hw-test/scripts/wrapper-badblocks

#!/bin/bash

source test-variables

DISK=$1
if [[ -z $DISK ]] ; then
	echo "Missing DISK parameter"
	exit 11
fi

SERIAL=$2
if [[ -z $SERIAL ]] ; then
	echo "Missing SERIAL parameter"
	exit 12
fi

MY_PROC=$3
if [[ -z $MY_PROC ]] ; then
	echo "Missing parameter 2: MY_PROC"
	exit 16
fi

if [[ ! -b /dev/$DISK ]] ; then
	echo "Device /dev/$DISK is not a block special file"
	exit 13
fi

DISK_SERIAL=`smartctl -i /dev/$DISK \
	| grep -i 'serial number' \
	| cut -d ':' -f 2 \
	| sed 's/\s\+//'`

if [[ -z $DISK_SERIAL ]] ; then
	echo "Empty serial number in disk /dev/$DISK"
	exit 14
fi

if [[ $DISK_SERIAL != $SERIAL ]] ; then
	echo "Serial numbers $SERIAL and $DISK_SERIAL for $DISK do not match"
	exit 15
fi

#TPATTERN=0x55
TPATTERN=0xaa
# TODO: Debug limit. Comment out for production
#LST_BLCK=10000000
SMARTLOG=300 # check every 5 minutes
#SMARTLOG=60 # check every minute
# TODO: Debug variable. Comment out for production
#SMARTLOG=15 # Check every second
BLK_SIZE=`cat /sys/block/$DISK/queue/physical_block_size`
MAX_SAME=`cat /sys/block/$DISK/queue/write_same_max_bytes`
CONCU=$(( MAX_SAME / ( BLK_SIZE / 2 ) ))
[[ $CONCU == 0 ]] && CONCU=32768

BEGIN=`date +%s`
# Check SMART overall self test
ERROR=`smartctl -H /dev/$DISK \
	| grep '\(overall-health\|Health Status\)' \
	| grep -v '\(PASSED\|OK\)'
`
if [[ -n $ERROR ]] ; then
	smartctl -H /dev/$DISK \
		> $OUTPUT_DIR/disk/$SERIAL.smart.error
else
	ERROR=`smartctl -A -f brief /dev/$DISK \
		| parse-smart-attr \
		| grep ERROR \
		| cut -d ':' -f 2
	`
	# TODO: Debug variable. Comment out for production
	#ERROR=
	if [[ -n $ERROR ]] ; then
		RAW_ERROR=$ERROR
		ERROR=''
		for TOKEN in $RAW_ERROR ; do
			ERROR="$ERROR $TOKEN"
		done
		smartctl -A -f brief /dev/$DISK \
		| parse-smart-attr \
		| sed "s/ERROR/$RES_ERROR/g" \
			> $OUTPUT_DIR/disk/$SERIAL.smart.error
	fi
fi
if [[ -z $ERROR ]] ; then
	badblocks -b $BLK_SIZE -c $CONCU -e 1 -t $TPATTERN \
		-o $OUTPUT_DIR/disk/$SERIAL.badblocks \
		-svw /dev/$DISK $LST_BLCK \
		2> /run/$SERIAL.badblocks.out &
	TEST_PID=`jobs -p %+`
	echo $TEST_PID > /run/$SERIAL.badblocks.pid
	while [[ -d /proc/$TEST_PID ]] ; do
		# Check SMART error log
		ERROR=`smartctl -l error /dev/$DISK \
			| parse-smart-error \
			| sed 's/://g'
		`
		if [[ -n $ERROR ]] ; then
			smartctl -l error /dev/$DISK \
				> $OUTPUT_DIR/disk/$SERIAL.smart.error
			kill $TEST_PID
			sleep 5
		fi
		# TODO: Debug procedure. unnecessary for production
		#BADBLOCK=`cat $OUTPUT_DIR/disk/$SERIAL.badblocks`
		#if [[ -n $BADBLOCK ]] ; then
		#	kill $TEST_PID
		#	sleep 5
		#fi
		sleep $SMARTLOG
	done
	if [[ -z $ERROR ]] ; then
		# Check smart attributes after the badblocks test
		ERROR=`smartctl -A -f brief /dev/$DISK \
			| parse-smart-attr \
			| grep ERROR \
			| cut -d ':' -f 2
		`
		# TODO: Debug variable. Comment out for production
		#ERROR=
		if [[ -n $ERROR ]] ; then
			RAW_ERROR=$ERROR
			ERROR=''
			for TOKEN in $RAW_ERROR ; do
				ERROR="$ERROR $TOKEN"
			done
			smartctl -A -f brief /dev/$DISK \
			| parse-smart-attr \
			| sed "s/ERROR/$RES_ERROR/g" \
				> $OUTPUT_DIR/disk/$SERIAL.smart.error
		fi
	fi
	#RET=$?
	#echo $RET > $OUTPUT_DIR/disk/$SERIAL.badblocks.ret
	rm -f /run/$SERIAL.badblocks.pid
fi
END=`date +%s`

