#!/usr/bin/perl
#
# Check an bmc equiped server
# So far, this means x86 models
# also works over the lan to a Dell idrac6
# 
# $Header: /opt/home/doke/work/nagios/RCS/check_bmc,v 1.23 2015/12/08 23:38:42 doke Exp $


use strict;   
use warnings;   
use Getopt::Long;
#use Data::Dumper;

use vars qw( $host $username $passwordfile $interface
    $verbose $help @crits @warns @unknowns @oks @ignores );

$ENV{PATH}='/usr/local/bin:/opt/sfw/bin:/usr/sfw/bin:/usr/bin:/bin:/usr/sbin:/sbin';


$host = undef;
$username = undef;
$passwordfile = undef;
$interface = 'auto';
$verbose = 0;
$help = 0;


sub usage {
    my( $rc ) = @_;
    print "Usage: $0 [-vh] [-H host] [-U username] [-f pwfile]
    -H s  host
    -U s  username
    -f s  file containing password
    -I s  interface: bmc, open, lan, lanplus [$interface]
    -v    verbose
    -h    help
";
    exit $rc;
    }

Getopt::Long::Configure ("bundling");
GetOptions(
    'H=s' => \$host,
    'U=s' => \$username,
    'f=s' => \$passwordfile,
    'I=s' => \$interface,
    'v+' => \$verbose,
    'h' => \$help,
    );
&usage( 0 ) if ( $help );

&check();

my $rc = 0;
my $sep = '';
if ( $#crits >= 0 ) {
    $rc = 2;
    print "CRITICAL ", join( ", ", @crits );
    $sep = '; ';
    }
if ( $#warns >= 0 ) {
    $rc = 1 if ( $rc == 0 );
    print $sep, "Warning ", join( ", ", @warns );
    $sep = '; ';
    }
if ( $#unknowns >= 0 ) {
    $rc = -1 if ( $rc == 0 );
    print $sep, "Unknown ", join( ", ", @unknowns );
    $sep = '; ';
    }
if ( $rc == 0 ) {
    print "Ok ", join( ", ", @oks );
    $sep = '; ';
    }
if ( $#ignores >= 0 ) {
    print $sep, "Ignoring ", join( ", ", @ignores );
    }
print "\n";
exit $rc;


##################


sub check {
    my( $ipmicmd, $dev, $sdr_cache, $cmd, $nok, $key, $val, $descr, $junk,
	%frus, $fru, $mfg, $model );

    $cmd = "ipmitool -V > /dev/null"; 
    $verbose && print "+ $cmd\n"; 
    system( $cmd );
    if ( ( $? >> 8 ) != 0 ) { 
	push @unknowns, "unable to run ipmitool: $!";
	return;
	}

    if ( $host ) { 
	# remote host
	# first make sure we can ping it.  
	# ipmitool has some really long timeouts
	if ( ! ping( $host ) ) { 
	    push @unknowns, "can't ping $host";
	    return;
	    }

	if ( $interface eq 'auto' || $interface eq 'bmc' ) { 
	    $interface = 'lan';
	    }

	$ipmicmd = "ipmitool -I $interface -H '$host'"; 
	$ipmicmd .= " -U '$username'" if $username;
	$ipmicmd .= " -f '$passwordfile'" if $passwordfile;

	# In ipmitool 1.8.8, the -c option for csv seperated output is very broken
	# It runs lines together.

	$sdr_cache = "/tmp/sdr_cache.$host.$<";
	if ( ! ( -f $sdr_cache ) || ( -M _ ) > 1.0 ) { 
	    $cmd = "$ipmicmd sdr dump $sdr_cache > /dev/null"; 
	    $verbose && print "+ $cmd\n";
	    system( $cmd );
	    if ( ( $? >> 8 ) == 0 ) { 
		$ipmicmd .= " -S '$sdr_cache'";
		}
	    else { 
		push @warns, "unable to create sdr cache";
		}
	    }
	}
    else { 
	# no host, we're local

	if ( $interface eq 'auto' ) { 
	    if ( -c "/dev/bmc" ) { 
		$dev = "/dev/bmc";
		$interface = 'bmc';
		}
	    elsif ( -c "/dev/ipmi0" ) { 
		$dev = "/dev/ipmi0";
		$interface = "open"; 
		}
	    elsif ( -c "/dev/ipmi/0" ) { 
		$dev = "/dev/ipmi/0";
		$interface = "open"; 
		}
	    elsif ( -c "/dev/ipmidev/0" ) { 
		$dev = "/dev/ipmidev/0";
		$interface = "open"; 
		}
	    else { 
		push @unknowns, "can't find bmc or ipmi device";
		return;
		}
	    }

	if ( -w $dev )  { 
	    $ipmicmd = "ipmitool -I $interface"; 
	    }
	else { 
	    $ipmicmd = "sudo -n ipmitool -I $interface"; 
	    }
	}


