X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=4ef21eb60461cb05e31635d8acf7aee9361510cd;hb=3f5fc6412d3ae2d3a57f8fdb8a457f35e9d9576e;hp=62164fb636d3f201aa27ba256bb97b68e5888541;hpb=3ca3500252765e7afbfe99cbbb7a0e7eb9df01bc;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index 62164fb..4ef21eb 100644
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,11 +11,11 @@ SRCDIR=$(dirname $0)
 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test:
+# bug number for skipped test:    3815
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
+# bug number for skipped test:4178         4176
+ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 221 223b 31a"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-# skip test cases failed before landing - Jinshan
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 12a 12b 12n 13 24 30a 31a 34 35 36 58 59"
-ALWAYS_EXCEPT="$ALWAYS_EXCEPT 110a 200 201 221 222a 223a 223b 225"
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 
@@ -30,9 +30,9 @@ MCREATE=${MCREATE:-mcreate}
 MOUNT_2=${MOUNT_2:-"yes"}
 FAIL_ON_ERROR=false
 
-if [ $MDSCOUNT -ge 2 ]; then
-	skip_env "Only run with single MDT for now" && exit
-fi
+# script only handles up to 10 MDTs (because of MDT_PREFIX)
+[ $MDSCOUNT -gt 9 ] &&
+	error "script cannot handle more than 9 MDTs, please fix" && exit
 
 check_and_setup_lustre
 
@@ -41,112 +41,269 @@ if [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.53) ]]; then
 fi
 
 # $RUNAS_ID may get set incorrectly somewhere else
-[ $UID -eq 0 -a $RUNAS_ID -eq 0 ] &&
-	error "\$RUNAS_ID set to 0, but \$UID is also 0!"
-
+if [[ $UID -eq 0 && $RUNAS_ID -eq 0 ]]; then
+	skip_env "\$RUNAS_ID set to 0, but \$UID is also 0!" && exit
+fi
 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
 
 build_test_filter
 
-# the standard state when starting a test is
-# - no copytool
-# - MOUNT2 done
-# as some test changes the default, we need to re-make it
+#
+# In order to test multiple remote HSM agents, a new facet type named "AGT" and
+# the following associated variables are added:
+#
+# AGTCOUNT: number of agents
+# AGTDEV{N}: target HSM mount point (root path of the backend)
+# agt{N}_HOST: hostname of the agent agt{N}
+# SINGLEAGT: facet of the single agent
+#
+# The number of agents is initialized as the number of remote client nodes.
+# By default, only single copytool is started on a remote client/agent. If there
+# was no remote client, then the copytool will be started on the local client.
+#
+init_agt_vars() {
+	local n
+	local agent
+
+	export AGTCOUNT=${AGTCOUNT:-$((CLIENTCOUNT - 1))}
+	[[ $AGTCOUNT -gt 0 ]] || AGTCOUNT=1
+
+	export SHARED_DIRECTORY=${SHARED_DIRECTORY:-$TMP}
+	if [[ $CLIENTCOUNT -gt 1 ]] &&
+		! check_shared_dir $SHARED_DIRECTORY $CLIENTS; then
+		skip_env "SHARED_DIRECTORY should be accessible"\
+			 "on all client nodes"
+		exit 0
+	fi
+
+	for n in $(seq $AGTCOUNT); do
+		eval export AGTDEV$n=\$\{AGTDEV$n:-"$SHARED_DIRECTORY/arc$n"\}
+		agent=CLIENT$((n + 1))
+		if [[ -z "${!agent}" ]]; then
+			[[ $CLIENTCOUNT -eq 1 ]] && agent=CLIENT1 ||
+				agent=CLIENT2
+		fi
+		eval export agt${n}_HOST=\$\{agt${n}_HOST:-${!agent}\}
+	done
+
+	export SINGLEAGT=${SINGLEAGT:-agt1}
+
+	export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
+	export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
+	export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
+	HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
+	HSM_ARCHIVE_NUMBER=2
+
+	# The test only support up to 10 MDTs
+	MDT_PREFIX="mdt.$FSNAME-MDT000"
+	HSM_PARAM="${MDT_PREFIX}0.hsm"
+
+	# archive is purged at copytool setup
+	HSM_ARCHIVE_PURGE=true
+}
+
+# Get the backend root path for the given agent facet.
+copytool_device() {
+	local facet=$1
+	local dev=AGTDEV$(facet_number $facet)
+
+	echo -n ${!dev}
+}
+
+# Stop copytool and unregister an existing changelog user.
 cleanup() {
 	copytool_cleanup
-	if ! is_mounted $MOUNT2
-	then
-		mount_client $MOUNT2
-	fi
 	changelog_cleanup
+	cdt_set_sanity_policy
 }
 
-export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
-export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
-export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
-HSM_ARCHIVE=${HSM_ARCHIVE:-$TMP/arc}
-HSM_ARCHIVE_NUMBER=2
-
-MDT_PARAM="mdt.$FSNAME-MDT0000"
-HSM_PARAM="$MDT_PARAM.hsm"
+get_mdt_devices() {
+	local mdtno
+	# get MDT device for each mdc
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		MDT[$idx]=$($LCTL get_param -n \
+			mdc.$FSNAME-MDT000${idx}-mdc-*.mds_server_uuid |
+			awk '{gsub(/_UUID/,""); print $1}' | head -1)
+	done
+}
 
-# archive is purged at copytool setup
-HSM_ARCHIVE_PURGE=true
+search_copytools() {
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
+}
 
 search_and_kill_copytool() {
-	echo "Killing existing copy tools"
-	killall -q $HSMTOOL_BASE || true
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+
+	echo "Killing existing copytools on $agents"
+	do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
 }
 
 copytool_setup() {
-	if pkill -CONT -x $HSMTOOL_BASE; then
-		echo "Wakeup copytool"
-		return
+	local facet=${1:-$SINGLEAGT}
+	local lustre_mntpnt=${2:-$MOUNT}
+	local arc_id=$3
+	local hsm_root=${4:-$(copytool_device $facet)}
+	local agent=$(facet_active_host $facet)
+
+	if [[ -z "$arc_id" ]] &&
+		do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
+			echo "Wakeup copytool $facet on $agent"
+			return 0
 	fi
 
 	if $HSM_ARCHIVE_PURGE; then
-		echo "Purging archive"
-		rm -rf $HSM_ARCHIVE/*
+		echo "Purging archive on $agent"
+		do_facet $facet "rm -rf $hsm_root/*"
 	fi
 
-	echo "Starting copytool"
-	mkdir -p $HSM_ARCHIVE
+	echo "Starting copytool $facet on $agent"
+	do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
 	# bandwidth is limited to 1MB/s so the copy time is known and
 	# independent of hardware
-	local CMD="$HSMTOOL $HSMTOOL_VERBOSE --hsm-root $HSM_ARCHIVE"
-	CMD=$CMD" --daemon --bandwidth 1 $MOUNT"
-	[[ -z "$1" ]] || CMD+=" --archive $1"
-
-	echo "$CMD"
-	$CMD  &
+	local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
+	[[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
+	cmd+=" --bandwidth 1 $lustre_mntpnt"
+
+	# Redirect the standard output and error to a log file which
+	# can be uploaded to Maloo.
+	local prefix=$TESTLOG_PREFIX
+	[[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
+	local copytool_log=$prefix.copytool${arc_id}_log.$agent.log
+
+	do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1" ||
+		error "start copytool $facet on $agent failed"
 	trap cleanup EXIT
 }
 
 copytool_cleanup() {
 	trap - EXIT
-	pkill -INT -x $HSMTOOL_BASE || return 0
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	local mdtno
+	local idx
+	local oldstate
+	local mdt_hsmctrl
+
+	do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
 	sleep 1
-	echo "Copytool is stopped"
+	echo "Copytool is stopped on $agents"
+
+	# clean all CDTs orphans requests from previous tests
+	# that would otherwise need to timeout to clear.
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		idx=$(($mdtno - 1))
+		mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
+		oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
+				   "$mdt_hsmctrl")
+		# skip already stop[ed,ing] CDTs
+		echo $oldstate | grep stop && continue
+
+		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
+		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+			"stopped" 20 ||
+			error "mds${mdtno} cdt state is not stopped"
+		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
+		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+			"$oldstate" 20 ||
+			error "mds${mdtno} cdt state is not $oldstate"
+	done
 }
 
 copytool_suspend() {
-	pkill -STOP -x $HSMTOOL_BASE || return 0
-	echo "Copytool is suspended"
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+
+	do_nodesv $agents "pkill -STOP -x $HSMTOOL_BASE" || return 0
+	echo "Copytool is suspended on $agents"
 }
 
 copytool_remove_backend() {
 	local fid=$1
-	local be=$(find $HSM_ARCHIVE -name $fid)
+	local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
 	echo "Remove from backend: $fid = $be"
-	rm -f $be
+	do_facet $SINGLEAGT rm -f $be
 }
 
 import_file() {
-	$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE \
-		--import $1 $2 $MOUNT || error "import of $1 to $2 failed"
+	do_facet $SINGLEAGT \
+		"$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
+		--import $1 $2 $MOUNT" ||
+		error "import of $1 to $2 failed"
 }
 
 make_archive() {
 	local file=$HSM_ARCHIVE/$1
-	mkdir -p $(dirname $file)
-	dd if=/dev/urandom of=$file count=32 bs=1000000 ||
+	do_facet $SINGLEAGT mkdir -p $(dirname $file)
+	do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
 		error "cannot create $file"
 }
 
