X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Freplay-single.sh;h=3b53bac1ec9349fda26fe14bc54097cb2e9ca68c;hb=1e63c2f85da17947d78d4b2fb79cab2bd04b2ca5;hp=9998498710fc1b58695cf114b4674d4f0b0126f7;hpb=1bf48948a99f51ede84a467a23c59c55eb055d7b;p=fs%2Flustre-release.git

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 9998498..3b53bac 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -22,23 +22,23 @@ require_dsh_mds || exit 0
 
 # Skip these tests
 # bug number for skipped tests:
-# b=17466/LU-472 : 61d
-# LU-5319 : 53a 53d
-ALWAYS_EXCEPT="61d 53a 53d $REPLAY_SINGLE_EXCEPT"
+ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT "
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
-case "$(lsb_release -sr)" in	# only disable tests for el7
-7*)	# bug number:  LU-6455-----
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 28"
-	;;
-esac
+# time in minutes:                 7.5"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="44b"
 
-#                                                  63 min  7 min  AT AT AT AT"
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a      44b    65 66 67 68"
-
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:        LU-1867	LU-3127
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89	73b"
+if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
+# bug number for skipped test: LU-11388
+	ALWAYS_EXCEPT+="131b"
+fi
+if $SHARED_KEY; then
+# bug number for skipped tests: LU-9795 (all below)
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT	0b	0c	0d	34	45"
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT	47	58b	58c	71a	85a"
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT	85b	86	88	89	90"
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT	93a	100a	100b	120"
+fi
 
 build_test_filter
 
@@ -127,6 +127,27 @@ test_2b() {
 }
 run_test 2b "touch"
 
+test_2c() {
+	replay_barrier $SINGLEMDS
+	$LFS setstripe -c $OSTCOUNT $DIR/$tfile
+	fail $SINGLEMDS
+	$CHECKSTAT -t file $DIR/$tfile ||
+		error "$CHECKSTAT $DIR/$tfile check failed"
+}
+run_test 2c "setstripe replay"
+
+test_2d() {
+	[[ $mds1_FSTYPE = "zfs" ]] &&
+		[[ $MDS1_VERSION -lt $(version_code 2.12.51) ]] &&
+		skip "requires LU-10143 fix on MDS"
+	replay_barrier $SINGLEMDS
+	$LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir
+	fail $SINGLEMDS
+	$CHECKSTAT -t dir $DIR/$tdir ||
+		error "$CHECKSTAT $DIR/$tdir check failed"
+}
+run_test 2d "setdirstripe replay"
+
 test_3a() {
 	local file=$DIR/$tfile
 	replay_barrier $SINGLEMDS
@@ -456,35 +477,47 @@ run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans
 
 test_20b() { # bug 10480
 	local wait_timeout=$((TIMEOUT * 4))
-	local BEFOREUSED
-	local AFTERUSED
+	local extra=$(fs_log_size)
+	local n_attempts=1
+
+	sync_all_data
+	$LFS setstripe -i 0 -c 1 $DIR
+
+	local beforeused=$(df -P $DIR | tail -1 | awk '{ print $3 }')
 
-	BEFOREUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
 	dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
 	while [ ! -e $DIR/$tfile ] ; do
-		usleep 60                      # give dd a chance to start
+		usleep 60			# give dd a chance to start
 	done
 
-	$GETSTRIPE $DIR/$tfile || error "$GETSTRIPE $DIR/$tfile failed"
+	$LFS getstripe $DIR/$tfile || error "$LFS getstripe $DIR/$tfile failed"
 	# make it an orphan
 	rm -f $DIR/$tfile || error "rm -f $DIR/$tfile failed"
 	mds_evict_client
-	client_up || client_up || true    # reconnect
+	client_up || client_up || true		# reconnect
 
-	do_facet $SINGLEMDS "lctl set_param -n osd*.*MDT*.force_sync 1"
+	do_facet $SINGLEMDS "lctl set_param -n osd*.*MDT*.force_sync=1"
 
-	fail $SINGLEMDS                            # start orphan recovery
+	fail $SINGLEMDS				# start orphan recovery
 	wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
-	wait_delete_completed_mds $wait_timeout ||
-		error "delete did not complete"
+	wait_delete_completed $wait_timeout || error "delete did not finish"
+	sync_all_data
 
-	AFTERUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
-	log "before $BEFOREUSED, after $AFTERUSED"
-	(( $AFTERUSED > $BEFOREUSED + $(fs_log_size) )) &&
-		error "after $AFTERUSED > before $BEFOREUSED"
-	return 0
+	while true; do
+		local afterused=$(df -P $DIR | tail -1 | awk '{ print $3 }')
+		log "before $beforeused, after $afterused"
+
+		(( $beforeused + $extra >= $afterused )) && break
+		n_attempts=$((n_attempts + 1))
+		[ $n_attempts -gt 3 ] &&
+			error "after $afterused > before $beforeused + $extra"
+
+		wait_zfs_commit $SINGLEMDS 5
+		sync_all_data
+	done
 }
-run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
+
+run_test 20b "write, unlink, eviction, replay (test mds_cleanup_orphans)"
 
 test_20c() { # bug 10480
 	multiop_bg_pause $DIR/$tfile Ow_c ||
@@ -917,13 +950,11 @@ run_test 40 "cause recovery in ptlrpc, ensure IO continues"
 # the page, guarnateeing that the unlock from the RPC completion would
 # assert on trying to unlock the unlocked page.
 test_41() {
-    [ $OSTCOUNT -lt 2 ] &&
-        skip_env "skipping test 41: we don't have a second OST to test with" &&
-        return
+	[ $OSTCOUNT -lt 2 ] && skip_env "needs >= 2 OSTs" && return
 
 	local f=$MOUNT/$tfile
 	# make sure the start of the file is ost1
-	$SETSTRIPE -S $((128 * 1024)) -i 0 $f
+	$LFS setstripe -S $((128 * 1024)) -i 0 $f
 	do_facet client dd if=/dev/zero of=$f bs=4k count=1 ||
 		error "dd on client failed"
 	cancel_lru_locks osc
@@ -1007,7 +1038,7 @@ test_44a() { # was test_44
 		do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
 		# lctl below may fail, it is valid case
 		$LCTL --device $mdcdev recover
-		df $MOUNT
+		$LFS df $MOUNT
 	done
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 	[ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
@@ -1125,8 +1156,7 @@ run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
 
 test_48() {
 	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
-	[ "$OSTCOUNT" -lt "2" ] &&
-		skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
+	[ "$OSTCOUNT" -lt "2" ] && skip_env "needs >= 2 OSTs" && return
 
 	replay_barrier $SINGLEMDS
 	createmany -o $DIR/$tfile 20  ||
@@ -1136,6 +1166,10 @@ test_48() {
 	do_facet ost1 "lctl set_param fail_loc=0x80000216"
 	client_up || error "client_up failed"
 
+	# let the MDS discover the OST failure, attempt to recover, fail
+	# and recover again.
+	sleep $((3 * TIMEOUT))
+
 	createmany -o $DIR/$tfile 20 20 ||
 		error "createmany recraete $DIR/$tfile failed"
 	unlinkmany $DIR/$tfile 40 || error "unlinkmany $DIR/$tfile failed"
@@ -1186,6 +1220,10 @@ run_test 52 "time out lock replay (3764)"
 
