X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ftest-framework.sh;h=0da57f60de7839520d6bb125da3ec3dfa0b2255c;hp=989031b87241f7083d36d64526b538bbbf000abb;hb=d16729c7e2e7d8b825181686ad42def3121f8445;hpb=fa423043b3bb152730f698cb5bf6126e2df882c2

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 989031b..0da57f6 100644
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -15,8 +15,8 @@ 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.
@@ -185,6 +185,7 @@ 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}
@@ -297,6 +298,7 @@ init_test_env() {
     fi
 
     export SHUTDOWN_ATTEMPTS=${SHUTDOWN_ATTEMPTS:-3}
+	export OSD_TRACK_DECLARES_LBUG=${OSD_TRACK_DECLARES_LBUG:-"yes"}
 
     # command line
 
@@ -490,6 +492,7 @@ 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
@@ -585,6 +588,18 @@ unload_modules() {
     return 0
 }
 
+fs_log_size() {
+	local facet=${1:-$SINGLEMDS}
+	local fstype=$(facet_fstype $facet)
+	local size=0
+	case $fstype in
+		ldiskfs) size=50;; # largest seen is 44, leave some headroom
+		zfs)     size=400;; # largest seen is 384
+	esac
+
+	echo -n $size
+}
+
 check_gss_daemon_nodes() {
     local list=$1
     dname=$2
@@ -796,6 +811,140 @@ 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
+}
+
+#
+# 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
+}
+
+#
+# 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 $poolname"
+	fi
+}
+
+#
 # This and set_osd_param() shall be used to access OSD parameters
 # once existed under "obdfilter":
 #
@@ -923,6 +1072,11 @@ mount_facet() {
 		opts=$(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
@@ -969,39 +1123,6 @@ 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
@@ -1016,9 +1137,14 @@ 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)
@@ -1476,7 +1602,6 @@ reboot_facet() {
 	if [ "$FAILURE_MODE" = HARD ]; then
 		reboot_node $(facet_active_host $facet)
 	else
-		refresh_disk ${facet}
 		sleep 10
 	fi
 }
@@ -1748,30 +1873,43 @@ 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
+
+	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() {
@@ -1781,7 +1919,10 @@ 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'
 }
 
@@ -1897,30 +2038,44 @@ wait_recovery_complete () {
 }
 
 wait_mds_ost_sync () {
-    # just because recovery is done doesn't mean we've finished
-    # orphan cleanup. Wait for llogs to get synchronized.
-    echo "Waiting for orphan cleanup..."
-    # MAX value includes time needed for MDS-OST reconnection
-    local MAX=$(( TIMEOUT * 2 ))
-    local WAIT=0
-    while [ $WAIT -lt $MAX ]; do
-        local -a sync=($(do_nodes $(comma_list $(mdts_nodes)) \
-            "$LCTL get_param -n osp.*osc*.old_sync_processed"))
-        local con=1
-        local i
-        for ((i=0; i<${#sync[@]}; i++)); do
-            [ ${sync[$i]} -eq 1 ] && continue
-            # there is a not finished MDS-OST synchronization
-            con=0
-            break;
-        done
-        sleep 2 # increase waiting time and cover statfs cache
-        [ ${con} -eq 1 ] && return 0
-        echo "Waiting $WAIT secs for $facet mds-ost sync done."
-        WAIT=$((WAIT + 2))
-    done
-    echo "$facet recovery not done in $MAX sec. $STATUS"
-    return 1
+	# just because recovery is done doesn't mean we've finished
+	# orphan cleanup. Wait for llogs to get synchronized.
+	echo "Waiting for orphan cleanup..."
+	# MAX value includes time needed for MDS-OST reconnection
+	local MAX=$(( TIMEOUT * 2 ))
+	local WAIT=0
+	local new_wait=true
+	local list=$(comma_list $(mdts_nodes))
+	local cmd="$LCTL get_param -n osp.*osc*.old_sync_processed"
+	if ! do_facet $SINGLEMDS \
+		"$LCTL list_param osp.*osc*.old_sync_processed 2> /dev/null"
+	then
+		# old way, use mds_sync
+		new_wait=false
+		list=$(comma_list $(osts_nodes))
+		cmd="$LCTL get_param -n obdfilter.*.mds_sync"
+	fi
+	while [ $WAIT -lt $MAX ]; do
+		local -a sync=($(do_nodes $list "$cmd"))
+		local con=1
+		local i
+		for ((i=0; i<${#sync[@]}; i++)); do
+			if $new_wait; then
+				[ ${sync[$i]} -eq 1 ] && continue
+			else
+				[ ${sync[$i]} -eq 0 ] && continue
+			fi
+			# there is a not finished MDS-OST synchronization
+			con=0
+			break;
+		done
+		sleep 2 # increase waiting time and cover statfs cache
+		[ ${con} -eq 1 ] && return 0
+		echo "Waiting $WAIT secs for $facet mds-ost sync done."
+		WAIT=$((WAIT + 2))
+	done
+	echo "$facet recovery not done in $MAX sec. $STATUS"
+	return 1
 }
 
