--- loncom/lond 2008/07/07 15:23:41 1.407 +++ loncom/lond 2009/06/11 19:19:51 1.416 @@ -2,7 +2,7 @@ # The LearningOnline Network # lond "LON Daemon" Server (port "LOND" 5663) # -# $Id: lond,v 1.407 2008/07/07 15:23:41 raeburn Exp $ +# $Id: lond,v 1.416 2009/06/11 19:19:51 raeburn Exp $ # # Copyright Michigan State University Board of Trustees # @@ -59,7 +59,7 @@ my $DEBUG = 0; # Non zero to ena my $status=''; my $lastlog=''; -my $VERSION='$Revision: 1.407 $'; #' stupid emacs +my $VERSION='$Revision: 1.416 $'; #' stupid emacs my $remoteVERSION; my $currenthostid="default"; my $currentdomainid; @@ -142,6 +142,16 @@ my @adderrors = ("ok", "lcuseradd Password mismatch"); +# This array are the errors from lcinstallfile: + +my @installerrors = ("ok", + "Initial user id of client not that of www", + "Usage error, not enough command line arguments", + "Source file name does not exist", + "Destination file name does not exist", + "Some file operation failed", + "Invalid table filename." + ); # # Statistics that are maintained and dislayed in the status line. @@ -398,6 +408,7 @@ sub isClient { # sub ReadManagerTable { + &Debug("Reading manager table"); # Clean out the old table first.. foreach my $key (keys %managers) { @@ -520,11 +531,9 @@ sub AdjustHostContents { } # # InstallFile: Called to install an administrative file: -# - The file is created with .tmp -# - The .tmp file is then mv'd to -# This lugubrious procedure is done to ensure that we are never without -# a valid, even if dated, version of the file regardless of who crashes -# and when the crash occurs. +# - The file is created int a temp directory called .tmp +# - lcinstall file is called to install the file. +# since the web app has no direct write access to the table directory # # Parameters: # Name of the file @@ -532,11 +541,16 @@ sub AdjustHostContents { # Return: # nonzero - success. # 0 - failure and $! has an errno. +# Assumptions: +# File installtion is a relatively infrequent # sub InstallFile { my ($Filename, $Contents) = @_; - my $TempFile = $Filename.".tmp"; +# my $TempFile = $Filename.".tmp"; + my $exedir = $perlvar{'lonDaemons'}; + my $tmpdir = $exedir.'/tmp/'; + my $TempFile = $tmpdir."TempTableFile.tmp"; # Open the file for write: @@ -550,11 +564,27 @@ sub InstallFile { print $fh ($Contents); $fh->close; # In case we ever have a filesystem w. locking - chmod(0660, $TempFile); + chmod(0664, $TempFile); # Everyone can write it. - # Now we can move install the file in position. - - move($TempFile, $Filename); + # Use lcinstall file to put the file in the table directory... + + &Debug("Opening pipe to $exedir/lcinstallfile $TempFile $Filename"); + my $pf = IO::File->new("| $exedir/lcinstallfile $TempFile $Filename > $exedir/logs/lcinstallfile.log"); + close $pf; + my $err = $?; + &Debug("Status is $err"); + if ($err != 0) { + my $msg = $err; + if ($err < @installerrors) { + $msg = $installerrors[$err]; + } + &logthis("Install failed for table file $Filename : $msg"); + return 0; + } + + # Remove the temp file: + + unlink($TempFile); return 1; } @@ -562,8 +592,13 @@ sub InstallFile { # # ConfigFileFromSelector: converts a configuration file selector -# (one of host or domain at this point) into a -# configuration file pathname. +# into a configuration file pathname. +# It's probably no longer necessary to preserve +# special handling of hosts or domain as those +# files have been superceded by dns_hosts, dns_domain. +# The default action is just to prepend the directory +# and append .tab +# # # Parameters: # selector - Configuration file selector. @@ -580,7 +615,7 @@ sub ConfigFileFromSelector { } elsif ($selector eq "domain") { $tablefile = $tabledir."domain.tab"; } else { - return undef; + $tablefile = $tabledir.$selector.'.tab'; } return $tablefile; @@ -603,6 +638,7 @@ sub ConfigFileFromSelector { sub PushFile { my $request = shift; my ($command, $filename, $contents) = split(":", $request, 3); + &Debug("PushFile"); # At this point in time, pushes for only the following tables are # supported: @@ -619,20 +655,7 @@ sub PushFile { if(! (defined $tablefile)) { return "refused"; } - # - # >copy< the old table to the backup table - # don't rename in case system crashes/reboots etc. in the time - # window between a rename and write. - # - my $backupfile = $tablefile; - $backupfile =~ s/\.tab$/.old/; - if(!CopyFile($tablefile, $backupfile)) { - &logthis(' CopyFile from '.