X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=a36eeb1d3422189bea99c11ef4ce9c36b9275f39;hb=9c4156e6fc146a198bb342e28eb246f1076889bd;hp=33c74591449acee25525c2885e842d0776de79ff;hpb=0209add4a5099817111c8576afe930d1e2daef03;p=fs%2Flustre-release.git

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 33c7459..a36eeb1 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -16,12 +16,12 @@ init_logging
 require_dsh_mds || exit 0
 
 # also long tests: 19, 21a, 21e, 21f, 23, 27
-#                                   1  2.5  2.5    4    4          (min)"
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="17  26a  26b    50   51     57"
+
+[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
 
 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:	      LU-2194 LU-2547
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b     24a 24b"
+# bug number for skipped test:	      LU-2547
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 24a 24b"
 
 build_test_filter
 
@@ -149,21 +149,140 @@ test_9() {
 run_test 9 "pause bulk on OST (bug 1420)"
 
 #bug 1521
-test_10() {
-	do_facet client mcreate $DIR/$tfile ||
-		{ error "mcreate failed: $?"; return 1; }
-	drop_bl_callback "chmod 0777 $DIR/$tfile"  || echo "evicted as expected"
-	# wait for the mds to evict the client
-	#echo "sleep $(($TIMEOUT*2))"
-	#sleep $(($TIMEOUT*2))
-	do_facet client touch $DIR/$tfile || echo "touch failed, evicted"
-	do_facet client checkstat -v -p 0777 $DIR/$tfile ||
-		{ error "client checkstat failed: $?"; return 3; }
-	do_facet client "munlink $DIR/$tfile"
-	# allow recovery to complete
-	client_up || client_up || sleep $TIMEOUT
+test_10a() {
+	local before=$(date +%s)
+	local evict
+
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_bl_callback "chmod 0777 $DIR" ||
+		error "failed to chmod $DIR: $?"
+
+	# let the client reconnect
+	client_reconnect
+	evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+	[ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "no eviction: $evict before:$before")
+
+	do_facet client checkstat -v -p 0777 $DIR ||
+		error "client checkstat failed: $?"
 }
-run_test 10 "finish request on server after client eviction (bug 1521)"
+run_test 10a "finish request on server after client eviction (bug 1521)"
+
+test_10b() {
+	local before=$(date +%s)
+	local evict
+
+	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
+		skip "Need MDS version at least 2.6.53" && return
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_bl_callback_once "chmod 0777 $DIR" ||
+		error "failed to chmod $DIR: $?"
+
+	# let the client reconnect
+	client_reconnect
+	evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+
+	[ -z "$evict" ] || [[ $evict -le $before ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "eviction happened: $evict before:$before")
+
+	do_facet client checkstat -v -p 0777 $DIR ||
+ 		error "client checkstat failed: $?"
+}
+run_test 10b "re-send BL AST"
+
+test_10c() {
+	local before=$(date +%s)
+	local evict
+	local mdccli
+	local mdcpath
+	local conn_uuid
+	local workdir
+	local pid
+	local rc
+
+	workdir="${DIR}/${tdir}"
+	mkdir -p ${workdir} || error "can't create workdir $?"
+	stat ${workdir} > /dev/null ||
+		error "failed to stat ${workdir}: $?"
+	mdtidx=$($LFS getdirstripe -i ${workdir})
+	mdtname=$($LFS mdts ${workdir} | grep -e "^$mdtidx:" |
+		  awk '{sub("_UUID", "", $2); print $2;}')
+	#assume one client
+	mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
+	conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+	mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
+
+	drop_bl_callback_once "chmod 0777 ${workdir}" &
+	pid=$!
+
+	# let chmod blocked
+	sleep 1
+	# force client reconnect
+	$LCTL set_param "${mdcpath}"
+
+	# wait client reconnect
+	client_reconnect
+	wait $pid
+	rc=$?
+	evict=$($LCTL get_param mdc.${mdccli}.state |
+	   awk -F"[ [,]" '/EVICTED]$/ { if (t<$4) {t=$4;} } END { print t }')
+
+	[[ $evict -le $before ]] ||
+		( $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "eviction happened: $EVICT before:$BEFORE" )
+
+	[ $rc -eq 0 ] || error "chmod must finished OK"
+	checkstat -v -p 0777 "${workdir}" ||
+		error "client checkstat failed: $?"
+}
+run_test 10c "re-send BL AST vs reconnect race (LU-5569)"
+
+test_10d() {
+	local before=$(date +%s)
+	local evict
+
+	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
+		skip "Need MDS version at least 2.6.90" && return
+
+	# sleep 1 is to make sure that BEFORE is not equal to EVICTED below
+	sleep 1
+	rm -f $TMP/$tfile
+	echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
+
+	mount_client $MOUNT2
+
+	cancel_lru_locks osc
+	$LFS setstripe -i 0 -c 1 $DIR1/$tfile
+	echo -n hello > $DIR1/$tfile
+
+	stat $DIR2/$tfile >& /dev/null
+	$LCTL set_param fail_err=71
+	drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
+
+	client_reconnect
+
+	cmp $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
+	cmp $DIR1/$tfile $TMP/$tfile || error "wrong content found"
+
+	evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
+		tr -d '\-\[\] ' | \
+	  awk -F"[ [,]" '/EVICTED$/ { if (mx<$1) {mx=$1;} } END { print mx }')
+
+	[[ $evict -gt $before ]] ||
+		(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
+		    error "no eviction: $evict before:$before")
+
+	$LCTL set_param fail_err=0
+	rm $TMP/$tfile
+	umount_client $MOUNT2
+}
+run_test 10d "test failed blocking ast"
 
