {
                   1069: 	    FailTransaction($ActiveTransactions{$Socket});
1.56      foxr     1070: 	} else {
                   1071: 	    #  In the process of conneting, so need to turn that off.
                   1072: 	    
                   1073: 	    $LondConnecting = 0;
1.43      albertel 1074: 	}
                   1075: 	$Watcher->cancel();
                   1076: 	KillSocket($Socket);
                   1077: 	return;
1.42      foxr     1078:     }
                   1079: 
                   1080: 
                   1081: 
1.41      albertel 1082:     if      ($State eq "Connected")         {
1.1       foxr     1083: 
1.41      albertel 1084: 	#  "init" is being sent...
1.42      foxr     1085:  
1.41      albertel 1086:     } elsif ($State eq "Initialized")       {
1.4       foxr     1087: 
1.41      albertel 1088: 	# Now that init was sent, we switch 
                   1089: 	# to watching for readability:
1.1       foxr     1090: 
1.41      albertel 1091: 	$Watcher->cb(\&LondReadable);
                   1092: 	$Watcher->poll("r");
                   1093: 	
                   1094:     } elsif ($State eq "ChallengeReceived") {
                   1095: 	# We received the challenge, now we 
                   1096: 	# are echoing it back. This is a no-op,
                   1097: 	# we're waiting for the state to change
1.1       foxr     1098: 	
1.41      albertel 1099:     } elsif ($State eq "ChallengeReplied")  {
                   1100: 	# The echo was sent back, so we switch
                   1101: 	# to watching readability.
                   1102: 
                   1103: 	$Watcher->cb(\&LondReadable);
                   1104: 	$Watcher->poll("r");
                   1105:     } elsif ($State eq "RequestingVersion") {
                   1106: 	# Sending the peer a version request...
1.42      foxr     1107: 
1.41      albertel 1108:     } elsif ($State eq "ReadingVersionString") {
                   1109: 	# Transition to read since we have sent the
                   1110: 	# version command and now just need to read the
                   1111: 	# version string from the peer:
1.40      foxr     1112:       
1.41      albertel 1113: 	$Watcher->cb(\&LondReadable);
                   1114: 	$Watcher->poll("r");
1.40      foxr     1115:       
1.41      albertel 1116:     } elsif ($State eq "SetHost") {
                   1117: 	#  Setting the remote domain...
1.42      foxr     1118: 
1.41      albertel 1119:     } elsif ($State eq "HostSet") {
                   1120: 	# Back to readable to get the ok.
1.40      foxr     1121:       
1.41      albertel 1122: 	$Watcher->cb(\&LondReadable);
                   1123: 	$Watcher->poll("r");
1.40      foxr     1124:       
                   1125: 
1.41      albertel 1126:     } elsif ($State eq "RequestingKey")     {
                   1127: 	# At this time we're requesting the key.
                   1128: 	# again, this is essentially a no-op.
                   1129: 
                   1130:     } elsif ($State eq "ReceivingKey")      {
                   1131: 	# Now we need to wait for the key
                   1132: 	# to come back from the peer:
                   1133: 
                   1134: 	$Watcher->cb(\&LondReadable);
                   1135: 	$Watcher->poll("r");
                   1136: 
                   1137:     } elsif ($State eq "SendingRequest")    {
1.40      foxr     1138:  
1.41      albertel 1139: 	# At this time we are sending a request to the
1.1       foxr     1140: 	# peer... write the next chunk:
                   1141: 
1.41      albertel 1142: 
                   1143:     } elsif ($State eq "ReceivingReply")    {
                   1144: 	# The send has completed.  Wait for the
                   1145: 	# data to come in for a reply.
                   1146: 	Debug(8,"Writable sent request/receiving reply");
                   1147: 	$Watcher->cb(\&LondReadable);
                   1148: 	$Watcher->poll("r");
1.1       foxr     1149: 
1.41      albertel 1150:     } else {
                   1151: 	#  Control only passes here on an error: 
                   1152: 	#  the socket state does not match any
                   1153: 	#  of the known states... so an error
                   1154: 	#  must be logged.
1.1       foxr     1155: 
1.41      albertel 1156: 	&Debug(4, "Invalid socket state ".$State."\n");
                   1157:     }
1.1       foxr     1158:     
                   1159: }
1.81      albertel 1160: 
1.6       foxr     1161: =pod
                   1162:     
                   1163: =cut
1.69      matthew  1164: 
1.81      albertel 1165: 
1.6       foxr     1166: sub QueueDelayed {
1.8       foxr     1167:     Debug(3,"QueueDelayed called");
                   1168: 
1.6       foxr     1169:     my $path = "$perlvar{'lonSockDir'}/delayed";
1.8       foxr     1170: 
                   1171:     Debug(4, "Delayed path: ".$path);
1.6       foxr     1172:     opendir(DIRHANDLE, $path);
1.75      albertel 1173: 
1.82      albertel 1174:     my $host_id_re = '(?:'.join('|',map {quotemeta($_)} (@all_host_ids)).')';
1.75      albertel 1175:     my @alldelayed = grep(/\.$host_id_re$/, readdir(DIRHANDLE));
1.6       foxr     1176:     closedir(DIRHANDLE);
1.75      albertel 1177:     foreach my $dfname (sort(@alldelayed)) {
                   1178: 	my $reqfile = "$path/$dfname";
                   1179: 	my ($host_id) = ($dfname =~ /\.([^.]*)$/);
                   1180: 	Debug(4, "queueing ".$reqfile." for $host_id");
1.6       foxr     1181: 	my $Handle = IO::File->new($reqfile);
                   1182: 	my $cmd    = <$Handle>;
1.8       foxr     1183: 	chomp $cmd;		# There may or may not be a newline...
1.12      foxr     1184: 	$cmd = $cmd."\n";	# now for sure there's exactly one newline.
1.75      albertel 1185: 	my $Transaction = LondTransaction->new("sethost:$host_id:$cmd");
1.7       foxr     1186: 	$Transaction->SetDeferred($reqfile);
                   1187: 	QueueTransaction($Transaction);
1.6       foxr     1188:     }
                   1189:     
                   1190: }
1.1       foxr     1191: 
                   1192: =pod
1.3       albertel 1193: 
1.1       foxr     1194: =head2 MakeLondConnection
1.3       albertel 1195: 
                   1196: Create a new lond connection object, and start it towards its initial
                   1197: idleness.  Once idle, it becomes elligible to receive transactions
                   1198: from the work queue.  If the work queue is not empty when the
                   1199: connection is completed and becomes idle, it will dequeue an entry and
                   1200: start off on it.
                   1201: 
1.1       foxr     1202: =cut
1.3       albertel 1203: 
1.1       foxr     1204: sub MakeLondConnection {     
                   1205:     Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
                   1206: 	  .GetServerPort());
                   1207: 
                   1208:     my $Connection = LondConnection->new(&GetServerHost(),
1.81      albertel 1209: 					 &GetServerPort(),
                   1210: 					 &GetHostId());
1.1       foxr     1211: 
1.90      foxr     1212:     if($Connection eq undef) {	
1.9       foxr     1213: 	Log("CRITICAL","Failed to make a connection with lond.");
1.10      foxr     1214: 	$ConnectionRetriesLeft--;
                   1215: 	return 0;		# Failure.
1.5       foxr     1216:     }  else {
1.22      foxr     1217: 
1.82      albertel 1218: 	$LondConnecting = 1;	# Connection in progress.
1.5       foxr     1219: 	# The connection needs to have writability 
                   1220: 	# monitored in order to send the init sequence
                   1221: 	# that starts the whole authentication/key
                   1222: 	# exchange underway.
