X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=586fbce5bf735806b0639dbe38722eb58f4c4ba3;hp=c20f7f7b24f67e4264ee72cdcf99da693b996c10;hb=62b57e34d9a0df1ce4b82650d7e328db5d048b39;hpb=c75aa6c74cd86c69de893395735a6725571f11f4

diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh
index c20f7f7..586fbce 100755
--- a/lustre/tests/sanityn.sh
+++ b/lustre/tests/sanityn.sh
@@ -3,8 +3,9 @@
 set -e
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test: 9977/LU-7105 LU-9452
-ALWAYS_EXCEPT="                28           29      $SANITYN_EXCEPT"
+# bug number for skipped test: 9977/LU-7105
+#		LU-7105
+ALWAYS_EXCEPT=" 28	$SANITYN_EXCEPT"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 SRCDIR=$(dirname $0)
@@ -48,6 +49,8 @@ TRACE=${TRACE:-""}
 
 check_and_setup_lustre
 
+OSC=${OSC:-"osc"}
+
 assert_DIR
 rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}*
 
@@ -82,7 +85,8 @@ test_2a() {
 	touch $DIR1/f2a
 	ls -l $DIR2/f2a
 	chmod 777 $DIR2/f2a
-	$CHECKSTAT -t file -p 0777 $DIR1/f2a || error
+	$CHECKSTAT -t file -p 0777 $DIR1/f2a ||
+		error "Either not file type or perms not 0777"
 }
 run_test 2a "check cached attribute updates on 2 mtpt's ========"
 
@@ -90,20 +94,23 @@ test_2b() {
 	touch $DIR1/f2b
 	ls -l $DIR2/f2b
 	chmod 777 $DIR1/f2b
-	$CHECKSTAT -t file -p 0777 $DIR2/f2b || error
+	$CHECKSTAT -t file -p 0777 $DIR2/f2b ||
+		error "Either not file type or perms not 0777"
 }
 run_test 2b "check cached attribute updates on 2 mtpt's ========"
 
 # NEED TO SAVE ROOT DIR MODE
 test_2c() {
 	chmod 777 $DIR1
-	$CHECKSTAT -t dir -p 0777 $DIR2 || error
+	$CHECKSTAT -t dir -p 0777 $DIR2 ||
+		error "Either not dir type or perms not 0777"
 }
 run_test 2c "check cached attribute updates on 2 mtpt's root ==="
 
 test_2d() {
 	chmod 755 $DIR1
-	$CHECKSTAT -t dir -p 0755 $DIR2 || error
+	$CHECKSTAT -t dir -p 0755 $DIR2 ||
+		error "Either not file type or perms not 0775"
 }
 run_test 2d "check cached attribute updates on 2 mtpt's root ==="
 
@@ -112,7 +119,8 @@ test_2e() {
         ls -l $DIR1
         ls -l $DIR2
         chmod 777 $DIR1
-        $RUNAS dd if=/dev/zero of=$DIR2/$tfile count=1 || error
+		$RUNAS dd if=/dev/zero of=$DIR2/$tfile count=1 ||
+			error "dd failed"
 }
 run_test 2e "check chmod on root is propagated to others"
 
@@ -159,7 +167,7 @@ test_2g() {
 	local block2=$(stat $DIR2/$tfile | awk '/Blocks/ {print $4} ')
 	echo "$DIR1/$tfile has $block1 blocks"
 	echo "$DIR2/$tfile has $block2 blocks"
-	[ $block1 -eq $block2 ] || error
+	[ $block1 -eq $block2 ] || error "$block1 not equal to $block2"
 }
 run_test 2g "check blocks update on sync write"
 
@@ -179,7 +187,8 @@ run_test 4 "fstat validation on multiple mount points =========="
 test_5() {
 	mcreate $DIR1/f5
 	$TRUNCATE $DIR2/f5 100
-	$CHECKSTAT -t file -s 100 $DIR1/f5 || error
+	$CHECKSTAT -t file -s 100 $DIR1/f5 ||
+		error "Either not file type or size not equal to 100 bytes"
 	rm $DIR1/f5
 }
 run_test 5 "create a file on one mount, truncate it on the other"
@@ -203,7 +212,7 @@ test_8() {
 }
 run_test 8 "remove of open special file on other node =========="
 
-test_9() {
+test_9a() {
 	MTPT=1
 	local dir
 	> $DIR2/f9
@@ -215,7 +224,23 @@ test_9() {
 	[ "`cat $DIR1/f9`" = "abcdefghijkl" ] || \
 		error "`od -a $DIR1/f9` != abcdefghijkl"
 }
-run_test 9 "append of file with sub-page size on multiple mounts"
+run_test 9a "append of file with sub-page size on multiple mounts"
+
+#LU-10681 - tiny writes & appending to sparse striped file
+test_9b() {
+	[[ $OSTCOUNT -ge 2 ]] || { skip "needs >= 2 OSTs"; return; }
+
+	$LFS setstripe -c 2 -S 1M $DIR/$tfile
+	echo "foo" >> $DIR/$tfile
+	dd if=/dev/zero of=$DIR2/$tfile bs=1M count=1 seek=1 conv=notrunc ||
+		error "sparse dd $DIR2/$tfile failed"
+	echo "foo" >> $DIR/$tfile
+
+	data=$(dd if=$DIR2/$tfile bs=1 count=3 skip=$((2 * 1048576)) conv=notrunc)
+	echo "Data read (expecting 'foo')": $data
+	[ "$data" = "foo" ] || error "append to sparse striped file failed"
+}
+run_test 9b "append to striped sparse file"
 
 test_10a() {
 	MTPT=1
@@ -258,7 +283,7 @@ test_11() {
 	$DIR2/d11/f
 	RC=$?
 	kill -USR1 $MULTIPID
-	wait $MULTIPID || error
+	wait $MULTIPID || error "wait for PID $MULTIPID failed"
 	[ $RC -eq 0 ] && error || true
 }
 run_test 11 "execution of file opened for write should return error ===="
@@ -270,16 +295,16 @@ run_test 12 "test lock ordering (link, stat, unlink)"
 
 test_13() {	# bug 2451 - directory coherency
 	test_mkdir $DIR1/d13
-	cd $DIR1/d13 || error
+	cd $DIR1/d13 || error "cd to $DIR1/d13 failed"
 	ls
 	( touch $DIR1/d13/f13 ) # needs to be a separate shell
 	ls
-	rm -f $DIR2/d13/f13 || error
+	rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13"
 	ls 2>&1 | grep f13 && error "f13 shouldn't return an error (1)" || true
 	# need to run it twice
 	( touch $DIR1/d13/f13 ) # needs to be a separate shell
 	ls
-	rm -f $DIR2/d13/f13 || error
+	rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13"
 	ls 2>&1 | grep f13 && error "f13 shouldn't return an error (2)" || true
 }
 run_test 13 "test directory page revocation"
@@ -377,14 +402,16 @@ else
 	FSXP=100
 fi
 