 #bug 2460
 # wake up a thread waiting for completion after eviction
@@ -177,7 +296,8 @@ test_11(){
 
 	do_facet client $MULTIOP $DIR/$tfile or  ||
 		{ error "multiop read failed: $?"; return 3; }
-	drop_bl_callback $MULTIOP $DIR/$tfile Ow || echo "evicted as expected"
+	drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
+		echo "evicted as expected"
 
 	do_facet client munlink $DIR/$tfile ||
 		{ error "munlink failed: $?"; return 4; }
@@ -188,7 +308,7 @@ run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
 
 #b=2494
 test_12(){
-	$LCTL mark $MULTIOP $DIR/$tfile OS_c
+	$LCTL mark "$HOSTNAME: $MULTIOP $DIR/$tfile OS_c"
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
 	clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
 	multiop_bg_pause $DIR/$tfile OS_c ||
@@ -428,8 +548,9 @@ test_19a() {
 
 	# let the client reconnect
 	client_reconnect
-	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state | \
-	    awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+		awk -F"[ [,]" '/EVICTED ]$/ \
+			{ if (mx<$5) {mx=$5;} } END { print mx }')
 
 	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
 		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
@@ -457,8 +578,9 @@ test_19b() {
 
 	# let the client reconnect
 	client_reconnect
-	EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state | \
-	    awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
+		awk -F"[ [,]" '/EVICTED ]$/ \
+			{ if (mx < $5) {mx = $5;} } END { print mx }')
 
 	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
 		(do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
@@ -490,7 +612,7 @@ test_19c() {
 	# let the client reconnect
 	sleep 5
 	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
-	   awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
 
 	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
 }
@@ -909,7 +1031,8 @@ run_test 27 "fail LOV while using OSC's"
 
 test_28() {      # bug 6086 - error adding new clients
 	do_facet client mcreate $DIR/$tfile       || return 1
-	drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected"
+	drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
+		echo "evicted as expected"
 	#define OBD_FAIL_MDS_CLIENT_ADD 0x12f
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
 	# fail once (evicted), reconnect fail (fail_loc), ok
@@ -1123,11 +1246,11 @@ test_56() { # b=11277
 #define OBD_FAIL_MDS_RESEND      0x136
         touch $DIR/$tfile
         do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
-        stat $DIR/$tfile
+        stat $DIR/$tfile || error "stat failed"
         do_facet $SINGLEMDS "lctl set_param fail_loc=0"
         rm -f $DIR/$tfile
 }
-run_test 56 "do not allow reconnect to busy exports"
+run_test 56 "do not fail on getattr resend"
 
