X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ftest-framework.sh;h=0292bc2dc3ddc9b6883ec5432e3205bc1404b03a;hp=346332723ae269027bcd49f837228b535f9b66f9;hb=8ce48257490f20ec7f11929a42ab3891139a9d8c;hpb=77601ac36b81ab7c659b98c3613367482a31ecad

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 3463327..0292bc2 100644
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -5,6 +5,7 @@ trap 'print_summary && touch $TF_FAIL && \
 set -e
 #set -x
 
+export LANG=en_US
 export EJOURNAL=${EJOURNAL:-""}
 export REFORMAT=${REFORMAT:-""}
 export WRITECONF=${WRITECONF:-""}
@@ -15,8 +16,8 @@ export GSS_KRB5=false
 export GSS_PIPEFS=false
 export IDENTITY_UPCALL=default
 export QUOTA_AUTO=1
-export JOBSTATS_AUTO=${JOBSTATS_AUTO:-1}
-export JOBID_VAR=${JOBID_VAR:-"procname_uid"}
+# specify environment variable containing batch job name for server statistics
+export JOBID_VAR=${JOBID_VAR:-"procname_uid"}  # or "existing" or "disable"
 
 # LOAD_LLOOP: LU-409: only load llite_lloop module if kernel < 2.6.32 or
 #             LOAD_LLOOP is true. LOAD_LLOOP is false by default.
@@ -178,6 +179,15 @@ init_test_env() {
         fi
     fi
 
+	export RESIZE2FS=$RESIZE2FS
+	if [ -z "$RESIZE2FS" ]; then
+		if which resizefs.ldiskfs >/dev/null 2>&1; then
+			export RESIZE2FS=resizefs.ldiskfs
+		else
+			export RESIZE2FS=resize2fs
+		fi
+	fi
+
     export LFSCK_BIN=${LFSCK_BIN:-lfsck}
     export LFSCK_ALWAYS=${LFSCK_ALWAYS:-"no"} # check fs after each test suite
     export FSCK_MAX_ERR=4   # File system errors left uncorrected
@@ -185,6 +195,7 @@ init_test_env() {
 	export ZFS=${ZFS:-zfs}
 	export ZPOOL=${ZPOOL:-zpool}
 	export ZDB=${ZDB:-zdb}
+	export PARTPROBE=${PARTPROBE:-partprobe}
 
     #[ -d /r ] && export ROOT=${ROOT:-/r}
     export TMP=${TMP:-$ROOT/tmp}
@@ -290,13 +301,15 @@ init_test_env() {
     export RLUSTRE=${RLUSTRE:-$LUSTRE}
     export RPWD=${RPWD:-$PWD}
     export I_MOUNTED=${I_MOUNTED:-"no"}
-    if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mdt.ko -a \
-        ! -f /lib/modules/$(uname -r)/updates/kernel/fs/lustre/mdt.ko -a \
-        ! -f `dirname $0`/../mdt/mdt.ko ]; then
-        export CLIENTMODSONLY=yes
-    fi
+	if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mdt.ko -a \
+	     ! -f /lib/modules/$(uname -r)/updates/kernel/fs/lustre/mdt.ko -a \
+	     ! -f /lib/modules/$(uname -r)/extra/kernel/fs/lustre/mdt.ko -a \
+	     ! -f $LUSTRE/mdt/mdt.ko ]; then
+	    export CLIENTMODSONLY=yes
+	fi
 
-    export SHUTDOWN_ATTEMPTS=${SHUTDOWN_ATTEMPTS:-3}
+	export SHUTDOWN_ATTEMPTS=${SHUTDOWN_ATTEMPTS:-3}
+	export OSD_TRACK_DECLARES_LBUG=${OSD_TRACK_DECLARES_LBUG:-"yes"}
 
     # command line
 
@@ -473,7 +486,6 @@ load_modules_local() {
     load_module ../lnet/lnet/lnet
     LNETLND=${LNETLND:-"socklnd/ksocklnd"}
     load_module ../lnet/klnds/$LNETLND
-    load_module lvfs/lvfs
     load_module obdclass/obdclass
     load_module ptlrpc/ptlrpc
     load_module ptlrpc/gss/ptlrpc_gss
@@ -490,24 +502,21 @@ load_modules_local() {
         grep -q crc16 $SYMLIST || { modprobe crc16 2>/dev/null || true; }
         grep -q -w jbd $SYMLIST || { modprobe jbd 2>/dev/null || true; }
         grep -q -w jbd2 $SYMLIST || { modprobe jbd2 2>/dev/null || true; }
+		load_module lfsck/lfsck
 		[ "$LQUOTA" != "no" ] && load_module quota/lquota $LQUOTAOPTS
 		if [[ $(node_fstypes $HOSTNAME) == *zfs* ]]; then
 			modprobe zfs
 			load_module osd-zfs/osd_zfs
 		fi
-		load_module mgs/mgs
-		load_module mdd/mdd
 		if [[ $(node_fstypes $HOSTNAME) == *ldiskfs* ]]; then
-			#
-			# This block shall be moved up beside osd-zfs as soon
-			# as osd-ldiskfs stops using mdd symbols.
-			#
 			grep -q exportfs_decode_fh $SYMLIST ||
 				{ modprobe exportfs 2> /dev/null || true; }
-			load_module ../ldiskfs/ldiskfs/ldiskfs
-			load_module lvfs/fsfilt_ldiskfs
+			load_module ../ldiskfs/ldiskfs
 			load_module osd-ldiskfs/osd_ldiskfs
 		fi
+		load_module nodemap/nodemap
+		load_module mgs/mgs
+		load_module mdd/mdd
 		load_module mdt/mdt
 		load_module ost/ost
 		load_module lod/lod
@@ -591,7 +600,7 @@ fs_log_size() {
 	local size=0
 	case $fstype in
 		ldiskfs) size=50;; # largest seen is 44, leave some headroom
-		zfs)     size=256;;
+		zfs)     size=400;; # largest seen is 384
 	esac
 
 	echo -n $size
@@ -709,6 +718,13 @@ cleanup_gss() {
     fi
 }
 
+facet_svc() {
+	local facet=$1
+	local var=${facet}_svc
+
+	echo -n ${!var}
+}
+
 facet_type() {
 	local facet=$1
 
@@ -808,6 +824,183 @@ ostdevlabel() {
 }
 
 #
+# Get the device of a facet.
+#
+facet_device() {
+	local facet=$1
+	local device
+
+	case $facet in
+		mgs) device=$(mgsdevname) ;;
+		mds*) device=$(mdsdevname $(facet_number $facet)) ;;
+		ost*) device=$(ostdevname $(facet_number $facet)) ;;
+		fs2mds) device=$(mdsdevname 1_2) ;;
+		fs2ost) device=$(ostdevname 1_2) ;;
+		fs3ost) device=$(ostdevname 2_2) ;;
+		*) ;;
+	esac
+
+	echo -n $device
+}
+
+#
+# Get the virtual device of a facet.
+#
+facet_vdevice() {
+	local facet=$1
+	local device
+
+	case $facet in
+		mgs) device=$(mgsvdevname) ;;
+		mds*) device=$(mdsvdevname $(facet_number $facet)) ;;
+		ost*) device=$(ostvdevname $(facet_number $facet)) ;;
+		fs2mds) device=$(mdsvdevname 1_2) ;;
+		fs2ost) device=$(ostvdevname 1_2) ;;
+		fs3ost) device=$(ostvdevname 2_2) ;;
+		*) ;;
+	esac
+
+	echo -n $device
+}
+
+#
+# Re-read the partition table on failover partner host.
+# After a ZFS storage pool is created on a shared device, the partition table
+# on the device may change. However, the operating system on the failover
+# host may not notice the change automatically. Without the up-to-date partition
+# block devices, 'zpool import ..' cannot find the labels, whose positions are
+# relative to partition rather than disk beginnings.
+#
+# This function performs partprobe on the failover host to make it re-read the
+# partition table.
+#
+refresh_partition_table() {
+	local facet=$1
+	local device=$2
+	local host
+
+	host=$(facet_passive_host $facet)
+	if [[ -n "$host" ]]; then
+		do_node $host "$PARTPROBE $device"
+	fi
+}
+
+#
+# Get ZFS storage pool name.
+#
+zpool_name() {
+	local facet=$1
+	local device
+	local poolname
+
+	device=$(facet_device $facet)
+	# poolname is string before "/"
+	poolname="${device%%/*}"
+
+	echo -n $poolname
+}
+
+#
+# Create ZFS storage pool.
+#
+create_zpool() {
+	local facet=$1
+	local poolname=$2
+	local vdev=$3
+	shift 3
+	local opts=${@:-"-o cachefile=none"}
+
+	do_facet $facet "$ZPOOL list -H $poolname >/dev/null 2>&1 ||
+		$ZPOOL create -f $opts $poolname $vdev"
+}
+
+#
+# Create ZFS file system.
+#
+create_zfs() {
+	local facet=$1
+	local dataset=$2
+	shift 2
+	local opts=${@:-"-o mountpoint=legacy"}
+
+	do_facet $facet "$ZFS list -H $dataset >/dev/null 2>&1 ||
+		$ZFS create $opts $dataset"
+}
+
+#
+# Export ZFS storage pool.
+# Before exporting the pool, all datasets within the pool should be unmounted.
+#
+export_zpool() {
+	local facet=$1
+	shift
+	local opts="$@"
+	local poolname
+
+	poolname=$(zpool_name $facet)
+
+	if [[ -n "$poolname" ]]; then
+		do_facet $facet "! $ZPOOL list -H $poolname >/dev/null 2>&1 ||
+			grep -q ^$poolname/ /proc/mounts ||
+			$ZPOOL export $opts $poolname"
+	fi
+}
+
+#
+# Destroy ZFS storage pool.
+# Destroy the given pool and free up any devices for other use. This command
+# tries to unmount any active datasets before destroying the pool.
+# -f    Force any active datasets contained within the pool to be unmounted.
+#
+destroy_zpool() {
+	local facet=$1
+	local poolname=${2:-$(zpool_name $facet)}
+
+	if [[ -n "$poolname" ]]; then
+		do_facet $facet "! $ZPOOL list -H $poolname >/dev/null 2>&1 ||
+			$ZPOOL destroy -f $poolname"
+	fi
+}
+
+#
+# Import ZFS storage pool.
+# Force importing, even if the pool appears to be potentially active.
+#
+import_zpool() {
+	local facet=$1
+	shift
+	local opts=${@:-"-o cachefile=none"}
+	local poolname
+
+	poolname=$(zpool_name $facet)
+
+	if [[ -n "$poolname" ]]; then
+		opts+=" -d $(dirname $(facet_vdevice $facet))"
+		do_facet $facet "$ZPOOL list -H $poolname >/dev/null 2>&1 ||
+			$ZPOOL import -f $opts $poolname"
+	fi
+}
+
+#
+# Set the "cachefile=none" property on ZFS storage pool so that the pool
+# is not automatically imported on system startup.
+#
+# In a failover environment, this will provide resource level fencing which
+# will ensure that the same ZFS storage pool will not be imported concurrently
+# on different nodes.
+#
+disable_zpool_cache() {
+	local facet=$1
+	local poolname
+
+	poolname=$(zpool_name $facet)
+
+	if [[ -n "$poolname" ]]; then
+		do_facet $facet "$ZPOOL set cachefile=none $poolname"
+	fi
+}
+
+#
 # This and set_osd_param() shall be used to access OSD parameters
 # once existed under "obdfilter":
 #
@@ -935,6 +1128,11 @@ mount_facet() {
 		opts=$(csa_add "$opts" -o loop)
 	fi
 
