LU-16036 test: make sanity-lfsck 15d more robust

[fs/lustre-release.git] / lustre / tests / sanity-lfsck.sh

diff --git a/lustre/tests/sanity-lfsck.sh b/lustre/tests/sanity-lfsck.sh
index bcd1a19..960b4d3 100644 (file)
--- a/lustre/tests/sanity-lfsck.sh
+++ b/lustre/tests/sanity-lfsck.sh
@@ -10,7 +10,7 @@ ONLY=${ONLY:-"$*"}
 
 LUSTRE=${LUSTRE:-$(dirname $0)/..}
 . $LUSTRE/tests/test-framework.sh
-init_test_env $@
+init_test_env "$@"
 init_logging
 
 # bug number for skipped test:
@@ -51,9 +51,8 @@ cleanupall
 # build up a clean test environment.
 REFORMAT="yes" check_and_setup_lustre
 
-MDT_DEV="${FSNAME}-MDT0000"
+MDT_DEV=$(devicelabel $SINGLEMDS $(facet_device $SINGLEMDS))
 OST_DEV="${FSNAME}-OST0000"
-MDT_DEVNAME=$(mdsdevname ${SINGLEMDS//mds/})
 START_NAMESPACE="do_facet $SINGLEMDS \
                $LCTL lfsck_start -M ${MDT_DEV} -t namespace"
 START_LAYOUT="do_facet $SINGLEMDS \
@@ -104,17 +103,31 @@ lfsck_prep() {
        echo "prepared $(date)."
 }
 
-run_e2fsck_on_mdt0() {
+start_facet () {
+       local facet=$1
+       local opts=$2
+       local err=$3
+       local dev=$(facet_device $facet)
+
+       start $facet $dev $opts > /dev/null ||
+               error "($err) Fail to start $facet!"
+}
+
+run_e2fsck_on_mds_facet() {
        [ $mds1_FSTYPE == ldiskfs ] || return 0
 
-       stop $SINGLEMDS > /dev/null || error "(0) Fail to the stop MDT0"
-       run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n" |
+       local mds=$1
+
+       stop $mds > /dev/null || error "(0) Fail to the stop $mds"
+       local host=$(facet_active_host $mds)
+       local dev=$(facet_device $mds)
+
+       run_e2fsck $host $dev "-n" |
                grep "Fix? no" && {
-               run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n"
-               error "(2) Detected inconsistency on MDT0"
+               run_e2fsck $host $dev "-n"
+               error "(2) Detected inconsistency on $mds"
        }
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
-               error "(3) Fail to start MDT0"
+       start_facet $mds "$MOUNT_OPTS_NOSCRUB" 3
 }
 
 wait_all_targets_blocked() {
@@ -227,7 +240,7 @@ test_1a() {
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed FID-in-dirent: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -273,7 +286,7 @@ test_1b()
                error "(5) Fail to repair the missing FID-in-LMA: $repaired"
 
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -312,7 +325,7 @@ test_1c() {
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair lost FID-in-dirent: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -415,7 +428,7 @@ test_2a() {
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -452,7 +465,7 @@ test_2b()
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -492,7 +505,7 @@ test_2c()
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -532,7 +545,7 @@ test_2d()
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(6) Fail to start client!"
 
@@ -629,12 +642,11 @@ test_4()
 
        lfsck_prep 3 3
        cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
-       stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop MDS!"
+       stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop $SINGLEMDS!"
 
        mds_backup_restore $SINGLEMDS || error "(1) Fail to backup/restore!"
        echo "start $SINGLEMDS with disabling OI scrub"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
-               error "(2) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_NOSCRUB" 2
 
        #define OBD_FAIL_LFSCK_DELAY2           0x1601
        do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
@@ -658,7 +670,7 @@ test_4()
                error "(7) unexpected status"
        }
 
-       FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
+       local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
        [ -z "$FLAGS" ] || error "(8) Expect empty flags, but got '$FLAGS'"
 
        local repaired=$($SHOW_NAMESPACE |
@@ -671,7 +683,7 @@ test_4()
        [ $repaired -ge 9 ] ||
                error "(9) Fail to re-generate FID-in-dirent: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(10) Fail to start client!"
 
