testing/ocfs2-tools: fix init script

3 files changed, 210 insertions, 85 deletions

diff --git a/testing/ocfs2-tools/APKBUILD b/testing/ocfs2-tools/APKBUILD
index 653918006a..34b613418c 100644
--- a/testing/ocfs2-tools/APKBUILD
+++ b/testing/ocfs2-tools/APKBUILD
@@ -2,7 +2,7 @@
 # Maintainer: Leonardo Arena <rnalrd@alpinelinux.org>
 pkgname=ocfs2-tools
 pkgver=1.4.4
-pkgrel=1
+pkgrel=2
 pkgdesc="Oracle Cluster File System 2 utilities"
 url="http://oss.oracle.com/projects/ocfs2-tools"
 arch="all"
@@ -51,12 +51,12 @@ package() {
 	make DESTDIR="$pkgdir" install
 
 	# remove the 2 lines below (and this) if there is no init.d script
-	install -m755 -D "$srcdir"/$pkgname.initd "$pkgdir"/etc/init.d/$pkgname
-	install -m644 -D "$srcdir"/$pkgname.confd "$pkgdir"/etc/conf.d/$pkgname
+	install -m755 -D "$srcdir"/$pkgname.initd "$pkgdir"/etc/init.d/o2cb
+	install -m644 -D "$srcdir"/$pkgname.confd "$pkgdir"/etc/conf.d/o2cb
 }
 
 md5sums="f7ae245e8baa499aa56d7af25a7885d5  ocfs2-tools-1.4.4.tar.gz
 5dba18ca5596b1de0f2eeb30f5ed1656  gcc45-ftbfs.patch
 c938040f0e77a4bb23e952c8f99b6cc5  build.patch
-5bc9de63e86fe8bd50c9c782e4a10076  ocfs2-tools.initd
-5c8e7b397fec3df784963eb55028cc8e  ocfs2-tools.confd"
+601fbd79acdc52a9046293aa977b1547  ocfs2-tools.initd
+d81b6ab068ec92a137b58c9ad56c4637  ocfs2-tools.confd"
diff --git a/testing/ocfs2-tools/ocfs2-tools.confd b/testing/ocfs2-tools/ocfs2-tools.confd
index 29b236dec6..29bc08406a 100644
--- a/testing/ocfs2-tools/ocfs2-tools.confd
+++ b/testing/ocfs2-tools/ocfs2-tools.confd
@@ -1,6 +1,38 @@
 # Copyright 1999-2006 Gentoo Foundation
 # Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/sys-fs/ocfs2-tools/files/Attic/ocfs2.conf,v 1.2 2009/10/08 09:03:35 flameeyes dead $
+# $Header: /var/cvsroot/gentoo-x86/sys-fs/ocfs2-tools/files/ocfs2.conf,v 1.1 2006/07/20 05:13:14 dberkholz Exp $
 
 # Put your cluster names here, separated by space, ie.
 # OCFS2_CLUSTER="cluster1 admincluster cluster2" 
+OCFS2_CLUSTER="ocfs2"
+
+# 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.
+OCFS2_IDLE_TIMEOUT_MS="30000"
+# How often we should attempt to send heartbeats.
+OCFS2_KEEPALIVE_DELAY_MS="2000"
+OCFS2_RECONNECT_DELAY_MS="2000"
+# How often we should attempt to send heartbeats.
+# How many interations before a node is considered dead from lack of IO activity.
+# (dead_threshold - 1) * 2s
+OCFS2_DEAD_THRESHOLD="31"
+
+# Default: "-fy" (force check on double-unmounted boot). I beleave this is best.
+# Native way is "-y", but once I have twice reboot/powerOFF with bad result...
+# Affected only fstab.
+OCFS2_FSCK="-fy"
+
+# http://oss.oracle.com/projects/ocfs2-tools/news/article_8.html
+# starting from 1.4.3, fsck.ocfs2 "aggressively cache the metadata blocks"
+# this is useful on lots of free memory, but I have heavy slowdown on x86_64
+# with 1G of RAM.
+# "yes" will "swapoff -a" & "swapon -a" around fsck to avoid caching over swap
+OCFS2_FSCK_SWAPOFF="yes"
+
+# Signal to kill processes on stop/umount on busy device (empty to not kill)
+#OCFS2_UMOUNT_KILL="KILL"
+
+# "yes" to force stop
+#OCFS2_FORCE_STOP="no"
diff --git a/testing/ocfs2-tools/ocfs2-tools.initd b/testing/ocfs2-tools/ocfs2-tools.initd
index 32abd47e48..677b498f88 100644
--- a/testing/ocfs2-tools/ocfs2-tools.initd
+++ b/testing/ocfs2-tools/ocfs2-tools.initd
@@ -1,100 +1,193 @@
 #!/sbin/runscript
-# Copyright 1999-2006 Gentoo Foundation
-# Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/sys-fs/ocfs2-tools/files/Attic/ocfs2.init,v 1.2 2009/10/08 09:03:35 flameeyes dead $
+
+: ${OCFS2_FSCK:="-fy"}
 
 depend() {
 	need net localmount
-	before netmount 
+	before netmount
+	after drbd
 }
 