+	if [[ $(facet_fstype $facet) == zfs ]]; then
+		# import ZFS storage pool
+		import_zpool $facet || return ${PIPESTATUS[0]}
+	fi
+
 	echo "Starting ${facet}: $opts ${!dev} $mntpt"
 	# for testing LU-482 error handling in mount_facets() and test_0a()
 	if [ -f $TMP/test-lu482-trigger ]; then
@@ -981,39 +1179,6 @@ start() {
     return $RC
 }
 
-#
-# When a ZFS OSD is made read-only by replay_barrier(), its pool is "freezed".
-# Because stopping corresponding target may not clear this in-memory state, we
-# need to zap the pool from memory by exporting and reimporting the pool.
-#
-# Although the uberblocks are not updated when a pool is freezed, transactions
-# are still written to the disks.  Modified blocks may be cached in memory when
-# tests try reading them back.  The export-and-reimport process also evicts any
-# cached pool data from memory to provide the correct "data loss" semantics.
-#
-refresh_disk() {
-	local facet=$1
-	local fstype=$(facet_fstype $facet)
-	local _dev
-	local dev
-	local poolname
-
-	if [ "${fstype}" == "zfs" ]; then
-		_dev=$(facet_active $facet)_dev
-		dev=${!_dev} # expand _dev to its value, e.g. ${mds1_dev}
-		poolname="${dev%%/*}" # poolname is string before "/"
-
-		if [ "${poolname}" == "" ]; then
-			echo "invalid dataset name: $dev"
-			return
-		fi
-		do_facet $facet "cp /etc/zfs/zpool.cache /tmp/zpool.cache.back"
-		do_facet $facet "$ZPOOL export ${poolname}"
-		do_facet $facet "$ZPOOL import -f -c /tmp/zpool.cache.back \
-		                 ${poolname}"
-	fi
-}
-
 stop() {
     local running
     local facet=$1
@@ -1028,9 +1193,14 @@ stop() {
         do_facet ${facet} umount -d $@ $mntpt
     fi
 
-    # umount should block, but we should wait for unrelated obd's
-    # like the MGS or MGC to also stop.
-    wait_exit_ST ${facet}
+	# umount should block, but we should wait for unrelated obd's
+	# like the MGS or MGC to also stop.
+	wait_exit_ST ${facet} || return ${PIPESTATUS[0]}
+
+	if [[ $(facet_fstype $facet) == zfs ]]; then
+		# export ZFS storage pool
+		export_zpool $facet
+	fi
 }
 
 # save quota version (both administrative and operational quotas)
@@ -1176,6 +1346,23 @@ lfs_df() {
 	$LFS df $* | sed -e 's/filesystem /filesystem_/'
 }
 
+# Get free inodes on the MDT specified by mdt index, free indoes on
+# the whole filesystem will be returned when index == -1.
+mdt_free_inodes() {
+	local index=$1
+	local free_inodes
+	local mdt_uuid
+
+	if [ $index -eq -1 ]; then
+		mdt_uuid="summary"
+	else
+		mdt_uuid=$(mdtuuid_from_index $index)
+	fi
+
+	free_inodes=$(lfs_df -i $MOUNT | grep $mdt_uuid | awk '{print $4}')
+	echo $free_inodes
+}
+
 setup_quota(){
 	if [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.50) ]; then
 		setup_quota_old $1
@@ -1488,7 +1675,6 @@ reboot_facet() {
 	if [ "$FAILURE_MODE" = HARD ]; then
 		reboot_node $(facet_active_host $facet)
 	else
-		refresh_disk ${facet}
 		sleep 10
 	fi
 }
@@ -1551,6 +1737,7 @@ TESTLOG_PREFIX=$TESTLOG_PREFIX \
 TESTNAME=$TESTNAME \
 DBENCH_LIB=$DBENCH_LIB \
 DBENCH_SRC=$DBENCH_SRC \
+CLIENT_COUNT=$((CLIENTCOUNT - 1)) \
 LFS=$LFS \
 run_${load}.sh" &
     local ppid=$!
@@ -1760,30 +1947,44 @@ cleanup_check() {
 }
 
 wait_update () {
-    local node=$1
-    local TEST=$2
-    local FINAL=$3
-    local MAX=${4:-90}
-
-        local RESULT
-        local WAIT=0
-        local sleep=1
-        local print=10
-        while [ true ]; do
-            RESULT=$(do_node $node "$TEST")
-            if [ "$RESULT" == "$FINAL" ]; then
-                [ -z "$RESULT" -o $WAIT -le $sleep ] ||
-                    echo "Updated after ${WAIT}s: wanted '$FINAL' got '$RESULT'"
-                return 0
-            fi
-            [ $WAIT -ge $MAX ] && break
-            [ $((WAIT % print)) -eq 0 ] &&
-                echo "Waiting $((MAX - WAIT)) secs for update"
-            WAIT=$((WAIT + sleep))
-            sleep $sleep
-        done
-        echo "Update not seen after ${MAX}s: wanted '$FINAL' got '$RESULT'"
-        return 3
+	local verbose=false
+	if [[ "$1" == "--verbose" ]]; then
+		shift
+		verbose=true
+	fi
+
+	local node=$1
+	local TEST=$2
+	local FINAL=$3
+	local MAX=${4:-90}
+	local RESULT
+	local PREV_RESULT
+	local WAIT=0
+	local sleep=1
+	local print=10
+
+	PREV_RESULT=$(do_node $node "$TEST")
+	while [ true ]; do
+		RESULT=$(do_node $node "$TEST")
+		if [[ "$RESULT" == "$FINAL" ]]; then
+			[[ -z "$RESULT" || $WAIT -le $sleep ]] ||
+				echo "Updated after ${WAIT}s: wanted '$FINAL'"\
+				     "got '$RESULT'"
+			return 0
+		fi
+		if [[ $verbose && "$RESULT" != "$PREV_RESULT" ]]; then
+			echo "Changed after ${WAIT}s: from '$PREV_RESULT'"\
+			     "to '$RESULT'"
+			PREV_RESULT=$RESULT
+		fi
+		[[ $WAIT -ge $MAX ]] && break
+		[[ $((WAIT % print)) -eq 0 ]] &&
+			echo "Waiting $((MAX - WAIT)) secs for update"
+		WAIT=$((WAIT + sleep))
+		sleep $sleep
+	done
+	echo "Update not seen after ${MAX}s: wanted '$FINAL' got '$RESULT'"
+	return 3
 }
 
 wait_update_facet() {
@@ -1793,7 +1994,10 @@ wait_update_facet() {
 }
 
 sync_all_data() {
-	do_node $(osts_nodes) "lctl set_param -n osd*.*OS*.force_sync 1" 2>&1 |
+	do_nodes $(comma_list $(mdts_nodes)) \
+	    "lctl set_param -n osd*.*MDT*.force_sync 1"
+	do_nodes $(comma_list $(osts_nodes)) \
+	    "lctl set_param -n osd*.*OS*.force_sync 1" 2>&1 |
 		grep -v 'Found no match'
 }
 
