X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=a69302a8d41779dd3e4aa1b354b2c59ac78f58be;hp=d3c9e89646c0dc4c60bbc0a7971fefc3d4460774;hb=cff9f1e7c6a41bfa05d1455b8964860803d12612;hpb=38695729d61958ab10e9e108175298f8a7d40536;ds=sidebyside

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
old mode 100644
new mode 100755
index d3c9e89..a69302a
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,8 +11,8 @@ SRCDIR=$(dirname $0)
 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test:    3815     3939
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36 40"
+# bug number for skipped test:
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
@@ -24,13 +24,13 @@ init_logging
 
 MULTIOP=${MULTIOP:-multiop}
 OPENFILE=${OPENFILE:-openfile}
-MCREATE=${MCREATE:-mcreate}
+MMAP_CAT=${MMAP_CAT:-mmap_cat}
 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
 
@@ -46,6 +46,14 @@ check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
 
 build_test_filter
 
+# if there is no CLIENT1 defined, some tests can be ran on localhost
+CLIENT1=${CLIENT1:-$HOSTNAME}
+# if CLIENT2 doesn't exist then use CLIENT1 instead
+# All tests should use CLIENT2 with MOUNT2 only therefore it will work if
+# $CLIENT2 == CLIENT1
+# Exception is the test which need two separate nodes
+CLIENT2=${CLIENT2:-$CLIENT1}
+
 #
 # In order to test multiple remote HSM agents, a new facet type named "AGT" and
 # the following associated variables are added:
@@ -74,8 +82,11 @@ init_agt_vars() {
 		exit 0
 	fi
 
+	# We used to put the HSM archive in $SHARED_DIRECTORY but that
+	# meant NFS issues could hose sanity-hsm sessions. So now we
+	# use $TMP instead.
 	for n in $(seq $AGTCOUNT); do
-		eval export AGTDEV$n=\$\{AGTDEV$n:-"$SHARED_DIRECTORY/arc$n"\}
+		eval export AGTDEV$n=\$\{AGTDEV$n:-"$TMP/arc$n"\}
 		agent=CLIENT$((n + 1))
 		if [[ -z "${!agent}" ]]; then
 			[[ $CLIENTCOUNT -eq 1 ]] && agent=CLIENT1 ||
@@ -88,15 +99,31 @@ init_agt_vars() {
 
 	export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
 	export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
+	export HSMTOOL_UPDATE_INTERVAL=${HSMTOOL_UPDATE_INTERVAL:=""}
+	export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
+	export HSMTOOL_TESTDIR
 	export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
+	# $hsm_root/$HSMTMP Makes $hsm_root dir path less generic to ensure
+	# rm -rf $hsm_root/* is safe even if $hsm_root becomes unset to avoid
+	# deleting everything in filesystem, independent of any copytool.
+	export HSMTMP=${HSMTMP:-"shsm"}
+
 	HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
+
+	[ -z "${HSM_ARCHIVE// /}" ] && error "HSM_ARCHIVE is empty!"
+	HSM_ARCHIVE=$HSM_ARCHIVE/$HSMTMP
+
 	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
+
+	# Don't allow copytool error upon start/setup
+	HSMTOOL_NOERROR=false
 }
 
 # Get the backend root path for the given agent facet.
@@ -109,42 +136,148 @@ copytool_device() {
 
 # Stop copytool and unregister an existing changelog user.
 cleanup() {
+	copytool_monitor_cleanup
 	copytool_cleanup
 	changelog_cleanup
 	cdt_set_sanity_policy
 }
 
-search_and_kill_copytool() {
-	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+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 -n1)
+	done
+}
 
-	echo "Killing existing copytools on $agents"
-	do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
+search_copytools() {
+	local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+	do_nodesv $hosts "pgrep -x $HSMTOOL_BASE"
 }
 
-copytool_setup() {
+kill_copytools() {
+	local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+
+	echo "Killing existing copytools on $hosts"
+	do_nodesv $hosts "killall -q $HSMTOOL_BASE" || true
+}
+
+wait_copytools() {
+	local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+	local wait_timeout=200
+	local wait_start=$SECONDS
+	local wait_end=$((wait_start + wait_timeout))
+	local sleep_time=100000 # 0.1 second
+
+	while ((SECONDS < wait_end)); do
+		if ! search_copytools $hosts; then
+			echo "copytools stopped in $((SECONDS - wait_start))s"
+			return 0
+		fi
+
+		echo "copytools still running on $hosts"
+		usleep $sleep_time
+		[ $sleep_time -lt 32000000 ] && # 3.2 seconds
+			sleep_time=$(bc <<< "$sleep_time * 2")
+	done
+
+	# try to dump Copytool's stack
+	do_nodesv $hosts "echo 1 >/proc/sys/kernel/sysrq ; " \
+			 "echo t >/proc/sysrq-trigger"
+
+	echo "copytools failed to stop in ${wait_timeout}s"
+
+	return 1
+}
+
+copytool_monitor_setup() {
 	local facet=${1:-$SINGLEAGT}
-	local lustre_mntpnt=${2:-$MOUNT}
-	local arc_id=$3
-	local hsm_root=$(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
+	local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
+	local test_dir=$(do_node $agent "$cmd") ||
+		error "Failed to create tempdir on $agent"
+	export HSMTOOL_MONITOR_DIR=$test_dir
+
+	# Create the fifo and a monitor (cat dies when copytool dies)
+	do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
+		error "failed to create copytool fifo on $agent"
+	cmd="cat $test_dir/fifo > $test_dir/events &"
+	cmd+=" echo \\\$! > $test_dir/monitor_pid"
+
+	if [[ $PDSH == *Rmrsh* ]]; then
+		# This is required for pdsh -Rmrsh and its handling of remote
+		# shells.
+		# Regular ssh and pdsh -Rssh work fine without this
+		# backgrounded subshell nonsense.
+		(do_node $agent "$cmd") &
+		export HSMTOOL_MONITOR_PDSH=$!
+
+		# Slightly racy, but just making a best-effort to catch obvious
+		# problems.
+		sleep 1
+		ps -p $HSMTOOL_MONITOR_PDSH > /dev/null ||
+			error "Failed to start copytool monitor on $agent"
+	else
+		do_node $agent "$cmd"
+		if [ $? != 0 ]; then
+			error "Failed to start copytool monitor on $agent"
+		fi
+	fi
+}
+
+copytool_monitor_cleanup() {
+	local facet=${1:-$SINGLEAGT}
+	local agent=$(facet_active_host $facet)
+
+	if [ -n "$HSMTOOL_MONITOR_DIR" ]; then
+		# Should die when the copytool dies, but just in case.
+		local cmd="kill \\\$(cat $HSMTOOL_MONITOR_DIR/monitor_pid)"
+		cmd+=" 2>/dev/null || true"
+		do_node $agent "$cmd"
+		do_node $agent "rm -fr $HSMTOOL_MONITOR_DIR"
+		export HSMTOOL_MONITOR_DIR=
+	fi
+
+	# The pdsh should die on its own when the monitor dies. Just
+	# in case, though, try to clean up to avoid any cruft.
+	if [ -n "$HSMTOOL_MONITOR_PDSH" ]; then
+		kill $HSMTOOL_MONITOR_PDSH 2>/dev/null
+		export HSMTOOL_MONITOR_PDSH=
 	fi
+}
+
+copytool_setup() {
+	local facet=${1:-$SINGLEAGT}
+	# Use MOUNT2 by default if defined
+	local lustre_mntpnt=${2:-${MOUNT2:-$MOUNT}}
+	local arc_id=$3
+	local hsm_root=${4:-$(copytool_device $facet)}
+
+	[ -z "${hsm_root// /}" ] && error "copytool_setup: hsm_root empty!"
+
+	local agent=$(facet_active_host $facet)
 
 	if $HSM_ARCHIVE_PURGE; then
 		echo "Purging archive on $agent"
-		do_facet $facet "rm -rf $hsm_root/*"
+		do_facet $facet "rm -rf $hsm_root/$HSMTMP/*"
 	fi
 
 	echo "Starting copytool $facet on $agent"
-	do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
+	do_facet $facet "mkdir -p $hsm_root/$HSMTMP/" ||
+			error "mkdir '$hsm_root/$HSMTMP' failed"
 	# bandwidth is limited to 1MB/s so the copy time is known and
 	# independent of hardware
-	local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
+	local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon"
+	cmd+=" --hsm-root $hsm_root/$HSMTMP"
 	[[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
+	[[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
+		cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
+	[[ -z "$HSMTOOL_EVENT_FIFO" ]] ||
+		cmd+=" --event-fifo $HSMTOOL_EVENT_FIFO"
 	cmd+=" --bandwidth 1 $lustre_mntpnt"
 
 	# Redirect the standard output and error to a log file which
@@ -153,18 +286,85 @@ copytool_setup() {
 	[[ -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"
+	do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1"
+	if [[ $? !=  0 ]]; then
+		[[ $HSMTOOL_NOERROR == true ]] ||
+			error "start copytool $facet on $agent failed"
+		echo "start copytool $facet on $agent failed"
+	fi
+
 	trap cleanup EXIT
 }
 
+get_copytool_event_log() {
+	local facet=${1:-$SINGLEAGT}
+	local agent=$(facet_active_host $facet)
+
+	[ -z "$HSMTOOL_MONITOR_DIR" ] &&
+		error "Can't get event log: No monitor directory!"
+
+	do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
+		error "Could not collect event log from $agent"
+}
+
 copytool_cleanup() {
 	trap - EXIT
-	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	local agt_facet=$SINGLEAGT
+	local agt_hosts=${1:-$(facet_active_host $agt_facet)}
+	local hsm_root=$(copytool_device $agt_facet)
 
