contrib/check_smart.pl

   1 #!/usr/bin/perl -w
   2
   3 # chec_smart
   4 # Check S.M.A.R.T. enabled disks status.
   5 #
   6 # This uses smartmontools to check for disk status.
   7 # This is NOT compatible with smartsuite
   8 # Please use smartctl --smart=on --offlineauto=on --saveauto=on /dev/something
   9 # or similar to enable automatic data collection, and RTFM about it.
  10 #
  11 # this uses sudo to access the smartctl program, so please add a line in sudoers
  12 # nagios  computername = NOPASSWD:/usr/sbin/smartctl
  13 #
  14 # CopyLeft Roy Sigurd Karlsbakk <roy@karlsbakk.net>
  15 # Developed under Debian/SID
  16 # No warranties what so ever. If this toasts your PC, or your wife
  17 # runs away with your girlfriend, or even me, don't blame me.
  18 #
  19 # Licenced under GPL
  20 #
  21
  22 use strict;
  23 use Getopt::Long;
  24
  25 my (
  26         $s, $i, $out, $retcode, $errtxt, $exitcode,
  27         $device, $text_mode, $exitcode_mode, $help, $verbose, $type,
  28 );
  29 my $smartctl = "sudo /usr/sbin/smartctl";
  30 my $e_commandline = 0;
  31 my $e_devopen = 0;
  32 my $e_chksum = 0;
  33 my $e_disk_failing = 0;
  34 my $e_prefail = 0;
  35 my $e_mayprefail = 0;
  36 my $e_errlog = 0;
  37 my $e_selftestlog = 0;
  38
  39 sub end {
  40         $s = shift;
  41         $i = shift;
  42         if ($i == 0) {
  43                 $s = "OK: $s";
  44         } elsif ($i == 1) {
  45                 $s = "WARNNG: $s";
  46         } elsif ($i == 2) {
  47                 $s = "CRITICAL: $s";
  48         } elsif ($i == 3) {
  49                 $s = "UNKNOWN: $s";
  50         } else {
  51                 $s = "OUT OF RANGE: $s";
  52         }
  53         print "$s\n";
  54         exit($i);
  55 }
  56
  57 sub syntax {
  58         $s = shift or $s = 'Unknown';
  59         printf STDERR ("Error: $s\n") unless ($help);
  60         printf STDERR ("Syntax: %s (-t|-e) -d <device> [-vh]\n", $0);
  61         printf STDERR ("  --text-mode -t        check by parsing smartctl's output\n");
  62         printf STDERR ("  --exitcode-mode -e    check smartctl's exitcode (only works on IDE)\n");
  63         printf STDERR ("  --device -d           disk device to check\n");
  64         printf STDERR ("  --type -T             drive type. See the -d flag in the smartctl manual\n");
  65         printf STDERR ("  --verbose -v          verbose\n");
  66         printf STDERR ("  --help -h             this help\n");
  67         exit(0) if $help;
  68         exit(3);
  69 }
  70
  71 Getopt::Long::Configure('bundling');
  72 GetOptions(
  73         "d=s" => \$device, "device=s" => \$device,
  74         "T=s" => \$type, "type=s" => \$type,
  75         "t" => \$text_mode, "text-mode" => \$text_mode,
  76         "e" => \$exitcode_mode, "exitcode-mode" => \$exitcode_mode,
  77         "h" => \$help, "help" => \$help,
  78         "v" => \$verbose, "verbose" => \$verbose
  79 ) || syntax("RTFM!");
  80
  81 syntax if ($help);
  82 syntax("Need device to check") unless ($device);
  83 syntax("Conflicting modes") if ($text_mode && $exitcode_mode);
  84 syntax("Need test mode") unless ($text_mode || $exitcode_mode);
  85 syntax("Exitcode mode only works on ATA drives") if ($exitcode_mode && ! ($device =~ /\/dev\/hd./));
  86
  87 if ($type) {
  88         $type =~ s/[\r\n]*?//g;
  89         print "type: '$type'\n" if ($verbose);
  90         syntax("Valid --type entries include ata, scsi and 3ware,n")
  91                 unless (($type =~ /^ata$/) || ($type =~ /^scsi$/) || ($type =~ /^3ware,\d+$/));
  92 }
  93 if (defined($type)) {
  94         $type = "--device=$type";
  95 } else {
  96         $type = "";
  97 }
