X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-lfsck.sh;h=4456975d2ab8595ac1cab65433bcb0bbf128f1f2;hp=79e130ee767f76eec723f930df8fdd11bc8312b3;hb=077570483e75e0610fd45149b926097547c434b8;hpb=62f64a1077468e01a41df35212b5e5b1fe6b31d5;ds=sidebyside

diff --git a/lustre/tests/sanity-lfsck.sh b/lustre/tests/sanity-lfsck.sh
index 79e130e..4456975 100644
--- a/lustre/tests/sanity-lfsck.sh
+++ b/lustre/tests/sanity-lfsck.sh
@@ -40,13 +40,16 @@ SAVED_OSTCOUNT=${OSTCOUNT}
 # use small MDS + OST size to speed formatting time
 # do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
 MDSSIZE=100000
+[ $(facet_fstype $SINGLEMDS) == zfs ] && MDSSIZE=300000
 OSTSIZE=100000
+[ $(facet_fstype ost1) == zfs ] && OSTSIZE=300000
+
 # no need too many OSTs, to reduce the format/start/stop overhead
+cleanupall
 [ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
 
 # build up a clean test environment.
-formatall
-setupall
+REFORMAT="yes" check_and_setup_lustre
 
 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
 	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
@@ -213,9 +216,6 @@ test_0() {
 run_test 0 "Control LFSCK manually"
 
 test_1a() {
-	[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
-		skip "OI Scrub not implemented for ZFS" && return
-
 	lfsck_prep 1 1
 
 	#define OBD_FAIL_FID_INDIR	0x1501
@@ -300,6 +300,45 @@ test_1b()
 }
 run_test 1b "LFSCK can find out and repair the missing FID-in-LMA"
 
+test_1c() {
+	lfsck_prep 1 1
+
+	#define OBD_FAIL_FID_IGIF	0x1504
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1504
+	touch $DIR/$tdir/dummy
+
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+	umount_client $MOUNT
+	$START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
+	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+		mdd.${MDT_DEV}.lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+		$SHOW_NAMESPACE
+		error "(4) unexpected status"
+	}
+
+	local repaired=$($SHOW_NAMESPACE |
+			 awk '/^dirent_repaired/ { print $2 }')
+	# for interop with old server
+	[ -z "$repaired" ] &&
+		repaired=$($SHOW_NAMESPACE |
+			 awk '/^updated_phase1/ { print $2 }')
+
+	[ $repaired -eq 1 ] ||
+		error "(5) Fail to repair lost FID-in-dirent: $repaired"
+
+	run_e2fsck_on_mdt0
+
+	mount_client $MOUNT || error "(6) Fail to start client!"
+
+	#define OBD_FAIL_FID_LOOKUP	0x1505
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1505
+	ls $DIR/$tdir/ > /dev/null || error "(7) no FID-in-dirent."
+
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+}
+run_test 1c "LFSCK can find out and repair lost FID-in-dirent"
+
 test_2a() {
 	lfsck_prep 1 1
 
@@ -454,8 +493,7 @@ run_test 2d "LFSCK can recover the missing linkEA entry"
 
 test_2e()
 {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	check_mount_and_prep
 
@@ -1022,9 +1060,6 @@ test_9a() {
 		return 0
 	fi
 
-	[[ $server_version -ge $(version_code 2.7.50) ]] ||
-		{ skip "Need MDS version >= 2.7.50"; return; }
-
 	check_mount_and_prep
 	$LFS mkdir -i 0 $DIR/$tdir/lfsck || error "(1) Fail to mkdir lfsck"
 	$LFS setstripe -c 1 -i -1 $DIR/$tdir/lfsck
@@ -1045,11 +1080,14 @@ test_9a() {
 	# There may be time error, normally it should be less than 2 seconds.
 	# We allow another 20% schedule error.
 	local TIME_DIFF=2
-	# MAX_MARGIN = 1.2 = 12 / 10
+	# MAX_MARGIN = 1.3 = 13 / 10
 	local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
-			   RUN_TIME1 * 12 / 10))
-	[ $SPEED -lt $MAX_SPEED ] ||
-		error "(4) Got speed $SPEED, expected less than $MAX_SPEED"
+			   RUN_TIME1 * 13 / 10))
+	[ $SPEED -lt $MAX_SPEED ] || {
+		$SHOW_LAYOUT
+		log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
+		error "(4) Speed $SPEED, expected < $MAX_SPEED"
+	}
 
 	# adjust speed limit
 	local BASE_SPEED2=300
@@ -1059,10 +1097,10 @@ test_9a() {
 	sleep $RUN_TIME2
 
 	SPEED=$($SHOW_LAYOUT | awk '/^average_speed_phase1/ { print $2 }')
-	# MIN_MARGIN = 0.8 = 8 / 10
+	# MIN_MARGIN = 0.7 = 7 / 10
 	local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
 			    BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
-			   (RUN_TIME1 + RUN_TIME2) * 8 / 10))
+			   (RUN_TIME1 + RUN_TIME2) * 7 / 10))
 	[ $SPEED -gt $MIN_SPEED ] || {
 		if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
 			error_ignore LU-5624 \
@@ -1073,15 +1111,21 @@ test_9a() {
 		fi
 	}
 
-	# MAX_MARGIN = 1.2 = 12 / 10
+	# MAX_MARGIN = 1.3 = 13 / 10
 	MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
 		      BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
-		     (RUN_TIME1 + RUN_TIME2) * 12 / 10))
-	[ $SPEED -lt $MAX_SPEED ] ||
-		error "(6) Got speed $SPEED, expected less than $MAX_SPEED"
+		     (RUN_TIME1 + RUN_TIME2) * 13 / 10))
+	[ $SPEED -lt $MAX_SPEED ] || {
+		$SHOW_LAYOUT
+		log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
+		log "speed2: $BASE_SPEED2 time2: $RUN_TIME2"
+		error "(6) Speed $SPEED, expected < $MAX_SPEED"
+	}
 
