scripts/xymon/plugins/client/ext/smartoverall

#!/bin/sh
# .. vim: foldmarker=[[[,]]]:foldmethod=marker

# NOTE: Must be run as root, so you probably need to setup sudo for this.

# This script is mostly intend to be used with Xymon and rather for devices unknown to the smartmontools base.
# Based on xymon.com's script : https://www.xymon.com/xymon-cgi/viewconf.sh?smart
# The script will scan all devices compatible with SMART and for each disk, it will : [[[
#	* try to guess the expected TYPE (even megaraid,…).
#	* display health status.
#	* set a "clear" state for incompatible device.
#	* display last selftests.
#	* set a "error" state if no selftest is recorded.
#	* display basic informations.
#	* recommend a more advanced SMART script if the disk is known of smartmontools's database (drivedb.h) or redirect to smartmontools's FAQ if not.
# ]]]
# Things the script CAN'T do : [[[
#	* ensure a recent selftest was run.
#	* compare current value with vendor's one (for failure prediction or error).
#	* give detail about errors.
#	* Take a look to this more advance script for such features : https://github.com/skazi0/xymon-plugins/blob/master/client/ext/smart
# ]]]

# Vars [[[
debug="1"

## Colors [[[
c_redb='\033[1;31m'
c_magentab='\033[1;35m'
c_reset='\033[0m'
## ]]]

plugin_name=$(basename "${0}")

plugin_result="${XYMONTMP}/${MACHINEDOTS}.smartoverall.plugin_result"
plugin_state="${XYMONTMP}/${MACHINEDOTS}.smartoverall.plugin_state"
device_list="${XYMONTMP}/${MACHINEDOTS}.smartoverall.dscan"
## List of devices known from the smartmontools base and compatible with test logging
## This file might be used by a more advanced script such as skazi0's one
drivedb_list="${XYMONTMP}/${MACHINEDOTS}.smart.drivedb.list"

# By default, don't empty files newer than 10hours (600 minutes)
default_mtime_minutes="600"

xymon_username="xymon"
xymon_groupname="xymon"
# ]]]

# Functions
## Create or empty a file if it's too old [[[
## First argument (required): Absolut path to the file
## Second argument (optionnal): Maximum number of minutes since last modification
regenerate_if_too_old() {
	## Set variables according to the number of passed arguments [[[
	case $# in
		0 )
			[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : regenerate_if_too_old func − Need at least 1 argument."
			exit 1
			;;
		1 )
			_file="${1}"
			[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : regenerate_if_too_old func − Use default_mtime_minutes value: ${default_mtime_minutes}."
			_max_mtime_minutes="${default_mtime_minutes}"
			;;
		2 )
			_file="${1}"
			_max_mtime_minutes="${2}"
			;;
	esac
	## ]]]
	_current_timestamp=$(date "+%s")
	_file_mtime_timestamp=$(stat --format="%Y" -- "${_file}")

	## Substract last modification timestamp of the file to current timestamp
	: $(( _file_mtime_seconds=_current_timestamp-_file_mtime_timestamp ))
	## Get maximum allowed mtime in seconds
	: $(( _max_mtime_seconds=_max_mtime_minutes*60 ))

	## Compare last modification mtime with the maximum allowed
	if [ "${_file_mtime_seconds}" -gt "${_max_mtime_seconds}" ]; then
		[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : regenerate_if_too_old func − Need to empty or create ${_file} last modification happened ${_file_mtime_seconds} seconds ago (maximum is ${_max_mtime_seconds})."
		true > "${_file}"
		chown -- "${xymon_username}":"${xymon_groupname}" "${_file}"
	else
		[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : regenerate_if_too_old func − Don't need to empty ${_file} last modification happened ${_file_mtime_seconds} seconds ago (maximum is ${_max_mtime_seconds})."
	fi

}
## ]]]
## Test if a disk really support SMART [[[
## Smartctl can give an health status even without a full support
## of SMART for some type (eg. scsi or megaraid).
## Exemple : SMART support is: Unavailable - device lacks SMART capability.
is_disk_support_smart() {
	_disk="${1}"
	_type="${2}"

	_smarctl_support_result="${XYMONTMP}/${MACHINEDOTS}.smartoverall.support.$(basename "${_disk}").${_type}"

	smart_support_msg=""

	[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : is_disk_support_smart func − check if SMART is supported on : ${_disk}."

