source: npl/overig/smartmontools/syn3-smartcheck @ 105afb5

#!/bin/bash
#(C)2007 DatuX. All rights reserved.
#Automatic harddisk smart monitoring via Syn-3 monitoring framework.

mkdir -p /etc/smart
touch /etc/smart/config
source /etc/smart/config

#return 0 if selftest of a device is disabled.
is_selftest_disabled()
{
        eval SELFTEST_DISABLED=\$SELFTEST_DISABLED_${1};
        if [ "$SELFTEST_DISABLED" == "1" ]; then
                return 0;
        fi
        return 1;
}

#return 0 if monitoring of a device is disabled.
is_monitoring_disabled()
{
        eval MONITORING_DISABLED=\$MONITORING_DISABLED_${1};
        if [ "$MONITORING_DISABLED" == "1" ]; then
                return 0;
        fi
        return 1;
}

#runs a smartctl commandline and syn3-state logs result
run_smartctl()
{
        #is this smart drive accesible this way?
        $SMARTCMD -i > /tmp/smart.$$ 2>/dev/null || return $?
            
        #yes, so get info
        INFO=`cat /tmp/smart.$$ | egrep '(Device Model:|Device:|Model Family:|Serial Number:)' | sed -r 's/ +/ /g' | tr "\n" " "`
        ID="`egrep '(^Device:|Serial )' /tmp/smart.$$ | md5sum`"
        
        #we already saw this disk?
        #(sometimes disks are accesible in more then one way, for example with smart array controlers)
        if echo "$IDS" | grep "$ID" >/dev/null; then
            return 0
        fi
        IDS="$IDS $ID"
        
        #re-enable smart if this is not yet the case:
        $SMARTCMD -s on &>/dev/null

        #get current status
        OUTPUT=`$SMARTCMD -H -i -A -l error -l selftest -q errorsonly 2>&1`
        EXIT=$?

        if [ "$OUTPUT" ]; then
                OUTPUT="Extra output: [ $OUTPUT ]";
        fi
        
        #List devices here
        if [ "$LIST" == 1 ]; then
                echo "`echo "$ID" | cut -d ' ' -f1`     $FACILITY       $INFO"; 
                return 0;
        fi

        #graph of temperature, if any
        TEMPERATURE=`$SMARTCMD -A|grep Temperature|head -1 | cut -c88- | egrep -o '^[0-9]+'`
        if [ "$TEMPERATURE" ]; then
                #TODO: add temperature_max_alert and caution?
                syn3-graphcreate "$FACILITY" --step=60 --use=MAX temperature=GAUGE:120:0:U
                syn3-graph "$FACILITY" temperature="$TEMPERATURE"
        fi

        #if monitoring is disabled in config file, skip current device.
        is_monitoring_disabled `echo "$ID" | cut -d ' ' -f1`;
        if [ "$?" == 0 ]; then
                syn3-state "$FACILITY" DELETE
                return 0
        fi

        #analyse status code we've got:
        if    [ "$(( $EXIT & 8 ))" != 0 ]; then
                syn3-state "$FACILITY" ALERT "Harddisk $INFO is FAILING! $OUTPUT"
        elif [ "$(( $EXIT & 16 ))" != 0 ]; then
                syn3-state "$FACILITY" ALERT "Harddisk $INFO is almost failing. $OUTPUT"
        elif [ "$(( $EXIT & 32 ))" != 0 ]; then
                syn3-state "$FACILITY" CAUTION "Harddisk $INFO has almost failed in the past. $OUTPUT"
        elif [ "$(( $EXIT & 128 ))" != 0 ]; then
                syn3-state "$FACILITY" CAUTION "Harddisk $INFO selftest failed. $OUTPUT"
        elif [ "$(( $EXIT & 64 ))" != 0 ]; then
                syn3-state "$FACILITY" OK "Harddisk $INFO is healty, but has logged errors: $OUTPUT"
        elif [ "$(( $EXIT & 2 ))" != 0 ]; then
                syn3-state "$FACILITY" DELETE #note: this cant be called anymore
        elif [ "$EXIT" == 0 ]; then
                syn3-state "$FACILITY" OK "Hardisk $INFO is healty. $OUTPUT"
        else
                syn3-state "$FACILITY" CAUTION "Hardisk $INFO has unknown SMART-status. $OUTPUT"
        fi

        #if selftest is disabled in config file, skip current device.
        is_selftest_disabled `echo "$ID" | cut -d ' ' -f1`;
        if [ "$?" == 0 ]; then
                return 0
        fi

        #Start a long self-test at mid-night
        if [ `date +%H%M` == "0000" ]; then
                $SMARTCMD -t long &>/dev/null
        fi
}

if [ "$1" == "--list" ]; then
        LIST=1;
fi

#traverse all blockdevices
cd /sys/block
for SYS in *; do
        DEV=/dev/`echo $SYS| sed s@!@/@g`

        #range=1 (things like dm, loop, fd etc)?
        [ "`cat $SYS/range`" == "1" ] && continue;

        #try default first (this works for ata and scsi):
        SMARTCMD="smartctl $DEV"
        FACILITY="SMART-`echo $DEV| sed s#.*/##`"
        run_smartctl && continue

        #SAT
        SMARTCMD="smartctl $DEV -d sat"
        FACILITY="SMART-`echo $DEV| sed s#.*/##`"
        run_smartctl && continue

        #try cciss Smart Array controller mode:
        FOUND=
        for DRIVE in `seq 0 15`; do
                SMARTCMD="smartctl $DEV -d cciss,$DRIVE"
                FACILITY="SMART-`echo $DEV| sed s#.*/##`-$DRIVE"
                run_smartctl && FOUND=1
        done
        [ "$FOUND" ] && continue;

        #try 3ware controller mode:
        FOUND=
        for DRIVE in `seq 0 15`; do
                SMARTCMD="smartctl $DEV -d 3ware,$DRIVE"
                FACILITY="SMART-`echo $DEV| sed s#.*/##`-$DRIVE"
                run_smartctl && FOUND=1
        done
        [ "$FOUND" ] && continue;

done

rm /tmp/smart.$$