-	do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
-	sleep 1
-	echo "Copytool is stopped on $agents"
+	[ -z "${hsm_root// /}" ] && error "copytool_cleanup: hsm_root empty!"
+
+	local i
+	local facet
+	local param
+	local -a state
+
+	kill_copytools $agt_hosts
+	wait_copytools $agt_hosts || error "copytools failed to stop"
+
+	# Clean all CDTs orphans requests from previous tests that
+	# would otherwise need to timeout to clear.
+	for ((i = 0; i < MDSCOUNT; i++)); do
+		facet=mds$((i + 1))
+		param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
+		state[$i]=$(do_facet $facet "$LCTL get_param -n $param")
+
+		# Skip already stopping or stopped CDTs.
+		[[ "${state[$i]}" =~ ^stop ]] && continue
+
+		do_facet $facet "$LCTL set_param $param=shutdown"
+	done
+
+	for ((i = 0; i < MDSCOUNT; i++)); do
+		# Only check and restore CDTs that we stopped in the first loop.
+		[[ "${state[$i]}" =~ ^stop ]] && continue
+
+		facet=mds$((i + 1))
+		param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
+
+		wait_result $facet "$LCTL get_param -n $param" stopped 20 ||
+			error "$facet CDT state is not stopped"
+
+		# Restore old CDT state.
+		do_facet $facet "$LCTL set_param $param=${state[$i]}"
+	done
+
+	for ((i = 0; i < MDSCOUNT; i++)); do
+		# Only check CDTs that we stopped in the first loop.
+		[[ "${state[$i]}" =~ ^stop ]] && continue
+
+		facet=mds$((i + 1))
+		param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
+
+		# Check that the old CDT state was restored.
+		wait_result $facet "$LCTL get_param -n $param" "${state[$i]}" \
+			20 || error "$facet CDT state is not '${state[$i]}'"
+	done
+
+	if do_facet $agt_facet "df $hsm_root" >/dev/null 2>&1 ; then
+		do_facet $agt_facet "rm -rf $hsm_root/$HSMTMP/*"
+	fi
 }
 
 copytool_suspend() {
@@ -174,9 +374,16 @@ copytool_suspend() {
 	echo "Copytool is suspended on $agents"
 }
 
+copytool_continue() {
+	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+
+	do_nodesv $agents "pkill -CONT -x $HSMTOOL_BASE" || return 0
+	echo "Copytool is continued 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"
 	do_facet $SINGLEAGT rm -f $be
 }
@@ -192,7 +399,7 @@ make_archive() {
 	local file=$HSM_ARCHIVE/$1
 	do_facet $SINGLEAGT mkdir -p $(dirname $file)
 	do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
-		error "cannot create $file"
+		file_creation_failure dd $file $?
 }
 
 copy2archive() {
@@ -201,20 +408,66 @@ copy2archive() {
 	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() {
@@ -235,64 +488,64 @@ set_hsm_param() {
 	local param=$1
 	local value=$2
 	local opt=$3
-	if [[ "$value" != "" ]]; then
-		value="=$value"
-	fi
-	do_facet $SINGLEMDS $LCTL set_param $opt -n $HSM_PARAM.$param$value
+	mdts_set_param "$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() {
 	if [[ "$CDT_POLICY_HAD_CHANGED" ]]
 	then
 		# clear all
-		do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NRA
-		do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NBR
+		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_non_blocking_restore() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NBR
+	mdts_set_param "" hsm.policy "+NBR"
 	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_non_blocking_restore() {
-	do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NBR
+	mdts_set_param "" hsm.policy "-NBR"
 	CDT_POLICY_HAD_CHANGED=true
 }
 
 cdt_clear_mount_state() {
-	do_facet $SINGLEMDS $LCTL set_param -d -P $MDT_PARAM.hsm_control
+	mdts_set_param "-P -d" hsm_control ""
 }
 
 cdt_set_mount_state() {
-	do_facet $SINGLEMDS $LCTL set_param -P $MDT_PARAM.hsm_control=$1
+	mdts_set_param "-P" hsm_control "$1"
+	# set_param -P is asynchronous operation and could race with set_param.
+	# In such case configs could be retrieved and applied at mgc after
+	# set_param -P completion. Sleep here to avoid race with set_param.
+	# We need at least 20 seconds. 10 for mgc_requeue_thread to wake up
+	# MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC(5 + 5)
+	# and 10 seconds to retrieve config from server.
+	sleep 20
 }
 
 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() {
@@ -328,17 +581,19 @@ needclients() {
 
 path2fid() {
 	$LFS path2fid $1 | tr -d '[]'
+	return ${PIPESTATUS[0]}
 }
 
 get_hsm_flags() {
 	local f=$1
 	local u=$2
+	local st
 
 	if [[ $u == "user" ]]; then
-		local st=$($RUNAS $LFS hsm_state $f)
+		st=$($RUNAS $LFS hsm_state $f)
 	else
-		local st=$($LFS hsm_state $f)
 		u=root
+		st=$($LFS hsm_state $f)
 	fi
 
 	[[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)"
@@ -349,7 +604,8 @@ get_hsm_flags() {
 
 get_hsm_archive_id() {
 	local f=$1
-	local st=$($LFS hsm_state $f)
+	local st
+	st=$($LFS hsm_state $f)
 	[[ $? == 0 ]] || error "$LFS hsm_state $f failed"
 
 	local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " |
@@ -373,6 +629,15 @@ check_hsm_flags_user() {
 	[[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
 }
 
+file_creation_failure() {
+	local cmd=$1
+	local f=$2
+	local err=$3
+
+	df $MOUNT $MOUNT2 >&2
+	error "cannot create $f with $cmd, status=$err"
+}
+
 copy_file() {
 	local f=
 
@@ -386,69 +651,53 @@ copy_file() {
 		f=${f/$DIR/$DIR2}
 	fi
 	rm -f $f
-	cp $1 $f || error "cannot copy $1 to $f"
+	cp $1 $f || file_creation_failure cp $f $?
+
 	path2fid $f || error "cannot get fid on $f"
 }
 
 make_small() {
         local file2=${1/$DIR/$DIR2}
         dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync ||
-		error "cannot create $file2"
-        path2fid $1 || error "cannot get fid on $1"
-}
+		file_creation_failure dd $file2 $?
 
-cleanup_large_files() {
-	local ratio=$(df $MOUNT |awk '{print $5}' |sed 's/%//g' |grep -v Use)
-	[ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
+        path2fid $1 || error "cannot get fid on $1"
 }
 
-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"
+make_small_sync() {
+	dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
+		file_creation_failure dd $1 $?
 	path2fid $1 || error "cannot get fid on $1"
 }
 
-make_large_for_progress() {
-	local file2=${1/$DIR/$DIR2}
-
-	cleanup_large_files
-
-	# big file is large enough, so copy time is > 30s
-	# so copytool make 1 progress
-	# size is not a multiple of 1M to avoid stripe
-	# aligment
-	dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync ||
-		error "cannot create $file2"
-	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_progress_aligned() {
-	local file2=${1/$DIR/$DIR2}
-
-	cleanup_large_files
-
-	# big file is large enough, so copy time is > 30s
-	# so copytool make 1 progress
-	# size is a multiple of 1M to have stripe
-	# aligment
-	dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync ||
-		error "cannot create $file2"
-	path2fid $1 || error "cannot get fid on $1"
+check_enough_free_space() {
+	local nb=$1
+	local unit=$2
+	local need=$((nb * unit /1024))
+	local free=$(df -kP $MOUNT | tail -1 | awk '{print $4}')
+	(( $need >= $free )) && return 1
+	return 0
 }
 
-make_large_for_cancel() {
+make_custom_file_for_progress() {
 	local file2=${1/$DIR/$DIR2}
+	local fsize=${2:-"39"}
+	local blksz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
+	blksz=${3:-$blksz}
 
-	cleanup_large_files
+	[[ $fsize -gt  0 ]] || error "Invalid file size"
+	[[ $blksz -gt 0 ]] || error "Invalid stripe size"
 
-	# Copy timeout is 100s. 105MB => 105s
-	dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
-		error "cannot create $file2"
+	cleanup_large_files
+	check_enough_free_space $fsize $blksz || return $?
+	dd if=/dev/zero of=$file2 count=$fsize bs=$blksz conv=fsync ||
+		file_creation_failure dd $file2 $?
 	path2fid $1 || error "cannot get fid on $1"
 }
 
@@ -462,12 +711,15 @@ 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 $HSM_PARAM.actions"
+	local cmd="$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 200 ||
+		error "request on $fid is not $state on $mds"
 }
 
 get_request_state() {
@@ -486,10 +738,21 @@ get_request_count() {
 		"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
 }
 
+# Ensure the number of HSM request for a given FID is correct
+# assert_request_count FID REQUEST_TYPE COUNT [ERROR_MSG]
+assert_request_count() {
+	local request_count=$(get_request_count $1 $2)
+	local default_error_msg=("expected $3 '$2' request(s) for '$1', found "
+				"'$request_count'")
+	[ $request_count -eq $3 ] || error "${4:-"${default_error_msg[@]}"}"
+}
+
 wait_all_done() {
 	local timeout=$1
+	local fid=$2
 
 	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
+	[[ -n $fid ]] && cmd+=" | grep '$fid'"
 	cmd+=" | egrep 'WAITING|STARTED'"
 
 	wait_result $SINGLEMDS "$cmd" "" $timeout ||
@@ -501,14 +764,91 @@ wait_for_grace_delay() {
 	sleep $val
 }
 
-MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
-	awk '{gsub(/_UUID/,""); print $1}' | head -1)
+wait_for_loop_period() {
+	local val=$(get_hsm_param loop_period)
+	sleep $val
+}
+
+parse_json_event() {
+	local raw_event=$1
+
+	# python2.6 in EL6 includes an internal json module
+	local json_parser='import json; import fileinput;'
+	json_parser+=' print "\n".join(["local %s=\"%s\"" % tuple for tuple in '
+	json_parser+='json.loads([line for line in '
+	json_parser+='fileinput.input()][0]).items()])'
+
+	echo $raw_event | python -c "$json_parser"
+}
+
+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 $((mdsno - 1))
+	done
+}
+
+get_agent_uuid() {
+	local agent=${1:-$(facet_active_host $SINGLEAGT)}
+
+	# Lustre mount-point is mandatory and last parameter on
+	# copytool cmd-line.
+	local mntpnt=$(do_rpc_nodes $agent ps -C $HSMTOOL_BASE -o args= |
+		       awk '{print $NF}')
+	[ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
+				  "parameter"
+	do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
+}
 
 # initiate variables
 init_agt_vars
 
+# populate MDT device array
+get_mdt_devices
+
 # cleanup from previous bad setup
-search_and_kill_copytool
+kill_copytools
 
 # for recovery tests, coordinator needs to be started at mount
 # so force it
@@ -517,9 +857,6 @@ echo "Set HSM on and start"
 cdt_set_mount_state enabled
 cdt_check_state enabled
 
-echo "Start copytool"
-copytool_setup
-
 echo "Set sanity-hsm HSM policy"
 cdt_set_sanity_policy
 
@@ -561,6 +898,59 @@ test_1() {
 }
 run_test 1 "lfs hsm flags root/non-root access"
 
+test_1a() {
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(make_small $f)
+
+	copytool_setup
+
+	$LFS hsm_archive $f || error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	# Release and check states
+	$LFS hsm_release $f || error "could not release file"
+	echo -n "Verifying released state: "
+	check_hsm_flags $f "0x0000000d"
+
+	$MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
+
+	copytool_cleanup
+}
+run_test 1a "mmap & cat a HSM released file"
+
+test_1b() {
+	mkdir -p $DIR/$tdir
+	$LFS setstripe -E 1M -E 64M -c 2 -E -1 -c 4 $DIR/$tdir ||
+		error "failed to set default stripe"
+	local f=$DIR/$tdir/$tfile
+	rm -f $f
+
+	dd if=/dev/random of=$f bs=1M count=1 conv=sync ||
+		error "failed to create file"
+	local fid=$(path2fid $f)
+
+	copytool_setup
+
+	echo "archive $f"
+	$LFS hsm_archive $f || error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	echo "release $f"
+	$LFS hsm_release $f || error "could not release file"
+	echo "verify released state: "
+	check_hsm_flags $f "0x0000000d" && echo "pass"
+
+	echo "restore $f"
+	$LFS hsm_restore $f || error "could not restore file"
+	wait_request_state $fid RESTORE SUCCEED
+	echo "verify restored state: "
+	check_hsm_flags $f "0x00000009" && echo "pass"
+
+	copytool_cleanup
+}
+run_test 1b "Archive, Release & Restore composite file"
+
 test_2() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
@@ -616,7 +1006,7 @@ test_3() {
 		error "user could not change hsm flags"
 	dd if=/etc/passwd of=$f.append bs=1 count=3\
 	   conv=notrunc oflag=append status=noxfer ||
-		error "could not append to test file"
+		file_creation_failure dd $f.append $?
 	check_hsm_flags $f.append "0x00000003"
 
 	# Modify a file sets it dirty
@@ -625,7 +1015,7 @@ test_3() {
 		error "user could not change hsm flags"
 	dd if=/dev/zero of=$f.modify bs=1 count=3\
 	   conv=notrunc status=noxfer ||
-		error "could not modify test file"
+		file_creation_failure dd $f.modify $?
 	check_hsm_flags $f.modify "0x00000003"
 
 	# Open O_TRUNC sets dirty
@@ -672,13 +1062,19 @@ test_8() {
 run_test 8 "Test default archive number"
 
 test_9() {
-	mkdir -p $DIR/$tdir
-	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))
 	copytool_cleanup
 	copytool_setup $SINGLEAGT $MOUNT $new_an
+
+	# give time for CT to register with MDTs
+	sleep $(($MDSCOUNT*2))
+	local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
+	check_agent_registered $uuid
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
 	$LFS hsm_archive --archive $new_an $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -686,7 +1082,7 @@ test_9() {
 
 	copytool_cleanup
 }
-run_test 9 "Use of explict archive number, with dedicated copytool"
+run_test 9 "Use of explicit archive number, with dedicated copytool"
 
 test_9a() {
 	needclients 3 || return 0
@@ -799,7 +1195,7 @@ test_10d() {
 }
 run_test 10d "Archive a file on the default archive id"
 
-test_11() {
+test_11a() {
 	mkdir -p $DIR/$tdir
 	copy2archive /etc/hosts $tdir/$tfile
 	local f=$DIR/$tdir/$tfile
@@ -823,7 +1219,31 @@ test_11() {
 	local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
 		error "fid $fid not in archive $HSM_ARCHIVE"
 }
-run_test 11 "Import a file"
+run_test 11a "Import a file"
+
+test_11b() {
+	# 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 -a $HSM_ARCHIVE_NUMBER $f ||
+		error "hsm_archive failed"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local FILE_HASH=$(md5sum $f)
+	rm -f $f
+
+	import_file $fid $f
+
+	echo "$FILE_HASH" | md5sum -c
+
+	[[ $? -eq 0 ]] || error "Restored file differs"
+
+	copytool_cleanup
+}
+run_test 11b "Import a deleted file using its FID"
 
 test_12a() {
 	# test needs a running copytool
@@ -834,16 +1254,16 @@ test_12a() {
 
 	local f=$DIR/$tdir/$tfile
 	import_file $tdir/$tfile $f
-	local f=$DIR2/$tdir/$tfile
+	local f2=$DIR2/$tdir/$tfile
 	echo "Verifying released state: "
-	check_hsm_flags $f "0x0000000d"
+	check_hsm_flags $f2 "0x0000000d"
 
-	local fid=$(path2fid $f)
-	$LFS hsm_restore $f
+	local fid=$(path2fid $f2)
+	$LFS hsm_restore $f2
 	wait_request_state $fid RESTORE SUCCEED
 
 	echo "Verifying file state: "
-	check_hsm_flags $f "0x00000009"
+	check_hsm_flags $f2 "0x00000009"
 
 	do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
 
@@ -887,7 +1307,10 @@ test_12c() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	$LFS setstripe -c 2 $f
-	local fid=$(make_large_for_striping $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 5)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	local FILE_CRC=$(md5sum $f)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -1063,43 +1486,185 @@ test_12n() {
 }
 run_test 12n "Import/implicit restore/release"
 
-test_13() {
+test_12o() {
 	# test needs a running copytool
 	copytool_setup
 
-	local ARC_SUBDIR="import.orig"
-	local d=""
-	local f=""
-
-	# populate directory to be imported
-	for d in $(seq 1 10); do
-		local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
-		do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
-		for f in $(seq 1 10); do
-			CURR_FILE="$CURR_DIR/$tfile.$f"
-			# write file-specific data
-			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=$(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)
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/hosts $f)
 
-	[ $cnt_diff -eq 0 ] ||
-		error "$cnt_diff imported files differ from read data"
-	[ $cnt_ok -eq 100 ] ||
-		error "not enough identical files ($cnt_ok != 100)"
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f || error "release of $f failed"
 
-	copytool_cleanup
-}
-run_test 13 "Recursively import and restore a directory"
+#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
+
+	# to be sure previous RESTORE result is gone
+	cdt_purge
+	wait_for_grace_delay
+
+	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_12p() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir $DIR/$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
+	do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
+	$LFS hsm_release $f || error "cannot release $f"
+	do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
+	$LFS hsm_release $f || error "cannot release $f"
+	do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
+
+	copytool_cleanup
+}
+run_test 12p "implicit restore of a file on copytool mount point"
+
+cleanup_test_12q() {
+	trap 0
+	zconf_umount $(facet_host $SINGLEAGT) $MOUNT3 ||
+		error "cannot umount $MOUNT3 on $SINGLEAGT"
+}
+
+test_12q() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.58) ] &&
+		skip "need MDS version at least 2.7.58" && return 0
+
+	zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
+		error "cannot mount $MOUNT3 on $SINGLEAGT"
+
+	trap cleanup_test_12q EXIT
+
+	# test needs a running copytool
+	copytool_setup $SINGLEAGT $MOUNT3
+
+	mkdir $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local f2=$DIR2/$tdir/$tfile
+	local fid=$(make_small $f)
+	local orig_size=$(stat -c "%s" $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	$LFS hsm_release $f || error "could not release file"
+	check_hsm_flags $f "0x0000000d"
+
+	kill_copytools
+	wait_copytools || error "copytool failed to stop"
+
+	cat $f > /dev/null &
+
+	# wait a bit to allow implicit restore request to be handled.
+	# if not, next stat would also block on layout-lock.
+	sleep 5
+
+	local size=$(stat -c "%s" $f2)
+	[ $size -eq $orig_size ] ||
+		error "$f2: wrong size after archive: $size != $orig_size"
+
+	HSM_ARCHIVE_PURGE=false copytool_setup $SINGLEAGT /mnt/lustre3
+
+	wait
+
+	size=$(stat -c "%s" $f)
+	[ $size -eq $orig_size ] ||
+		error "$f: wrong size after restore: $size != $orig_size"
+
+	size=$(stat -c "%s" $f2)
+	[ $size -eq $orig_size ] ||
+		error "$f2: wrong size after restore: $size != $orig_size"
+
+	:>$f
+
+	size=$(stat -c "%s" $f)
+	[ $size -eq 0 ] ||
+		error "$f: wrong size after overwrite: $size != 0"
+
+	size=$(stat -c "%s" $f2)
+	[ $size -eq 0 ] ||
+		error "$f2: wrong size after overwrite: $size != 0"
+
+	copytool_cleanup
+	zconf_umount $(facet_host $SINGLEAGT) $MOUNT3 ||
+		error "cannot umount $MOUNT3 on $SINGLEAGT"
+}
+run_test 12q "file attributes are refreshed after restore"
+
+test_13() {
+	# test needs a running copytool
+	copytool_setup
+
+	local ARC_SUBDIR="import.orig"
+	local d=""
+	local f=""
+
+	# populate directory to be imported
+	for d in $(seq 1 10); do
+		local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
+		do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
+		for f in $(seq 1 10); do
+			CURR_FILE="$CURR_DIR/$tfile.$f"
+			# write file-specific data
+			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=$(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"
+	[ $cnt_ok -eq 100 ] ||
+		error "not enough identical files ($cnt_ok != 100)"
+
+	copytool_cleanup
+}
+run_test 13 "Recursively import and restore a directory"
 
 test_14() {
 	# test needs a running copytool
@@ -1208,9 +1773,10 @@ test_16() {
 	$LFS hsm_archive $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	local end=$(date +%s)
-	local duration=$((end - start))
+	# Add 1 to account for rounding errors between start and end (LU-8155)
+	local duration=$((end - start + 1))
 
