A little dialog wrapper for smartctl

check_disk.sh

#!/bin/bash

smartctl="${SMARTCTL:-smartctl}"
dialog="${DIALOG:-dialog}"
jq="${JQ:-jq}"
# Debian workaround
blockdev="${BLOCKDEV:-$(PATH="$PATH:/sbin:/usr/sbin" which blockdev)}"

sce_parse_bit=1
sce_open_bit=2
sce_cmd_bit=4
sce_fail_bit=8
sce_badattr_bit=16
sce_histfail_bit=32
sce_errlog_bit=64
sce_stfail_bit=128

poll_interval=10  #sec
temp_file=/tmp/smartout

dle_extra=3

ui() {
	$dialog --backtitle check_disk --colors "$@" 3>&1 1>&2 2>&3 3>&-
}

errui() {
	DIALOGRC=<(echo "screen_color=(RED,RED,ON)") ui "$@"
}

warnui() {
	DIALOGRC=<(echo "screen_color=(YELLOW,YELLOW,ON)") ui "$@"
}

dangui() {
	DIALOGRC=<(echo "screen_color=(MAGENTA,MAGENTA,ON)") ui "$@"
}

testui() {
	DIALOGRC=<(echo "screen_color=(GREEN,GREEN,ON)") ui "$@"
}

progrui() {
	DIALOGRC=<(echo "screen_color=(CYAN,CYAN,ON)") ui "$@"
}

flash() {
	local times="$1"
	local wait="$2"
	shift 2

	for i in $(seq 1 $times); do
		local color="RED"
		if (( i % 2 == 0 )); then color="MAGENTA"; fi
		DIALOGRC=<(echo "screen_color=($color,$color,ON)") ui "$@"
		sleep "$wait"
	done
}

exec_test() {
	local dev="$1"
	local test="$2"
	local esttime="$3"

	while true; do
		local out
		out="$($smartctl $dev -t "$test" 2>&1)"
		local ret="$?"
		if (( ret & sce_cmd_bit )); then
			local choice
			if ! choice=$(warnui --title "$dev, $test test: Possibly Ongoing Test" --no-cancel --menu \
				"Failed to submit the command to begin the test (code $ret).\n\n"\
"Ouput: $out\n\n"\
"This could be due to an ongoing test. Would you like to:" \
				0 0 10 \
				abort "Abort the current test and try again?" \
				status "Monitor the status of the ongoing test?" \
				cancel "Cancel this test and return to the menu?" \
			); then
				choice="cancel"
			fi
			case "$choice" in
				abort)
					if ! $smartctl $dev -X; then
						warnui --title "$dev, $test test: Abort Failed" --msgbox "Abort failed! (The test may not be cancellable.)" 0 0
					fi
					continue
					;;
				status)
					break  # from the while
					;;
				cancel)
					return 3
					;;
			esac
		elif (( ret )); then
			errui --title "$dev, $test test: Failed to Start" --msgbox "Failed to start test of $dev (exit code $ret)" 0 0
			return 2
		fi
		break
	done

	local start="$(date +"%s")"
	while true; do
		local stat="$($jq -c '.ata_smart_data.self_test.status' <<< "$($smartctl $dev -j -c)")"
		case "$($jq '.passed' <<< "$stat")" in
			true)
				testui --title "$dev, $test test: Passed" --timeout 5 --msgbox "$test test of $dev passed!" 0 0
				return 0
				;;
			false)
				errui --title "$dev, $test test: Failed" --timeout 5 --msgbox "$test test of $dev failed!" 0 0
				return 1;
				;;
			null)	;;
			*)
				echo "unexpected stat: $stat"
				exit 1
				;;
		esac
		local rempct="$($jq '.remaining_percent' <<< "$stat")"
		local string="$($jq '.string' <<< "$stat")"
		local now="$(date +"%s")"
		string="$string\n$(( (now - start) / 60 )) min $(( (now - start) % 60 )) sec elapsed / $esttime min estimated"
		if [ "$rempct" = "null" ]; then
			warnui --title "$dev, $test test: Progress Unknown" --infobox "$string" 10 -1
		else
			progrui --title "$dev, $test test: Ongoing" --gauge "$string" 10 -1 $(( 100 - rempct )) < /dev/null
		fi
		sleep "$poll_interval"
	done
}

show_info() {
	local info="$1"
	shift 1

	cat > "$temp_file" <<< "$info"
	ui --hline "HOME | END | PGUP | PGDN | Arrow keys | / to search" "$@" --exit-label Continue --textbox "$temp_file" -1 -1
}

classify_smartctl_error() {
	local stat="$1"
	local out="$2"
	local continue="$3"
	local before="$4"