+copy2archive() {
+	local file=$HSM_ARCHIVE/$2
+	do_facet $SINGLEAGT mkdir -p $(dirname $file)
+	do_facet $SINGLEAGT cp -p $1 $file || error "cannot copy $1 to $file"
+}
+
+mdts_set_param() {
+	local arg=$1
+	local key=$2
+	local value=$3
+	local mdtno
+	local rc=0
+	if [[ "$value" != "" ]]; then
+		value="=$value"
+	fi
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		local facet=mds${mdtno}
+		# if $arg include -P option, run 1 set_param per MDT on the MGS
+		# else, run set_param on each MDT
+		[[ $arg = *"-P"* ]] && facet=mgs
+		do_facet $facet $LCTL set_param $arg mdt.${MDT[$idx]}.$key$value
+		[[ $? != 0 ]] && rc=1
+	done
+	return $rc
+}
+
+mdts_check_param() {
+	local key="$1"
+	local target="$2"
+	local timeout="$3"
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		wait_result mds${mdtno} \
+			"$LCTL get_param -n $MDT_PREFIX${idx}.$key" "$target" \
+			$timeout ||
+			error "$key state is not '$target' on mds${mdtno}"
+	done
+}
+
 changelog_setup() {
-	CL_USER=$(do_facet $SINGLEMDS $LCTL --device $MDT0\
-		  changelog_register -n)
-	do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog_mask="+hsm"
-	$LFS changelog_clear $MDT0 $CL_USER 0
+	CL_USERS=()
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		local cl_user=$(do_facet mds${mdtno} $LCTL \
+			     --device ${MDT[$idx]} \
+			     changelog_register -n)
+		CL_USERS+=($cl_user)
+		do_facet mds${mdtno} lctl set_param \
+			mdd.${MDT[$idx]}.changelog_mask="+hsm"
+		$LFS changelog_clear ${MDT[$idx]} $cl_user 0
+	done
 }
 
 changelog_cleanup() {
-#	$LFS changelog $MDT0
-	[[ -n "$CL_USER" ]] || return 0
-
-	$LFS changelog_clear $MDT0 $CL_USER 0
-	do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $CL_USER
-	CL_USER=
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		local idx=$(($mdtno - 1))
+		[[ -z  ${CL_USERS[$idx]} ]] && continue
+		$LFS changelog_clear ${MDT[$idx]} ${CL_USERS[$idx]} 0
+		do_facet mds${mdtno} lctl --device ${MDT[$idx]} \
+			changelog_deregister ${CL_USERS[$idx]}
+	done
+	CL_USERS=()
 }
 
 changelog_get_flags() {
@@ -166,58 +323,58 @@ get_hsm_param() {
 set_hsm_param() {
 	local param=$1
 	local value=$2
-	do_facet $SINGLEMDS $LCTL set_param -n $HSM_PARAM.$param=$value
+	local opt=$3
+	mdts_set_param "$opt -n" "hsm.$param" "$value"
 	return $?
 }
 
 set_test_state() {
 	local cmd=$1
 	local target=$2
-	do_facet $SINGLEMDS $LCTL set_param $MDT_PARAM.hsm_control=$cmd
-	wait_result $SINGLEMDS "$LCTL get_param -n $MDT_PARAM.hsm_control"\
-		$target 10 || error "cdt state is not $target"
+	mdts_set_param "" hsm_control "$cmd"
+	mdts_check_param hsm_control "$target" 10
 }
 
 cdt_set_sanity_policy() {
-	# clear all
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-nra
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-nbr
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-gc
+	if [[ "$CDT_POLICY_HAD_CHANGED" ]]
+	then
+		# clear all
+		mdts_set_param "" hsm.policy "+NRA"
+		mdts_set_param "" hsm.policy "-NBR"
+		CDT_POLICY_HAD_CHANGED=
+	fi
 }
 
 cdt_set_no_retry() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+nra
+	mdts_set_param "" hsm.policy "+NRA"
+	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_no_retry() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-nra
+	mdts_set_param "" hsm.policy "-NRA"
+	CDT_POLICY_HAD_CHANGED=true
 }
 
-cdt_set_no_blocking_restore() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+nbr
+cdt_set_non_blocking_restore() {
+	mdts_set_param "" hsm.policy "+NBR"
+	CDT_POLICY_HAD_CHANGED=true
 }
 
-cdt_clear_no_blocking_restore() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-nbr
+cdt_clear_non_blocking_restore() {
+	mdts_set_param "" hsm.policy "-NBR"
+	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_mount_state() {
-	# /!\ conf_param and set_param syntax differ +> we cannot use
-	# $MDT_PARAM
-	do_facet $SINGLEMDS $LCTL conf_param -d $FSNAME-MDT0000.mdt.hsm_control
+	mdts_set_param "-P -d" hsm_control ""
 }
 
 cdt_set_mount_state() {
-	# /!\ conf_param and set_param syntax differ +> we cannot use
-	# $MDT_PARAM
-	do_facet $SINGLEMDS $LCTL conf_param $FSNAME-MDT0000.mdt.hsm_control=$1
+	mdts_set_param "-P" hsm_control "$1"
 }
 
 cdt_check_state() {
-	local target=$1
-	wait_result $SINGLEMDS\
-		"$LCTL get_param -n $MDT_PARAM.hsm_control" "$target" 20 ||
-			error "cdt state is not $target"
+	mdts_check_param hsm_control "$1" 20
 }
 
 cdt_disable() {
@@ -242,9 +399,10 @@ cdt_restart() {
 	cdt_set_sanity_policy
 }
 
-need2clients() {
-	if [[ $CLIENTCOUNT -lt 2 ]]; then
-		skip "Need two or more clients, have $CLIENTCOUNT"
+needclients() {
+	local client_count=$1
+	if [[ $CLIENTCOUNT -lt $client_count ]]; then
+		skip "Need $client_count or more clients, have $CLIENTCOUNT"
 		return 1
 	fi
 	return 0
@@ -258,8 +416,7 @@ get_hsm_flags() {
 	local f=$1
 	local u=$2
 
-	if [[ $u == "user" ]]
-	then
+	if [[ $u == "user" ]]; then
 		local st=$($RUNAS $LFS hsm_state $f)
 	else
 		local st=$($LFS hsm_state $f)
@@ -301,15 +458,13 @@ check_hsm_flags_user() {
 copy_file() {
 	local f=
 
-	if [[ -d $2 ]]
-	then
+	if [[ -d $2 ]]; then
 		f=$2/$(basename $1)
 	else
 		f=$2
 	fi
 
-	if [[ "$3" != 1 ]]
-	then
+	if [[ "$3" != 1 ]]; then
 		f=${f/$DIR/$DIR2}
 	fi
 	rm -f $f
@@ -324,9 +479,24 @@ make_small() {
         path2fid $1 || error "cannot get fid on $1"
 }
 
+make_small_sync() {
+	dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
+		error "cannot create $1"
+	path2fid $1 || error "cannot get fid on $1"
+}
+
+cleanup_large_files() {
+	local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
+		      sed 's/%//g')
+	[ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
+}
+
 make_large_for_striping() {
 	local file2=${1/$DIR/$DIR2}
 	local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -1)
+
+	cleanup_large_files
+
 	dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync ||
 		error "cannot create $file2"
 	path2fid $1 || error "cannot get fid on $1"
@@ -334,6 +504,9 @@ make_large_for_striping() {
 
 make_large_for_progress() {
 	local file2=${1/$DIR/$DIR2}
+
+	cleanup_large_files
+
 	# big file is large enough, so copy time is > 30s
 	# so copytool make 1 progress
 	# size is not a multiple of 1M to avoid stripe
@@ -345,6 +518,9 @@ make_large_for_progress() {
 
 make_large_for_progress_aligned() {
 	local file2=${1/$DIR/$DIR2}
+
+	cleanup_large_files
+
 	# big file is large enough, so copy time is > 30s
 	# so copytool make 1 progress
 	# size is a multiple of 1M to have stripe
@@ -356,6 +532,9 @@ make_large_for_progress_aligned() {
 
 make_large_for_cancel() {
 	local file2=${1/$DIR/$DIR2}
+
+	cleanup_large_files
+
 	# Copy timeout is 100s. 105MB => 105s
 	dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
 		error "cannot create $file2"
@@ -368,54 +547,57 @@ wait_result() {
 	wait_update --verbose $(facet_active_host $facet) "$@"
 }
 