 # bug 3462 - simultaneous MDC requests
 test_53a() {
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
 	cancel_lru_locks mdc    # cleanup locks from former test cases
 	mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
 	mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
@@ -1290,6 +1328,10 @@ test_53c() {
 run_test 53c "|X| open request and close request while two MDC requests in flight"
 
 test_53d() {
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
 	cancel_lru_locks mdc    # cleanup locks from former test cases
 
 	mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
@@ -1497,7 +1539,8 @@ test_57() {
 	touch $DIR/$tfile || error "touch $DIR/$tfile failed"
 	replay_barrier $SINGLEMDS
 	fail $SINGLEMDS
-	sleep 1
+	wait_recovery_complete $SINGLEMDS || error "MDS recovery is not done"
+	wait_mds_ost_sync || error "wait_mds_ost_sync failed"
 	$CHECKSTAT -t file $DIR/$tfile ||
 		error "$CHECKSTAT $DIR/$tfile attribute check failed"
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
@@ -1804,36 +1847,39 @@ run_test 65a "AT: verify early replies"
 
 test_65b() #bug 3055
 {
-    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
-
-    at_start || return 0
-    # turn on D_ADAPTTO
-    debugsave
-    $LCTL set_param debug="other trace"
-    $LCTL dk > /dev/null
-    # Slow down a request to the current service time, this is critical
-    # because previous tests may have caused this value to increase.
-    $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
-    multiop $DIR/$tfile Ow1yc
-    REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
-               awk '/portal 6/ {print $5}'`
-    REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
-
-    do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
-#define OBD_FAIL_OST_BRW_PAUSE_PACK      0x224
-    do_facet ost1 $LCTL set_param fail_loc=0x224
+	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
 
-    rm -f $DIR/$tfile
-    $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
-    # force some real bulk transfer
-    multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+	at_start || return 0
+	# turn on D_ADAPTTO
+	debugsave
+	$LCTL set_param debug="other trace"
+	$LCTL dk > /dev/null
+	# Slow down a request to the current service time, this is critical
+	# because previous tests may have caused this value to increase.
+	$LFS setstripe --stripe-index=0 --stripe-count=1 $DIR/$tfile ||
+		error "$LFS setstripe failed for $DIR/$tfile"
+
+	multiop $DIR/$tfile Ow1yc
+	REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
+		   awk '/portal 6/ {print $5}'`
+	REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
+
+	do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
+	#define OBD_FAIL_OST_BRW_PAUSE_PACK      0x224
+	do_facet ost1 $LCTL set_param fail_loc=0x224
 
-    do_facet ost1 $LCTL set_param fail_loc=0
-    # check for log message
-    $LCTL dk | grep "Early reply #" || error "No early reply"
-    debugrestore
-    # client should show REQ_DELAY estimates
-    lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
+	rm -f $DIR/$tfile
+	$LFS setstripe --stripe-index=0 --stripe-count=1 $DIR/$tfile ||
+		error "$LFS setstripe failed"
+	# force some real bulk transfer
+	multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+
+	do_facet ost1 $LCTL set_param fail_loc=0
+	# check for log message
+	$LCTL dk | grep "Early reply #" || error "No early reply"
+	debugrestore
+	# client should show REQ_DELAY estimates
+	lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
 }
 run_test 65b "AT: verify early replies on packed reply / bulk"
 
@@ -1877,7 +1923,7 @@ test_66b() #bug 3055
 	$LCTL set_param fail_val=$(($ORIG + 5))
 	#define OBD_FAIL_PTLRPC_PAUSE_REP      0x50c
 	$LCTL set_param fail_loc=0x50c
-	ls $DIR/$tfile > /dev/null 2>&1
+	touch $DIR/$tfile > /dev/null 2>&1
 	$LCTL set_param fail_loc=0
 	CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
 		awk '/network/ {print $4}')
@@ -1927,7 +1973,8 @@ test_67b() #bug 3055
         osc.$mdtosc.prealloc_next_id)
 
 	mkdir -p $DIR/$tdir/${OST} || error "mkdir $DIR/$tdir/${OST} failed"
-	$SETSTRIPE -i 0 -c 1 $DIR/$tdir/${OST} || error "$SETSTRIPE failed"
+	$LFS setstripe -i 0 -c 1 $DIR/$tdir/${OST} ||
+		error "$LFS setstripe failed"
 	echo "Creating to objid $last_id on ost $OST..."
 #define OBD_FAIL_OST_PAUSE_CREATE        0x223
     do_facet ost1 "$LCTL set_param fail_val=20000"
@@ -1969,20 +2016,21 @@ test_68 () #bug 13813
 	do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
 
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
-	$SETSTRIPE --stripe-index=0 --count=1 $DIR/$tdir
-#define OBD_FAIL_LDLM_PAUSE_CANCEL       0x312
-    $LCTL set_param fail_val=$(($TIMEOUT - 1))
-    $LCTL set_param fail_loc=0x80000312
-    cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
-    $LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
-    $LCTL set_param fail_loc=0x80000312
-    cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
-    $LCTL set_param fail_loc=0
-
-    echo $ENQ_MIN >> $ldlm_enqueue_min
-    do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
-    rm -rf $DIR/$tdir
-    return 0
+	$LFS setstripe --stripe-index=0 -c 1 $DIR/$tdir ||
+		error "$LFS setstripe failed for $DIR/$tdir"
+	#define OBD_FAIL_LDLM_PAUSE_CANCEL       0x312
+	$LCTL set_param fail_val=$(($TIMEOUT - 1))
+	$LCTL set_param fail_loc=0x80000312
+	cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
+	$LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
+	$LCTL set_param fail_loc=0x80000312
+	cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
+	$LCTL set_param fail_loc=0
+
+	echo $ENQ_MIN >> $ldlm_enqueue_min
+	do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
+	rm -rf $DIR/$tdir
+	return 0
 }
 run_test 68 "AT: verify slowing locks"
 
@@ -2092,6 +2140,395 @@ test_70b () {
 run_test 70b "dbench ${MDSCOUNT}mdts recovery; $CLIENTCOUNT clients"
 # end multi-client tests
 