@@ -689,12 +701,11 @@ test_5()
 
        lfsck_prep 1 1 1
        cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
-       stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop MDS!"
+       stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop $SINGLEMDS!"
 
        mds_backup_restore $SINGLEMDS 1 || error "(1) Fail to backup/restore!"
        echo "start $SINGLEMDS with disabling OI scrub"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
-               error "(2) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_NOSCRUB" 2
 
        #define OBD_FAIL_LFSCK_DELAY2           0x1601
        do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
@@ -718,7 +729,7 @@ test_5()
                error "(7) unexpected status"
        }
 
-       FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
+       local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
        [ -z "$FLAGS" ] || error "(8) Expect empty flags, but got '$FLAGS'"
 
        local repaired=$($SHOW_NAMESPACE |
@@ -731,7 +742,7 @@ test_5()
        [ $repaired -ge 2 ] ||
                error "(9) Fail to generate FID-in-dirent for IGIF: $repaired"
 
-       run_e2fsck_on_mdt0
+       run_e2fsck_on_mds_facet $SINGLEMDS
 
        mount_client $MOUNT || error "(10) Fail to start client!"
 
@@ -881,12 +892,11 @@ test_7a()
        # Sleep 3 sec to guarantee at least one object processed by LFSCK
        sleep 3
        echo "stop $SINGLEMDS"
-       stop $SINGLEMDS > /dev/null || error "(4) Fail to stop MDS!"
+       stop $SINGLEMDS > /dev/null || error "(4) Fail to stop $SINGLEMDS!"
 
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
        echo "start $SINGLEMDS"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
-               error "(5) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_SCRUB" 5
 
        wait_update_facet $SINGLEMDS "$LCTL get_param -n \
                mdd.${MDT_DEV}.lfsck_namespace |
@@ -919,12 +929,11 @@ test_7b()
 
        umount_client $MOUNT
        echo "stop $SINGLEMDS"
-       stop $SINGLEMDS > /dev/null || error "(5) Fail to stop MDS!"
+       stop $SINGLEMDS > /dev/null || error "(5) Fail to stop $SINGLEMDS!"
 
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
        echo "start $SINGLEMDS"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
-               error "(6) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_SCRUB" 6
 
        wait_update_facet $SINGLEMDS "$LCTL get_param -n \
                mdd.${MDT_DEV}.lfsck_namespace |
@@ -935,6 +944,12 @@ test_7b()
 }
 run_test 7b "non-stopped LFSCK should auto restarts after MDS remount (2)"
 
+namespace_error()
+{
+       $SHOW_NAMESPACE
+       error "$@"
+}
+
 test_8()
 {
        echo "formatall"
@@ -946,7 +961,7 @@ test_8()
 
        local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "init" ] ||
-               error "(2) Expect 'init', but got '$STATUS'"
+               namespace_error "(2) Expect 'init', but got '$STATUS'"
 
        #define OBD_FAIL_LFSCK_LINKEA_CRASH     0x1603
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
@@ -962,23 +977,25 @@ test_8()
 
        #define OBD_FAIL_LFSCK_DELAY2           0x1601
        do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1601
-       $START_NAMESPACE || error "(4) Fail to start LFSCK for namespace!"
+       $START_NAMESPACE ||
+               namespace_error "(4) Fail to start LFSCK for namespace!"
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "scanning-phase1" ] ||
-               error "(5) Expect 'scanning-phase1', but got '$STATUS'"
+               namespace_error "(5) Expect 'scanning-phase1', but got '$STATUS'"
 
-       $STOP_LFSCK || error "(6) Fail to stop LFSCK!"
+       $STOP_LFSCK || namespace_error "(6) Fail to stop LFSCK!"
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "stopped" ] ||
-               error "(7) Expect 'stopped', but got '$STATUS'"
+               namespace_error "(7) Expect 'stopped', but got '$STATUS'"
 