@@ -1914,6 +2118,7 @@ wait_mds_ost_sync () {
 	echo "Waiting for orphan cleanup..."
 	# MAX value includes time needed for MDS-OST reconnection
 	local MAX=$(( TIMEOUT * 2 ))
+	local WAIT_TIMEOUT=${1:-$MAX}
 	local WAIT=0
 	local new_wait=true
 	local list=$(comma_list $(mdts_nodes))
@@ -1926,7 +2131,9 @@ wait_mds_ost_sync () {
 		list=$(comma_list $(osts_nodes))
 		cmd="$LCTL get_param -n obdfilter.*.mds_sync"
 	fi
-	while [ $WAIT -lt $MAX ]; do
+
+	echo "wait $WAIT_TIMEOUT secs maximumly for $list mds-ost sync done."
+	while [ $WAIT -lt $WAIT_TIMEOUT ]; do
 		local -a sync=($(do_nodes $list "$cmd"))
 		local con=1
 		local i
@@ -1942,10 +2149,13 @@ wait_mds_ost_sync () {
 		done
 		sleep 2 # increase waiting time and cover statfs cache
 		[ ${con} -eq 1 ] && return 0
-		echo "Waiting $WAIT secs for $facet mds-ost sync done."
+		echo "Waiting $WAIT secs for $list $i mds-ost sync done."
 		WAIT=$((WAIT + 2))
 	done
-	echo "$facet recovery not done in $MAX sec. $STATUS"
+
+	# show which nodes are not finished.
+	do_nodes $list "$cmd"
+	echo "$facet recovery node $i not done in $WAIT_TIMEOUT sec. $STATUS"
 	return 1
 }
 
@@ -2148,7 +2358,7 @@ replay_barrier() {
 	do_facet $facet "sync; sync; sync"
 	df $MOUNT
 
-        # make sure there will be no seq change
+	# make sure there will be no seq change
 	local clients=${CLIENTS:-$HOSTNAME}
 	local f=fsa-\\\$\(hostname\)
 	do_nodes $clients "mcreate $MOUNT/$f; rm $MOUNT/$f"
@@ -2156,6 +2366,21 @@ replay_barrier() {
 
 	local svc=${facet}_svc
 	do_facet $facet $LCTL --device ${!svc} notransno
+	#
+	# If a ZFS OSD is made read-only here, its pool is "freezed". This
+	# in-memory state has to be cleared by either rebooting the host or
+	# exporting and reimporting the pool.
+	#
+	# Although the uberblocks are not updated when a pool is freezed,
+	# transactions are still written to the disks. Modified blocks may be
+	# cached in memory when tests try reading them back. The
+	# export-and-reimport process also evicts any cached pool data from
+	# memory to provide the correct "data loss" semantics.
+	#
+	# In the test framework, the exporting and importing operations are
+	# handled by stop() and mount_facet() separately, which are used
+	# inside fail() and fail_abort().
+	#
 	do_facet $facet $LCTL --device ${!svc} readonly
 	do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
 	$LCTL mark "local REPLAY BARRIER on ${!svc}"
@@ -2182,14 +2407,32 @@ replay_barrier_nosync() {
 	$LCTL mark "local REPLAY BARRIER on ${!svc}"
 }
 
+#
+# Get Lustre client uuid for a given Lustre mount point.
+#
+get_client_uuid() {
+	local mntpnt=${1:-$MOUNT}
+
+	local name=$($LFS getname $mntpnt | cut -d' ' -f1)
+	local uuid=$($LCTL get_param -n llite.$name.uuid)
+
+	echo -n $uuid
+}
+
 mds_evict_client() {
-    UUID=`lctl get_param -n mdc.${mds1_svc}-mdc-*.uuid`
-    do_facet mds1 "lctl set_param -n mdt.${mds1_svc}.evict_client $UUID"
+	local mntpnt=${1:-$MOUNT}
+	local uuid=$(get_client_uuid $mntpnt)
+
+	do_facet $SINGLEMDS \
+		"$LCTL set_param -n mdt.${mds1_svc}.evict_client $uuid"
 }
 
 ost_evict_client() {
-    UUID=`lctl get_param -n devices| grep ${ost1_svc}-osc- | egrep -v 'MDT' | awk '{print $5}'`
-    do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.evict_client $UUID"
+	local mntpnt=${1:-$MOUNT}
+	local uuid=$(get_client_uuid $mntpnt)
+
+	do_facet ost1 \
+		"$LCTL set_param -n obdfilter.${ost1_svc}.evict_client $uuid"
 }
 
 fail() {
@@ -2209,7 +2452,6 @@ fail_nodf() {
 fail_abort() {
 	local facet=$1
 	stop $facet
-	refresh_disk ${facet}
 	change_active $facet
 	wait_for_facet $facet
 	mount_facet $facet -o abort_recovery
@@ -2235,42 +2477,43 @@ host_nids_address() {
 }
 
 h2name_or_ip() {
-    if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
-        echo $1"@$2"
-    fi
+	if [ "$1" = "'*'" ]; then echo \'*\'; else
+		echo $1"@$2"
+	fi
 }
 
 h2ptl() {
-   if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
-       ID=`xtprocadmin -n $1 2>/dev/null | egrep -v 'NID' | awk '{print $1}'`
-       if [ -z "$ID" ]; then
-           echo "Could not get a ptl id for $1..."
-           exit 1
-       fi
-       echo $ID"@ptl"
-   fi
+	if [ "$1" = "'*'" ]; then echo \'*\'; else
+		ID=`xtprocadmin -n $1 2>/dev/null | egrep -v 'NID' | \
+							awk '{print $1}'`
+		if [ -z "$ID" ]; then
+			echo "Could not get a ptl id for $1..."
+			exit 1
+		fi
+		echo $ID"@ptl"
+	fi
 }
 declare -fx h2ptl
 
 h2tcp() {
-    h2name_or_ip "$1" "tcp"
+	h2name_or_ip "$1" "tcp"
 }
 declare -fx h2tcp
 
 h2elan() {
-    if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
-        if type __h2elan >/dev/null 2>&1; then
-            ID=$(__h2elan $1)
-        else
-            ID=`echo $1 | sed 's/[^0-9]*//g'`
-        fi
-        echo $ID"@elan"
-    fi
+	if [ "$1" = "'*'" ]; then echo \'*\'; else
+		if type __h2elan >/dev/null 2>&1; then
+			ID=$(__h2elan $1)
+		else
+			ID=`echo $1 | sed 's/[^0-9]*//g'`
+		fi
+		echo $ID"@elan"
+	fi
 }
 declare -fx h2elan
 
 h2o2ib() {
-    h2name_or_ip "$1" "o2ib"
+	h2name_or_ip "$1" "o2ib"
 }
 declare -fx h2o2ib
 
@@ -2371,7 +2614,11 @@ facet_host() {
 	varname=${facet}_HOST
 	if [ -z "${!varname}" ]; then
 		if [ "${facet:0:3}" == "ost" ]; then
-			eval export ${facet}_HOST=${ost_HOST}
+			local fh=${facet%failover}_HOST
+			eval export ${facet}_HOST=${!fh}
+			if [ -z "${!varname}" ]; then
+				eval export ${facet}_HOST=${ost_HOST}
+			fi
 		elif [ "${facet:0:3}" == "mdt" -o \
 			"${facet:0:3}" == "mds" -o \
 			"${facet:0:3}" == "mgs" ]; then
@@ -2432,6 +2679,24 @@ facet_active_host() {
     fi
 }
 
+# Get the passive failover partner host of facet.
+facet_passive_host() {
+	local facet=$1
+	[[ $facet = client ]] && return
+
+	local host=${facet}_HOST
+	local failover_host=${facet}failover_HOST
+	local active_host=$(facet_active_host $facet)
+
+	[[ -z ${!failover_host} || ${!failover_host} = ${!host} ]] && return
+
+	if [[ $active_host = ${!host} ]]; then
+		echo -n ${!failover_host}
+	else
+		echo -n ${!host}
+	fi
+}
+
 change_active() {
     local facetlist=$1
     local facet
@@ -2617,18 +2882,32 @@ do_nodesv() {
 }
 
 add() {
-    local facet=$1
-    shift
-    # make sure its not already running
-    stop ${facet} -f
-    rm -f $TMP/${facet}active
-    [[ $facet = mds1 ]] && combined_mgs_mds && rm -f $TMP/mgsactive
-    do_facet ${facet} $MKFS $*
+	local facet=$1
+	shift
+	# make sure its not already running
+	stop ${facet} -f
+	rm -f $TMP/${facet}active
+	[[ $facet = mds1 ]] && combined_mgs_mds && rm -f $TMP/mgsactive
+	do_facet ${facet} $MKFS $* || return ${PIPESTATUS[0]}
+
+	if [[ $(facet_fstype $facet) == zfs ]]; then
+		#
+		# After formatting a ZFS target, "cachefile=none" property will
+		# be set on the ZFS storage pool so that the pool is not
+		# automatically imported on system startup. And then the pool
+		# will be exported so as to leave the importing and exporting
+		# operations handled by mount_facet() and stop() separately.
+		#
+		refresh_partition_table $facet $(facet_vdevice $facet)
+		disable_zpool_cache $facet
+		export_zpool $facet
+	fi
 }
 
+# Device formatted as ost
 ostdevname() {
-    num=$1
-    DEVNAME=OSTDEV$num
+	local num=$1
+	local DEVNAME=OSTDEV$num
 
 	local fstype=$(facet_fstype ost$num)
 
@@ -2637,8 +2916,9 @@ ostdevname() {
 			#if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
 			eval DEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}};;
 		zfs )
-			#dataset name is independent of vdev device names
-			eval DEVPTR=${FSNAME}-ost${num}/ost${num};;
+			#try $OSTZFSDEVn - independent of vdev
+			DEVNAME=OSTZFSDEV$num
+			eval DEVPTR=${!DEVNAME:=${FSNAME}-ost${num}/ost${num}};;
 		* )
 			error "unknown fstype!";;
 	esac
@@ -2646,9 +2926,11 @@ ostdevname() {
     echo -n $DEVPTR
 }
 
