#! /usr/bin/perl -w

#==========================================================================
# This script parses files generated by slurpd (OpenLDAP replication server) 
# displays if errors occured and if master and slaves are up2date
# Script designed for Nagios/NRPE ( http://www.nagios.org )
#
# Usage :
# You have to set the replica's name and port (as written in slapd.conf)
# ./slurpd_status -w warning_level -c critical_level [-h hostname] [-p hostport] [-v]
# The level must be given with 3 integers, for exemple 100,5,30
# The first integer are the entries in transition (not picked up by slurpd)
# The second integer are the rejected entries (not accepted by the slave)
# The third integer are the waiting entries (picked up by slurpd but not sent to the slave)
# If only one integer is given, the level will be the max of all entries (in transition + rejected + waiting)
# If no port is declared, 0 will be used
# If no host is declared, localhost will be used
# Option -v displays a lot of message, use for debug only
# 
# 2004 (c) Clement OUDOT <clement.oudot@linagora.com>
# 
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# as published by the Free Software Foundation; either version 2
# of the License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# GPL Licence : http://www.gnu.org/licenses/gpl.txt
#
# More contributions on http://www.linagora.org
#==========================================================================
 
#==========================================================================
# Modules
#==========================================================================
use strict ;
use Getopt::Std ;
use List::Util qw(max) ;

#==========================================================================
# Internal Parameters
#==========================================================================
# Directory where slurpd stores his files
my $slurpd_tempdir = "/var/openldap-slurp/replica" ;
# Replog file beetween slapd and slurpd, as written in slapd.conf
my $slapd_replog_file = "/var/replog" ;
# Nagios Codes
my %code = ( 'Ok' => 0, 'Warning' => 1, 'Critical' => 2, 'Unknown' => 3) ;

#==========================================================================
# Command Line Parameters
#==========================================================================
my %opts ;
getopt( 'wchpv' , \%opts ) ;
my $debug = 1 if ( exists ( $opts{"v"} ) ) ;
&usage unless ( exists ( $opts{"w"} ) ) ;
&usage unless ( exists ( $opts{"c"} ) ) ;
$opts{"h"} = "localhost" unless ( exists ( $opts{"h"} ) ) ;
$opts{"p"} = 0 unless ( exists ( $opts{"p"} ) ) ;

my $slave_name = $opts{"h"} ;
my $slave_port = $opts{"p"} ;
my $warning_level = $opts{"w"} ;
my $critical_level = $opts{"c"} ;

# See if warning_level and critical_level are given with 3 integers or just one

my @warning_level ;

if ( $warning_level =~ /^(\d*)$/ ) {
	# This is just one integer
	$warning_level = $1 ;
	if ( $debug ) {
		print "Warning level set to $warning_level entries\n" ;
	}
}
elsif ( $warning_level =~ /^(\d*),(\d*),(\d*)$/ ) {
	# Three integers, comma separated
	@warning_level = ( $1, $2 , $3 ) ;
	if ( $debug ) {
		print "Warning level set to $1 entries in transition, $2 rejected entries and $3 waiting entries\n" ;
	}
	$warning_level = 0 ;
}
else {
	# Bad usage
	&usage ;
}

my @critical_level ;

if ( $critical_level =~ /^(\d*)$/ ) {
	# This is just one integer
	$critical_level = $1 ;
	if ( $debug ) {
		print "Critical level set to $critical_level entries\n" ;
	}
}
elsif ( $critical_level =~ /^(\d*),(\d*),(\d*)$/ ) {
	# Three integers, comma separated
	@critical_level = ( $1, $2 , $3 ) ;
	if ( $debug ) {
		print "Critical level set to $1 entries in transition, $2 rejected entries and $3 waiting entries\n" ;
	}
	$critical_level = 0 ;
}
else {
	# Bad usage
	&usage ;
}

#==========================================================================
# Usage function
#==========================================================================
sub usage {
	if ( $debug ) {
		print "Usage: $0 -w warning_level -c critical_level [-h hostname] [-p hostport] [-v]\n\n" ;
		print "The level (warning or critical) can be given as 3 comma separated integers, for example 100,5,30\n" ;
		print "The first integer is the number of entries in transition (not picked up by slurpd)\n" ;
		print "The second integer is the number of rejected entries (not accepted by the slave)\n" ;
		print "The third integer is the number of waiting entries (picked up by slurpd but not sent to the slave)\n" ;
		print "If only one integer is given, the level will be the max (in transition, rejected and waiting)\n\n" ;
		print "hostname and hostport must fit what is written in slapd.conf\n" ;
		print "If no hostport is declared, 0 will be used\n" ;
		print "If no hostname is declared, localhost will be used\n\n" ;
		print "Option -v displays a lot of message, use for debug only !\n" ;
	}
	else {
		print "Bad script usage. Use -v for more inforations.\n" ;
	}
	exit $code{'Unknown'} ;
}