                   1223: 	#
                   1224: 	my $Socket = $Connection->GetSocket();
1.30      foxr     1225: 	if($Socket eq undef) {
1.64      foxr     1226: 	    &child_exit(-1, "did not get a socket from the connection");
1.5       foxr     1227: 	} else {
                   1228: 	    &Debug(9,"MakeLondConnection got socket: ".$Socket);
                   1229: 	}
1.1       foxr     1230: 	
1.21      foxr     1231: 	$Connection->SetTimeoutCallback(\&SocketTimeout);
                   1232: 
1.23      foxr     1233: 	my $event = Event->io(fd       => $Socket,
1.5       foxr     1234: 			   poll     => 'w',
                   1235: 			   cb       => \&LondWritable,
1.8       foxr     1236: 			   data     => $Connection,
1.5       foxr     1237: 			   desc => 'Connection to lond server');
                   1238: 	$ActiveConnections{$Connection} = $event;
1.52      foxr     1239: 	if ($ConnectionCount == 0) {
                   1240: 	    &SetupTimer;	# Need to handle timeouts with connections...
                   1241: 	}
1.5       foxr     1242: 	$ConnectionCount++;
1.8       foxr     1243: 	Debug(4, "Connection count = ".$ConnectionCount);
1.6       foxr     1244: 	if($ConnectionCount == 1) { # First Connection:
                   1245: 	    QueueDelayed;
                   1246: 	}
1.9       foxr     1247: 	Log("SUCESS", "Created connection ".$ConnectionCount
                   1248: 	    ." to host ".GetServerHost());
1.10      foxr     1249: 	return 1;		# Return success.
1.1       foxr     1250:     }
                   1251:     
                   1252: }
1.3       albertel 1253: 
1.1       foxr     1254: =pod
1.3       albertel 1255: 
1.1       foxr     1256: =head2 StartRequest
1.3       albertel 1257: 
                   1258: Starts a lond request going on a specified lond connection.
                   1259: parameters are:
                   1260: 
                   1261: =item $Lond
                   1262: 
                   1263: Connection to the lond that will send the transaction and receive the
                   1264: reply.
                   1265: 
                   1266: =item $Client
                   1267: 
                   1268: Connection to the client that is making this request We got the
                   1269: request from this socket, and when the request has been relayed to
                   1270: lond and we get a reply back from lond it will get sent to this
                   1271: socket.
                   1272: 
                   1273: =item $Request
                   1274: 
                   1275: The text of the request to send.
                   1276: 
1.1       foxr     1277: =cut
                   1278: 
                   1279: sub StartRequest {
1.47      foxr     1280: 
                   1281:     my ($Lond, $Request) = @_;
1.1       foxr     1282:     
1.7       foxr     1283:     Debug(6, "StartRequest: ".$Request->getRequest());
1.1       foxr     1284: 
                   1285:     my $Socket = $Lond->GetSocket();
                   1286:     
1.7       foxr     1287:     $Request->Activate($Lond);
                   1288:     $ActiveTransactions{$Lond} = $Request;
1.1       foxr     1289: 
1.7       foxr     1290:     $Lond->InitiateTransaction($Request->getRequest());
1.23      foxr     1291:     my $event = Event->io(fd      => $Socket,
1.1       foxr     1292: 		       poll    => "w",
                   1293: 		       cb      => \&LondWritable,
                   1294: 		       data    => $Lond,
                   1295: 		       desc    => "lond transaction connection");
                   1296:     $ActiveConnections{$Lond} = $event;
                   1297:     Debug(8," Start Request made watcher data with ".$event->data."\n");
                   1298: }
                   1299: 
                   1300: =pod
1.3       albertel 1301: 
1.1       foxr     1302: =head2 QueueTransaction
1.3       albertel 1303: 
                   1304: If there is an idle lond connection, it is put to work doing this
                   1305: transaction.  Otherwise, the transaction is placed in the work queue.
                   1306: If placed in the work queue and the maximum number of connections has
                   1307: not yet been created, a new connection will be started.  Our goal is
                   1308: to eventually have a sufficient number of connections that the work
                   1309: queue will typically be empty.  parameters are:
                   1310: 
                   1311: =item Socket
                   1312: 
                   1313: open on the lonc client.
                   1314: 
                   1315: =item Request
                   1316: 
                   1317: data to send to the lond.
1.1       foxr     1318: 
                   1319: =cut
1.3       albertel 1320: 
1.1       foxr     1321: sub QueueTransaction {
                   1322: 
1.7       foxr     1323:     my $requestData   = shift;	# This is a LondTransaction.
                   1324:     my $cmd           = $requestData->getRequest();
                   1325: 
                   1326:     Debug(6,"QueueTransaction: ".$cmd);
1.1       foxr     1327: 
                   1328:     my $LondSocket    = $IdleConnections->pop();
                   1329:     if(!defined $LondSocket) {	# Need to queue request.
1.29      foxr     1330: 	Debug(5,"Must queue...");
1.1       foxr     1331: 	$WorkQueue->enqueue($requestData);
1.56      foxr     1332: 	Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
                   1333: 	if(($ConnectionCount < $MaxConnectionCount)   && (! $LondConnecting)) {
                   1334: 
1.22      foxr     1335: 	    if($ConnectionRetriesLeft > 0) {
1.29      foxr     1336: 		Debug(5,"Starting additional lond connection");
1.56      foxr     1337: 		if(&MakeLondConnection() == 0) {
1.22      foxr     1338: 		    EmptyQueue();	# Fail transactions, can't make connection.
1.42      foxr     1339: 		    CloseAllLondConnections; # Should all be closed but...
1.22      foxr     1340: 		}
                   1341: 	    } else {
                   1342: 		ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
1.56      foxr     1343: 		$LondConnecting = 0;
1.22      foxr     1344: 		EmptyQueue();	# It's worse than that ... he's dead Jim.
1.42      foxr     1345: 		CloseAllLondConnections; # Should all be closed but..
1.17      foxr     1346: 	    }
1.1       foxr     1347: 	}
                   1348:     } else {			# Can start the request:
                   1349: 	Debug(8,"Can start...");
1.7       foxr     1350: 	StartRequest($LondSocket,  $requestData);
1.1       foxr     1351:     }
                   1352: }
                   1353: 
                   1354: #-------------------------- Lonc UNIX socket handling ---------------------
1.3       albertel 1355: 
1.1       foxr     1356: =pod
1.3       albertel 1357: 
1.1       foxr     1358: =head2 ClientRequest
1.3       albertel 1359: Callback that is called when data can be read from the UNIX domain
                   1360: socket connecting us with an apache server process.
1.1       foxr     1361: 
                   1362: =cut
                   1363: 
                   1364: sub ClientRequest {
                   1365:     Debug(6, "ClientRequest");
                   1366:     my $event   = shift;
                   1367:     my $watcher = $event->w;
                   1368:     my $socket  = $watcher->fd;
                   1369:     my $data    = $watcher->data;
                   1370:     my $thisread;
                   1371: 
                   1372:     Debug(9, "  Watcher named: ".$watcher->desc);
                   1373: 
                   1374:     my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
                   1375:     Debug(8, "rcv:  data length = ".length($thisread)
                   1376: 	  ." read =".$thisread);
1.29      foxr     1377:     unless (defined $rv  && length($thisread)) {
1.1       foxr     1378: 	 # Likely eof on socket.
                   1379: 	Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
                   1380: 	close($socket);
                   1381: 	$watcher->cancel();
                   1382: 	delete($ActiveClients{$socket});
1.10      foxr     1383: 	return;
1.1       foxr     1384:     }
                   1385:     Debug(8,"Data: ".$data." this read: ".$thisread);
                   1386:     $data = $data.$thisread;	# Append new data.
                   1387:     $watcher->data($data);
1.44      albertel 1388:     if($data =~ /\n$/) {	# Request entirely read.
