X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=8949a08c2a53c1eda67b7cd3c575c64fe82fecae;hp=721cb2c17f69aea2b751782687e478783a453462;hb=37a5157b84bce367e31743cb8648a15618492531;hpb=73bca6c1f4923cdf673fa11486aec04ec3576051

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index 721cb2c..8949a08 100755
--- 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:    LU-3815
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
+# bug number for skipped test:
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
@@ -24,6 +24,7 @@ init_logging
 
 MULTIOP=${MULTIOP:-multiop}
 OPENFILE=${OPENFILE:-openfile}
+MMAP_CAT=${MMAP_CAT:-mmap_cat}
 MOUNT_2=${MOUNT_2:-"yes"}
 FAIL_ON_ERROR=false
 
@@ -102,7 +103,16 @@ init_agt_vars() {
 	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
 
 	# The test only support up to 10 MDTs
@@ -144,15 +154,43 @@ get_mdt_devices() {
 }
 
 search_copytools() {
-	local agents=${1:-$(facet_active_host $SINGLEAGT)}
-	do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
+	local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+	do_nodesv $hosts "pgrep -x $HSMTOOL_BASE"
 }
 
-search_and_kill_copytool() {
-	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+kill_copytools() {
+	local hosts=${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
 
-	echo "Killing existing copytools on $agents"
-	do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
+	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() {
@@ -218,24 +256,23 @@ copytool_setup() {
 	local lustre_mntpnt=${2:-${MOUNT2:-$MOUNT}}
 	local arc_id=$3
 	local hsm_root=${4:-$(copytool_device $facet)}
-	local agent=$(facet_active_host $facet)
 
-	if [[ -z "$arc_id" ]] &&
-		do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
-			echo "Only wakeup running copytool $facet on $agent"
-			return 0
-	fi
+	[ -z "${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"
@@ -272,54 +309,61 @@ get_copytool_event_log() {
 
 copytool_cleanup() {
 	trap - EXIT
-	local facet=$SINGLEAGT
-	local agents=${1:-$(facet_active_host $facet)}
-	local mdtno
-	local idx
-	local oldstate
-	local mdt_hsmctrl
-	local hsm_root=$(copytool_device $facet)
-	local end_wait=$(( SECONDS + TIMEOUT ))
-
-	do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
-
-	while (( SECONDS < end_wait )); do
-		sleep 2
-		do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
-		if [ $? -ne 0 ]; then
-			echo "Copytool is stopped on $agents"
-			break
-		fi
-		echo "Copytool still running on $agents"
+	local agt_facet=$SINGLEAGT
+	local agt_hosts=${1:-$(facet_active_host $agt_facet)}
+	local hsm_root=$(copytool_device $agt_facet)
+
+	[ -z "${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
-	if do_nodesv $agents "pgrep -x $HSMTOOL_BASE"; then
-		error "Copytool failed to stop in ${TIMEOUT}s ..."
-	else
-		echo "Copytool has stopped in " \
-		     "$((TIMEOUT - (end_wait - SECONDS)))s."
-	fi
 
-	# clean all CDTs orphans requests from previous tests
-	# that would otherwise need to timeout to clear.
-	for mdtno in $(seq 1 $MDSCOUNT); do
-		idx=$(($mdtno - 1))
-		mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
-		oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
-				   "$mdt_hsmctrl")
-		# skip already stop[ed,ing] CDTs
-		echo $oldstate | grep stop && continue
-
-		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
-		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
-			"stopped" 20 ||
-			error "mds${mdtno} cdt state is not stopped"
-		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
-		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
-			"$oldstate" 20 ||
-			error "mds${mdtno} cdt state is not $oldstate"
+	for ((i = 0; i < MDSCOUNT; i++)); do
+		# Only check and restore CDTs that we stopped in the first loop.
+		[[ "${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 $facet "df $hsm_root" >/dev/null 2>&1 ; then
-		do_facet $facet "rm -rf $hsm_root/*"
+
+	if do_facet $agt_facet "df $hsm_root" >/dev/null 2>&1 ; then
+		do_facet $agt_facet "rm -rf $hsm_root/$HSMTMP/*"
 	fi
 }
 
