X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=lustre%2Ftests%2Fsanity-hsm.sh;h=0d43e33b53bad60642f49474dbec24222624c619;hb=26cc6c4228ae19a8dd522042060852f50bbb621a;hp=971a38a0491d71b40b206fe6b6455bc1c03f29b2;hpb=69756f524a1b488f64512c6b20a6d4a94652c208;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index 971a38a..0d43e33 100644
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,12 +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     3939
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36 40"
+# bug number for skipped test:4178         4176
+ALWAYS_EXCEPT="$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="$SANITY_HSM_EXCEPT 12a 12b 12n 13 30a 31a 34 35 36 58 59"
-ALWAYS_EXCEPT="$ALWAYS_EXCEPT 110a 200 201 221 222a 223a 223b 225"
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 
@@ -31,9 +30,13 @@ 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
+
+[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
+# bug number for skipped test:        LU-3700
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 51b"
 
 check_and_setup_lustre
 
@@ -95,8 +98,9 @@ init_agt_vars() {
 	HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
 	HSM_ARCHIVE_NUMBER=2
 
-	MDT_PARAM="mdt.$FSNAME-MDT0000"
-	HSM_PARAM="$MDT_PARAM.hsm"
+	# 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
@@ -114,6 +118,23 @@ copytool_device() {
 cleanup() {
 	copytool_cleanup
 	changelog_cleanup
+	cdt_set_sanity_policy
+}
+
+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 $1}' | head -1)
+	done
+}
+
+search_copytools() {
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
 }
 
 search_and_kill_copytool() {
@@ -163,10 +184,34 @@ copytool_setup() {
 copytool_cleanup() {
 	trap - EXIT
 	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 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() {
@@ -178,9 +223,9 @@ copytool_suspend() {
 
 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() {
@@ -192,25 +237,77 @@ import_file() {
 
 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() {
@@ -230,58 +327,58 @@ 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() {
@@ -306,9 +403,10 @@ 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
@@ -322,8 +420,7 @@ 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)
@@ -365,15 +462,13 @@ 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
@@ -389,7 +484,8 @@ make_small() {
 }
 
 cleanup_large_files() {
-	local ratio=$(df $MOUNT |awk '{print $5}' |sed 's/%//g' |grep -v Use)
+	local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
+		      sed 's/%//g')
 	[ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
 }
 
@@ -453,19 +549,22 @@ wait_request_state() {
 	local fid=$1
 	local request=$2
 	local state=$3
+	# 4th arg (mdt index) is optional
+	local mdtidx=${4:-0}
+	local mds=mds$(($mdtidx + 1))
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions"
 	cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 
-	wait_result $SINGLEMDS "$cmd" $state 100 ||
-		error "request on $fid is not $state"
+	wait_result $mds "$cmd" $state 100 ||
+		error "request on $fid is not $state on $mds"
 }
 
 get_request_state() {
 	local fid=$1
 	local request=$2
 
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
 		"awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 }
 
@@ -473,14 +572,14 @@ get_request_count() {
 	local fid=$1
 	local request=$2
 
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
 		"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
 }
 
 wait_all_done() {
 	local timeout=$1
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
 	cmd+=" | egrep 'WAITING|STARTED'"
 
 	wait_result $SINGLEMDS "$cmd" "" $timeout ||
@@ -492,8 +591,8 @@ wait_for_grace_delay() {
 	sleep $val
 }
 
-MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
-	awk '{gsub(/_UUID/,""); print $1}' | head -1)
+# populate MDT device array
+get_mdt_devices
 
 # initiate variables
 init_agt_vars
@@ -511,6 +610,9 @@ 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
 
@@ -677,8 +779,7 @@ test_9() {
 run_test 9 "Use of explict archive number, with dedicated copytool"
 
 test_9a() {
-	[[ $CLIENTCOUNT -ge 3 ]] ||
-		{ skip "Need three or more clients"; return 0; }
+	needclients 3 || return 0
 
 	local n
 	local file
@@ -700,7 +801,7 @@ test_9a() {
 
 		$LFS hsm_archive $file || error "could not archive file $file"
 		wait_request_state $fid ARCHIVE SUCCEED
-		check_hsm_flags $file "0x00000001"
+		check_hsm_flags $file "0x00000009"
 	done
 
 	trap - EXIT
@@ -719,10 +820,10 @@ 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"
 
@@ -740,7 +841,7 @@ 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"
@@ -759,7 +860,7 @@ 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
@@ -780,7 +881,7 @@ test_10d() {
 	wait_request_state $fid ARCHIVE SUCCEED
 
 	local ar=$(get_hsm_archive_id $f)
-	local dflt=$(get_hsm_param archive_id)
+	local dflt=$(get_hsm_param default_archive_id)
 	[[ $ar == $dflt ]] ||
 		error "archived file is not on default archive: $ar != $dflt"
 
@@ -789,8 +890,8 @@ test_10d() {
 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
@@ -798,7 +899,7 @@ 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 != $ASZ"
@@ -809,7 +910,7 @@ test_11() {
 	local fid=$(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"
@@ -818,8 +919,9 @@ 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
@@ -833,7 +935,7 @@ 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"
 
@@ -845,8 +947,9 @@ 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: "
@@ -857,7 +960,7 @@ 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"
 
@@ -893,7 +996,8 @@ 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"
@@ -985,7 +1089,7 @@ 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
@@ -1034,12 +1138,14 @@ 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"
 