1.87      albertel 1389: 	if ($data eq "close_connection_exit\n") {
1.9       foxr     1390: 	    Log("CRITICAL",
                   1391: 		"Request Close Connection ... exiting");
                   1392: 	    CloseAllLondConnections();
                   1393: 	    exit;
1.87      albertel 1394: 	} elsif ($data eq "reset_retries\n") {
                   1395: 	    Log("INFO", "Resetting Connection Retries.");
                   1396: 	    $ConnectionRetriesLeft = $ConnectionRetries;
                   1397: 	    &UpdateStatus();
                   1398: 	    my $Transaction = LondTransaction->new($data);
                   1399: 	    $Transaction->SetClient($socket);
                   1400: 	    StartClientReply($Transaction, "ok\n");
                   1401: 	    $watcher->cancel();
                   1402: 	    return;
1.9       foxr     1403: 	}
1.1       foxr     1404: 	Debug(8, "Complete transaction received: ".$data);
1.87      albertel 1405: 	if ($LogTransactions) {
1.39      foxr     1406: 	    Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
                   1407: 	}
1.8       foxr     1408: 	my $Transaction = LondTransaction->new($data);
1.7       foxr     1409: 	$Transaction->SetClient($socket);
                   1410: 	QueueTransaction($Transaction);
1.1       foxr     1411: 	$watcher->cancel();	# Done looking for input data.
                   1412:     }
                   1413: 
                   1414: }
                   1415: 
1.62      foxr     1416: #
                   1417: #     Accept a connection request for a client (lonc child) and
                   1418: #    start up an event watcher to keep an eye on input from that 
                   1419: #    Event.  This can be called both from NewClient and from
1.80      albertel 1420: #    ChildProcess.
1.62      foxr     1421: # Parameters:
                   1422: #    $socket       - The listener socket.
                   1423: # Returns:
                   1424: #   NONE
                   1425: # Side Effects:
                   1426: #    An event is made to watch the accepted connection.
                   1427: #    Active clients hash is updated to reflect the new connection.
                   1428: #    The client connection count is incremented.
                   1429: #
                   1430: sub accept_client {
                   1431:     my ($socket) = @_;
                   1432: 
                   1433:     Debug(8, "Entering accept for lonc UNIX socket\n");
                   1434:     my $connection = $socket->accept();	# Accept the client connection.
                   1435:     Debug(8,"Connection request accepted from "
                   1436: 	  .GetPeername($connection, AF_UNIX));
                   1437: 
                   1438: 
                   1439:     my $description = sprintf("Connection to lonc client %d",
                   1440: 			      $ClientConnection);
                   1441:     Debug(9, "Creating event named: ".$description);
                   1442:     Event->io(cb      => \&ClientRequest,
                   1443: 	      poll    => 'r',
                   1444: 	      desc    => $description,
                   1445: 	      data    => "",
                   1446: 	      fd      => $connection);
                   1447:     $ActiveClients{$connection} = $ClientConnection;
                   1448:     $ClientConnection++;
                   1449: }
1.1       foxr     1450: 
                   1451: =pod
1.3       albertel 1452: 
1.1       foxr     1453: =head2  NewClient
1.3       albertel 1454: 
                   1455: Callback that is called when a connection is received on the unix
                   1456: socket for a new client of lonc.  The callback is parameterized by the
                   1457: event.. which is a-priori assumed to be an io event, and therefore has
                   1458: an fd member that is the Listener socket.  We Accept the connection
                   1459: and register a new event on the readability of that socket:
                   1460: 
1.1       foxr     1461: =cut
1.3       albertel 1462: 
1.1       foxr     1463: sub NewClient {
                   1464:     Debug(6, "NewClient");
                   1465:     my $event      = shift;		# Get the event parameters.
                   1466:     my $watcher    = $event->w; 
                   1467:     my $socket     = $watcher->fd;	# Get the event' socket.
                   1468: 
1.62      foxr     1469:     &accept_client($socket);
1.1       foxr     1470: }
1.3       albertel 1471: 
                   1472: =pod
                   1473: 
                   1474: =head2 GetLoncSocketPath
                   1475: 
                   1476: Returns the name of the UNIX socket on which to listen for client
                   1477: connections.
1.1       foxr     1478: 
1.58      foxr     1479: =head2 Parameters:
                   1480: 
                   1481:     host (optional)  - Name of the host socket to return.. defaults to
                   1482:                        the return from GetServerHost().
                   1483: 
1.1       foxr     1484: =cut
1.3       albertel 1485: 
1.1       foxr     1486: sub GetLoncSocketPath {
1.58      foxr     1487: 
                   1488:     my $host = GetServerHost();	# Default host.
                   1489:     if (@_) {
                   1490: 	($host)  = @_;		# Override if supplied.
                   1491:     }
                   1492:     return $UnixSocketDir."/".$host;
1.1       foxr     1493: }
                   1494: 
1.3       albertel 1495: =pod
                   1496: 
                   1497: =head2 GetServerHost
                   1498: 
                   1499: Returns the host whose lond we talk with.
                   1500: 
1.1       foxr     1501: =cut
1.3       albertel 1502: 
1.7       foxr     1503: sub GetServerHost {
1.1       foxr     1504:     return $RemoteHost;		# Setup by the fork.
                   1505: }
1.3       albertel 1506: 
                   1507: =pod
                   1508: 
1.81      albertel 1509: =head2 GetServerId
                   1510: 
                   1511: Returns the hostid whose lond we talk with.
                   1512: 
                   1513: =cut
                   1514: 
                   1515: sub GetHostId {
                   1516:     return $RemoteHostId;		# Setup by the fork.
                   1517: }
                   1518: 
                   1519: =pod
                   1520: 
1.3       albertel 1521: =head2 GetServerPort
                   1522: 
                   1523: Returns the lond port number.
                   1524: 
1.1       foxr     1525: =cut
1.3       albertel 1526: 
1.7       foxr     1527: sub GetServerPort {
1.1       foxr     1528:     return $perlvar{londPort};
                   1529: }
1.3       albertel 1530: 
                   1531: =pod
                   1532: 
                   1533: =head2 SetupLoncListener
                   1534: 
                   1535: Setup a lonc listener event.  The event is called when the socket
                   1536: becomes readable.. that corresponds to the receipt of a new
                   1537: connection.  The event handler established will accept the connection
                   1538: (creating a communcations channel), that int turn will establish
                   1539: another event handler to subess requests.
1.1       foxr     1540: 
1.58      foxr     1541: =head2  Parameters:
                   1542: 
                   1543:    host (optional)   Name of the host to set up a unix socket to.
                   1544: 
1.1       foxr     1545: =cut
1.3       albertel 1546: 
1.1       foxr     1547: sub SetupLoncListener {
1.78      albertel 1548:     my ($host,$SocketName) = @_;
                   1549:     if (!$host) { $host = &GetServerHost(); }
                   1550:     if (!$SocketName) { $SocketName = &GetLoncSocketPath($host); }
1.1       foxr     1551: 
1.78      albertel 1552: 
                   1553:     unlink($SocketName);
1.58      foxr     1554: 
1.1       foxr     1555:     my $socket;
1.7       foxr     1556:     unless ($socket =IO::Socket::UNIX->new(Local  => $SocketName,
1.55      albertel 1557: 					    Listen => 250, 
1.1       foxr     1558: 					    Type   => SOCK_STREAM)) {
1.64      foxr     1559: 	if($I_am_child) {
                   1560: 	    &child_exit(-1, "Failed to create a lonc listener socket");
                   1561: 	} else {
                   1562: 	    die "Failed to create a lonc listner socket";
                   1563: 	}
1.1       foxr     1564:     }
1.59      foxr     1565:     return $socket;
1.1       foxr     1566: }
                   1567: 
1.39      foxr     1568: #
                   1569: #   Toggle transaction logging.