-	do_facet $SINGLEMDS \
-		$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param -n mdd.*.lfsck_speed_limit 0
+	do_nodes $(comma_list $(osts_nodes)) \
+		$LCTL set_param -n obdfilter.*.lfsck_speed_limit 0
 
 	wait_update_facet $SINGLEMDS \
 		"$LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout |
@@ -1096,9 +1140,6 @@ test_9b() {
 		return 0
 	fi
 
-	[[ $server_version -ge $(version_code 2.7.50) ]] ||
-		{ skip "Need MDS version >= 2.7.50"; return; }
-
 	lfsck_prep 0 0
 
 	echo "Preparing another 50 * 50 files (with error) at $(date)."
@@ -1137,11 +1178,14 @@ test_9b() {
 	# There may be time error, normally it should be less than 2 seconds.
 	# We allow another 20% schedule error.
 	local TIME_DIFF=2
-	# MAX_MARGIN = 1.2 = 12 / 10
+	# MAX_MARGIN = 1.3 = 13 / 10
 	local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
-			  RUN_TIME1 * 12 / 10))
-	[ $SPEED -lt $MAX_SPEED ] ||
-		error "(8) Got speed $SPEED, expected less than $MAX_SPEED"
+			  RUN_TIME1 * 13 / 10))
+	[ $SPEED -lt $MAX_SPEED ] || {
+		$SHOW_NAMESPACE
+		log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
+		error "(8) Speed $SPEED, expected < $MAX_SPEED"
+	}
 
 	# adjust speed limit
 	local BASE_SPEED2=150
@@ -1151,10 +1195,10 @@ test_9b() {
 	sleep $RUN_TIME2
 
 	SPEED=$($SHOW_NAMESPACE | awk '/^average_speed_phase2/ { print $2 }')
-	# MIN_MARGIN = 0.8 = 8 / 10
+	# MIN_MARGIN = 0.7 = 7 / 10
 	local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
 			    BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
-			   (RUN_TIME1 + RUN_TIME2) * 8 / 10))
+			   (RUN_TIME1 + RUN_TIME2) * 7 / 10))
 	[ $SPEED -gt $MIN_SPEED ] || {
 		if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
 			error_ignore LU-5624 \
@@ -1165,15 +1209,21 @@ test_9b() {
 		fi
 	}
 
-	# MAX_MARGIN = 1.2 = 12 / 10
+	# MAX_MARGIN = 1.3 = 13 / 10
 	MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
 		      BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
-		     (RUN_TIME1 + RUN_TIME2) * 12 / 10))
-	[ $SPEED -lt $MAX_SPEED ] ||
-		error "(10) Got speed $SPEED, expected less than $MAX_SPEED"
+		     (RUN_TIME1 + RUN_TIME2) * 13 / 10))
+	[ $SPEED -lt $MAX_SPEED ] || {
+		$SHOW_NAMESPACE
+		log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
+		log "speed2: $BASE_SPEED2 time2: $RUN_TIME2"
+		error "(10) Speed $SPEED, expected < $MAX_SPEED"
+	}
 
-	do_facet $SINGLEMDS \
-		$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param -n mdd.*.lfsck_speed_limit 0
+	do_nodes $(comma_list $(osts_nodes)) \
+		$LCTL set_param -n obdfilter.*.lfsck_speed_limit 0
 	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
 		mdd.${MDT_DEV}.lfsck_namespace |
 		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
@@ -1245,8 +1295,10 @@ test_10()
 	[ "$STATUS" == "scanning-phase1" ] ||
 		error "(15) Expect 'scanning-phase1', but got '$STATUS'"
 
-	do_facet $SINGLEMDS \
-		$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
+	do_nodes $(comma_list $(mdts_nodes)) \
+		$LCTL set_param -n mdd.*.lfsck_speed_limit 0
+	do_nodes $(comma_list $(osts_nodes)) \
+		$LCTL set_param -n obdfilter.*.lfsck_speed_limit 0
 	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
 		mdd.${MDT_DEV}.lfsck_namespace |
 		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
@@ -1383,8 +1435,7 @@ test_11b() {
 run_test 11b "LFSCK can rebuild crashed last_id"
 
 test_12a() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for test_12a" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	check_mount_and_prep
 	for k in $(seq $MDSCOUNT); do
@@ -1485,7 +1536,7 @@ test_13() {
 	#define OBD_FAIL_LFSCK_BAD_LMMOI	0x160f
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160f
 	createmany -o $DIR/$tdir/f 1
-	$LFS setstripe -E 1M -E -1 $DIR/$tdir/f1 ||
+	$LFS setstripe -E 1M -S 1M -E -1 $DIR/$tdir/f1 ||
 		error "(0) Fail to create PFL $DIR/$tdir/f1"
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
 
@@ -1771,8 +1822,7 @@ test_15b() {
 run_test 15b "LFSCK can repair unmatched MDT-object/OST-object pairs (2)"
 
 test_15c() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.55) ] &&
 		skip "Skip the test after 2.7.55 see LU-6437" && return
@@ -1832,6 +1882,11 @@ test_16() {
 	dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
 	cancel_lru_locks osc
 