-wait_request_state()
-{
+wait_request_state() {
 	local fid=$1
 	local request=$2
 	local state=$3
-	wait_result $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |\
-				grep $fid | grep action=$request |\
-				cut -f 13 -d ' ' | cut -f 2 -d =" $state 100 ||
-		error "request on $fid is not $state"
+	# 4th arg (mdt index) is optional
+	local mdtidx=${4:-0}
+	local mds=mds$(($mdtidx + 1))
+
+	local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions"
+	cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
+
+	wait_result $mds "$cmd" $state 100 ||
+		error "request on $fid is not $state on $mds"
 }
 
-get_request_state()
-{
+get_request_state() {
 	local fid=$1
 	local request=$2
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |\
-				grep $fid | grep action=$request |\
-				cut -f 13 -d ' ' | cut -f 2 -d ="
+
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
+		"awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 }
 
-get_request_count()
-{
+get_request_count() {
 	local fid=$1
 	local request=$2
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |\
-				grep $fid | grep action=$request | wc -l"
+
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
+		"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
 }
 
-wait_all_done()
-{
+wait_all_done() {
 	local timeout=$1
-	wait_result $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |\
-		egrep 'WAITING|STARTED' " "" $timeout ||
-	error "requests did not complete"
+
+	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
+	cmd+=" | egrep 'WAITING|STARTED'"
+
+	wait_result $SINGLEMDS "$cmd" "" $timeout ||
+		error "requests did not complete"
 }
 
-wait_for_grace_delay()
-{
+wait_for_grace_delay() {
 	local val=$(get_hsm_param grace_delay)
 	sleep $val
 }
 
-my_uuid() {
-	$LCTL get_param -n llite.$FSNAME-*.uuid
-}
+# populate MDT device array
+get_mdt_devices
 
-MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
-	awk '{gsub(/_UUID/,""); print $1}' | head -1)
+# initiate variables
+init_agt_vars
 
 # cleanup from previous bad setup
 search_and_kill_copytool
@@ -430,6 +612,9 @@ cdt_check_state enabled
 echo "Start copytool"
 copytool_setup
 
+echo "Set sanity-hsm HSM policy"
+cdt_set_sanity_policy
+
 # finished requests are quickly removed from list
 set_hsm_param grace_delay 10
 
@@ -583,9 +768,9 @@ test_9() {
 	local f=$DIR/$tdir/$tfile
 	local fid=$(copy_file /etc/passwd $f)
 	# we do not use the default one to be sure
-	local new_an=$((HSM_ARCHIVE_NUMBER+ 1))
+	local new_an=$((HSM_ARCHIVE_NUMBER + 1))
 	copytool_cleanup
-	copytool_setup $new_an
+	copytool_setup $SINGLEAGT $MOUNT $new_an
 	$LFS hsm_archive --archive $new_an $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -595,6 +780,37 @@ test_9() {
 }
 run_test 9 "Use of explict archive number, with dedicated copytool"
 
+test_9a() {
+	needclients 3 || return 0
+
+	local n
+	local file
+	local fid
+
+	copytool_cleanup $(comma_list $(agts_nodes))
+
+	# start all of the copytools
+	for n in $(seq $AGTCOUNT); do
+		copytool_setup agt$n
+	done
+
+	trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
+	# archive files
+	mkdir -p $DIR/$tdir
+	for n in $(seq $AGTCOUNT); do
+		file=$DIR/$tdir/$tfile.$n
+		fid=$(make_small $file)
+
+		$LFS hsm_archive $file || error "could not archive file $file"
+		wait_request_state $fid ARCHIVE SUCCEED
+		check_hsm_flags $file "0x00000009"
+	done
+
+	trap - EXIT
+	copytool_cleanup $(comma_list $(agts_nodes))
+}
+run_test 9a "Multiple remote agents"
+
 test_10a() {
 	# test needs a running copytool
 	copytool_setup
@@ -606,10 +822,10 @@ test_10a() {
 		error "hsm_archive failed"
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	local AFILE=$(ls $HSM_ARCHIVE/*/*/*/*/*/*/$fid) ||
+	local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
 		error "fid $fid not in archive $HSM_ARCHIVE"
 	echo "Verifying content"
-	diff $f $AFILE || error "archived file differs"
+	do_facet $SINGLEAGT diff $f $AFILE || error "archived file differs"
 	echo "Verifying hsm state "
 	check_hsm_flags $f "0x00000009"
 
@@ -627,7 +843,7 @@ test_10b() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
+	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid=$(copy_file /etc/hosts $f)
 	$LFS hsm_archive $f || error "archive request failed"
@@ -646,7 +862,7 @@ test_10c() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
+	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid=$(copy_file /etc/hosts $f)
 	$LFS hsm_set --noarchive $f
@@ -656,9 +872,28 @@ test_10c() {
 }
 run_test 10c "Check forbidden archive"
 
+test_10d() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/hosts $f)
+	$LFS hsm_archive $f || error "cannot archive $f"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local ar=$(get_hsm_archive_id $f)
+	local dflt=$(get_hsm_param default_archive_id)
+	[[ $ar == $dflt ]] ||
+		error "archived file is not on default archive: $ar != $dflt"
+
+	copytool_cleanup
+}
+run_test 10d "Archive a file on the default archive id"
+
 test_11() {
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp /etc/hosts $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/hosts $tdir/$tfile
 	local f=$DIR/$tdir/$tfile
 
 	import_file $tdir/$tfile $f
@@ -666,7 +901,7 @@ test_11() {
 	check_hsm_flags $f "0x0000000d"
 
 	local LSZ=$(stat -c "%s" $f)
-	local ASZ=$(stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
+	local ASZ=$(do_facet $SINGLEAGT stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
 
 	echo "Verifying imported size $LSZ=$ASZ"
 	[[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
@@ -677,7 +912,7 @@ test_11() {
 	local fid=$(path2fid $f)
 	echo "Verifying new fid $fid in archive"
 
-	local AFILE=$(ls $HSM_ARCHIVE/*/*/*/*/*/*/$fid) ||
+	local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
 		error "fid $fid not in archive $HSM_ARCHIVE"
 }
 run_test 11 "Import a file"
@@ -686,8 +921,9 @@ test_12a() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp /etc/hosts $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/hosts $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
 	import_file $tdir/$tfile $f
 	local f=$DIR2/$tdir/$tfile
@@ -701,7 +937,7 @@ test_12a() {
 	echo "Verifying file state: "
 	check_hsm_flags $f "0x00000009"
 
-	diff -q $HSM_ARCHIVE/$tdir/$tfile $f
+	do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
 
 	[[ $? -eq 0 ]] || error "Restored file differs"
 
@@ -713,8 +949,9 @@ test_12b() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp /etc/hosts $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/hosts $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
 	import_file $tdir/$tfile $f
 	echo "Verifying released state: "
@@ -725,7 +962,7 @@ test_12b() {
 	echo "Verifying file state after restore: "
 	check_hsm_flags $f "0x00000009"
 
-	diff -q $HSM_ARCHIVE/$tdir/$tfile $f
+	do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
 
 	[[ $? -eq 0 ]] || error "Restored file differs"
 
@@ -761,7 +998,8 @@ test_12d() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
+	mkdir -p $DIR/$tdir
+
 	local f=$DIR/$tdir/$tfile
 	local fid=$(copy_file /etc/hosts $f)
 	$LFS hsm_restore $f || error "restore of non archived file failed"
@@ -853,7 +1091,7 @@ test_12g() {
 run_test 12g "Restore a released file implicitly"
 
 test_12h() {
-	need2clients || return 0
+	needclients 2 || return 0
 
 	# test needs a running copytool
 	copytool_setup
@@ -902,12 +1140,14 @@ test_12n() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp /etc/hosts $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/hosts $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
 	import_file $tdir/$tfile $f
 
-	cmp /etc/hosts $f || error "Restored file differs"
+	do_facet $SINGLEAGT cmp /etc/hosts $f ||
+		error "Restored file differs"
 
 	$LFS hsm_release $f || error "release of $f failed"
 
@@ -915,6 +1155,53 @@ test_12n() {
 }
 run_test 12n "Import/implicit restore/release"
 