#==========================================================================
# Slurpd Files
#==========================================================================
my $slurpd_replog_file = $slurpd_tempdir."/slurpd.replog" ;
my $status_file = $slurpd_tempdir."/slurpd.status" ;
my $rej_file = $slurpd_tempdir."/".$slave_name.":".$slave_port.".rej" ;

#==========================================================================
# Welcome
#==========================================================================
if ( $debug ) {
	print "Determining replication status for $slave_name at port $slave_port\n\n" ;
}

#==========================================================================
# Preliminary Tests
#==========================================================================
# If one of the 4 files exists but we can't read it, it's a problem
foreach ( $slapd_replog_file , $slurpd_replog_file , $status_file , $rej_file ) {

	if ( ( -e $_ ) && ( ! -r $_ ) ) {
	
		if ( $debug ) {
			print "File $_ exists but is not readable for the user running this script\n" ;
		}
		else {
			print "Unable to read $_\n" ;
		}
		exit $code{'Unknown'} ;
	}
}

# If slapd replog file is not present, it can meen that :
# - the internal script config is bad
# - the config of OpenLDAP is bad
# - no modifications were done to slapd, so replog file was not created
# - slapd is not running well...
# => an Unknown status is returned

unless ( -e $slapd_replog_file ) {
	if ( $debug ) {
		print "File $slapd_replog_file does not exist\n" ;
	}
	else {
		print "No slapd replog file\n" ;
	}
	exit $code{'Unknown'} ;
}

# If slurpd status file is not present, it can meen that :
# - the internal script config is bad
# - slurpd was launched with an alternate tempdir (option -t)
# - slurpd was never launched
# - slurpd is not running well...
# => a Critical status is returned

unless ( -e $status_file ) {
	if ( $debug ) {
		print "File $status_file does not exist\n" ;
	}
	else {
		print "No slurpd status file\n" ;
	}
	exit $code{'Critical'} ;
}

# If slurpd replog file is not present, it can meen that :
# - the internal script config is bad
# - slurpd is not running well...
# => a Critical status is returned

unless ( -e $slurpd_replog_file ) {
	if ( $debug ) {
		print "File $slurpd_replog_file does not exists\n" ;
	}
	else {
		print "No slurpd replog file\n" ;
	}
	exit $code{'Critical'} ;	
}

# If slurpd status file is empty, it can meen that :
# - it was manually cleared
# - slurpd is not running well...
# => a Critical status is returned

if ( -z $status_file ) {
	if ( $debug ) {
		print "File $status_file is empty\n" ;
	}
	else {
		print "No data in slurpd status file\n" ;
	}
	exit $code{'Critical'} ;	
}

# If the slave is not present in status file, it can meen that :
# - the command line parameters are bad
# - the config of OpenLDAP is bad
# - the config of OpenLDAP is good but slurpd was not restarted
# => a Critical status is returned
# If it is present :
# Get the time and the range in status file in order to count how many
# entries are waiting in slurpd replog file

# Variables for further comparaison
my ( $status_time, $status_range ) ;

# Open status file for reading

if ( open ( STATUS , $status_file ) ) {
	# Parse the file
	# lines in status file are like :
	# servername:serverport:time:range
	while ( <STATUS> ) {
		if ( /^$slave_name:$slave_port:(\d*):(\d*)$/ ) {
			( $status_time , $status_range ) = ( $1 , $2 ) ;
		}
	}
	# Close file
	close ( STATUS ) ;
	# Test if something was found
	if ( defined $status_time ) {
		# Just print a message for debug
		if ( $debug ) {
			print "$slave_name (port $slave_port) is listed in $status_file\n\n" ;
			print "Time found in status file : $status_time\n\n" ;
			print "Range found in status file : $status_range\n\n" ;
		}
	}
	else {
		if ( $debug ) {
			print "$slave_name (port $slave_port) is not listed in $status_file\n";
			print "Maybe slurpd needs to be restarted to read OpenLDAP configuration\n" ;
		}
		else {
			print "No status for $slave_name:$slave_port\n" ; 
		}
		exit $code{'Critical'} ;
	}
}

# Error while opening the file
# => an Unknown status is returned

else {
	if ( $debug ) {
		print "Unable to open $status_file for reading\n" ;
	}
	else {
		print "Error while reading status file\n" ;
	}
	exit $code{'Unknown'} ;
}

#==========================================================================
# Waiting entries in Slapd Replog File
#==========================================================================
# If there are waiting entries in slapd replog file, it meens that slurpd
# has not yet get these entries ou that slurpd is disturbed.