@@ -330,6 +374,13 @@ 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=$1
 	local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
@@ -634,66 +685,18 @@ check_enough_free_space() {
 	return 0
 }
 
-make_large_for_striping() {
-	local file2=${1/$DIR/$DIR2}
-	local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
-
-	cleanup_large_files
-
-	check_enough_free_space 5 $sz
-	[ $? != 0 ] && return $?
-
-	dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync ||
-		file_creation_failure dd $file2 $?
-
-	path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_progress() {
-	local file2=${1/$DIR/$DIR2}
-
-	cleanup_large_files
-
-	check_enough_free_space 39 1000000
-	[ $? != 0 ] && return $?
-
-	# big file is large enough, so copy time is > 30s
-	# so copytool make 1 progress
-	# size is not a multiple of 1M to avoid stripe
-	# aligment
-	dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync ||
-		file_creation_failure dd $file2 $?
-
-	path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_progress_aligned() {
+make_custom_file_for_progress() {
 	local file2=${1/$DIR/$DIR2}
+	local fsize=${2:-"39"}
+	local blksz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
+	blksz=${3:-$blksz}
 
-	cleanup_large_files
-
-	check_enough_free_space 33 1048576
-	[ $? != 0 ] && return $?
-
-	# big file is large enough, so copy time is > 30s
-	# so copytool make 1 progress
-	# size is a multiple of 1M to have stripe
-	# aligment
-	dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync ||
-		file_creation_failure dd $file2 $?
-	path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_cancel() {
-	local file2=${1/$DIR/$DIR2}
+	[[ $fsize -gt  0 ]] || error "Invalid file size"
+	[[ $blksz -gt 0 ]] || error "Invalid stripe size"
 
 	cleanup_large_files
-
-	check_enough_free_space 103 1048576
-	[ $? != 0 ] && return $?
-
-	# Copy timeout is 100s. 105MB => 105s
-	dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
+	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 $1"
 }
@@ -735,6 +738,15 @@ get_request_count() {
 		"awk -vn=0 '/'$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
@@ -752,6 +764,11 @@ wait_for_grace_delay() {
 	sleep $val
 }
 
+wait_for_loop_period() {
+	local val=$(get_hsm_param loop_period)
+	sleep $val
+}
+
 parse_json_event() {
 	local raw_event=$1
 
@@ -764,14 +781,74 @@ parse_json_event() {
 	echo $raw_event | python -c "$json_parser"
 }
 
-# populate MDT device array
-get_mdt_devices
+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
@@ -780,9 +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
 
@@ -824,6 +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
@@ -935,13 +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
 
@@ -1175,7 +1308,7 @@ test_12c() {
 	local f=$DIR/$tdir/$tfile
 	$LFS setstripe -c 2 $f
 	local fid
-	fid=$(make_large_for_striping $f)
+	fid=$(make_custom_file_for_progress $f 5)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	local FILE_CRC=$(md5sum $f)
@@ -1431,6 +1564,8 @@ cleanup_test_12q() {
 }
 
 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"
@@ -1452,9 +1587,8 @@ test_12q() {
 	$LFS hsm_release $f || error "could not release file"
 	check_hsm_flags $f "0x0000000d"
 
-	search_and_kill_copytool
-	sleep 5
-	search_copytools && error "Copytool should have stopped"
+	kill_copytools
+	wait_copytools || error "copytool failed to stop"
 
 	cat $f > /dev/null &
 
@@ -1639,9 +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
@@ -1873,8 +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
 
@@ -2001,7 +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"
 
@@ -2015,7 +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"
 
@@ -2029,7 +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"
 
@@ -2045,6 +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() {
@@ -2053,16 +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'"
@@ -2071,15 +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"
@@ -2192,7 +2334,7 @@ test_26() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2207,61 +2349,277 @@ test_26() {
 }
 run_test 26 "Remove the archive of a valid file"
 