+	# created but no setattr or write to the file.
+	mkdir $DIR/$tdir/d1
+	chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/d1
+	$RUNAS createmany -o $DIR/$tdir/d1/o 100 || error "create failed"
+
 	echo "Inject failure stub to skip OST-object owner changing"
 	#define OBD_FAIL_LFSCK_BAD_OWNER	0x1613
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1613
@@ -2579,8 +2634,7 @@ test_18e() {
 run_test 18e "Find out orphan OST-object and repair it (5)"
 
 test_18f() {
-	[ $OSTCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 OSTs" && return
+	[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
 
 	echo "#####"
 	echo "The target MDT-object is lost. The LFSCK should re-create the"
@@ -2795,7 +2849,7 @@ test_18h() {
 
 	check_mount_and_prep
 
-	$LFS setstripe -E 2M -c 1 -E -1 $DIR/$tdir/f0 ||
+	$LFS setstripe -E 2M -S 1M -c 1 -E -1 $DIR/$tdir/f0 ||
 		error "(0) Fail to create PFL $DIR/$tdir/f0"
 
 	cat $LUSTRE/tests/test-framework.sh > $DIR/$tdir/f0 ||
@@ -2932,11 +2986,10 @@ test_19b() {
 run_test 19b "OST-object inconsistency self repair"
 
 PATTERN_WITH_HOLE="40000001"
-PATTERN_WITHOUT_HOLE="1"
+PATTERN_WITHOUT_HOLE="raid0"
 
 test_20a() {
-	[ $OSTCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 OSTs" && return
+	[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
 
 	echo "#####"
 	echo "The target MDT-object and some of its OST-object are lost."
@@ -3275,8 +3328,7 @@ test_20a() {
 run_test 20a "Handle the orphan with dummy LOV EA slot properly"
 
 test_20b() {
-	[ $OSTCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 OSTs" && return
+	[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
 
 	echo "#####"
 	echo "The target MDT-object and some of its OST-object are lost."
@@ -3612,8 +3664,7 @@ test_21() {
 run_test 21 "run all LFSCK components by default"
 
 test_22a() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "The parent_A references the child directory via some name entry,"
@@ -3655,8 +3706,7 @@ test_22a() {
 run_test 22a "LFSCK can repair unmatched pairs (1)"
 
 test_22b() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "The parent_A references the child directory via the name entry_B,"
@@ -3705,8 +3755,7 @@ test_22b() {
 run_test 22b "LFSCK can repair unmatched pairs (2)"
 
 test_23a() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "The name entry is there, but the MDT-object for such name "
@@ -3845,6 +3894,18 @@ test_23b() {
 }
 run_test 23b "LFSCK can repair dangling name entry (2)"
 
+cleanup_23c() {
+	do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
+	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+		mdd.${MDT_DEV}.lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+		$SHOW_NAMESPACE
+		error "(10) unexpected status"
+	}
+
+	stop_full_debug_logging
+}
+
 test_23c() {
 	echo "#####"
 	echo "The objectA has multiple hard links, one of them corresponding"
@@ -3864,34 +3925,34 @@ test_23c() {
 	check_mount_and_prep
 
 	$LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
-	$LFS path2fid $DIR/$tdir/d0
+	parent_fid="$($LFS path2fid $DIR/$tdir/d0)"
+	echo "parent_fid=$parent_fid"
 
 	createmany -o $DIR/$tdir/d0/t 10 || error "(1.5) Fail to creatmany"
 
 	echo "dummy" > $DIR/$tdir/d0/f0 || error "(2) Fail to touch on MDT0"
-	$LFS path2fid $DIR/$tdir/d0/f0
+	f0_fid="$($LFS path2fid $DIR/$tdir/d0/f0)"
+	echo "f0_fid=$f0_fid"
 
 	echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
-	$LFS path2fid $DIR/$tdir/d0/f1
-
-	local SEQ0=$($LFS path2fid $DIR/$tdir/d0/f0 | awk -F':' '{print $1}')
-	local SEQ1=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $1}')
+	f1_fid="$($LFS path2fid $DIR/$tdir/d0/f1)"
+	echo "f1_fid=$f1_fid"
 
-	if [ "$SEQ0" != "$SEQ1" ]; then
+	if [ "${fid_f0/:.*/}" != "${fid_f1/:.*/}" ]; then
 		# To guarantee that the f0 and f1 are in the same FID seq
 		rm -f $DIR/$tdir/d0/f0 ||
 			error "(3.1) Fail to unlink $DIR/$tdir/d0/f0"
 		echo "dummy" > $DIR/$tdir/d0/f0 ||
 			error "(3.2) Fail to touch on MDT0"
-		$LFS path2fid $DIR/$tdir/d0/f0
+		f0_fid="$($LFS path2fid $DIR/$tdir/d0/f0)"
+		echo "f0_fid=$f0_fid (replaced)"
 	fi
 
-	local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
-	OID=$(printf %d $OID)
+	local oid=$(awk -F':' '{ printf $2 }' <<< $f1_fid)
 
 	echo "Inject failure stub on MDT0 to simulate dangling name entry"
 	#define OBD_FAIL_LFSCK_DANGLING3	0x1621
-	do_facet $SINGLEMDS $LCTL set_param fail_val=$OID fail_loc=0x1621
+	do_facet $SINGLEMDS $LCTL set_param fail_val=$oid fail_loc=0x1621
 	ln $DIR/$tdir/d0/f0 $DIR/$tdir/d0/foo || error "(4) Fail to hard link"
 	do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
 
@@ -3916,8 +3977,17 @@ test_23c() {
 	$START_NAMESPACE -r -C ||
 		error "(7) Fail to start LFSCK for namespace"
 