-	[[ $duration -ge $goal ]] ||
+	[[ $duration -ge $((goal - 1)) ]] ||
 		error "Transfer is too fast $duration < $goal"
 
 	copytool_cleanup
@@ -1257,19 +1823,41 @@ 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" $f) -ne "1" ] || error "wrong block number"
-	local sz=$(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 "1" ] || 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
@@ -1287,6 +1875,8 @@ 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"
@@ -1418,8 +2008,7 @@ test_24a() {
 	[ $ctime0 -eq $ctime1 ] ||
 		error "release changed ctime from $ctime0 to $ctime1"
 
-	# Restore should not change atime or mtime and should not
-	# decrease ctime.
+	# Restore should not change any timestamps.
 	$LFS hsm_restore $file
 	wait_request_state $fid RESTORE SUCCEED
 
@@ -1433,7 +2022,7 @@ test_24a() {
 	[ $mtime0 -eq $mtime1 ] ||
 		error "restore changed mtime from $mtime0 to $mtime1"
 
-	[ $ctime0 -le $ctime1 ] ||
+	[ $ctime0 -eq $ctime1 ] ||
 		error "restore changed ctime from $ctime0 to $ctime1"
 
 	copytool_cleanup
@@ -1452,7 +2041,7 @@ test_24a() {
 	[ $mtime0 -eq $mtime1 ] ||
 		error "remount changed mtime from $mtime0 to $mtime1"
 
-	[ $ctime0 -le $ctime1 ] ||
+	[ $ctime0 -eq $ctime1 ] ||
 		error "remount changed ctime from $ctime0 to $ctime1"
 }
 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
@@ -1468,7 +2057,7 @@ test_24b() {
 	copytool_setup
 	mkdir -p $DIR/$tdir
 
-	# Check that root can do HSM actions on a ordinary user's file.
+	# Check that root can do HSM actions on a regular user's file.
 	rm -f $file
 	fid=$(make_small $file)
 	sum0=$(md5sum $file)
@@ -1546,7 +2135,6 @@ test_24c() {
 	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"
 
@@ -1560,7 +2148,6 @@ test_24c() {
 	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"
 
@@ -1574,7 +2161,6 @@ test_24c() {
 	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"
 
@@ -1590,6 +2176,7 @@ run_test 24c "check that user,group,other request masks work"
 cleanup_test_24d() {
 	trap 0
 	mount -o remount,rw $MOUNT2
+	zconf_umount $(facet_host $SINGLEAGT) "$MOUNT3"
 }
 
 test_24d() {
@@ -1598,16 +2185,24 @@ test_24d() {
 	local fid1
 	local fid2
 
-	copytool_setup
-
 	mkdir -p $DIR/$tdir
 	rm -f $file1
 	fid1=$(make_small $file1)
 
+	echo $fid1
+	$LFS getstripe $file1
+
 	trap cleanup_test_24d EXIT
+	zconf_mount $(facet_host $SINGLEAGT) "$MOUNT3" ||
+		error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
+	copytool_setup $SINGLEAGT "$MOUNT3" ||
+		error "unable to setup a copytool for the test"
 
 	mount -o remount,ro $MOUNT2
 
+	do_nodes $(comma_list $(nodes_list)) $LCTL clear
+	start_full_debug_logging
+
 	fid2=$(path2fid $file2)
 	[ "$fid1" == "$fid2" ] ||
 		error "FID mismatch '$fid1' != '$fid2'"
@@ -1616,15 +2211,17 @@ test_24d() {
 		error "archive should fail on read-only mount"
 	check_hsm_flags $file1 "0x00000000"
 
-	$LFS hsm_archive $file1
+	$LFS hsm_archive $file1 || error "Fail to archive $file1"
 	wait_request_state $fid1 ARCHIVE SUCCEED
 
+	stop_full_debug_logging
+
 	$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'"
+	dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
 
 	$LFS hsm_release $file2 &&
 		error "release should fail on read-only mount"
@@ -1634,6 +2231,55 @@ test_24d() {
 }
 run_test 24d "check that read-only mounts are respected"
 
+test_24e() {
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	local fid
+
+	fid=$(make_small $f) || error "cannot create $f"
+	$LFS hsm_archive $f || error "cannot archive $f"
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f || error "cannot release $f"
+	while ! $LFS hsm_state $f | grep released; do
+		sleep 1
+	done
+
+	tar -cf $TMP/$tfile.tar $DIR/$tdir || error "cannot tar $DIR/$tdir"
+
+	copytool_cleanup
+}
+run_test 24e "tar succeeds on HSM released files" # LU-6213
+
+test_24f() {
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir/d1
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/hosts $f)
+	sum0=$(md5sum $f)
+	echo $sum0
+	$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
+		error "hsm_archive failed"
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f || error "cannot release $f"
+	tar --xattrs -cvf $f.tar -C $DIR/$tdir $tfile
+	rm -f $f
+	sync
+	tar --xattrs -xvf $f.tar -C $DIR/$tdir ||
+		error "Can not recover the tar contents"
+	sum1=$(md5sum $f)
+	echo "Sum0 = $sum0, sum1 = $sum1"
+	[ "$sum0" == "$sum1" ] || error "md5sum mismatch for '$tfile'"
+
+	copytool_cleanup
+}
+run_test 24f "root can archive, release, and restore tar files"
+
 test_25a() {
 	# test needs a running copytool
 	copytool_setup
@@ -1687,7 +2333,10 @@ test_26() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
@@ -1700,93 +2349,457 @@ test_26() {
 }
 run_test 26 "Remove the archive of a valid file"
 
-test_27a() {
+cleanup_test_26a() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	set_hsm_param loop_period $orig_loop_period
+	set_hsm_param grace_delay $orig_grace_delay
+	copytool_cleanup
+}
+
+test_26a() {
+	local raolu=$(get_hsm_param remove_archive_on_last_unlink)
+	[[ $raolu -eq 0 ]] || error "RAoLU policy should be off"
+
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
-	make_archive $tdir/$tfile
 	local f=$DIR/$tdir/$tfile
-	import_file $tdir/$tfile $f
-	local fid=$(path2fid $f)
+	local fid=$(copy_file /etc/passwd $f)
 
-	$LFS hsm_remove $f
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
 
-	[[ $? != 0 ]] || error "Remove of a released file should fail"
+	local f2=$DIR/$tdir/${tfile}_2
+	local fid2=$(copy_file /etc/passwd $f2)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
+	wait_request_state $fid2 ARCHIVE SUCCEED
+
+	local f3=$DIR/$tdir/${tfile}_3
+	local fid3=$(copy_file /etc/passwd $f3)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f3
+	wait_request_state $fid3 ARCHIVE SUCCEED
 
+	trap cleanup_test_26a EXIT
+
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(get_hsm_param loop_period)
+	local orig_grace_delay=$(get_hsm_param grace_delay)
+	set_hsm_param loop_period 10
+	set_hsm_param grace_delay 100
+
+	rm -f $f
+
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	ln "$f3" "$f3"_bis || error "Unable to create hard-link"
+	rm -f $f3
+
+	rm -f $f2
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid2 REMOVE SUCCEED
+
+	assert_request_count $fid REMOVE 0 \
+		"Unexpected archived data remove request for $f"
+	assert_request_count $fid3 REMOVE 0 \
+		"Unexpected archived data remove request for $f3"
+
+	cleanup_test_26a
+}
+run_test 26a "Remove Archive On Last Unlink (RAoLU) policy"
+
+cleanup_test_26b() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
 	copytool_cleanup
 }
-run_test 27a "Remove the archive of an imported file (Operation not permitted)"
 
-test_27b() {
+test_26b() {
+
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid=$(copy_file /etc/passwd $f)
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
-	$LFS hsm_release $f
 
-	$LFS hsm_remove $f
+	trap cleanup_test_26b EXIT
 
-	[[ $? != 0 ]] || error "Remove of a released file should fail"
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	cdt_shutdown
+	cdt_check_state stopped
+
+	rm -f $f
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid REMOVE WAITING
+
+	cdt_enable
+	# copytool must re-register
+	kill_copytools
+	wait_copytools || error "copytool failed to stop"
+	HSM_ARCHIVE_PURGE=false copytool_setup
+
+	wait_request_state $fid REMOVE SUCCEED
+
+	cleanup_test_26b
+}
+run_test 26b "RAoLU policy when CDT off"
 