-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"
-
+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
 }
-run_test 27a "Remove the archive of an imported file (Operation not permitted)"
 
-test_27b() {
+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
 	local f=$DIR/$tdir/$tfile
-	local fid
-	fid=$(make_large_for_progress $f)
-	[ $? != 0 ] && skip "not enough free space" && return
+	local fid=$(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 $f
+	local f2=$DIR/$tdir/${tfile}_2
+	local fid2=$(copy_file /etc/passwd $f2)
 
-	[[ $? != 0 ]] || error "Remove of a released file should fail"
+	$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 27b "Remove the archive of a relased file (Operation not permitted)"
 
-test_28() {
+test_26b() {
+
 	# test needs a running copytool
 	copytool_setup
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid
-	fid=$(make_large_for_progress $f)
-	[ $? != 0 ] && skip "not enough free space" && return
+	local fid=$(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
+	trap cleanup_test_26b EXIT
+
+	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
+}
+
+test_26c() {
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/passwd $f)
+
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	local f2=$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
+
+	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
+}
+
+test_26d() {
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/$tfile
+	local fid=$(copy_file /etc/motd $f 1)
+
+	$LFS hsm_archive $f || error "could not archive file"
+	wait_request_state $fid ARCHIVE SUCCEED
+
+	trap cleanup_test_26d EXIT
+
+	# set a long grace_delay vs short loop_period
+	local orig_loop_period=$(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 $f O_c || error "multiop failed"
+	local MULTIPID=$!
+
+	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
 
@@ -2343,6 +2701,78 @@ test_29c() {
 }
 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 $(agts_nodes))
+
+}
+run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
+	     "request to be sent once for each registered archive_id"
+
 test_30a() {
 	# restore at exec cannot work on agent node (because of Linux kernel
 	# protection of executables)
@@ -2460,19 +2890,18 @@ restore_and_check_size() {
 	while [[ "$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 [[ $cpt -lt 10 ]]; then
+	if [[ "$st" = "0x00000009" ]]; then
 		echo " "done
 	else
 		echo " restore is too long"
@@ -2511,7 +2940,7 @@ test_31b() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2535,7 +2964,7 @@ test_31c() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress_aligned $f)
+	fid=$(make_custom_file_for_progress $f 33 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2559,7 +2988,7 @@ test_33() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2626,7 +3055,7 @@ test_34() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2662,7 +3091,7 @@ test_35() {
 	local f=$DIR/$tdir/$tfile
 	local f1=$DIR/$tdir/$tfile-1
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	local fid1=$(copy_file /etc/passwd $f1)
@@ -2701,7 +3130,7 @@ test_36() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -2745,6 +3174,9 @@ test_37() {
 	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"
 
@@ -2766,6 +3198,12 @@ multi_archive() {
 	echo "$count archive requests submitted"
 }
 
+cleanup_test_40() {
+	trap 0
+	set_hsm_param max_requests $max_requests
+	copytool_cleanup
+}
+
 test_40() {
 	local stream_count=4
 	local file_count=100
@@ -2774,6 +3212,17 @@ test_40() {
 	local i=""
 	local p=""
 	local fid=""
+	local max_requests=$(get_hsm_param max_requests)
+
+	# Increase the number of HSM request that can be performed in
+	# parallel. With the coordinator running once per second, this
+	# also limits the number of requests per seconds that can be
+	# performed, so we pick a decent number. But we also need to keep
+	# that number low because the copytool has no rate limit and will
+	# fail some requests if if gets too many at once.
+	set_hsm_param max_requests 300
+
+	trap cleanup_test_40 EXIT
 
 	for i in $(seq 1 $file_count); do
 		for p in $(seq 1 $stream_count); do
@@ -2800,7 +3249,8 @@ test_40() {
 	wait ${pids[*]}
 	echo OK
 	wait_all_done 100
-	copytool_cleanup
+
+	cleanup_test_40
 }
 run_test 40 "Parallel archive requests"
 