	if (( stat & (sce_parse_bit | sce_open_bit) )); then
		local oncemsg=""
		if [ -n "$before" ]; then
			oncemsg=$'\n\n'"This device worked before! Check your hardware (e.g. dock),"$'\n'"and perhaps reset or reconnect the device."
		fi
		local msg="$(printf "\\Z1smartctl could not open device (exit code %d)%s\\Zn\n\n%s" "$stat" "$oncemsg" "$out")"
		errui --title "$dev: Open Failed" --msgbox "$msg" 25 75
		return 1
	fi
	local failmsg=() faillvl=()
	if (( stat & sce_fail_bit )); then faillvl+=( 2 ); failmsg+=( "- Bit 3: Device reports it is failing NOW" ); fi
	if (( stat & sce_badattr_bit )); then faillvl+=( 2 ); failmsg+=( "- Bit 4: Device has an attribute below threshold (indicating imminent failure)" ); fi
	if (( stat & sce_histfail_bit )); then faillvl+=( 1 ); failmsg+=( "- Bit 5: Device has detected an attribute previously below threshold (indicating failure is likely)" ); fi
	if (( stat & sce_errlog_bit )); then faillvl+=( 2 ); failmsg+=( "- Bit 6: The error log is not empty" ); fi
	if (( stat & sce_stfail_bit )); then faillvl+=( 2 ); failmsg+=( "- Bit 7: Device failed to self test" ); fi
	local failmax=0
	for i in "${faillvl[@]}"; do if (( i > failmax )); then failmax=$i; fi; done
	local failcmd="" failcolor=0 failtitle="" failbutton=""
	case "$failmax" in
		0) ;;
		1) failcmd="warnui"; failcolor=3; failtitle="Caution"; failbutton="ok" ;;
		*) failcmd="errui"; failcolor=1; failtitle="Warning"; failbutton="cancel" ;;
	esac
	if [ -n "$failcmd" ]; then
		local msgs=""
		for msg in "${failmsg[@]}"; do msgs="${msgs}$msg\n\n"; done
		while true; do
			# Bug with labels when using extra with yes/no?
			if [ -n "$continue" ]; then
				$failcmd --title "$dev: $failtitle" --extra-button --extra-label "Details" \
					--cancel-label "Yes" --ok-label "No" --default-button "$failbutton" \
					--yesno "\Z${failcolor}This device reports failures:\n\n${msgs}\ZnDo you want to abort tests?" 25 75
				stat="$?"
			else
				$failcmd --title "$dev: $failtitle" --extra-button --extra-label "Details" --default-button extra \
					--msgbox "This device reports failures:\n\n${msgs}" 25 75
				stat="$?"
			fi
			if (( stat == dle_extra )); then
				show_info "$out"
				continue
			elif (( stat )); then
				return 1
			fi
			break
		done
	fi
}

run_tests() {
	local dev="$1"
	local once=""
	local out stat  # Predeclared, otherwise local eats the $? from the substitution
	while true; do
		out="$($smartctl $dev -a 2>&1)"
		stat="$?"
		if ! classify_smartctl_error "$stat" "$out" "1" "$once"; then
			return 1
		fi
		once="1"
		local tests
		declare -A testtimes=()
		local timedata="$($jq -c '.ata_smart_data.self_test.polling_minutes' <<< "$($smartctl $dev -j -c)")"
		for testkind in short extended conveyance; do
			local tname="$testkind"
			if [ "$tname" = "extended" ]; then tname="long"; fi
			testtimes["$tname"]="$($jq -r ".$testkind" <<< "$timedata")"
		done
		tests=( $(testui --title "$dev" --extra-button --extra-label "Details" --cancel-label "Back" \
			--checklist "Select tests to perform. (Times given are estimates by the target device only.)" 0 0 0 \
			short "Short (${testtimes[short]} min)" on \
			long "Long / Extended (${testtimes[long]} min)" off \
			conveyance "Conveyance (${testtimes[conveyance]} min)" off \
		) )
		stat="$?"
		if (( stat == dle_extra )); then
			cat > "$temp_file" <<< "$out"
			ui --title "$dev" --exit-label Continue --textbox "$temp_file" -1 -1
			continue
		elif (( stat )); then
			return
		fi
		local i=0
		local aborted=""
		while (( i < ${#tests[@]} )); do
			local res
			exec_test "$dev" "${tests[$i]}" "${testtimes[${tests[$i]}]}"
			res="$?"
			if (( res )); then
				case "$res" in
					1)
						errui --title "$dev Self-Test Failed" --msgbox "$dev failed a self-test; no further tests will be attempted." 0 0
						;;
					*)
						warnui --title "$dev Self-Test Cancelled" --msgbox "A test for $dev was cancelled or failed for an internal error (code $res)." 0 0
						;;
				esac
				aborted="1"
				break
			fi
			i=$(( i + 1 ))
		done
		if [ -z "$aborted" ]; then
			testui --title "$dev: All Tests Run" --msgbox "All scheduled tests for $dev were successfully run." 0 0
		fi
	done
}