    $cmd = "$ipmicmd chassis status"; 
    $verbose && print "+ $cmd\n";
    if ( ! open( pH, "$cmd |" ) ) { 
	push @unknowns, "can't run $cmd: $!";
	return;
	}
    while( <pH> ) { 
	$verbose && print;
	chomp;
	( $key, $val ) = split( m/\s*:\s*/, $_, 2 );
	if ( $key eq 'System Power' && $val ne 'on' ) { 
	    # wtf?  So how is this program running?
	    push @crits, "$key: $val";
	    }
	elsif ( $key eq 'Power Restore Policy' && $val !~ m/always-on|previous/i ) { 
	    push @ignores, "$key: $val";
	    }
	elsif ( $key =~ m/Main Power Fault/i && $val =~ m/true/i ) { 
	    # sometimes this gets left on after the problem clears
	    # so only report it if there are other problems
	    # we've seen this on ironchef, an X4600 with ilom v2.0.2.5
	    push @ignores, "main power fault $val";
	    }
	elsif ( $key =~ m/fault|lockout|overload/i && $val !~ m/false|inactive/i ) { 
	    push @warns, "$key: $val";
	    }
	}
    close pH;


    $frus{ 'mainboard' }{ 'product_manufacturer' } = '';

    $cmd = "$ipmicmd fru list"; 
    $verbose && print "+ $cmd\n";
    if ( ! open( pH, "$cmd 2>&1 |" ) ) { 
	push @unknowns, "can't run $cmd: $!";
	return;
	}
    while( <pH> ) { 
	$verbose && print;
	chomp;
	next unless ( m/\s*(\S*?)\s*:\s*(\S.*)\s*$/ ); 
	( $key, $val ) = split( m/\s*:\s*/, $_, 2 );
	$verbose && print "key $key, val $val\n";
	if ( $key =~ m/FRU Device Description/i ) { 
	    $fru = $val;
	    $verbose && print "fru $fru\n";
	    }
	elsif ( $key =~ m/Product Manufacturer/i ) { 
	    $verbose && print "fru $fru, product_manufacturer $val\n";
	    $frus{ $fru }{ 'product_manufacturer' } = $val;
	    }
	elsif ( $key =~ m/Product Name/i ) { 
	    $verbose && print "fru $fru, product_name $val\n";
	    $frus{ $fru }{ 'product_name' } = $val;
	    }
	elsif ( $key =~ m/Product Serial/i ) { 
	    $frus{ $fru }{ 'product_serial' } = $val;
	    }
	elsif ( $key =~ m/Chassis Serial/i ) { 
	    $frus{ $fru }{ 'chassis_serial' } = $val;
	    }
	elsif ( $key =~ m/Board Mfg/i ) { 
	    $frus{ $fru }{ 'board_mfg' } = $val;
	    }
	elsif ( $key =~ m/Board Product/i ) { 
	    $frus{ $fru }{ 'board_product' } = $val;
	    }
	elsif ( $key =~ m/Board Serial/i ) { 
	    $frus{ $fru }{ 'board_serial' } = $val;
	    }
	elsif ( $verbose > 1 )  { 
	    print "unparsed: $_\n";
	    }
	}
    close pH;


    $mfg = '';
    $model = '';