+test_12o() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/hosts $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f || error "release of $f failed"
+
+#define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS		0x152
+	do_facet $SINGLEMDS lctl set_param fail_loc=0x152
+
+	# set no retry action mode
+	cdt_set_no_retry
+
+	diff -q /etc/hosts $f
+	local st=$?
+
+	# we check we had a restore failure
+	wait_request_state $fid RESTORE FAILED
+
+	[[ $st -eq 0 ]] && error "Restore must fail"
+
+	# remove no retry action mode
+	cdt_clear_no_retry
+
+	# check file is still released
+	check_hsm_flags $f "0x0000000d"
+
+	# retry w/o failure injection
+	do_facet $SINGLEMDS lctl set_param fail_loc=0
+
+	diff -q /etc/hosts $f
+	st=$?
+
+	# we check we had a restore done
+	wait_request_state $fid RESTORE SUCCEED
+
+	[[ $st -eq 0 ]] || error "Restored file differs"
+
+	copytool_cleanup
+}
+run_test 12o "Layout-swap failure during Restore leaves file released"
+
 test_13() {
 	# test needs a running copytool
 	copytool_setup
@@ -926,24 +1213,23 @@ test_13() {
 	# populate directory to be imported
 	for d in $(seq 1 10); do
 		local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
-		mkdir -p "$CURR_DIR"
+		do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
 		for f in $(seq 1 10); do
 			CURR_FILE="$CURR_DIR/$tfile.$f"
 			# write file-specific data
-			echo "d=$d, f=$f, dir=$CURR_DIR, file=$CURR_FILE"\
-				> $CURR_FILE
+			do_facet $SINGLEAGT \
+				"echo d=$d, f=$f, dir=$CURR_DIR, "\
+					"file=$CURR_FILE > $CURR_FILE"
 		done
 	done
 	# import to Lustre
 	import_file "$ARC_SUBDIR" $DIR/$tdir
 	# diff lustre content and origin (triggers file restoration)
 	# there must be 10x10 identical files, and no difference
-	local cnt_ok=$(diff -rs $HSM_ARCHIVE/$ARC_SUBDIR \
-		       $DIR/$tdir/$ARC_SUBDIR |
-		grep identical | wc -l)
-	local cnt_diff=$(diff -r $HSM_ARCHIVE/$ARC_SUBDIR \
-			 $DIR/$tdir/$ARC_SUBDIR |
-		wc -l)
+	local cnt_ok=$(do_facet $SINGLEAGT diff -rs $HSM_ARCHIVE/$ARC_SUBDIR \
+		       $DIR/$tdir/$ARC_SUBDIR | grep identical | wc -l)
+	local cnt_diff=$(do_facet $SINGLEAGT diff -r $HSM_ARCHIVE/$ARC_SUBDIR \
+			 $DIR/$tdir/$ARC_SUBDIR | wc -l)
 
 	[ $cnt_diff -eq 0 ] ||
 		error "$cnt_diff imported files differ from read data"
@@ -976,9 +1262,10 @@ test_14() {
 
 	# rebind the archive to the newly created file
 	echo "rebind $fid to $fid2"
-	$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root="$HSM_ARCHIVE"\
-	 --rebind $fid $fid2 $DIR ||
-		error "could not rebind file"
+
+	do_facet $SINGLEAGT \
+		"$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
+		 --rebind $fid $fid2 $DIR" || error "could not rebind file"
 
 	# restore file and compare md5sum
 	local sum2=$(md5sum $f | awk '{print $1}')
@@ -997,7 +1284,7 @@ test_15() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local count=5
-	local tmpfile=$TMP/tmp.$$
+	local tmpfile=$SHARED_DIRECTORY/tmp.$$
 
 	local fids=()
 	local sums=()
@@ -1026,9 +1313,9 @@ test_15() {
 	[[ $nl == $count ]] || error "$nl files in list, $count expected"
 
 	echo "rebind list of files"
-	$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root="$HSM_ARCHIVE"\
-	 --rebind $tmpfile $DIR ||
-		error "could not rebind file list"
+	do_facet $SINGLEAGT \
+		"$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
+		 --rebind $tmpfile $DIR" || error "could not rebind file list"
 
 	# restore files and compare md5sum
 	for i in $(seq 1 $count); do
@@ -1109,19 +1396,41 @@ test_21() {
 	local fid=$(make_small $f)
 	check_hsm_flags $f "0x00000000"
 
+	# LU-4388/LU-4389 - ZFS does not report full number of blocks
+	# used until file is flushed to disk
+	if [  $(facet_fstype ost1) == "zfs" ]; then
+	    # this causes an OST_SYNC rpc to be sent
+	    dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
+	    # clear locks to reread file data
+	    cancel_lru_locks osc
+	fi
+
+	local orig_size=$(stat -c "%s" $f)
+	local orig_blocks=$(stat -c "%b" $f)
+
+	start_full_debug_logging
+
 	$LFS hsm_archive $f || error "could not archive file"
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	[ $(stat -c "%b" $f) -ne "0" ] || error "wrong block number"
-	local sz=$(stat -c "%s" $f)
-	[ $sz -ne "0" ] || error "file size should not be zero"
+	local blocks=$(stat -c "%b" $f)
+	[ $blocks -eq $orig_blocks ] ||
+		error "$f: wrong block number after archive: " \
+		      "$blocks != $orig_blocks"
+	local size=$(stat -c "%s" $f)
+	[ $size -eq $orig_size ] ||
+		error "$f: wrong size after archive: $size != $orig_size"
 
 	# Release and check states
 	$LFS hsm_release $f || error "could not release file"
 	check_hsm_flags $f "0x0000000d"
 
-	[ $(stat -c "%b" $f) -eq "0" ] || error "wrong block number"
-	[ $(stat -c "%s" $f) -eq $sz ] || error "wrong file size"
+	blocks=$(stat -c "%b" $f)
+	[ $blocks -gt 5 ] &&
+		error "$f: too many blocks after release: $blocks > 5"
+	size=$(stat -c "%s" $f)
+	[ $size -ne $orig_size ] &&
+		error "$f: wrong size after release: $size != $orig_size"
 
 	# Check we can release an file without stripe info
 	f=$f.nolov
@@ -1139,6 +1448,8 @@ test_21() {
 	$LFS hsm_release $f || fail "second release should succeed"
 	check_hsm_flags $f "0x0000000d"
 
+	stop_full_debug_logging
+
 	copytool_cleanup
 }
 run_test 21 "Simple release tests"
@@ -1202,47 +1513,297 @@ test_23() {
 }
 run_test 23 "Release does not change a/mtime (utime)"
 
-test_24() {
+test_24a() {
+	local file=$DIR/$tdir/$tfile
+	local fid
+	local atime0
+	local atime1
+	local mtime0
+	local mtime1
+	local ctime0
+	local ctime1
+
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
-
-	local f=$DIR/$tdir/test_mtime
+	rm -f $file
+	fid=$(make_small $file)
 
 	# Create a file and check its states
-	local fid=$(make_small $f)
-	check_hsm_flags $f "0x00000000"
+	check_hsm_flags $file "0x00000000"
 
-	# make mtime is different
+	# Ensure atime is less than mtime and ctime.
 	sleep 1
-	echo "append" >> $f
-	local MTIME=$(stat -c "%Y" $f)
-	local ATIME=$(stat -c "%X" $f)
+	echo >> $file
 