-	wait_update_facet client "stat $DIR/$tdir/d0/foo |
-		awk '/Size/ { print \\\$2 }'" "0" $LTIME || {
+	wait_update_facet client "stat -c%s $DIR/$tdir/d0/foo" "0" $LTIME || {
+		# While unexpected by the test, it is valid for LFSCK to repair
+		# the link to the original object before any data is written.
+		local size=$(stat -c %s $DIR/$tdir/d0/foo)
+
+		if [ "$size" = "6" -a "$(<$DIR/$tdir/d0/foo)" = "dummy" ]; then
+			log "LFSCK repaired file prematurely"
+			cleanup_23c
+			return 0
+		fi
+
 		stat $DIR/$tdir/d0/foo
 		$SHOW_NAMESPACE
 		error "(8) unexpected size"
@@ -3926,15 +3996,7 @@ test_23c() {
 	echo "data" >> $DIR/$tdir/d0/foo || error "(9) Fail to write"
 	cancel_lru_locks osc
 
-	do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
-	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
-		mdd.${MDT_DEV}.lfsck_namespace |
-		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
-		$SHOW_NAMESPACE
-		error "(10) unexpected status"
-	}
-
-	stop_full_debug_logging
+	cleanup_23c
 
 	local repaired=$($SHOW_NAMESPACE |
 			 awk '/^dangling_repaired/ { print $2 }')
@@ -3948,8 +4010,7 @@ test_23c() {
 run_test 23c "LFSCK can repair dangling name entry (3)"
 
 test_24() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "Two MDT-objects back reference the same name entry via their"
@@ -4025,7 +4086,7 @@ run_test 24 "LFSCK can repair multiple-referenced name entry"
 
 test_25() {
 	[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
-		skip "Only support to inject failure on ldiskfs" && return
+		skip "ldiskfs only test" && return
 
 	echo "#####"
 	echo "The file type in the name entry does not match the file type"
@@ -4113,8 +4174,7 @@ test_26a() {
 run_test 26a "LFSCK can add the missing local name entry back to the namespace"
 
 test_26b() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "The remote name entry back referenced by the MDT-object is lost."
@@ -4213,8 +4273,7 @@ test_27a() {
 run_test 27a "LFSCK can recreate the lost local parent directory as orphan"
 
 test_27b() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "We need at least 2 MDSes for this test" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "The remote parent referenced by the MDT-object linkEA is lost."
@@ -4267,8 +4326,7 @@ test_27b() {
 run_test 27b "LFSCK can recreate the lost remote parent directory as orphan"
 
 test_28() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "The target name entry is lost. The LFSCK should insert the"
@@ -4482,8 +4540,8 @@ test_29c()
 
 	cancel_lru_locks mdc
 	if [ $MDSCOUNT -ge 2 ]; then
-		$LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null ||
-			error "(3.1) Migrate failure"
+		$LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null &&
+			error "(3.1) Migrate should fail"
 
 		echo "The object with linkEA overflow should NOT be migrated"
 		local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
@@ -4497,8 +4555,8 @@ test_29c()
 	unlinkmany $DIR/$tdir/foo/ttttttttttt 100 || error "(4) Fail to unlink"
 
 	if [ $MDSCOUNT -ge 2 ]; then
-		$LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null ||
-			error "(5.1) Migrate failure"
+		$LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null &&
+			error "(5.1) Migrate should fail"
 
 		# The overflow timestamp is still there, so migration will fail.
 		local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
@@ -4545,7 +4603,7 @@ run_test 29c "verify linkEA size limitation"
 
 test_30() {
 	[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
-		skip "Only support backend /lost+found for ldiskfs" && return
+		skip "ldiskfs only test" && return
 
 	echo "#####"
 	echo "The namespace LFSCK will move the orphans from backend"
@@ -4567,6 +4625,9 @@ test_30() {
 	mkdir $DIR/$tdir/foo/d0 || error "(3) Fail to mkdir d0"
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
 
+	local pfid=$($LFS path2fid $DIR/$tdir/foo)
+	local cfid=$($LFS path2fid $DIR/$tdir/foo/d0)
+
 	touch $DIR/$tdir/foo/d0/f1 || error "(4) Fail to touch f1"
 	mkdir $DIR/$tdir/foo/d0/d1 || error "(5) Fail to mkdir d1"
 
@@ -4617,17 +4678,16 @@ test_30() {
 
 	ls -ail $MOUNT/.lustre/lost+found/MDT0000/
 
-	cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-*-D-*)
+	local cname=$MOUNT/.lustre/lost+found/MDT0000/${cfid}-${pfid}-D-0
 	[ ! -z "$cname" ] || error "(20) d0 is not recovered"
 
-	stat ${cname}/d1 || error "(21) d0 is not recovered"
+	stat ${cname}/d1 || error "(21) d1 is not recovered"
 	stat ${cname}/f1 || error "(22) f1 is not recovered"
 }
 run_test 30 "LFSCK can recover the orphans from backend /lost+found"
 
 test_31a() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For the name entry under a striped directory, if the name"
@@ -4677,8 +4737,7 @@ test_31a() {
 run_test 31a "The LFSCK can find/repair the name entry with bad name hash (1)"
 
 test_31b() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For the name entry under a striped directory, if the name"
@@ -4729,8 +4788,7 @@ test_31b() {
 run_test 31b "The LFSCK can find/repair the name entry with bad name hash (2)"
 
 test_31c() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For some reason, the master MDT-object of the striped directory"
@@ -4773,8 +4831,7 @@ test_31c() {
 run_test 31c "Re-generate the lost master LMV EA for striped directory"
 
 test_31d() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For some reason, the master MDT-object of the striped directory"
@@ -4828,8 +4885,7 @@ test_31d() {
 run_test 31d "Set broken striped directory (modified after broken) as read-only"
 
 test_31e() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For some reason, the slave MDT-object of the striped directory"
@@ -4866,8 +4922,7 @@ test_31e() {
 run_test 31e "Re-generate the lost slave LMV EA for striped directory (1)"
 
 test_31f() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For some reason, the slave MDT-object of the striped directory"
@@ -4905,8 +4960,7 @@ test_31f() {
 run_test 31f "Re-generate the lost slave LMV EA for striped directory (2)"
 
 test_31g() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For some reason, the stripe index in the slave LMV EA is"
@@ -4951,8 +5005,7 @@ test_31g() {
 run_test 31g "Repair the corrupted slave LMV EA"
 
 test_31h() {
-	[ $MDSCOUNT -lt 2 ] &&
-		skip "The test needs at least 2 MDTs" && return
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	echo "#####"
 	echo "For some reason, the shard's name entry in the striped"
@@ -4997,29 +5050,70 @@ test_31h() {
 }
 run_test 31h "Repair the corrupted shard's name entry"
 
