#!/usr/bin/perl -w
#
# NAGIOS Plugin: check_3ware_disk
# Check the status of the 3ware raid controller disks with tw_cli
# Author: Marius Hein <mhein@netways.de>
# NETWAYS GmbH, www.netways.de, info@netways.de
#
# Additions by Ianaré Sévi <ianare@gmail.com> :
# SSH addon-on, newer style report, more keywords

# Mauled over by John Jore, j o h n @ j o r e . n o
# Removed stuff I dont want or need, added Disk Temp and ReAllocated Sectors checks.
# Most severe error "wins" and is reported
#
# One of the problems with the 3ware controller is the disks can, and do, move around from one port to another. How do you support command line parameters
# when this happens? My Seagate disks run 4-6 degress C cooler than the WD disks. Using a command line parameter would make sense, but not possible.
# Maybe creating some kind of database/lookup with different models and "acceptable" temperature ranges is the way forward?


use strict;
use File::Basename;
use Getopt::Long;
use vars qw(
	$opt_disk
	$opt_control
	$opt_help
	$opt_usage
	$progname
	@state_ok
	@state_warning
	@state_critical
	$temp_ok
	$temp_warning
	$ra_ok
	$ra_warning
	$return
	$command
	%conf
	$tmp1
	$tmp2
	$state_stat
	$state_temp
	$state_ra
	$state_ncq
	$desc_stat
	$desc_temp
	$desc_ra
	$desc_ncq
        $desc_model
	$exit_stat
	$exit_temp
	$exit_ra
	$exit_ncq
	$state_out
	$exit_out
	$stat
	);

sub print_help();
sub print_usage();

$progname = basename($0);

@state_ok=("OK","VERIFYING");
@state_warning=("DEGRADED","INITIALIZING","MIGRATING");
@state_critical=("OFFLINE","DEVICE-ERROR");

$temp_ok       = 51;	#Upto this is ok
$temp_warning  = 55;    #Between ok and this is warning
#$temp_critical= xx;	#Anyting else is critical


$ra_ok        = 39;	#Upto 20 is ok
$ra_warning   = 40;	#Upto 30 is warning
#$ra_critical = xx;	#Anything else is critical

$conf{'bin'} = "tw_cli";
$conf{'fullbin'} = "/usr/sbin/$conf{'bin'}";
$conf{'cmd_part_status'} = "show status";
$conf{'cmd_part_temperature'} = "show temperature";
$conf{'cmd_part_rasect'} = "show rasect";
$conf{'cmd_part_ncq'} = "show ncq";
$conf{'cmd_part_model'} = "show model";


Getopt::Long::Configure('bundling');
GetOptions
	(
	"controller=s"	=>	\$opt_control,
	"C=s"		=>	\$opt_control,
	
	"disk=s"	=>	\$opt_disk,
	"P=s"		=>	\$opt_disk,

	"h"		=>	\$opt_help,
	"help"		=>	\$opt_help,
	"usage"		=>	\$opt_usage
	) || die "try '$progname --help' for informations.\n";

#Default controller and disk (My 3ware starts on P8 with the latest firmware. Go figure)
if (!$opt_control) { $opt_control=0; }
if (!$opt_disk) { $opt_disk=8; }

sub print_help() {
  print "\n";
  print "check 3ware >>HELP<<\n";
  print "\n";
  print "\t --help, -h\t\t\t help screen.\n";
  print "\t --usage\t\t\t little usage\n";
  print "\n";
  print "\t --controller, -C\t\t Controller ID\n";
  print "\t --disk, -P\t\t\t Disk ID\n";
}

sub print_usage() {
  print "\n";
  print "check 3ware >>USAGE<<\n";
  print "\n";
  print "\t$progname -C 0 -P 8 (checks Controller 0 Disk 8 for status)\n";
  print "\t$progname --help (Displays the Helpmessage)\n";
  print "\n";
}

if ($opt_help) {
  print_help();
  exit;
}

if ($opt_usage) {
  print_usage();
  exit;
}

if ($opt_control >= 0 && $opt_disk >= 0)
{
  #Disk Status
  $command = "sudo ".$conf{'fullbin'}." /c$opt_control/p$opt_disk ".$conf{'cmd_part_status'}; 
  #print "$command\n";
  $return = qx ( $command );
  #print $return;
 
  ($tmp1,$tmp2) = split(/= /,$return);
  $tmp2 =~ s/\s+$//;
  $stat = $tmp2;

  $desc_stat = "Disk /c$opt_control/p$opt_disk is $stat";
  foreach (@state_ok) {
    if ($stat eq $_) {
      $state_stat = "OK";
      $exit_stat = 0;
    }
  }
  
  if (!$state_stat) {
    foreach (@state_warning) {
      if ($stat eq $_) {
        $state_stat = "WARNING";
        $exit_stat = 1;
      }
    }
  }

  if (!$state_stat) {
    foreach (@state_critical) {
      if ($stat eq $_) {
        $state_stat = "CRITICAL";
        $exit_stat = 2;
      }
    }
  }
                                                 
  if (!$state_stat) {
    $state_stat = "UNKNOWN";
    $exit_stat = 3;
  }
  #print "$progname: $state_stat ($desc_stat)\n";