 test_57_helper() {
         # no oscs means no client or mdt 
@@ -1161,10 +1284,10 @@ test_58() { # bug 11546
         pid=$!
         sleep 1
         lctl set_param fail_loc=0
-        drop_bl_callback rm -f $DIR/$tfile
+        drop_bl_callback_once rm -f $DIR/$tfile
         wait $pid
         # the first 'df' could tigger the eviction caused by
-        # 'drop_bl_callback', and it's normal case.
+        # 'drop_bl_callback_once', and it's normal case.
         # but the next 'df' should return successfully.
         do_facet client "df $DIR" || do_facet client "df $DIR"
 }
@@ -1198,8 +1321,8 @@ err17935 () {
 }
 
 test_60() {
-        MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
-	    awk '{gsub(/_UUID/,""); print $1}' | head -1)
+	MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
+		awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
 
 	NUM_FILES=15000
 	mkdir -p $DIR/$tdir
@@ -1263,12 +1386,14 @@ test_61()
 
 	replay_barrier $SINGLEMDS
 	createmany -o $DIR/$tdir/$tfile-%d 10 
-	local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
+	local oid=$(do_facet ost1 "lctl get_param -n \
+		obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
 
 	fail_abort $SINGLEMDS
-	
+
 	touch $DIR/$tdir/$tfile
-	local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
+	local id=$($LFS getstripe $DIR/$tdir/$tfile |
+		awk '$1 == 0 { print $2 }')
 	[ $id -le $oid ] && error "the orphan objid was reused, failed"
 
 	# Cleanup
@@ -1286,6 +1411,72 @@ run_test 61 "Verify to not reuse orphan objects - bug 17025"
 #}
 #run_test 62 "Verify connection flags race - bug LU-1716"
 
+test_66()
+{
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
+		{ skip "Need MDS version at least 2.7.51"; return 0; }
+
+	local list=$(comma_list $(osts_nodes))
+
+	# modify dir so that next revalidate would not obtain UPDATE lock
+	touch $DIR
+
+	# drop 1 reply with UPDATE lock
+	mcreate $DIR/$tfile || error "mcreate failed: $?"
+	drop_ldlm_reply_once "stat $DIR/$tfile" &
+	sleep 2
+
+	# make the re-sent lock to sleep
+#define OBD_FAIL_MDS_RESEND              0x136
+	do_nodes $list $LCTL set_param fail_loc=0x80000136
+
+	#initiate the re-connect & re-send
+	local mdccli=$($LCTL dl | awk '/-mdc-/ {print $4;}')
+	local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+	$LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
+	sleep 2
+
+	#initiate the client eviction while enqueue re-send is in progress
+	mds_evict_client
+
+	client_reconnect
+	wait
+}
+run_test 66 "lock enqueue re-send vs client eviction"
+
+test_65() {
+	mount_client $DIR2
+
+	#grant lock1, export2
+	$SETSTRIPE -i -0 $DIR2/$tfile || return 1
+	$MULTIOP $DIR2/$tfile Ow  || return 2
+
+#define OBD_FAIL_LDLM_BL_EVICT            0x31e
+	do_facet ost $LCTL set_param fail_loc=0x31e
+	#get waiting lock2, export1
+	$MULTIOP $DIR/$tfile Ow &
+	PID1=$!
+	# let enqueue to get asleep
+	sleep 2
+
+	#get lock2 blocked
+	$MULTIOP $DIR2/$tfile Ow &
+	PID2=$!
+	sleep 2
+
+	#evict export1
+	ost_evict_client
+
+	sleep 2
+	do_facet ost $LCTL set_param fail_loc=0
+
+	wait $PID1
+	wait $PID2
+
+	umount_client $DIR2
+}
+run_test 65 "lock enqueue for destroyed export"
+
 check_cli_ir_state()
 {
         local NODE=${1:-$HOSTNAME}
@@ -1385,8 +1576,8 @@ target_instance_match()
 
         local target=${srv}_svc
         local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
-        local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
-                  awk '/instance/{ print $2 }' |head -1)
+	local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
+		awk '/instance/{ print $2 }' | head -n1)
 
         return $([ $si -eq $ci ])
 }
@@ -1566,10 +1757,9 @@ test_105()
         # get one of the clients from client list
         local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
 