+cleanup_test_26c() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	set_hsm_param loop_period $orig_loop_period
+	set_hsm_param grace_delay $orig_grace_delay
 	copytool_cleanup
 }
-run_test 27b "Remove the archive of a relased file (Operation not permitted)"
 
-test_28() {
+test_26c() {
+
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid=$(copy_file /etc/passwd $f)
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	cdt_disable
-	$LFS hsm_remove $f
+	local f2=$DIR/$tdir/${tfile}_2
+	local fid2=$(copy_file /etc/passwd $f2)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
+	wait_request_state $fid2 ARCHIVE SUCCEED
+
+	trap cleanup_test_26c EXIT
+
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(get_hsm_param loop_period)
+	local orig_grace_delay=$(get_hsm_param grace_delay)
+	set_hsm_param loop_period 10
+	set_hsm_param grace_delay 100
+
+	set_hsm_param remove_archive_on_last_unlink 1
+
+	multiop_bg_pause $f O_c || error "open $f failed"
+	local pid=$!
 
 	rm -f $f
+	rm -f $f2
 
-	cdt_enable
+	wait_request_state $fid2 REMOVE SUCCEED
+	assert_request_count $fid REMOVE 0 \
+		"Unexpected archived data remove request for $f"
+
+	kill -USR1 $pid || error "multiop early exit"
+	# should reach autotest timeout if multiop fails to trap
+	# signal, close file, and exit ...
+	wait $pid || error
+
+	set_hsm_param remove_archive_on_last_unlink 0
 
 	wait_request_state $fid REMOVE SUCCEED
 
+	cleanup_test_26c
+}
+run_test 26c "RAoLU effective when file closed"
+
+cleanup_test_26d() {
+	trap 0
+	set_hsm_param remove_archive_on_last_unlink 0
+	set_hsm_param loop_period $orig_loop_period
+	set_hsm_param grace_delay $orig_grace_delay
 	copytool_cleanup
 }
-run_test 28 "Concurrent archive/file remove"
 
-test_30a() {
-	# restore at exec cannot work on agent node (because of Linux kernel
-	# protection of executables)
-	needclients 2 || return 0
+test_26d() {
 
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
-	copy2archive /bin/true $tdir/$tfile
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/motd $f 1)
 
-	local f=$DIR/$tdir/true
-	import_file $tdir/$tfile $f
+	$LFS hsm_archive $f || error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
 
-	local fid=$(path2fid $f)
+	trap cleanup_test_26d EXIT
 
-	# set no retry action mode
-	cdt_set_no_retry
-	do_node $CLIENT2 $f
-	local st=$?
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(get_hsm_param loop_period)
+	local orig_grace_delay=$(get_hsm_param grace_delay)
+	set_hsm_param loop_period 10
+	set_hsm_param grace_delay 100
 
-	# cleanup
-	# remove no try action mode
-	cdt_clear_no_retry
-	$LFS hsm_state $f
+	set_hsm_param remove_archive_on_last_unlink 1
 
-	[[ $st == 0 ]] || error "Failed to exec a released file"
+	multiop_bg_pause $f O_c || error "multiop failed"
+	local MULTIPID=$!
+
+	rm -f $f
+
+	mds_evict_client
+
+	set_hsm_param remove_archive_on_last_unlink 0
+
+	wait_request_state $fid REMOVE SUCCEED
+
+	client_up || client_up || true
+
+	kill -USR1 $MULTIPID
+	wait $MULTIPID || error "multiop close failed"
+
+	cleanup_test_26d
+}
+run_test 26d "RAoLU when Client eviction"
+
+test_27a() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	make_archive $tdir/$tfile
+	local f=$DIR/$tdir/$tfile
+	import_file $tdir/$tfile $f
+	local fid=$(path2fid $f)
+
+	$LFS hsm_remove $f
+
+	[[ $? != 0 ]] || error "Remove of a released file should fail"
+
+	copytool_cleanup
+}
+run_test 27a "Remove the archive of an imported file (Operation not permitted)"
+
+test_27b() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f
+
+	$LFS hsm_remove $f
+
+	[[ $? != 0 ]] || error "Remove of a released file should fail"
+
+	copytool_cleanup
+}
+run_test 27b "Remove the archive of a relased file (Operation not permitted)"
+
+test_28() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	cdt_disable
+	$LFS hsm_remove $f
+
+	rm -f $f
+
+	cdt_enable
+
+	wait_request_state $fid REMOVE SUCCEED
+
+	copytool_cleanup
+}
+run_test 28 "Concurrent archive/file remove"
+
+test_29a() {
+	# Tests --mntpath and --archive options
+
+	local archive_id=7
+	copytool_setup $SINGLEAGT $MOUNT $archive_id
+
+	# Bad archive number
+	$LFS hsm_remove -m $MOUNT -a 33 0x857765760:0x8:0x2 2>&1 |
+		grep "Invalid argument" ||
+		error "unexpected hsm_remove failure (1)"
+
+	# mntpath is present but file is given
+	$LFS hsm_remove --mntpath $MOUNT --archive 30 /qwerty/uyt 2>&1 |
+		grep "hsm: '/qwerty/uyt' is not a valid FID" ||
+		error "unexpected hsm_remove failure (2)"
+
+	copytool_cleanup
+}
+run_test 29a "Tests --mntpath and --archive options"
+
+test_29b() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(make_small $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	rm -f $f
+
+	$LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER $fid
+	wait_request_state $fid REMOVE SUCCEED
+
+	copytool_cleanup
+}
+run_test 29b "Archive/delete/remove by FID from the archive."
+
+test_29c() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local fid1=$(make_small $DIR/$tdir/$tfile-1)
+	local fid2=$(make_small $DIR/$tdir/$tfile-2)
+	local fid3=$(make_small $DIR/$tdir/$tfile-3)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$tfile-[1-3]
+	wait_request_state $fid1 ARCHIVE SUCCEED
+	wait_request_state $fid2 ARCHIVE SUCCEED
+	wait_request_state $fid3 ARCHIVE SUCCEED
+
+	rm -f $DIR/$tdir/$tfile-[1-3]
+
+	echo $fid1 > $DIR/$tdir/list
+	echo $fid2 >> $DIR/$tdir/list
+	echo $fid3 >> $DIR/$tdir/list
+
+	$LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER \
+		--filelist $DIR/$tdir/list
+	wait_request_state $fid1 REMOVE SUCCEED
+	wait_request_state $fid2 REMOVE SUCCEED
+	wait_request_state $fid3 REMOVE SUCCEED
+
+	copytool_cleanup
+}
+run_test 29c "Archive/delete/remove by FID, using a file list."
+
+test_29d() {
+	# test needs more than one CT
+	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 $MOUNT2 $n
+	done
+
+	trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
+	# archive files
+	mkdir -p $DIR/$tdir
+	file=$DIR/$tdir/$tfile
+	fid=$(make_small $file)
+
+	$LFS hsm_archive $file
+	wait_request_state $fid ARCHIVE SUCCEED
+	check_hsm_flags $file "0x00000009"
+
+	rm -f $file
+
+	$LFS hsm_remove --mntpath "$MOUNT" -a 0 $fid ||
+		error "cannot hsm_remove '$fid'"
+
+	# give time for CDT to handle remove request and create broadcasted
+	sleep 2
+
+	# remove request has been broadcasted ?
+	local cnt=$(get_request_count $fid REMOVE)
+	# broadcasted requests + original
+	[[ $cnt -eq $((AGTCOUNT + 1)) ]] ||
+		error "remove not broadcasted to all CTs"
+
+	# give time for CDT and CTs to handle broadcasted
+	wait_for_loop_period
+
+	# each agent serves one different archive_id, so broadcasted
+	# hsm_remove request should only succeed once and fail at all others
+	local res
+	local scnt=0
+	local fcnt=0
+	for n in $(seq $AGTCOUNT); do
+		res=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+			       $HSM_PARAM.actions | awk \
+			       '/'$fid'.*action=REMOVE archive#='$n'/ \
+			       {print \\\$13}' | cut -f2 -d=")
+		if [[ "$res" == "SUCCEED" ]]; then
+			scnt=$((scnt + 1))
+		elif [[ "$res" == "FAILED" ]]; then
+			fcnt=$((fcnt + 1))
+		fi
+	done
+
+	[[ $scnt -ne 1 ]] &&
+		error "one and only CT should have removed successfully"
+
+	[[ $AGTCOUNT -ne $((scnt + fcnt)) ]] &&
+		error "all but one CT should have failed to remove"
+
+	trap - EXIT
+	copytool_cleanup $(comma_list $(agts_nodes))
+
+}
+run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
+	     "request to be sent once for each registered archive_id"
+
+test_30a() {
+	# restore at exec cannot work on agent node (because of Linux kernel
+	# protection of executables)
+	needclients 2 || return 0
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	copy2archive /bin/true $tdir/$tfile
+
+	local f=$DIR/$tdir/true
+	import_file $tdir/$tfile $f
+
+	local fid=$(path2fid $f)
+
+	# set no retry action mode
+	cdt_set_no_retry
+	do_node $CLIENT2 $f
+	local st=$?
+
+	# cleanup
+	# remove no try action mode
+	cdt_clear_no_retry
+	$LFS hsm_state $f
+
+	[[ $st == 0 ]] || error "Failed to exec a released file"
 
 	copytool_cleanup
 }
@@ -1824,6 +2837,47 @@ 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 slp_sum1=$(md5sum /bin/sleep)
+	local fid=$(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
+	if [[ $? != 0 ]]; then
+		local slp_sum2=$(md5sum /bin/sleep)
+		# in case sleep file is modified during the test
+		[[ $slp_sum1 == $slp_sum2 ]] &&
+			error "Binary overwritten during exec"
+	fi
+
+	# 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=$2
@@ -1836,24 +2890,23 @@ restore_and_check_size() {
 	while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
 	do
 		n=$(stat -c "%s" $f)
-		# we echo in both cases to show stat is not
-		# hang
+		# we echo in both cases to show stat is not hang
 		if [[ $n != $s ]]; then
 			echo "size seen is $n != $s"
 			err=1
 		else
 			echo "size seen is right: $n == $s"
 		fi
-		st=$(get_hsm_flags $f)
 		sleep 10
 		cpt=$((cpt + 1))
+		st=$(get_hsm_flags $f)
 	done
-	if [[ $cpt -lt 10 ]]; then
-		echo " restore is too long"
-	else
+	if [[ "$st" = "0x00000009" ]]; then
 		echo " "done
+	else
+		echo " restore is too long"
+		wait_request_state $fid RESTORE SUCCEED
 	fi
-	wait_request_state $fid RESTORE SUCCEED
 	return $err
 }
 