                   1570: #  Implicit inputs:  
                   1571: #     LogTransactions
                   1572: #  Implicit Outputs:
                   1573: #     LogTransactions
                   1574: sub ToggleTransactionLogging {
                   1575:     print STDERR "Toggle transaction logging...\n";
                   1576:     if(!$LogTransactions) {
                   1577: 	$LogTransactions = 1;
                   1578:     } else {
                   1579: 	$LogTransactions = 0;
                   1580:     }
                   1581: 
                   1582: 
                   1583:     Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
                   1584: }
                   1585: 
1.14      foxr     1586: =pod 
                   1587: 
                   1588: =head2 ChildStatus
                   1589:  
                   1590: Child USR1 signal handler to report the most recent status
                   1591: into the status file.
                   1592: 
1.22      foxr     1593: We also use this to reset the retries count in order to allow the
                   1594: client to retry connections with a previously dead server.
1.69      matthew  1595: 
1.14      foxr     1596: =cut
1.46      albertel 1597: 
1.14      foxr     1598: sub ChildStatus {
                   1599:     my $event = shift;
                   1600:     my $watcher = $event->w;
                   1601: 
                   1602:     Debug(2, "Reporting child status because : ".$watcher->data);
                   1603:     my $docdir = $perlvar{'lonDocRoot'};
1.67      albertel 1604:     
                   1605:     open(LOG,">>$docdir/lon-status/loncstatus.txt");
                   1606:     flock(LOG,LOCK_EX);
                   1607:     print LOG $$."\t".$RemoteHost."\t".$Status."\t".
1.14      foxr     1608: 	$RecentLogEntry."\n";
1.38      foxr     1609:     #
                   1610:     #  Write out information about each of the connections:
                   1611:     #
1.46      albertel 1612:     if ($DebugLevel > 2) {
1.67      albertel 1613: 	print LOG "Active connection statuses: \n";
1.46      albertel 1614: 	my $i = 1;
                   1615: 	print STDERR  "================================= Socket Status Dump:\n";
                   1616: 	foreach my $item (keys %ActiveConnections) {
                   1617: 	    my $Socket = $ActiveConnections{$item}->data;
                   1618: 	    my $state  = $Socket->GetState();
1.67      albertel 1619: 	    print LOG "Connection $i State: $state\n";
1.46      albertel 1620: 	    print STDERR "---------------------- Connection $i \n";
1.48      foxr     1621: 	    $Socket->Dump(-1);	# Ensure it gets dumped..
1.46      albertel 1622: 	    $i++;	
                   1623: 	}
1.38      foxr     1624:     }
1.67      albertel 1625:     flock(LOG,LOCK_UN);
                   1626:     close(LOG);
1.22      foxr     1627:     $ConnectionRetriesLeft = $ConnectionRetries;
1.70      albertel 1628:     UpdateStatus();
1.14      foxr     1629: }
                   1630: 
1.1       foxr     1631: =pod
1.3       albertel 1632: 
1.10      foxr     1633: =head2 SignalledToDeath
                   1634: 
                   1635: Called in response to a signal that causes a chid process to die.
                   1636: 
                   1637: =cut
                   1638: 
                   1639: 
                   1640: sub SignalledToDeath {
1.14      foxr     1641:     my $event  = shift;
                   1642:     my $watcher= $event->w;
                   1643: 
                   1644:     Debug(2,"Signalled to death! via ".$watcher->data);
1.17      foxr     1645:     my ($signal) = $watcher->data;
1.10      foxr     1646:     chomp($signal);
                   1647:     Log("CRITICAL", "Abnormal exit.  Child $$ for $RemoteHost "
                   1648: 	."died through "."\"$signal\"");
1.68      albertel 1649:     #LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
                   1650: #	    ."\"$signal\"");
1.12      foxr     1651:     exit 0;
1.10      foxr     1652: 
                   1653: }
1.16      foxr     1654: 
1.69      matthew  1655: =pod
                   1656: 
1.16      foxr     1657: =head2 ToggleDebug
                   1658: 
                   1659: This sub toggles trace debugging on and off.
                   1660: 
                   1661: =cut
                   1662: 
                   1663: sub ToggleDebug {
                   1664:     my $Current    = $DebugLevel;
                   1665:        $DebugLevel = $NextDebugLevel;
                   1666:        $NextDebugLevel = $Current;
                   1667: 
                   1668:     Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
                   1669: 
                   1670: }
                   1671: 
1.69      matthew  1672: =pod
                   1673: 
1.1       foxr     1674: =head2 ChildProcess
                   1675: 
                   1676: This sub implements a child process for a single lonc daemon.
1.61      foxr     1677: Optional parameter:
                   1678:    $socket  - if provided, this is a socket already open for listen
                   1679:               on the client socket. Otherwise, a new listen is set up.
1.1       foxr     1680: 
                   1681: =cut
                   1682: 
                   1683: sub ChildProcess {
1.80      albertel 1684:     #  We've inherited all the
1.62      foxr     1685:     #  events of our parent and those have to be cancelled or else
                   1686:     #  all holy bloody chaos will result.. trust me, I already made
                   1687:     #  >that< mistake.
                   1688: 
                   1689:     my $host = GetServerHost();
                   1690:     foreach my $listener (keys %parent_dispatchers) {
                   1691: 	my $watcher = $parent_dispatchers{$listener};
                   1692: 	my $s       = $watcher->fd;
                   1693: 	if ($listener ne $host) { # Close everyone but me.
                   1694: 	    Debug(5, "Closing listen socket for $listener");
                   1695: 	    $s->close();
                   1696: 	}
                   1697: 	Debug(5, "Killing watcher for $listener");
                   1698: 
                   1699: 	$watcher->cancel();
1.65      foxr     1700: 	delete($parent_dispatchers{$listener});
1.62      foxr     1701: 
                   1702:     }
1.65      foxr     1703: 
                   1704:     #  kill off the parent's signal handlers too!  
                   1705:     #
                   1706: 
                   1707:     for my $handler (keys %parent_handlers) {
                   1708: 	my $watcher = $parent_handlers{$handler};
                   1709: 	$watcher->cancel();
                   1710: 	delete($parent_handlers{$handler});
                   1711:     }
                   1712: 
1.64      foxr     1713:     $I_am_child    = 1;		# Seems like in spite of it all I may still getting
                   1714:                                 # parent event dispatches.. flag I'm a child.
1.1       foxr     1715: 
                   1716: 
1.14      foxr     1717:     #
                   1718:     #  Signals must be handled by the Event framework...
1.61      foxr     1719:     #
1.14      foxr     1720: 
                   1721:     Event->signal(signal   => "QUIT",
                   1722: 		  cb       => \&SignalledToDeath,
                   1723: 		  data     => "QUIT");
                   1724:     Event->signal(signal   => "HUP",
                   1725: 		  cb       => \&ChildStatus,
                   1726: 		  data     => "HUP");
                   1727:     Event->signal(signal   => "USR1",
                   1728: 		  cb       => \&ChildStatus,
                   1729: 		  data     => "USR1");
1.39      foxr     1730:     Event->signal(signal   => "USR2",
                   1731: 		  cb       => \&ToggleTransactionLogging);
1.16      foxr     1732:     Event->signal(signal   => "INT",
                   1733: 		  cb       => \&ToggleDebug,
                   1734: 		  data     => "INT");
1.1       foxr     1735: 
1.62      foxr     1736:     #  Figure out if we got passed a socket or need to open one to listen for
                   1737:     #  client requests.
                   1738: 
1.61      foxr     1739:     my ($socket) = @_;
                   1740:     if (!$socket) {
                   1741: 
                   1742: 	$socket =  SetupLoncListener();
                   1743:     }
1.62      foxr     1744:     #  Establish an event to listen for client connection requests.
                   1745: 
                   1746: 
1.59      foxr     1747:     Event->io(cb   => \&NewClient,
                   1748: 	      poll => 'r',
                   1749: 	      desc => 'Lonc Listener Unix Socket',
                   1750: 	      fd   => $socket);
1.1       foxr     1751:     
1.76      albertel 1752:     $Event::DebugLevel = $DebugLevel;
1.1       foxr     1753:     
                   1754:     Debug(9, "Making initial lond connection for ".$RemoteHost);
                   1755: 
                   1756: # Setup the initial server connection:
                   1757:     
1.62      foxr     1758:      # &MakeLondConnection(); // let first work request do it.