-        local old_MOUNTOPT=$MOUNTOPT
-        MOUNTOPT=${MOUNTOPT},noir
+        local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
         zconf_umount $rcli $MOUNT || error "umount failed"
-        zconf_mount $rcli $MOUNT || error "mount failed"
+        zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
 
         # make sure lustre mount at $rcli disabling IR
         local ir_state=$(check_cli_ir_state $rcli)
@@ -1591,8 +1781,7 @@ test_105()
 	[ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
 		error "IR status on ost1 should be DISABLED"
 
-        # restore it
-        MOUNTOPT=$old_MOUNTOPT
+        # remount with the default MOUNT_OPTS
         zconf_umount $rcli $MOUNT || error "umount failed"
         zconf_mount $rcli $MOUNT || error "mount failed"
 
@@ -1681,13 +1870,38 @@ test_107 () {
 }
 run_test 107 "drop reint reply, then restart MDT"
 
+test_108() {
+	mkdir -p $DIR/$tdir
+	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
+
+	dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
+	local dd_pid=$!
+	sleep 0.1
+
+	ost_evict_client
+
+	wait $dd_pid
+
+	client_up || error "reconnect failed"
+	rm -f $DIR/$tdir/$tfile
+}
+run_test 108 "client eviction don't crash"
+
 test_110a () {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	local remote_dir=$DIR/$tdir/remote_dir
 	local MDTIDX=1
+	local num
+
+	#prepare for 110 test, which need set striped dir on remote MDT.
+	for num in $(seq $MDSCOUNT); do
+		do_facet mds$num \
+			lctl set_param -n mdt.${FSNAME}*.enable_remote_dir=1 \
+				2>/dev/null
+	done
 
 	mkdir -p $DIR/$tdir
-	drop_request "$LFS mkdir -i $MDTIDX $remote_dir" ||
+	drop_request "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
 					error "lfs mkdir failed"
 	local diridx=$($GETSTRIPE -M $remote_dir)
 	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
@@ -1702,7 +1916,7 @@ test_110b () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	drop_reint_reply "$LFS mkdir -i $MDTIDX $remote_dir" ||
+	drop_reint_reply "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
 					error "lfs mkdir failed"
 
 	diridx=$($GETSTRIPE -M $remote_dir)
@@ -1718,7 +1932,7 @@ test_110c () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX $remote_dir" ||
+	drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
 						error "lfs mkdir failed"
 
 	diridx=$($GETSTRIPE -M $remote_dir)
@@ -1734,7 +1948,7 @@ test_110d () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+	$LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
 
 	drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
 
@@ -1750,7 +1964,7 @@ test_110e () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	$LFS mkdir -i $MDTIDX $remote_dir  || error "lfs mkdir failed"
+	$LFS mkdir -i $MDTIDX -c2 $remote_dir  || error "lfs mkdir failed"
 	drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
@@ -1765,7 +1979,7 @@ test_110f () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+	$LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
 	drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
 					error "rm remote dir failed"
 
@@ -1780,11 +1994,11 @@ test_110g () {
 
 	mkdir -p $remote_dir
 
-	createmany -o $remote_dir/f 5000
+	createmany -o $remote_dir/f 100
 
-	#define OBD_FAIL_MIGRATE_NET_REP	0x1702
-	do_facet mds$MDTIDX lctl set_param fail_loc=0x1702
-	$LFS mv -M $MDTIDX $remote_dir || error "migrate failed"
+	#define OBD_FAIL_MIGRATE_NET_REP		0x1800
+	do_facet mds$MDTIDX lctl set_param fail_loc=0x1800
+	$LFS migrate -m $MDTIDX $remote_dir || error "migrate failed"
 	do_facet mds$MDTIDX lctl set_param fail_loc=0x0
 
 	for file in $(find $remote_dir); do
@@ -1797,6 +2011,77 @@ test_110g () {
 }
 run_test 110g "drop reply during migration"
 