-       $START_NAMESPACE || error "(8) Fail to start LFSCK for namespace!"
+       $START_NAMESPACE ||
+               namespace_error "(8) Fail to start LFSCK for namespace!"
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "scanning-phase1" ] ||
-               error "(9) Expect 'scanning-phase1', but got '$STATUS'"
+               namespace_error "(9) Expect 'scanning-phase1', but got '$STATUS'"
 
        #define OBD_FAIL_LFSCK_FATAL2           0x1609
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001609
@@ -986,30 +1003,30 @@ test_8()
                mdd.${MDT_DEV}.lfsck_namespace |
                awk '/^status/ { print \\\$2 }'" "failed" 32 || {
                $SHOW_NAMESPACE
-               error "(10) unexpected status"
+               namespace_error "(10) unexpected status"
        }
 
        #define OBD_FAIL_LFSCK_DELAY1           0x1600
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1600
-       $START_NAMESPACE || error "(11) Fail to start LFSCK for namespace!"
+       $START_NAMESPACE ||
+               namespace_error "(11) Fail to start LFSCK for namespace!"
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "scanning-phase1" ] ||
-               error "(12) Expect 'scanning-phase1', but got '$STATUS'"
+               namespace_error "(12) Expect 'scanning-phase1', but got '$STATUS'"
 
        #define OBD_FAIL_LFSCK_CRASH            0x160a
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160a
        sleep 5
 
        echo "stop $SINGLEMDS"
-       stop $SINGLEMDS > /dev/null || error "(13) Fail to stop MDS!"
+       stop $SINGLEMDS > /dev/null || namespace_error "(13) Fail to stop MDS!"
 
        #define OBD_FAIL_LFSCK_NO_AUTO          0x160b
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160b
 
        echo "start $SINGLEMDS"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
-               error "(14) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_SCRUB" 14
 
        local timeout=$(max_recovery_time)
        local timer=0
@@ -1023,30 +1040,33 @@ test_8()
                timer=$((timer + 1))
        done
 
-       [ $timer != $timeout ] ||
+       [ $timer != $timeout ] || (
+               do_facet $SINGLEMDS "$LCTL get_param -n \
+                       mdt.${MDT_DEV}.recovery_status"
                error "(14.1) recovery timeout"
+               )
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "crashed" ] ||
-               error "(15) Expect 'crashed', but got '$STATUS'"
+               namespace_error "(15) Expect 'crashed', but got '$STATUS'"
 
        #define OBD_FAIL_LFSCK_DELAY2           0x1601
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
-       $START_NAMESPACE || error "(16) Fail to start LFSCK for namespace!"
+       $START_NAMESPACE ||
+               namespace_error "(16) Fail to start LFSCK for namespace!"
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "scanning-phase1" ] ||
-               error "(17) Expect 'scanning-phase1', but got '$STATUS'"
+               namespace_error "(17) Expect 'scanning-phase1', but got '$STATUS'"
 
        echo "stop $SINGLEMDS"
-       stop $SINGLEMDS > /dev/null || error "(18) Fail to stop MDS!"
+       stop $SINGLEMDS > /dev/null || error "(18) Fail to stop $SINGLEMDS!"
 
        #define OBD_FAIL_LFSCK_NO_AUTO          0x160b
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160b
 
        echo "start $SINGLEMDS"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
-               error "(19) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_SCRUB" 19
 
        timer=0
        while [ $timer -lt $timeout ]; do
@@ -1058,19 +1078,21 @@ test_8()
                timer=$((timer + 1))
        done
 
-       [ $timer != $timeout ] ||
+       [ $timer != $timeout ] || (
+               do_facet $SINGLEMDS "$LCTL get_param -n \
+                       mdt.${MDT_DEV}.recovery_status"
                error "(19.1) recovery timeout"
+               )
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "paused" ] ||
-               error "(20) Expect 'paused', but got '$STATUS'"
+               namespace_error "(20) Expect 'paused', but got '$STATUS'"
 
        echo "stop $SINGLEMDS"
        stop $SINGLEMDS > /dev/null || error "(20.1) Fail to stop MDS!"
 
