#!/sbin/runscript

: ${OCFS2_FSCK:="-fy"}

depend() {
	need net localmount
	before netmount
	after drbd sshd
}

pseudofs() {
	[ -n "`mount -l -t $1`" ] && return 0
	ewarn "OCFS2: Pseudo-filesystem $1 are not mounted."
	ewarn "Make sure you have following lines in your /etc/fstab:"
	ewarn "none         $2         $1        defaults                0 0"

	# Why not?
	ebegin "Mounting $1"
	mount -t $1 none $2 && [ -n "`mount -l -t $1`" ]
	eend $? || return 1
}

fmod() {
	[[ -e "$1" ]] && return 0
	modprobe -s $2 && [[ -e "$1" ]] &&  return 0
	eerror "OCFS2: Module '$2' failed, '$1' not found"
	return 1
}

# unsure about possibility to keep heartbeat unclean after correct umount
# but Oracle do so
clean_heartbeat(){
	local err=0 id
	for i in "/sys/kernel/config/cluster/$1/heartbeat"/*; do
		[[ -d "$i" ]] || continue
		id="${i##*/}"
		ebegin "Cleaning OCFS2 heartbeat region $1/$id"
		[[ "`/sbin/ocfs2_hb_ctl -I -u "$id" | grep -o " [0-9]* refs$"`" == " 0 refs" ]] && /sbin/ocfs2_hb_ctl -K -u "$id"  || ! [[ -d "$i" ]]
		eend $? || err=1
	done
	return $err
}

UUID(){
	local dev fs stack uuid label
	/sbin/mounted.ocfs2 -d|while read dev fs stack uuid label; do
		[[ "$stack" == "o2cb" ]] && echo "$uuid"
	done|sort -u
}

clusters(){
	[[ -z "${OCFS2_CLUSTER}" ]] && for i in /sys/kernel/config/cluster/*; do
		OCFS2_CLUSTER="${OCFS2_CLUSTER} ${i##*/}"
	done
	echo "${OCFS2_CLUSTER}"
}

stop_cluster(){
	if clean_heartbeat $1 || [[ "${OCFS2_FORCE_STOP}" == yes ]]; then
		ebegin "Stopping OCFS2 cluster '$1'"
		/sbin/o2cb_ctl -H -n $1 -t cluster -a online=no >/dev/null
		eend $? || return 1
	else
		return 1
	fi
}

online(){
	local cluster
	for cluster in ${*:-$(clusters)}; do
		grep -q "^1\$" /sys/kernel/config/cluster/$cluster/node/*/local 2>/dev/null || return 1
	done
	return 0
}

start() {
	local i
	(	fmod /sys/fs/ocfs2 ocfs2 &&
		fmod /sys/fs/ocfs2/cluster_stack ocfs2_stackglue &&
		fmod /sys/kernel/dlm dlm &&
		pseudofs configfs /sys/kernel/config &&
		pseudofs ocfs2_dlmfs /sys/kernel/dlm
	) || return 1
	i=`mount -l -t ocfs2_dlmfs`
	if [ -z "${i##* /dlm *}" -a -n "${i##* /sys/kernel/dlm *}" ]; then
		mount -B /dlm /sys/kernel/dlm
#	elif [ -n "${i##* /dlm *}" -a -z "${i##* /sys/kernel/dlm *}" ]; then
#		mount -B /sys/kernel/dlm /dlm
	fi
	echo o2cb >/sys/fs/ocfs2/cluster_stack 2>/dev/null
	if [[ "`cat /sys/fs/ocfs2/cluster_stack`" != "o2cb" ]]; then
		eerror "OCFS2: Stack 'o2cb' not loaded. Check your kernel config."
		return 1
	fi

	# autodetect
	: ${OCFS2_CLUSTER:=$(UUID)}

	for cluster in $(clusters); do
		ebegin "Starting OCFS2 cluster '${cluster}'"
		online $cluster || /sbin/o2cb_ctl -H -n ${cluster} -t cluster -a online=yes >/dev/null
		if ! eend $? ; then
			local OCFS2_FORCE_STOP=no
			stop_cluster $cluster
			return 1
		fi

		# Some heartbeat tweaks to prevent self-fencing quite so much during heavy load.
		# http://oss.oracle.com/projects/ocfs2/dist/documentation/ocfs2_faq.html