@@ -1058,24 +1164,23 @@ 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"
@@ -1242,19 +1347,32 @@ test_21() {
 	local fid=$(make_small $f)
 	check_hsm_flags $f "0x00000000"
 
+	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
@@ -1272,6 +1390,8 @@ 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"
@@ -1467,8 +1587,8 @@ test_24b() {
 	$LFS hsm_archive $file
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	$LFS hsm_release $file ||
-		check_hsm_flags $file "0x0000000d"
+	$LFS hsm_release $file
+	check_hsm_flags $file "0x0000000d"
 
 	$LFS hsm_restore $file
 	wait_request_state $fid RESTORE SUCCEED
@@ -1477,8 +1597,8 @@ test_24b() {
 	$RUNAS $LFS hsm_state $file ||
 		error "user '$RUNAS_ID' cannot get HSM state of '$file'"
 
-	$LFS hsm_release $file ||
-		check_hsm_flags $file "0x0000000d"
+	$LFS hsm_release $file
+	check_hsm_flags $file "0x0000000d"
 
 	# Check that ordinary user can accessed released file.
 	sum1=$($RUNAS md5sum $file) ||
@@ -1491,12 +1611,141 @@ test_24b() {
 }
 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 $MOUNT2
+
+	fid2=$(path2fid $file2)
+	[ "$fid1" == "$fid2" ] ||
+		error "FID mismatch '$fid1' != '$fid2'"
+
+	$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 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
@@ -1619,13 +1868,14 @@ 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
 
@@ -1650,7 +1900,7 @@ 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
@@ -1679,6 +1929,41 @@ test_30b() {
 }
 run_test 30b "Restore at exec (release case)"
 
+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=$2
@@ -1693,8 +1978,7 @@ restore_and_check_size() {
 		n=$(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 != $s"
 			err=1
 		else
@@ -1704,13 +1988,12 @@ restore_and_check_size() {
 		sleep 10
 		cpt=$((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
 }
 
@@ -2098,7 +2381,7 @@ 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
@@ -2127,13 +2410,15 @@ 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"
@@ -2141,63 +2426,46 @@ 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
@@ -2243,8 +2511,7 @@ double_verify_reset_hsm_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
 }
@@ -2252,9 +2519,9 @@ double_verify_reset_hsm_param() {
 test_100() {
 	double_verify_reset_hsm_param loop_period
 	double_verify_reset_hsm_param grace_delay
-	double_verify_reset_hsm_param request_timeout
+	double_verify_reset_hsm_param active_request_timeout
 	double_verify_reset_hsm_param max_requests
-	double_verify_reset_hsm_param archive_id
+	double_verify_reset_hsm_param default_archive_id
 }
 run_test 100 "Set coordinator /proc tunables"
 
@@ -2282,7 +2549,7 @@ 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"
@@ -2304,7 +2571,7 @@ 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
 
@@ -2325,12 +2592,12 @@ 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
@@ -2340,28 +2607,70 @@ 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 $((mdsno - 1))
+	done
+}
+
 test_106() {
-	# test needs a running copytool
+	local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
+		get_client_uuid $MOUNT | cut -d' ' -f2)
+
 	copytool_setup
+	check_agent_registered $uuid
+
+	search_copytools || error "No copytool found"
 
-	local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
-		get_client_uuid | cut -d' ' -f2)
-	local agent=$(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)
+	check_agent_registered $uuid
+
 	copytool_cleanup
-	[[ ! -z "$agent" ]] ||
-		error "My uuid $uuid not found in agent list after"\
-		      " copytool restart"
 }
 run_test 106 "Copytool register/unregister"
 
@@ -2386,23 +2695,60 @@ 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 ]] ||
@@ -2424,13 +2770,13 @@ 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 ]] ||
@@ -2445,9 +2791,11 @@ 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)
 
@@ -2598,7 +2946,7 @@ 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
@@ -2625,7 +2973,7 @@ 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"
@@ -2638,9 +2986,10 @@ 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)
 
@@ -2649,7 +2998,7 @@ 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"
@@ -2675,7 +3024,7 @@ 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"
@@ -2704,7 +3053,7 @@ 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 ]] ||
@@ -2733,7 +3082,7 @@ 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 ]] ||
@@ -2759,7 +3108,7 @@ 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 ]] ||
@@ -2795,9 +3144,9 @@ 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 ]] ||
@@ -2829,7 +3178,7 @@ 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 ]] ||
@@ -2837,7 +3186,7 @@ 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 ]] ||
@@ -2857,7 +3206,7 @@ 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 ]] ||
@@ -2868,7 +3217,7 @@ 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 ]] ||
@@ -2904,6 +3253,41 @@ test_227() {
 }
 run_test 227 "changelog when explicit setting of HSM flags"
 
+test_228() {
+	# test needs a running copytool
+	copytool_setup
+
+	dd if=/dev/urandom of=$DIR/$tfile bs=1M count=1 conv=sync ||
+		error "creating $DIR/$tfile"
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
+	wait_request_state $(path2fid $DIR/$tfile) 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 $DIR/$tdir
+
+	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"
+
+	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
@@ -2928,12 +3312,12 @@ 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
@@ -2952,8 +3336,8 @@ 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)
@@ -2965,7 +3349,7 @@ 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
@@ -2999,6 +3383,209 @@ 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