-test_16() {
+test_16a() {
 	local file1=$DIR1/$tfile
 	local file2=$DIR2/$tfile
+	local stripe_size=$(do_facet $SINGLEMDS \
+		"$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
 
 	# to allocate grant because it may run out due to test_15.
 	$LFS setstripe -c -1 $file1
-	dd if=/dev/zero of=$file1 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync
-	dd if=/dev/zero of=$file2 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync
+	dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync
+	dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
 	rm -f $file1
 
 	$LFS setstripe -c -1 $file1 # b=10919
@@ -396,7 +423,62 @@ test_16() {
 	fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r 4096 \
 		-w 4096 $file1 $file2 || error "fsx with O_DIRECT failed."
 }
-run_test 16 "$FSXNUM iterations of dual-mount fsx"
+run_test 16a "$FSXNUM iterations of dual-mount fsx"
+
+# Consistency check for tiny writes, LU-9409
+test_16b() {
+	local file1=$DIR1/$tfile
+	local file2=$DIR2/$tfile
+
+	# to allocate grant because it may run out due to test_15.
+	lfs setstripe -c -1 $file1
+	dd if=/dev/zero of=$file1 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync
+	dd if=/dev/zero of=$file2 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync
+	rm -f $file1
+
+	lfs setstripe -c -1 $file1 # b=10919
+	# -o is set to 8192 because writes < 1 page and between 1 and 2 pages
+	# create a mix of tiny writes & normal writes
+	fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -o 8192 -S 0 $file1 \
+	$file2
+}
+run_test 16b "$FSXNUM iterations of dual-mount fsx at small size"
+
+test_16c() {
+	local file1=$DIR1/$tfile
+	local file2=$DIR2/$tfile
+	local stripe_size=$(do_facet $SINGLEMDS \
+		"$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
+
+	[ $(facet_fstype ost1) != ldiskfs ] && skip "dio on ldiskfs only"
+
+	# to allocate grant because it may run out due to test_15.
+	$LFS setstripe -c -1 $file1
+	dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync
+	dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
+	rm -f $file1
+	wait_delete_completed
+
+	local list=$(comma_list $(osts_nodes))
+	if ! get_osd_param $list '' read_cache_enable >/dev/null; then
+		skip "not cache-capable obdfilter"
+	fi
+
+	set_osd_param $list '' read_cache_enable 0
+	set_osd_param $list '' writethrough_cache_enable 0
+
+	$LFS setstripe -c -1 $file1 # b=10919
+	fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 \
+		|| error "fsx failed"
+	rm -f $file1
+
+	set_osd_param $list '' read_cache_enable 1
+	set_osd_param $list '' writethrough_cache_enable 1
+
+	return 0
+}
+run_test 16c "verify data consistency on ldiskfs with cache disabled (b=17397)"
+
 
 test_17() { # bug 3513, 3667
 	remote_ost_nodsh && skip "remote OST with nodsh" && return
@@ -432,21 +514,20 @@ run_test 18 "mmap sanity check ================================="
 test_19() { # bug3811
 	local node=$(facet_active_host ost1)
 
+	[ "x$DOM" = "xyes" ] && node=$(facet_active_host $SINGLEMDS)
+
 	# check whether obdfilter is cache capable at all
-	if ! get_osd_param $node '' read_cache_enable >/dev/null; then
-		echo "not cache-capable obdfilter"
-		return 0
-	fi
+	get_osd_param $node '' read_cache_enable >/dev/null ||
+		skip "not cache-capable obdfilter"
 
-	local MAX=$(get_osd_param $node '' readcache_max_filesize | \
-		    head -n 1)
+	local MAX=$(get_osd_param $node '' readcache_max_filesize | head -n 1)
 	set_osd_param $node '' readcache_max_filesize 4096
 	dd if=/dev/urandom of=$TMP/$tfile bs=512k count=32
 	local SUM=$(cksum $TMP/$tfile | cut -d" " -f 1,2)
 	cp $TMP/$tfile $DIR1/$tfile
 	for i in `seq 1 20`; do
 		[ $((i % 5)) -eq 0 ] && log "$testname loop $i"
-		cancel_lru_locks osc > /dev/null
+		cancel_lru_locks $OSC > /dev/null
 		cksum $DIR1/$tfile | cut -d" " -f 1,2 > $TMP/sum1 & \
 		cksum $DIR2/$tfile | cut -d" " -f 1,2 > $TMP/sum2
 		wait
@@ -462,12 +543,12 @@ run_test 19 "test concurrent uncached read races ==============="
 
 test_20() {
 	test_mkdir $DIR1/d20
-	cancel_lru_locks osc
+	cancel_lru_locks $OSC
 	CNT=$((`lctl get_param -n llite.*.dump_page_cache | wc -l`))
 	$MULTIOP $DIR1/f20 Ow8190c
 	$MULTIOP $DIR2/f20 Oz8194w8190c
 	$MULTIOP $DIR1/f20 Oz0r8190c
-	cancel_lru_locks osc
+	cancel_lru_locks $OSC
 	CNTD=$((`lctl get_param -n llite.*.dump_page_cache | wc -l` - $CNT))
 	[ $CNTD -gt 0 ] && \
 	    error $CNTD" page left in cache after lock cancel" || true
@@ -498,7 +579,7 @@ test_23() { # Bug 5972
 	echo "atime should be updated while another read" > $DIR1/$tfile
 
 	# clear the lock(mode: LCK_PW) gotten from creating operation
-	cancel_lru_locks osc
+	cancel_lru_locks $OSC
 	time1=$(date +%s)
 	echo "now is $time1"
 	sleep $((at_diff + 1))
@@ -530,9 +611,9 @@ test_24a() {
 
 	OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1`
 #	OSC=`lctl dl | awk '/-osc-/ {print $4}' | head -n 1`
-	lctl --device %$OSC deactivate
+	lctl --device %osc deactivate
 	lfs df -i || error "lfs df -i with deactivated OSC failed"
-	lctl --device %$OSC activate
+	lctl --device %osc activate
 	lfs df || error "lfs df with reactivated OSC failed"
 }
 run_test 24a "lfs df [-ih] [path] test ========================="
@@ -602,7 +683,7 @@ test_25b() {
 run_test 25b "change ACL under remote dir on one mountpoint be seen on another"
 
 test_26a() {
-        utime $DIR1/f26a -s $DIR2/f26a || error
+	utime $DIR1/f26a -s $DIR2/f26a || error "utime failed for $DIR1/f26a"
 }
 run_test 26a "allow mtime to get older"
 