+random_fail_mdt() {
+	local max_index=$1
+	local duration=$2
+	local monitor_pid=$3
+	local elapsed
+	local start_ts=$(date +%s)
+	local num_failovers=0
+	local fail_index
+
+	elapsed=$(($(date +%s) - start_ts))
+	while [ $elapsed -lt $duration ]; do
+		fail_index=$((RANDOM%max_index+1))
+		kill -0 $monitor_pid ||
+			error "$monitor_pid stopped"
+		sleep 120
+		replay_barrier mds$fail_index
+		sleep 10
+		# Increment the number of failovers
+		num_failovers=$((num_failovers+1))
+		log "$TESTNAME fail mds$fail_index $num_failovers times"
+		fail mds$fail_index
+		elapsed=$(($(date +%s) - start_ts))
+	done
+}
+
+cleanup_70c() {
+	trap 0
+	rm -f $DIR/replay-single.70c.lck
+	rm -rf /$DIR/$tdir
+}
+
+test_70c () {
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=600
+
+	local elapsed
+	local start_ts=$(date +%s)
+
+	trap cleanup_70c EXIT
+	(
+		while [ ! -e $DIR/replay-single.70c.lck ]; do
+			test_mkdir -p -c$MDSCOUNT $DIR/$tdir || break
+			if [ $MDSCOUNT -ge 2 ]; then
+				$LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir ||
+				error "set default dirstripe failed"
+			fi
+			cd $DIR/$tdir || break
+			tar cf - /etc | tar xf - || error "tar failed in loop"
+		done
+	)&
+	tar_70c_pid=$!
+	echo "Started tar $tar_70c_pid"
+
+	random_fail_mdt $MDSCOUNT $duration $tar_70c_pid
+	kill -0 $tar_70c_pid || error "tar $tar_70c_pid stopped"
+
+	touch $DIR/replay-single.70c.lck
+	wait $tar_70c_pid || error "$?: tar failed"
+
+	cleanup_70c
+	true
+}
+run_test 70c "tar ${MDSCOUNT}mdts recovery"
+
+cleanup_70d() {
+	trap 0
+	kill -9 $mkdir_70d_pid
+}
+
+test_70d () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	mkdir -p $DIR/$tdir
+
+	local elapsed
+	local start_ts=$(date +%s)
+
+	trap cleanup_70d EXIT
+	(
+		while true; do
+			$LFS mkdir -i0 -c2 $DIR/$tdir/test || {
+				echo "mkdir fails"
+				break
+			}
+			$LFS mkdir -i1 -c2 $DIR/$tdir/test1 || {
+				echo "mkdir fails"
+				break
+			}
+
+			touch $DIR/$tdir/test/a || {
+				echo "touch fails"
+				break;
+			}
+			mkdir $DIR/$tdir/test/b || {
+				echo "mkdir fails"
+				break;
+			}
+			rm -rf $DIR/$tdir/test || {
+				echo "rmdir fails"
+				ls -lR $DIR/$tdir
+				break
+			}
+
+			touch $DIR/$tdir/test1/a || {
+				echo "touch fails"
+				break;
+			}
+			mkdir $DIR/$tdir/test1/b || {
+				echo "mkdir fails"
+				break;
+			}
+
+			rm -rf $DIR/$tdir/test1 || {
+				echo "rmdir fails"
+				ls -lR $DIR/$tdir/test1
+				break
+			}
+		done
+	)&
+	mkdir_70d_pid=$!
+	echo "Started  $mkdir_70d_pid"
+
+	random_fail_mdt $MDSCOUNT $duration $mkdir_70d_pid
+	kill -0 $mkdir_70d_pid || error "mkdir/rmdir $mkdir_70d_pid stopped"
+
+	cleanup_70d
+	true
+}
+run_test 70d "mkdir/rmdir striped dir ${MDSCOUNT}mdts recovery"
+
+test_70e () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	lctl set_param debug=+ha
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i0 $DIR/$tdir/test_0
+	$LFS mkdir -i0 $DIR/$tdir/test_1
+	touch $DIR/$tdir/test_0/a
+	touch $DIR/$tdir/test_1/b
+	(
+	while true; do
+		mrename $DIR/$tdir/test_0/a $DIR/$tdir/test_1/b > /dev/null || {
+			echo "a->b fails"
+			break;
+		}
+
+		checkstat $DIR/$tdir/test_0/a && {
+			echo "a still exists"
+			break
+		}
+
+		checkstat $DIR/$tdir/test_1/b || {
+			echo "b still  exists"
+			break
+		}
+
+		touch $DIR/$tdir/test_0/a || {
+			echo "touch a fails"
+			break
+		}
+
+		mrename $DIR/$tdir/test_1/b $DIR/$tdir/test_0/a > /dev/null || {
+			echo "a->a fails"
+			break;
+		}
+	done
+	)&
+	rename_70e_pid=$!
+	stack_trap "kill -9 $rename_70e_pid" EXIT
+	echo "Started PID=$rename_70e_pid"
+
+	random_fail_mdt 2 $duration $rename_70e_pid
+	kill -0 $rename_70e_pid || error "rename $rename_70e_pid stopped"
+}
+run_test 70e "rename cross-MDT with random fails"
+
+test_70f_write_and_read(){
+	local srcfile=$1
+	local stopflag=$2
+	local client
+
+	echo "Write/read files in: '$DIR/$tdir', clients: '$CLIENTS' ..."
+	for client in ${CLIENTS//,/ }; do
+		[ -f $stopflag ] || return
+
+		local tgtfile=$DIR/$tdir/$tfile.$client
+		do_node $client dd $DD_OPTS bs=1M count=10 if=$srcfile \
+			of=$tgtfile 2>/dev/null ||
+			error "dd $DD_OPTS bs=1M count=10 if=$srcfile " \
+			      "of=$tgtfile failed on $client, rc=$?"
+	done
+
+	local prev_client=$(echo ${CLIENTS//,/ } | awk '{ print $NF }')
+	local index=0
+
+	for client in ${CLIENTS//,/ }; do
+		[ -f $stopflag ] || return
+
+		# flush client cache in case test is running on only one client
+		# do_node $client cancel_lru_locks osc
+		do_node $client $LCTL set_param ldlm.namespaces.*.lru_size=clear
+
+		tgtfile=$DIR/$tdir/$tfile.$client
+		local md5=$(do_node $prev_client "md5sum $tgtfile")
+		[ ${checksum[$index]// */} = ${md5// */} ] ||
+			error "$tgtfile: checksum doesn't match on $prev_client"
+		index=$((index + 1))
+		prev_client=$client
+	done
+}
+
+test_70f_loop(){
+	local srcfile=$1
+	local stopflag=$2
+	DD_OPTS=
+
+	mkdir -p $DIR/$tdir || error "cannot create $DIR/$tdir directory"
+	$LFS setstripe -c -1 $DIR/$tdir ||
+		error "cannot $LFS setstripe $DIR/$tdir"
+
+	touch $stopflag
+	while [ -f $stopflag ]; do
+		test_70f_write_and_read $srcfile $stopflag
+		# use direct IO and buffer cache in turns if loop
+		[ -n "$DD_OPTS" ] && DD_OPTS="" || DD_OPTS="oflag=direct"
+	done
+}
+
+test_70f_cleanup() {
+	trap 0
+	rm -f $TMP/$tfile.stop
+	do_nodes $CLIENTS rm -f $TMP/$tfile
+	rm -f $DIR/$tdir/$tfile.*
+}
+
+test_70f() {
+#	[ x$ost1failover_HOST = x$ost_HOST ] &&
+#		{ skip "Failover host not defined" && return; }
+#	[ -z "$CLIENTS" ] &&
+#		{ skip "CLIENTS are not specified." && return; }
+#	[ $CLIENTCOUNT -lt 2 ] &&
+#		{ skip "Need 2 or more clients, have $CLIENTCOUNT" && return; }
+
+	[[ $(lustre_version_code ost1) -lt $(version_code 2.9.53) ]] &&
+		skip "Need server version at least 2.9.53" && return
+
+	echo "mount clients $CLIENTS ..."
+	zconf_mount_clients $CLIENTS $MOUNT
+
+	local srcfile=$TMP/$tfile
+	local client
+	local index=0
+
+	trap test_70f_cleanup EXIT
+	# create a different source file local to each client node so we can
+	# detect if the file wasn't written out properly after failover
+	do_nodes $CLIENTS dd bs=1M count=10 if=/dev/urandom of=$srcfile \
+		2>/dev/null || error "can't create $srcfile on $CLIENTS"
+	for client in ${CLIENTS//,/ }; do
+		checksum[$index]=$(do_node $client "md5sum $srcfile")
+		index=$((index + 1))
+	done
+
+	local duration=120
+	[ "$SLOW" = "no" ] && duration=60
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	local stopflag=$TMP/$tfile.stop
+	test_70f_loop $srcfile $stopflag &
+	local pid=$!
+
+	local elapsed=0
+	local num_failovers=0
+	local start_ts=$SECONDS
+	while [ $elapsed -lt $duration ]; do
+		sleep 3
+		replay_barrier ost1
+		sleep 1
+		num_failovers=$((num_failovers + 1))
+		log "$TESTNAME failing OST $num_failovers times"
+		fail ost1
+		sleep 2
+		elapsed=$((SECONDS - start_ts))
+	done
+
+	rm -f $stopflag
+	wait $pid
+	test_70f_cleanup
+}
+run_test 70f "OSS O_DIRECT recovery with $CLIENTCOUNT clients"
+
+cleanup_71a() {
+	trap 0
+	kill -9 $mkdir_71a_pid
+}
+
+random_double_fail_mdt() {
+	local max_index=$1
+	local duration=$2
+	local monitor_pid=$3
+	local elapsed
+	local start_ts=$(date +%s)
+	local num_failovers=0
+	local fail_index
+	local second_index
+
+	elapsed=$(($(date +%s) - start_ts))
+	while [ $elapsed -lt $duration ]; do
+		fail_index=$((RANDOM%max_index + 1))
+		if [ $fail_index -eq $max_index ]; then
+			second_index=1
+		else
+			second_index=$((fail_index + 1))
+		fi
+		kill -0 $monitor_pid ||
+			error "$monitor_pid stopped"
+		sleep 120
+		replay_barrier mds$fail_index
+		replay_barrier mds$second_index
+		sleep 10
+		# Increment the number of failovers
+		num_failovers=$((num_failovers+1))
+		log "fail mds$fail_index mds$second_index $num_failovers times"
+		fail mds${fail_index},mds${second_index}
+		elapsed=$(($(date +%s) - start_ts))
+	done
+}
+
+test_71a () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local clients=${CLIENTS:-$HOSTNAME}
+	local rc=0
+
+	zconf_mount_clients $clients $MOUNT
+
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=180
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	mkdir -p $DIR/$tdir
+
+	local elapsed
+	local start_ts=$(date +%s)
+
+	trap cleanup_71a EXIT
+	(
+		while true; do
+			$LFS mkdir -i0 -c2 $DIR/$tdir/test
+			rmdir $DIR/$tdir/test
+		done
+	)&
+	mkdir_71a_pid=$!
+	echo "Started  $mkdir_71a_pid"
+
+	random_double_fail_mdt 2 $duration $mkdir_71a_pid
+	kill -0 $mkdir_71a_pid || error "mkdir/rmdir $mkdir_71a_pid stopped"
+
+	cleanup_71a
+	true
+}
+run_test 71a "mkdir/rmdir striped dir with 2 mdts recovery"
+
 test_73a() {
 	multiop_bg_pause $DIR/$tfile O_tSc ||
 		error "multiop_bg_pause $DIR/$tfile failed"
@@ -2116,8 +2553,8 @@ test_73b() {
 	rm -f $DIR/$tfile
 