@@ -1886,7 +2939,10 @@ test_31b() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1907,7 +2963,10 @@ test_31c() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress_aligned $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 33 1048576)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1928,11 +2987,23 @@ test_33() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
 
+	# to be sure wait_all_done will not be mislead by previous tests
+	# and ops.
+	cdt_purge
+	wait_for_grace_delay
+	# Also raise grace_delay significantly so the Canceled
+	# Restore action will stay enough long avail.
+	local old_grace=$(get_hsm_param grace_delay)
+	set_hsm_param grace_delay 100
+
 	md5sum $f >/dev/null &
 	local pid=$!
 	wait_request_state $fid RESTORE STARTED
@@ -1945,8 +3016,29 @@ test_33() {
 
 	$LFS hsm_cancel $f
 
-	wait_request_state $fid RESTORE CANCELED
-	wait_request_state $fid CANCEL SUCCEED
+	# instead of waiting+checking both Restore and Cancel ops
+	# sequentially, wait for both to be finished and then check
+	# each results.
+	wait_all_done 100 $fid
+	local rstate=$(get_request_state $fid RESTORE)
+	local cstate=$(get_request_state $fid CANCEL)
+
+	# restore orig grace_delay.
+	set_hsm_param grace_delay $old_grace
+
+	if [[ "$rstate" == "CANCELED" ]] ; then
+		[[ "$cstate" == "SUCCEED" ]] ||
+			error "Restore state is CANCELED and Cancel state " \
+			       "is not SUCCEED but $cstate"
+		echo "Restore state is CANCELED, Cancel state is SUCCEED"
+	elif [[ "$rstate" == "SUCCEED" ]] ; then
+		[[ "$cstate" == "FAILED" ]] ||
+			error "Restore state is SUCCEED and Cancel state " \
+				"is not FAILED but $cstate"
+		echo "Restore state is SUCCEED, Cancel state is FAILED"
+	else
+		error "Restore state is $rstate and Cancel state is $cstate"
+	fi
 
 	[ -z $killed ] ||
 		error "Cannot kill process waiting for restore ($killed)"
@@ -1962,7 +3054,10 @@ test_34() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -1995,7 +3090,10 @@ test_35() {
 
 	local f=$DIR/$tdir/$tfile
 	local f1=$DIR/$tdir/$tfile-1
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	local fid1=$(copy_file /etc/passwd $f1)
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
@@ -2031,7 +3129,10 @@ test_36() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 	$LFS hsm_release $f
@@ -2058,6 +3159,34 @@ test_36() {
 }
 run_test 36 "Move file during restore"
 
+test_37() {
+	# LU-5683: check that an archived dirty file can be rearchived.
+	copytool_cleanup
+	copytool_setup $SINGLEAGT $MOUNT2
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid
+
+	fid=$(make_small $f) || error "cannot create small file"
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f || error "cannot release $f"
+
+	# Allow previous archive request to expire from the actions log.
+	wait_for_grace_delay
+
+	# Dirty file.
+	dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	copytool_cleanup
+}
+run_test 37 "re-archive a dirty file"
+
 multi_archive() {
 	local prefix=$1
 	local count=$2
@@ -2083,7 +3212,13 @@ test_40() {
 			fid=$(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 do_facet $SINGLEAGT "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
@@ -2162,7 +3297,7 @@ test_54() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_small $f)
+	local fid=$(make_custom_file_for_progress $f 39 1000000)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -2190,7 +3325,7 @@ test_55() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_small $f)
+	local fid=$(make_custom_file_for_progress $f 39 1000000)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -2218,7 +3353,9 @@ test_56() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -2327,10 +3464,352 @@ test_58() {
 }
 run_test 58 "Truncate a released file will trigger restore"
 