-	$LFS hsm_archive $f || error "could not archive file"
+	atime0=$(stat -c "%X" $file)
+	mtime0=$(stat -c "%Y" $file)
+	ctime0=$(stat -c "%Z" $file)
+
+	[ $atime0 -lt $mtime0 ] ||
+		error "atime $atime0 is not less than mtime $mtime0"
+
+	[ $atime0 -lt $ctime0 ] ||
+		error "atime $atime0 is not less than ctime $ctime0"
+
+	# Archive should not change any timestamps.
+	$LFS hsm_archive $file || error "cannot archive '$file'"
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	# Release and check states
-	$LFS hsm_release $f || error "could not release file"
-	check_hsm_flags $f "0x0000000d"
+	atime1=$(stat -c "%X" $file)
+	mtime1=$(stat -c "%Y" $file)
+	ctime1=$(stat -c "%Z" $file)
+
+	[ $atime0 -eq $atime1 ] ||
+		error "archive changed atime from $atime0 to $atime1"
+
+	[ $mtime0 -eq $mtime1 ] ||
+		error "archive changed mtime from $mtime0 to $mtime1"
+
+	[ $ctime0 -eq $ctime1 ] ||
+		error "archive changed ctime from $ctime0 to $ctime1"
+
+	# Release should not change any timestamps.
+	$LFS hsm_release $file || error "cannot release '$file'"
+	check_hsm_flags $file "0x0000000d"
+
+	atime1=$(stat -c "%X" $file)
+	mtime1=$(stat -c "%Y" $file)
+	ctime1=$(stat -c "%Z" $file)
+
+	[ $atime0 -eq $atime1 ] ||
+		error "release changed atime from $atime0 to $atime1"
+
+	[ $mtime0 -eq $mtime1 ] ||
+		error "release changed mtime from $mtime0 to $mtime1"
+
+	[ $ctime0 -eq $ctime1 ] ||
+		error "release changed ctime from $ctime0 to $ctime1"
+
+	# Restore should not change atime or mtime and should not
+	# decrease ctime.
+	$LFS hsm_restore $file
+	wait_request_state $fid RESTORE SUCCEED
+
+	atime1=$(stat -c "%X" $file)
+	mtime1=$(stat -c "%Y" $file)
+	ctime1=$(stat -c "%Z" $file)
+
+	[ $atime0 -eq $atime1 ] ||
+		error "restore changed atime from $atime0 to $atime1"
+
+	[ $mtime0 -eq $mtime1 ] ||
+		error "restore changed mtime from $mtime0 to $mtime1"
+
+	[ $ctime0 -le $ctime1 ] ||
+		error "restore changed ctime from $ctime0 to $ctime1"
+
+	copytool_cleanup
+
+	# Once more, after unmount and mount.
+	umount_client $MOUNT || error "cannot unmount '$MOUNT'"
+	mount_client $MOUNT || error "cannot mount '$MOUNT'"
+
+	atime1=$(stat -c "%X" $file)
+	mtime1=$(stat -c "%Y" $file)
+	ctime1=$(stat -c "%Z" $file)
+
+	[ $atime0 -eq $atime1 ] ||
+		error "remount changed atime from $atime0 to $atime1"
+
+	[ $mtime0 -eq $mtime1 ] ||
+		error "remount changed mtime from $mtime0 to $mtime1"
+
+	[ $ctime0 -le $ctime1 ] ||
+		error "remount changed ctime from $ctime0 to $ctime1"
+}
+run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
+
+test_24b() {
+	local file=$DIR/$tdir/$tfile
+	local fid
+	local sum0
+	local sum1
+	# LU-3811
+
+	# Test needs a running copytool.
+	copytool_setup
+	mkdir -p $DIR/$tdir
+
+	# Check that root can do HSM actions on a ordinary user's file.
+	rm -f $file
+	fid=$(make_small $file)
+	sum0=$(md5sum $file)
+
+	chown $RUNAS_ID:$RUNAS_GID $file ||
+		error "cannot chown '$file' to '$RUNAS_ID'"
+
+	chmod ugo-w $DIR/$tdir ||
+		error "cannot chmod '$DIR/$tdir'"
+
+	$LFS hsm_archive $file
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	$LFS hsm_release $file
+	check_hsm_flags $file "0x0000000d"
+
+	$LFS hsm_restore $file
+	wait_request_state $fid RESTORE SUCCEED
+
+	# Check that ordinary user can get HSM state.
+	$RUNAS $LFS hsm_state $file ||
+		error "user '$RUNAS_ID' cannot get HSM state of '$file'"
+
+	$LFS hsm_release $file
+	check_hsm_flags $file "0x0000000d"
+
+	# Check that ordinary user can accessed released file.
+	sum1=$($RUNAS md5sum $file) ||
+		error "user '$RUNAS_ID' cannot read '$file'"
+
+	[ "$sum0" == "$sum1" ] ||
+		error "md5sum mismatch for '$file'"
+
+	copytool_cleanup
+}
+run_test 24b "root can archive, release, and restore user files"
+
+cleanup_test_24c() {
+	trap 0
+	set_hsm_param user_request_mask RESTORE
+	set_hsm_param group_request_mask RESTORE
+	set_hsm_param other_request_mask RESTORE
+}
+
+test_24c() {
+	local file=$DIR/$tdir/$tfile
+	local action=archive
+	local user_save
+	local group_save
+	local other_save
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	# Save the default masks and check that cleanup_24c will
+	# restore the request masks correctly.
+	user_save=$(get_hsm_param user_request_mask)
+	group_save=$(get_hsm_param group_request_mask)
+	other_save=$(get_hsm_param other_request_mask)
+
+	[ "$user_save" == RESTORE ] ||
+		error "user_request_mask is '$user_save' expected 'RESTORE'"
+	[ "$group_save" == RESTORE ] ||
+		error "group_request_mask is '$group_save' expected 'RESTORE'"
+	[ "$other_save" == RESTORE ] ||
+		error "other_request_mask is '$other_save' expected 'RESTORE'"
+
+	trap cleanup_test_24c EXIT
+
+	# User.
+	rm -f $file
+	make_small $file
+	chown $RUNAS_ID:nobody $file ||
+		error "cannot chown '$file' to '$RUNAS_ID:nobody'"
+
+	set_hsm_param user_request_mask ""
+	$RUNAS $LFS hsm_$action $file &&
+		error "$action by user should fail"
+
+	set_hsm_param user_request_mask $action
+	$RUNAS $LFS hsm_$action $file ||
+		error "$action by user should succeed"
+
+	# Group.
+	rm -f $file
+	make_small $file
+	chown nobody:$RUNAS_GID $file ||
+		error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
+
+	set_hsm_param group_request_mask ""
+	$RUNAS $LFS hsm_$action $file &&
+		error "$action by group should fail"
+
+	set_hsm_param group_request_mask $action
+	$RUNAS $LFS hsm_$action $file ||
+		error "$action by group should succeed"
+
+	# Other.
+	rm -f $file
+	make_small $file
+	chown nobody:nobody $file ||
+		error "cannot chown '$file' to 'nobody:nobody'"
+
+	set_hsm_param other_request_mask ""
+	$RUNAS $LFS hsm_$action $file &&
+		error "$action by other should fail"
+
+	set_hsm_param other_request_mask $action
+	$RUNAS $LFS hsm_$action $file ||
+		error "$action by other should succeed"
+
+	copytool_cleanup
+	cleanup_test_24c
+}
+run_test 24c "check that user,group,other request masks work"
+
+cleanup_test_24d() {
+	trap 0
+	mount -o remount,rw $MOUNT2
+}
+
+test_24d() {
+	local file1=$DIR/$tdir/$tfile
+	local file2=$DIR2/$tdir/$tfile
+	local fid1
+	local fid2
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	rm -f $file1
+	fid1=$(make_small $file1)
+
+	trap cleanup_test_24d EXIT
+
+	mount -o remount,ro $MOUNT2
 
-	[ "$(stat -c "%Y" $f)" -eq "$MTIME" ] ||
-		error "mtime should be $MTIME"
+	fid2=$(path2fid $file2)
+	[ "$fid1" == "$fid2" ] ||
+		error "FID mismatch '$fid1' != '$fid2'"
 
-	[ "$(stat -c "%X" $f)" -eq "$ATIME" ] ||
-		error "atime should be $ATIME"
+	$LFS hsm_archive $file2 &&
+		error "archive should fail on read-only mount"
+	check_hsm_flags $file1 "0x00000000"
+
+	$LFS hsm_archive $file1
+	wait_request_state $fid1 ARCHIVE SUCCEED
+
+	$LFS hsm_release $file1
+	$LFS hsm_restore $file2
+	wait_request_state $fid1 RESTORE SUCCEED
+
+	$LFS hsm_release $file1 || error "cannot release '$file1'"
+	dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
+
+	$LFS hsm_release $file2 &&
+		error "release should fail on read-only mount"
 
 	copytool_cleanup
+	cleanup_test_24d
 }
-run_test 24 "Release does not change a/mtime (i/o)"
+run_test 24d "check that read-only mounts are respected"
 
 test_25a() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp /etc/hosts $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/hosts $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
 
 	import_file $tdir/$tfile $f
@@ -1365,13 +1926,14 @@ run_test 28 "Concurrent archive/file remove"
 test_30a() {
 	# restore at exec cannot work on agent node (because of Linux kernel
 	# protection of executables)
-	need2clients || return 0
+	needclients 2 || return 0
 
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp -p /bin/true $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+	copy2archive /bin/true $tdir/$tfile
+
 	local f=$DIR/$tdir/true
 	import_file $tdir/$tfile $f
 
@@ -1396,7 +1958,7 @@ run_test 30a "Restore at exec (import case)"
 test_30b() {
 	# restore at exec cannot work on agent node (because of Linux kernel
 	# protection of executables)
-	need2clients || return 0
+	needclients 2 || return 0
 