 	replay_barrier $SINGLEMDS
-	#define OBD_FAIL_LDLM_REPLY       0x30c
-	do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
+	#define OBD_FAIL_MDS_LDLM_REPLY_NET       0x157
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000157"
 	fail $SINGLEMDS
 	kill -USR1 $pid
 	wait $pid || error "multiop pid failed"
@@ -2143,18 +2580,18 @@ test_74() {
 run_test 74 "Ensure applications don't fail waiting for OST recovery"
 
 remote_dir_check_80() {
-	local MDTIDX=1
+	local mdtidx=1
 	local diridx
 	local fileidx
 
-	diridx=$($GETSTRIPE -M $remote_dir) ||
-		error "$GETSTRIPE -M $remote_dir failed"
-	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+	diridx=$($LFS getstripe -m $remote_dir) ||
+		error "$LFS getstripe -m $remote_dir failed"
+	[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
 
 	createmany -o $remote_dir/f-%d 20 || error "creation failed"
-	fileidx=$($GETSTRIPE -M $remote_dir/f-1) ||
-		error "$GETSTRIPE -M $remote_dir/f-1 failed"
-	[ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
+	fileidx=$($LFS getstripe -m $remote_dir/f-1) ||
+		error "$LFS getstripe -m $remote_dir/f-1 failed"
+	[ $fileidx -eq $mdtidx ] || error "$fileidx != $mdtidx"
 
 	return 0
 }
@@ -2198,7 +2635,7 @@ test_80b() {
 	local remote_dir=$DIR/$tdir/remote_dir
 
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
-	#define OBD_FAIL_UPDATE_OBJ_NET_REP	0x1701
+	#define OBD_FAIL_OUT_UPDATE_NET_REP	0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
@@ -2217,6 +2654,9 @@ test_80b() {
 run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
 
 test_80c() {
+	[[ $mds1_FSTYPE = "zfs" ]] &&
+		[[ $MDS1_VERSION -lt $(version_code 2.12.51) ]] &&
+		skip "requires LU-10143 fix on MDS"
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
@@ -2227,7 +2667,7 @@ test_80c() {
 	local remote_dir=$DIR/$tdir/remote_dir
 
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
-	#define OBD_FAIL_UPDATE_OBJ_NET_REP	0x1701
+	#define OBD_FAIL_OUT_UPDATE_NET_REP	0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
@@ -2247,12 +2687,15 @@ test_80c() {
 run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
 
 test_80d() {
+	[[ $mds1_FSTYPE = "zfs" ]] &&
+		[[ $MDS1_VERSION -lt $(version_code 2.12.51) ]] &&
+		skip "requires LU-10143 fix on MDS"
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	local MDTIDX=1
 	local remote_dir=$DIR/$tdir/remote_dir
 
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
-	#define OBD_FAIL_UPDATE_OBJ_NET_REP	0x1701
+	#define OBD_FAIL_OUT_UPDATE_NET_REP	0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	$LFS mkdir -i $MDTIDX $remote_dir &
 	local CLIENT_PID=$!
@@ -2409,7 +2852,7 @@ test_81a() {
 	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
 
 	touch $remote_dir || error "touch $remote_dir failed"
-	# OBD_FAIL_OBJ_UPDATE_NET_REP       0x1701
+	# OBD_FAIL_OUT_UPDATE_NET_REP       0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
@@ -2419,7 +2862,7 @@ test_81a() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2439,7 +2882,7 @@ test_81b() {
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
 
-	# OBD_FAIL_OBJ_UPDATE_NET_REP       0x1701
+	# OBD_FAIL_OUT_UPDATE_NET_REP       0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
@@ -2449,7 +2892,7 @@ test_81b() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2470,7 +2913,7 @@ test_81c() {
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
 
-	# OBD_FAIL_OBJ_UPDATE_NET_REP       0x1701
+	# OBD_FAIL_OUT_UPDATE_NET_REP       0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
@@ -2482,7 +2925,7 @@ test_81c() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2498,7 +2941,7 @@ test_81d() {
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
 
-	# OBD_FAIL_OBJ_UPDATE_NET_REP       0x1701
+	# OBD_FAIL_OUT_UPDATE_NET_REP       0x1701
 	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
 	rmdir $remote_dir &
 	local CLIENT_PID=$!
@@ -2509,7 +2952,7 @@ test_81d() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2541,7 +2984,7 @@ test_81e() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2572,7 +3015,7 @@ test_81f() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2605,7 +3048,7 @@ test_81g() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2632,7 +3075,7 @@ test_81h() {
 
 	wait $CLIENT_PID || error "rm remote dir failed"
 