 wait_destroy_complete () {
@@ -2122,7 +2277,7 @@ 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"
@@ -2130,6 +2285,21 @@ 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}"
@@ -2167,8 +2337,12 @@ ost_evict_client() {
 }
 
 fail() {
-    facet_failover $* || error "failover: $?"
-    clients_up || error "post-failover df: $?"
+	local facets=$1
+	local clients=${CLIENTS:-$HOSTNAME}
+
+	facet_failover $* || error "failover: $?"
+	wait_clients_import_state "$clients" "$facets" FULL
+	clients_up || error "post-failover df: $?"
 }
 
 fail_nodf() {
@@ -2179,7 +2353,6 @@ 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
@@ -2402,6 +2575,24 @@ 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
@@ -2587,13 +2778,26 @@ 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
 }
 
 ostdevname() {
@@ -2853,6 +3057,9 @@ mkfs_opts() {
 
 		if [ $fstype == ldiskfs ]; then
 			fs_mkfs_opts+=${MDSJOURNALSIZE:+" -J size=$MDSJOURNALSIZE"}
+			if [ ! -z $EJOURNAL ]; then
+				fs_mkfs_opts+=${MDSJOURNALSIZE:+" device=$EJOURNAL"}
+			fi
 			fs_mkfs_opts+=${MDSISIZE:+" -i $MDSISIZE"}
 		fi
 	fi
@@ -2973,7 +3180,7 @@ 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
 }
@@ -3155,18 +3362,15 @@ init_facets_vars () {
 	if ! remote_mds_nodsh; then
 		for num in $(seq $MDSCOUNT); do
 			DEVNAME=`mdsdevname $num`
-			eval export MDSDEV${num}=$DEVNAME
 			init_facet_vars mds$num $DEVNAME $MDS_MOUNT_OPTS
 		done
 	fi
 
-	eval export MGSDEV=$(mgsdevname)
 	combined_mgs_mds || init_facet_vars mgs $(mgsdevname) $MGS_MOUNT_OPTS
 
 	if ! remote_ost_nodsh; then
 		for num in $(seq $OSTCOUNT); do
 			DEVNAME=$(ostdevname $num)
-			eval export OSTDEV${num}=$DEVNAME
 			init_facet_vars ost$num $DEVNAME $OST_MOUNT_OPTS
 		done
 	fi
@@ -3219,26 +3423,20 @@ 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 == $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
+		if [ $JOBID_VAR = "existing" ]; then
+			echo "keeping jobstats as $current_jobid_var"
+		elif [ $current_jobid_var != $JOBID_VAR ]; then
+			echo "seting jobstats to $JOBID_VAR"
+
+			set_conf_param_and_check client			\
+				"$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
@@ -3342,17 +3540,24 @@ is_mounted () {
 }
 
 is_empty_dir() {
-    [ $(find $1 -maxdepth 1 -print | wc -l) = 1 ] && return 0
-    return 1
+	[ $(find $1 -maxdepth 1 -print | wc -l) = 1 ] && return 0
+	return 1
 }
 
 # empty lustre filesystem may have empty directories lost+found and .lustre
 is_empty_fs() {
-    [ $(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 0
-    [ ! -d $1/.lustre ] || is_empty_dir $1/.lustre && return 0
-    return 1
+	# 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
+	if [ $(lustre_version_code $SINGLEMDS) -lt $(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
 }
 
 check_and_setup_lustre() {
@@ -3413,6 +3618,15 @@ check_and_setup_lustre() {
         set_default_debug_nodes $(comma_list $(nodes_list))
     fi
 
+	if [ $(lower $OSD_TRACK_DECLARES_LBUG) == 'yes' ] ; then
+		local facets="$(get_facets OST),$(get_facets MDS),mgs"
+		local nodes="$(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
@@ -3543,17 +3757,55 @@ generate_db() {
     done
 }
 