+# Physical device location of data
 ostvdevname() {
-	num=$1
-	DEVNAME=OSTDEV$num
+	local num=$1
+	local DEVNAME
+	local VDEVPTR
 
 	local fstype=$(facet_fstype ost$num)
 
@@ -2657,7 +2939,9 @@ ostvdevname() {
 			# vdevs are not supported by ldiskfs
 			eval VDEVPTR="";;
 		zfs )
-			#if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
+			#if $OSTDEVn isn't defined, default is $OSTDEVBASE{n}
+			# Device formated by zfs
+			DEVNAME=OSTDEV$num
 			eval VDEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}};;
 		* )
 			error "unknown fstype!";;
@@ -2666,19 +2950,21 @@ ostvdevname() {
 	echo -n $VDEVPTR
 }
 
+# Logical device formated for lustre
 mdsdevname() {
-    num=$1
-    DEVNAME=MDSDEV$num
+	local num=$1
+	local DEVNAME=MDSDEV$num
 
 	local fstype=$(facet_fstype mds$num)
 
 	case $fstype in
 		ldiskfs )
-			#if $MDSDEVn isn't defined, default is $MDSDEVBASE + num
+			#if $MDSDEVn isn't defined, default is $MDSDEVBASE{n}
 			eval DEVPTR=${!DEVNAME:=${MDSDEVBASE}${num}};;
 		zfs )
-			#dataset name is independent of vdev device names
-			eval DEVPTR=${FSNAME}-mdt${num}/mdt${num};;
+			# try $MDSZFSDEVn - independent of vdev
+			DEVNAME=MDSZFSDEV$num
+			eval DEVPTR=${!DEVNAME:=${FSNAME}-mdt${num}/mdt${num}};;
 		* )
 			error "unknown fstype!";;
 	esac
@@ -2686,10 +2972,10 @@ mdsdevname() {
 	echo -n $DEVPTR
 }
 
+# Physical location of data
 mdsvdevname() {
-	num=$1
-	DEVNAME=MDSDEV$num
-
+	local VDEVPTR=""
+	local num=$1
 	local fstype=$(facet_fstype mds$num)
 
 	case $fstype in
@@ -2697,7 +2983,9 @@ mdsvdevname() {
 			# vdevs are not supported by ldiskfs
 			eval VDEVPTR="";;
 		zfs )
-			#if $MDSDEVn isn't defined, default is $MDSDEVBASE + num
+			# if $MDSDEVn isn't defined, default is $MDSDEVBASE{n}
+			# Device formated by ZFS
+			local DEVNAME=MDSDEV$num
 			eval VDEVPTR=${!DEVNAME:=${MDSDEVBASE}${num}};;
 		* )
 			error "unknown fstype!";;
@@ -2720,10 +3008,11 @@ mgsdevname() {
 		fi;;
 	zfs )
 		if [ $(facet_host mgs) = $(facet_host mds1) ] &&
-		   ( [ -z "$MGSDEV" ] || [ $MGSDEV = $(mdsvdevname 1) ] ); then
+		    ( [ -z "$MGSZFSDEV" ] &&
+			[ -z "$MGSDEV" -o "$MGSDEV" = $(mdsvdevname 1) ] ); then
 			DEVPTR=$(mdsdevname 1)
 		else
-			DEVPTR=${FSNAME}-mgs/mgs
+			DEVPTR=${MGSZFSDEV:-${FSNAME}-mgs/mgs}
 		fi;;
 	* )
 		error "unknown fstype!";;
@@ -2733,8 +3022,7 @@ mgsdevname() {
 }
 
 mgsvdevname() {
-	local VDEVPTR
-	DEVNAME=MGSDEV
+	local VDEVPTR=""
 
 	local fstype=$(facet_fstype mgs)
 
@@ -2744,9 +3032,10 @@ mgsvdevname() {
 		;;
 	zfs )
 		if [ $(facet_host mgs) = $(facet_host mds1) ] &&
-		   ( [ -z "$MGSDEV" ] || [ $MGSDEV = $(mdsvdevname 1) ] ); then
+		   ( [ -z "$MGSDEV" ] &&
+		       [ -z "$MGSZFSDEV" -o "$MGSZFSDEV" = $(mdsdevname 1) ]); then
 			VDEVPTR=$(mdsvdevname 1)
-		else
+		elif [ -n "$MGSDEV" ]; then
 			VDEVPTR=$MGSDEV
 		fi;;
 	* )
@@ -2766,6 +3055,84 @@ facet_mntpt () {
     echo -n $mntpt
 }
 
+mount_ldiskfs() {
+	local facet=$1
+	local dev=$(facet_device $facet)
+	local mnt=$(facet_mntpt $facet)
+	local opts
+
+	if ! do_facet $facet test -b $dev; then
+		opts="-o loop"
+	fi
+	do_facet $facet mount -t ldiskfs $opts $dev $mnt
+}
+
+unmount_ldiskfs() {
+	local facet=$1
+	local dev=$(facet_device $facet)
+	local mnt=$(facet_mntpt $facet)
+
+	do_facet $facet umount -d $mnt
+}
+
+var_name() {
+	echo -n "$1" | tr -c '[:alnum:]\n' '_'
+}
+
+mount_zfs() {
+	local facet=$1
+	local ds=$(facet_device $facet)
+	local mnt=$(facet_mntpt $facet)
+	local canmnt
+	local mntpt
+
+	import_zpool $facet
+	canmnt=$(do_facet $facet $ZFS get -H -o value canmount $ds)
+	mntpt=$(do_facet $facet $ZFS get -H -o value mountpoint $ds)
+	do_facet $facet $ZFS set canmount=noauto $ds
+	#
+	# The "legacy" mount method is used here because "zfs unmount $mnt"
+	# calls stat(2) on $mnt/../*, which may include $MOUNT.  If certain
+	# targets are not available at the time, the stat(2) on $MOUNT will
+	# hang.
+	#
+	do_facet $facet $ZFS set mountpoint=legacy $ds
+	do_facet $facet mount -t zfs $ds $mnt
+	eval export mz_$(var_name ${facet}_$ds)_canmount=$canmnt
+	eval export mz_$(var_name ${facet}_$ds)_mountpoint=$mntpt
+}
+
+unmount_zfs() {
+	local facet=$1
+	local ds=$(facet_device $facet)
+	local mnt=$(facet_mntpt $facet)
+	local var_mntpt=mz_$(var_name ${facet}_$ds)_mountpoint
+	local var_canmnt=mz_$(var_name ${facet}_$ds)_canmount
+	local mntpt=${!var_mntpt}
+	local canmnt=${!var_canmnt}
+
+	unset $var_mntpt
+	unset $var_canmnt
+	do_facet $facet umount $mnt
+	do_facet $facet $ZFS set mountpoint=$mntpt $ds
+	do_facet $facet $ZFS set canmount=$canmnt $ds
+	export_zpool $facet
+}
+
+mount_fstype() {
+	local facet=$1
+	local fstype=$(facet_fstype $facet)
+
+	mount_$fstype $facet
+}
+
+unmount_fstype() {
+	local facet=$1
+	local fstype=$(facet_fstype $facet)
+
+	unmount_$fstype $facet
+}
+
 ########
 ## MountConf setup
 
@@ -2882,7 +3249,16 @@ mkfs_opts() {
 		opts+=${L_GETIDENTITY:+" --param=mdt.identity_upcall=$L_GETIDENTITY"}
 
 		if [ $fstype == ldiskfs ]; then
+			# Check for wide striping
+			if [ $OSTCOUNT -gt 160 ]; then
+				MDSJOURNALSIZE=${MDSJOURNALSIZE:-4096}
+				fs_mkfs_opts+="-O large_xattr"
+			fi
+
 			fs_mkfs_opts+=${MDSJOURNALSIZE:+" -J size=$MDSJOURNALSIZE"}
+			if [ ! -z $EJOURNAL ]; then
+				fs_mkfs_opts+=${MDSJOURNALSIZE:+" device=$EJOURNAL"}
+			fi
 			fs_mkfs_opts+=${MDSISIZE:+" -i $MDSISIZE"}
 		fi
 	fi
@@ -3003,7 +3379,7 @@ writeconf_facet() {
 	local dev=$2
 
 	stop ${facet} -f
-	rm -f ${facet}active
+	rm -f $TMP/${facet}active
 	do_facet ${facet} "$TUNEFS --quiet --writeconf $dev" || return 1
 	return 0
 }
@@ -3246,26 +3622,20 @@ init_param_vars () {
 	osc_ensure_active $SINGLEMDS $TIMEOUT
 	osc_ensure_active client $TIMEOUT
 
-	local jobid_var
-	if [ -z "$(lctl get_param -n mdc.*.connect_flags | grep jobstats)" ]; then
-		jobid_var="none"
-	elif [ $JOBSTATS_AUTO -ne 0 ]; then
-		echo "enable jobstats, set job scheduler as $JOBID_VAR"
-		jobid_var=$JOBID_VAR
-	else
-		jobid_var=`$LCTL get_param -n jobid_var`
-		if [ $jobid_var != "disable" ]; then
-			echo "disable jobstats as required"
-			jobid_var="disable"
-		else
-			jobid_var="none"
-		fi
-	fi
+	if [ -n "$(lctl get_param -n mdc.*.connect_flags|grep jobstats)" ]; then
+		local current_jobid_var=$($LCTL get_param -n jobid_var)
+
+		if [ $JOBID_VAR = "existing" ]; then
+			echo "keeping jobstats as $current_jobid_var"
+		elif [ $current_jobid_var != $JOBID_VAR ]; then
+			echo "seting jobstats to $JOBID_VAR"
 