1.10      foxr     1759: 
1.80      albertel 1760:     #  need to accept the connection since the event may  not fire.
1.62      foxr     1761: 
1.80      albertel 1762:     &accept_client($socket);
1.5       foxr     1763: 
1.1       foxr     1764:     Debug(9,"Entering event loop");
                   1765:     my $ret = Event::loop();		#  Start the main event loop.
                   1766:     
                   1767:     
1.64      foxr     1768:     &child_exit (-1,"Main event loop exited!!!");
1.1       foxr     1769: }
                   1770: 
                   1771: #  Create a new child for host passed in:
                   1772: 
                   1773: sub CreateChild {
1.81      albertel 1774:     my ($host, $hostid) = @_;
1.52      foxr     1775: 
1.12      foxr     1776:     my $sigset = POSIX::SigSet->new(SIGINT);
                   1777:     sigprocmask(SIG_BLOCK, $sigset);
1.1       foxr     1778:     $RemoteHost = $host;
1.91      foxr     1779:     ShowStatus('Parent keeping the flock'); # Update time in status message.
1.9       foxr     1780:     Log("CRITICAL", "Forking server for ".$host);
1.23      foxr     1781:     my $pid          = fork;
1.1       foxr     1782:     if($pid) {			# Parent
1.17      foxr     1783: 	$RemoteHost = "Parent";
1.83      albertel 1784: 	$ChildPid{$pid} = $host;
1.12      foxr     1785: 	sigprocmask(SIG_UNBLOCK, $sigset);
1.82      albertel 1786: 	undef(@all_host_ids);
1.1       foxr     1787:     } else {			# child.
1.81      albertel 1788: 	$RemoteHostId = $hostid;
1.5       foxr     1789: 	ShowStatus("Connected to ".$RemoteHost);
1.23      foxr     1790: 	$SIG{INT} = 'DEFAULT';
1.12      foxr     1791: 	sigprocmask(SIG_UNBLOCK, $sigset);
1.81      albertel 1792: 	&ChildProcess();		# Does not return.
1.1       foxr     1793:     }
1.61      foxr     1794: }
1.1       foxr     1795: 
1.61      foxr     1796: # parent_client_connection:
                   1797: #    Event handler that processes client connections for the parent process.
                   1798: #    This sub is called when the parent is listening on a socket and
                   1799: #    a connection request arrives.  We must:
                   1800: #     Start a child process to accept the connection request.
                   1801: #     Kill our listen on the socket.
                   1802: # Parameter:
                   1803: #    event       - The event object that was created to monitor this socket.
                   1804: #                  event->w->fd is the socket.
                   1805: # Returns:
                   1806: #    NONE
                   1807: #
                   1808: sub parent_client_connection {
1.62      foxr     1809:     if ($I_am_child) {
                   1810: 	#  Should not get here, but seem to anyway:
                   1811: 	&Debug(5," Child caught parent client connection event!!");
                   1812: 	my ($event) = @_;
                   1813: 	my $watcher = $event->w;
                   1814: 	$watcher->cancel();	# Try to kill it off again!!
                   1815:     } else {
                   1816: 	&Debug(9, "parent_client_connection");
                   1817: 	my ($event)   = @_;
                   1818: 	my $watcher   = $event->w;
                   1819: 	my $socket    = $watcher->fd;
1.81      albertel 1820: 	my $connection = $socket->accept();	# Accept the client connection.
                   1821: 	Event->io(cb      => \&get_remote_hostname,
                   1822: 		  poll    => 'r',
                   1823: 		  data    => "",
                   1824: 		  fd      => $connection);
1.77      albertel 1825:     }
                   1826: }
                   1827: 
                   1828: sub get_remote_hostname {
1.82      albertel 1829:     my ($event)   = @_;
                   1830:     my $watcher   = $event->w;
                   1831:     my $socket    = $watcher->fd;
                   1832: 
                   1833:     my $thisread;
                   1834:     my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
                   1835:     Debug(8, "rcv:  data length = ".length($thisread)." read =".$thisread);
                   1836:     if (!defined($rv) || length($thisread) == 0) {
                   1837: 	# Likely eof on socket.
                   1838: 	Debug(5,"Client Socket closed on lonc for p_c_c");
                   1839: 	close($socket);
                   1840: 	$watcher->cancel();
                   1841: 	return;
                   1842:     }
                   1843: 
                   1844:     my $data    = $watcher->data().$thisread;
                   1845:     $watcher->data($data);
                   1846:     if($data =~ /\n$/) {	# Request entirely read.
                   1847: 	chomp($data);
                   1848:     } else {
                   1849: 	return;
                   1850:     }
1.77      albertel 1851: 
1.82      albertel 1852:     &Debug(5,"Creating child for $data (parent_client_connection)");
                   1853:     (my $hostname,my $lonid,@all_host_ids) = split(':',$data);
1.83      albertel 1854:     $ChildHost{$hostname}++;
                   1855:     if ($ChildHost{$hostname} == 1) {
                   1856: 	&CreateChild($hostname,$lonid);
                   1857:     } else {
                   1858: 	&Log('WARNING',"Request for a second child on $hostname");
                   1859:     }
1.82      albertel 1860:     # Clean up the listen since now the child takes over until it exits.
                   1861:     $watcher->cancel();		# Nolonger listening to this event
                   1862:     $socket->send("done\n");
                   1863:     $socket->close();
1.61      foxr     1864: }
                   1865: 
                   1866: # parent_listen:
                   1867: #    Opens a socket and starts a listen for the parent process on a client UNIX
                   1868: #    domain socket.
                   1869: #
                   1870: #    This involves:
                   1871: #       Creating a socket for listen.
                   1872: #       Removing any socket lock file
                   1873: #       Adding an event handler for this socket becoming readable
                   1874: #         To the parent's event dispatcher.
                   1875: # Parameters:
                   1876: #    loncapa_host    - LonCAPA cluster name of the host represented by the client
                   1877: #                      socket.
                   1878: # Returns:
                   1879: #    NONE
                   1880: #
                   1881: sub parent_listen {
                   1882:     my ($loncapa_host) = @_;
                   1883:     Debug(5, "parent_listen: $loncapa_host");
                   1884: 
1.78      albertel 1885:     my ($socket,$file);
                   1886:     if (!$loncapa_host) {
                   1887: 	$loncapa_host = 'common_parent';
                   1888: 	$file         = $perlvar{'lonSockCreate'};
                   1889:     } else {
                   1890: 	$file         = &GetLoncSocketPath($loncapa_host);
                   1891:     }
                   1892:     $socket = &SetupLoncListener($loncapa_host,$file);
                   1893: 
1.62      foxr     1894:     $listening_to{$socket} = $loncapa_host;
1.61      foxr     1895:     if (!$socket) {
                   1896: 	die "Unable to create a listen socket for $loncapa_host";
                   1897:     }
                   1898:     
1.78      albertel 1899:     my $lock_file = $file.".lock";
1.61      foxr     1900:     unlink($lock_file);		# No problem if it doesn't exist yet [startup e.g.]
                   1901: 
1.77      albertel 1902:     my $watcher = 
                   1903: 	Event->io(cb    => \&parent_client_connection,
                   1904: 		  poll  => 'r',
                   1905: 		  desc  => "Parent listener unix socket ($loncapa_host)",
                   1906: 		  data => "",
                   1907: 		  fd    => $socket);
1.62      foxr     1908:     $parent_dispatchers{$loncapa_host} = $watcher;
1.61      foxr     1909: 
                   1910: }
                   1911: 
1.77      albertel 1912: sub parent_clean_up {
                   1913:     my ($loncapa_host) = @_;
1.87      albertel 1914:     Debug(1, "parent_clean_up: $loncapa_host");
1.77      albertel 1915: 
                   1916:     my $socket_file = &GetLoncSocketPath($loncapa_host);
                   1917:     unlink($socket_file);	# No problem if it doesn't exist yet [startup e.g.]