	## Create or empty previous file only if older than 24h (1440 minutes)
	regenerate_if_too_old "${_smarctl_support_result}" 1440

	## Grep only "support" lines from disk's informations only if the file was emptied
	if test ! -s "${_smarctl_support_result}"; then
		smartctl -d "${_type}" -i -- "${_disk}" | grep -E "^SMART support is:" -- >> "${_smarctl_support_result}"
	fi

	## If the file is not empty
	if test -s "${_smarctl_support_result}"; then
		## Parse all "support" lines
		while IFS= read -r _LINE; do
			if ! printf -- '%s' "${_LINE}" | grep -q -E -- "(Enabled|Available)"
			then
				smart_support_msg="${_LINE}"
			fi
		done < "${_smarctl_support_result}"
	else
		smart_support_msg="smartctl was not able to open ${_disk} DEVICE with ${_type} TYPE."
	fi

	if [ -z "${smart_support_msg}" ]; then
		[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : is_disk_support_smart func − SMART seems fully supported on : ${_disk} with ${_type} type."
	else
		[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : is_disk_support_smart func − SMART is not fully supported on : ${_disk} with ${_type} type. See smartctl informations :\n${smart_support_msg}"
	fi

## Clean temp files
### As the Xymon's tmpdir is used to store log files, no need to delete them at
### the end of the script. They will be emptied, reused or regenerate (if oldest
### than the expected interval) at the next run.

}
## ]]]
## Test the type of disk with smartctl [[[
## Cause the scanned one might not be the one to use
choose_correct_type() {
	_disk="${1}"
	_scanned_type="${2}"
	_default_type="auto"

	TYPE=""
	SMART_SUPPORT_MSG=""

	for test_type in "${_scanned_type}" "${_default_type}"; do
		is_disk_support_smart "${_disk}" "${test_type}"

		## If no message, the type is correct
		if [ -z "${smart_support_msg}" ]; then
			TYPE="${test_type}"
			SMART_SUPPORT_MSG=""
			return
		else
			SMART_SUPPORT_MSG="${smart_support_msg}"
		fi

	done

}
## ]]]

# Create or empty previous files
true > "${plugin_result}"
chown -- "${xymon_username}":"${xymon_groupname}" "${plugin_result}"
true > "${plugin_state}"
chown -- "${xymon_username}":"${xymon_groupname}" "${plugin_state}"
## Create or empty previous file only if older than 24h (1440 minutes)
regenerate_if_too_old "${device_list}" 1440
regenerate_if_too_old "${drivedb_list}" 1440

# Get the list of all available devices if the previous list was emptied
if test ! -s "${device_list}"; then
	smartctl --scan >> "${device_list}"
fi

# If the file is not empty
if test -s "${device_list}"; then
	while IFS= read -r LINE; do
		## Get device path
		DISK=$(echo "${LINE}" | cut -d" " -f1)
		## Try to determine the best type
		SCANNED_TYPE=$(echo "${LINE}" | cut -d" " -f3)
		choose_correct_type "${DISK}" "${SCANNED_TYPE}"

		## If no correct type was found for this device
		if [ -z "${TYPE}" ]; then
			[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : SMART is not fully supported."
			DRES=$(printf '%s' "SMART Health Status can't be determine because of:\n${SMART_SUPPORT_MSG}")
			DCODE="2"
			TYPE="unsupported"
			### Still try to display informations about unsupported device (eg. RAID controller,…)
			DID="unsupported-${DISK}"
			DINFO=$(smartctl -i -d "${SCANNED_TYPE}" "${DISK}" | grep -v -E "^smartctl|^Copyright|^$" || printf '%s' "Can't get informations due to no SMART support.")
			DDRIVEDB_MSG=""
			DSELFTEST=""
		else
			[ "${debug}" -eq "0" ] && printf "${c_magentab}%-6b${c_reset}\n" "DEBUG : SMART seems fully supported, proceed normally."
			### Get SMART Health Status and return code
			DRES=$(/usr/sbin/smartctl -H -d "${TYPE}" -n standby "${DISK}")
			DCODE=$?
			### Get disk's serial number and informations
			DID=$(smartctl -i -d "${TYPE}" "${DISK}" | awk '/.erial .umber:/ { print $NF }')
			DINFO=$(smartctl -i -d "${TYPE}" "${DISK}" | grep -v -E "^smartctl|^Copyright|^$")