-test_32()
+test_32a()
 {
 	lfsck_prep 5 5
 	umount_client $MOUNT
 
-	#define OBD_FAIL_LFSCK_ASSISTANT_DIRECT	0x162d
+	#define OBD_FAIL_LFSCK_ENGINE_DELAY	0x162d
 	do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
-	$START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
+	$START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
 
 	local STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
 	[ "$STATUS" == "scanning-phase1" ] ||
-		error "(3) Expect 'scanning-phase1', but got '$STATUS'"
+		error "(2) Expect 'scanning-phase1', but got '$STATUS'"
 
 	echo "stop ost1"
-	stop ost1 > /dev/null || error "(4) Fail to stop OST1!"
+	stop ost1 > /dev/null || error "(3) Fail to stop OST1!"
 
 	do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
-	sleep 1
+	sleep 4
 
 	echo "stop LFSCK"
-	$STOP_LFSCK || error "(5) Fail to stop LFSCK!"
+	$STOP_LFSCK || error "(4) Fail to stop LFSCK!"
+
+	start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB > /dev/null ||
+		error "(5) Fail to start ost1"
 }
-run_test 32 "stop LFSCK when some OST failed"
+run_test 32a "stop LFSCK when some OST failed"
+
+test_32b()
+{
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	lfsck_prep 5 5
+	$LFS mkdir -i 1 $DIR/$tdir/dp ||
+		error "(1) Fail to create $DIR/$tdir/dp"
+	$LFS mkdir -i 0 -c $MDSCOUNT $DIR/$tdir/dp/dc1 ||
+		error "(2) Fail to create $DIR/$tdir/dp/dc1"
+	$LFS mkdir -i 0 -c $MDSCOUNT $DIR/$tdir/dp/dc2 ||
+		error "(3) Fail to create $DIR/$tdir/dp/dc2"
+	umount_client $MOUNT
+
+	#define OBD_FAIL_LFSCK_ENGINE_DELAY	0x162d
+	do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
+	$START_NAMESPACE -r -A || error "(4) Fail to start LFSCK for namespace!"
+
+	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+		mdd.${MDT_DEV}.lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "scanning-phase1" 32 || {
+		$SHOW_NAMESPACE
+		error "(5) unexpected status"
+	}
+
+	echo "stop mds2"
+	stop mds2 > /dev/null || error "(6) Fail to stop MDT2!"
+
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
+	sleep 4
+
+	echo "stop LFSCK"
+	$STOP_LFSCK || error "(7) Fail to stop LFSCK!"
+
+	start mds2 $(mdsdevname 2) $MOUNT_OPTS_NOSCRUB > /dev/null ||
+		error "(8) Fail to start MDT2"
+}
+run_test 32b "stop LFSCK when some MDT failed"
 
 test_33()
 {
@@ -5043,13 +5137,432 @@ test_33()
 }
 run_test 33 "check LFSCK paramters"
 