@@ -622,7 +703,7 @@ test_26b() {
 run_test 26b "sync mtime between ost and mds"
 
 test_27() {
-	cancel_lru_locks osc
+	cancel_lru_locks $OSC
 	lctl clear
 	dd if=/dev/zero of=$DIR2/$tfile bs=$((4096+4))k conv=notrunc count=4 seek=3 &
 	DD2_PID=$!
@@ -666,32 +747,21 @@ test_28() { # bug 9977
 	EOF
 
 	# reading of 1st stripe should pass
-	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error
+	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed"
 	# reading of 2nd stripe should fail (this stripe was destroyed)
 	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 && error
 
 	# now, recreating test file
-	dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error
+	dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error "dd failed"
 	# reading of 1st stripe should pass
-	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error
+	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed"
 	# reading of 2nd stripe should pass
-	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 || error
+	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 ||
+		error "dd failed"
 }
 run_test 28 "read/write/truncate file with lost stripes"
 
-test_29() { # bug 10999
-	touch $DIR1/$tfile
-	#define OBD_FAIL_LDLM_GLIMPSE  0x30f
-	lctl set_param fail_loc=0x8000030f
-	ls -l $DIR2/$tfile &
-	usleep 500
-	dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1
-	wait
-}
-#bug 11549 - permanently turn test off in b1_5
-run_test 29 "lock put race between glimpse and enqueue ========="
-
-test_30() { #bug #11110, LU-2523
+test_30() { #b=11110, LU-2523
 	test_mkdir $DIR1/$tdir
 	cp -f /bin/bash $DIR1/$tdir/bash
 	/bin/sh -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' &
@@ -700,7 +770,6 @@ test_30() { #bug #11110, LU-2523
 	wait
 	true
 }
-
 run_test 30 "recreate file race"
 
 test_31a() {
@@ -741,43 +810,45 @@ run_test 31b "voluntary OST cancel / blocking ast race=============="
 
 # enable/disable lockless truncate feature, depending on the arg 0/1
 enable_lockless_truncate() {
-        lctl set_param -n osc.*.lockless_truncate $1
+	lctl set_param -n $OSC.*.lockless_truncate $1
 }
 
 test_32a() { # bug 11270
-	local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
-	save_lustre_params client "osc.*.lockless_truncate" > $p
-	cancel_lru_locks osc
+	local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
+	local stripe_size=$(do_facet $SINGLEMDS \
+		"$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
+
+	save_lustre_params client "$OSC.*.lockless_truncate" > $save
+	# restore lockless_truncate default values on exit
+	stack_trap "restore_lustre_params < $save; rm -f $save" EXIT
+	cancel_lru_locks $OSC
 	enable_lockless_truncate 1
 	rm -f $DIR1/$tfile
 	lfs setstripe -c -1 $DIR1/$tfile
-	dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \
+	dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$stripe_size > \
 		/dev/null 2>&1
-	clear_stats osc.*.osc_stats
+	clear_stats $OSC.*.${OSC}_stats
 
 	log "checking cached lockless truncate"
 	$TRUNCATE $DIR1/$tfile 8000000
 	$CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size"
-	[ $(calc_stats osc.*.osc_stats lockless_truncate) -ne 0 ] ||
+	[ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -ne 0 ] ||
 		error "cached truncate isn't lockless"
 
 	log "checking not cached lockless truncate"
 	$TRUNCATE $DIR2/$tfile 5000000
 	$CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size"
-	[ $(calc_stats osc.*.osc_stats lockless_truncate) -ne 0 ] ||
+	[ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -ne 0 ] ||
 		error "not cached truncate isn't lockless"
 
 	log "disabled lockless truncate"
 	enable_lockless_truncate 0
-	clear_stats osc.*.osc_stats
+	clear_stats $OSC.*.${OSC}_stats
 	$TRUNCATE $DIR2/$tfile 3000000
 	$CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size"
-	[ $(calc_stats osc.*.osc_stats lockless_truncate) -eq 0 ] ||
+	[ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -eq 0 ] ||
 		error "lockless truncate disabling failed"
-	rm $DIR1/$tfile
-	# restore lockless_truncate default values
-	restore_lustre_params < $p
-	rm -f $p
+	rm -f $DIR1/$tfile
 }
 run_test 32a "lockless truncate"
 
@@ -795,21 +866,21 @@ test_32b() { # bug 11270
 		"ldlm.namespaces.filter-*.contended_locks" >> $p
 	save_lustre_params $facets \
 		"ldlm.namespaces.filter-*.contention_seconds" >> $p
-	clear_stats osc.*.osc_stats
+	clear_stats $OSC.*.${OSC}_stats
 
 	# agressive lockless i/o settings
 	do_nodes $(comma_list $(osts_nodes)) \
 		"lctl set_param -n ldlm.namespaces.*.max_nolock_bytes=2000000 \
 			ldlm.namespaces.filter-*.contended_locks=0 \
 			ldlm.namespaces.filter-*.contention_seconds=60"
-	lctl set_param -n osc.*.contention_seconds=60
+	lctl set_param -n $OSC.*.contention_seconds=60
 	for i in {1..5}; do
 		dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
 			/dev/null 2>&1
 		dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
 			/dev/null 2>&1
 	done
-	[ $(calc_stats osc.*.osc_stats lockless_write_bytes) -ne 0 ] ||
+	[ $(calc_stats $OSC.*.${OSC}_stats lockless_write_bytes) -ne 0 ] ||
 		error "lockless i/o was not triggered"
 	# disable lockless i/o (it is disabled by default)
 	do_nodes $(comma_list $(osts_nodes)) \
@@ -818,15 +889,15 @@ test_32b() { # bug 11270
 			ldlm.namespaces.filter-*.contention_seconds=0"
 	# set contention_seconds to 0 at client too, otherwise Lustre still
 	# remembers lock contention
-	lctl set_param -n osc.*.contention_seconds=0
-	clear_stats osc.*.osc_stats
+	lctl set_param -n $OSC.*.contention_seconds=0
+	clear_stats $OSC.*.${OSC}_stats
 	for i in {1..1}; do
 		dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
 			/dev/null 2>&1
 		dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
 			/dev/null 2>&1
 	done
-	[ $(calc_stats osc.*.osc_stats lockless_write_bytes) -eq 0 ] ||
+	[ $(calc_stats $OSC.*.${OSC}_stats lockless_write_bytes) -eq 0 ] ||
 		error "lockless i/o works when disabled"
 	rm -f $DIR1/$tfile
 	restore_lustre_params <$p
@@ -969,39 +1040,57 @@ test_33c() {
 	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
 		skip "DNE CoS not supported" && return
 
