From 871d2987f9c877a776cd8fff9513c37d601ca2d1 Mon Sep 17 00:00:00 2001
From: hadaq <hadaq>
Date: Wed, 12 May 2010 15:36:15 +0000
Subject: [PATCH] New script for hard disks checks. Sergey.

---
 plugins/check_hdisks.pl | 354 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 354 insertions(+)
 create mode 100755 plugins/check_hdisks.pl

diff --git a/plugins/check_hdisks.pl b/plugins/check_hdisks.pl
new file mode 100755
index 0000000..71140a5
--- /dev/null
+++ b/plugins/check_hdisks.pl
@@ -0,0 +1,354 @@
+#!/usr/bin/perl -w
+
+use strict;
+
+######################################################################
+#                                                                    #
+# Nagios plugin to check hard disks and raid arrays.                 #
+# To be able to run smartctl under normal user account               #
+# you should open permissions under root account:                    #                                              
+# chmod u+s /usr/sbin/smartctl                                       #
+#                                                                    #
+# Long self-tests should be executed in crontab:                     #
+# 0 23 * * 0 /usr/sbin/smartctl -t long /dev/sda > /dev/null         #
+#                                                                    #
+# Sergey Yurevich                                                    #
+#                                                                    #
+######################################################################
+
+use threads;
+use threads::shared;
+use IO::Socket;
+use IO::Select;
+
+use Data::Dumper;
+use FileHandle;
+use Getopt::Long;
+
+our $server_port = '50999';
+our $protocol    = 'tcp';
+
+my $opt_help   = 0;
+my $opt_verb   = 0;
+my $opt_daemon = 0;
+my $opt_smart  = 1;     # Check disks with smartctl tool
+my $opt_raid   = 1;     # Check RAID Arrays
+my $opt_sleep  = 3600;  # 1 hour
+
+GetOptions ('p|port=i'      => \$server_port,
+	    'd|daemon'      => \$opt_daemon,
+	    's|smart=i'     => \$opt_smart,
+	    'r|raid=i'      => \$opt_raid,
+	    'v|verb'        => \$opt_verb,
+	    'l|sleep=i'     => \$opt_sleep,
+	    'h|help'        => \$opt_help);
+
+if($opt_help){
+    &help();
+    exit(0);
+}
+
+my $status : shared = "OK";
+
+my @disks;                        # array with the hard disks
+my $disks_aref = \@disks;
+my %status_hash;                  # hash with status info for each hard disk
+my $status_href = \%status_hash;
+my @md_list;                      # array with RAIDs problematic info 
+
+#my $MDSTAT = "/home/icinga/nagios/plugins/mdstat";
+my $MDSTAT   = "/proc/mdstat";
+my $SMARTCTL = "/usr/sbin/smartctl";
+
+#- Daemonize
+if($opt_daemon){
+    open(STDIN,  '>/dev/null');
+    open(STDOUT, '>/dev/null');
+    open(STDERR, '>/dev/null');
+}
+
+threads->new(\&statusServer);
+
+&main();
+
+exit(0);
+
+####################### END OF MAIN ######################
+
+sub help()
+{
+    print "\n";
+    print << 'EOF';
+check_hdisks.pl
+
+   This script checks RAID status in $MDSTAT. It also checks disks
+   using smartctl tool. Smart checks must be of course supported 
+   on the given PC. Long smart tests should run as cron jobs:
+   0 23 * * 0 /usr/sbin/smartctl -t long /dev/sda > /dev/null  
+
+Usage:
+
+   Command line:  check_hdisks.pl 
+      [h|help]                : Show this help.
+      [p|port <port>]         : Port for the Status Server (default: 50999).
+      [d|daemon]              : Run as a daemon.
+      [s|smart <1|0>]         : Enable|disable smart checks (default: 1).
+      [r|raid  <1|0>]         : Enable|disable RAID checks (default: 1).
+      [v|verb]                : More verbouse.
+      [l|sleep <time>]        : Sleep time in seconds between checks (default: 3600).