-test_90() {
-	file_count=57
+test_59() {
+	local fid
+	local server_version=$(lustre_version_code $SINGLEMDS)
+	[[ $server_version -lt $(version_code 2.7.63) ]] &&
+		skip "Need MDS version at least 2.7.63" && return
+
+	copytool_setup
+	$MCREATE $DIR/$tfile || error "mcreate failed"
+	$TRUNCATE $DIR/$tfile 42 || error "truncate failed"
+	$LFS hsm_archive $DIR/$tfile || error "archive request failed"
+	fid=$(path2fid $DIR/$tfile)
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $DIR/$tfile || error "release failed"
+	copytool_cleanup
+}
+run_test 59 "Release stripeless file with non-zero size"
+
+test_60() {
+	# This test validates the fix for LU-4512. Ensure that the -u
+	# option changes the progress reporting interval from the
+	# default (30 seconds) to the user-specified interval.
+	local interval=5
+	local progress_timeout=$((interval * 4))
+
+	# test needs a new running copytool
+	copytool_cleanup
+	HSMTOOL_UPDATE_INTERVAL=$interval copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid
+	fid=$(make_custom_file_for_progress $f 10)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	local mdtidx=0
+	local mdt=${MDT_PREFIX}${mdtidx}
+	local mds=mds$((mdtidx + 1))
+
+	# Wait for copytool to register
+	wait_update_facet $mds \
+		"$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
+		uuid 100 || error "coyptool failed to register with $mdt"
+
+	local start_at=$(date +%s)
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
+		error "could not archive file"
+
+	local agent=$(facet_active_host $SINGLEAGT)
+	local prefix=$TESTLOG_PREFIX
+	[[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
+	local copytool_log=$prefix.copytool_log.$agent.log
+
+
+	wait_update $agent \
+	    "grep -o start.copy $copytool_log" "start copy" 100 ||
+		error "copytool failed to start"
+
+	local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
+	cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
+
+	local RESULT
+	local WAIT=0
+	local sleep=1
+
+	echo -n "Expecting a progress update within $progress_timeout seconds... "
+	while [ true ]; do
+		RESULT=$(do_node $(facet_active_host $mds) "$cmd")
+		if [ $RESULT -gt 0 ]; then
+			echo "$RESULT bytes copied in $WAIT seconds."
+			break
+		elif [ $WAIT -ge $progress_timeout ]; then
+			error "Timed out waiting for progress update!"
+			break
+		fi
+		WAIT=$((WAIT + sleep))
+		sleep $sleep
+	done
+
+	local finish_at=$(date +%s)
+	local elapsed=$((finish_at - start_at))
+
+	# Ensure that the progress update occurred within the expected window.
+	if [ $elapsed -lt $((interval - 1)) ]; then
+		error "Expected progress update after at least $interval seconds"
+	fi
+
+	echo "Wait for on going archive hsm action to complete"
+	wait_update $agent "grep -o copied $copytool_log" "copied" 10 ||
+		echo "File archiving not completed even after 10 secs"
+
+	cdt_clear_no_retry
+	copytool_cleanup
+}
+run_test 60 "Changing progress update interval from default"
+
+test_61() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+	cdt_disable
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	rm -f $f
+	cdt_enable
+	wait_request_state $fid ARCHIVE FAILED
+
+	copytool_cleanup
+}
+run_test 61 "Waiting archive of a removed file should fail"
+
+test_70() {
+	# test needs a new running copytool
+	copytool_cleanup
+	copytool_monitor_setup
+	HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
+
+	# Just start and stop the copytool to generate events.
+	cdt_clear_no_retry
+
+	# Wait for the copytool to register.
+	wait_update --verbose $(facet_active_host mds1) \
+		"$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
+		uuid 100 ||
+		error "copytool failed to register with MDT0000"
+
+	copytool_cleanup
+
+	local REGISTER_EVENT
+	local UNREGISTER_EVENT
+	while read event; do
+		local parsed=$(parse_json_event "$event")
+		if [ -z "$parsed" ]; then
+			error "Copytool sent malformed event: $event"
+		fi
+		eval $parsed
+
+		if [ $event_type == "REGISTER" ]; then
+			REGISTER_EVENT=$event
+		elif [ $event_type == "UNREGISTER" ]; then
+			UNREGISTER_EVENT=$event
+		fi
+	done < <(echo $"$(get_copytool_event_log)")
+
+	if [ -z "$REGISTER_EVENT" ]; then
+		error "Copytool failed to send register event to FIFO"
+	fi
+
+	if [ -z "$UNREGISTER_EVENT" ]; then
+		error "Copytool failed to send unregister event to FIFO"
+	fi
+
+	copytool_monitor_cleanup
+	echo "Register/Unregister events look OK."
+}
+run_test 70 "Copytool logs JSON register/unregister events to FIFO"
+
+test_71() {
+	# Bump progress interval for livelier events.
+	local interval=5
+
+	# test needs a new running copytool
+	copytool_cleanup
+	copytool_monitor_setup
+	HSMTOOL_UPDATE_INTERVAL=$interval \
+	HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
+		error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local expected_fields="event_time data_fid source_fid"
+	expected_fields+=" total_bytes current_bytes"
+
+	local START_EVENT
+	local FINISH_EVENT
+	while read event; do
+		# Make sure we're not getting anything from previous events.
+		for field in $expected_fields; do
+			unset $field
+		done
+
+		local parsed=$(parse_json_event "$event")
+		if [ -z "$parsed" ]; then
+			error "Copytool sent malformed event: $event"
+		fi
+		eval $parsed
+
+		if [ $event_type == "ARCHIVE_START" ]; then
+			START_EVENT=$event
+			continue
+		elif [ $event_type == "ARCHIVE_FINISH" ]; then
+			FINISH_EVENT=$event
+			continue
+		elif [ $event_type != "ARCHIVE_RUNNING" ]; then
+			continue
+		fi
+
+		# Do some simple checking of the progress update events.
+		for expected_field in $expected_fields; do
+			if [ -z ${!expected_field+x} ]; then
+				error "Missing $expected_field field in event"
+			fi
+		done
+
+		if [ $total_bytes -eq 0 ]; then
+			error "Expected total_bytes to be > 0"
+		fi
+
+		# These should be identical throughout an archive
+		# operation.
+		if [ $source_fid != $data_fid ]; then
+			error "Expected source_fid to equal data_fid"
+		fi
+	done < <(echo $"$(get_copytool_event_log)")
+
+	if [ -z "$START_EVENT" ]; then
+		error "Copytool failed to send archive start event to FIFO"
+	fi
+
+	if [ -z "$FINISH_EVENT" ]; then
+		error "Copytool failed to send archive finish event to FIFO"
+	fi
+
+	echo "Archive events look OK."
+
+	cdt_clear_no_retry
+	copytool_cleanup
+	copytool_monitor_cleanup
+}
+run_test 71 "Copytool logs JSON archive events to FIFO"
+
+test_72() {
+	# Bump progress interval for livelier events.
+	local interval=5
+
+	# test needs a new running copytool
+	copytool_cleanup
+	copytool_monitor_setup
+	HSMTOOL_UPDATE_INTERVAL=$interval \
+	HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
+	local test_file=$HSMTOOL_MONITOR_DIR/file
+
+	local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
+	cmd+="conv=fsync"
+	do_facet $SINGLEAGT "$cmd" ||
+		error "cannot create $test_file on $SINGLEAGT"
+	copy2archive $test_file $tdir/$tfile
+
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
+	import_file $tdir/$tfile $f
+	f=$DIR2/$tdir/$tfile
+	echo "Verifying released state: "
+	check_hsm_flags $f "0x0000000d"
+
+	local fid=$(path2fid $f)
+	$LFS hsm_restore $f
+	wait_request_state $fid RESTORE SUCCEED
+
+	local expected_fields="event_time data_fid source_fid"
+	expected_fields+=" total_bytes current_bytes"
+
+	local START_EVENT
+	local FINISH_EVENT
+	while read event; do
+		# Make sure we're not getting anything from previous events.
+		for field in $expected_fields; do
+			unset $field
+		done
+
+		local parsed=$(parse_json_event "$event")
+		if [ -z "$parsed" ]; then
+			error "Copytool sent malformed event: $event"
+		fi
+		eval $parsed
+
+		if [ $event_type == "RESTORE_START" ]; then
+			START_EVENT=$event
+			if [ $source_fid != $data_fid ]; then
+				error "source_fid should == data_fid at start"
+			fi
+			continue
+		elif [ $event_type == "RESTORE_FINISH" ]; then
+			FINISH_EVENT=$event
+			if [ $source_fid != $data_fid ]; then
+				error "source_fid should == data_fid at finish"
+			fi
+			continue
+		elif [ $event_type != "RESTORE_RUNNING" ]; then
+			continue
+		fi
+
+		# Do some simple checking of the progress update events.
+		for expected_field in $expected_fields; do
+			if [ -z ${!expected_field+x} ]; then
+				error "Missing $expected_field field in event"
+			fi
+		done
+
+		if [ $total_bytes -eq 0 ]; then
+			error "Expected total_bytes to be > 0"
+		fi
+
+		# When a restore starts out, the data fid is the same as the
+		# source fid. After the restore has gotten going, we learn
+		# the new data fid. Once the restore has finished, the source
+		# fid is set to the new data fid.
+		#
+		# We test this because some monitoring software may depend on
+		# this behavior. If it changes, then the consumers of these
+		# events may need to be modified.
+		if [ $source_fid == $data_fid ]; then
+			error "source_fid should != data_fid during restore"
+		fi
+	done < <(echo $"$(get_copytool_event_log)")
+
+	if [ -z "$START_EVENT" ]; then
+		error "Copytool failed to send restore start event to FIFO"
+	fi
+
+	if [ -z "$FINISH_EVENT" ]; then
+		error "Copytool failed to send restore finish event to FIFO"
+	fi
+
+	echo "Restore events look OK."
+
+	cdt_clear_no_retry
+	copytool_cleanup
+	copytool_monitor_cleanup
+
+	rm -rf $test_dir
+}
+run_test 72 "Copytool logs JSON restore events to FIFO"
+
+test_90() {
+	file_count=51 # Max number of files constrained by LNET message size
+	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	local f=$DIR/$tdir/$tfile
 	local FILELIST=/tmp/filelist.txt
 	local i=""
 
@@ -2339,7 +3818,14 @@ test_90() {
 		fid=$(copy_file /etc/hosts $f.$i)
 		echo $f.$i >> $FILELIST
 	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 do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
+		copytool_setup
+	else
+		local dai=$(get_hsm_param default_archive_id)
+		copytool_setup $SINGLEAGT $MOUNT $dai $TMP/$tdir
+	fi
 	# to be sure wait_all_done will not be mislead by previous tests
 	cdt_purge
 	wait_for_grace_delay
@@ -2426,7 +3912,10 @@ test_104() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	# if cdt is on, it can serve too quickly the request
 	cdt_disable
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
@@ -2471,24 +3960,20 @@ test_106() {
 	# test needs a running copytool
 	copytool_setup
 
-	local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
-		get_client_uuid $MOUNT | cut -d' ' -f2)
-	local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
-		grep $uuid)
+	local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
+
+	check_agent_registered $uuid
+
+	search_copytools || error "No copytool found"
+
 	copytool_cleanup
-	[[ ! -z "$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 $uuid)
+	uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
+	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"
 
@@ -2695,10 +4180,15 @@ test_200() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_cancel $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 103 1048576)
+	[ $? != 0 ] && skip "not enough free space" && return
+
 	# test with cdt on is made in test_221
 	cdt_disable
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	# wait archive to register at CDT
+	wait_request_state $fid ARCHIVE WAITING
 	$LFS hsm_cancel $f
 	cdt_enable
 	wait_request_state $fid ARCHIVE CANCELED
@@ -2721,6 +4211,8 @@ test_201() {
 	# test with cdt on is made in test_222
 	cdt_disable
 	$LFS hsm_restore $f
+	# wait restore to register at CDT
+	wait_request_state $fid RESTORE WAITING
 	$LFS hsm_cancel $f
 	cdt_enable
 	wait_request_state $fid RESTORE CANCELED
@@ -2736,21 +4228,50 @@ test_202() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	cdt_disable
+	$LFS hsm_remove $f
+	# wait remove to register at CDT
+	wait_request_state $fid REMOVE WAITING
+	$LFS hsm_cancel $f
+	cdt_enable
+	wait_request_state $fid REMOVE CANCELED
+
+	copytool_cleanup
+}
+run_test 202 "Register/Cancel remove"
+
+test_220() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	changelog_setup
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE SUCCEED
 
-	cdt_disable
-	$LFS hsm_remove $f
-	$LFS hsm_cancel $f
-	cdt_enable
-	wait_request_state $fid REMOVE CANCELED
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+	changelog_cleanup
+
+	local target=0x0
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
 
 	copytool_cleanup
 }
-run_test 202 "Register/Cancel remove"
+run_test 220 "Changelog for archive"
 
-test_220() {
+test_220a() {
 	# test needs a running copytool
 	copytool_setup
 
@@ -2761,18 +4282,28 @@ test_220() {
 
 	changelog_setup
 
+	# block copytool operations to allow for HSM request to be
+	# submitted and file be unlinked (CDT will find object removed)
+	copytool_suspend
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
-	wait_request_state $fid ARCHIVE SUCCEED
 
-	local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid ARCHIVE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
 	changelog_cleanup
 
-	local target=0x0
+	# HE_ARCHIVE|ENOENT
+	local target=0x2
 	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
 
 	copytool_cleanup
 }
-run_test 220 "Changelog for archive"
+run_test 220a "Changelog for failed archive"
 
 test_221() {
 	# test needs a running copytool
@@ -2781,7 +4312,9 @@ test_221() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_cancel $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 103 1048576)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
 
@@ -2791,7 +4324,7 @@ 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"
@@ -2816,7 +4349,7 @@ 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"
@@ -2842,7 +4375,7 @@ 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"
@@ -2851,6 +4384,68 @@ test_222b() {
 }
 run_test 222b "Changelog for implicit restore"
 
+test_222c() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	copy2archive /etc/passwd $tdir/$tfile
+
+	local f=$DIR/$tdir/$tfile
+	import_file $tdir/$tfile $f
+	local fid=$(path2fid $f)
+
+	changelog_setup
+
+	# block copytool operations to allow for HSM request to be
+	# submitted and file be unlinked (CDT will find object removed)
+	copytool_suspend
+
+	$LFS hsm_restore $f
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid RESTORE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+
+	# HE_RESTORE|ENOENT
+	local target=0x82
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
+
+	cleanup
+}
+run_test 222c "Changelog for failed explicit restore"
+
+test_222d() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	changelog_setup
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f
+
+	copytool_remove_backend $fid
+	md5sum $f
+
+	wait_request_state $fid RESTORE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+
+	# HE_RESTORE|ENOENT
+	local target=0x82
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
+
+	cleanup
+}
+run_test 222d "Changelog for failed implicit restore"
+
 test_223a() {
 	# test needs a running copytool
 	copytool_setup
@@ -2871,7 +4466,7 @@ 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 ]] ||
@@ -2888,7 +4483,9 @@ test_223b() {
 	mkdir -p $DIR/$tdir
 