                   1918:     my $lock_file   = $socket_file.".lock";
                   1919:     unlink($lock_file);		# No problem if it doesn't exist yet [startup e.g.]
                   1920: }
                   1921: 
1.61      foxr     1922: 
1.83      albertel 1923: 
                   1924: #    This sub initiates a listen on the common unix domain lonc client socket.
                   1925: #    loncnew starts up with no children, and only spawns off children when a
                   1926: #    connection request occurs on the common client unix socket.  The spawned
                   1927: #    child continues to run until it has been idle a while at which point it
                   1928: #    eventually exits and once more the parent picks up the listen.
1.61      foxr     1929: #
                   1930: #  Parameters:
                   1931: #      NONE
                   1932: #  Implicit Inputs:
                   1933: #    The configuration file that has been read in by LondConnection.
                   1934: #  Returns:
                   1935: #     NONE
                   1936: #
1.77      albertel 1937: sub listen_on_common_socket {
                   1938:     Debug(5, "listen_on_common_socket");
1.78      albertel 1939:     &parent_listen();
1.77      albertel 1940: }
                   1941: 
1.63      foxr     1942: #   server_died is called whenever a child process exits.
                   1943: #   Since this is dispatched via a signal, we must process all
                   1944: #   dead children until there are no more left.  The action
                   1945: #   is to:
                   1946: #      - Remove the child from the bookeeping hashes
                   1947: #      - Re-establish a listen on the unix domain socket associated
                   1948: #        with that host.
                   1949: # Parameters:
                   1950: #    The event, but we don't actually care about it.
                   1951: sub server_died {
                   1952:     &Debug(9, "server_died called...");
                   1953:     
                   1954:     while(1) {			# Loop until waitpid nowait fails.
                   1955: 	my $pid = waitpid(-1, WNOHANG);
                   1956: 	if($pid <= 0) {
                   1957: 	    return;		# Nothing left to wait for.
                   1958: 	}
                   1959: 	# need the host to restart:
                   1960: 
1.83      albertel 1961: 	my $host = $ChildPid{$pid};
1.63      foxr     1962: 	if($host) {		# It's for real...
                   1963: 	    &Debug(9, "Caught sigchild for $host");
1.83      albertel 1964: 	    delete($ChildPid{$pid});
                   1965: 	    delete($ChildHost{$host});
1.81      albertel 1966: 	    &parent_clean_up($host);
                   1967: 
1.63      foxr     1968: 	} else {
                   1969: 	    &Debug(5, "Caught sigchild for pid not in hosts hash: $pid");
                   1970: 	}
                   1971:     }
                   1972: 
                   1973: }
                   1974: 
1.1       foxr     1975: #
                   1976: #  Parent process logic pass 1:
                   1977: #   For each entry in the hosts table, we will
                   1978: #  fork off an instance of ChildProcess to service the transactions
                   1979: #  to that host.  Each pid will be entered in a global hash
                   1980: #  with the value of the key, the host.
                   1981: #  The parent will then enter a loop to wait for process exits.
                   1982: #  Each exit gets logged and the child gets restarted.
                   1983: #
                   1984: 
1.5       foxr     1985: #
                   1986: #   Fork and start in new session so hang-up isn't going to 
                   1987: #   happen without intent.
                   1988: #
                   1989: 
                   1990: 
1.6       foxr     1991: 
                   1992: 
1.8       foxr     1993: 
1.6       foxr     1994: 
                   1995: ShowStatus("Forming new session");
                   1996: my $childpid = fork;
                   1997: if ($childpid != 0) {
                   1998:     sleep 4;			# Give child a chacne to break to
                   1999:     exit 0;			# a new sesion.
                   2000: }
1.8       foxr     2001: #
                   2002: #   Write my pid into the pid file so I can be located
                   2003: #
                   2004: 
                   2005: ShowStatus("Parent writing pid file:");
1.23      foxr     2006: my $execdir = $perlvar{'lonDaemons'};
1.8       foxr     2007: open (PIDSAVE, ">$execdir/logs/lonc.pid");
                   2008: print PIDSAVE "$$\n";
                   2009: close(PIDSAVE);
1.6       foxr     2010: 
1.17      foxr     2011: 
                   2012: 
1.6       foxr     2013: if (POSIX::setsid() < 0) {
                   2014:     print "Could not create new session\n";
                   2015:     exit -1;
                   2016: }
1.5       foxr     2017: 
                   2018: ShowStatus("Forking node servers");
                   2019: 
1.9       foxr     2020: Log("CRITICAL", "--------------- Starting children ---------------");
                   2021: 
1.31      foxr     2022: LondConnection::ReadConfig;               # Read standard config files.
1.1       foxr     2023: 
1.80      albertel 2024: $RemoteHost = "[parent]";
1.81      albertel 2025: &listen_on_common_socket();
1.60      foxr     2026: 
1.12      foxr     2027: $RemoteHost = "Parent Server";
1.1       foxr     2028: 
                   2029: # Maintain the population:
1.5       foxr     2030: 
                   2031: ShowStatus("Parent keeping the flock");
1.1       foxr     2032: 
1.12      foxr     2033: 
1.80      albertel 2034: # We need to setup a SIGChild event to handle the exit (natural or otherwise)
                   2035: # of the children.
1.61      foxr     2036: 
1.80      albertel 2037: Event->signal(cb       => \&server_died,
                   2038: 	      desc     => "Child exit handler",
                   2039: 	      signal   => "CHLD");
                   2040: 
                   2041: 
                   2042: # Set up all the other signals we set up.
                   2043: 
                   2044: $parent_handlers{INT} = Event->signal(cb       => \&Terminate,
                   2045: 				      desc     => "Parent INT handler",
                   2046: 				      signal   => "INT");
                   2047: $parent_handlers{TERM} = Event->signal(cb       => \&Terminate,
                   2048: 				       desc     => "Parent TERM handler",
                   2049: 				       signal   => "TERM");
1.81      albertel 2050: $parent_handlers{HUP}  = Event->signal(cb       => \&KillThemAll,
                   2051: 				       desc     => "Parent HUP handler.",
                   2052: 				       signal   => "HUP");
1.80      albertel 2053: $parent_handlers{USR1} = Event->signal(cb       => \&CheckKids,
                   2054: 				       desc     => "Parent USR1 handler",
                   2055: 				       signal   => "USR1");
                   2056: $parent_handlers{USR2} = Event->signal(cb       => \&UpdateKids,
                   2057: 				       desc     => "Parent USR2 handler.",
                   2058: 				       signal   => "USR2");
                   2059: 
                   2060: #  Start procdesing events.
                   2061: 
                   2062: $Event::DebugLevel = $DebugLevel;
                   2063: Debug(9, "Parent entering event loop");
                   2064: my $ret = Event::loop();
                   2065: die "Main Event loop exited: $ret";
1.14      foxr     2066: 
                   2067: =pod
                   2068: 
                   2069: =head1 CheckKids
                   2070: 
                   2071:   Since kids do not die as easily in this implementation
                   2072: as the previous one, there  is no need to restart the
                   2073: dead ones (all dead kids get restarted when they die!!)
                   2074: The only thing this function does is to pass USR1 to the
                   2075: kids so that they report their status.