-	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+	stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 
@@ -2660,7 +3103,7 @@ test_85a() { #bug 16774
 	done
 
 	lov_id=$(lctl dl | grep "clilov")
-	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
+	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
 	count=$(lctl get_param -n \
 		ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
 	echo "before recovery: unused locks count = $count"
@@ -2678,33 +3121,35 @@ test_85a() { #bug 16774
 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
 
 test_85b() { #bug 16774
+	rm -rf $DIR/$tdir
+	mkdir $DIR/$tdir
+
 	lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
 
-	do_facet mgs $LCTL pool_new $FSNAME.$TESTNAME ||
-		error "unable to create pool $TESTNAME"
-	do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
-		error "unable to add pool $TESTNAME"
+	if ! combined_mgs_mds ; then
+		mount_mgs_client
+	fi
 
-	$SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
+	$LFS setstripe -c 1 -i 0 $DIR/$tdir
 
 	for i in $(seq 100); do
-		dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 \
+		dd if=/dev/urandom of=$DIR/$tdir/$tfile-$i bs=4096 \
 			count=32 >/dev/null 2>&1
 	done
 
 	cancel_lru_locks osc
 
 	for i in $(seq 100); do
-		dd if=$DIR/$tfile-$i of=/dev/null bs=4096 \
+		dd if=$DIR/$tdir/$tfile-$i of=/dev/null bs=4096 \
 			count=32 >/dev/null 2>&1
 	done
 
 	lov_id=$(lctl dl | grep "clilov")
-	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
-	count=$(lctl get_param \
-		-n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
+	addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
+	count=$(lctl get_param -n \
+			  ldlm.namespaces.*OST0000*$addr.lock_unused_count)
 	echo "before recovery: unused locks count = $count"
-	[ $count != 0 ] || error "unused locks ($count) should be zero"
+	[ $count -ne 0 ] || error "unused locks ($count) should be zero"
 
 	fail ost1
 
@@ -2712,14 +3157,15 @@ test_85b() { #bug 16774
 		 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
 	echo "after recovery: unused locks count = $count2"
 
-	do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 ||
-		error "unable to remove pool $TESTNAME"
-	do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
-		error "unable to destroy the pool $TESTNAME"
+	if ! combined_mgs_mds ; then
+		umount_mgs_client
+	fi
 
 	if [ $count2 -ge $count ]; then
 		error "unused locks are not canceled"
 	fi
+
+	rm -rf $DIR/$tdir
 }
 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
 
@@ -2733,11 +3179,11 @@ test_86() {
 }
 run_test 86 "umount server after clear nid_stats should not hit LBUG"
 
-test_87() {
+test_87a() {
 	do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
 
 	replay_barrier ost1
-	$SETSTRIPE -i 0 -c 1 $DIR/$tfile
+	$LFS setstripe -i 0 -c 1 $DIR/$tfile
 	dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
 		error "dd to $DIR/$tfile failed"
 	cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
@@ -2749,13 +3195,13 @@ test_87() {
 		error "New checksum $cksum2 does not match original $cksum"
 	fi
 }
-run_test 87 "write replay"
+run_test 87a "write replay"
 
 test_87b() {
 	do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
 
 	replay_barrier ost1
-	$SETSTRIPE -i 0 -c 1 $DIR/$tfile
+	$LFS setstripe -i 0 -c 1 $DIR/$tfile
 	dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
 		error "dd to $DIR/$tfile failed"
 	sleep 1 # Give it a chance to flush dirty data
@@ -2775,7 +3221,7 @@ test_88() { #bug 17485
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	mkdir -p $TMP/$tdir || error "mkdir $TMP/$tdir failed"
 
-	$SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
+	$LFS setstripe -i 0 -c 1 $DIR/$tdir || error "$LFS setstripe failed"
 
 	replay_barrier ost1
 	replay_barrier $SINGLEMDS
@@ -2853,15 +3299,24 @@ test_88() { #bug 17485
 }
 run_test 88 "MDS should not assign same objid to different files "
 
+function calc_osc_kbytes_used() {
+	local kbtotal=$(calc_osc_kbytes kbytestotal)
+	local kbfree=$(calc_osc_kbytes kbytesfree)
+	echo $((kbtotal-kbfree))
+}
+
 test_89() {
 	cancel_lru_locks osc
 	mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
 	rm -f $DIR/$tdir/$tfile
-	wait_mds_ost_sync
-	wait_delete_completed
-	BLOCKS1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
-	$SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
-	dd if=/dev/zero bs=1M count=10 of=$DIR/$tdir/$tfile
+	wait_mds_ost_sync || error "initial MDS-OST sync timed out"
+	wait_delete_completed || error "initial wait delete timed out"
+	local blocks1=$(calc_osc_kbytes_used)
+	local write_size=$(fs_log_size)
+
+	$LFS setstripe -i 0 -c 1 $DIR/$tdir/$tfile
+	[ $write_size -lt 1024 ] && write_size=1024
+	dd if=/dev/zero bs=${write_size}k count=10 of=$DIR/$tdir/$tfile
 	sync
 	stop ost1
 	facet_failover $SINGLEMDS
@@ -2870,23 +3325,30 @@ test_89() {
 	mount_facet ost1
 	zconf_mount $(hostname) $MOUNT || error "mount fails"
 	client_up || error "client_up failed"
-	wait_mds_ost_sync
-	wait_delete_completed
-	BLOCKS2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
-	[ $((BLOCKS2 - BLOCKS1)) -le 4  ] ||
-		error $((BLOCKS2 - BLOCKS1)) blocks leaked
-}
 
+	# wait for the remounted client to connect to ost1
+	local target=$(get_osc_import_name client ost1)
+	wait_import_state "FULL" "osc.${target}.ost_server_uuid" \
+		$(max_recovery_time)
+
+	wait_mds_ost_sync || error "MDS-OST sync timed out"
+	wait_delete_completed || error "wait delete timed out"
+	local blocks2=$(calc_osc_kbytes_used)
+
+	[ $((blocks2 - blocks1)) -le $(fs_log_size)  ] ||
+		error $((blocks2 - blocks1)) blocks leaked
+}
 run_test 89 "no disk space leak on late ost connection"
 
 cleanup_90 () {
-    local facet=$1
-    trap 0
-    reboot_facet $facet
-    change_active $facet
-    wait_for_facet $facet
-    mount_facet $facet || error "Restart of $facet failed"
-    clients_up
+	local facet=$1
+
+	trap 0
+	reboot_facet $facet
+	change_active $facet
+	wait_for_facet $facet
+	mount_facet $facet || error "Restart of $facet failed"
+	clients_up
 }
 
 test_90() { # bug 19494
@@ -2900,30 +3362,37 @@ test_90() { # bug 19494
             return 0
         fi
     fi
+	# ensure all OSTs are active to allow allocations
+	wait_osts_up
 
 	mkdir $dir || error "mkdir $dir failed"
 
 	echo "Create the files"
 
-    # file "f${index}" striped over 1 OST
-    # file "all" striped over all OSTs
+	# file "f${index}" striped over 1 OST
+	# file "all" striped over all OSTs
 
-    $SETSTRIPE -c $OSTCOUNT $dir/all ||
-        error "setstripe failed to create $dir/all"
+	$LFS setstripe -c $OSTCOUNT $dir/all ||
+		error "setstripe failed to create $dir/all"
 