$tablefile." to ".$backupfile." failed "); - return "error:$!"; - } - &logthis(' Pushfile: backed up ' - .$tablefile." to $backupfile"); - + # If the file being pushed is the host file, we adjust the entry for ourself so that the # IP will be our current IP as looked up in dns. Note this is only 99% good as it's possible # to conceive of conditions where we don't have a DNS entry locally. This is possible in a @@ -645,6 +668,7 @@ sub PushFile { # Install the new file: + &logthis("Installing new $tablefile contents:\n$contents"); if(!InstallFile($tablefile, $contents)) { &logthis(' Pushfile: unable to install ' .$tablefile." $! "); @@ -1198,7 +1222,7 @@ sub user_authorization_type { # a reply is written to the client. sub push_file_handler { my ($cmd, $tail, $client) = @_; - + &Debug("In push file handler"); my $userinput = "$cmd:$tail"; # At this time we only know that the IP of our partner is a valid manager @@ -1206,7 +1230,8 @@ sub push_file_handler { # spoofing). my $cert = &GetCertificate($userinput); - if(&ValidManager($cert)) { + if(&ValidManager($cert)) { + &Debug("Valid manager: $client"); # Now presumably we have the bona fides of both the peer host and the # process making the request. @@ -1215,6 +1240,7 @@ sub push_file_handler { &Reply($client, \$reply, $userinput); } else { + &logthis("push_file_handler $client is not valid"); &Failure( $client, "refused\n", $userinput); } return 1; @@ -1538,7 +1564,7 @@ sub ls3_handler { } elsif ($alternate_root ne '') { $dir_root = $alternate_root; } - if ($dir_root ne '') { + if (($dir_root ne '') && ($dir_root ne '/')) { if ($ulsdir =~ /^\//) { $ulsdir = $dir_root.$ulsdir; } else { @@ -1587,6 +1613,46 @@ sub ls3_handler { } ®ister_handler("ls3", \&ls3_handler, 0, 1, 0); +sub server_timezone_handler { + my ($cmd,$tail,$client) = @_; + my $userinput = "$cmd:$tail"; + my $timezone; + my $clockfile = '/etc/sysconfig/clock'; # Fedora/CentOS/SuSE + my $tzfile = '/etc/timezone'; # Debian/Ubuntu + if (-e $clockfile) { + if (open(my $fh,"<$clockfile")) { + while (<$fh>) { + next if (/^[\#\s]/); + if (/^(?:TIME)?ZONE\s*=\s*['"]?\s*([\w\/]+)/) { + $timezone = $1; + last; + } + } + close($fh); + } + } elsif (-e $tzfile) { + if (open(my $fh,"<$tzfile")) { + $timezone = <$fh>; + close($fh); + chomp($timezone); + if ($timezone =~ m{^Etc/(\w+)$}) { + $timezone = $1; + } + } + } + &Reply($client,\$timezone,$userinput); # This supports debug logging. + return 1; +} +®ister_handler("servertimezone", \&server_timezone_handler, 0, 1, 0); + +sub server_loncaparev_handler { + my ($cmd,$tail,$client) = @_; + my $userinput = "$cmd:$tail"; + &Reply($client,\$perlvar{'lonVersion'},$userinput); + return 1; +} +®ister_handler("serverloncaparev", \&server_loncaparev_handler, 0, 1, 0); + # Process a reinit request. Reinit requests that either # lonc or lond be reinitialized so that an updated # host.tab or domain.tab can be processed. @@ -4277,12 +4343,12 @@ sub dump_domainroles_handler { my ($start,$end) = split(/:/,&unescape($value)); my ($trole,$uname,$udom,$runame,$rudom,$rsec) = split(/:/,&unescape($key)); unless ($startfilter eq '.' || !defined($startfilter)) { - if ($start >= $startfilter) { + if ((defined($start)) && ($start >= $startfilter)) { $match = 0; } } unless ($endfilter eq '.' || !defined($endfilter)) { - if ($end <= $endfilter) { + if ((defined($end)) && ($end <= $endfilter)) { $match = 0; } } @@ -4862,6 +4928,37 @@ sub get_institutional_defaults_handler { ®ister_handler("autoinstcodedefaults", \&get_institutional_defaults_handler,0,1,0); +sub get_possible_instcodes_handler { + my ($cmd, $tail, $client) = @_; + my $userinput = "$cmd:$tail"; + + my $reply; + my $cdom = $tail; + my (@codetitles,%cat_titles,%cat_order); + my $formatreply = &localenroll::possible_instcodes($cdom, + \@codetitles, + \%cat_titles, + \%cat_order); + if ($formatreply eq 'ok') { + my $result = join('&',map {&escape($_);} (@codetitles)).':'; + foreach my $key (keys(%cat_titles)) { + $result .= &escape($key).'='.&Apache::lonnet::freeze_escape($cat_titles{$key}).'&'; + } + $result =~ s/\&$//; + $result .= ':'; + foreach my $key (keys(%cat_order)) { + $result .= &escape($key).'='.&Apache::lonnet::freeze_escape($cat_order{$key}).'