		# How long to wait before a node is considered dead from lack of network activity.
		echo $OCFS2_IDLE_TIMEOUT_MS > /sys/kernel/config/cluster/${cluster}/idle_timeout_ms
		# How often we should attempt to send heartbeats.
		echo $OCFS2_KEEPALIVE_DELAY_MS > /sys/kernel/config/cluster/${cluster}/keepalive_delay_ms
		echo $OCFS2_RECONNECT_DELAY_MS > /sys/kernel/config/cluster/${cluster}/reconnect_delay_ms
		# How many interations before a node is considered dead from lack of IO activity.
		# (dead_threshold - 1) * 2s
		echo $OCFS2_DEAD_THRESHOLD|tee /sys/kernel/config/cluster/${cluster}/heartbeat/*threshold >/dev/null
	done

	for i in 9 8 7 6 5 4 3 2 1 0; do
		online && break
		echo -n "$i"
		sleep 1
	done
	# usure: IMHO locking not starts here
#	let i=OCFS2_RECONNECT_DELAY_MS*2/1000
#	sleep $i

	# Voluntary fsck. Will be happened only ondemand -
	# on both unmounted/double fault, placed in fstab.
	# mount -at broken in util-linux 2.22
	[[ "$OCFS2_FSCK_SWAPOFF" == "yes" ]] && swapoff -a
	local r=0
	sed -e 's:#.*$::g' </etc/fstab|while read d m t i; do
#		[ "$t" = auto ] # blkid ... broken too!
		[ "$t" = ocfs2 -a -e "$d" ] || continue
		einfo "OCFS2: Trying 'fsck.ocfs2 $OCFS2_FSCK $d' (fs check if possible)"
		/sbin/fsck.ocfs2 $OCFS2_FSCK "$d" 2>&1 |tee /var/log/ocfs2-fsck.log
	done
	[[ "$OCFS2_FSCK_SWAPOFF" == "yes" ]] && swapon -a
	if [[ "$OCFS2_NOMOUNT" != "yes" ]]; then
		ebegin "Mounting OCFS2 filesystems"
		mount -at ocfs2
		eend $?
	else
		true
	fi
}

_fuser(){
	fuser -"${OCFS2_UMOUNT_KILL}" -v"${OCFS2_UMOUNT_KILL:+k}" $*
	return $?
}

_umount(){
	local i m x=0
	i=`umount $* 2>&1` && {
		echo -n "$i"
		return 0
	}
	echo "$i"
	local p=`echo "$i"|grep "^umount: .* \(device\|target\) is busy\.\$"|sed -e 's%^umount: \([^ 	]*\): [^ ]* is busy\.$%\1%g'`
	while [ -n "$p" -a "$x" -lt 3 ]; do
		_fuser -mM "$p" || return 1
		[[ -z "${OCFS2_UMOUNT_KILL}" ]] && break
		umount $* && return 0
		sleep 1
		umount $* && return 0
		let x=[x+1]
	done
	return 1
}

stop() {
	local ret=0 dev m i
	# now umount only clusters - under heartbeat. others later
#	ebegin "Unmounting OCFS2 filesystems"
#	_umount -at ocfs2
#	eend $?
#	ret=$?

	for cluster in $(clusters); do
		for dev in `cat /sys/kernel/config/cluster/$cluster/heartbeat/*/dev 2>/dev/null`; do
			m=`grep "/dev/$dev " /proc/mounts|sed -e 's:^[^ ]* \([^ ]*\) .*$:\1:g'`
			[[ -z "$m" ]] && continue
			ebegin "Unmounting OCFS2 cluster '$cluster' filesystems '$dev' from '$m'"
			if _umount $m; then
				eend 0
				continue
			elif umount -l $m; then
				einfon "Lazy unmounted. Waiting "
				for i in 9 8 7 6 5 4 3 2 1 0; do
					( /sbin/ocfs2_hb_ctl -I -d /dev/$dev || break ) | grep -q " 0 refs\$" && break
					echo -n "$i"
					_fuser $m
					sleep 1
				done
				echo
				[[ "$i" != 0 ]] && continue
			fi
			eend 1
			ret=1
		done
		stop_cluster $cluster || ret=1
	done
	return $ret
}