  98
  99 if ($text_mode) {
 100         print "running $smartctl $type -H $device" if ($verbose);
 101         unless (open SMARTCTL,"$smartctl $type -H $device|") {
 102                 print STDERR "Can't execute $smartctl: $!\n";
 103                 exit(3);
 104         }
 105         while (<SMARTCTL>) {
 106                 last if (/=== START OF READ SMART DATA SECTION ===/);
 107         }
 108         $out = <SMARTCTL>;
 109         print $out;
 110         exit(0) if ($out =~ /PASSED/);
 111         exit(2) if ($out =~ /SAVE ALL DATA/ || $out =~ /FAILED/);
 112         exit(3);
 113 } elsif ($exitcode_mode) {
 114         print "Running $smartctl $type -q silent $device\n" if ($verbose);
 115         system("$smartctl $type -q silent $device");
 116         $retcode = $?;
 117         $e_commandline = 1 if ($retcode & 0x0100);
 118         $e_devopen = 1 if ($retcode & 0x0200);
 119         $e_chksum = 1 if ($retcode & 0x0400);
 120         $e_disk_failing = 1 if ($retcode & 0x0800);
 121         $e_prefail = 1 if ($retcode & 0x1000);
 122         $e_mayprefail = 1 if ($retcode & 0x2000);
 123         $e_errlog = 1 if ($retcode & 0x4000);
 124         $e_selftestlog = 1 if ($retcode & 0x8000);
 125
 126         print "$e_commandline $e_devopen $e_chksum $e_disk_failing $e_prefail $e_mayprefail $e_errlog $e_selftestlog\n"
 127                 if ($verbose);
 128
 129         $exitcode = 0;
 130         $errtxt = "";
 131         if ($exitcode) {
 132                 if ($e_commandline) {
 133                         $errtxt .= "Commandline didn't parse, ";
 134                         $exitcode = 3 if ($exitcode == 0);
 135                 }
 136                 if ($e_devopen) {
 137                         $errtxt .= "Device could not be opened, ";
 138                         $exitcode = 3 if ($exitcode == 0);
 139                 }
 140                 if ($e_chksum) {
 141                         $errtxt .= "Checksum failure somewhere, ";
 142                         $exitcode = 1 if ($exitcode != 2);
 143                 }
 144                 if ($e_disk_failing) {
 145                         $errtxt .= "Disk is failing!, ";
 146                         $exitcode = 2;
 147                 }
 148                 if ($e_prefail) {
 149                         $errtxt .= "Disk is in prefail, ";
 150                         $exitcode = 1 if ($exitcode != 2);
 151                 }
 152                 if ($e_mayprefail) {
 153                         $errtxt .= "Disk is close to prefail. Please check manually, ";
 154                         $exitcode = 1 if ($exitcode != 2);
 155                 }
 156                 if ($e_errlog) {
 157                         $errtxt .= "The device error log contains records of errors, ";
 158                         $exitcode = 1 if ($exitcode != 2);
 159                 }
 160                 if ($e_selftestlog) {
 161                         $errtxt .= "The device self-test log contains records of errors, ";
 162                         $exitcode = 1 if ($exitcode != 2);
 163                 }
 164                 if ($exitcode == 1) {
 165                         $errtxt = "WARNNG: $errtxt";
 166                 } elsif ($exitcode == 2) {
 167                         $errtxt = "CRITICAL: $errtxt";
 168                 } else {
 169                         $errtxt = "UNKNOWN: $errtxt";
 170                 }
 171         } else {
 172                 $errtxt = "OK";
 173         }
 174         print "$errtxt\n";
 175         exit($exitcode);
 176 } else {
 177         print STDERR "Something's strange is going on :~|\n";
 178         exit(3);
 179 }
 180 # vim:ts=4:sw=4:cindent