&'; + } + $result =~ s/\&$//; + &Reply($client,\$result,$userinput); + } else { + &Reply($client, "format_error\n", $userinput); + } + return 1; +} +®ister_handler("autopossibleinstcodes", + \&get_possible_instcodes_handler,0,1,0); + sub get_institutional_user_rules { my ($cmd, $tail, $client) = @_; my $userinput = "$cmd:$tail"; @@ -5944,7 +6041,7 @@ sub make_new_child { if ($clientip eq '127.0.0.1') { $outsideip=&Apache::lonnet::get_host_ip($perlvar{'lonHostID'}); } - + &ReadManagerTable(); my $clientrec=defined(&Apache::lonnet::get_hosts_from_ip($outsideip)); my $ismanager=($managers{$outsideip} ne undef); $clientname = "[unknonwn]"; @@ -7118,3 +7215,406 @@ linux Server/Process =cut + + +=pod + +=head1 LOG MESSAGES + +The messages below can be emitted in the lond log. This log is located +in ~httpd/perl/logs/lond.log Many log messages have HTML encapsulation +to provide coloring if examined from inside a web page. Some do not. +Where color is used, the colors are; Red for sometihhng to get excited +about and to follow up on. Yellow for something to keep an eye on to +be sure it does not get worse, Green,and Blue for informational items. + +In the discussions below, sometimes reference is made to ~httpd +when describing file locations. There isn't really an httpd +user, however there is an httpd directory that gets installed in the +place that user home directories go. On linux, this is usually +(always?) /home/httpd. + + +Some messages are colorless. These are usually (not always) +Green/Blue color level messages. + +=over 2 + +=item (Red) LocalConnection rejecting non local: ne 127.0.0.1 + +A local connection negotiation was attempted by +a host whose IP address was not 127.0.0.1. +The socket is closed and the child will exit. +lond has three ways to establish an encyrption +key with a client: + +=over 2 + +=item local + +The key is written and read from a file. +This is only valid for connections from localhost. + +=item insecure + +The key is generated by the server and +transmitted to the client. + +=item ssl (secure) + +An ssl connection is negotiated with the client, +the key is generated by the server and sent to the +client across this ssl connection before the +ssl connectionis terminated and clear text +transmission resumes. + +=back + +=item (Red) LocalConnection: caller is insane! init = and type = + +The client is local but has not sent an initialization +string that is the literal "init:local" The connection +is closed and the child exits. + +=item Red CRITICAL Can't get key file + +SSL key negotiation is being attempted but the call to +lonssl::KeyFile failed. This usually means that the +configuration file is not correctly defining or protecting +the directories/files lonCertificateDirectory or +lonnetPrivateKey + is a string that describes the reason that +the key file could not be located. + +=item (Red) CRITICAL Can't get certificates + +SSL key negotiation failed because we were not able to retrives our certificate +or the CA's certificate in the call to lonssl::CertificateFile + is the textual reason this failed. Usual reasons: + +=over 2 + +=item Apache config file for loncapa incorrect: + +one of the variables +lonCertificateDirectory, lonnetCertificateAuthority, or lonnetCertificate +undefined or incorrect + +=item Permission error: + +The directory pointed to by lonCertificateDirectory is not readable by lond + +=item Permission error: + +Files in the directory pointed to by lonCertificateDirectory are not readable by lond. + +=item Installation error: + +Either the certificate authority file or the certificate have not +been installed in lonCertificateDirectory. + +=item (Red) CRITICAL SSL Socket promotion failed: + +The promotion of the connection from plaintext to SSL failed + is the reason for the failure. There are two +system calls involved in the promotion (one of which failed), +a dup to produce +a second fd on the raw socket over which the encrypted data +will flow and IO::SOcket::SSL->new_from_fd which creates +the SSL connection on the duped fd. + +=item (Blue) WARNING client did not respond to challenge + +This occurs on an insecure (non SSL) connection negotiation request. +lond generates some number from the time, the PID and sends it to +the client. The client must respond by echoing this information back. +If the client does not do so, that's a violation of the challenge +protocols and the connection will be failed. + +=item (Red) No manager table. Nobody can manage!! + +lond has the