+# Run lfsck on server node if lfsck can't be found on client (LU-2571)
+run_lfsck_remote() {
+	local cmd="$LFSCK_BIN -c -l --mdsdb $MDSDB --ostdb $OSTDB_LIST $MOUNT"
+	local client=$1
+	local mounted=true
+	local rc=0
+
+	#Check if lustre is already mounted
+	do_rpc_nodes $client is_mounted $MOUNT || mounted=false
+	if ! $mounted; then
+		zconf_mount $client $MOUNT ||
+			error "failed to mount Lustre on $client"
+	fi
+	#Run lfsck
+	echo $cmd
+	do_node $node $cmd || rc=$?
+	#Umount if necessary
+	if ! $mounted; then
+		zconf_umount $client $MOUNT ||
+			error "failed to unmount Lustre on $client"
+	fi
+
+	[ $rc -le $FSCK_MAX_ERR ] ||
+		error "$cmd returned $rc, should be <= $FSCK_MAX_ERR"
+	echo "lfsck finished with rc=$rc"
+
+	return $rc
+}
+
 run_lfsck() {
-    local cmd="$LFSCK_BIN -c -l --mdsdb $MDSDB --ostdb $OSTDB_LIST $MOUNT"
-    echo $cmd
-    local rc=0
-    eval $cmd || rc=$?
-    [ $rc -le $FSCK_MAX_ERR ] || \
-        error "$cmd returned $rc, should be <= $FSCK_MAX_ERR"
-    echo "lfsck finished with rc=$rc"
+	local facets="client $SINGLEMDS"
+	local found=false
+	local facet
+	local node
+	local rc=0
 
-    rm -rvf $MDSDB* $OSTDB* || true
-    return 0
+	for facet in $facets; do
+		node=$(facet_active_host $facet)
+		if check_progs_installed $node $LFSCK_BIN; then
+			found=true
+			break
+		fi
+	done
+	! $found && error "None of \"$facets\" supports lfsck"
+
+	run_lfsck_remote $node || rc=$?
+
+	rm -rvf $MDSDB* $OSTDB* || true
+	return $rc
 }
 
 check_and_cleanup_lustre() {
@@ -3795,11 +4047,11 @@ drop_reint_reply() {
 }
 
 drop_update_reply() {
-# OBD_FAIL_MDS_OBJ_UPDATE_NET
+# OBD_FAIL_UPDATE_OBJ_NET_REP
 	local index=$1
 	shift 1
 	RC=0
-	do_facet mds${index} lctl set_param fail_loc=0x188
+	do_facet mds${index} lctl set_param fail_loc=0x1701
 	do_facet client "$@" || RC=$?
 	do_facet mds${index} lctl set_param fail_loc=0
 	return $RC
@@ -3947,28 +4199,33 @@ 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" > $LOGDIR/err
+	else
+		echo "$@" > $LOGDIR/err
+	fi
 }
 
 exit_status () {
@@ -3980,21 +4237,21 @@ 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"
 error_ignore() {
-    local TYPE="IGNORE (bz$1)"
-    shift
-    error_noexit "$@"
+	local TYPE="IGNORE (bz$1)"
+	shift
+	error_noexit "$@"
 }
 
 error_and_remount() {
@@ -4004,7 +4261,7 @@ error_and_remount() {
 }
 
 skip_env () {
-    $FAIL_ON_SKIP_ENV && error false $@ || skip $@
+	$FAIL_ON_SKIP_ENV && error false $@ || skip $@
 }
 
 skip() {
@@ -4084,7 +4341,6 @@ run_test() {
             return $?
         fi
         LAST_SKIPPED="y"
-        echo -n "."
         return 0
     fi
 
@@ -4461,64 +4717,121 @@ 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
 
-    echo $NODES_sort
+	nodes_sort=$(for i in $nodes; do echo $i; done | sort -u)
+	echo -n $nodes_sort
 }
 
-osts_nodes () {
-    local facets=$(get_facets OST)
-    local nodes=$(facets_nodes $facets)
+# Get all of the active MDS nodes.
+mdts_nodes () {
+	echo -n $(facets_nodes $(get_facets MDS))
+}
 
-    echo $nodes
+# Get all of the active OSS nodes.
+osts_nodes () {
+	echo -n $(facets_nodes $(get_facets OST))
 }
 
+# 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 () {
@@ -4599,10 +4912,11 @@ 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 () {
@@ -4819,21 +5133,33 @@ 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() {
@@ -4843,9 +5169,17 @@ check_catastrophe() {
 
 	[ -z "$rnodes" ] && return 0
 
-	do_nodes "$rnodes" "rc=\\\$([ -f $C ] && echo \\\$(< $C) || echo 0);
+	local data
+	data=$(do_nodes "$rnodes" "rc=\\\$([ -f $C ] &&
+		echo \\\$(< $C) || echo 0);
 		if [ \\\$rc -ne 0 ]; then echo \\\$(hostname): \\\$rc; fi
-		exit \\\$rc;"
+		exit \\\$rc")
+	local rc=$?
+	if [ -n "$data" ]; then
+	    echo $data
+	    return $rc
+	fi
+	return 0
 }
 
