Hard drive health testing on linux

disktest.sh

#!/bin/bash -u

usage()
{
    echo "usage: $0 [OPTIONS] DEVICE [DEVICE...]"
    echo
    echo "Options:"
    echo "    -s|--smart          do a smart short test"
    echo "    -e|--smarterrors    do a smart errors test"
    echo "    -d|--devstat        do a smart devstat test"
    echo "    -b|--badblocks      do a write but non-destructive test"
    echo "    -a|--all            do all the tests above"
    echo
    echo "(example: $0 -d sda)"
}

do_smarttest()
{
    device="$1"
    echo "[+] Stopping SMART tests on ${device}..."
    smartctl -X "/dev/${device}" >/dev/null
    echo "[+] Starting short SMART test on ${device}..."
    smartctl -t short "/dev/${device}" >/dev/null
    while true; do
        output="$(smartctl -l selftest /dev/${device})"
        if grep -q "Self-test routine in progress" <<< "$output"; then
            echo -n "."
            sleep 1
        else
            echo
            break
        fi
    done
    result="$(smartctl -H "/dev/${device}" | grep 'test result')"
    if grep -q "PASSED" <<< "$result"; then
        echo "[+] Success"
    else
        echo "[-] Failure"
        return 1
    fi
}

do_blocktest()
{
    device="$1"
    lastblock="${2:-1000000}"
    badblocksfile="$(mktemp)"
    echo "[+] Starting bad blocks test on ${device}..."
    badblocks -sne 1 -o "${badblocksfile}" "/dev/${device}" "${lastblock}"
    if [ -s "${badblocksfile}" ]; then
        echo "[-] Failure"
        result=1
    else
        echo "[+] Success"
        result=0
    fi
    rm -f "${badblocksfile}"
    return $result
}

do_smarterrorstest()
{
    device="$1"
    echo "[+] Starting SMART errors test on ${device}..."
    output="$(smartctl -l error /dev/${device})"
    errors=$(grep 'ATA Error Count:' <<< "$output" | cut -d' ' -f4)
    if grep -q 'No Errors Logged' <<< "$output"; then
        echo "[+] Success"
        return 0
    elif [ -z "$errors" ]; then
        echo "[+] Error retrieving error count"
        return 2
    elif [ "$errors" -gt "0" ]; then
        echo "[-] Failure ($errors errors)"
        return 1
    elif [ "$errors" -eq "0" ]; then
        echo "[+] Success"
        return 0
    else
        echo "[-] Error retrieving error count"
        return 2
    fi
}

do_devstattest()
{
    device="$1"
    echo "[+] Starting SMART devstat test on ${device}..."
    output="$(smartctl -l devstat /dev/${device})"

    declare -A statdesc
    statdesc[uncorrectable_errors]='Number of Reported Uncorrectable Errors'
    statdesc[reallocated_sectors]='Number of Reallocated Logical Sectors'
    statdesc[mechanical_failures]='Number of Mechanical Start Failures'
    statdesc[cmd_resets]='Resets Between Cmd Acceptance and Completion'

    declare -A devstat
    devstat[uncorrectable_errors]=$(grep 'Number of Reported Uncorrectable Errors' <<< "$output" | awk '{print $4}')
    devstat[reallocated_sectors]=$(grep 'Number of Reallocated Logical Sectors' <<< "$output" | awk '{print $4}')
    devstat[mechanical_failures]=$(grep 'Number of Mechanical Start Failures' <<< "$output" | awk '{print $4}')
    devstat[cmd_resets]=$(grep 'Resets Between Cmd Acceptance and Completion' <<< "$output" | awk '{print $4}')

    status=0
    for stype in uncorrectable_errors reallocated_sectors mechanical_failures cmd_resets; do
        if [ -z "${devstat[$stype]}" ]; then
            echo "warning: no info for: ${statdesc[$stype]}"
        elif [ "${devstat[$stype]}" -ne 0 ]; then
            echo "error: ${statdesc[$stype]}: ${devstat[$stype]}"
            status=1
        fi
    done

    if [ "$status" -eq "0" ]; then
        echo "[+] Success"
    else
        echo "[-] Failure"
    fi
    return "$status"
}

if [ $# -lt 2 ]; then
    usage >&2
    exit 1
fi

SMARTTEST=0
BLOCKTEST=0
ERRORTEST=0
DEVSTATTEST=0

while true; do
    case "$1" in
        -s|--smart)
            SMARTTEST=1
            shift
        ;;
        -e|--smarterrors)
            ERRORTEST=1
            shift
        ;;
        -b|--badblocks)
            BLOCKTEST=1
            shift
        ;;
        -d|--devstat)
            DEVSTATTEST=1
            shift
        ;;
        -a|--all)
            SMARTTEST=1
            ERRORTEST=1
            BLOCKTEST=1
            DEVSTATTEST=1
            shift
        ;;
        --)
            shift
            break
        ;;
        -*)
            usage
            exit 1
        ;;
        *)
            break
        ;;
    esac
done

if [ $# -lt 1 ]; then
    usage >&2
    exit 1
fi

overall_status=0

for device in $@; do

    if [ ! -b "/dev/${device}" ]; then
        echo "[-] error: device $device not found or not a block device" >&2
        exit 1
    fi

    if [ "$BLOCKTEST" = "1" ]; then
        if ! command -v badblocks >/dev/null; then
            echo "[-] error: badblocks: command not found"
            exit 1
        fi
        do_blocktest "$device" || overall_status=1
    fi

    if [ "$SMARTTEST" = "1" ]; then
        if ! command -v smartctl >/dev/null; then
            echo "[-] error: smartctl: command not found"
            exit 1
        fi
        do_smarttest "$device" || overall_status=1
    fi

    if [ "$ERRORTEST" = "1" ]; then
        if ! command -v smartctl >/dev/null; then
            echo "[-] error: smartctl: command not found"
            exit 1
        fi
        do_smarterrorstest "$device" || overall_status=1
    fi

    if [ "$DEVSTATTEST" = "1" ]; then
        if ! command -v smartctl >/dev/null; then
            echo "[-] error: smartctl: command not found"
            exit 1
        fi
        do_devstattest "$device" || overall_status=1
    fi

done

if [ "$overall_status" -eq "0" ]; then
    echo "[+] Overall Result: Success"
else
    echo "[-] Overall Result: Failure"
    exit 1
fi