    # sun style, works on SUN FIRE X4600, SPARC T3-1, Sun Fire X2200 M2
    foreach $fru ( keys %frus ) { 
	if ( $fru =~ m%Mainboard|^/SYS \(ID%i ) { 
	    foreach $key ( keys %{$frus{ $fru }} ) { 
		$val = $frus{ $fru }{ $key };
		$verbose && print "fru $fru, key $key, val $val, mfg $mfg, model $model\n";
		if ( ! $mfg && $key =~ m/Product.Manufacturer/i ) {
		    $mfg = $val;
		    }
		elsif ( ! $model && $key =~ m/Product.Name/i ) {
		    $model = $val;
		    }
		}
	    }
	}
    if ( ! $mfg || ! $model ) { 
	# Dell style, works on PowerEdge R515, PowerEdge R720 
	foreach $fru ( keys %frus ) { 
	    if ( $fru =~ m/Builtin FRU Device/i ) { 
		foreach $key ( keys %{$frus{ $fru }} ) { 
		    $val = $frus{ $fru }{ $key };
		    $verbose && print "fru $fru, key $key, val $val, mfg $mfg, model $model\n";
		    if ( ! $mfg && $key =~ m/Board.Mfg/i ) {
			$mfg = $val;
			}
		    elsif ( ! $model && $key =~ m/Board.Product/i ) {
			$model = $val;
			}
		    }
		}
	    }
	}
		
    $verbose && print "mfg $mfg, model $model\n";



    $cmd = "$ipmicmd sdr elist all"; 
    $verbose && print "+ $cmd\n";
    if ( ! open( pH, "$cmd |" ) ) { 
	push @unknowns, "can't run $cmd: $!";
	return;
	}
    $nok = 0;
    while( <pH> ) { 
	$verbose && print;
	chomp;
	( $key, $junk, $val, $junk, $descr ) = split( m/\s*\|\s*/, $_, 5 );
	if ( $val eq 'ok' ) { 
	    $nok++;
	    }
	# I have no idea what these values mean, but they seem to be useless
	elsif ( $val eq 'ns' || $val eq 'lcr' || $val eq 'lnc' ) { 
	    # ignore it
	    }
	elsif ( $descr =~ '0 unspecified' ) { 
	    # ignore it
	    }
	else { 
	    push @crits, "$key $val $descr";
	    }
	}
    push @oks, "$nok ok sensors";
    close pH;


    if ( $mfg =~ m/Sun/i ) { 
	# sunoem commands need to be run as root, no idea why
	if ( $ipmicmd =~ m/^sudo / ) { 
	    $cmd = "$ipmicmd sunoem led get"; 
	    }
	else { 
	    $cmd = "sudo -n $ipmicmd sunoem led get"; 
	    }
	$verbose && print "+ $cmd\n";
	if ( ! open( pH, "$cmd 2>&1 |" ) ) { 
	    push @unknowns, "can't run $cmd: $!";
	    return;
	    }
	$nok = 0;
	while( <pH> ) { 
	    $verbose && print;
	    chomp;
	    next if ( m/command failed/i );
	    ( $key, $val ) = split( m/\s*\|\s*/, $_, 2 );
	    if ( ( $key eq 'sys.power.led' 
			|| $key eq 'bp.power.led'    # specific to Sun X4100?
			|| $key eq 'fp.power.led'    # specific to Sun X4100?
			|| $key =~ '(ft[0-4]|sc)\.act\.led' )  # specific to Sun X4500?
		    && $val eq 'ON' ) {
		$nok++;
		}
	    elsif ( $key eq 'sys.locate.led' ) { 
		# don't care about the locating led
		$nok++;
		next;
		}
	    elsif ( $val eq 'OFF' ) { 
		$nok++;
		next;
		}
	    elsif ( $key eq 'bmc_send_cmd: Permission denied' ) { 
		# ignore it
		}
	    else { 
		push @crits, "$key $val";
		}
	    }

	if ( $nok ) { 
	    push @oks, "$nok ok leds";
	    }
	else { 
	    # lots of systems produce no led output, all x2200s?
	    #push @unknowns, "no leds could be checked";
	    }
	close pH;
	}
    #elsif ( $mfg =~ m/Dell/i ) { 
	# There's a delloem command, but so far I havn't found anything
	# useful to monitor in it.  You can see the mac address, or set the
	# led states, but you can't get the led states. 
	#}

    if ( $mfg ) { 
	push @oks, $mfg;
	}
    if ( $model ) { 
	push @oks, $model;
	}

    }




# return true if can ping host
sub ping {
    my( $host ) = @_;
    my( $cmd, $rc );

    $verbose && print "pinging host '$host'\n";

    $cmd = "/usr/local/sbin/fping -A -q -t 100 -p 100 '$host'";
    $verbose > 1 && print "+ $cmd\n";
    $rc = system( $cmd ) >> 8;
    if ( $rc ) {
	# failed
	$verbose && print "can't ping host '$host'\n";
	return 0;
	}
    else { 
	# worked
	$verbose && print "successfully pinged host '$host'\n";
	return 1;
	}

    }