-	if [ $jobid_var == $JOBID_VAR -o $jobid_var == "disable" ]; then
-		do_facet mgs $LCTL conf_param $FSNAME.sys.jobid_var=$jobid_var
-		wait_update $HOSTNAME "$LCTL get_param -n jobid_var" \
-			$jobid_var || return 1
+			set_conf_param_and_check client			\
+				"$LCTL get_param -n jobid_var"		\
+				"$FSNAME.sys.jobid_var" $JOBID_VAR
+		fi
+	else
+		echo "jobstats not supported by server"
 	fi
 
 	if [ $QUOTA_AUTO -ne 0 ]; then
@@ -3375,10 +3745,17 @@ is_empty_dir() {
 
 # empty lustre filesystem may have empty directories lost+found and .lustre
 is_empty_fs() {
+	# exclude .lustre & lost+found
 	[ $(find $1 -maxdepth 1 -name lost+found -o -name .lustre -prune -o \
 		-print | wc -l) = 1 ] || return 1
 	[ ! -d $1/lost+found ] || is_empty_dir $1/lost+found || return 1
-	[ ! -d $1/.lustre ] || is_empty_dir $1/.lustre || return 1
+	if [ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.4.0) ]; then
+		# exclude .lustre/fid (LU-2780)
+		[ $(find $1/.lustre -maxdepth 1 -name fid -prune -o \
+			-print | wc -l) = 1 ] || return 1
+	else
+		[ ! -d $1/.lustre ] || is_empty_dir $1/.lustre || return 1
+	fi
 	return 0
 }
 
@@ -3440,6 +3817,21 @@ check_and_setup_lustre() {
         set_default_debug_nodes $(comma_list $(nodes_list))
     fi
 
+	if [ $(lower $OSD_TRACK_DECLARES_LBUG) == 'yes' ] ; then
+		local facets=""
+		[ "$(facet_fstype ost1)" = "ldiskfs" ] &&
+			facets="$(get_facets OST)"
+		[ "$(facet_fstype mds1)" = "ldiskfs" ] &&
+			facets="$facets,$(get_facets MDS)"
+		[ "$(facet_fstype mgs)" = "ldiskfs" ] &&
+			facets="$facets,mgs"
+		local nodes="$(facets_hosts ${facets})"
+		if [ -n "$nodes" ] ; then
+			do_nodes $nodes "$LCTL set_param \
+				 osd-ldiskfs.track_declares_assert=1 || true"
+		fi
+	fi
+
 	init_gss
 	if $GSS; then
 		set_flavor_all $SEC
@@ -3458,10 +3850,10 @@ restore_mount () {
 }
 
 cleanup_mount () {
-    local clients=${CLIENTS:-$HOSTNAME}
-    local mntpt=$1
+	local clients=${CLIENTS:-$HOSTNAME}
+	local mntpt=$1
 
-    zconf_umount_clients $clients $mntpt    
+	zconf_umount_clients $clients $mntpt
 }
 
 cleanup_and_setup_lustre() {
@@ -3485,8 +3877,7 @@ get_mnt_devs() {
 	if [ "$type" == ost ]; then
 		devs=$(get_osd_param $node "" mntdev)
 	else
-		devs=$(do_node $node \
-		       "lctl get_param -n osd-*.$FSNAME-M*.mntdev")
+		devs=$(do_node $node $LCTL get_param -n osd-*.$FSNAME-M*.mntdev)
 	fi
 	for dev in $devs; do
 		case $dev in
@@ -3499,63 +3890,101 @@ get_mnt_devs() {
 
 # Get all of the server target devices.
 get_svr_devs() {
-    local i
+	local node
+	local i
 
-    # MDT device
-    MDTDEV=$(get_mnt_devs $(mdts_nodes) mdt)
+	# Master MDS parameters used by lfsck
+	MDTNODE=$(facet_active_host $SINGLEMDS)
+	MDTDEV=$(echo $(get_mnt_devs $MDTNODE mdt) | awk '{print $1}')
 
-    # OST devices
-    i=0
-    for node in $(osts_nodes); do
-        OSTDEVS[i]=$(get_mnt_devs $node ost)
-        i=$((i + 1))
-    done
+	# MDT devices
+	i=0
+	for node in $(mdts_nodes); do
+		MDTDEVS[i]=$(get_mnt_devs $node mdt)
+		i=$((i + 1))
+	done
+
+	# OST devices
+	i=0
+	for node in $(osts_nodes); do
+		OSTDEVS[i]=$(get_mnt_devs $node ost)
+		i=$((i + 1))
+	done
 }
 
 # Run e2fsck on MDT or OST device.
 run_e2fsck() {
-    local node=$1
-    local target_dev=$2
-    local extra_opts=$3
+	local node=$1
+	local target_dev=$2
+	local extra_opts=$3
+	local cmd="$E2FSCK -d -v -t -t -f $extra_opts $target_dev"
+	local log=$TMP/e2fsck.log
+	local rc=0
 
-    df > /dev/null      # update statfs data on disk
-    local cmd="$E2FSCK -d -v -t -t -f $extra_opts $target_dev"
-    echo $cmd
-    local rc=0
-    do_node $node $cmd || rc=$?
-    [ $rc -le $FSCK_MAX_ERR ] || \
-        error "$cmd returned $rc, should be <= $FSCK_MAX_ERR"
-    return 0
+	echo $cmd
+	do_node $node $cmd 2>&1 | tee $log
+	rc=${PIPESTATUS[0]}
+	if [ -n "$(grep "DNE mode isn't supported" $log)" ]; then
+		rm -f $log
+		if [ $MDSCOUNT -gt 1 ]; then
+			skip "DNE mode isn't supported!"
+			cleanupall
+			exit_status
+		else
+			error "It's not DNE mode."
+		fi
+	fi
+	rm -f $log
+
+	[ $rc -le $FSCK_MAX_ERR ] ||
+		error "$cmd returned $rc, should be <= $FSCK_MAX_ERR"
+
+	return 0
+}
+
+#
+# Run resize2fs on MDT or OST device.
+#
+run_resize2fs() {
+	local facet=$1
+	local device=$2
+	local size=$3
+	shift 3
+	local opts="$@"
+
+	do_facet $facet "$RESIZE2FS $opts $device $size"
 }
 
 # verify a directory is shared among nodes.
 check_shared_dir() {
 	local dir=$1
+	local list=${2:-$(comma_list $(nodes_list))}
 
 	[ -z "$dir" ] && return 1
-	do_rpc_nodes "$(comma_list $(nodes_list))" check_logdir $dir
-	check_write_access $dir || return 1
+	do_rpc_nodes "$list" check_logdir $dir
+	check_write_access $dir "$list" || return 1
 	return 0
 }
 
 # Run e2fsck on MDT and OST(s) to generate databases used for lfsck.
 generate_db() {
-    local i
-    local ostidx
-    local dev
+	local i
+	local ostidx
+	local dev
+	local node
 
 	[[ $(lustre_version_code $SINGLEMDS) -ne $(version_code 2.2.0) ]] ||
 		{ skip "Lustre 2.2.0 lacks the patch for LU-1255"; exit 0; }
 
-    check_shared_dir $SHARED_DIRECTORY ||
-        error "$SHARED_DIRECTORY isn't a shared directory"
-
-    export MDSDB=$SHARED_DIRECTORY/mdsdb
-    export OSTDB=$SHARED_DIRECTORY/ostdb
+	check_shared_dir $SHARED_DIRECTORY ||
+		error "$SHARED_DIRECTORY isn't a shared directory"
 
-    [ $MDSCOUNT -eq 1 ] || error "CMD is not supported"
+	export MDSDB=$SHARED_DIRECTORY/mdsdb
+	export OSTDB=$SHARED_DIRECTORY/ostdb
 
-    run_e2fsck $(mdts_nodes) $MDTDEV "-n --mdsdb $MDSDB"
+	# DNE is not supported, so when running e2fsck on a DNE filesystem,
+	# we only pass master MDS parameters.
+	run_e2fsck $MDTNODE $MDTDEV "-n --mdsdb $MDSDB"
 
     i=0
     ostidx=0
@@ -3585,7 +4014,7 @@ run_lfsck_remote() {
 	fi
 	#Run lfsck
 	echo $cmd
-	do_node $node $cmd || rc=$?
+	do_node $client $cmd || rc=$?
 	#Umount if necessary
 	if ! $mounted; then
 		zconf_umount $client $MOUNT ||
@@ -3769,7 +4198,7 @@ get_facets () {
 
         case $type in
                 MGS ) list="$list $name";;
-            MDS|OST ) local count=${type}COUNT
+            MDS|OST|AGT ) local count=${type}COUNT
                        for ((i=1; i<=${!count}; i++)) do
                           list="$list ${name}$i"
                       done;;
@@ -3872,12 +4301,16 @@ drop_update_reply() {
 
 pause_bulk() {
 #define OBD_FAIL_OST_BRW_PAUSE_BULK      0x214
-    RC=0
-    do_facet ost1 lctl set_param fail_loc=0x214
-    do_facet client "$1" || RC=$?
-    do_facet client "sync"
-    do_facet ost1 lctl set_param fail_loc=0
-    return $RC
+	RC=0
+
+	local timeout=${2:-0}
+	# default is (obd_timeout / 4) if unspecified
+	echo "timeout is $timeout/$2"
+	do_facet ost1 lctl set_param fail_val=$timeout fail_loc=0x80000214
+	do_facet client "$1" || RC=$?
+	do_facet client "sync"
+	do_facet ost1 lctl set_param fail_loc=0
+	return $RC
 }
 
 drop_ldlm_cancel() {
@@ -3919,7 +4352,7 @@ clear_failloc() {
 }
 
 set_nodes_failloc () {
-    do_nodes $(comma_list $1)  lctl set_param fail_loc=$2
+	do_nodes $(comma_list $1)  lctl set_param fail_val=0 fail_loc=$2
 }
 
 cancel_lru_locks() {
@@ -4012,28 +4445,37 @@ log_trace_dump() {
 ##################################
 
 error_noexit() {
-    local TYPE=${TYPE:-"FAIL"}
+	local TYPE=${TYPE:-"FAIL"}
 
-    local dump=true
-    # do not dump logs if $1=false
-    if [ "x$1" = "xfalse" ]; then
-        shift
-        dump=false
-    fi
+	local dump=true
+	# do not dump logs if $1=false
+	if [ "x$1" = "xfalse" ]; then
+		shift
+		dump=false
+	fi
 