zero_warning="\Zb\ZuEVEN IF CANCELED,\Zn zeroing a disk is an essentially \Zb\ZuIRRECOVERABLE OPERATION.\Zn\n\n\
The partition table and filesystem administrative structure, usually at the beginning of the disk, will be the first to be destroyed. Recovering even an aborted zero usually requires forensic methods.\n\n\
\Z1Do \Zb\ZuNOT\Zn\Z1 proceed unless you are sure you want to erase \Zb\ZuALL\Zn\Z1 data on this device. This cannot be undone."

try_zero_disk() {
	flash 5 0.2 --title "ZERO $dev" --infobox "$zero_warning" 25 75
	if dangui --title "ZERO $dev" --yes-label "Erase Disk" --no-label "Cancel" --default-button no \
		--yesno "$zero_warning" 25 75; then
		if dangui --title "ZERO $dev: Confirm" --yes-label "Yes, Erase Disk" --no-label "Cancel" --default-button no \
			--yesno "$zero_warning\n\n\Z5\Zb\ZuAre you really sure?" 25 75; then
			really_zero_disk "$dev"
		fi
	fi
}

mem_info() {
	grep "$1" /proc/meminfo | while read nm val rest; do echo "$val"; done
}

really_zero_disk() {
	local dev="$1"
	local total="$($blockdev --getsize64 "$dev")"
	local totalhr=""
	if [ -n "$total" ]; then totalhr="$(numfmt --to=iec --format='%.1f' <<< "$total")"; fi

	coproc dd if=/dev/zero of="$dev" bs=4M 2>&1
	echo
	ps

	local progmon
	while true; do
		kill -USR1 "$COPROC_PID"
		sleep 0.25
	done &
	progmon="$!"

	while read bytes rest; do
		if [[ $bytes =~ .*\+.* ]]; then continue; fi
		if [ -z "$bytes" ]; then break; fi
		local hr="$(numfmt --to=iec --format='%.1f' <<< "$bytes")"
		local pc="$(bc <<< "100 * $bytes / $total")"
		if [ -z "$total" ]; then
			dangui --title "$dev: Zeroing" --infobox "$bytes bytes ($hr) written, progress unknown" 10 -1
		else
			dangui --title "$dev: Zeroing" --gauge "$bytes/$total bytes ($hr/$totalhr) written" 10 -1 "$pc" < /dev/null
		fi
	done <&"${COPROC[0]}"

	kill "$progmon"

	local memtot="$(mem_info MemTotal)"

	local dots=1
	local suffix
	while true; do
		if kill -0 "$COPROC_PID"; then
			case dots in
				1) suffix="."; dots=2 ;;
				2) suffix=".."; dots=3 ;;
				3) suffix="..."; dots=1 ;;
				*) dots=1 ;;
			esac
			local membuf="$(mem_info Buffers)"
			dangui --title "$dev: Waiting for Buffers" --gauge "Waiting for buffers to clear (this may take a while)$suffix\n\n$membuf/$memtot buffers/total memory" 10 -1 $(( 100 - 100 * membuf / memtot ))
			sleep 1
		else
			break
		fi
	done

	dots=1
	while true; do
		suffix=""
		case dots in
			1) suffix="."; dots=2 ;;
			2) suffix=".."; dots=3 ;;
			3) suffix="..."; dots=1 ;;
			*) dots=1 ;;
		esac
		dangui --title "$dev: Syncing" --infobox "Syncing (this may take a moment)$suffix"
		sleep 1
	done &
	progmon="$!"

	sync
	kill "$progmon"

	dangui --title "$dev: Zeroed" --msgbox "$dev has been zeroed." 0 0
}

last_dmesg_check=""
check_dmesg() {
	local buffer

	if [ -z "$last_dmesg_check"]; then
		while read line; do
			buffer+=( "$line" )
		done <<< "$(dmesg)"
	else
		local time
		while read line; do
			time="${line#[}"
			time="${line%%]*}"
			time="${line/.*/}"
			if (( time >= last_dmesg_check )); then
				buffer+=( "$line" )
			fi
		done <<< "$(dmesg)"
	fi
	last_dmesg_check="$(cat /proc/uptime)"
	last_dmesg_check="${last_dmesg_check##.*}"

	local problems=()
	for line in "${buffer[@]}"; do
		if [[ $line =~ .*Buffer\ I/O\ error\ on\ dev\ ([^,]*),.* ]]; then
			problems+=( "${BASH_REMATCH[1]}" )
		fi
	done

