#!/usr/bin/bash
# (Solaris)
#--!/bin/bash
# (Linux)
# (Use the applicable line)
#
# A Nagios plugin to check Oracle StorageTek ACSLS status.
# Tested with SL8500 libraries, but possibly also works on other SL libraries as long as ACSLS is used.
# Tested on Solaris 11 x86 / ACSLS 8.5.0 and Solaris 10 x86 / ACSLS 8.3.0
# but should work on Linux and ACSLS 8.0 and higher.
#
# by Tristan Zondag
# tzondag@beeldengeluid.nl
#
# version 1.0 2019-06-07
# version 1.1 2019-10-31
#  - fixed typo in portcheck() function
#

me="$(basename $0)"

usage () {
cat <<EOF
Usage: $me <-l,--lsm|-s,--services|-p,--ports|-w,--watchvols,-h,--help>
A Nagios plugin to check Oracle StorageTek ACSLS status.

Options:
	--lsm, -l	Checks for any LSMs in offline (critical) or recovery (warning) state
	--ports, -p	Checks for offline ports (one warning two critical)
	--services, -s	Checks if ACSDB and ACSLS services are running
	--watchvols, -w Checks if watch_vols is running
	--help, -h	Print this help text.
EOF
}


## exit statuses recognized by Nagios
OK=0
WARNING=1
CRITICAL=2
UNKNOWN=3

TMPFILE=/tmp/$$check_acsls.tmp
ACSLSENV=/var/tmp/acsls/.acsls_env

# source acsss environment variables
if [ -e $ACSLSENV ]; then
	.  $ACSLSENV
else
	echo "Could not source $ACSLSENV!"
	exit $UNKNOWN
fi

## helper functions
die () {
  rc="$1"
  txt="$2"
  shift
  echo "$me: ERROR: $txt";
  exit $rc 
}

warn () {
  (echo -n "$me: WARNING: ";
      if [ $# -gt 0 ]; then echo "$@"; else cat; fi) 1>&2
}

have_command () {
  type "$1" >/dev/null 2>/dev/null
}

require_command () {
  if ! have_command "$1"; then
    die 1 "Could not find required command '$1' in system PATH. Aborting."
  fi
}

is_absolute_path () {
    expr match "$1" '/' >/dev/null 2>/dev/null
}


servicecheck () {
	STATUS=$OK
	require_command acsss

	ACSDBSTAT=$(acsss status | grep acsdb | awk '{print $2}')
	if [ $? -ne 0 ]
	then
			echo "acsss command failed!"
			die $UNKNOWN
	fi

	ACSLSSTAT=$(acsss status | grep acsls | awk '{print $2}')
	if [ $? -ne 0 ]
	then
			echo "acsss command failed!"
			die $UNKNOWN
	fi

	if [ $ACSDBSTAT != 'online' ]
	then
			STATUS=$CRITICAL
			MESSAGE="ACSDB not running!"
	fi

	if [ $ACSLSSTAT != 'online' ]
	then
			STATUS=$CRITICAL
			MESSAGE="${message} ACSLS not running!"
	fi

	if [ $STATUS -eq $OK ]
	then
			MESSAGE="ACSLS services online."
	fi
}

exitwithstatus () {
	# RETURN OUTPUT TO NAGIOS
	if [ $STATUS -eq $OK ]
	then
			echo "OK - ${MESSAGE}"
			exit $OK
	else
			echo "NOK - ${MESSAGE}"
			exit $STATUS
	fi
}

querycmdproc () {

	require_command cmd_proc
	echo $1 | cmd_proc -lq > $TMPFILE 2>&1
	if [ $? -ne 0 ]
	then
		echo "cmd_proc command failed!"
		exit $UNKNOWN
	fi
}

lsmcheck () {
	querycmdproc "query lsm all"
	ONLINE=$(grep online $TMPFILE | wc -l | tr -s " ")
	OFFLINE=$(grep offline $TMPFILE | wc -l | tr -s " ")
	RECOVERY=$(grep recovery $TMPFILE | wc -l | tr -s " ")
	DIAGNOSTIC=$(grep diagnostic $TMPFILE | wc -l | tr -s " ")

	STATUS=$OK
	MESSAGE="${ONLINE} LSM(s) are online."

	if [ $RECOVERY -ne 0 ]; then
		STATUS=$WARNING
		MESSAGE="${RECOVERY} LSM(s) are in recovery mode!"
	fi
	if [ $DIAGNOSTIC -ne 0 ]; then
		STATUS=$WARNING
		MESSAGE="$DIAGNOSTIC LSM(s) are in diagnostic mode!"
	fi

	if [ $OFFLINE -ne 0 ]; then
		STATUS=$CRITICAL
		MESSAGE="${MESSAGE} ${OFFLINE} LSM(s) are offline!"
	fi	

	rm -f $TMPFILE

}

portcheck () {
	querycmdproc "query port all"
	ONLINE=$(grep online $TMPFILE | wc -l | tr -s " ")
	OFFLINE=$(grep offline $TMPFILE | wc -l | tr -s " ")
	AMOUNTPORTS=$(grep 0\, $TMPFILE | wc -l | tr -s " ")
	
	STATUS=$OK
	MESSAGE="$ONLINE Port(s) are online."
	
	if [ $OFFLINE -gt 0 ]; then
		STATUS=$WARNING
		MESSAGE="${OFFLINE} port is offline!"
	fi
	if [ $OFFLINE -eq $AMOUNTPORTS ]; then
		STATUS=$CRITICAL
		MESSAGE="All $AMOUNTPORTS ports are offline!"
	fi	
	rm -f $TMPFILE
}	

watchvolscheck () {
	WATCHVOLS=$(watch_vols)
	STATUS=$OK
	MESSAGE=$WATCHVOLS
	if [ $(echo $WATCHVOLS | grep off | wc -l) -eq 1 ]; then
		STATUS=$WARNING
		MESSAGE=$WATCHVOLS
	fi
}

## parse command-line
# allowed options
short_opts='lspwh'
long_opts='lsm,services,ports,watchvols,help'
if [ $# -eq 0 ]
then
	usage; exit 0
fi
# test which `getopt` version is available:
# - GNU `getopt` will generate no output and exit with status 4
# - POSIX `getopt` will output `--` and exit with status 0
getopt -T > /dev/null
rc=$?
if [ "$rc" -eq 4 ]; then
    # GNU getopt
    args=$(getopt --name "$me" --shell sh -l "$long_opts" -o "$short_opts" -- "$@")
    if [ $? -ne 0 ]; then
        die 1 "Type '$me --help' to get usage information."
    fi
    # use 'eval' to remove getopt quoting
    eval set -- $args
else
    # old-style getopt, use compatibility syntax
    args=$(getopt "$short_opts" "$@")
    if [ $? -ne 0 ]; then
        die 1 "Type '$me --help' to get usage information."
    fi
    set -- $args
fi

while [ $# -gt 0 ]; do
    case "$1" in
	--lsm|-l)		lsmcheck;;
	--services|-s)  servicecheck;;
	--ports|-p)		portcheck;;
	--watchvols|-w)	watchvolscheck;;
        --help|-h)    	usage; exit 0 ;;
        --)           	shift; break ;;
    esac
    shift
done

exitwithstatus