 	# test needs a running copytool
 	copytool_setup
@@ -1425,8 +1987,42 @@ test_30b() {
 }
 run_test 30b "Restore at exec (release case)"
 
-restore_and_check_size()
-{
+test_30c() {
+	needclients 2 || return 0
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/SLEEP
+	local fid=$(copy_file /bin/sleep $f)
+	chmod 755 $f
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f
+	check_hsm_flags $f "0x0000000d"
+	# set no retry action mode
+	cdt_set_no_retry
+	do_node $CLIENT2 "$f 10" &
+	local pid=$!
+	sleep 3
+	echo 'Hi!' > $f
+	[[ $? == 0 ]] && error "Update during exec of released file must fail"
+	wait $pid
+	[[ $? == 0 ]] || error "Execution failed during run"
+	cmp /bin/sleep $f
+	[[ $? == 0 ]] || error "Binary overwritten during exec"
+
+	# cleanup
+	# remove no try action mode
+	cdt_clear_no_retry
+	check_hsm_flags $f "0x00000009"
+
+	copytool_cleanup
+}
+run_test 30c "Update during exec of released file must fail"
+
+restore_and_check_size() {
 	local f=$1
 	local fid=$2
 	local s=$(stat -c "%s" $f)
@@ -1440,8 +2036,7 @@ restore_and_check_size()
 		n=$(stat -c "%s" $f)
 		# we echo in both cases to show stat is not
 		# hang
-		if [[ $n != $s ]]
-		then
+		if [[ $n != $s ]]; then
 			echo "size seen is $n != $s"
 			err=1
 		else
@@ -1451,13 +2046,12 @@ restore_and_check_size()
 		sleep 10
 		cpt=$((cpt + 1))
 	done
-	if [[ $cpt -lt 10 ]]
-	then
-		echo " restore is too long"
-	else
+	if [[ $cpt -lt 10 ]]; then
 		echo " "done
+	else
+		echo " restore is too long"
+		wait_request_state $fid RESTORE SUCCEED
 	fi
-	wait_request_state $fid RESTORE SUCCEED
 	return $err
 }
 
@@ -1687,7 +2281,13 @@ test_40() {
 			fid=$(copy_file /etc/hosts $f.$p.$i)
 		done
 	done
-	copytool_setup
+	# force copytool to use a local/temp archive dir to ensure best
+	# performance vs remote/NFS mounts used in auto-tests
+	if df --local $HSM_ARCHIVE >/dev/null 2>&1 ; then
+		copytool_setup
+	else
+		copytool_setup $SINGLEAGT $MOUNT $HSM_ARCHIVE_NUMBER $TMP/$tdir
+	fi
 	# to be sure wait_all_done will not be mislead by previous tests
 	cdt_purge
 	wait_for_grace_delay
@@ -1709,9 +2309,6 @@ test_52() {
 	# test needs a running copytool
 	copytool_setup
 
-	# Test behave badly if 2 mount points are present
-	umount_client $MOUNT2
-
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid=$(copy_file /etc/motd $f 1)
@@ -1731,9 +2328,6 @@ test_52() {
 
 	check_hsm_flags $f "0x0000000b"
 
-	# Restore test environment
-	mount_client $MOUNT2
-
 	copytool_cleanup
 }
 run_test 52 "Opened for write file on an evicted client should be set dirty"
@@ -1742,9 +2336,6 @@ test_53() {
 	# test needs a running copytool
 	copytool_setup
 
-	# Checks are wrong with 2 mount points
-	umount_client $MOUNT2
-
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid=$(copy_file /etc/motd $f 1)
@@ -1765,8 +2356,6 @@ test_53() {
 
 	check_hsm_flags $f "0x00000009"
 
-	mount_client $MOUNT2
-
 	copytool_cleanup
 }
 run_test 53 "Opened for read file on an evicted client should not be set dirty"
@@ -1856,7 +2445,7 @@ run_test 56 "Setattr during an archive is ok"
 
 test_57() {
 	# Need one client for I/O, one for request
-	need2clients || return 0
+	needclients 2 || return 0
 
 	# test needs a running copytool
 	copytool_setup
@@ -1885,13 +2474,15 @@ test_57() {
 }
 run_test 57 "Archive a file with dirty cache on another node"
 
-test_58() {
-	# test needs a running copytool
-	copytool_setup
+truncate_released_file() {
+	local src_file=$1
+	local trunc_to=$2
 
-	mkdir -p $DIR/$tdir
+	local sz=$(stat -c %s $src_file)
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_small $f)
+	local fid=$(copy_file $1 $f)
+	local ref=$f-ref
+	cp $f $f-ref
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -1899,63 +2490,46 @@ test_58() {
 
 	$LFS hsm_release $f || error "could not release file"
 
-	$TRUNCATE $f 0 || error "truncate failed"
+	$TRUNCATE $f $trunc_to || error "truncate failed"
 	sync
 
-	local sz=$(stat -c %s $f)
-	[[ $sz == 0 ]] || error "size after truncate is $sz != 0"
+	local sz1=$(stat -c %s $f)
+	[[ $sz1 == $trunc_to ]] ||
+		error "size after trunc: $sz1 expect $trunc_to, original $sz"
 
 	$LFS hsm_state $f
-
 	check_hsm_flags $f "0x0000000b"
 
 	local state=$(get_request_state $fid RESTORE)
-	[[ "$state" == "" ]] ||
-		error "truncate 0 trigs a restore, state = $state"
+	[[ "$state" == "SUCCEED" ]] ||
+		error "truncate $sz does not trig restore, state = $state"
 
-	copytool_cleanup
+	$TRUNCATE $ref $trunc_to
+	cmp $ref $f || error "file data wrong after truncate"
+
+	rm -f $f $f-ref
 }
-run_test 58 "Truncate 0 on a released file must not trigger restore"
 
-test_59() {
+test_58() {
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
-	local f=$DIR/$tdir/$tfile
-	local fid=$(copy_file /etc/passwd $f)
-	local ref=$f-ref
-	cp $f $ref
-	local sz=$(stat -c %s $ref)
-	sz=$((sz / 2))
-	$TRUNCATE $ref $sz
-
-	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
-		error "could not archive file"
-	wait_request_state $fid ARCHIVE SUCCEED
 
-	$LFS hsm_release $f || error "could not release file"
-
-	$TRUNCATE $f $sz || error "truncate failed"
-	sync
+	local sz=$(stat -c %s /etc/passwd)
 
-	local sz1=$(stat -c %s $f)
-	[[ $sz1 == $sz ]] || error "size after truncate is $sz1 != $sz"
+	echo "truncate up from $sz to $((sz*2))"
+	truncate_released_file /etc/passwd $((sz*2))
 
-	$LFS hsm_state $f
+	echo "truncate down from $sz to $((sz/2))"
+	truncate_released_file /etc/passwd $((sz/2))
 
-	check_hsm_flags $f "0x0000000b"
-
-	local state=$(get_request_state $fid RESTORE)
-	[[ "$state" == "SUCCEED" ]] ||
-		error "truncate $sz does not trig a successfull restore,"\
-		      " state = $state"
-
-	cmp $ref $f || error "file data wrong after truncate"
+	echo "truncate to 0"
+	truncate_released_file /etc/passwd 0
 
 	copytool_cleanup
 }
-run_test 59 "Truncate != 0 on a released file"
+run_test 58 "Truncate a released file will trigger restore"
 
 test_90() {
 	file_count=57
@@ -1985,7 +2559,7 @@ test_90() {
 }
 run_test 90 "Archive/restore a file list"
 
-double_verify_reset_ham_param() {
+double_verify_reset_hsm_param() {
 	local p=$1
 	echo "Testing $HSM_PARAM.$p"
 	local val=$(get_hsm_param $p)
@@ -2001,17 +2575,17 @@ double_verify_reset_ham_param() {
 	# restore value
 	set_hsm_param $p $save
 
-	if [[ $rc == 0 ]]
-	then
+	if [[ $rc == 0 ]]; then
 		error "we must not be able to set $HSM_PARAM.$p to 0"
 	fi
 }
 
 test_100() {
-	double_verify_reset_ham_param loop_period
-	double_verify_reset_ham_param grace_delay
-	double_verify_reset_ham_param request_timeout
-	double_verify_reset_ham_param max_requests
+	double_verify_reset_hsm_param loop_period
+	double_verify_reset_hsm_param grace_delay
+	double_verify_reset_hsm_param active_request_timeout
+	double_verify_reset_hsm_param max_requests
+	double_verify_reset_hsm_param default_archive_id
 }
 run_test 100 "Set coordinator /proc tunables"
 
@@ -2039,7 +2613,7 @@ test_103() {
 
 	echo "Current requests"
 	local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
 