-    log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
-    log_trace_dump
 
-    mkdir -p $LOGDIR
-    # We need to dump the logs on all nodes
-    if $dump; then
-        gather_logs $(comma_list $(nodes_list))
-    fi
+	log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
+	log_trace_dump
+
+	mkdir -p $LOGDIR
+	# We need to dump the logs on all nodes
+	if $dump; then
+		gather_logs $(comma_list $(nodes_list))
+	fi
 
 	debugrestore
 	[ "$TESTSUITELOG" ] &&
 		echo "$TESTSUITE: $TYPE: $TESTNAME $@" >> $TESTSUITELOG
-	echo "$@" > $LOGDIR/err
+	if [ -z "$*" ]; then
+		echo "error() without useful message, please fix" > $LOGDIR/err
+	else
+		if [[ `echo $TYPE | grep ^IGNORE` ]]; then
+			echo "$@" > $LOGDIR/ignore
+		else
+			echo "$@" > $LOGDIR/err
+		fi
+	fi
 }
 
 exit_status () {
@@ -4045,21 +4487,22 @@ exit_status () {
 }
 
 error() {
-    error_noexit "$@"
-    exit 1
+	error_noexit "$@"
+	exit 1
 }
 
 error_exit() {
-    error "$@"
+	error "$@"
 }
 
 # use only if we are ignoring failures for this test, bugno required.
 # (like ALWAYS_EXCEPT, but run the test and ignore the results.)
-# e.g. error_ignore 5494 "your message"
+# e.g. error_ignore bz5494 "your message" or
+# error_ignore LU-5494 "your message"
 error_ignore() {
-    local TYPE="IGNORE (bz$1)"
-    shift
-    error_noexit "$@"
+	local TYPE="IGNORE ($1)"
+	shift
+	error_noexit "$@"
 }
 
 error_and_remount() {
@@ -4069,7 +4512,7 @@ error_and_remount() {
 }
 
 skip_env () {
-    $FAIL_ON_SKIP_ENV && error false $@ || skip $@
+	$FAIL_ON_SKIP_ENV && error false $@ || skip $@
 }
 
 skip() {
@@ -4090,10 +4533,16 @@ skip() {
 build_test_filter() {
     EXCEPT="$EXCEPT $(testslist_filter)"
 
-    [ "$ONLY" ] && log "only running test `echo $ONLY`"
-    for O in $ONLY; do
-        eval ONLY_${O}=true
-    done
+	for O in $ONLY; do
+		if [[ $O = [0-9]*-[0-9]* ]]; then
+			for num in $(seq $(echo $O | tr '-' ' ')); do
+				eval ONLY_$num=true
+			done
+		else
+			eval ONLY_${O}=true
+		fi
+	done
+
     [ "$EXCEPT$ALWAYS_EXCEPT" ] && \
         log "excepting tests: `echo $EXCEPT $ALWAYS_EXCEPT`"
     [ "$EXCEPT_SLOW" ] && \
@@ -4269,34 +4718,35 @@ banner() {
 }
 
 #
-# Run a single test function and cleanup after it.  
+# Run a single test function and cleanup after it.
 #
 # This function should be run in a subshell so the test func can
 # exit() without stopping the whole script.
 #
 run_one() {
-    local testnum=$1
-    local message=$2
-    tfile=f.${TESTSUITE}.${testnum}
-    export tdir=d0.${TESTSUITE}/d${base}
-    export TESTNAME=test_$testnum
-    local SAVE_UMASK=`umask`
-    umask 0022
-
-    banner "test $testnum: $message"
-    test_${testnum} || error "test_$testnum failed with $?"
-    cd $SAVE_PWD
-    reset_fail_loc
-    check_grant ${testnum} || error "check_grant $testnum failed with $?"
-    check_catastrophe || error "LBUG/LASSERT detected"
+	local testnum=$1
+	local message=$2
+	export tfile=f${testnum}.${TESTSUITE}
+	export tdir=d${testnum}.${TESTSUITE}
+	export TESTNAME=test_$testnum
+	local SAVE_UMASK=`umask`
+	umask 0022
+
+	banner "test $testnum: $message"
+	test_${testnum} || error "test_$testnum failed with $?"
+	cd $SAVE_PWD
+	reset_fail_loc
+	check_grant ${testnum} || error "check_grant $testnum failed with $?"
+	check_catastrophe || error "LBUG/LASSERT detected"
 	if [ "$PARALLEL" != "yes" ]; then
 		ps auxww | grep -v grep | grep -q multiop &&
 					error "multiop still running"
 	fi
-    unset TESTNAME
-    unset tdir
-    umask $SAVE_UMASK
-    return 0
+	unset TESTNAME
+	unset tdir
+	unset tfile
+	umask $SAVE_UMASK
+	return 0
 }
 
 #
@@ -4311,6 +4761,7 @@ run_one_logged() {
 	local name=${TESTSUITE}.test_${1}.test_log.$(hostname -s).log
 	local test_log=$LOGDIR/$name
 	rm -rf $LOGDIR/err
+	rm -rf $LOGDIR/ignore
 	rm -rf $LOGDIR/skip
 	local SAVE_UMASK=`umask`
 	umask 0022
@@ -4320,7 +4771,7 @@ run_one_logged() {
 	(run_one $1 "$2") 2>&1 | tee -i $test_log
 	local RC=${PIPESTATUS[0]}
 
-	[ $RC -ne 0 ] && [ ! -f $LOGDIR/err ] && \
+	[ $RC -ne 0 ] && [ ! -f $LOGDIR/err ] &&
 		echo "test_$1 returned $RC" | tee $LOGDIR/err
 
 	duration=$((`date +%s` - $BEFORE))
@@ -4328,11 +4779,17 @@ run_one_logged() {
 
 	if [[ -f $LOGDIR/err ]]; then
 		TEST_ERROR=$(cat $LOGDIR/err)
+	elif [[ -f $LOGDIR/ignore ]]; then
+		TEST_ERROR=$(cat $LOGDIR/ignore)
 	elif [[ -f $LOGDIR/skip ]]; then
 		TEST_ERROR=$(cat $LOGDIR/skip)
 	fi
 	log_sub_test_end $TEST_STATUS $duration "$RC" "$TEST_ERROR"
 
+	if [[ "$TEST_STATUS" != "SKIP" ]] && [[ -f $TF_SKIP ]]; then
+		rm -f $TF_SKIP
+	fi
+
 	if [ -f $LOGDIR/err ]; then
 		$FAIL_ON_ERROR && exit $RC
 	fi
@@ -4525,64 +4982,126 @@ local_mode ()
         $(single_local_node $(comma_list $(nodes_list)))
 }
 
-mdts_nodes () {
-    local MDSNODES
-    local NODES_sort
-    for num in `seq $MDSCOUNT`; do
-        MDSNODES="$MDSNODES $(facet_host mds$num)"
-    done
-    NODES_sort=$(for i in $MDSNODES; do echo $i; done | sort -u)
-
-    echo $NODES_sort
-}
-
 remote_servers () {
     remote_ost && remote_mds
 }
 
+# Get the active nodes for facets.
 facets_nodes () {
-    local facets=$1
-    local nodes
-    local NODES_sort
+	local facets=$1
+	local facet
+	local nodes
+	local nodes_sort
+	local i
 
-    for facet in ${facets//,/ }; do
-        if [ "$FAILURE_MODE" = HARD ]; then
-            nodes="$nodes $(facet_active_host $facet)"
-        else
-            nodes="$nodes $(facet_host $facet)"
-        fi
-    done
-    NODES_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	for facet in ${facets//,/ }; do
+		nodes="$nodes $(facet_active_host $facet)"
+	done
+
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
+}
 
-    echo $NODES_sort
+# Get all of the active MDS nodes.
+mdts_nodes () {
+	echo -n $(facets_nodes $(get_facets MDS))
 }
 
+# Get all of the active OSS nodes.
 osts_nodes () {
-    local facets=$(get_facets OST)
-    local nodes=$(facets_nodes $facets)
+	echo -n $(facets_nodes $(get_facets OST))
+}
 
-    echo $nodes
+# Get all of the active AGT (HSM agent) nodes.
+agts_nodes () {
+	echo -n $(facets_nodes $(get_facets AGT))
 }
 
+# Get all of the client nodes and active server nodes.
 nodes_list () {
-    # FIXME. We need a list of clients
-    local myNODES=$HOSTNAME
-    local myNODES_sort
-
-    # CLIENTS (if specified) contains the local client
-    [ -n "$CLIENTS" ] && myNODES=${CLIENTS//,/ }
+	local nodes=$HOSTNAME
+	local nodes_sort
+	local i
 