-    for (( i=0; i<$OSTCOUNT; i++ )); do
-        local f=$dir/f$i
-        $SETSTRIPE -i $i -c 1 $f || error "$SETSTRIPE failed to create $f"
+	for ((i = 0; i < $OSTCOUNT; i++)); do
+		local f=$dir/f$i
 
-        # confirm setstripe actually created the stripe on the requested OST
-        local uuid=$(ostuuid_from_index $i)
-        for file in f$i all; do
-            if [[ $dir/$file != $($LFS find --obd $uuid --name $file $dir) ]]; then
-		$GETSTRIPE $dir/$file
-		error wrong stripe: $file, uuid: $uuid
-            fi
-        done
-    done
+		$LFS setstripe -i $i -c 1 $f ||
+			error "$LFS setstripe failed to create $f"
+
+		# confirm setstripe actually created stripe on requested OST
+		local uuid=$(ostuuid_from_index $i)
+
+		for file in f$i all; do
+			local found=$($LFS find --obd $uuid --name $file $dir)
+
+			if [[ $dir/$file != $found ]]; then
+				$LFS getstripe $dir/$file
+				error "wrong stripe: $file, uuid: $uuid"
+			fi
+		done
+	done
 
 	# Before failing an OST, get its obd name and index
 	local varsvc=${ostfail}_svc
@@ -2959,9 +3428,9 @@ test_90() { # bug 19494
     [[ $(echo $list | wc -w) -eq 2 ]] ||
         error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
 
-    echo "Check getstripe: $GETSTRIPE -r --obd $obd"
-    list=$($GETSTRIPE -r --obd $obd $dir)
-    echo "$list"
+	echo "Check getstripe: $LFS getstripe -r --obd $obd"
+	list=$($LFS getstripe -r --obd $obd $dir)
+	echo "$list"
     for file in all f$index; do
         echo "$list" | grep $dir/$file ||
             error_noexit "lfs getsripe does not report the affected $obd for $file"
@@ -2971,7 +3440,7 @@ test_90() { # bug 19494
 }
 run_test 90 "lfs find identifies the missing striped file segments"
 
-test_93() {
+test_93a() {
 	local server_version=$(lustre_version_code $SINGLEMDS)
 		[[ $server_version -ge $(version_code 2.6.90) ]] ||
 		[[ $server_version -ge $(version_code 2.5.4) &&
@@ -2980,8 +3449,8 @@ test_93() {
 
 	cancel_lru_locks osc
 
-	$SETSTRIPE -i 0 -c 1 $DIR/$tfile ||
-		error "$SETSTRIPE  $DIR/$tfile failed"
+	$LFS setstripe -i 0 -c 1 $DIR/$tfile ||
+		error "$LFS setstripe  $DIR/$tfile failed"
 	dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 ||
 		error "dd to $DIR/$tfile failed"
 	#define OBD_FAIL_TGT_REPLAY_RECONNECT     0x715
@@ -2993,7 +3462,28 @@ test_93() {
 	do_facet ost1 "$LCTL set_param fail_loc=0x715"
 	fail ost1
 }
-run_test 93 "replay + reconnect"
+run_test 93a "replay + reconnect"
+
+test_93b() {
+	local server_version=$(lustre_version_code $SINGLEMDS)
+		[[ $server_version -ge $(version_code 2.7.90) ]] ||
+		{ skip "Need MDS version 2.7.90+"; return; }
+
+	cancel_lru_locks mdc
+
+	createmany -o $DIR/$tfile 20 ||
+			error "createmany -o $DIR/$tfile failed"
+
+	#define OBD_FAIL_TGT_REPLAY_RECONNECT     0x715
+	# We need to emulate a state that MDT is waiting for other clients
+	# not completing the recovery. Final ping is queued, but reply will be
+	# sent on the recovery completion. It is done by sleep before
+	# processing final pings
+	do_facet mds1 "$LCTL set_param fail_val=80"
+	do_facet mds1 "$LCTL set_param fail_loc=0x715"
+	fail mds1
+}
+run_test 93b "replay + reconnect on mds"
 
 striped_dir_check_100() {
 	local striped_dir=$DIR/$tdir/striped_dir
@@ -3088,6 +3578,245 @@ test_101() { #LU-5648
 }
 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
 
+test_102a() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	# drop request on MDT to force resend
+	#define OBD_FAIL_MDS_REINT_MULTI_NET 0x159
+	do_facet $facet "$LCTL set_param fail_loc=0x159"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+	do_facet $facet "$LCTL set_param fail_loc=0"
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102a "check resend (request lost) with multiple modify RPCs in flight"
+
+test_102b() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	# drop reply on MDT to force reconstruction
+	#define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
+	do_facet $facet "$LCTL set_param fail_loc=0x15a"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+	do_facet $facet "$LCTL set_param fail_loc=0"
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102b "check resend (reply lost) with multiple modify RPCs in flight"
+
+test_102c() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	replay_barrier $facet
+
+	# drop reply on MDT
+	#define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
+	do_facet $facet "$LCTL set_param fail_loc=0x15a"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+	do_facet $facet "$LCTL set_param fail_loc=0"
+
+	# fail MDT
+	fail $facet
+
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102c "check replay w/o reconstruction with multiple mod RPCs in flight"
+
+test_102d() {
+	local idx
+	local facet
+	local num
+	local i
+	local pids pid
+
+	[[ $(lctl get_param mdc.*.import |
+	     grep "connect_flags:.*multi_mod_rpc") ]] ||
+		{ skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
+
+	$LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
+	facet="mds$((0x$idx + 1))"
+
+	# get current value of max_mod_rcps_in_flight
+	num=$($LCTL get_param -n \
+		mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
+	# set default value if client does not support multi mod RPCs
+	[ -z "$num" ] && num=1
+
+	echo "creating $num files ..."
+	umask 0022
+	for i in $(seq $num); do
+		touch $DIR/$tdir/file-$i
+	done
+
+	# drop reply on MDT
+	#define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
+	do_facet $facet "$LCTL set_param fail_loc=0x15a"
+	echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
+	for i in $(seq $num); do
+		chmod 0600 $DIR/$tdir/file-$i &
+		pids="$pids $!"
+	done
+	sleep 1
+
+	# write MDT transactions to disk
+	do_facet $facet "sync; sync; sync"
+
+	do_facet $facet "$LCTL set_param fail_loc=0"
+
+	# fail MDT
+	fail $facet
+
+	for pid in $pids; do
+		wait $pid || error "chmod failed"
+	done
+	echo "done ($(date +%H:%M:%S))"
+
+	# check chmod succeed
+	for i in $(seq $num); do
+		checkstat -vp 0600 $DIR/$tdir/file-$i
+	done
+
+	rm -rf $DIR/$tdir
+}
+run_test 102d "check replay & reconstruction with multiple mod RPCs in flight"
+
+test_103() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	local mds_version=$(lustre_version_code $SINGLEMDS)
+	[[ $mds_version -gt $(version_code 2.8.54) ]] ||
+		{ skip "Need MDS version 2.8.54+"; return; }
+
+#define OBD_FAIL_MDS_TRACK_OVERFLOW 0x162
+	do_facet mds1 $LCTL set_param fail_loc=0x80000162
+
+	mkdir -p $DIR/$tdir
+	createmany -o $DIR/$tdir/t- 30 ||
+		error "create files on remote directory failed"
+	sync
+	rm -rf $DIR/$tdir/t-*
+	sync
+#MDS should crash with tr->otr_next_id overflow
+	fail mds1
+}
+run_test 103 "Check otr_next_id overflow"
+
+
 check_striped_dir_110()
 {
 	$CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
@@ -3099,6 +3828,9 @@ check_striped_dir_110()
 