+
+Examples:
+
+   Run as a daemon and do only RAID checks:
+      check_hdisks.pl -d -s 0 &
+
+EOF
+}
+
+sub main()
+{
+    my $MDSTAT = "/proc/mdstat";
+
+    #- Get all hard disks
+    &getHardDisks($disks_aref);
+
+    while(1){
+
+	$status = "OK";
+
+	&checkRaid() if($opt_raid);
+	&checkSmart() if($opt_smart);
+	&getStatus();
+	&init2zero();
+
+	sleep($opt_sleep);  # 1 hour
+    }
+}
+
+sub getStatus()
+{
+    
+    my @crit_list = ();
+    my @warn_list = ();
+    my @ok_list   = ();
+    
+    #- Check info collected by smartctl
+    foreach my $disk ( %$status_href ){
+	if( defined $status_href->{$disk}->{'FAIL'} ){
+	    my $info = $disk . " " . $status_href->{$disk}->{'FAIL'};
+	    push(@crit_list, $info);
+	}
+	
+	if( defined $status_href->{$disk}->{'WARN'} ){
+	    my $info = $disk . " " . $status_href->{$disk}->{'WARN'};
+	    push(@warn_list, $info);
+	}
+	
+	if( defined $status_href->{$disk}->{'SELF'} ){
+	    my $info = $disk . " " . $status_href->{$disk}->{'SELF'};
+	    push(@warn_list, $info);
+	}
+	
+	if( defined $status_href->{$disk}->{'TEMP'} ){
+	    my $temperature = "temp: " . $disk . " " . $status_href->{$disk}->{'TEMP'};
+	    push(@ok_list, $temperature);
+	}
+    }
+    
+    #- Look at messages from RAID array check
+    my $msg = "";
+    if( @md_list ){
+	foreach my $entry (@md_list){
+	    $msg = $msg . " " . $entry;
+	}
+	
+	$msg = "Problem with " . $msg;
+	push(@crit_list, $msg);
+    }
+    
+    #- Build status message
+    if( @crit_list ){
+	my $msg = "";
+	foreach my $crit (@crit_list){
+	    $msg = $msg . " " . $crit . ",";
+	}
+	
+	$status = "CRITICAL - " . $msg;
+    }
+    elsif( @warn_list ){
+	my $msg = "";
+	foreach my $warn (@warn_list){
+	    $msg = $msg . " " . $warn . ",";
+	}
+	
+	$status = "WARNING - " . $msg;
+    }
+    else{
+	my $msg = "";
+	if( @ok_list ){
+	    foreach my $ok (@ok_list){
+		$msg = $msg . " " . $ok . ",";
+	    }
+	}
+	
+	$status = "OK - " . $msg;
+    }
+
+
+    print "status: $status\n" if($opt_verb);
+}
+
+sub init2zero()
+{
+    #- Reinitialize
+    %status_hash = ();
+    @md_list = ();
+}
+
+sub checkSmart()
+{
+    foreach my $disk (@disks){
+	&checkSmartShort($disk);
+	&checkSmartTemp($disk);
+    }
+}
+
+sub checkSmartShort()
+{
+    my ($disk) = @_;
+
+    my $disk_status  = `$SMARTCTL -H $disk`;
+    
+    if( $disk_status =~ /\n(.*?test result: PASSED\n)/i ||   
+	$disk_status =~ /\n(.*?Sense: Ok!\n)/i ){
+	$status_href->{$disk}->{'OK'} = $1;
+    }
+    
+    if( $disk_status =~ /\n(.*?OLD[^_age][^\n]*)/i ){
+	$status_href->{$disk}->{'WARN'} = $1;
+    }
+    
+    if( $disk_status =~ /\n(.*?[^WHEN_]FAIL[^ED][^\n]*)/i ){
+	$status_href->{$disk}->{'FAIL'} = $1;
+    }
+
+    #- Check results of the long test
+    $disk_status  = `$SMARTCTL -l selftest $disk`;
+
+    if( $disk_status =~ /\n*(No\s+self-tests\s+have\s+been\s+logged)*/ ){
+	$status_href->{$disk}->{'SELF'} = $1;
+    }
+    elsif( ! ($disk_status =~ /\n*(Completed\s+without\s+error)*/) ){
+	$status_href->{$disk}->{'SELF'} = "Self-test failed";
+    }
+}
+
+sub checkSmartTemp()
+{
+    my ($disk) = @_;
+
+    my @disk_status  = `$SMARTCTL -A $disk`;
+
+    foreach my $line ( @disk_status ){
+	if( $line =~ "194 Temperature_Celsius" ){                                       
+	    my @words = split(/ +/, $line);     
+	    my $temperature = $words[9];
+	    chomp($temperature);
+	    $status_href->{$disk}->{'TEMP'} = $temperature;
+	}
+    }
+}
+
+sub getHardDisks()
+{
+    my ($disks_aref) = @_;
+
+    my $disk;
+
+    foreach my $letter ('a'..'z'){
+	$disk = "/dev/sd" . $letter;
+	push( @$disks_aref, $disk ) if( -e $disk );
+	$disk = "/dev/hd" . $letter;
+	push( @$disks_aref, $disk ) if( -e $disk );
+    }
+}
+
+sub checkRaid()
+{
+    my ($md, $active);
+
+    open(FH, $MDSTAT);
+
+    #- Now check the mdstat file...
+    while (<FH>) {               
+	my $line= $_;  
+	if( $line =~ /(md\d{1})\s+:\s+(\w+)/ ){
+	    #- Found the raid array
+	    $md     = $1;
+	    $active = $2;
+	}
+	elsif( defined $md ){
+	    #- Check if all disks are UP for the current raid array
+	    if( $line =~ /\[(\d+)\/(\d+)\]\s+\[(\w+)\]/ ){
+		my $disks_all    = $1;
+		my $disks_up     = $2;
+		my $disks_status = $3;
+
+		if( $disks_all != $disks_up || $disks_status =~ /_/ ){
+		    push(@md_list, $md);
+		}
+
+		$md = "";
+	    }
+	}
+    }
+
+    close (FH);
+
+    if( @md_list ){
+	my $msg = "";
+	foreach my $entry (@md_list){
+	    $msg = $msg . " " . $entry;
+	}
+	
+	$status = "CRITICAL - Problem with $msg";
+    }
+}
+
+sub statusServer{ my $server_socket;
+    my $client_socket;
+    my $selector;
+
+    unless (defined( $server_socket =
+                     IO::Socket::INET->new( LocalPort => $server_port,
+                                            Proto     => 'tcp',
+                                            Listen    => SOMAXCONN ) ))
+    {
+        print "ERROR: Cannot start status server!\n";
+    }
+
+    $selector = new IO::Select( $server_socket );
+
+    while(1) {
+
+        # wait 5 seconds for connections
+        while (my @file_handles = $selector->can_read( 5 )) {
+
+            foreach my $file_handle (@file_handles) {
+
+                if($file_handle == $server_socket) {
+
+                    # create a new socket for this transaction
+                    unless (defined( $client_socket = $server_socket->accept() ))
+                    {
+                        print "ERROR: Cannot open socket to send status!\n";
+                    }
+
+                    print $client_socket $status;
+
+                    close( $client_socket );
+                }
+            }
+        }
+    }
+}
-- 
2.43.0