SMART disk self-test script

disk_tests.sh

#!/bin/bash
#
###################################################################
#
# Run short SMART self-test on all drives and email the result.
#   Populate DISKLIST with your drive devices (w/o the '/dev/' part)
#   and MailTo, MailFrom, MailSubject with your email header info.
#
# Your system needs a mailer set up for this to work as-is.
# Home Assistant parts require a suitable MQTT sensor definition.
#   BROKER is the MQTT broker's IP or FQDN
#   IFACE is the interface IP of this host to use when connecting
#   to the MQTT broker
#   TOPIC is the MQTT message topic defined by your HA sensor.
#
###################################################################
#
# Jeff Thompson, March 2023
#
# Licence: Copy and adapt freely!
#
###################################################################
TMP=/tmp/disktests.$$
BROKER=mqtt.boolie.net
IFACE=192.168.11.16
MSG=/tmp/msg.$$
MQTEMP=/tmp/mqtt.$$
PARAM=$1
HOST=`hostname`
TOPIC="systems/$HOST"
[email protected]
[email protected]
MailSubject="SMART test results from $HOST"
DISKLIST="sda sdb sdc sdd sde sdf sdg nvme0 nvme1"
NASTY=0

if [ "$PARAM" == "" ]
then
    PARAM="ALL"
fi

echo >$MSG

if [ "$PARAM" == "ALL" ]
then
# request short self-tests
    for DISK in ${DISKLIST}
    do
	if [ -e /dev/${DISK} ]
	then
		/usr/sbin/smartctl -t short /dev/${DISK} >/dev/null 2>&1
	fi
    done

    # test takes typically 2 mins; wait 3 mins to be sure
    sleep 180

    # check the results
    for DISK in ${DISKLIST}
    do
	if [ -e /dev/${DISK} ]
	then
	    RES=`/usr/sbin/smartctl -a /dev/${DISK} 2>/dev/null | grep "test result:" | sed "s/.*test result: //"`
	    if [ "$RES" == "PASSED" ]
	    then
		echo "SMART: $DISK passed short self-test" >>$MSG
		echo "{ \"${DISK}_smart_test\": \"Okay\" }" >$MQTEMP
	    elif [ "$RES" == ""]
	    then
		echo "SMART: $DISK did not return a self-test result" >>$MSG
		echo "{ \"${DISK}_smart_test\": \"Fail\" }" >$MQTEMP
		NASTY=1
	    else
		echo "SMART: $DISK did not pass short self-test. Result=$RES.  Full result follows." >>$MSG
		/usr/sbin/smartctl -a /dev/${DISK} >>$MSG 2>&1
		echo "-------------------------------------------------------------------------" >>$MSG
		echo "{ \"${DISK}_smart_test\": \"Fail\" }" >$MQTEMP
		NASTY=1
	    fi
	    # send to HA (retain to avoid misleading 'unknown' values after restarts)
            /usr/bin/mosquitto_pub -A $IFACE -h $BROKER -t $TOPIC -f $MQTEMP --retain
	fi
    done
fi

# send to email if there's anything to say
if [ $NASTY -eq 1 ]
then
    mail -s "${MailSubject}" "${MailTo}" -r "${MailFrom}" <$MSG
fi

# tidy up
rm -f $MSG $TMP $MQTEMP

exit 0