			## If the model of the disk is known from smartmontools database
			if smartctl -d "${TYPE}" -P show "${DISK}" | grep -qi -- "drive found in"; then
				DDRIVEDB_MSG="&green Device is known in smartmontools database. You might consider using a more advanced plugin such as:
https://github.com/skazi0/xymon-plugins/blob/master/client/ext/smart"
			else
				DDRIVEDB_MSG="&clear Device is unknown or not complete in smartmontools database. Please take a look to the FAQ:
https://www.smartmontools.org/wiki/FAQ#SmartmontoolsDatabase"
			fi

			DSELFTEST=$(smartctl -d "${TYPE}" -l selftest "${DISK}" | grep -v -E -- "^smartctl|^Copyright|^$")
			## If no selftest have been recorded
			if smartctl -d "${TYPE}" -l selftest "${DISK}" | grep -qi -- "No self-tests"; then
				DSELFTEST_MSG="&red No self-tests recorded:"
				DCODE="8"
			## If the device doesn't support test logging
			elif smartctl -d "${TYPE}" -l selftest "${DISK}" | grep -qEi -- "(does not support.*logging|Log not supported)"; then
				DSELFTEST_MSG="&clear Test logging are not supported:"
			else
				DSELFTEST_MSG=""
				### If the device is also known from smartmontools database
				###    and not already present in the list of compatible disk
				if printf -- '%s' "${DDRIVEDB_MSG}" | grep -q -E -- "green" &&
					! grep -q -- "${DISK}" "${drivedb_list}"
				then
					echo "${DISK}" >> "${drivedb_list}"
				fi
			fi
		fi

		## Test health status
		DSTBY=$(( DCODE & 2 ))
		DFAIL=$(( DCODE & 8 ))
		DWARN=$(( DCODE & 32 ))

		## According to health, give a weight to each color to easily get the page status
		if test $DSTBY -ne 0
		then
			COLOR="4&clear"
		elif test $DFAIL -ne 0
		then
			COLOR="1&red"
		elif test $DWARN -ne 0
		then
			COLOR="2&yellow"
		else
			COLOR="3&green"
		fi

		## Avoid duplicate device
		if ! grep -q "${DID}" "${plugin_result}"; then
			## For summary
			echo "${COLOR} $DISK ${TYPE}"

			## For detailed informations
			{
				echo "${COLOR} $DISK ${TYPE}" | cut -c2-
				echo ""
				echo "$DRES" | grep -v -E "^smartctl|^Copyright|^$|^==="
				echo "${DDRIVEDB_MSG}"
				echo "${DINFO}"
				echo "${DSELFTEST_MSG}"
				echo "${DSELFTEST}" | head -n12
				echo "------------------------------------------------------------"
				echo ""
				echo ""
			} >> "${plugin_result}"
	fi
	done < "${device_list}" >> "${plugin_state}"

# If the file is empty
else
	echo "1&red Error while scanning devices with smartctl" >> "${plugin_state}"
fi

# Set the global color according to the highest alert
COLOR=$(< "${plugin_state}" awk '{print $1}' | sort | uniq | head -1 | cut -c3-)

# Send informations to Xymon server
$XYMON "${XYMSRV}" "status ${MACHINE}.${plugin_name} ${COLOR} SMART health check

$(< "${plugin_state}" cut -c2-)

==================== Detailed status ====================

$(cat "${plugin_result}")
"

## Clean temp files
### As the Xymon's tmpdir is used to store log files, no need to delete them at
### the end of the script. They will be emptied, reused or regenerate (if oldest
### than the expected interval) at the next run.

exit 0