+test_34()
+{
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+	[ $(facet_fstype $SINGLEMDS) != zfs ] &&
+		skip "Only valid for ZFS backend" && return
+
+	lfsck_prep 1 1
+
+	#define OBD_FAIL_LFSCK_NO_AGENTOBJ	0x1630
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1630
+	$LFS mkdir -i 1 $DIR/$tdir/dummy ||
+		error "(1) Fail to create $DIR/$tdir/dummy"
+
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+	$START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
+	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+		mdd.${MDT_DEV}.lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+		$SHOW_NAMESPACE
+		error "(3) unexpected status"
+	}
+
+	local repaired=$($SHOW_NAMESPACE |
+			 awk '/^dirent_repaired/ { print $2 }')
+	[ $repaired -eq 1 ] ||
+		error "(4) Fail to repair the lost agent object: $repaired"
+
+	$START_NAMESPACE -r || error "(5) Fail to start LFSCK for namespace!"
+	wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+		mdd.${MDT_DEV}.lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+		$SHOW_NAMESPACE
+		error "(6) unexpected status"
+	}
+
+	repaired=$($SHOW_NAMESPACE | awk '/^dirent_repaired/ { print $2 }')
+	[ $repaired -eq 0 ] ||
+		error "(7) Unexpected repairing: $repaired"
+}
+run_test 34 "LFSCK can rebuild the lost agent object"
+
+test_35()
+{
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+	lfsck_prep 1 1
+
+	#define OBD_FAIL_LFSCK_NO_AGENTENT	0x1631
+	do_facet mds2 $LCTL set_param fail_loc=0x1631
+	$LFS mkdir -i 1 $DIR/$tdir/dummy ||
+		error "(1) Fail to create $DIR/$tdir/dummy"
+
+	sync; sleep 3
+	do_facet mds2 $LCTL set_param fail_loc=0
+	$START_NAMESPACE -A -r || error "(2) Fail to start LFSCK for namespace!"
+	wait_update_facet mds2 "$LCTL get_param -n \
+		mdd.$(facet_svc mds2).lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
+		error "(3) MDS${k} is not the expected 'completed'"
+
+	local repaired=$(do_facet mds2 $LCTL get_param -n \
+			 mdd.$(facet_svc mds2).lfsck_namespace |
+			 awk '/^agent_entries_repaired/ { print $2 }')
+	[ $repaired -eq 1 ] ||
+		error "(4) Fail to repair the lost agent entry: $repaired"
+
+	echo "stopall to cleanup object cache"
+	stopall > /dev/null
+	echo "setupall"
+	setupall > /dev/null
+
+	$START_NAMESPACE -A -r || error "(5) Fail to start LFSCK for namespace!"
+	wait_update_facet mds2 "$LCTL get_param -n \
+		mdd.$(facet_svc mds2).lfsck_namespace |
+		awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
+		error "(6) MDS${k} is not the expected 'completed'"
+
+	repaired=$(do_facet mds2 $LCTL get_param -n \
+		   mdd.$(facet_svc mds2).lfsck_namespace |
+		   awk '/^agent_entries_repaired/ { print $2 }')
+	[ $repaired -eq 0 ] ||
+		error "(7) Unexpected repairing: $repaired"
+}
+run_test 35 "LFSCK can rebuild the lost agent entry"
+
+test_36a() {
+	[ $OSTCOUNT -lt 3 ] && skip "needs >= 3 OSTs" && return
+
+	echo "#####"
+	echo "The target MDT-object's LOV EA corrupted as to lose one of the "
+	echo "mirrors information. The layout LFSCK should rebuild the LOV EA "
+	echo "with the PFID EA of related OST-object(s) belong to the mirror."
+	echo "#####"
+
+	check_mount_and_prep
+
+	$LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+		-N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f0 ||
+		error "(0) Fail to create mirror file $DIR/$tdir/f0"
+	$LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+		-N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f1 ||
+		error "(1) Fail to create mirror file $DIR/$tdir/f1"
+	$LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+		-N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f2 ||
+		error "(2) Fail to create mirror file $DIR/$tdir/f2"
+
+	dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+		error "(3) Fail to write $DIR/$tdir/f0"
+	dd if=/dev/zero of=$DIR/$tdir/f1 bs=1M count=4 ||
+		error "(4) Fail to write $DIR/$tdir/f1"
+	dd if=/dev/zero of=$DIR/$tdir/f2 bs=1M count=4 ||
+		error "(5) Fail to write $DIR/$tdir/f2"
+
+	$LFS mirror resync $DIR/$tdir/f0 ||
+		error "(6) Fail to resync $DIR/$tdir/f0"
+	$LFS mirror resync $DIR/$tdir/f1 ||
+		error "(7) Fail to resync $DIR/$tdir/f1"
+	$LFS mirror resync $DIR/$tdir/f2 ||
+		error "(8) Fail to resync $DIR/$tdir/f2"
+
+	cancel_lru_locks mdc
+	cancel_lru_locks osc
+
+	$LFS getstripe $DIR/$tdir/f0 ||
+		error "(9) Fail to getstripe for $DIR/$tdir/f0"
+	$LFS getstripe $DIR/$tdir/f1 ||
+		error "(10) Fail to getstripe for $DIR/$tdir/f1"
+	$LFS getstripe $DIR/$tdir/f2 ||
+		error "(11) Fail to getstripe for $DIR/$tdir/f2"
+
+	echo "Inject failure, to simulate the case of missing one mirror in LOV"
+	#define OBD_FAIL_LFSCK_LOST_MDTOBJ	0x1616
+	do_facet mds1 $LCTL set_param fail_loc=0x1616
+
+	$LFS mirror split --mirror-id 1 -d $DIR/$tdir/f0 ||
+		error "(12) Fail to split 1st mirror from $DIR/$tdir/f0"
+	$LFS mirror split --mirror-id 2 -d $DIR/$tdir/f1 ||
+		error "(13) Fail to split 2nd mirror from $DIR/$tdir/f1"
+	$LFS mirror split --mirror-id 3 -d $DIR/$tdir/f2 ||