concept of privileged hosts that +can perform remote management function such +as update the hosts.tab. The manager hosts +are described in the +~httpd/lonTabs/managers.tab file. +this message is logged if this file is missing. + + +=item (Green) Registering manager as with + +Reports the successful parse and registration +of a specific manager. + +=item Green existing host + +The manager host is already defined in the hosts.tab +the information in that table, rather than the info in the +manager table will be used to determine the manager's ip. + +=item (Red) Unable to craete + +lond has been asked to create new versions of an administrative +file (by a manager). When this is done, the new file is created +in a temp file and then renamed into place so that there are always +usable administrative files, even if the update fails. This failure +message means that the temp file could not be created. +The update is abandoned, and the old file is available for use. + +=item (Green) CopyFile from to failed + +In an update of administrative files, the copy of the existing file to a +backup file failed. The installation of the new file may still succeed, +but there will not be a back up file to rever to (this should probably +be yellow). + +=item (Green) Pushfile: backed up to + +See above, the backup of the old administrative file succeeded. + +=item (Red) Pushfile: Unable to install + +The new administrative file could not be installed. In this case, +the old administrative file is still in use. + +=item (Green) Installed new < filename>. + +The new administrative file was successfullly installed. + +=item (Red) Reinitializing lond pid= + +The lonc child process will be sent a USR2 +signal. + +=item (Red) Reinitializing self + +We've been asked to re-read our administrative files,and +are doing so. + +=item (Yellow) error:Invalid process identifier + +A reinit command was received, but the target part of the +command was not valid. It must be either +'lond' or 'lonc' but was + +=item (Green) isValideditCommand checking: Command = Key = newline = + +Checking to see if lond has been handed a valid edit +command. It is possible the edit command is not valid +in that case there are no log messages to indicate that. + +=item Result of password change for pwchange_success + +The password for was +successfully changed. + +=item Unable to open passwd to change password + +Could not rewrite the +internal password file for a user + +=item Result of password change for : + +A unix password change for was attempted +and the pipe returned + +=item LWP GET: for () + +The lightweight process fetch for a resource failed +with the local filename that should +have existed/been created was the +corresponding URI: This is emitted in several +places. + +=item Unable to move to + +From fetch_user_file_handler - the user file was replicated but could not +be mv'd to its final location. + +=item Looking for + +From user_has_session_handler - This should be a Debug call instead +it indicates lond is about to check whether the specified user has a +session active on the specified domain on the local host. + +=item Client () hanging up: + +lond has been asked to exit by its client. The and identify the +client systemand is the full exit command sent to the server. + +=item Red CRITICAL: ABNORMAL EXIT. child for server died through a crass with this error->[]. + +A lond child terminated. NOte that this termination can also occur when the +child receives the QUIT or DIE signals. is the process id of the child, + the host lond is working for, and the reason the child died +to the best of our ability to get it (I would guess that any numeric value +represents and errno value). This is immediately followed by + +=item Famous last words: Catching exception - + +Where log is some recent information about the state of the child. + +=item Red CRITICAL: TIME OUT + +Some timeout occured for server . THis is normally a timeout on an LWP +doing an HTTP::GET. + +=item child died + +The reaper caught a SIGCHILD for the lond child process +This should be modified to also display the IP of the dying child +$children{$pid} + +=item Unknown child 0 died +A child died but the wait for it returned a pid of zero which really should not +ever happen. + +=item Child - looks like we missed it's death + +When a sigchild is received, the reaper process checks