One Hat Cyber Team

View File Name : check_ssd_health

#!/bin/bash

# Getting parameters:
while getopts "d:w:c:L:h" OPT; do
    case $OPT in
        "d") _DISK=$OPTARG;;
        "w") _WARN=$OPTARG;;
        "c") _CRIT=$OPTARG;;
        "L") _LIFETIME=$OPTARG;;
        "h") help;;
    esac
done

STATE_OK=0
STATE_WARNING=1
STATE_CRITICAL=2
STATE_UNKNOWN=3

RETVAL=${STATE_UNKNOWN}

function help {
    echo "Usage: $0 -d <DEVICE> -w <WARNING PERCENT> -c <CRITICAL PERCENT> [-L LIFETIME TBW OVERRIDE]"
    echo "Example: $0 -d /dev/sda -w 80 -c 90 -L 140"
    echo "Example: $0 -d '-d sat+megaraid,3 /dev/sda' -w 80 -c 90 -L 140"
    exit ${STATE_UNKNOWN}
}


if [ -z "${_DISK}" ]; then
    help
fi

sudo /usr/sbin/smartctl -a ${_DISK} 2>&1 >/dev/null
_RV=$?
if [[ $((${_RV} & 2)) -gt 0 ]]; then
    echo "Failed to open device (${_DISK})"
    RETVAL=${STATE_CRITICAL}
else

    _DATA="$( sudo /usr/sbin/smartctl -a ${_DISK} | grep '\(NAND_Writes_32MiB\|Total_LBAs_Written\|Data Units Written:\|Device Model\|Model Number\)' | head -n2 | tr -d '[]' | sed 's/ GB//g' | rev | cut -d' ' -f1,2 | rev )"

    if [ $( echo "${_DATA}" | wc -l ) -ne 2 ]; then
        echo "Error in smartctl output: ${_DATA}"
        RETVAL=${STATE_UNKNOWN}
    else
        _MODEL=$( echo "${_DATA}" | head -n1 | sed 's/^ //g' | tr ' -' '_' )

        if [ -n "${_LIFETIME}" ]; then
            _MODEL_MLTW=${_LIFETIME}
        else
            LTW_=0
            LTW_INTEL_SSDSC2BB150G7=412
            LTW_INTEL_SSDSC2BB240G4=140
            LTW_INTEL_SSDSC2BB240G6=140
            LTW_INTEL_SSDSC2KB240G7=620
            LTW_INTEL_SSDSC2BB480G4=275
            LTW_INTEL_SSDSC2BB480G6=275
            LTW_INTEL_SSDSC2BB480G7=945
            LTW_INTEL_SSDSC2BB800G6=450
            LTW_INTEL_SSDSC2BB800G7=1663
            LTW_INTEL_SSDSC2KB960G7=1860
            LTW_INTEL_SSDSC2KB960G8=3400
            LTW_INTEL_SSDSC2BB012T7=2455
            LTW_INTEL_SSDSC2KB019T7=3270
            LTW_Micron_5300_MTFDDAK1T9TDT=17520
            LTW_Micron_5400_MTFDDAK1T9TGB=17520
            LTW_INTEL_SSDPF2KE016T1=8400
            LTW_SAMSUNG_MZ7LH3T8HMLT_00005=5466

            
            _MODEL_MLTW=$( eval echo \$$( eval echo $( echo LTW_${_MODEL}) ) )
            if [ -z "${_MODEL_MLTW}" ]; then
                echo "UNKNOWN: Unknown SSD model maximum lifetime writes (${_MODEL})"
                exit ${STATE_UNKNOWN}
            fi
        fi

        if [[ ${_MODEL} =~ "Micron" ]]; then
            _CLTW=$(( $( echo "${_DATA}" | tail -n1 ) /1000/1000/1000/2 ))
        elif [[ ${_MODEL} =~ "INTEL_SSDPF" ]]; then
            # intel nvme
            _CLTW="$( printf '%.3f' "$( echo "scale=3 ; $( echo "${_DATA}" | tail -n1 | cut -d' ' -f2 )/1000"  | bc )" )"
        elif [[ ${_MODEL} =~ "SSDSC" ]]; then
            # intel ssd
            _CLTW=$(( $( echo "${_DATA}" | tail -n1 ) *32/1024/1024 ))
        elif [[ ${_MODEL} =~ "SAMSUNG" ]]; then
            # samsung ssd
            _CLTW=$(( $( echo "${_DATA}" | tail -n1 ) *512 /1024 /1024 /1024 /1024 ))
        fi

        _CURR_PERCENT=$( echo "scale=0 ; ${_CLTW} * 100 / ${_MODEL_MLTW}" | bc )

        if [ ${_CURR_PERCENT} -ge ${_CRIT} ]; then
            echo "${_CLTW} TB (${_CURR_PERCENT}%) written of max ${_MODEL_MLTW} TB by \"${_DISK} ${_MODEL}\"|terabytes_written=${_CLTW};${_MODEL_MLTW};"
            RETVAL=${STATE_CRITICAL}
        elif [ ${_CURR_PERCENT} -ge ${_WARN} ]; then
            echo "${_CLTW} TB (${_CURR_PERCENT}%) written of max ${_MODEL_MLTW} TB by \"${_DISK} ${_MODEL}\"|terabytes_written=${_CLTW};${_MODEL_MLTW};"
            RETVAL=${STATE_WARNING}
        else
            echo "${_CLTW} TB (${_CURR_PERCENT}%) written of max ${_MODEL_MLTW} TB by \"${_DISK} ${_MODEL}\"|terabytes_written=${_CLTW};${_MODEL_MLTW};"
            RETVAL=${STATE_OK}
        fi
    fi
fi

exit ${RETVAL}