+		error "(14) Fail to split 3rd mirror from $DIR/$tdir/f2"
+
+	sync
+	sleep 2
+	do_facet mds1 $LCTL set_param fail_loc=0
+
+	$LFS getstripe $DIR/$tdir/f0 | grep "lcme_mirror_id:.*1" &&
+		error "(15) The 1st of mirror is not destroyed"
+	$LFS getstripe $DIR/$tdir/f1 | grep "lcme_mirror_id:.*2" &&
+		error "(16) The 2nd of mirror is not destroyed"
+	$LFS getstripe $DIR/$tdir/f2 | grep "lcme_mirror_id:.*3" &&
+		error "(17) The 3rd of mirror is not destroyed"
+
+	local mirrors
+
+	mirrors=$($LFS getstripe -N $DIR/$tdir/f0)
+	[ $mirrors -eq 2 ] || error "(18) $DIR/$tdir/f0 has $mirrors mirrors"
+	mirrors=$($LFS getstripe -N $DIR/$tdir/f1)
+	[ $mirrors -eq 2 ] || error "(19) $DIR/$tdir/f1 has $mirrors mirrors"
+	mirrors=$($LFS getstripe -N $DIR/$tdir/f2)
+	[ $mirrors -eq 2 ] || error "(20) $DIR/$tdir/f2 has $mirrors mirrors"
+
+	echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+	$START_LAYOUT -r -o || error "(21) Fail to start LFSCK for layout!"
+
+	for k in $(seq $MDSCOUNT); do
+		# The LFSCK status query internal is 30 seconds. For the case
+		# of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+		# time to guarantee the status sync up.
+		wait_update_facet mds${k} "$LCTL get_param -n \
+			mdd.$(facet_svc mds${k}).lfsck_layout |
+			awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+			error "(22) MDS${k} is not the expected 'completed'"
+	done
+
+	for k in $(seq $OSTCOUNT); do
+		local cur_status=$(do_facet ost${k} $LCTL get_param -n \
+				obdfilter.$(facet_svc ost${k}).lfsck_layout |
+				awk '/^status/ { print $2 }')
+		[ "$cur_status" == "completed" ] ||
+		error "(23) OST${k} Expect 'completed', but got '$cur_status'"
+	done
+
+	local repaired=$(do_facet mds1 $LCTL get_param -n \
+			 mdd.$(facet_svc mds1).lfsck_layout |
+			 awk '/^repaired_orphan/ { print $2 }')
+	[ $repaired -eq 9 ] ||
+		error "(24) Expect 9 fixed on mds1, but got: $repaired"
+
+	mirrors=$($LFS getstripe -N $DIR/$tdir/f0)
+	[ $mirrors -eq 3 ] || error "(25) $DIR/$tdir/f0 has $mirrors mirrors"
+	mirrors=$($LFS getstripe -N $DIR/$tdir/f1)
+	[ $mirrors -eq 3 ] || error "(26) $DIR/$tdir/f1 has $mirrors mirrors"
+	mirrors=$($LFS getstripe -N $DIR/$tdir/f2)
+	[ $mirrors -eq 3 ] || error "(27) $DIR/$tdir/f2 has $mirrors mirrors"
+
+	$LFS getstripe $DIR/$tdir/f0 | grep "lcme_mirror_id:.*1" || {
+		$LFS getstripe $DIR/$tdir/f0
+		error "(28) The 1st of mirror is not recovered"
+	}
+
+	$LFS getstripe $DIR/$tdir/f1 | grep "lcme_mirror_id:.*2" || {
+		$LFS getstripe $DIR/$tdir/f1
+		error "(29) The 2nd of mirror is not recovered"
+	}
+
+	$LFS getstripe $DIR/$tdir/f2 | grep "lcme_mirror_id:.*3" || {
+		$LFS getstripe $DIR/$tdir/f2
+		error "(30) The 3rd of mirror is not recovered"
+	}
+}
+run_test 36a "rebuild LOV EA for mirrored file (1)"
+
+test_36b() {
+	[ $OSTCOUNT -lt 3 ] && skip "needs >= 3 OSTs" && return
+
+	echo "#####"
+	echo "The mirrored file lost its MDT-object, but relatd OST-objects "
+	echo "are still there. The layout LFSCK should rebuild the LOV EA "
+	echo "with the PFID EA of related OST-object(s) belong to the file. "
+	echo "#####"
+
+	check_mount_and_prep
+
+	$LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+		-N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f0 ||
+		error "(0) Fail to create mirror file $DIR/$tdir/f0"
+
+	local fid=$($LFS path2fid $DIR/$tdir/f0)
+
+	dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+		error "(1) Fail to write $DIR/$tdir/f0"
+	$LFS mirror resync $DIR/$tdir/f0 ||
+		error "(2) Fail to resync $DIR/$tdir/f0"
+
+	cancel_lru_locks mdc
+	cancel_lru_locks osc
+
+	$LFS getstripe $DIR/$tdir/f0 ||
+		error "(3) Fail to getstripe for $DIR/$tdir/f0"
+
+	echo "Inject failure, to simulate the case of missing the MDT-object"
+	#define OBD_FAIL_LFSCK_LOST_MDTOBJ	0x1616
+	do_facet mds1 $LCTL set_param fail_loc=0x1616
+	rm -f $DIR/$tdir/f0 || error "(4) Fail to remove $DIR/$tdir/f0"
+
+	sync
+	sleep 2
+	do_facet mds1 $LCTL set_param fail_loc=0
+
+	echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+	$START_LAYOUT -r -o || error "(5) Fail to start LFSCK for layout!"
+
+	for k in $(seq $MDSCOUNT); do
+		# The LFSCK status query internal is 30 seconds. For the case
+		# of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+		# time to guarantee the status sync up.
+		wait_update_facet mds${k} "$LCTL get_param -n \
+			mdd.$(facet_svc mds${k}).lfsck_layout |
+			awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+			error "(6) MDS${k} is not the expected 'completed'"
+	done
+
+	for k in $(seq $OSTCOUNT); do
+		local cur_status=$(do_facet ost${k} $LCTL get_param -n \
+				obdfilter.$(facet_svc ost${k}).lfsck_layout |
+				awk '/^status/ { print $2 }')
+		[ "$cur_status" == "completed" ] ||
+		error "(7) OST${k} Expect 'completed', but got '$cur_status'"
+	done
+
+	local count=$(do_facet mds1 $LCTL get_param -n \
+		      mdd.$(facet_svc mds1).lfsck_layout |
+		      awk '/^repaired_orphan/ { print $2 }')
+	[ $count -eq 9 ] || error "(8) Expect 9 fixed on mds1, but got: $count"
+