-	sync
+	local sync_count
 
 	mkdir $DIR/$tdir
-	# remote mkdir is done on MDT2, which enqueued lock of $tdir on MDT1
-	$LFS mkdir -i 1 $DIR/$tdir/d1
+	sync_all_data
 	do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
-	mkdir $DIR/$tdir/d2
-	local sync_count=$(do_facet mds1 \
-		"lctl get_param -n mdt.*MDT0000.sync_count")
-	[ $sync_count -eq 1 ] || error "Sync-Lock-Cancel not triggered"
+	# do twice in case transaction is committed before unlock, see LU-8200
+	for i in 1 2; do
+		# remote dir is created on MDT1, which enqueued lock of $tdir on
+		# MDT0
+		$LFS mkdir -i 1 $DIR/$tdir/remote.$i
+		mkdir $DIR/$tdir/local.$i
+	done
+	sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count")
+	echo "sync_count $sync_count"
+	[ $sync_count -eq 0 ] && error "Sync-Lock-Cancel not triggered"
 
-	$LFS mkdir -i 1 $DIR/$tdir/d3
+	sync_all_data
 	do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
+	$LFS mkdir -i 1 $DIR/$tdir/remote.3
 	# during sleep remote mkdir should have been committed and canceled
 	# remote lock spontaneously, which shouldn't trigger sync
 	sleep 6
-	mkdir $DIR/$tdir/d4
-	local sync_count=$(do_facet mds1 \
-		"lctl get_param -n mdt.*MDT0000.sync_count")
+	mkdir $DIR/$tdir/local.3
+	sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count")
+	echo "sync_count $sync_count"
 	[ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered"
 }
 run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel"
 
-ops_do_cos() {
+# arg1 is operations done before CoS, arg2 is the operation that triggers CoS
+op_trigger_cos() {
+	local commit_nr
+	local total=0
 	local nodes=$(comma_list $(mdts_nodes))
-	do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
-	sh -c "$@"
-	local async_commit_count=$(do_nodes $nodes \
-		"lctl get_param -n mdt.*.async_commit_count" | calc_sum)
-	[ $async_commit_count -gt 0 ] || error "CoS not triggerred"
 
-	rm -rf $DIR/$tdir
-	sync
+	sync_all_data
+
+	# trigger CoS twice in case transaction commit before unlock
+	for i in 1 2; do
+		sh -c "$1"
+		do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
+		sh -c "$2"
+		commit_nr=$(do_nodes $nodes \
+			"lctl get_param -n mdt.*.async_commit_count" | calc_sum)
+		total=$((total + commit_nr));
+		rm -rf $DIR/$tdir
+		sync_all_data
+	done
+
+	echo "CoS count $total"
+	[ $total -gt 0 ] || error "$2 didn't trigger CoS"
 }
 
 test_33d() {
@@ -1009,39 +1098,34 @@ test_33d() {
 	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
 		skip "DNE CoS not supported" && return
 
-	sync
 	# remote directory create
-	mkdir $DIR/$tdir
-	ops_do_cos "$LFS mkdir -i 1 $DIR/$tdir/subdir"
+	op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -i 1 $DIR/$tdir/subdir"
 	# remote directory unlink
-	$LFS mkdir -i 1 $DIR/$tdir
-	ops_do_cos "rmdir $DIR/$tdir"
+	op_trigger_cos "$LFS mkdir -i 1 $DIR/$tdir" "rmdir $DIR/$tdir"
 	# striped directory create
-	mkdir $DIR/$tdir
-	ops_do_cos "$LFS mkdir -c 2 $DIR/$tdir/subdir"
+	op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -c 2 $DIR/$tdir/subdir"
 	# striped directory setattr
-	$LFS mkdir -c 2 $DIR/$tdir
-	touch $DIR/$tdir
-	ops_do_cos "chmod 713 $DIR/$tdir"
+	op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \
+		"chmod 713 $DIR/$tdir"
 	# striped directory unlink
-	$LFS mkdir -c 2 $DIR/$tdir
-	touch $DIR/$tdir
-	ops_do_cos "rmdir $DIR/$tdir"
+	op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \
+		"rmdir $DIR/$tdir"
 	# cross-MDT link
-	$LFS mkdir -c 2 $DIR/$tdir
-	$LFS mkdir -i 0 $DIR/$tdir/d1
-	$LFS mkdir -i 1 $DIR/$tdir/d2
-	touch $DIR/$tdir/d1/tgt
-	ops_do_cos "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src"
+	op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \
+			$LFS mkdir -i 0 $DIR/$tdir/d1; \
+			$LFS mkdir -i 1 $DIR/$tdir/d2; \
+			touch $DIR/$tdir/d1/tgt" \
+		"ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src"
 	# cross-MDT rename
-	$LFS mkdir -c 2 $DIR/$tdir
-	$LFS mkdir -i 0 $DIR/$tdir/d1
-	$LFS mkdir -i 1 $DIR/$tdir/d2
-	touch $DIR/$tdir/d1/src
-	ops_do_cos "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt"
+	op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \
+			$LFS mkdir -i 0 $DIR/$tdir/d1; \
+			$LFS mkdir -i 1 $DIR/$tdir/d2; \
+			touch $DIR/$tdir/d1/src" \
+		"mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt"
 	# migrate
-	$LFS mkdir -i 0 $DIR/$tdir
-	ops_do_cos "$LFS migrate -m 1 $DIR/$tdir"
+	op_trigger_cos "$LFS mkdir -i 0 $DIR/$tdir" \
+		"$LFS migrate -m 1 $DIR/$tdir"
+
 	return 0
 }
 run_test 33d "DNE distributed operation should trigger COS"
@@ -1093,7 +1177,7 @@ cleanup_34() {
 	do_nodes $(comma_list $(osts_nodes)) \
 		"lctl set_param -n fail_loc=0 2>/dev/null || true"
 	for i in $(seq $OSTCOUNT); do
-		wait_osc_import_state client ost$i FULL
+		wait_osc_import_ready client ost$i
 	done
 }
 
@@ -1380,7 +1464,7 @@ test_39d() { # LU-7310
 
 	$LCTL set_param fail_loc=0
 
-	cancel_lru_locks osc
+	cancel_lru_locks $OSC
 
 	local mtime2=$(stat -c %Y $DIR2/$tfile)
 	[ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] ||
@@ -2391,11 +2475,13 @@ run_test 47f "pdirops: remote mkdir and rename (src)"
 
 test_47g() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+	sync
+	sync_all_data
 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK    0x145
 	do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
 	$LFS mkdir -i 1 $DIR1/$tfile &
 	PID1=$!
-	sleep 1
+	sleep 2
 	stat $DIR2/$tfile > /dev/null || error "stat must succeed"
 	check_pdo_conflict $PID1 && { wait $PID1;
 					error "getattr isn't blocked"; }
@@ -2527,14 +2613,14 @@ test_51d() {
 	local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
 	echo "Before revoking layout lock: $br KB mapped"
 
-	# delete the file will revoke layout lock
-	rm -f $DIR2/$tfile
+	# cancel layout lock manually
+	cancel_lru_locks mdc
 
 	# rss after revoking
 	local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
 
 	kill -USR1 $PID
-	wait $PID || error
+	wait $PID || error "wait PID $PID failed"
 