        echo "start $SINGLEMDS without resume LFSCK"
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SKIP_LFSCK > /dev/null ||
-               error "(20.2) Fail to start MDS!"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_SKIP_LFSCK" 20.2
 
        timer=0
        while [ $timer -lt $timeout ]; do
@@ -1082,22 +1104,26 @@ test_8()
                timer=$((timer + 1))
        done
 
-       [ $timer != $timeout ] ||
+       [ $timer != $timeout ] || (
+               do_facet $SINGLEMDS "$LCTL get_param -n \
+                       mdt.${MDT_DEV}.recovery_status"
                error "(20.3) recovery timeout"
+               )
 
        STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
        [ "$STATUS" == "paused" ] ||
-               error "(20.4) Expect 'paused', but got '$STATUS'"
+               namespace_error "(20.4) Expect 'paused', but got '$STATUS'"
 
        #define OBD_FAIL_LFSCK_DELAY3           0x1602
        do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1602
 
-       $START_NAMESPACE || error "(21) Fail to start LFSCK for namespace!"
+       $START_NAMESPACE ||
+               namespace_error "(21) Fail to start LFSCK for namespace!"
        wait_update_facet $SINGLEMDS "$LCTL get_param -n \
                mdd.${MDT_DEV}.lfsck_namespace |
                awk '/^status/ { print \\\$2 }'" "scanning-phase2" 32 || {
                $SHOW_NAMESPACE
-               error "(22) unexpected status"
+               namespace_error "(22) unexpected status"
        }
 
        # wait to process one inode at least (OBD_FAIL_LFSCK_DELAY3)
@@ -1105,18 +1131,19 @@ test_8()
 
        local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
        [ "$FLAGS" == "scanned-once,inconsistent" ] ||
-               error "(23) Expect 'scanned-once,inconsistent',but got '$FLAGS'"
+               namespace_error "(23) Expect 'scanned-once,inconsistent',but got '$FLAGS'"
 
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
        wait_update_facet $SINGLEMDS "$LCTL get_param -n \
                mdd.${MDT_DEV}.lfsck_namespace |
                awk '/^status/ { print \\\$2 }'" "completed" 32 || {
                $SHOW_NAMESPACE
-               error "(24) unexpected status"
+               namespace_error "(24) unexpected status"
        }
 
        FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
-       [ -z "$FLAGS" ] || error "(25) Expect empty flags, but got '$FLAGS'"
+       [ -z "$FLAGS" ] ||
+               namespace_error "(25) Expect empty flags, but got '$FLAGS'"
 }
 run_test 8 "LFSCK state machine"
 
@@ -1147,7 +1174,7 @@ test_9a() {
        # We allow another 20% schedule error.
        local TIME_DIFF=2
        # MAX_MARGIN = 1.3 = 13 / 10
-       local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
+       local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) /
                           RUN_TIME1 * 13 / 10))
        [ $SPEED -lt $MAX_SPEED ] || {
                $SHOW_LAYOUT
@@ -1164,8 +1191,8 @@ test_9a() {
 
        SPEED=$($SHOW_LAYOUT | awk '/^average_speed_phase1/ { print $2 }')
        # MIN_MARGIN = 0.7 = 7 / 10
-       local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
-                           BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
+       local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) +
+                           BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) /
                           (RUN_TIME1 + RUN_TIME2) * 7 / 10))
        [ $SPEED -gt $MIN_SPEED ] || {
                if [ $mds1_FSTYPE != ldiskfs ]; then
@@ -1178,8 +1205,8 @@ test_9a() {
        }
 
        # MAX_MARGIN = 1.3 = 13 / 10
-       MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
-                     BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
+       MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) +
+                     BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) /
                     (RUN_TIME1 + RUN_TIME2) * 13 / 10))
        [ $SPEED -lt $MAX_SPEED ] || {
                $SHOW_LAYOUT
@@ -1245,7 +1272,7 @@ test_9b() {
        # We allow another 20% schedule error.
        local TIME_DIFF=2
        # MAX_MARGIN = 1.3 = 13 / 10
-       local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
+       local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) /
                          RUN_TIME1 * 13 / 10))
        [ $SPEED -lt $MAX_SPEED ] || {
                $SHOW_NAMESPACE
@@ -1262,8 +1289,8 @@ test_9b() {
 