  #Disk Model
  $command = "sudo ".$conf{'fullbin'}." /c$opt_control/p$opt_disk ".$conf{'cmd_part_model'};
  $return = qx ( $command );
 
  ($tmp1,$tmp2) = split(/= /,$return);
  $tmp2 =~ s/\s+$//;
  $desc_model = $tmp2;
  #print "$desc_model\n";
    

  #NCQ
  $command = "sudo ".$conf{'fullbin'}." /c$opt_control/p$opt_disk ".$conf{'cmd_part_ncq'};
  $return = qx ( $command );
  ($tmp1,$tmp2) = split(/Enabled /,$return);
  ($tmp1,$return) = split(/= /,$tmp2);
  $return =~ s/\s+$//;
  $stat = substr($return,0,3);

  #print "Stat: $stat \n";
  $desc_ncq = "NCQ is $stat";
  
  if ( "$stat" eq "No" ) { 
    $state_ncq = "OK";
    $exit_ncq = 0;
  }
  
  if (!$state_ncq) {
    if ( $desc_model eq "ST3750330NS" ) {
      $state_ncq = "OK";
      $exit_ncq = 0;
    }
    else 
    {
      $state_ncq = "CRITICAL";
      $exit_ncq = 2;
    }
  }

  if (!$state_ncq) {
    $state_ncq = "UNKNOWN";
    $exit_ncq = 3;
  }
  #print "$state_ncq ($desc_ncq)\n";
  
    
  #Disk Temperature  
  $command = "sudo ".$conf{'fullbin'}." /c$opt_control/p$opt_disk ".$conf{'cmd_part_temperature'};
  #print "$command\n";
  $return = qx ( $command );
  #print $return;
 
  ($tmp1,$tmp2) = split(/= /,$return);
  $tmp2 =~ s/\s+$//;
  $stat = substr($tmp2,0,2);
    
  $desc_temp = "Temp: is $stat";
  if ($stat >= 0 and $stat <= $temp_ok) {
    $state_temp = "OK";
    $exit_temp = 0;
  }
   
  if (!$state_temp) {
    if ($stat > $temp_ok and $stat < $temp_warning) {
      $state_temp = "WARNING";
      $exit_temp = 1;
    }
  }

  if (!$state_temp) {
    if ($stat >= $temp_warning) {
      $state_temp = "CRITICAL";
      $exit_temp = 2;
    }
  }

  if (!$state_temp) {
    $state_temp = "UNKNOWN";
    $exit_temp = 3;
  }
  #print "$progname: $state_temp ($desc_temp)\n";


  

  #Reallocated Sectors
  $command = "sudo ".$conf{'fullbin'}." /c$opt_control/p$opt_disk ".$conf{'cmd_part_rasect'};
  #print "$command\n";
  $return = qx ( $command );
  #print $return;

  ($tmp1,$tmp2) = split(/= /,$return);
  $tmp2 =~ s/\s+$//;
  $stat = $tmp2;

  $desc_ra = "RA Sect: is $stat";
  if ($stat >= 0 and $stat <= $ra_ok) {
    $state_ra = "OK";
    $exit_ra = 0;
  }
   
  if (!$state_ra) {
    if ($stat > $ra_ok and $stat < $ra_warning) {
      $state_ra = "WARNING";
      $exit_ra = 1;
    }
   }

  if (!$state_ra) {
    if ($stat >= $ra_warning) {
      $state_ra = "CRITICAL";
      $exit_ra = 2;
    }
  }
                                                 
  if (!$state_ra) {
    $state_ra = "UNKNOWN";
    $exit_ra = 3;
  }
  #print "$progname: $state_ra ($desc_ra)\n";
  
  # print "$exit_stat $exit_temp $exit_ra\n";
    
  #Most severe exit code
  my @exit_status = ($exit_stat, $exit_ncq, $exit_temp, $exit_ra,);
  @exit_status = sort{$a <=> $b} @exit_status;
  $exit_out  = pop (@exit_status);
  #print "largest is $exit_out\n";


  #Which is the most severe exit status?
  if ($exit_out > 0) {
    if ($exit_stat == $exit_out) {
      $state_out = "$state_stat (Status)";
      #print "1\n";
    }

    if ($exit_ncq == $exit_out) {
      $state_out = "$state_ncq (NCQ)";
      #print "2\n";
    }
    
    if ($exit_temp == $exit_out) {
      $state_out = "$state_temp (Temp)";
      #print "3\n";
    }

    if ($exit_ra == $exit_out) {
      $state_out = "$state_ra (RA)";
      #print "4\n";
    }
  }
  else
  {
    $state_out = "OK";
  }
 
  #We have a winner:
  print "$state_out ($desc_stat, $desc_ncq, $desc_temp, $desc_ra)\n";
  exit $exit_out;   
}
else
{
  print_help();
  exit;
}