 	[ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained"
 }
@@ -2796,7 +2882,7 @@ test_71a() {
 		{ skip "checkfiemap not runnable: $?" && return; }
 	# write data this way: hole - data - hole - data
 	dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
-	[ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 1)))" = \
+	[ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tfile) + 1)))" = \
 		"zfs" ] &&
 		skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
 	dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
@@ -2844,7 +2930,7 @@ test_71b() {
 
 	$LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed"
 	dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=40K count=1
-	[ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tdir/$tfile) + 1)))" = \
+	[ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tdir/$tfile) + 1)))" = \
 		"zfs" ] &&
 		skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
 	checkfiemap $DIR1/$tdir/$tfile 40960 || error "checkfiemap failed"
@@ -3065,8 +3151,13 @@ nrs_write_read() {
 }
 
 test_77a() { #LU-3266
+	local rc
+
 	oss=$(comma_list $(osts_nodes))
-	do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo"
+	do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo" ||
+		rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS exists" && return
+	[[ $rc -ne 0 ]] && error "failed to set fifo policy"
 	nrs_write_read
 
 	return 0
@@ -3074,21 +3165,29 @@ test_77a() { #LU-3266
 run_test 77a "check FIFO NRS policy"
 
 test_77b() { #LU-3266
+	local rc
+
 	oss=$(comma_list $(osts_nodes))
 
 	do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="crrn" \
-			   ost.OSS.*.nrs_crrn_quantum=1
+		ost.OSS.*.nrs_crrn_quantum=1 || rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS exists" && return
+	[[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 1"
 
 	echo "policy: crr-n, crrn_quantum 1"
 	nrs_write_read
 
-	do_nodes $oss lctl set_param ost.OSS.*.nrs_crrn_quantum=64
+	do_nodes $oss lctl set_param \
+		ost.OSS.*.nrs_crrn_quantum=64 || rc=$?
+	[[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 64"
 
 	echo "policy: crr-n, crrn_quantum 64"
 	nrs_write_read
 
 	# cleanup
-	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+	do_nodes $oss lctl set_param \
+		ost.OSS.ost_io.nrs_policies="fifo" || rc=$?
+	[[ $rc -ne 0 ]] && error "failed to set fifo policy"
 	return 0
 }
 run_test 77b "check CRR-N NRS policy"
@@ -3099,37 +3198,49 @@ orr_trr() {
 	oss=$(comma_list $(osts_nodes))
 
 	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies=$policy \
-				     ost.OSS.*.nrs_"$policy"_quantum=1 \
-				     ost.OSS.*.nrs_"$policy"_offset_type="physical" \
-				     ost.OSS.*.nrs_"$policy"_supported="reads"
+		ost.OSS.*.nrs_"$policy"_quantum=1 \
+		ost.OSS.*.nrs_"$policy"_offset_type="physical" \
+		ost.OSS.*.nrs_"$policy"_supported="reads" || return $?
 
-	echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type physical, ${policy}_supported reads"
+	echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type " \
+		"physical, ${policy}_supported reads"
 	nrs_write_read
 
-	do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="writes" \
-				     ost.OSS.*.nrs_${policy}_quantum=64
+	do_nodes $oss lctl set_param \
+		ost.OSS.*.nrs_${policy}_supported="writes" \
+		ost.OSS.*.nrs_${policy}_quantum=64 || return $?
 
-	echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type physical, ${policy}_supported writes"
+	echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \
+		"physical, ${policy}_supported writes"
 	nrs_write_read
 
-	do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \
-				     ost.OSS.*.nrs_${policy}_offset_type="logical"
-	echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type logical, ${policy}_supported reads_and_writes"
+	do_nodes $oss lctl set_param \
+		ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \
+		ost.OSS.*.nrs_${policy}_offset_type="logical" || return $?
+	echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \
+		"logical, ${policy}_supported reads_and_writes"
 	nrs_write_read
 
 	# cleanup
-	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" ||
+		return $?
 	return 0
 }
 
 test_77c() { #LU-3266
-	orr_trr "orr"
+	local rc
+	orr_trr "orr" || rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS exists" && return
+	[[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc"
 	return 0
 }
 run_test 77c "check ORR NRS policy"
 
 test_77d() { #LU-3266
-	orr_trr "trr"
+	local rc
+	orr_trr "trr" || rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS exists" && return
+	[[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc"
 	return 0
 }
 run_test 77d "check TRR nrs policy"
@@ -3161,47 +3272,55 @@ tbf_verify() {
 	local client1=${CLIENT1:-$(hostname)}
 	local myRUNAS="$3"
 
+	local np=$(check_cpt_number ost1)
+	[ $np -gt 0 ] || error "CPU partitions should not be $np."
+	echo "cpu_npartitions on ost1 is $np"
+
 	mkdir $dir || error "mkdir $dir failed"
-	$LFS setstripe -c 1 $dir || error "setstripe to $dir failed"
+	$LFS setstripe -c 1 -i 0 $dir || error "setstripe to $dir failed"
 	chmod 777 $dir
 
 	trap cleanup_tbf_verify EXIT
 	echo "Limited write rate: $1, read rate: $2"
 	echo "Verify the write rate is under TBF control"
-	local runtime=$(do_node $client1 $myRUNAS dd if=/dev/zero of=$dir/tbf \
-		bs=1M count=100 oflag=direct 2>&1 | awk '/bytes/ {print $6}')
+	local start=$SECONDS
+	do_node $client1 $myRUNAS dd if=/dev/zero of=$dir/tbf \
+		bs=1M count=100 oflag=direct 2>&1
+	local runtime=$((SECONDS - start + 1))
 	local rate=$(bc <<< "scale=6; 100 / $runtime")
 	echo "Write runtime is $runtime s, speed is $rate IOPS"
 
-	# verify the write rate does not exceed 110% of TBF limited rate
-	[ $(bc <<< "$rate < 1.1 * $1") -eq 1 ] ||
-		error "The write rate ($rate) exceeds 110% of preset rate ($1)"
+	# verify the write rate does not exceed TBF rate limit
+	[ $(bc <<< "$rate < 1.1 * $np * $1") -eq 1 ] ||
+		error "The write rate ($rate) exceeds 110% of rate limit ($1 * $np)"
 
 	cancel_lru_locks osc
 
 	echo "Verify the read rate is under TBF control"
-	runtime=$(do_node $client1 $myRUNAS dd if=$dir/tbf of=/dev/null \
-		bs=1M count=100 iflag=direct 2>&1 | awk '/bytes/ {print $6}')
+	start=$SECONDS
+	do_node $client1 $myRUNAS dd if=$dir/tbf of=/dev/null \
+		bs=1M count=100 iflag=direct 2>&1
+	runtime=$((SECONDS - start + 1))
 	rate=$(bc <<< "scale=6; 100 / $runtime")
 	echo "Read runtime is $runtime s, speed is $rate IOPS"
 