@@ -2865,7 +3315,7 @@ test_54() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid=$(make_custom_file_for_progress $f 39 1000000)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -2893,7 +3343,7 @@ test_55() {
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
-	local fid=$(make_large_for_progress $f)
+	local fid=$(make_custom_file_for_progress $f 39 1000000)
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
 		error "could not archive file"
@@ -2922,7 +3372,7 @@ test_56() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
@@ -3032,6 +3482,23 @@ test_58() {
 }
 run_test 58 "Truncate a released file will trigger restore"
 
+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
@@ -3046,7 +3513,7 @@ test_60() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 10)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	local mdtidx=0
@@ -3097,15 +3564,36 @@ test_60() {
 	local elapsed=$((finish_at - start_at))
 
 	# Ensure that the progress update occurred within the expected window.
-	if [ $elapsed -lt $interval ]; then
+	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
@@ -3165,7 +3653,7 @@ test_71() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
@@ -3437,34 +3925,45 @@ run_test 103 "Purge all requests"
 DATA=CEA
 DATAHEX='[434541]'
 test_104() {
-	# test needs a running copytool
-	copytool_setup
 
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(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 $f
 	local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
 			$HSM_PARAM.actions |\
 			grep $fid | cut -f16 -d=")
-	cdt_enable
 
 	[[ "$data1" == "$DATAHEX" ]] ||
 		error "Data field in records is ($data1) and not ($DATAHEX)"
 
+	# archive the file
+	copytool_setup
+
+	wait_request_state $fid ARCHIVE SUCCEED
+
 	copytool_cleanup
 }
 run_test 104 "Copy tool data field"
 
+cleanup_test_105() {
+	trap 0
+	set_hsm_param max_requests $max_requests
+	copytool_cleanup
+}
+
 test_105() {
+	local max_requests=$(get_hsm_param max_requests)
 	mkdir -p $DIR/$tdir
 	local i=""
 
+	set_hsm_param max_requests 300
+
+	trap cleanup_test_105 EXIT
+
 	cdt_disable
 	for i in $(seq -w 1 10); do
 		cp /etc/passwd $DIR/$tdir/$i
@@ -3474,6 +3973,7 @@ test_105() {
 			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_restart
+
 	cdt_disable
 	local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
 			$HSM_PARAM.actions |\
@@ -3483,68 +3983,10 @@ test_105() {
 	[[ "$reqcnt1" == "$reqcnt2" ]] ||
 		error "Requests count after shutdown $reqcnt2 != "\
 		      "before shutdown $reqcnt1"
-}
-run_test 105 "Restart of coordinator"
-
-get_agent_by_uuid_mdt() {
-	local uuid=$1
-	local mdtidx=$2
-	local mds=mds$(($mdtidx + 1))
-	do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
-		 grep $uuid"
-}
 
-check_agent_registered_by_mdt() {
-	local uuid=$1
-	local mdtidx=$2
-	local mds=mds$(($mdtidx + 1))
-	local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
-	if [[ ! -z "$agent" ]]; then
-		echo "found agent $agent on $mds"
-	else
-		error "uuid $uuid not found in agent list on $mds"
-	fi
-}
-
-check_agent_unregistered_by_mdt() {
-	local uuid=$1
-	local mdtidx=$2
-	local mds=mds$(($mdtidx + 1))
-	local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
-	if [[ -z "$agent" ]]; then
-		echo "uuid not found in agent list on $mds"
-	else
-		error "uuid found in agent list on $mds: $agent"
-	fi
-}
-
-check_agent_registered() {
-	local uuid=$1
-	local mdsno
-	for mdsno in $(seq 1 $MDSCOUNT); do
-		check_agent_registered_by_mdt $uuid $((mdsno - 1))
-	done
-}
-
-check_agent_unregistered() {
-	local uuid=$1
-	local mdsno
-	for mdsno in $(seq 1 $MDSCOUNT); do
-		check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
-	done
-}
-
-get_agent_uuid() {
-	local agent=${1:-$(facet_active_host $SINGLEAGT)}
-
-	# Lustre mount-point is mandatory and last parameter on
-	# copytool cmd-line.
-	local mntpnt=$(do_rpc_nodes $agent pgrep -fl $HSMTOOL_BASE |
-		       grep -v pgrep | 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
+	cleanup_test_105
 }
+run_test 105 "Restart of coordinator"
 
 test_106() {
 	# test needs a running copytool
@@ -3771,12 +4213,14 @@ test_200() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_cancel $f)
+	fid=$(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
@@ -3799,6 +4243,8 @@ 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
@@ -3815,7 +4261,7 @@ test_202() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
@@ -3823,6 +4269,8 @@ test_202() {
 