 # CMD: determine mds index where directory inode presents
@@ -4907,7 +5241,7 @@ convert_facet2label() {
 }
 
 get_clientosc_proc_path() {
-    echo "${1}-osc-[^M]*"
+    echo "${1}-osc-*"
 }
 
 get_lustre_version () {
@@ -4971,7 +5305,7 @@ _wait_import_state () {
     local CONN_STATE
     local i=0
 
-    CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+	CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2 | uniq)
     while [ "${CONN_STATE}" != "${expected}" ]; do
         if [ "${expected}" == "DISCONN" ]; then
             # for disconn we can check after proc entry is removed
@@ -4984,7 +5318,8 @@ _wait_import_state () {
             error "can't put import for $CONN_PROC into ${expected} state after $i sec, have ${CONN_STATE}" && \
             return 1
         sleep 1
-        CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2)
+	# Add uniq for multi-mount case
+	CONN_STATE=$($LCTL get_param -n $CONN_PROC 2>/dev/null | cut -f2 | uniq)
         i=$(($i + 1))
     done
 
@@ -5003,6 +5338,14 @@ wait_import_state() {
     done
 }
 
+wait_import_state_mount() {
+	if ! is_mounted $MOUNT && ! is_mounted $MOUNT2; then
+		return 0
+	fi
+
+	wait_import_state $*
+}
+
 # One client request could be timed out because server was not ready
 # when request was sent by client.
 # The request timeout calculation details :
@@ -5043,20 +5386,42 @@ request_timeout () {
     echo $(( init_connect_timeout + at_min ))
 }
 
-wait_osc_import_state() {
+_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.${ost}.ost_server_uuid"
+	local param="osc.${ost}.ost_server_uuid"
+	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)))
 
-	if ! do_rpc_nodes "$(facet_host $facet)" \
-			_wait_import_state $expected $param $maxtime; then
+	#During setup time, the osc might not be setup, it need wait
+	#until list_param can return valid value. And also if there
+	#are mulitple osc entries we should list all of them before
+	#go to wait.
+	local params=$($LCTL list_param $param 2>/dev/null || true)
+	while [ -z "$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
+				return 1
+			fi
+		fi
+		sleep 1
+		i=$((i + 1))
+		params=$($LCTL list_param $param 2>/dev/null || true)
+	done
+
+	if ! do_rpc_nodes "$(facet_active_host $facet)" \
+			wait_import_state $expected "$params" $maxtime; then
 		error "import is not in ${expected} state"
 		return 1
 	fi
@@ -5064,6 +5429,21 @@ 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-*"
 }
@@ -5075,7 +5455,9 @@ 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 $@ "
 }
 
@@ -5101,7 +5483,7 @@ wait_clients_import_state () {
     local params=$(expand_list $params $proc_path)
     done
 
-	if ! do_rpc_nodes "$list" wait_import_state $expected $params; then
+	if ! do_rpc_nodes "$list" wait_import_state_mount $expected $params; then
 		error "import is not in ${expected} state"
 		return 1
 	fi
@@ -5704,6 +6086,33 @@ 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
@@ -5719,7 +6128,7 @@ 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=$?; break; }
 	done
@@ -5880,6 +6289,7 @@ mds_backup_restore() {
 	local metadata=${TMP}/backup_restore.tgz
 	local opts=${MDS_MOUNT_OPTS}
 	local svc=${SINGLEMDS}_svc
+	local igif=$1
 
 	if ! ${rcmd} test -b ${devname}; then
 		opts=$(csa_add "$opts" -o loop)
@@ -5893,6 +6303,10 @@ mds_backup_restore() {
 	${rcmd} rm -f $metaea $metadata
 	# step 3: mount dev
 	${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 1
+	if [ ! -z $igif ]; then
+		# step 3.5: rm .lustre
+		${rcmd} rm -rf $mntpt/ROOT/.lustre || return 1
+	fi
 	# step 4: backup metaea
 	echo "backup EA"
 	${rcmd} "cd $mntpt && getfattr -R -d -m '.*' -P . > $metaea && cd -" ||
@@ -5906,8 +6320,9 @@ mds_backup_restore() {
 	reformat_external_journal || return 5
 	# step 8: reformat dev
 	echo "reformat new device"
-	add ${SINGLEMDS} $(mkfs_opts ${SINGLEMDS}) --backfstype ldiskfs \
-		--reformat $devname > /dev/null || return 6
+	add ${SINGLEMDS} $(mkfs_opts ${SINGLEMDS} ${devname}) --backfstype \
+		ldiskfs --reformat ${devname} $(mdsvdevname 1) > /dev/null ||
+		exit 6
 	# step 9: mount dev
 	${rcmd} mount -t ldiskfs $opts $devname $mntpt || return 7
 	# step 10: restore metadata
@@ -5971,41 +6386,28 @@ generate_logname() {
 	echo "$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}
@@ -6015,19 +6417,18 @@ 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
 }