 	[[ -z "$res" ]] || error "Some request have not been canceled"
@@ -2061,7 +2635,7 @@ test_104() {
 	cdt_disable
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
 	local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep $fid | cut -f16 -d=")
 	cdt_enable
 
@@ -2082,12 +2656,12 @@ test_105() {
 		$LFS hsm_archive $DIR/$tdir/$i
 	done
 	local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_restart
 	cdt_disable
 	local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_enable
 	cdt_purge
@@ -2097,33 +2671,70 @@ test_105() {
 }
 run_test 105 "Restart of coordinator"
 
+get_agent_by_uuid_mdt() {
+	local uuid=$1
+	local mdtidx=$2
+	local mds=mds$(($mdtidx + 1))
+	do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
+		 grep $uuid"
+}
+
+check_agent_registered_by_mdt() {
+	local uuid=$1
+	local mdtidx=$2
+	local mds=mds$(($mdtidx + 1))
+	local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
+	if [[ ! -z "$agent" ]]; then
+		echo "found agent $agent on $mds"
+	else
+		error "uuid $uuid not found in agent list on $mds"
+	fi
+}
+
+check_agent_unregistered_by_mdt() {
+	local uuid=$1
+	local mdtidx=$2
+	local mds=mds$(($mdtidx + 1))
+	local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
+	if [[ -z "$agent" ]]; then
+		echo "uuid not found in agent list on $mds"
+	else
+		error "uuid found in agent list on $mds: $agent"
+	fi
+}
+
+check_agent_registered() {
+	local uuid=$1
+	local mdsno
+	for mdsno in $(seq 1 $MDSCOUNT); do
+		check_agent_registered_by_mdt $uuid $((mdsno - 1))
+	done
+}
+
+check_agent_unregistered() {
+	local uuid=$1
+	local mdsno
+	for mdsno in $(seq 1 $MDSCOUNT); do
+		check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
+	done
+}
+
 test_106() {
-	# Test behave badly if 2 mount points are present
-	umount_client $MOUNT2
+	local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
+		get_client_uuid $MOUNT | cut -d' ' -f2)
 
-	# test needs a running copytool
 	copytool_setup
+	check_agent_registered $uuid
+
+	search_copytools || error "No copytool found"
 
-	local uuid=$(my_uuid)
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
 	copytool_cleanup
-	[[ ! -z "$agent" ]] || error "My uuid $uuid not found in agent list"
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
-	[[ -z "$agent" ]] ||
-		error "My uuid $uuid still found in agent list,"\
-		      " after copytool shutdown"
+	check_agent_unregistered $uuid
+
 	copytool_setup
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
-	copytool_cleanup
-	[[ ! -z "$agent" ]] ||
-		error "My uuid $uuid not found in agent list after"\
-		      " copytool restart"
+	check_agent_registered $uuid
 
-	# Restore test environment
-	mount_client $MOUNT2
+	copytool_cleanup
 }
 run_test 106 "Copytool register/unregister"
 
@@ -2148,23 +2759,60 @@ test_107() {
 }
 run_test 107 "Copytool re-register after MDS restart"
 
+policy_set_and_test()
+{
+	local change="$1"
+	local target="$2"
+	do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
+	local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
+	[[ "$policy" == "$target" ]] ||
+		error "Wrong policy after '$change': '$policy' != '$target'"
+}
+
+test_109() {
+	# to force default policy setting if error
+	CDT_POLICY_HAD_CHANGED=true
+
+	local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
+	local default="NonBlockingRestore [NoRetryAction]"
+	[[ "$policy" == "$default" ]] ||
+		error "default policy has changed,"\
+		      " '$policy' != '$default' update the test"
+	policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
+	policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
+	policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
+	policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
+	policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
+	# useless bacause we know but safer for futur changes to use real value
+	local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
+	echo "Next set_param must failed"
+	policy_set_and_test "wrong" "$policy"
+
+	# return to default
+	echo "Back to default policy"
+	cdt_set_sanity_policy
+}
+run_test 109 "Policy display/change"
+
 test_110a() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
-	cp /etc/passwd $HSM_ARCHIVE/$tdir/$tfile
+	mkdir -p $DIR/$tdir
+
+	copy2archive /etc/passwd $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
 	import_file $tdir/$tfile $f
 	local fid=$(path2fid $f)
 
-	cdt_set_no_blocking_restore
+	cdt_set_non_blocking_restore
 	md5sum $f
 	local st=$?
 
 	# cleanup
 	wait_request_state $fid RESTORE SUCCEED
-	cdt_clear_no_blocking_restore
+	cdt_clear_non_blocking_restore
 
 	# Test result
 	[[ $st == 1 ]] ||
@@ -2186,13 +2834,13 @@ test_110b() {
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
 
-	cdt_set_no_blocking_restore
+	cdt_set_non_blocking_restore
 	md5sum $f
 	local st=$?
 
 	# cleanup
 	wait_request_state $fid RESTORE SUCCEED
-	cdt_clear_no_blocking_restore
+	cdt_clear_non_blocking_restore
 
 	# Test result
 	[[ $st == 1 ]] ||
@@ -2207,9 +2855,11 @@ test_111a() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/passwd $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
-	cp /etc/passwd $HSM_ARCHIVE/$tdir/$tfile
+
 	import_file $tdir/$tfile $f
 	local fid=$(path2fid $f)
 
@@ -2360,7 +3010,7 @@ test_220() {
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 	changelog_cleanup
 
 	local target=0x0
@@ -2387,13 +3037,12 @@ test_221() {
 	wait_request_state $fid ARCHIVE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0x7d
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 221 "Changelog for archive canceled"
 
@@ -2401,9 +3050,10 @@ test_222a() {
 	# test needs a running copytool
 	copytool_setup
 
-	mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/passwd $tdir/$tfile
+
 	local f=$DIR/$tdir/$tfile
-	cp /etc/passwd $HSM_ARCHIVE/$tdir/$tfile
 	import_file $tdir/$tfile $f
 	local fid=$(path2fid $f)
 
@@ -2412,13 +3062,12 @@ test_222a() {
 	$LFS hsm_restore $f
 	wait_request_state $fid RESTORE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0x80
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 222a "Changelog for explicit restore"
 
@@ -2439,13 +3088,12 @@ test_222b() {
 
 	wait_request_state $fid RESTORE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0x80
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 222b "Changelog for implicit restore"
 
@@ -2469,14 +3117,13 @@ test_223a() {
 	wait_request_state $fid RESTORE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0xfd
 	[[ $flags == $target ]] ||
 		error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 223a "Changelog for restore canceled (import case)"
 
@@ -2499,14 +3146,13 @@ test_223b() {
 	wait_request_state $fid RESTORE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 
 	local target=0xfd
 	[[ $flags == $target ]] ||
 		error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 223b "Changelog for restore canceled (release case)"
 
@@ -2526,14 +3172,13 @@ test_224() {
 	$LFS hsm_remove $f
 	wait_request_state $fid REMOVE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
 
 	local target=0x200
 	[[ $flags == $target ]] ||
 		error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 224 "Changelog for remove"
 
@@ -2563,16 +3208,15 @@ test_225() {
 	wait_request_state $fid REMOVE CANCELED
 	wait_request_state $fid CANCEL SUCCEED
 
-	flags=$(changelog_get_flags $MDT0 RENME $fid2)
-	local flags=$($LFS changelog $MDT0 | grep HSM | grep $fid | tail -1 |
-		awk '{print $5}')
+	flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
+	local flags=$($LFS changelog ${MDT[0]} | grep HSM | grep $fid |
+		tail -n 1 | awk '{print $5}')
 
 	local target=0x27d
 	[[ $flags == $target ]] ||
 		error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 225 "Changelog for remove canceled"
 
@@ -2598,7 +3242,7 @@ test_226() {
 
 	rm $f1 || error "rm $f1 failed"
 
-	local flags=$(changelog_get_flags $MDT0 UNLNK $fid1)
+	local flags=$(changelog_get_flags ${MDT[0]} UNLNK $fid1)
 
 	local target=0x3
 	[[ $flags == $target ]] ||
@@ -2606,14 +3250,13 @@ test_226() {
 
 	mv $f3 $f2 || error "mv $f3 $f2 failed"
 
-	flags=$(changelog_get_flags $MDT0 RENME $fid2)
+	flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
 
 	target=0x3
 	[[ $flags == $target ]] ||
 		error "Changelog flag is $flags not $target"
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 226 "changelog for last rm/mv with exiting archive"
 