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2900,7 +4497,7 @@ 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 ]] ||
@@ -2926,7 +4523,7 @@ 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 ]] ||
@@ -2936,6 +4533,43 @@ test_224() {
 }
 run_test 224 "Changelog for remove"
 
+test_224a() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	changelog_setup
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	copytool_remove_backend $fid
+
+	# block copytool operations to allow for HSM request to be
+	# submitted and file be unlinked (CDT will find object removed)
+	copytool_suspend
+
+	$LFS hsm_remove $f
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid REMOVE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
+
+	# HE_REMOVE|ENOENT
+	local target=0x202
+	[[ $flags == $target ]] ||
+		error "Changelog flag is $flags not $target"
+
+	cleanup
+}
+run_test 224a "Changelog for failed remove"
+
 test_225() {
 	# test needs a running copytool
 	copytool_setup
@@ -2948,7 +4582,9 @@ test_225() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 39 1000000)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2962,9 +4598,9 @@ 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 ]] ||
@@ -2996,7 +4632,7 @@ 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 ]] ||
@@ -3004,7 +4640,7 @@ 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 ]] ||
@@ -3024,7 +4660,7 @@ 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 ]] ||
@@ -3035,7 +4671,7 @@ 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 ]] ||
@@ -3075,10 +4711,9 @@ 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"
+	local fid=$(make_small_sync $DIR/$tfile)
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
-	wait_request_state $(path2fid $DIR/$tfile) ARCHIVE SUCCEED
+	wait_request_state $fid ARCHIVE SUCCEED
 
 	$LFS hsm_release $DIR/$tfile
 	check_hsm_flags $DIR/$tfile "0x0000000d"
@@ -3095,13 +4730,15 @@ test_228() {
 	$LFS hsm_release $DIR/$tfile
 	check_hsm_flags $DIR/$tfile "0x0000000d"
 
-	mkdir $DIR/$tdir
+	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"
@@ -3150,7 +4787,9 @@ test_251() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_cancel $f)
+	local fid
+	fid=$(make_custom_file_for_progress $f 103 1048576)
+	[ $? != 0 ] && skip "not enough free space" && return
 
 	cdt_disable
 	# to have a short test
@@ -3162,6 +4801,10 @@ test_251() {
 	set_hsm_param loop_period 2
 	cdt_enable
 
+	# clear locks to avoid extra delay caused by flush/cancel
+	# and thus prevent early copytool death to timeout.
+	cancel_lru_locks osc
+
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
 	wait_request_state $fid ARCHIVE STARTED
 	sleep 5
@@ -3174,6 +4817,43 @@ test_251() {
 }
 run_test 251 "Coordinator request timeout"
 
+test_252() {
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(make_custom_file_for_progress $f 103 1048576)
+
+	cdt_disable
+	# to have a short test
+	local old_to=$(get_hsm_param active_request_timeout)
+	set_hsm_param active_request_timeout 20
+	# 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)
+	set_hsm_param loop_period 2
+	cdt_enable
+
+	# clear locks to avoid extra delay caused by flush/cancel
+	# and thus prevent early copytool death to timeout.
+	cancel_lru_locks osc
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE STARTED
+	rm -f $f
+
+	# wait but less than active_request_timeout+grace_delay
+	sleep 25
+	wait_request_state $fid ARCHIVE CANCELED
+
+	set_hsm_param active_request_timeout $old_to
+	set_hsm_param loop_period $old_loop
+
+	copytool_cleanup
+}
+run_test 252 "Timeout'ed running archive of a removed file should be canceled"
+
 test_300() {
 	# the only way to test ondisk conf is to restart MDS ...
 	echo "Stop coordinator and remove coordinator state at mount"
@@ -3224,7 +4904,11 @@ test_302() {
 	cdt_shutdown
 
 	set_hsm_param default_archive_id $new -P
-	fail $SINGLEMDS
+
+	local mdtno
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		fail mds${mdtno}
+	done
 
 	# check cdt is on
 	cdt_check_state enabled
@@ -3238,6 +4922,318 @@ test_302() {
 }
 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_402a() {
+	# make sure there is no running copytool
+	copytool_cleanup
+
+	# deactivate all mdc on agent1
+	mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
+
+	HSMTOOL_NOERROR=true 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 $SINGLEAGT "$FSNAME-MDT000." "activate"
+}
+run_test 402a "Copytool start fails if all MDTs are inactive"
+
+test_402b() {
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	touch $f || error "touch $f failed"
+	local fid=$(path2fid $f)
+
+#define OBD_FAIL_MDS_HSM_CT_REGISTER_NET	0x14d
+	do_facet $SINGLEAGT lctl set_param fail_loc=0x14d
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+
+	# give time for CDT to send request and to keep it for retry
+	wait_for_loop_period
+
+	wait_request_state $fid ARCHIVE WAITING
+
+	do_facet $SINGLEAGT lctl set_param fail_loc=0
+
+	# request should succeed now
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	copytool_cleanup
+}
+run_test 402b "CDT must retry request upon slow start of CT"
+
+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)
+
+	# deactivate all mdc for MDT0001
+	mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
+
+	copytool_setup
+	local uuid=$(get_agent_uuid $agent)
+	# 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 $SINGLEAGT "$FSNAME-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 $SINGLEAGT "$FSNAME-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 $SINGLEAGT "$FSNAME-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"
+
+test_405() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local striped_dir=$DIR/$tdir/striped_dir
+
+	# create striped dir on all of MDTs
+	$LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
+
+	local fid1=$(make_small_sync $striped_dir/${tfile}_0)
+	local fid2=$(make_small_sync $striped_dir/${tfile}_1)
+	local fid3=$(make_small_sync $striped_dir/${tfile}_2)
+	local fid4=$(make_small_sync $striped_dir/${tfile}_3)
+
+	local idx1=$($LFS getstripe -M $striped_dir/${tfile}_0)
+	local idx2=$($LFS getstripe -M $striped_dir/${tfile}_1)
+	local idx3=$($LFS getstripe -M $striped_dir/${tfile}_2)
+	local idx4=$($LFS getstripe -M $striped_dir/${tfile}_3)
+
+	# check that compound requests are shunt to the rights MDTs
+	$LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1  \
+			 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
+		error "lfs hsm_archive"
+
+	wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
+		echo "archive successful on $fid1"
+	wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
+		echo "archive successful on $fid2"
+	wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
+		echo "archive successful on $fid3"
+	wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
+		echo "archive successful on $fid4"
+
+	$LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
+	$LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
+	$LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
+	$LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
+
+	cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
+	cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
+	cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
+	cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
+
+	copytool_cleanup
+}
+run_test 405 "archive and release under striped directory"
+
+test_406() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "need MDS version at least 2.7.64" && return 0
+
+	local fid
+	local mdt_index
+
+	copytool_setup
+	mkdir -p $DIR/$tdir
+	fid=$(make_small $DIR/$tdir/$tfile)
+	echo "old fid $fid"
+
+	$LFS hsm_archive $DIR/$tdir/$tfile
+	wait_request_state "$fid" ARCHIVE SUCCEED
+	$LFS hsm_release $DIR/$tdir/$tfile
+
+	# Should migrate $tdir but not $tfile.
+	$LFS mv -M1 $DIR/$tdir &&
+		error "migrating HSM an archived file should fail"
+
+	$LFS hsm_restore $DIR/$tdir/$tfile
+	wait_request_state "$fid" RESTORE SUCCEED
+
+	$LFS hsm_remove $DIR/$tdir/$tfile
+	wait_request_state "$fid" REMOVE SUCCEED
+
+	cat $DIR/$tdir/$tfile > /dev/null ||
+		error "cannot read $DIR/$tdir/$tfile"
+
+	$LFS mv -M1 $DIR/$tdir ||
+		error "cannot complete migration after HSM remove"
+
+	mdt_index=$($LFS getstripe -M $DIR/$tdir)
+	if ((mdt_index != 1)); then
+		error "expected MDT index 1, got $mdt_index"
+	fi
+
+	# Refresh fid after migration.
+	fid=$(path2fid $DIR/$tdir/$tfile)
+	echo "new fid $fid"
+
+	$LFS hsm_archive $DIR/$tdir/$tfile
+	wait_request_state "$fid" ARCHIVE SUCCEED 1
+
+	lctl set_param debug=+trace
+	$LFS hsm_release $DIR/$tdir/$tfile ||
+		error "cannot release $DIR/$tdir/$tfile"
+
+	$LFS hsm_restore $DIR/$tdir/$tfile
+	wait_request_state "$fid" RESTORE SUCCEED 1
+
+	cat $DIR/$tdir/$tfile > /dev/null ||
+		error "cannot read $DIR/$tdir/$tfile"
+
+	copytool_cleanup
+}
+run_test 406 "attempting to migrate HSM archived files is safe"
+
+test_500()
+{
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&
+		skip "HSM migrate is not supported" && return
+
+	# Stop the existing copytool
+	copytool_cleanup
+
+	test_mkdir -p $DIR/$tdir
+	llapi_hsm_test -d $DIR/$tdir || error "One llapi HSM test failed"
+}
+run_test 500 "various LLAPI HSM tests"
+
 copytool_cleanup
 
 complete $SECONDS