-check_modules() {
-	local MODULES="ocfs2_dlmfs ocfs2 ocfs2_dlm ocfs2_nodemanager configfs"
-	local MODULE
-	local MODPROBE="modprobe -l"
-	local retval=0
-	
-	for MODULE in ${MODULES}; do
-		if [ -z "`${MODPROBE} ${MODULE}`" ] ; then
-			retval=1
-		fi
+pseudofs() {
+	[ -n "`mount -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 -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
-	if [ ${retval} -eq 1 ] ; then 
-		ewarn "One or more required modules are not loaded."
-                ewarn "Make sure you have "
-                ewarn " - placed ocfs, dlmfs and configfs into /etc/modules.autoload.d/kernel-2.6"
-                ewarn "For a (in)complete documentation, read /usr/share/doc/ocfs-<version>/INSTALL.GENTOO.gz"
-	fi
-	return ${retval}
+	return $err
 }
 
-check_pseudofs() { 
-	local retval=0
-	local HASMOUNT="mount -l -t"
-	if [ -z "`${HASMOUNT} configfs`" ] ; then
-		retval=1
-	fi
-	if [ -z "`${HASMOUNT} ocfs2_dlmfs`" ] ; then
-		retval=1
-	fi
+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
+}
 
-	if [ ${retval} -eq 1 ]; then
-		ewarn "One or more pseudo-filesystes are not mounted."
-		ewarn "Make sure you have following lines in your /etc/fstab:"
-		ewarn "none                    /config         configfs        defaults                0 0"
-		ewarn "none                    /dlm            ocfs2_dlmfs     defaults                0 0"
-		ewarn "For a (in)complete documentation, read /usr/share/doc/ocfs-<version>/INSTALL.GENTOO.gz"
-	fi
-	return ${retval}
+clusters(){
+	[[ -z "${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() {
-	check_modules || return $? 
-	check_pseudofs || return $?
-	
-	einfo "Starting OCFS2 cluster"
-	for cluster in ${OCFS2_CLUSTER}; do
-		ebegin " - ${cluster}"
-		/sbin/o2cb_ctl -H -n ${cluster} -t cluster -a online=yes >/dev/null 2>&1
-		eend $?
+	(	fmod /sys/fs/ocfs2 ocfs2 &&
+		fmod /sys/fs/ocfs2/cluster_stack ocfs2_stackglue &&
+		pseudofs configfs /sys/kernel/config &&
+		pseudofs ocfs2_dlmfs /dlm
+	) || return 1
+	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 $OCFS2_DEAD_THRESHOLD > /sys/kernel/config/cluster/${cluster}/heartbeat/dead_threshold
+	done
+
+	local i
+	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 $i
+
+	# Voluntary fsck. Will be happened only ondemand -
+	# on both unmounted/double fault, placed in fstab.
+	[[ "$OCFS2_FSCK_SWAPOFF" == "yes" ]] && swapoff -a
+	for i in `mount -invfat ocfs2 2>/dev/null|sed -e 's: .*::g'` ; do
+		[[ -e "$i" ]] || continue
+		einfo "OCFS2: Trying 'fsck.ocfs2 $OCFS2_FSCK $i' (fs check if possible)"
+		/sbin/fsck.ocfs2 $OCFS2_FSCK "$i" 2>/dev/null
+	done
+	[[ "$OCFS2_FSCK_SWAPOFF" == "yes" ]] && swapon -a
+
+	# Any behaviour
+	ebegin "Mounting OCFS2 filesystems"
+	mount -at ocfs2
+	eend $?
+	return $?
+}
+
+_fuser(){
+	fuser -"${OCFS2_UMOUNT_KILL}" -v"${OCFS2_UMOUNT_KILL:+k}" $*
+	return $?
+}
+
+_umount(){
+	local i m
+	i=`umount $* 2>&1` && {
+		echo -n "$i"
+		return 0
+	}
+	echo "$i"
+	_fuser -mM `echo "$i"|grep "^umount: .* device is busy\.\$"|sed -e 's%^umount: \([^ 	]*\): device is busy\.$%\1%g'` || return 1
+	[[ -z "${OCFS2_UMOUNT_KILL}" ]] && return 1
+	umount $* && return 0
+	sleep 1
+	umount $*
+	return $?
 }
 
 stop() {
-	# Shamelesly stolen from netmount
-        local ret
-        ebegin "Unmounting OCFS2 filesystems"
-        [ -z "$(umount -art ocfs2 2>&1)" ]
-        ret=$?
-        eend ${ret} "Failed to simply unmount filesystems"
-        [ ${ret} -eq 0 ] && return 0
-
-        declare -a siglist=( "TERM" "KILL" "KILL" )
-        local retry=0
-        local remaining="go"
-
-        while [ -n "${remaining}" -a ${retry} -lt 3 ]
-        do
-                remaining="$(awk '$3 ~ /'ocfs2'/ { if ($2 != "/") print $2 }' /proc/mounts | sort -r)"
-                IFS=$'\n'
-                set -- ${remaining//\\040/ }
-                unset IFS
-                [ -z "${remaining}" ] && break
-
-                ebegin $'\t'"Unmounting ocfs2 filesystems (retry #$((retry+1)))"
-                /bin/fuser -k -${siglist[$((retry++))]} -m "$@" &>/dev/null
-                sleep 5
-                umount "$@" &>/dev/null
-                eend $? $'\t'"Failed to unmount filesystems"
-        done
-
-	
-	einfo "Stopping OCFS2 cluster" 
-	for cluster in ${OCFS_CLUSTERS}; do
-		ebegin " - ${cluster}"
-		/sbin/o2cb_ctl -H -n ${cluster} -t cluster -a online=no >/dev/null 2>&1
-		eend $?
+	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 $m; then
+				einfon "Lazy unmounted. Waiting "
+				for i in 9 8 7 6 5 4 3 2 1 0; do
+					( /sbin/ocfs2_hb_ctl -I -d /dev/$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
 }