+	local name=$MOUNT/.lustre/lost+found/MDT0000/${fid}-R-0
+	count=$($LFS getstripe --mirror-count $name)
+	[ $count -eq 3 ] || error "(9) $DIR/$tdir/f0 has $count mirrors"
+
+	count=$($LFS getstripe --component-count $name)
+	[ $count -eq 6 ] || error "(10) $DIR/$tdir/f0 has $count components"
+
+	$LFS getstripe $name | grep "lcme_mirror_id:.*1" || {
+		$LFS getstripe $name
+		error "(11) The 1st of mirror is not recovered"
+	}
+
+	$LFS getstripe $name | grep "lcme_mirror_id:.*2" || {
+		$LFS getstripe $name
+		error "(12) The 2nd of mirror is not recovered"
+	}
+
+	$LFS getstripe $name | grep "lcme_mirror_id:.*3" || {
+		$LFS getstripe $name
+		error "(13) The 3rd of mirror is not recovered"
+	}
+}
+run_test 36b "rebuild LOV EA for mirrored file (2)"
+
+test_36c() {
+	[ $OSTCOUNT -lt 3 ] && skip "needs >= 3 OSTs" && return
+
+	echo "#####"
+	echo "The mirrored file has been modified, not resynced yet, then "
+	echo "lost its MDT-object, but relatd OST-objects are still there. "
+	echo "The layout LFSCK should rebuild the LOV EA and relatd status "
+	echo "with the PFID EA of related OST-object(s) belong to the file. "
+	echo "#####"
+
+	check_mount_and_prep
+
+	$LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+		$DIR/$tdir/f0 ||
+		error "(0) Fail to create mirror file $DIR/$tdir/f0"
+
+	local fid=$($LFS path2fid $DIR/$tdir/f0)
+
+	# The 1st dd && resync makes all related OST-objects have been written
+	dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+		error "(1.1) Fail to write $DIR/$tdir/f0"
+	$LFS mirror resync $DIR/$tdir/f0 ||
+		error "(1.2) Fail to resync $DIR/$tdir/f0"
+	# The 2nd dd makes one mirror to be stale
+	dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+		error "(1.3) Fail to write $DIR/$tdir/f0"
+
+	cancel_lru_locks mdc
+	cancel_lru_locks osc
+
+	$LFS getstripe $DIR/$tdir/f0 ||
+		error "(2) Fail to getstripe for $DIR/$tdir/f0"
+
+	local saved_flags1=$($LFS getstripe $DIR/$tdir/f0 | head -n 10 |
+			     awk '/lcme_flags/ { print $2 }')
+	local saved_flags2=$($LFS getstripe $DIR/$tdir/f0 | tail -n 10 |
+			     awk '/lcme_flags/ { print $2 }')
+
+	echo "Inject failure, to simulate the case of missing the MDT-object"
+	#define OBD_FAIL_LFSCK_LOST_MDTOBJ	0x1616
+	do_facet mds1 $LCTL set_param fail_loc=0x1616
+	rm -f $DIR/$tdir/f0 || error "(3) Fail to remove $DIR/$tdir/f0"
+
+	sync
+	sleep 2
+	do_facet mds1 $LCTL set_param fail_loc=0
+
+	echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+	$START_LAYOUT -r -o || error "(4) Fail to start LFSCK for layout!"
+
+	for k in $(seq $MDSCOUNT); do
+		# The LFSCK status query internal is 30 seconds. For the case
+		# of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+		# time to guarantee the status sync up.
+		wait_update_facet mds${k} "$LCTL get_param -n \
+			mdd.$(facet_svc mds${k}).lfsck_layout |
+			awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+			error "(5) MDS${k} is not the expected 'completed'"
+	done
+
+	for k in $(seq $OSTCOUNT); do
+		local cur_status=$(do_facet ost${k} $LCTL get_param -n \
+				obdfilter.$(facet_svc ost${k}).lfsck_layout |
+				awk '/^status/ { print $2 }')
+		[ "$cur_status" == "completed" ] ||
+		error "(6) OST${k} Expect 'completed', but got '$cur_status'"
+	done
+
+	local count=$(do_facet mds1 $LCTL get_param -n \
+		      mdd.$(facet_svc mds1).lfsck_layout |
+		      awk '/^repaired_orphan/ { print $2 }')
+	[ $count -eq 6 ] || error "(7) Expect 9 fixed on mds1, but got: $count"
+
+	local name=$MOUNT/.lustre/lost+found/MDT0000/${fid}-R-0
+	count=$($LFS getstripe --mirror-count $name)
+	[ $count -eq 2 ] || error "(8) $DIR/$tdir/f0 has $count mirrors"
+
+	count=$($LFS getstripe --component-count $name)
+	[ $count -eq 4 ] || error "(9) $DIR/$tdir/f0 has $count components"
+
+	local flags=$($LFS getstripe $name | head -n 10 |
+		awk '/lcme_flags/ { print $2 }')
+	[ "$flags" == "$saved_flags1" ] || {
+		$LFS getstripe $name
+		error "(10) expect flags $saved_flags1, got $flags"
+	}
+
+	flags=$($LFS getstripe $name | tail -n 10 |
+		awk '/lcme_flags/ { print $2 }')
+	[ "$flags" == "$saved_flags2" ] || {
+		$LFS getstripe $name
+		error "(11) expect flags $saved_flags2, got $flags"
+	}
+}
+run_test 36c "rebuild LOV EA for mirrored file (3)"
+
+test_37()
+{
+	local PID
+	local rc
+	local t_dir="$DIR/$tdir/d0"
+	check_mount_and_prep
+
+	$LFS mkdir -i 0 $t_dir || error "(2) Fail to mkdir $t_dir on MDT0"
+	multiop_bg_pause $t_dir D_c || { error "multiop failed: $?"; return 1; }
+	PID=$!
+	rmdir $t_dir
+
+	$START_NAMESPACE -r -A || {
+	    error "(3) Fail to start LFSCK for namespace!"; kill -USR1 $PID; }
+
+	wait_all_targets_blocked namespace completed 4
+	stat $t_dir && rc=1
+	kill -USR1 $PID
+	return $rc
+}
+run_test 37 "LFSCK must skip a ORPHAN"
+
+
 # restore MDS/OST size
 MDSSIZE=${SAVED_MDSSIZE}
 OSTSIZE=${SAVED_OSTSIZE}
 OSTCOUNT=${SAVED_OSTCOUNT}
 
 # cleanup the system at last
-formatall
+REFORMAT="yes" cleanup_and_setup_lustre
 
 complete $SECONDS
+check_and_cleanup_lustre
 exit_status