-	# verify the read rate does not exceed 110% of TBF limited rate
-	[ $(bc <<< "$rate < 1.1 * $2") -eq 1 ] ||
-		error "The read rate ($rate) exceeds 110% of preset rate ($2)"
+	# verify the read rate does not exceed TBF rate limit
+	[ $(bc <<< "$rate < 1.1 * $np * $2") -eq 1 ] ||
+		error "The read rate ($rate) exceeds 110% of rate limit ($2 * $np)"
 
 	cancel_lru_locks osc
 	cleanup_tbf_verify || error "rm -rf $dir failed"
 }
 
 test_77e() {
-	local server_version=$(lustre_version_code ost1)
-	[[ $server_version -ge $(version_code 2.7.58) ]] ||
-		{ skip "Need server version newer than 2.7.57"; return 0; }
+	local rc
 
 	oss=$(comma_list $(osts_nodes))
 
-	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
-	[ $? -ne 0 ] && error "failed to set TBF policy"
+	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" ||
+		rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
+	[[ $rc -ne 0 ]] && error "failed to set TBF NID policy"
 
 	local idis
 	local rateis
@@ -3240,23 +3359,25 @@ test_77e() {
 run_test 77e "check TBF NID nrs policy"
 
 test_77f() {
-	local server_version=$(lustre_version_code ost1)
-	[[ $server_version -ge $(version_code 2.7.58) ]] ||
-		{ skip "Need server version newer than 2.7.57"; return 0; }
+	local rc
 
 	oss=$(comma_list $(osts_nodes))
 
+	do_nodes $oss $LCTL set_param \
+		ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
+	[[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy"
+
 	# Configure jobid_var
 	local saved_jobid_var=$($LCTL get_param -n jobid_var)
+	rc=$?
+	[[ $rc -eq 3 ]] && skip "jobid_var not found" && return
+	[[ $rc -ne 0 ]] && error "failed to get param jobid_var"
 	if [ $saved_jobid_var != procname_uid ]; then
-		set_conf_param_and_check client			\
-			"$LCTL get_param -n jobid_var"		\
-			"$FSNAME.sys.jobid_var" procname_uid
+		set_persistent_param_and_check client \
+			"jobid_var" "$FSNAME.sys.jobid_var" procname_uid
 	fi
 
-	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid"
-	[ $? -ne 0 ] && error "failed to set TBF policy"
-
 	local idis
 	local rateis
 	if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
@@ -3289,27 +3410,28 @@ test_77f() {
 	nrs_write_read "$RUNAS"
 
 	local current_jobid_var=$($LCTL get_param -n jobid_var)
+	[[ $? -ne 0 ]] && error "failed to get param jobid_var"
 	if [ $saved_jobid_var != $current_jobid_var ]; then
-		set_conf_param_and_check client			\
-			"$LCTL get_param -n jobid_var"		\
-			"$FSNAME.sys.jobid_var" $saved_jobid_var
+		set_persistent_param_and_check client \
+			"jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var
 	fi
 	return 0
 }
 run_test 77f "check TBF JobID nrs policy"
 
 test_77g() {
-	local server_version=$(lustre_version_code ost1)
-	[[ $server_version -ge $(version_code 2.7.58) ]] ||
-		{ skip "Need server version newer than 2.7.57"; return 0; }
+	local rc=0
 
 	oss=$(comma_list $(osts_nodes))
 
-	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
-	[ $? -ne 0 ] && error "failed to set TBF policy"
+	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" ||
+		rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
+	[[ $rc -ne 0 ]] && error "failed to set TBF NID policy"
 
-	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid"
-	[ $? -ne 0 ] && error "failed to set TBF policy"
+	do_nodes $oss lctl set_param \
+		ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$?
+	[[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy"
 
 	local idis
 	local rateis
@@ -3385,7 +3507,7 @@ tbf_rule_check()
 }
 
 test_77i() {
-    [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] ||
+	[ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] ||
 		{ skip "Need OST version at least 2.8.55"; return 0; }
 
 	for i in $(seq 1 $OSTCOUNT)
@@ -3440,7 +3562,11 @@ run_test 77i "Change rank of TBF rule"
 test_77j() {
 	local idis
 	local rateis
-	if [ $(lustre_version_code ost1) -ge $(version_code 2.8.60) ]; then
+	local ost_version=$(lustre_version_code ost1)
+
+	[ $ost_version -ge $(version_code 2.9.53) ] ||
+		{ skip "Need OST version at least 2.9.53"; return 0; }
+	if [ $ost_version -ge $(version_code 2.8.60) ]; then
 		idis="opcode="
 		rateis="rate="
 	fi
@@ -3467,6 +3593,57 @@ test_77j() {
 }
 run_test 77j "check TBF-OPCode NRS policy"
 
+test_id() {
+	local idstr="${1}id"
+	local policy="${idstr}={$2}"
+	local rate="rate=$3"
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param jobid_var=procname_uid \
+			ost.OSS.ost_io.nrs_policies="tbf\ ${idstr}" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ ost_${idstr}\ ${policy}\ ${rate}"
+	[ $? -ne 0 ] && error "failed to set tbf ${idstr} policy"
+
+	nrs_write_read "runas $4"
+	tbf_verify $3 $3 "runas $4"
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_${idstr}" \
+			ost.OSS.ost_io.nrs_policies="fifo"
+
+	# sleep 3 seconds to wait the tbf policy stop completely,
+	# or the next test case is possible get -eagain when
+	# setting the tbf policy
+	sleep 3
+}
+
+test_77ja(){
+	if [ $(lustre_version_code ost1) -lt $(version_code 2.11.50) ]; then
+		skip "Need OST version at least 2.11.50"
+		return 0
+	fi
+	test_id "u" "500" "5" "-u 500"
+	test_id "g" "500" "5" "-u 500 -g 500"
+}
+run_test 77ja "check TBF-UID/GID NRS policy"
+
+cleanup_77k()
+{
+	local rule_lists=$1
+	local old_nrs=$2
+
+	trap 0
+	for rule in $rule_lists; do
+		do_nodes $(comma_list $(osts_nodes)) \
+			lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ $rule"
+	done
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_policies="$old_nrs"
+
+	sleep 3
+}
+
 test_77k() {
 	[[ $(lustre_version_code ost1) -ge $(version_code 2.9.53) ]] ||
 		{ skip "Need OST version at least 2.9.53"; return 0; }
@@ -3509,16 +3686,59 @@ test_77k() {
 	tbf_verify 10 10 "$RUNAS"
 	tbf_verify 20 10
 