 test_110a() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3118,6 +3850,9 @@ run_test 110a "DNE: create striped dir, fail MDT1"
 
 test_110b() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3141,6 +3876,9 @@ run_test 110b "DNE: create striped dir, fail MDT1 and client"
 
 test_110c() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3161,6 +3899,9 @@ run_test 110c "DNE: create striped dir, fail MDT2"
 
 test_110d() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3184,6 +3925,9 @@ run_test 110d "DNE: create striped dir, fail MDT2 and client"
 
 test_110e() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3208,6 +3952,9 @@ run_test 110e "DNE: create striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
 
 test_110f() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3217,7 +3964,7 @@ test_110f() {
 	replay_barrier mds1
 	replay_barrier mds2
 	$LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
-	fail mds1,mds2
+	fail mds2,mds1
 
 	check_striped_dir_110 || error "check striped_dir failed"
 
@@ -3229,6 +3976,9 @@ run_test 110f "DNE: create striped dir, fail MDT1/MDT2"
 
 test_110g() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3253,6 +4003,9 @@ run_test 110g "DNE: create striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
 
 test_111a() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3272,6 +4025,9 @@ run_test 111a "DNE: unlink striped dir, fail MDT1"
 
 test_111b() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3294,6 +4050,9 @@ run_test 111b "DNE: unlink striped dir, fail MDT2"
 
 test_111c() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3316,6 +4075,9 @@ run_test 111c "DNE: unlink striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
 
 test_111d() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3339,6 +4101,9 @@ run_test 111d "DNE: unlink striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
 
 test_111e() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3358,6 +4123,9 @@ run_test 111e "DNE: unlink striped dir, uncommit on MDT2, fail MDT1/MDT2"
 
 test_111f() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3377,6 +4145,9 @@ run_test 111f "DNE: unlink striped dir, uncommit on MDT1, fail MDT1/MDT2"
 
 test_111g() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3419,6 +4190,9 @@ test_112_check() {
 
 test_112a() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3438,6 +4212,9 @@ run_test 112a "DNE: cross MDT rename, fail MDT1"
 
 test_112b() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3458,6 +4235,9 @@ run_test 112b "DNE: cross MDT rename, fail MDT2"
 
 test_112c() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3478,6 +4258,9 @@ run_test 112c "DNE: cross MDT rename, fail MDT3"
 
 test_112d() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3498,6 +4281,9 @@ run_test 112d "DNE: cross MDT rename, fail MDT4"
 
 test_112e() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3519,6 +4305,9 @@ run_test 112e "DNE: cross MDT rename, fail MDT1 and MDT2"
 
 test_112f() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3540,6 +4329,9 @@ run_test 112f "DNE: cross MDT rename, fail MDT1 and MDT3"
 
 test_112g() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3561,6 +4353,9 @@ run_test 112g "DNE: cross MDT rename, fail MDT1 and MDT4"
 
 test_112h() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3582,6 +4377,9 @@ run_test 112h "DNE: cross MDT rename, fail MDT2 and MDT3"
 
 test_112i() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3603,6 +4401,9 @@ run_test 112i "DNE: cross MDT rename, fail MDT2 and MDT4"
 
 test_112j() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3624,6 +4425,9 @@ run_test 112j "DNE: cross MDT rename, fail MDT3 and MDT4"
 
 test_112k() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3646,6 +4450,9 @@ run_test 112k "DNE: cross MDT rename, fail MDT1,MDT2,MDT3"
 
 test_112l() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3668,6 +4475,9 @@ run_test 112l "DNE: cross MDT rename, fail MDT1,MDT2,MDT4"
 
 test_112m() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3690,6 +4500,9 @@ run_test 112m "DNE: cross MDT rename, fail MDT1,MDT3,MDT4"
 
 test_112n() {
 	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3712,6 +4525,9 @@ run_test 112n "DNE: cross MDT rename, fail MDT2,MDT3,MDT4"
 
 test_115() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	([ $FAILURE_MODE == "HARD" ] &&
 		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
 		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
@@ -3758,7 +4574,7 @@ test_116a() {
 
 	# OBD_FAIL_SPLIT_UPDATE_REC       0x1702
 	do_facet mds1 "lctl set_param fail_loc=0x80001702"
-	$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
+	$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir
 
 	fail mds1
 	$CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
@@ -3783,7 +4599,7 @@ test_116b() {
 
 	# OBD_FAIL_SPLIT_UPDATE_REC       0x1702
 	do_facet mds2 "lctl set_param fail_loc=0x80001702"
-	$LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
+	$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir
 
 	fail mds2
 	$CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
@@ -3791,6 +4607,313 @@ test_116b() {
 }
 run_test 116b "large update log slave MDT recovery"
 