                   2076: 
                   2077: =cut
                   2078: 
                   2079: sub CheckKids {
                   2080:     Debug(2, "Checking status of children");
                   2081:     my $docdir = $perlvar{'lonDocRoot'};
                   2082:     my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt");
                   2083:     my $now=time;
                   2084:     my $local=localtime($now);
                   2085:     print $fh "LONC status $local - parent $$ \n\n";
1.65      foxr     2086:     foreach my $host (keys %parent_dispatchers) {
                   2087: 	print $fh "LONC Parent process listening for $host\n";
                   2088:     }
1.83      albertel 2089:     foreach my $pid (keys %ChildPid) {
1.14      foxr     2090: 	Debug(2, "Sending USR1 -> $pid");
                   2091: 	kill 'USR1' => $pid;	# Tell Child to report status.
                   2092:     }
1.65      foxr     2093: 
1.14      foxr     2094: }
1.24      foxr     2095: 
                   2096: =pod
                   2097: 
                   2098: =head1  UpdateKids
                   2099: 
1.25      foxr     2100: parent's SIGUSR2 handler.  This handler:
1.24      foxr     2101: 
                   2102: =item
                   2103: 
                   2104: Rereads the hosts file.
                   2105: 
                   2106: =item
                   2107:  
                   2108: Kills off (via sigint) children for hosts that have disappeared.
                   2109: 
                   2110: =item
                   2111: 
1.27      foxr     2112: QUITs  children for hosts that already exist (this just forces a status display
1.24      foxr     2113: and resets the connection retry count for that host.
                   2114: 
                   2115: =item
                   2116: 
                   2117: Starts new children for hosts that have been added to the hosts.tab file since
                   2118: the start of the master program and maintains them.
                   2119: 
                   2120: =cut
                   2121: 
                   2122: sub UpdateKids {
1.27      foxr     2123: 
1.25      foxr     2124:     Log("INFO", "Updating connections via SIGUSR2");
1.27      foxr     2125: 
1.65      foxr     2126:     #  I'm not sure what I was thinking in the first implementation.
                   2127:     # someone will have to work hard to convince me the effect is any
                   2128:     # different than Restart, especially now that we don't start up 
                   2129:     # per host servers automatically, may as well just restart.
                   2130:     # The down side is transactions that are in flight will get timed out
                   2131:     # (lost unless they are critical).
1.27      foxr     2132: 
1.81      albertel 2133:     &KillThemAll();
1.24      foxr     2134: }
                   2135: 
1.14      foxr     2136: 
1.13      foxr     2137: =pod
                   2138: 
                   2139: =head1 Restart
                   2140: 
                   2141: Signal handler for HUP... all children are killed and
                   2142: we self restart.  This is an el-cheapo way to re read
                   2143: the config file.
                   2144: 
                   2145: =cut
                   2146: 
                   2147: sub Restart {
1.23      foxr     2148:     &KillThemAll;		# First kill all the children.
1.13      foxr     2149:     Log("CRITICAL", "Restarting");
                   2150:     my $execdir = $perlvar{'lonDaemons'};
                   2151:     unlink("$execdir/logs/lonc.pid");
1.65      foxr     2152:     exec("$executable");
1.10      foxr     2153: }
1.12      foxr     2154: 
                   2155: =pod
                   2156: 
                   2157: =head1 KillThemAll
                   2158: 
                   2159: Signal handler that kills all children by sending them a 
1.17      foxr     2160: SIGHUP.  Responds to sigint and sigterm.
1.12      foxr     2161: 
                   2162: =cut
                   2163: 
1.10      foxr     2164: sub KillThemAll {
1.12      foxr     2165:     Debug(2, "Kill them all!!");
1.85      albertel 2166:     
                   2167:     #local($SIG{CHLD}) = 'IGNORE';
                   2168:     # Our children >will< die.
                   2169:     # but we need to catch their death and cleanup after them in case this is 
                   2170:     # a restart set of kills
                   2171:     my @allpids = keys(%ChildPid);
                   2172:     foreach my $pid (@allpids) {
1.83      albertel 2173: 	my $serving = $ChildPid{$pid};
1.52      foxr     2174: 	ShowStatus("Nicely Killing lonc for $serving pid = $pid");
                   2175: 	Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
1.17      foxr     2176: 	kill 'QUIT' => $pid;
1.12      foxr     2177:     }
1.85      albertel 2178:     ShowStatus("Finished killing child processes off.");
1.1       foxr     2179: }
1.12      foxr     2180: 
1.52      foxr     2181: 
                   2182: #
                   2183: #  Kill all children via KILL.  Just in case the
                   2184: #  first shot didn't get them.
                   2185: 
                   2186: sub really_kill_them_all_dammit
                   2187: {
                   2188:     Debug(2, "Kill them all Dammit");
                   2189:     local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
1.83      albertel 2190:     foreach my $pid (keys %ChildPid) {
                   2191: 	my $serving = $ChildPid{$pid};
1.52      foxr     2192: 	&ShowStatus("Nastily killing lonc for $serving pid = $pid");
                   2193: 	Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
                   2194: 	kill 'KILL' => $pid;
1.83      albertel 2195: 	delete($ChildPid{$pid});
1.52      foxr     2196: 	my $execdir = $perlvar{'lonDaemons'};
                   2197: 	unlink("$execdir/logs/lonc.pid");
                   2198:     }
                   2199: }
1.69      matthew  2200: 
1.14      foxr     2201: =pod
                   2202: 
                   2203: =head1 Terminate
                   2204:  
                   2205: Terminate the system.
                   2206: 
                   2207: =cut
                   2208: 
                   2209: sub Terminate {
1.52      foxr     2210:     &Log("CRITICAL", "Asked to kill children.. first be nice...");
                   2211:     &KillThemAll;
                   2212:     #
                   2213:     #  By now they really should all be dead.. but just in case 
                   2214:     #  send them all SIGKILL's after a bit of waiting:
                   2215: 
                   2216:     sleep(4);
                   2217:     &Log("CRITICAL", "Now kill children nasty");
                   2218:     &really_kill_them_all_dammit;
1.17      foxr     2219:     Log("CRITICAL","Master process exiting");
                   2220:     exit 0;
1.14      foxr     2221: 
                   2222: }
1.81      albertel 2223: 
                   2224: sub my_hostname {
                   2225:     use Sys::Hostname;
                   2226:     my $name = &hostname();
                   2227:     &Debug(9,"Name is $name");
                   2228:     return $name;
                   2229: }
                   2230: 
1.12      foxr     2231: =pod
1.1       foxr     2232: 
                   2233: =head1 Theory
1.3       albertel 2234: 
                   2235: The event class is used to build this as a single process with an
                   2236: event driven model.  The following events are handled:
1.1       foxr     2237: 
                   2238: =item UNIX Socket connection Received
                   2239: 
                   2240: =item Request data arrives on UNIX data transfer socket.
                   2241: 
                   2242: =item lond connection becomes writable.
                   2243: 
                   2244: =item timer fires at 1 second intervals.
                   2245: 
                   2246: All sockets are run in non-blocking mode.  Timeouts managed by the timer
                   2247: handler prevents hung connections.
                   2248: 
                   2249: Key data structures:
                   2250: 
1.3       albertel 2251: =item RequestQueue
                   2252: 
                   2253: A queue of requests received from UNIX sockets that are
                   2254: waiting for a chance to be forwarded on a lond connection socket.
                   2255: 
                   2256: =item ActiveConnections
                   2257: 
                   2258: A hash of lond connections that have transactions in process that are
                   2259: available to be timed out.
                   2260: 
                   2261: =item ActiveTransactions
                   2262: 
                   2263: A hash indexed by lond connections that contain the client reply
                   2264: socket for each connection that has an active transaction on it.
                   2265: 
                   2266: =item IdleConnections
                   2267: 
                   2268: A hash of lond connections that have no work to do.  These connections
                   2269: can be closed if they are idle for a long enough time.
1.1       foxr     2270: 
                   2271: =cut
1.88      foxr     2272: 
                   2273: =pod
                   2274: 
                   2275: =head1 Log messages
                   2276: 
                   2277: The following is a list of log messages that can appear in the 
                   2278: lonc.log file.  Each log file has a severity and a message.