all children to see if they are +alive. If children are dying quite quickly, the lack of signal queuing can mean +that a signal hearalds the death of more than one child. If so this message indicates +which other one died. is the ip of a dead child + +=item Free socket: + +The HUNTSMAN sub was called due to a SIGINT in a child process. The socket is being shutdown. +for whatever reason, is printed but in fact shutdown() is not documented +to return anything. This is followed by: + +=item Red CRITICAL: Shutting down + +Just prior to exit. + +=item Free socket: + +The HUPSMAN sub was called due to a SIGHUP. all children get killsed, and lond execs itself. +This is followed by: + +=item (Red) CRITICAL: Restarting + +lond is about to exec itself to restart. + +=item (Blue) Updating connections + +(In response to a USR2). All the children (except the one for localhost) +are about to be killed, the hosts tab reread, and Apache reloaded via apachereload. + +=item (Blue) UpdateHosts killing child for ip + +Due to USR2 as above. + +=item (Green) keeping child for ip (pid = ) + +In response to USR2 as above, the child indicated is not being restarted because +it's assumed that we'll always need a child for the localhost. + + +=item Going to check on the children + +Parent is about to check on the health of the child processes. +Note that this is in response to a USR1 sent to the parent lond. +there may be one or more of the next two messages: + +=item is dead + +A child that we have in our child hash as alive has evidently died. + +=item Child did not respond + +In the health check the child did not update/produce a pid_.txt +file when sent it's USR1 signal. That process is killed with a 9 signal, as it's +assumed to be hung in some un-fixable way. + +=item Finished checking children + +Master processs's USR1 processing is cojmplete. + +=item (Red) CRITICAL: ------- Starting ------ + +(There are more '-'s on either side). Lond has forked itself off to +form a new session and is about to start actual initialization. + +=item (Green) Attempting to start child () + +Started a new child process for . Client is IO::Socket object +connected to the child. This was as a result of a TCP/IP connection from a client. + +=item Unable to determine who caller was, getpeername returned nothing + +In child process initialization. either getpeername returned undef or +a zero sized object was returned. Processing continues, but in my opinion, +this should be cause for the child to exit. + +=item Unable to determine clientip + +In child process initialization. The peer address from getpeername was not defined. +The client address is stored as "Unavailable" and processing continues. + +=item (Yellow) INFO: Connection connection type = + +In child initialization. A good connectionw as received from . + +=over 2 + +=item + +is the name of the client from hosts.tab. + +=item + +Is the connection type which is either + +=over 2 + +=item manager + +The connection is from a manager node, not in hosts.tab + +=item client + +the connection is from a non-manager in the hosts.tab + +=item both + +The connection is from a manager in the hosts.tab. + +=back + +=back + +=item (Blue) Certificates not installed -- trying insecure auth + +One of the certificate file, key file or +certificate authority file could not be found for a client attempting +SSL connection intiation. COnnection will be attemptied in in-secure mode. +(this would be a system with an up to date lond that has not gotten a +certificate from us). + +=item (Green) Successful local authentication + +A local connection successfully negotiated the encryption key. +In this case the IDEA key is in a file (that is hopefully well protected). + +=item (Green) Successful ssl authentication with + +The client ( is the peer's name in hosts.tab), has successfully +negotiated an SSL connection with this child process. + +=item (Green) Successful insecure authentication with + + +The client has successfully negotiated an insecure connection withthe child process. + +=item (Yellow) Attempted insecure connection disallowed + +The client attempted and failed to successfully negotiate a successful insecure +connection. This can happen either because the variable londAllowInsecure is false +or undefined, or becuse the child did not successfully echo back the challenge +string. + + +=back + + +=cut