X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=8b89a1852ffed17dc83464725bbe9729d766cc96;hp=37a66ce30d69191ee2ce836a51beb190e18ea9fe;hb=0ec1dbbea450375af070838ce6487e14c6595c83;hpb=57118830eb55ab43b4b6fc096ad40b2bd83c6de3

diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh
index 37a66ce..8b89a18 100644
--- a/lustre/tests/sanity-hsm.sh
+++ b/lustre/tests/sanity-hsm.sh
@@ -11,12 +11,11 @@ SRCDIR=$(dirname $0)
 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test:
+# bug number for skipped test:    3815     3939
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36 40"
+# bug number for skipped test:4178         4176
+ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 221 223b 31a"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-# skip test cases failed before landing - Jinshan
-
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 31a 34 35 36"
-ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 201 221 223a 223b 225"
 
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 
@@ -164,10 +163,34 @@ copytool_setup() {
 copytool_cleanup() {
 	trap - EXIT
 	local agents=${1:-$(facet_active_host $SINGLEAGT)}
+	local mdtno
+	local idx
+	local oldstate
+	local mdt_hsmctrl
 
 	do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
 	sleep 1
 	echo "Copytool is stopped on $agents"
+
+	# clean all CDTs orphans requests from previous tests
+	# that would otherwise need to timeout to clear.
+	for mdtno in $(seq 1 $MDSCOUNT); do
+		idx=$(($mdtno - 1))
+		mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
+		oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
+				   "$MDT_HSMCTRL")
+		# skip already stop[ed,ing] CDTs
+		echo $oldstate | grep stop || continue
+
+		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
+		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+			"stopped" 20 ||
+			error "mds${mdtno} cdt state is not stopped"
+		do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
+		wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+			"$oldstate" 20 ||
+			error "mds${mdtno} cdt state is not $oldstate"
+	done
 }
 
 copytool_suspend() {
@@ -179,7 +202,7 @@ copytool_suspend() {
 
 copytool_remove_backend() {
 	local fid=$1
-	local be=$(find $HSM_ARCHIVE -name $fid)
+	local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
 	echo "Remove from backend: $fid = $be"
 	do_facet $SINGLEAGT rm -f $be
 }
@@ -401,7 +424,8 @@ make_small() {
 }
 
 cleanup_large_files() {
-	local ratio=$(df $MOUNT |awk '{print $5}' |sed 's/%//g' |grep -v Use)
+	local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
+		      sed 's/%//g')
 	[ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
 }
 
@@ -466,7 +490,7 @@ wait_request_state() {
 	local request=$2
 	local state=$3
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
 	cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 
 	wait_result $SINGLEMDS "$cmd" $state 100 ||
@@ -477,7 +501,7 @@ get_request_state() {
 	local fid=$1
 	local request=$2
 
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
 		"awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
 }
 
@@ -485,14 +509,14 @@ get_request_count() {
 	local fid=$1
 	local request=$2
 
-	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+	do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
 		"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
 }
 
 wait_all_done() {
 	local timeout=$1
 
-	local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+	local cmd="$LCTL get_param -n $HSM_PARAM.actions"
 	cmd+=" | egrep 'WAITING|STARTED'"
 
 	wait_result $SINGLEMDS "$cmd" "" $timeout ||
@@ -1082,8 +1106,8 @@ test_13() {
 			CURR_FILE="$CURR_DIR/$tfile.$f"
 			# write file-specific data
 			do_facet $SINGLEAGT \
-				echo "d=$d, f=$f, dir=$CURR_DIR, "\
-				     "file=$CURR_FILE" > $CURR_FILE
+				"echo d=$d, f=$f, dir=$CURR_DIR, "\
+					"file=$CURR_FILE > $CURR_FILE"
 		done
 	done
 	# import to Lustre
@@ -1827,6 +1851,41 @@ test_30b() {
 }
 run_test 30b "Restore at exec (release case)"
 
+test_30c() {
+	needclients 2 || return 0
+
+	# test needs a running copytool
+	copytool_setup
+
+	mkdir -p $DIR/$tdir
+	local f=$DIR/$tdir/SLEEP
+	local fid=$(copy_file /bin/sleep $f)
+	chmod 755 $f
+	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+	wait_request_state $fid ARCHIVE SUCCEED
+	$LFS hsm_release $f
+	check_hsm_flags $f "0x0000000d"
+	# set no retry action mode
+	cdt_set_no_retry
+	do_node $CLIENT2 "$f 10" &
+	local pid=$!
+	sleep 3
+	echo 'Hi!' > $f
+	[[ $? == 0 ]] && error "Update during exec of released file must fail"
+	wait $pid
+	[[ $? == 0 ]] || error "Execution failed during run"
+	cmp /bin/sleep $f
+	[[ $? == 0 ]] || error "Binary overwritten during exec"
+
+	# cleanup
+	# remove no try action mode
+	cdt_clear_no_retry
+	check_hsm_flags $f "0x00000009"
+
+	copytool_cleanup
+}
+run_test 30c "Update during exec of released file must fail"
+
 restore_and_check_size() {
 	local f=$1
 	local fid=$2
@@ -2382,7 +2441,7 @@ double_verify_reset_hsm_param() {
 test_100() {
 	double_verify_reset_hsm_param loop_period
 	double_verify_reset_hsm_param grace_delay
-	double_verify_reset_hsm_param request_timeout
+	double_verify_reset_hsm_param active_request_timeout
 	double_verify_reset_hsm_param max_requests
 	double_verify_reset_hsm_param default_archive_id
 }
@@ -2412,7 +2471,7 @@ test_103() {
 
 	echo "Current requests"
 	local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
 
 	[[ -z "$res" ]] || error "Some request have not been canceled"
@@ -2434,7 +2493,7 @@ test_104() {
 	cdt_disable
 	$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
 	local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep $fid | cut -f16 -d=")
 	cdt_enable
 
@@ -2455,12 +2514,12 @@ test_105() {
 		$LFS hsm_archive $DIR/$tdir/$i
 	done
 	local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_restart
 	cdt_disable
 	local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 	cdt_enable
 	cdt_purge
@@ -3133,12 +3192,12 @@ test_250() {
 	while [[ $cnt != 0 || $wt != 0 ]]; do
 		sleep 1
 		cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep STARTED | grep -v CANCEL | wc -l")
 		[[ $cnt -le $maxrequest ]] ||
 			error "$cnt > $maxrequest too many started requests"
 		wt=$(do_facet $SINGLEMDS "$LCTL get_param\
-			$HSM_PARAM.agent_actions |\
+			$HSM_PARAM.actions |\
 			grep WAITING | wc -l")
 		echo "max=$maxrequest started=$cnt waiting=$wt"
 	done
@@ -3157,8 +3216,8 @@ test_251() {
 
 	cdt_disable
 	# to have a short test
-	local old_to=$(get_hsm_param request_timeout)
-	set_hsm_param request_timeout 4
+	local old_to=$(get_hsm_param active_request_timeout)
+	set_hsm_param active_request_timeout 4
 	# to be sure the cdt will wake up frequently so
 	# it will be able to cancel the "old" request
 	local old_loop=$(get_hsm_param loop_period)
@@ -3170,7 +3229,7 @@ test_251() {
 	sleep 5
 	wait_request_state $fid ARCHIVE CANCELED
 
-	set_hsm_param request_timeout $old_to
+	set_hsm_param active_request_timeout $old_to
 	set_hsm_param loop_period $old_loop
 
 	copytool_cleanup