+test_117() {
+	[ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local index
+	local mds_indexs
+
+	mkdir -p $DIR/$tdir
+	$LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/remote_dir
+	$LFS setdirstripe -i1 -c$MDSCOUNT $DIR/$tdir/remote_dir_1
+	sleep 2
+
+	# Let's set rdonly on all MDTs, so client will send
+	# replay requests on all MDTs and replay these requests
+	# at the same time. This test will verify the recovery
+	# will not be deadlock in this case, LU-7531.
+	for ((index = 0; index < $((MDSCOUNT)); index++)); do
+		replay_barrier mds$((index + 1))
+		if [ -z $mds_indexs ]; then
+			mds_indexs="${mds_indexs}mds$((index+1))"
+		else
+			mds_indexs="${mds_indexs},mds$((index+1))"
+		fi
+	done
+
+	rm -rf $DIR/$tdir/remote_dir
+	rm -rf $DIR/$tdir/remote_dir_1
+
+	fail $mds_indexs
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 117 "DNE: cross MDT unlink, fail MDT1 and MDT2"
+
+test_118() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "Do not support large update log before 2.7.64" &&
+		return 0
+
+	mkdir -p $DIR/$tdir
+
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir ||
+		error "setdirstripe fails"
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir1 ||
+		error "setdirstripe fails 1"
+	rm -rf $DIR/$tdir/striped_dir* || error "rmdir fails"
+
+	# OBD_FAIL_INVALIDATE_UPDATE       0x1705
+	do_facet mds1 "lctl set_param fail_loc=0x1705"
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
+	do_facet mds1 "lctl set_param fail_loc=0x0"
+
+	replay_barrier mds1
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir
+	$LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
+	fail mds1
+
+	true
+}
+run_test 118 "invalidate osp update will not cause update log corruption"
+
+test_119() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "Do not support large update log before 2.7.64" &&
+		return 0
+	local stripe_count
+	local hard_timeout=$(do_facet mds1 \
+		"lctl get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard")
+
+	local clients=${CLIENTS:-$HOSTNAME}
+	local time_min=$(recovery_time_min)
+
+	mkdir -p $DIR/$tdir
+	mkdir $DIR/$tdir/tmp
+	rmdir $DIR/$tdir/tmp
+
+	replay_barrier mds1
+	mkdir $DIR/$tdir/dir_1
+	for ((i = 0; i < 20; i++)); do
+		$LFS setdirstripe -i0 -c2 $DIR/$tdir/stripe_dir-$i
+	done
+
+	stop mds1
+	change_active mds1
+	wait_for_facet mds1
+
+	#define OBD_FAIL_TGT_REPLAY_DELAY  0x714
+	do_facet mds1 $LCTL set_param fail_loc=0x80000714
+	#sleep (timeout + 5), so mds will evict the client exports,
+	#but DNE update recovery will keep going.
+	do_facet mds1 $LCTL set_param fail_val=$((time_min + 5))
+
+	mount_facet mds1 "-o recovery_time_hard=$time_min"
+
+	wait_clients_import_state "$clients" mds1 FULL
+
+	clients_up || clients_up || error "failover df: $?"
+
+	#revert back the hard timeout
+	do_facet mds1 $LCTL set_param \
+		mdt.$FSNAME-MDT0000.recovery_time_hard=$hard_timeout
+
+	for ((i = 0; i < 20; i++)); do
+		stripe_count=$($LFS getdirstripe -c $DIR/$tdir/stripe_dir-$i)
+		[ $stripe_count == 2 ] || {
+			error "stripe_dir-$i creation replay fails"
+			break
+		}
+	done
+}
+run_test 119 "timeout of normal replay does not cause DNE replay fails  "
+
+test_120() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+		skip "Do not support large update log before 2.7.64" &&
+		return 0
+
+	mkdir $DIR/$tdir
+	replay_barrier_nosync mds1
+	for ((i = 0; i < 20; i++)); do
+		mkdir $DIR/$tdir/dir-$i || {
+			error "create dir-$i fails"
+			break
+		}
+		$LFS setdirstripe -i0 -c2 $DIR/$tdir/stripe_dir-$i || {
+			error "create stripe_dir-$i fails"
+			break
+		}
+	done
+
+	fail_abort mds1
+
+	for ((i = 0; i < 20; i++)); do
+		[ ! -e "$DIR/$tdir/dir-$i" ] || {
+			error "dir-$i still exists"
+			break
+		}
+		[ ! -e "$DIR/$tdir/stripe_dir-$i" ] || {
+			error "stripe_dir-$i still exists"
+			break
+		}
+	done
+}
+run_test 120 "DNE fail abort should stop both normal and DNE replay"
+
+test_121() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.90) ] &&
+		skip "Don't support it before 2.11" &&
+		return 0
+
+	local at_max_saved=$(at_max_get mds)
+
+	touch $DIR/$tfile || error "touch $DIR/$tfile failed"
+	cancel_lru_locks mdc
+
+	multiop_bg_pause $DIR/$tfile s_s || error "multiop $DIR/$tfile failed"
+	mpid=$!
+
+	lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
+
+	stop mds1
+	change_active mds1
+	wait_for_facet mds1
+
+	#define OBD_FAIL_TGT_RECOVERY_REQ_RACE	0x721
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x721 fail_val=0"
+	at_max_set 0 mds
+
+	mount_facet mds1
+	wait_clients_import_state "$clients" mds1 FULL
+	clients_up || clients_up || error "failover df: $?"
+
+	kill -USR1 $mpid
+	wait $mpid || error "multiop_bg_pause pid failed"
+
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
+	lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
+	at_max_set $at_max_saved mds
+	rm -f $DIR/$tfile
+}
+run_test 121 "lock replay timed out and race"
+
+test_130a() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.90) ] &&
+		skip "Do not support Data-on-MDT before 2.11"
+
+	replay_barrier $SINGLEMDS
+	$LFS setstripe -E 1M -L mdt -E EOF -c 2 $DIR/$tfile
+	fail $SINGLEMDS
+
+	[ $($LFS getstripe -L $DIR/$tfile) == "mdt" ] ||
+		error "Fail to replay DoM file creation"
+}
+run_test 130a "DoM file create (setstripe) replay"
+
+test_130b() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.90) ] &&
+		skip "Do not support Data-on-MDT before 2.11"
+
+	mkdir $DIR/$tdir
+	$LFS setstripe -E 1M -L mdt -E EOF -c 2 $DIR/$tdir
+	replay_barrier $SINGLEMDS
+	touch $DIR/$tdir/$tfile
+	fail $SINGLEMDS
+
+	[ $($LFS getstripe -L $DIR/$tdir/$tfile) == "mdt" ] ||
+		error "Fail to replay DoM file creation"
+}
+run_test 130b "DoM file create (inherited) replay"
+
+test_131a() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.90) ] &&
+		skip "Do not support Data-on-MDT before 2.11"
+
+	$LFS setstripe -E 1M -L mdt -E EOF -c 2 $DIR/$tfile
+	replay_barrier $SINGLEMDS
+	echo "dom_data" | dd of=$DIR/$tfile bs=8 count=1
+	# lock is not canceled and will be replayed
+	fail $SINGLEMDS
+
+	[ $(cat $DIR/$tfile) == "dom_data" ] ||
+		error "Wrong file content after failover"
+}
+run_test 131a "DoM file write lock replay"
+
+test_131b() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.90) ] &&
+		skip "Do not support Data-on-MDT before 2.11"
+
+	$LFS setstripe -E 1M -L mdt -E EOF -c 2 $DIR/$tfile
+	replay_barrier $SINGLEMDS
+	echo "dom_data" | dd of=$DIR/$tfile bs=8 count=1
+	cancel_lru_locks mdc
+
+	fail $SINGLEMDS
+
+	[ $(cat $DIR/$tfile) == "dom_data" ] ||
+		error "Wrong file content after failover"
+}
+run_test 131b "DoM file write replay"
+
+test_132a() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.9.90) ] &&
+		skip "Do not support PFL files before 2.10"
+
+	$LFS setstripe -E 1M -c 1 -E EOF -c 2 $DIR/$tfile
+	replay_barrier $SINGLEMDS
+	# write over the first component size cause next component instantiation
+	dd if=/dev/urandom of=$DIR/$tfile bs=1M count=1 seek=1 ||
+		error "dd to $DIR/$tfile failed"
+	lfs getstripe $DIR/$tfile
+
+	cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
+	$LFS getstripe -I2 $DIR/$tfile | grep -q lmm_objects ||
+		error "Component #1 was not instantiated"
+
+	fail $SINGLEMDS
+
+	lfs getstripe $DIR/$tfile
+	$LFS getstripe -I2 $DIR/$tfile | grep -q lmm_objects ||
+		error "Component #1 instantiation was not replayed"
+	cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
+	if [ $cksum != $cksum2 ] ; then
+		error_noexit "New cksum $cksum2 does not match original $cksum"
+	fi
+}
+run_test 132a "PFL new component instantiate replay"
+
+test_133() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir || error "mkdir $DIR/$tdir failed"
+	$LFS mkdir -i 1 $remote_dir
+
+	umount $MOUNT
+	do_facet mds2 $LCTL set_param seq.srv*MDT0001.space=clear
+
+	zconf_mount $(hostname) $MOUNT
+	client_up || return 1
+
+	#define OBD_FAIL_MDS_ALL_REQUEST_NET     0x123
+	# SEQ_QUERY                       = 700
+	do_facet mds1 $LCTL set_param fail_val=700 fail_loc=0x80000123
+	cp /etc/hosts $remote_dir/file &
+	local pid=$!
+	sleep 1
+
+	fail_nodf mds1
+
+	wait $pid || error "cp failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 133 "check resend of ongoing requests for lwp during failover"
 
 complete $SECONDS
 check_and_cleanup_lustre