        SPEED=$($SHOW_NAMESPACE | awk '/^average_speed_phase2/ { print $2 }')
        # MIN_MARGIN = 0.7 = 7 / 10
-       local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
-                           BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
+       local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) +
+                           BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) /
                           (RUN_TIME1 + RUN_TIME2) * 7 / 10))
        [ $SPEED -gt $MIN_SPEED ] || {
                if [ $mds1_FSTYPE != ldiskfs ]; then
@@ -1276,8 +1303,8 @@ test_9b() {
        }
 
        # MAX_MARGIN = 1.3 = 13 / 10
-       MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
-                     BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
+       MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) +
+                     BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) /
                     (RUN_TIME1 + RUN_TIME2) * 13 / 10))
        [ $SPEED -lt $MAX_SPEED ] || {
                $SHOW_NAMESPACE
@@ -1961,6 +1988,53 @@ test_15c() {
 }
 run_test 15c "LFSCK can repair unmatched MDT-object/OST-object pairs (3)"
 
+test_15d() {
+       (( $MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
+
+       check_mount_and_prep
+       rm -rf $DIR/$tdir
+       $LFS mkdir -c -1 $DIR/$tdir || error "create $tdir failed"
+       $LFS setdirstripe -D -i -1 -c 1 $DIR/$tdir ||
+               error "setdirstripe failed"
+
+       createmany -o $DIR/$tdir/f 100 || error "create sub files failed"
+       createmany -d $DIR/$tdir/s 100 || error "create sub dirs failed"
+
+       echo "Migrate $DIR/$tdir to MDT1"
+       $LFS migrate -m 1 $DIR/$tdir &
+       pid=$!
+
+       sleep 2
+       # fail sub transactions on random MDTs, which may cause some file
+       # inaccessible
+       #define OBD_FAIL_OUT_EIO                0x1709
+       for ((i = 0; i < $MDSCOUNT; i++)); do
+               do_facet mds$i $LCTL set_param fail_loc=0x1709
+               sleep 0.1
+               do_facet mds$i $LCTL set_param fail_loc=0
+       done
+
+       wait $pid
+
+       # LFSCK can't fully fix migrating directories, and may leave some
+       # files inaccessible, but it shouldn't cause crash
+       $START_NAMESPACE -A -r ||
+               error "Fail to start LFSCK for namespace"
+
+       wait_all_targets_blocked namespace completed 1
+
+       # resume migration may fail because some file may be inaccessible, but
+       # it shouldn't cause crash
+       $LFS migrate -m 1 $DIR/$tdir
+
+       # rm $tdir to avoid cleanup failure in the end
+       rm -rf $DIR/$tdir/*
+       $LFS rm_entry $DIR/$tdir/*
+       rm -rf $DIR/$tdir
+       REFORMAT="yes" cleanup_and_setup_lustre
+}
+run_test 15d "LFSCK don't crash upon dir migration failure"
+
 test_16() {
        (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
                skip "MDS older than 2.5.55, LU-3594"
@@ -4838,14 +4912,15 @@ test_30() {
 
        umount_client $MOUNT || error "(10) Fail to stop client!"
 
-       stop $SINGLEMDS || error "(11) Fail to stop MDT0"
+       stop $SINGLEMDS || error "(11) Fail to stop $SINGLEMDS"
+
+       local dev=$(facet_device $SINGLEMDS)
 
-       echo "run e2fsck"
-       run_e2fsck $(facet_host $SINGLEMDS) $MDT_DEVNAME "-y" ||
+       echo "run e2fsck on $SINGLEMDS"
+       run_e2fsck $(facet_active_host $SINGLEMDS) $dev "-y" ||
                error "(12) Fail to run e2fsck"
 
-       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
-               error "(13) Fail to start MDT0"
+       start_facet $SINGLEMDS "$MOUNT_OPTS_NOSCRUB" 13
 
        echo "Trigger namespace LFSCK to recover backend orphans"
        $START_NAMESPACE -r -A ||