-    if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then
-        myNODES="$myNODES $(facets_nodes $(get_facets))"
-    fi
+	# CLIENTS (if specified) contains the local client
+	[ -n "$CLIENTS" ] && nodes=${CLIENTS//,/ }
 
-    myNODES_sort=$(for i in $myNODES; do echo $i; done | sort -u)
+	if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then
+		nodes="$nodes $(facets_nodes $(get_facets))"
+	fi
 
-    echo $myNODES_sort
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
 }
 
+# Get all of the remote client nodes and remote active server nodes.
 remote_nodes_list () {
-	echo $(nodes_list) | sed -re "s/\<$HOSTNAME\>//g"
+	echo -n $(nodes_list) | sed -re "s/\<$HOSTNAME\>//g"
+}
+
+# Get all of the MDS nodes, including active and passive nodes.
+all_mdts_nodes () {
+	local host
+	local failover_host
+	local nodes
+	local nodes_sort
+	local i
+
+	for i in $(seq $MDSCOUNT); do
+		host=mds${i}_HOST
+		failover_host=mds${i}failover_HOST
+		nodes="$nodes ${!host} ${!failover_host}"
+	done
+
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
+}
+
+# Get all of the OSS nodes, including active and passive nodes.
+all_osts_nodes () {
+	local host
+	local failover_host
+	local nodes
+	local nodes_sort
+	local i
+
+	for i in $(seq $OSTCOUNT); do
+		host=ost${i}_HOST
+		failover_host=ost${i}failover_HOST
+		nodes="$nodes ${!host} ${!failover_host}"
+	done
+
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
+}
+
+# Get all of the server nodes, including active and passive nodes.
+all_server_nodes () {
+	local nodes
+	local nodes_sort
+	local i
+
+	nodes="$mgs_HOST $mgsfailover_HOST $(all_mdts_nodes) $(all_osts_nodes)"
+
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
+}
+
+# Get all of the client and server nodes, including active and passive nodes.
+all_nodes () {
+	local nodes=$HOSTNAME
+	local nodes_sort
+	local i
+
+	# CLIENTS (if specified) contains the local client
+	[ -n "$CLIENTS" ] && nodes=${CLIENTS//,/ }
+
+	if [ "$PDSH" -a "$PDSH" != "no_dsh" ]; then
+		nodes="$nodes $(all_server_nodes)"
+	fi
+
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
 }
 
 init_clients_lists () {
@@ -4663,10 +5182,11 @@ generate_machine_file() {
 }
 
 get_stripe () {
-    local file=$1/stripe
-    touch $file
-    $LFS getstripe -v $file || error
-    rm -f $file
+	local file=$1/stripe
+
+	touch $file
+	$LFS getstripe -v $file || error "getstripe $file failed"
+	rm -f $file
 }
 
 setstripe_nfsserver () {
@@ -4883,21 +5403,33 @@ calc_osc_kbytes () {
         $LCTL get_param -n osc.*[oO][sS][cC][-_][0-9a-f]*.$1 | calc_sum
 }
 
-# save_lustre_params(node, parameter_mask)
-# generate a stream of formatted strings (<node> <param name>=<param value>)
+# save_lustre_params(comma separated facet list, parameter_mask)
+# generate a stream of formatted strings (<facet> <param name>=<param value>)
 save_lustre_params() {
-        local s
-        do_nodesv $1 "lctl get_param $2 | while read s; do echo \\\$s; done"
+	local facets=$1
+	local facet
+	local nodes
+	local node
+
+	for facet in ${facets//,/ }; do
+		node=$(facet_active_host $facet)
+		[[ *\ $node\ * = " $nodes " ]] && continue
+		nodes="$nodes $node"
+
+		do_node $node "$LCTL get_param $2 |
+			while read s; do echo $facet \\\$s; done"
+	done
 }
 
 # restore lustre parameters from input stream, produces by save_lustre_params
 restore_lustre_params() {
-        local node
-        local name
-        local val
-        while IFS=" =" read node name val; do
-                do_node ${node//:/} "lctl set_param -n $name $val"
-        done
+	local facet
+	local name
+	local val
+
+	while IFS=" =" read facet name val; do
+		do_facet $facet "$LCTL set_param -n $name $val"
+	done
 }
 
 check_catastrophe() {
@@ -4949,7 +5481,7 @@ get_mds_dir () {
 
 mdsrate_cleanup () {
 	if [ -d $4 ]; then
-		mpi_run -np $1 ${MACHINEFILE_OPTION} $2 ${MDSRATE} --unlink \
+		mpi_run ${MACHINEFILE_OPTION} $2 -np $1 ${MDSRATE} --unlink \
 			--nfiles $3 --dir $4 --filefmt $5 $6
 		rmdir $4
 	fi
@@ -4962,7 +5494,7 @@ delayed_recovery_enabled () {
 
 ########################
 
-convert_facet2label() { 
+convert_facet2label() {
     local facet=$1
 
     if [ x$facet = xost ]; then
@@ -4973,7 +5505,7 @@ convert_facet2label() {
 
     if [ -n ${!varsvc} ]; then
         echo ${!varsvc}
-    else  
+    else
         error "No lablel for $facet!"
     fi
 }
@@ -5124,41 +5656,36 @@ request_timeout () {
     echo $(( init_connect_timeout + at_min ))
 }
 
-wait_osc_import_state() {
-    local facet=$1
-    local ost_facet=$2
-    local expected=$3
-    local ost=$(get_osc_import_name $facet $ost_facet)
-
+_wait_osc_import_state() {
+	local facet=$1
+	local ost_facet=$2
+	local expected=$3
+	local ost=$(get_osc_import_name $facet $ost_facet)
 	local param="osc.${ost}.ost_server_uuid"
+	local params=$param
 	local i=0
 
-    # 1. wait the deadline of client 1st request (it could be skipped)
-    # 2. wait the deadline of client 2nd request
-    local maxtime=$(( 2 * $(request_timeout $facet)))
-
-	#During setup time, the osc might not be setup, it need wait
-	#until list_param can return valid value. And also if there
-	#are mulitple osc entries we should list all of them before
-	#go to wait.
-	local params=$($LCTL list_param $param || true)
-	while [ -z "$params" ]; do
-		if [ $i -ge $maxtime ]; then
-			echo "can't get $param by list_param in $maxtime secs"
-			if [[ $facet != client* ]]; then
-				echo "Go with $param directly"
-				params=$param
-				break
-			else
+	# 1. wait the deadline of client 1st request (it could be skipped)
+	# 2. wait the deadline of client 2nd request
+	local maxtime=$(( 2 * $(request_timeout $facet)))
+
+	if [[ $facet == client* ]]; then
+		# During setup time, the osc might not be setup, it need wait
+		# until list_param can return valid value. And also if there
+		# are mulitple osc entries we should list all of them before
+		# go to wait.
+		params=$($LCTL list_param $param 2>/dev/null || true)
+		while [ -z "$params" ]; do
+			if [ $i -ge $maxtime ]; then
+				echo "can't get $param in $maxtime secs"
 				return 1
 			fi
-		fi
-		sleep 1
-		i=$((i + 1))
-		params=$($LCTL list_param $param || true)
-	done
-
-	if ! do_rpc_nodes "$(facet_host $facet)" \
+			sleep 1
+			i=$((i + 1))
+			params=$($LCTL list_param $param 2>/dev/null || true)
+		done
+	fi
+	if ! do_rpc_nodes "$(facet_active_host $facet)" \
 			wait_import_state $expected "$params" $maxtime; then
 		error "import is not in ${expected} state"
 		return 1
@@ -5167,6 +5694,21 @@ wait_osc_import_state() {
 	return 0
 }
 
+wait_osc_import_state() {
+	local facet=$1
+	local ost_facet=$2
+	local expected=$3
+	local num
+
+	if [[ $facet = mds ]]; then
+		for num in $(seq $MDSCOUNT); do
+			_wait_osc_import_state mds$num "$ost_facet" "$expected"
+		done
+	else
+		_wait_osc_import_state "$facet" "$ost_facet" "$expected"
+	fi
+}
+
 get_clientmdc_proc_path() {
     echo "${1}-mdc-*"
 }
@@ -5178,7 +5720,9 @@ do_rpc_nodes () {
 	[ -z "$list" ] && return 0
 
 	# Add paths to lustre tests for 32 and 64 bit systems.
-	local RPATH="PATH=$RLUSTRE/tests:/usr/lib/lustre/tests:/usr/lib64/lustre/tests:$PATH"
+	local LIBPATH="/usr/lib/lustre/tests:/usr/lib64/lustre/tests:"
+	local TESTPATH="$RLUSTRE/tests:"
+	local RPATH="PATH=${TESTPATH}${LIBPATH}${PATH}:/sbin:/bin:/usr/sbin:"
 	do_nodesv $list "${RPATH} NAME=${NAME} sh rpc.sh $@ "
 }
 
@@ -5716,19 +6260,20 @@ check_logdir() {
 }
 
 check_write_access() {
-    local dir=$1
-    local node
-    local file
+	local dir=$1
+	local list=${2:-$(comma_list $(nodes_list))}
+	local node
+	local file
 
-    for node in $(nodes_list); do
-        file=$dir/check_file.$(short_hostname $node)
-        if [[ ! -f "$file" ]]; then
-            # Logdir not accessible/writable from this node.
-            return 1
-        fi
-        rm -f $file || return 1
-    done
-    return 0
+	for node in ${list//,/ }; do
+		file=$dir/check_file.$(short_nodename $node)
+		if [[ ! -f "$file" ]]; then
+			# Logdir not accessible/writable from this node.
+			return 1
+		fi
+		rm -f $file || return 1
+	done
+	return 0
 }
 
 init_logging() {
@@ -5807,6 +6352,33 @@ run_llverfs()
         llverfs $partial_arg $llverfs_opts $dir
 }
 