                   2279: 
                   2280: =over 2
                   2281: 
                   2282: =item Warning  A socket timeout was detected
                   2283: 
                   2284: If there are pending transactions in the socket's queue,
                   2285: they are failed (saved if critical).  If the connection
                   2286: retry count gets exceeded by this, the
                   2287: remote host is marked as dead.
                   2288: Called when timeouts occured during the connection and
                   2289: connection dialog with a remote host.
                   2290: 
                   2291: =item Critical Host makred DEAD <hostname>   
                   2292: 
                   2293: The numer of retry counts for contacting a host was
                   2294: exceeded. The host is marked dead an no 
                   2295: further attempts will be made by that child.
                   2296: 
                   2297: =item Info lonc pipe client hung up on us     
                   2298: 
                   2299: Write to the client pipe indicated no data transferred
                   2300: Socket to remote host is shut down.  Reply to the client 
                   2301: is discarded.  Note: This is commented out in &ClientWriteable
                   2302: 
                   2303: =item Success  Reply from lond: <data>   
                   2304: 
                   2305: Can be enabled for debugging by setting LogTransactions to nonzero.
                   2306: Indicates a successful transaction with lond, <data> is the data received
                   2307: from the remote lond.
                   2308: 
                   2309: =item Success A delayed transaction was completed  
                   2310: 
                   2311: A transaction that must be reliable was executed and completed
                   2312: as lonc restarted.  This is followed by a mesage of the form
                   2313: 
                   2314:   S: client-name : request
                   2315: 
                   2316: =item WARNING  Failing transaction <cmd>:<subcmd>  
                   2317: 
                   2318: Transaction failed on a socket, but the failure retry count for the remote
                   2319: node has not yet been exhausted (the node is not yet marked dead).
                   2320: cmd is the command, subcmd is the subcommand.  This results from a con_lost
                   2321: when communicating with lond.
                   2322: 
                   2323: =item WARNING Shutting down a socket     
                   2324: 
                   2325: Called when a socket is being closed to lond.  This is emitted both when 
                   2326: idle pruning is being done and when the socket has been disconnected by the remote.
                   2327: 
                   2328: =item WARNING Lond connection lost.
                   2329: 
                   2330: Called when a read from lond's socket failed indicating lond has closed the 
                   2331: connection or died.  This should be followed by one or more
                   2332: 
                   2333:  "WARNING Failing transaction..." msgs for each in-flight or queued transaction.
                   2334: 
                   2335: =item INFO Connected to lond version:  <version> 
                   2336: 
                   2337: When connection negotiation is complete, the lond version is requested and logged here.
                   2338: 
                   2339: =item SUCCESS Connection n to host now ready for action
                   2340: 
                   2341: Emitted when connection has been completed with lond. n is then number of 
                   2342: concurrent connections and host, the host to which the connection has just
                   2343: been established.
                   2344: 
                   2345: =item WARNING Connection to host has been disconnected
                   2346: 
                   2347: Write to a lond resulted in failure status.  Connection to lond is dropped.
                   2348: 
                   2349: =item SUCCESS Created connection n to host host 
                   2350: 
                   2351: Initial connection request to host..(before negotiation).
                   2352: 
                   2353: =item CRITICAL Request Close Connection ... exiting
                   2354: 
                   2355: Client has sent "close_connection_exit"   The loncnew server is exiting.
                   2356: 
                   2357: =item INFO Resetting Connection Retries 
                   2358: 
                   2359: Client has sent "reset_retries" The lond connection retries are reset to zero for the
                   2360: corresponding lond.
                   2361: 
                   2362: =item SUCCESS Transaction <data>
                   2363: 
                   2364: Only emitted if the global variable $LogTransactions was set to true.
                   2365: A client has requested a lond transaction <data> is the contents of the request.
                   2366: 
                   2367: =item SUCCESS Toggled transaction logging <LogTransactions>
                   2368:                                     
                   2369: The state of the $LogTransactions global has been toggled, and its current value
                   2370: (after being toggled) is displayed.  When non zero additional logging of transactions
                   2371: is enabled for debugging purposes.  Transaction logging is toggled on receipt of a USR2
                   2372: signal.
                   2373: 
                   2374: =item CRITICAL Abnormal exit. Child <pid> for <host> died thorugh signal.
                   2375: 
                   2376: QUIT signal received.  lonc child process is exiting.
                   2377: 
                   2378: =item SUCCESS New debugging level for <RemoteHost> now <DebugLevel>
                   2379:                                     
                   2380: Debugging toggled for the host loncnew is talking with.
                   2381: Currently debugging is a level based scheme with higher number 
                   2382: conveying more information.  The daemon starts out at
                   2383: DebugLevel 0 and can toggle back and forth between that and
                   2384: DebugLevel 2  These are controlled by
                   2385: the global variables $DebugLevel and $NextDebugLevel
                   2386: The debug level can go up to 9.
                   2387: SIGINT toggles the debug level.  The higher the debug level the 
                   2388: more debugging information is spewed.  See the Debug
                   2389: sub in loncnew.
                   2390: 
                   2391: =item CRITICAL Forking server for host  
                   2392: 
                   2393: A child is being created to service requests for the specified host.
                   2394: 
                   2395: 
                   2396: =item WARNING Request for a second child on hostname
                   2397:                                     
                   2398: Somehow loncnew was asked to start a second child on a host that already had a child
                   2399: servicing it.  This request is not honored, but themessage is emitted.  This could happen
                   2400: due to a race condition.  When a client attempts to contact loncnew for a new host, a child
                   2401: is forked off to handle the requests for that server.  The parent then backs off the Unix
                   2402: domain socket leaving it for the child to service all requests.  If in the time between
                   2403: creating the child, and backing off, a new connection request comes in to the unix domain
                   2404: socket, this could trigger (unlikely but remotely possible),.
                   2405: 
                   2406: =item CRITICAL ------ Starting Children ----
                   2407: 
                   2408: This message should probably be changed to "Entering event loop"  as the loncnew only starts
                   2409: children as needed.  This message is emitted as new events are established and
                   2410: the event processing loop is entered.
                   2411: 
                   2412: =item INFO Updating connections via SIGUSR2
                   2413:                                     
                   2414: SIGUSR2 received. The original code would kill all clients, re-read the host file,
                   2415: then restart children for each host.  Now that childrean aree started on demand, this
                   2416: just kills all child processes and lets requests start them as needed again.
                   2417: 
                   2418: 
                   2419: =item CRITICAL Restarting
                   2420: 
                   2421: SigHUP received.  all the children are killed and the script exec's itself to start again.
                   2422: 
                   2423: =item CRITICAL Nicely killing lonc for host pid = <pid>
                   2424: 
                   2425: Attempting to kill the child that is serving the specified host (pid given) cleanly via
                   2426: SIGQUIT  The child should handle that, clean up nicely and exit.
                   2427: 
                   2428: =item CRITICAL Nastily killing lonc for host pid = <pid>
                   2429: 
                   2430: The child specified did not die when requested via SIGQUIT.  Therefore it is killed
                   2431: via SIGKILL.
                   2432: 
                   2433: =item CRITICAL Asked to kill children.. first be nice..
                   2434: 
                   2435: In the parent's INT handler.  INT kills the child processes.  This inidicate loncnew
                   2436: is about to attempt to kill all known children via SIGQUIT.  This message should be followed 
                   2437: by one "Nicely killing" message for each extant child.
                   2438: 
                   2439: =item CRITICAL Now kill children nasty 
                   2440: 
                   2441: In the parent's INT handler. remaining children are about to be killed via
                   2442: SIGKILL. Should be followed by a Nastily killing... for each lonc child that 
                   2443: refused to die.
                   2444: 
                   2445: =item CRITICAL Master process exiting
                   2446: 
                   2447: In the parent's INT handler. just prior to the exit 0 call.
                   2448: 
                   2449: =back
                   2450: 
                   2451: =cut

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