@@ -2627,7 +3270,7 @@ check_flags_changes() {
 	local target=0x280
 	$LFS hsm_set --$hsm_flag $f ||
 		error "Cannot set $hsm_flag on $f"
-	local flags=($(changelog_get_flags $MDT0 HSM $fid))
+	local flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
 	local seen=${#flags[*]}
 	cnt=$((fst + cnt))
 	[[ $seen == $cnt ]] ||
@@ -2638,7 +3281,7 @@ check_flags_changes() {
 
 	$LFS hsm_clear --$hsm_flag $f ||
 		error "Cannot clear $hsm_flag on $f"
-	flags=($(changelog_get_flags $MDT0 HSM $fid))
+	flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
 	seen=${#flags[*]}
 	cnt=$(($cnt + 1))
 	[[ $cnt == $seen ]] ||
@@ -2670,11 +3313,46 @@ test_227() {
 	wait_request_state $fid ARCHIVE SUCCEED
 	check_flags_changes $f $fid lost 3 1
 
-	changelog_cleanup
-	copytool_cleanup
+	cleanup
 }
 run_test 227 "changelog when explicit setting of HSM flags"
 
+test_228() {
+	# test needs a running copytool
+	copytool_setup
+
+	local fid=$(make_small_sync $DIR/$tfile)
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	$LFS hsm_release $DIR/$tfile
+	check_hsm_flags $DIR/$tfile "0x0000000d"
+
+	filefrag $DIR/$tfile | grep " 1 extent found" ||
+		error "filefrag on released file must return only one extent"
+
+	# only newer versions of cp detect sparse files by stat/FIEMAP
+	# (LU-2580)
+	cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
+		error "copying $DIR/$tfile"
+	cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
+
+	$LFS hsm_release $DIR/$tfile
+	check_hsm_flags $DIR/$tfile "0x0000000d"
+
+	mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
+
+	tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
+		error "tar failed"
+	cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
+		error "comparing untarred $DIR/$tfile"
+
+	rm -f $DIR/$tfile $DIR/$tfile.2 ||
+		error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
+	copytool_cleanup
+}
+run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
+
 test_250() {
 	# test needs a running copytool
 	copytool_setup
@@ -2699,12 +3377,12 @@ test_250() {
 	while [[ $cnt != 0 || $wt != 0 ]]; do
 		sleep 1
 		cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep STARTED | grep -v CANCEL | wc -l")
 		[[ $cnt -le $maxrequest ]] ||
 			error "$cnt > $maxrequest too many started requests"
 		wt=$(do_facet $SINGLEMDS "$LCTL get_param\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 		echo "max=$maxrequest started=$cnt waiting=$wt"
 	done
@@ -2723,8 +3401,8 @@ test_251() {
 
 	cdt_disable
 	# to have a short test
-	local old_to=$(get_hsm_param request_timeout)
-	set_hsm_param request_timeout 4
+	local old_to=$(get_hsm_param active_request_timeout)
+	set_hsm_param active_request_timeout 4
 	# to be sure the cdt will wake up frequently so
 	# it will be able to cancel the "old" request
 	local old_loop=$(get_hsm_param loop_period)
@@ -2736,7 +3414,7 @@ test_251() {
 	sleep 5
 	wait_request_state $fid ARCHIVE CANCELED
 
-	set_hsm_param request_timeout $old_to
+	set_hsm_param active_request_timeout $old_to
 	set_hsm_param loop_period $old_loop
 
 	copytool_cleanup
@@ -2770,6 +3448,209 @@ test_300() {
 }
 run_test 300 "On disk coordinator state kept between MDT umount/mount"
 
+test_301() {
+	local ai=$(get_hsm_param default_archive_id)
+	local new=$((ai + 1))
+
+	set_hsm_param default_archive_id $new -P
+	fail $SINGLEMDS
+	local res=$(get_hsm_param default_archive_id)
+
+	# clear value
+	set_hsm_param default_archive_id "" "-P -d"
+
+	[[ $new == $res ]] || error "Value after MDS restart is $res != $new"
+}
+run_test 301 "HSM tunnable are persistent"
+
+test_302() {
+	local ai=$(get_hsm_param default_archive_id)
+	local new=$((ai + 1))
+
+	# stop coordinator
+	cdt_shutdown
+
+	set_hsm_param default_archive_id $new -P
+
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		fail mds${mdtno}
+	done
+
+	# check cdt is on
+	cdt_check_state enabled
+
+	local res=$(get_hsm_param default_archive_id)
+
+	# clear value
+	set_hsm_param default_archive_id "" "-P -d"
+
+	[[ $new == $res ]] || error "Value after MDS restart is $res != $new"
+}
+run_test 302 "HSM tunnable are persistent when CDT is off"
+
+test_400() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local dir_mdt0=$DIR/$tdir/mdt0
+	local dir_mdt1=$DIR/$tdir/mdt1
+
+	# create 1 dir per MDT
+	$LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
+	$LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
+
+	# create 1 file in each MDT
+	local fid1=$(make_small $dir_mdt0/$tfile)
+	local fid2=$(make_small $dir_mdt1/$tfile)
+
+	# check that hsm request on mdt0 is sent to the right MDS
+	$LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
+	wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
+		echo "archive successful on mdt0"
+
+	# check that hsm request on mdt1 is sent to the right MDS
+	$LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
+	wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
+		echo "archive successful on mdt1"
+
+	copytool_cleanup
+	# clean test files and directories
+	rm -rf $dir_mdt0 $dir_mdt1
+}
+run_test 400 "Single request is sent to the right MDT"
+
+test_401() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local dir_mdt0=$DIR/$tdir/mdt0
+	local dir_mdt1=$DIR/$tdir/mdt1
+
+	# create 1 dir per MDT
+	$LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
+	$LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
+
+	# create 1 file in each MDT
+	local fid1=$(make_small $dir_mdt0/$tfile)
+	local fid2=$(make_small $dir_mdt1/$tfile)
+
+	# check that compound requests are shunt to the rights MDTs
+	$LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
+		error "lfs hsm_archive"
+	wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
+		echo "archive successful on mdt0"
+	wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
+		echo "archive successful on mdt1"
+
+	copytool_cleanup
+	# clean test files and directories
+	rm -rf $dir_mdt0 $dir_mdt1
+}
+run_test 401 "Compound requests split and sent to their respective MDTs"
+
+mdc_change_state() # facet, MDT_pattern, activate|deactivate
+{
+	local facet=$1
+	local pattern="$2"
+	local state=$3
+	local node=$(facet_active_host $facet)
+	local mdc
+	for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
+			awk '{print $4}'); do
+		echo "$3 $mdc on $node"
+		do_facet $facet "$LCTL --device $mdc $state" || return 1
+	done
+}
+
+test_402() {
+	# make sure there is no running copytool
+	copytool_cleanup
+
+	# deactivate all mdc on agent1
+	mdc_change_state $SINGLEAGT "MDT000." "deactivate"
+
+	copytool_setup $SINGLEAGT
+
+	check_agent_unregistered "uuid" # match any agent
+
+	# no expected running copytool
+	search_copytools $agent && error "Copytool start should have failed"
+
+	# reactivate MDCs
+	mdc_change_state $SINGLEAGT "MDT000." "activate"
+}
+run_test 402 "Copytool start fails if all MDTs are inactive"
+
+test_403() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	# make sure there is no running copytool
+	copytool_cleanup
+
+        local agent=$(facet_active_host $SINGLEAGT)
+	local uuid=$(do_rpc_nodes $agent get_client_uuid | cut -d' ' -f2)
+
+	# deactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "deactivate"
+
+	copytool_setup
+	# check the agent is registered on MDT0000, and not on MDT0001
+	check_agent_registered_by_mdt $uuid 0
+	check_agent_unregistered_by_mdt $uuid 1
+
+	# check running copytool process
+	search_copytools $agent || error "No running copytools on $agent"
+
+	# reactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "activate"
+
+	# make sure the copytool is now registered to all MDTs
+	check_agent_registered $uuid
+
+	copytool_cleanup
+}
+run_test 403 "Copytool starts with inactive MDT and register on reconnect"
+
+test_404() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	# create files on both MDT0000 and MDT0001
+	mkdir -p $DIR/$tdir
+
+	local dir_mdt0=$DIR/$tdir/mdt0
+	$LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
+
+	# create 1 file on mdt0
+	local fid1=$(make_small $dir_mdt0/$tfile)
+
+	# deactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "deactivate"
+
+	# send an HSM request for files in MDT0000
+	$LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
+
+	# check for completion of files in MDT0000
+	wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
+		echo "archive successful on mdt0"
+
+	# reactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "MDT0001" "activate"
+
+	copytool_cleanup
+	# clean test files and directories
+	rm -rf $dir_mdt0
+}
+run_test 404 "Inactive MDT does not block requests for active MDTs"
+
 copytool_cleanup
 
 complete $SECONDS