	for dev in "${problems[@]}"; do echo "$dev"; done | sort | uniq
}

partitions() {
	local dev="$1"

	echo p | fdisk "$dev" 2>&1
}

device_menu() {
	local dev="$1"
	local choice

	if ! partitions "$dev"; then
		while true; do
			warnui --extra-button --extra-label "Dmesg" \
				--ok-label "Yes" --cancel-label "No" --default-button no \
				--title "$dev: Input/Output Errors" \
				--yesno "The device $dev appears to be experiencing I/O errors, which may make any further tests unreliable. More information may be available in the kernel debug message ('dmesg') logs.\n\nDo you want to continue?" \
				25 75
			local stat="$?"
			if (( stat == dle_extra )); then
				show_info "$(dmesg --time-format iso)" --title "Dmesg"
				continue
			elif (( stat )); then
				return
			fi
			break
		done
	fi

	while true; do
		if ! choice="$(testui --title "$dev" --cancel-label "Back" \
			--menu "Choose an action for $dev:" 0 0 10 \
			"test" 'Test the disk' \
			"stat" 'Read the SMART status and details' \
			"part" 'Show the partition table' \
			"info" 'Show detailed drive information' \
			"zero" '\Z1\ZbZero the disk\Zn' \
			"erase" '\Z1\ZbSecure erase or sanitize the disk\Zn' \
		)"; then
			return
		fi

		case "$choice" in
			test)
				run_tests "$dev"
				;;

			stat)
				local stat out
				out="$($smartctl $dev -a)"
				stat="$?"
				if (( stat )); then
					classify_smartctl_error "$stat" "$out"
				else
					show_info "$out"
				fi
				;;

			part)
				show_info "$(partitions "$dev")" --title "$dev: Partitions"
				;;

			info)
				show_info "$(hdparm -I "$dev")" --title "$dev: Info"
				;;

			zero)
				try_zero_disk "$dev"
				;;

			erase)
				errui --title "TODO" --timeout 5 --msgbox "Not yet implemented" 0 0
				;;
		esac
	done
}

enum_devices() {
	declare -A devices
	while read line; do
		name="$($jq -r .name <<< "$line")"
		local devnode="/dev/$name"
		local sz
		if ! sz="$($blockdev --getsize64 /dev/$name)"; then
			extra='\Z1NOMEDIA\Zn'
		else
			extra='\Z5'"$(numfmt --to=iec --format='%.1f' <<< "$sz")"'\Zn'
		fi
		# Superfluous echo to trim ws
		extra="$extra "'\Z4'"$(echo $(cat /sys/class/block/$name/device/model))"'\Zn'
		local info
		if info="$($smartctl -i /dev/$name)"; then
			extra="$extra "'\Z6'"SN:$(grep Serial <<< "$info" | tr -d ' ' | cut -d: -f2)"'\Zn'
		else
			extra="$extra "'\Z6'"SN:"'\Z1\Zb'"???"'\Zn'
		fi
		if [ "$($jq '[.children?[]?.mountpoint? != null] | any' <<< "$line")" = "true" ]; then
			extra="$extra "'\Z3\Zb(mounted)\Zn'
		fi
		if ! [ "$($jq -r '.type' <<< "$line")" = "disk" ]; then
			extra="$extra "'\Z1\Zb(not a disk)\Zn'
		fi
		devices["/dev/$name"]="$extra"
	done <<< "$(lsblk -J | $jq -c '.blockdevices[]')"
	declare -p devices
}

main() {
	if [ -z "$CD_NO_ROOT" ] && (( EUID )); then
		errui --title "check_disk: Not Root" --msgbox \
			"This script is not running as root!\n\n"\
"Try using 'sudo' or equivalent.\n\n"\
"If you know what you're doing and can assure this script has sufficient privileges, you may set CD_NO_ROOT to a non-empty value." \
			0 0
		exit 1
	fi
	while true; do
		ui --title "check_disk" --infobox "Scanning devices, please wait..." 10 75
		eval "$(enum_devices)"
		local args=()
		local sorted=( $(for key in "${!devices[@]}"; do echo "$key"; done | sort) )
		for dev in "${sorted[@]}"; do
			args+=( "$dev" "${devices[$dev]}" )
		done
		local device
		if ! device=$(ui --title "check_disk" --no-cancel --menu \
			"Choose the device to analyze." \
			0 0 10 \
			"${args[@]}" \
			"refresh" '\Z2Refresh this list' \
			"exit" '\Z2Exit'
		); then break; fi
		if [ "$device" = "refresh" ]; then continue; fi
		if [ "$device" = "exit" ]; then break; fi
		device_menu "$device"
	done
}

main