my $waiting_entries = 0 ;

if ( open ( SLAPDREPLOG , $slapd_replog_file ) ) {
	# Count entries that must be added to slurpd replog file
	while ( <SLAPDREPLOG> ) {
		if ( /^replica:/ ) {
			if ( $slave_port ) {
				# The port is <> 0, so it is written after the name
				$waiting_entries++ if /^replica: $slave_name:$slave_port$/ ;
			}
			else {
				# The port is = 0, so only the name is written (or the port after the name)
				$waiting_entries++ if /^replica: $slave_name(:$slave_port)?$/ ;
			}
			next ;
		}
	}
	# Close file
	close ( SLAPDREPLOG ) ;
	# Print debug message
	if ( $debug ) {
		print "$waiting_entries entries in slapd replog file\n\n" ;
	}
}

else {
	if ( $debug ) {
		print "Unable to read $slapd_replog_file\n" ;
	}
}

#==========================================================================
# Rejection File
#==========================================================================
# Rejection file exists only if errors occured.
# Keep time and range for comparaison with slurpd replog file

my $rejected_entries = 0 ;

if ( -e $rej_file ) {
	# The rejection file exists
	if ( open ( REJ, $rej_file ) ) {
		# Count entries in error
		while ( <REJ> ) {
			$rejected_entries++ if ( /^time: / ) ;
			if ( ( $debug ) &&  (/^dn: (.*)$/ ) ) {
				print "Modification on $1 was rejected\n" ;
			}
		}
		# Close file
		close ( REJ ) ;
		# Print debug message
		if ( $debug ) {
			print "$rejected_entries entries rejected\n\n" ;
		}
	}
	
	else {
		# Error while opening the file
		if ( $debug ) {
			print "Unable to read $rej_file\n" ;
		}
	}
}

else {
	# No rejection file
	if ( $debug ) {
		print "There is no rejection file\n\n" ;
	}
}

#==========================================================================
# Slurpd Replog File
#==========================================================================
# Find all entries younger than time and range found in status file
# This entries must concern our slave
# => $slave_is_concerned is set to 1 if replica: $slave_name:$slave_port is found
# => Go to the next lines
# => Get the time and set $slave_is_concerned to 0, and continue the loop
my $slave_is_concerned = 0 ;
my $younger_entries = 0 ;

if ( open ( REPLOG , $slurpd_replog_file ) ) {
	# Parse the file
	while ( <REPLOG> ) {
		# If it is a "replica:" line
		if ( /^replica:/ ) {
			if ( $slave_port ) {
				# The port is <> 0, so it is written after the name
				$slave_is_concerned = 1 if /^replica: $slave_name:$slave_port$/ ;
			}
			else {
				# The port is = 0, so only the name is written (or the port after the name)
				$slave_is_concerned = 1 if /^replica: $slave_name(:$slave_port)?$/ ;
			}
			# Go to the next line
			next ;
		}
		
		if ( ( $slave_is_concerned ) && ( /^time: (\d*)\.?(\d*)?$/ ) ) {
			# Reset $slave_is_concerned
			$slave_is_concerned = 0 ;
			# Next if replog entry is older than status time
			next if ( $1 < $status_time ) ;
			# Count the entry if younger and go to next line
			if ( $1 > $status_time ) {
				$younger_entries++ ;
				if ( $debug ) {
					print "Entry with time $1 is young\n"
				}
				next ;
			}
			# Check the range if times are equal
			if ( $1 == $status_time ) {
				my $replog_range = $2 ? $2 : 0 ;
				# If replog range > status range, entry is younger
				if ( $replog_range > $status_range ) {
					$younger_entries++ ;
					if ( $debug ) {
						print "Entry with time $1 and range $replog_range is young\n" ;
					}
					next ;
				}
				# Else it is older, nothing to do
			}
		}
	}
	# Close file
	close ( REPLOG ) ;
	if ( $debug ) {
		print "$younger_entries entries waiting in slurpd replog file\n\n" ;
	}
}

# Error while opening the file
else {
	if ( $debug ) {
		print "Unable to read $slurpd_replog_file\n" ;
	}
}

#==========================================================================
# Results and exit with Nagios codes
#==========================================================================
# All errors
my $all_errors = $waiting_entries + $rejected_entries + $younger_entries ;
my $max_errors = max ( $waiting_entries , $rejected_entries , $younger_entries ) ;

if ( $debug ) {
	print "A total of $all_errors errors was found :\n" ;
	print "$waiting_entries entries are in transition\n" ;
	print "$rejected_entries entries are rejected\n" ;
	print "$younger_entries entries are waiting\n" ;
}

# Message to sent
print "$waiting_entries entries in transition, $rejected_entries entries rejected, $younger_entries entries waiting\n" ;

# exit code

if ( ( $critical_level && $max_errors >= $critical_level  ) || ( @critical_level && ( $critical_level[0] <= $waiting_entries || $critical_level[1] <= $rejected_entries || $critical_level[2] <= $younger_entries ) ) ) {
	exit $code{'Critical'} ;
}

if ( ( $warning_level && $max_errors >= $warning_level  ) || ( @warning_level && ( $warning_level[0] <= $waiting_entries || $warning_level[1] <= $rejected_entries || $warning_level[2] <= $younger_entries ) ) ) {
	exit $code{'Warning'} ;
}

exit $code{'Ok'} ;