 	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
@@ -3855,6 +4303,43 @@ test_220() {
 }
 run_test 220 "Changelog for archive"
 
+test_220a() {
+	# 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
+
+	# 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 to reach CT
+	wait_request_state $fid ARCHIVE STARTED
+
+	rm -f $f
+
+	copytool_continue
+
+	wait_request_state $fid ARCHIVE FAILED
+
+	local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
+	changelog_cleanup
+
+	# HE_ARCHIVE|ENOENT
+	local target=0x2
+	[[ $flags == $target ]] || error "Changelog flag is $flags not $target"
+
+	copytool_cleanup
+}
+run_test 220a "Changelog for failed archive"
+
 test_221() {
 	# test needs a running copytool
 	copytool_setup
@@ -3863,7 +4348,7 @@ test_221() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_cancel $f)
+	fid=$(make_custom_file_for_progress $f 103 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
@@ -3934,6 +4419,72 @@ 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
+
+	# wait request to reach CT
+	wait_request_state $fid RESTORE STARTED
+
+	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
@@ -3972,7 +4523,7 @@ test_223b() {
 
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
@@ -4021,6 +4572,47 @@ 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
+
+	# wait for request to reach CT
+	wait_request_state $fid REMOVE STARTED
+
+	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
@@ -4034,7 +4626,7 @@ test_225() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_progress $f)
+	fid=$(make_custom_file_for_progress $f 39 1000000)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	changelog_setup
@@ -4239,7 +4831,7 @@ test_251() {
 	mkdir -p $DIR/$tdir
 	local f=$DIR/$tdir/$tfile
 	local fid
-	fid=$(make_large_for_cancel $f)
+	fid=$(make_custom_file_for_progress $f 103 1048576)
 	[ $? != 0 ] && skip "not enough free space" && return
 
 	cdt_disable
@@ -4268,6 +4860,43 @@ 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"
@@ -4416,7 +5045,7 @@ mdc_change_state() # facet, MDT_pattern, activate|deactivate
 	done
 }
 
-test_402() {
+test_402a() {
 	# make sure there is no running copytool
 	copytool_cleanup
 
@@ -4433,7 +5062,34 @@ test_402() {
 	# reactivate MDCs
 	mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
 }
-run_test 402 "Copytool start fails if all MDTs are inactive"
+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
@@ -4548,6 +5204,105 @@ test_405() {
 }
 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_407() {
+	needclients 2 || return 0
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+
+	local f=$DIR/$tdir/$tfile
+	local f2=$DIR2/$tdir/$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
+
+#define OBD_FAIL_MDS_HSM_CDT_DELAY      0x164
+	do_facet $SINGLEMDS $LCTL set_param fail_val=5 fail_loc=0x164
+
+	md5sum $f &
+	# 1st request holds layout lock while appropriate
+	# RESTORE record is still not added to llog
+	md5sum $f2 &
+	sleep 2
+
+	# after umount hsm_actions->O/x/x log shouldn't have
+	# double RESTORE records like below
+	#[0x200000401:0x1:0x0]...0x58d03a0d/0x58d03a0c action=RESTORE...WAITING
+	#[0x200000401:0x1:0x0]...0x58d03a0c/0x58d03a0d action=RESTORE...WAITING
+	sleep 30 &&
+		do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"&
+	fail $SINGLEMDS
+
+	wait_request_state $fid RESTORE SUCCEED
+	copytool_cleanup
+}
+run_test 407 "Check for double RESTORE records in llog"
+
 test_500()
 {
 	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&