+	trap "cleanup_77k \"ext_a ext_b\" \"fifo\"" EXIT
+
+	[[ $(lustre_version_code ost1) -ge $(version_code 2.10.58) ]] ||
+		{ skip "Need OST version at least 2.10.58"; return 0; }
+
 	do_nodes $(comma_list $(osts_nodes)) \
 		lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \
 			ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \
-			ost.OSS.ost_io.nrs_policies="fifo"
+			ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ug\ uid={500}\&gid={1000}\ rate=5"
+	nrs_write_read "runas -u 500 -g 1000"
+	tbf_verify 5 5 "runas -u 500 -g 1000"
 
-	sleep 3
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ug" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ ext_uw\ uid={500}\&opcode={ost_write}\ rate=20" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ur\ uid={500}\&opcode={ost_read}\ rate=10"
+
+	nrs_write_read "runas -u 500"
+	tbf_verify 20 10 "runas -u 500"
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_uw" \
+			ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ur" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ uid={500},opcode={ost_write}\ rate=20" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ uid={500},opcode={ost_read}\ rate=10"
+	nrs_write_read "runas -u 500"
+	tbf_verify 10 10 "runas -u 500"
+	tbf_verify 20 10 "runas -u 500"
+	cleanup_77k "ext_a ext_b" "fifo"
 }
-run_test 77k "check the extended TBF policy with NID/JobID/OPCode expression"
+run_test 77k "check TBF policy with NID/JobID/OPCode expression"
 
 test_77l() {
+	[[ $(lustre_version_code ost1) -ge $(version_code 2.10.56) ]] ||
+		{ skip "Need OST version at least 2.10.56"; return 0; }
+
+	do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
+	do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf"
+
+	local output=$(do_facet ost1 lctl get_param \
+			ost.OSS.ost_io.nrs_policies | \
+			awk '/name: tbf/ {print;exit}' | \
+			awk -F ': ' '{print $2}')
+
+	if [ "$output" != "tbf" ]; then
+		error "The generic TBF output is '$output', not 'tbf'"
+	fi
+
+	do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+}
+run_test 77l "check the output of NRS policies for generic TBF"
+
+test_77m() {
 	if [ $(lustre_version_code ost1) -lt $(version_code 2.9.54) ]; then
 		skip "Need OST version at least 2.9.54"
 		return 0
@@ -3568,19 +3788,58 @@ test_77l() {
 
 	return 0
 }
-run_test 77l "check NRS Delay slows write RPC processing"
+run_test 77m "check NRS Delay slows write RPC processing"
 
-test_78() { #LU-6673
-	local server_version=$(lustre_version_code ost1)
-	[[ $server_version -ge $(version_code 2.7.58) ]] ||
-		{ skip "Need server version newer than 2.7.57"; return 0; }
+test_77n() { #LU-10802
+	if [ $(lustre_version_code ost1) -lt $(version_code 2.10.58) ]; then
+		skip "Need OST version at least 2.10.58"
+		return 0
+	fi
+
+	# Configure jobid_var
+	local saved_jobid_var=$($LCTL get_param -n jobid_var)
+	if [ $saved_jobid_var != procname_uid ]; then
+		set_persistent_param_and_check client \
+			"jobid_var" "$FSNAME.sys.jobid_var" procname_uid
+	fi
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" \
+			ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={*.$RUNAS_ID}\ rate=20"
+
+	nrs_write_read
+	tbf_verify 20 20 "$RUNAS"
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
+			ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={dd.*}\ rate=20"
+
+	nrs_write_read
+	tbf_verify 20 20
+
+	do_nodes $(comma_list $(osts_nodes)) \
+		lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
+			ost.OSS.ost_io.nrs_policies="fifo"
+
+	sleep 3
+
+	local current_jobid_var=$($LCTL get_param -n jobid_var)
+	if [ $saved_jobid_var != $current_jobid_var ]; then
+		set_persistent_param_and_check client \
+			"jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var
+	fi
+}
+run_test 77n "check wildcard support for TBF JobID NRS policy"
 
+test_78() { #LU-6673
 	local rc
 
 	oss=$(comma_list $(osts_nodes))
 	do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="orr" &
 	do_nodes $oss lctl set_param ost.OSS.*.nrs_orr_quantum=1
 	rc=$?
+	[[ $rc -eq 3 ]] && skip "no NRS exists" && return
 	# Valid return codes are:
 	# 0: Tuning succeeded
 	# ENODEV: Policy is still stopped
@@ -3607,7 +3866,7 @@ test_79() {
 		error "setfattr -n trusted.name1=value1 $DIR/$tdir failed"
 
 #define OBD_FAIL_MDS_INTENT_DELAY		0x160
-	local mdtidx=$($LFS getstripe -M $DIR/$tdir)
+	local mdtidx=$($LFS getstripe -m $DIR/$tdir)
 	local facet=mds$((mdtidx + 1))
 	stat $DIR/$tdir
 	set_nodes_failloc $(facet_active_host $facet) 0x80000160
@@ -3653,7 +3912,7 @@ test_80a() {
 
 	echo "Finish migration, then checking.."
 	for file in $(find $DIR1/$tdir); do
-		mdt_index=$($LFS getstripe -M $file)
+		mdt_index=$($LFS getstripe -m $file)
 		[ $mdt_index == $MDTIDX ] ||
 			error "$file is not on MDT${MDTIDX}"
 	done
@@ -3788,7 +4047,7 @@ test_80b() {
 }
 run_test 80b "Accessing directory during migration"
 
-test_81() {
+test_81a() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
 
 	rm -rf $DIR1/$tdir
@@ -3814,7 +4073,35 @@ test_81() {
 
 	return 0
 }
-run_test 81 "rename and stat under striped directory"
+run_test 81a "rename and stat under striped directory"
+
+test_81b() {
+	[ $MDSCOUNT -lt 2 ] &&
+		skip "We need at least 2 MDTs for this test"
+
+	local total
+	local setattr_pid
+
+	total=1000
+
+	$LFS mkdir -c $MDSCOUNT $DIR1/$tdir || error "$LFS mkdir"
+	createmany -o $DIR1/$tdir/$tfile. $total || error "createmany"
+
+	(
+		while true; do
+			touch $DIR1/$tdir
+		done
+	) &
+	setattr_pid=$!
+
+	for i in $(seq $total); do
+		mrename $DIR2/$tdir/$tfile.$i $DIR2/$tdir/$tfile-new.$i \
+			> /dev/null
+	done
+
+	kill -9 $setattr_pid
+}
+run_test 81b "rename under striped directory doesn't deadlock"
 
 test_82() {
 	[[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.6.91) ]] ||
@@ -3984,10 +4271,10 @@ test_93() {
 	#define OBD_FAIL_MDS_LOV_CREATE_RACE     0x163
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0x00000163"
 