DURATION=$(( END - BEGIN ))
if [[ -f $OUTPUT_DIR/disk/$SERIAL.badblocks ]] ; then
	BADBLOCK=`cat $OUTPUT_DIR/disk/$SERIAL.badblocks`
fi

echo -n "${BLD}Process $MY_PROC${OFF} : " \
	> $OUTPUT_DIR/disk/$SERIAL.badblocks.result
if [[ -n $ERROR ]] ; then
	echo -n "Disk $DISK $SERIAL $ERROR : " \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.result
	echo $RES_FAIL \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.result	
	echo "$SERVER_SERIAL|$DISK|$SERIAL|FAIL|$ERROR" \
		> $OUTPUT_DIR/disk/$SERIAL.badblocks.csv
elif [[ -z $BADBLOCK ]] ; then
	echo -n "disk $DISK $SERIAL tested in ${DURATION}s : " \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.result
	echo $RES_OK \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.result
	echo "$SERVER_SERIAL|$DISK|$SERIAL|OK|${DURATION}s" \
		> $OUTPUT_DIR/disk/$SERIAL.badblocks.csv
else
	echo -n "Bad block $BADBLOCK found in disk $DISK $SERIAL : " \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.result
	echo $RES_FAIL \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.result
	echo "$SERVER_SERIAL|$DISK|$SERIAL|FAIL|Bad block $BADBLOCK" \
		> $OUTPUT_DIR/disk/$SERIAL.badblocks.csv
fi

if [[ -f $OUTPUT_DIR/disk/$SERIAL.badblocks ]] ; then
	ln -sf $OUTPUT_DIR/disk/$SERIAL.badblocks \
		$OUTPUT_DIR/disk/$DISK.badblocks
fi
if [[ -f /run/$SERIAL.badblocks.out ]] ; then
#	strings /run/$SERIAL.badblocks.out \
#		| sed 's/\s\+[0-9.%]\+ done, [0-9:]\+ elapsed.//' \
#		| uniq > $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	date > $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	strings /run/$SERIAL.badblocks.out \
		| grep -B5 -A5 Testing \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	echo "..." >> $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	strings /run/$SERIAL.badblocks.out \
		| grep -B5 -A5 Reading \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	echo "..." >> $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	strings /run/$SERIAL.badblocks.out \
		| grep -B5 -A5 completed \
		>> $OUTPUT_DIR/disk/$SERIAL.badblocks.out
	ln -sf $OUTPUT_DIR/disk/$SERIAL.badblocks.out \
		$OUTPUT_DIR/disk/$DISK.badblocks.out
#	rm -f /run/$SERIAL.badblocks.out
fi
#ln -sf $OUTPUT_DIR/disk/$SERIAL.badblocks.ret \
#	$OUTPUT_DIR/disk/$DISK.badblocks.ret
ln -sf $OUTPUT_DIR/disk/$SERIAL.badblocks.result \
	$OUTPUT_DIR/disk/$DISK.badblocks.result
ln -sf $OUTPUT_DIR/disk/$SERIAL.badblocks.csv \
	$OUTPUT_DIR/disk/$DISK.badblocks.csv
if [[ -f $OUTPUT_DIR/disk/$SERIAL.smart.error ]] ; then
	ln -sf $OUTPUT_DIR/disk/$SERIAL.smart.error \
		$OUTPUT_DIR/disk/$DISK.smart.error
fi
	
cat $OUTPUT_DIR/disk/$SERIAL.badblocks.result
cat $OUTPUT_DIR/disk/$SERIAL.badblocks.result \
	>> $OUTPUT_DIR/badblocks.log