Send mail if a power controller isn't responding or is failing.

sensors/powermon/powermon.in

@@ -42,7 +42,8 @@ use POSIX qw(strftime);
 
 sub usage() {
     print << "END";
-Usage: $0 [-TPC] [ <type> ... ]
+Usage: $0 [-w] [-TPC] [ <type> ... ]
+ -w whiner flag, send mail to $TBOPS about errors
  -T turn on temperature (in degrees F) monitoring
  -P turn on power (watts) consumption monitoring
  -C turn on current (amps) used monitoring
@@ -55,10 +56,19 @@ Usage: $0 [-TPC] [ <type> ... ]
 END
 }
 
-my $optlist = "ANTPC";
+my $optlist = "wANTPC";
 my $dotemps = 1;
 my $dopower = 0;
 my $docurrent = 0;
+my $whiner = 0;
+my $whinefile = "/tmp/powermon.$$";
+my $childpid = -1;
+
+# un-taint path
+$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
+delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};
+
+$| = 1; #Turn off line buffering on output
 
 #
 # Parse command arguments. Once we return from getopts, all that should be
@@ -68,6 +78,13 @@ my %options = ();
 if (! getopts($optlist, \%options)) {
     usage();
 }
+if (defined($options{"w"})) {
+    if (open(WHINE, ">$whinefile")) {
+	$whiner = 1;
+    } else {
+	print "*** Could not open $whinefile, $TBOPS whining disabled\n";
+    }
+}
 if (defined($options{"A"})) {
     $dotemps = $dopower = $docurrent = 1;
 }
@@ -137,33 +154,52 @@ sub dostatus(@) {
     # We do this in a child process so we can time it out if
     # one of the controllers is not responding.
     #
-    my $syspid = fork();
-    if ($syspid) {
+    $childpid = fork();
+    if ($childpid) {
 	#
 	# Allow 5 seconds per controller
 	#
-	local $SIG{ALRM} = sub { kill("TERM", $syspid); };
+	local $SIG{ALRM} = sub { kill("TERM", $childpid); };
 	my $to = 5 * scalar(@wanted);
 	alarm $to;
-	waitpid($syspid, 0);
+	waitpid($childpid, 0);
 	alarm 0;
-	if ($?) {
-	    logit("*** Status fetch failed after $to seconds with exit val $?");
+	my $ecode = $?;
+	if ($ecode) {
+	    my $cause;
+
+	    if (($ecode & 0xff) > 0) {
+		$cause = "timed out after $to seconds";
+	    } else {
+		$cause = "failed with exit value ". ($ecode >> 8);
+	    }
+	    logit("*** Status fetch $cause");
+	    #
+	    # If pissin' and moanin' send the last run log to TBOPS
+	    #
+	    if ($whiner) {
+		SENDMAIL($TBOPS,
+			 "WARNING: power controller(s) $cause",
+			 "There were errors getting status from one or more ".
+			 "of the power controllers,\nthe log is appended.",
+			 $TBOPS, undef,
+			 ($whinefile));
+	    }
 	}
-	return $?;
+	return $ecode;
     }
 
     for my $ctrl (@wanted) {
 	my %status;
 
 	if (!defined($ctrls{$ctrl})) {
-	    warn "No such power controller '$ctrl', ignored\n";
+	    logit("No such power controller '$ctrl', ignored");
 	    $errors++;
 	    next;
 	}
 	if ($ctrls{$ctrl} =~ /^RPC/) {
 	    if (rpc27status($ctrl,\%status)) {
-		warn "Could not get status for $ctrl.\n";
+		logit("Could not get status for $ctrl.");
 		$errors++;
 		next;
 	    }
@@ -182,11 +218,12 @@ sub dostatus(@) {
 	    }
 	    logit($msg);
 	} elsif (!$doall) {
-	    warn "Cannot get status for $ctrl (type " .
-		$ctrls{$ctrl} . ") yet\n";
+	    logit("Cannot get status for $ctrl (type " .
+		$ctrls{$ctrl} . ") yet");
 	    $errors++;
 	}
     }
+    close(WHINE);
     exit($errors);
 }
 
@@ -194,4 +231,16 @@ sub logit($) {
     my ($msg) = @_;
 
     print strftime("%b %e %H:%M:%S", localtime)." powermon[$$]: $msg\n";
+    if ($whiner && $childpid == 0) {
+	print WHINE $msg, "\n";
+    }
+}
+
+END {
+    return
+	if ($childpid == 0 || !$whiner);
+
+    my $ecode = $?;
+    unlink($whinefile);
+    $? = $ecode;
 }