+#Remove objects from OST
+remove_ost_objects() {
+	local facet=$1
+	local ostdev=$2
+	local group=$3
+	shift 3
+	local objids="$@"
+	local mntpt=$(facet_mntpt $facet)
+	local opts=$OST_MOUNT_OPTS
+	local i
+	local rc
+
+	echo "removing objects from $ostdev on $facet: $objids"
+	if ! test -b $ostdev; then
+		opts=$(csa_add "$opts" -o loop)
+	fi
+	mount -t $(facet_fstype $facet) $opts $ostdev $mntpt ||
+		return $?
+	rc=0
+	for i in $objids; do
+		rm $mntpt/O/$group/d$((i % 32))/$i || { rc=$?; break; }
+	done
+	umount -f $mntpt || return $?
+	return $rc
+}
+
+#Remove files from MDT
 remove_mdt_files() {
 	local facet=$1
 	local mdtdev=$2
@@ -5822,7 +6394,7 @@ remove_mdt_files() {
 	fi
 	mount -t $(facet_fstype $facet) $opts $mdtdev $mntpt ||
 		return $?
-	rc=0;
+	rc=0
 	for f in $files; do
 		rm $mntpt/ROOT/$f || { rc=$?; break; }
 	done
@@ -5910,6 +6482,45 @@ min_ost_size () {
     $LCTL get_param -n osc.*.kbytesavail | sort -n | head -n1
 }
 
+#
+# Get the available size (KB) of a given obd target.
+#
+get_obd_size() {
+	local facet=$1
+	local obd=$2
+	local size
+
+	[[ $facet != client ]] || return 0
+
+	size=$(do_facet $facet $LCTL get_param -n *.$obd.kbytesavail | head -n1)
+	echo -n $size
+}
+
+#
+# Get the page size (bytes) on a given facet node.
+#
+get_page_size() {
+	local facet=$1
+	local size
+
+	size=$(do_facet $facet getconf PAGE_SIZE)
+	[[ ${PIPESTATUS[0]} = 0 && -n "$size" ]] || size=4096
+	echo -n $size
+}
+
+#
+# Get the block count of the filesystem.
+#
+get_block_count() {
+	local facet=$1
+	local device=$2
+	local count
+
+	count=$(do_facet $facet "$DUMPE2FS -h $device 2>&1" |
+		awk '/^Block count:/ {print $3}')
+	echo -n $count
+}
+
 # Get the block size of the filesystem.
 get_block_size() {
     local facet=$1
@@ -5927,7 +6538,8 @@ large_xattr_enabled() {
 
 	local mds_dev=$(mdsdevname ${SINGLEMDS//mds/})
 
-	do_facet $SINGLEMDS "$DUMPE2FS -h $mds_dev 2>&1 | grep -q large_xattr"
+	do_facet $SINGLEMDS "$DUMPE2FS -h $mds_dev 2>&1 |
+		grep -E -q '(ea_inode|large_xattr)'"
 	return ${PIPESTATUS[0]}
 }
 
@@ -5966,30 +6578,33 @@ generate_string() {
 }
 
 reformat_external_journal() {
+	local facet=$1
+
 	if [ ! -z ${EJOURNAL} ]; then
-		local rcmd="do_facet ${SINGLEMDS}"
+		local rcmd="do_facet $facet"
 
-		echo "reformat external journal on ${SINGLEMDS}:${EJOURNAL}"
+		echo "reformat external journal on $facet:${EJOURNAL}"
 		${rcmd} mke2fs -O journal_dev ${EJOURNAL} || return 1
 	fi
 }
 
 # MDT file-level backup/restore
 mds_backup_restore() {
-	local devname=$(mdsdevname ${SINGLEMDS//mds/})
+	local facet=$1
+	local igif=$2
+	local devname=$(mdsdevname $(facet_number $facet))
 	local mntpt=$(facet_mntpt brpt)
-	local rcmd="do_facet ${SINGLEMDS}"
+	local rcmd="do_facet $facet"
 	local metaea=${TMP}/backup_restore.ea
 	local metadata=${TMP}/backup_restore.tgz
 	local opts=${MDS_MOUNT_OPTS}
-	local svc=${SINGLEMDS}_svc
-	local igif=$1
+	local svc=${facet}_svc
 
 	if ! ${rcmd} test -b ${devname}; then
 		opts=$(csa_add "$opts" -o loop)
 	fi
 
-	echo "file-level backup/restore on ${SINGLEMDS}:${devname}"
+	echo "file-level backup/restore on $facet:${devname}"
 
 	# step 1: build mount point
 	${rcmd} mkdir -p $mntpt
@@ -6011,11 +6626,12 @@ mds_backup_restore() {
 	# step 6: umount
 	${rcmd} umount -d $mntpt || return 4
 	# step 7: reformat external journal if needed
-	reformat_external_journal || return 5
+	reformat_external_journal $facet || return 5
 	# step 8: reformat dev
 	echo "reformat new device"
-	add ${SINGLEMDS} $(mkfs_opts ${SINGLEMDS}) --backfstype ldiskfs \
-		--reformat $devname > /dev/null || return 6
+	add $facet $(mkfs_opts $facet ${devname}) --backfstype ldiskfs \
+		--reformat ${devname} $(mdsvdevname $(facet_number $facet)) \
+		> /dev/null || exit 6
 	# step 9: mount dev
 	${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 7
 	# step 10: restore metadata
@@ -6037,17 +6653,18 @@ mds_backup_restore() {
 
 # remove OI files
 mds_remove_ois() {
-	local devname=$(mdsdevname ${SINGLEMDS//mds/})
+	local facet=$1
+	local idx=$2
+	local devname=$(mdsdevname $(facet_number $facet))
 	local mntpt=$(facet_mntpt brpt)
-	local rcmd="do_facet ${SINGLEMDS}"
-	local idx=$1
+	local rcmd="do_facet $facet"
 	local opts=${MDS_MOUNT_OPTS}
 
 	if ! ${rcmd} test -b ${devname}; then
 		opts=$(csa_add "$opts" -o loop)
 	fi
 
-	echo "remove OI files: idx=${idx}"
+	echo "removing OI files on $facet: idx=${idx}"
 
 	# step 1: build mount point
 	${rcmd} mkdir -p $mntpt
@@ -6079,41 +6696,28 @@ generate_logname() {
 	echo "$TESTLOG_PREFIX.$TESTNAME.$logname.$(hostname -s).log"
 }
 
-# mkdir directory on different MDTs
+# make directory on different MDTs
 test_mkdir() {
 	local option
 	local parent
 	local child
 	local path
-	local dir
 	local rc=0
 
-	if [ $# -eq 2 ]; then
-		option=$1
-		path=$2
-	else
-		path=$1
-	fi
-
-	child=${path##*/}
-	parent=${path%/*}
+	case $# in
+		1) path=$1;;
+		2) option=$1
+		   path=$2;;
+		*) error "Only creating single directory is supported";;
+	esac
 
-	if [ "$parent" == "$child" ]; then
-		parent=$(pwd)
-	fi
+	child=$(basename $path)
+	parent=$(dirname $path)
 
-	if [ "$option" == "-p" -a -d ${parent}/${child} ]; then
+	if [ "$option" == "-p" -a -d $parent/$child ]; then
 		return $rc
 	fi
 
-	# it needs to check whether there is further / in child
-	dir=$(echo $child | awk -F '/' '{print $2}')
-	if [ ! -z "$dir" ]; then
-		local subparent=$(echo $child | awk -F '/' '{ print $1 }')
-		parent=${parent}"/"${subparent}
-		child=$dir
-	fi
-
 	if [ ! -d ${parent} ]; then
 		if [ "$option" == "-p" ]; then
 			mkdir -p ${parent}
@@ -6123,19 +6727,30 @@ test_mkdir() {
 	fi
 
 	if [ $MDSCOUNT -le 1 ]; then
-		mkdir $option ${parent}/${child} || rc=$?
+		mkdir $option $parent/$child || rc=$?
 	else
 		local mdt_idx=$($LFS getstripe -M $parent)
+		local test_num=$(echo $testnum | sed -e 's/[^0-9]*//g')
 
 		if [ "$mdt_idx" -ne 0 ]; then
-			mkdir $option ${parent}/${child} || rc=$?
-			return $rc
+			mkdir $option $parent/$child || rc=$?
+		else
+			mdt_idx=$((test_num % MDSCOUNT))
+			echo "mkdir $mdt_idx for $parent/$child"
+			$LFS setdirstripe -i $mdt_idx $parent/$child || rc=$?
 		fi
-
-		local test_num=$(echo $testnum | sed -e 's/[^0-9]*//g')
-		local mdt_idx=$((test_num % MDSCOUNT))
-		echo "mkdir $mdt_idx for ${parent}/${child}"
-		$LFS setdirstripe -i $mdt_idx ${parent}/${child} || rc=$?
 	fi
 	return $rc
 }
+
+# find the smallest and not in use file descriptor
+free_fd()
+{
+        local max_fd=$(ulimit -n)
+        local fd=3
+        while [[ $fd -le $max_fd && -e /proc/self/fd/$fd ]]; do
+                ((++fd))
+        done
+        [ $fd -lt $max_fd ] || error "finding free file descriptor failed"
+        echo $fd
+}