#!/usr/local/bin/perl

# checks a list of servers, if they do not ping, sends a message
#
# Created  01-Jun-1998, Brandon Gillespie <brandon@roguetrader.com>
# Modified 02-Jun-1998, Phil Humpherys <humphery@www.spire.com>
#          -- fixed spelling error,started ip-only lookup.
#          -- started uptime logger for clusters.
# Modified 03-Jun-1998, Brandon Gillespie <brandon@roguetrader.com>
#          -- changed it to use the exit status of ping
#          -- finished setup @check to be IPs, mapping to names 
#             only for messages
#          -- changed it to wait once on a dead ping response
# Modified 11-Jun-1998, Brandon Gillespie <brandon@roguetrader.com>
#          -- it will email out a msg now as well
#          -- fixed bug where it would say a host was 'back up'
#             when it had never reported it being down.

## startup delay
if ($ARGV[0] > 0) {
    print "Waiting $ARGV[0] seconds before starting...\n";
    sleep($ARGV[0]);
}

$config = "/usr/local/etc/pingcheck.conf";

$SIG{HUP} = 'sighup';

select(STDOUT); $| = 1;

my %deadt = ();
my %dead = ();
$rotate_checkpoint = 0;

&readcfg();

checkpoint();

for (;;) {
    for $host (@check) {
        $rc = 0xffff & system("($ping -c1 $host) 2>&1 > /dev/null");
        $nhost = $ncheck{$host};
        if ($rc == 0) {
            ## it ran Ok
            if ($dead{$host} >= 1) {
                if ($dead{$host} > 1) {
                    mailmsg("$nhost is back up");
                    $rotate_checkpoint = 1;
                }
                $dead{$host} = 0;
                $deadt{$host} = 0;
            }
        } elsif ($rc > 0x80) {
            ## it did not run Ok
            if ($dead{$host} > 1) {
                if ((time() - $deadt{$host}) >= $dead_interval) {
                    if ($dead{$host} < 6) {
                        mailmsg("$nhost is STILL not responding!");
                        $deadt{$host} = time();
                    }
                    $dead{$host}++;
                }
            } elsif ($dead{$host} == 1) {
                mailmsg("$nhost is not responding!");
                $dead{$host}++;
            } else {
                $deadt{$host} = time();
                $dead{$host} = 1;
            }
        } elsif ($rc == 0xff00) {
            die("failed to exec $ping\n");
        }
    }
    checkpoint();
    sleep($interval);
}

sub checkpoint {
    if (!$checkpoint) {
        return;
    }
   
    if ($rotate_checkpoint) {
        rename("/tmp/pingcheck.run", ("/tmp/pingcheck.run." . time()));
        $rotate_checkpoint = 0;
    }

    my ($k, $v);
    open(F, ">/tmp/pingcheck.run") || print ("open(pingcheck.run): $!\n");
    for $k (sort(keys %dead)) {
        $v = $dead{$k};
        print F "dead:  $k => $v\n";
    }
    for $k (sort(keys %deadt)) {
        $v = $deadt{$k};
        print F "deadt: $k => $v\n";
    }
    close(F);
}

sub mailmsg {
    my $msg = $_[0];

    ($sec,$min,$hour,$mday,$mon,$year,$wday,$yday,$isdst) = localtime();
    $mon++;
    if ($isdst) {
        $tz = "MDT";
    } else {
        $tz = "MST";
    }
    $date = sprintf("$mon/$mday/$year $hour:%02d $tz", $min);

    if ($qpage_server) {
        system($qpage, "-s", $qpage_server, "-p", $qpage_recip, "$msg $date");
    }

    open(MSG, "|$sendmail") || die "Unable to open sendmail: $!\n";

    print MSG<<END;
From: $mailfrom
To: $mailto
Subject: $msg $date
END
    close(MSG);

    print "[$date] $msg\n";
}

sub readcfg {
    do $config;

    ## @check should be IPs, lets map them to names.
    %ncheck = ();
    my $ips = join("\n", @check);
    my @list = split(/\n/, `echo "$ips" | /usr/sbin/nslookup | egrep '^Name'`);
    my ($s, $n, $x, @n);
    for $x (0 .. $#list) {
        $s = $list[$x];
        if ($s =~ /^Name:\s+([a-z0-9.]+)$/) {
            @n = split(/\./, $1);
            if ($#n > 2) {
                $n = $n[0];
            } else {
                $n = join(".", @n);
            }
            $ncheck{$check[$x]} = $n;
        }
    }
    ## this will drop anything we can't do an nslookup on
    @check = keys(%ncheck);

    for $x (@check) {
        $dead{$x} = 0;
        $deadt{$x} = 0;
    }

    ## print a msg
    print "[" . localtime() . "] Reading $config\n";
    print "Checking servers: " . join(", ", sort values %ncheck) . "\n";
    print "Using a $interval second interval.\n";
    print "Using group $qpage_recip at NPP server $qpage_server\n";
    print "Mailing to $mailto as $mailfrom\n";
    print "Checkpoint logfile: " . ($checkpoint ? "yes" : "no") . "\n";
}

sub sighup {
    &readcfg();
}