-	$SETSTRIPE -c -1 $DIR1/$tfile-1/file1 &
+	$LFS setstripe -c -1 $DIR1/$tfile-1/file1 &
 	local PID1=$!
 	sleep 1
-	$SETSTRIPE -c -1 $DIR2/$tfile-2/file2 &
+	$LFS setstripe -c -1 $DIR2/$tfile-2/file2 &
 	local PID2=$!
 	wait $PID2
 	wait $PID1
@@ -3995,11 +4282,11 @@ test_93() {
 	do_facet $SINGLEMDS "lctl set_param -n \
 		'lod.lustre-MDT*/qos_threshold_rr' $old_rr"
 
-	$GETSTRIPE $DIR1/$tfile-1/file1
-	rc1=$($GETSTRIPE -q $DIR1/$tfile-1/file1 |
+	$LFS getstripe $DIR1/$tfile-1/file1
+	rc1=$($LFS getstripe -q $DIR1/$tfile-1/file1 |
 		awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
-	$GETSTRIPE $DIR2/$tfile-2/file2
-	rc2=$($GETSTRIPE -q $DIR2/$tfile-2/file2 |
+	$LFS getstripe $DIR2/$tfile-2/file2
+	rc2=$($LFS getstripe -q $DIR2/$tfile-2/file2 |
 		awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
 	echo "rc1=$rc1 and rc2=$rc2 "
 	[ $rc1 -eq 0 ] && [ $rc2 -eq 0 ] ||
@@ -4007,6 +4294,196 @@ test_93() {
 }
 run_test 93 "alloc_rr should not allocate on same ost"
 
+# Data-on-MDT tests
+test_100a() {
+	skip "Reserved for glimpse-ahead" && return
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	mkdir -p $DIR/$tdir
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
+
+	lctl set_param -n mdc.*.stats=clear
+	dd if=/dev/zero of=$DIR2/$tdir/dom bs=4096 count=1 || return 1
+
+	$CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1"
+	# first stat from server should return size data and save glimpse
+	local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
+	[ $gls -eq 0 ] || error "Unexpected $gls glimpse RPCs"
+	# second stat to check size is NOT cached on client without IO lock
+	$CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2"
+
+	local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
+	[ $gls -ge 1 ] || error "Expect glimpse RPCs but none"
+	rm -f $dom
+}
+run_test 100a "DoM: glimpse RPCs for stat without IO lock (DoM only file)"
+
+test_100b() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	mkdir -p $DIR/$tdir
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
+
+	lctl set_param -n mdc.*.stats=clear
+	dd if=/dev/zero of=$DIR2/$tdir/dom bs=4096 count=1 || return 1
+	cancel_lru_locks mdc
+	# first stat data from server should have size
+	$CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1"
+	# second stat to check size is cached on client
+	$CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2"
+
+	local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
+	# both stats should cause no glimpse requests
+	[ $gls == 0 ] || error "Unexpected $gls glimpse RPCs"
+	rm -f $dom
+}
+run_test 100b "DoM: no glimpse RPC for stat with IO lock (DoM only file)"
+
+test_100c() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	mkdir -p $DIR/$tdir
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
+
+	lctl set_param -n mdc.*.stats=clear
+	lctl set_param -n osc.*.stats=clear
+	dd if=/dev/zero of=$DIR2/$tdir/dom bs=2048K count=1 || return 1
+
+	# check that size is merged from MDT and OST correctly
+	$CHECKSTAT -t file -s 2097152 $DIR/$tdir/dom ||
+		error "Wrong size from stat #1"
+
+	local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
+	[ $gls -eq 0 ] && error "Expect OST glimpse RPCs but got none"
+
+	rm -f $dom
+}
+run_test 100c "DoM: write vs stat without IO lock (combined file)"
+
+test_100d() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	mkdir -p $DIR/$tdir
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
+
+
+	dd if=/dev/zero of=$DIR2/$tdir/dom bs=2048K count=1 || return 1
+	lctl set_param -n mdc.*.stats=clear
+	$TRUNCATE $DIR2/$tdir/dom 4096
+
+	# check that reported size is valid after file grows to OST and
+	# is truncated back to MDT stripe size
+	$CHECKSTAT -t file -s 4096 $DIR/$tdir/dom ||
+		error "Wrong size from stat #1"
+
+	local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
+	[ $gls -eq 0 ] && error "Expect OST glimpse but got none"
+
+	rm -f $dom
+}
+run_test 100d "DoM: write+truncate vs stat without IO lock (combined file)"
+
+test_101a() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
+	# to get layout
+	$CHECKSTAT -t file $DIR1/$tfile
+
+	local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
+	sysctl -wq vm.dirty_writeback_centisecs=0
+
+	trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
+
+	# open + IO lock
+	dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 ||
+		error_noexit "Write fails"
+	# must discard pages
+	lctl set_param -n mdc.*.stats=clear
+	rm $DIR2/$tfile || error "Unlink fails"
+
+	local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
+	[ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
+}
+run_test 101a "Discard DoM data on unlink"
+
+test_101b() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
+	touch $DIR1/${tfile}_2
+	# to get layout
+	$CHECKSTAT -t file $DIR1/$tfile
+
+	local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
+	sysctl -wq vm.dirty_writeback_centisecs=0
+
+	trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
+
+	# open + IO lock
+	dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails"
+	# must discard pages
+	lctl set_param -n mdc.*.stats=clear
+	mv $DIR2/${tfile}_2 $DIR2/$tfile || error "Rename fails"
+
+	local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
+	[ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
+}
+run_test 101b "Discard DoM data on rename"
+
+test_101c() {
+	[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
+		skip "Need MDS version at least 2.10.55" && return
+
+	$LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
+	# to get layout
+	$CHECKSTAT -t file $DIR1/$tfile
+
+	local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
+	sysctl -wq vm.dirty_writeback_centisecs=0
+
+	trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
+
+	# open + IO lock
+	dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails"
+	$MULTIOP $DIR1/$tfile O_c &
+	MULTIOP_PID=$!
+	sleep 1
+	lctl set_param -n mdc.*.stats=clear
+	rm $DIR2/$tfile > /dev/null || error "Unlink fails for opened file"
+	kill -USR1 $MULTIOP_PID && wait $MULTIOP_PID || error "multiop failure"
+
+	local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
+	[ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
+}
+run_test 101c "Discard DoM data on close-unlink"
+
+# test to verify file handle related system calls
+# (name_to_handle_at/open_by_handle_at)
+# The new system calls are supported in glibc >= 2.14.
+
+# test to verify we can open by handle an unlinked file from > 1 client
+# This test opens the file normally on $DIR1, which is on one mount, and then
+# opens it by handle on $DIR2, which is on a different mount.
+test_102() {
+	echo "Test file_handle syscalls" > $DIR/$tfile ||
+		error "write failed"
+	check_fhandle_syscalls $DIR/$tfile $DIR2 ||
+		error "check_fhandle_syscalls failed"
+	rm -f $DIR2/$tfile
+}
+run_test 102 "Test open by handle of unlinked file"
+
 log "cleanup: ======================================================"
 
 # kill and wait in each test only guarentee script finish, but command in script