+test_110h () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local src_dir=$DIR/$tdir/source_dir
+	local tgt_dir=$DIR/$tdir/target_dir
+	local MDTIDX=1
+
+	mkdir -p $src_dir
+	$LFS mkdir -i $MDTIDX $tgt_dir
+
+	dd if=/etc/hosts of=$src_dir/src_file
+	touch $tgt_dir/tgt_file
+	drop_update_reply $MDTIDX \
+		"mrename $src_dir/src_file $tgt_dir/tgt_file" ||
+		error "mrename failed"
+
+	$CHECKSTAT -t file $src_dir/src_file &&
+				error "src_file present after rename"
+
+	diff /etc/hosts $tgt_dir/tgt_file ||
+			error "file changed after rename"
+
+}
+run_test 110h "drop update reply during cross-MDT file rename"
+
+test_110i () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local src_dir=$DIR/$tdir/source_dir
+	local tgt_dir=$DIR/$tdir/target_dir
+	local MDTIDX=1
+
+	mkdir -p $src_dir
+	$LFS mkdir -i $MDTIDX $tgt_dir
+
+	mkdir $src_dir/src_dir
+	touch $src_dir/src_dir/a
+	mkdir $tgt_dir/tgt_dir
+	drop_update_reply $MDTIDX \
+		"mrename $src_dir/src_dir $tgt_dir/tgt_dir" ||
+		error "mrename failed"
+
+	$CHECKSTAT -t dir $src_dir/src_dir &&
+			error "src_dir present after rename"
+
+	$CHECKSTAT -t dir $tgt_dir/tgt_dir ||
+				error "tgt_dir not present after rename"
+
+	$CHECKSTAT -t file $tgt_dir/tgt_dir/a ||
+				error "a not present after rename"
+}
+run_test 110i "drop update reply during cross-MDT dir rename"
+
+test_110j () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local local_dir=$DIR/$tdir/local_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	mkdir $DIR/$tdir/local_dir
+	$LFS mkdir -i $MDTIDX $remote_dir
+
+	touch $local_dir/local_file
+	drop_update_reply $MDTIDX \
+		"ln $local_dir/local_file $remote_dir/remote_file" ||
+		error "ln failed"
+
+	$CHECKSTAT -t file $remote_dir/remote_file ||
+				error "remote not present after ln"
+}
+run_test 110j "drop update reply during cross-MDT ln"
+
 # LU-2844 mdt prepare fail should not cause umount oops
 test_111 ()
 {
@@ -1833,6 +2118,232 @@ test_112a() {
 }
 run_test 112a "bulk resend while orignal request is in progress"
 
+# parameters: fail_loc CMD RC
+test_120_reply() {
+	local PID
+	local PID2
+	local rc=5
+	local fail
+
+	#define OBD_FAIL_LDLM_CP_CB_WAIT2	0x320
+	#define OBD_FAIL_LDLM_CP_CB_WAIT3	0x321
+	#define OBD_FAIL_LDLM_CP_CB_WAIT4	0x322
+	#define OBD_FAIL_LDLM_CP_CB_WAIT5	0x323
+
+	echo
+	echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
+	[ "$1" = "CLEANUP" ] &&
+		fail=0x80000320 && echo ", $1 cp first"
+	[ "$1" = "REPLY" ] &&
+		fail=0x80000321 && echo ", $1 cp first"
+	[ "$1" = "DEADLOCK CLEANUP" ] &&
+		fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
+	[ "$1" = "DEADLOCK REPLY" ] &&
+		fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
+
+	if [ x"$2" = x"get" ]; then
+		#for TEST lock, take a conflict in advance
+		# sleep longer than evictor to not confuse fail_loc: 2+2+4
+		echo "** Taking conflict **"
+		flocks_test 5 set read sleep 10 $DIR/$tfile &
+		PID2=$!
+
+		sleep 2
+	fi
+
+	$LCTL set_param fail_loc=$fail
+
+	flocks_test 5 $2 write $DIR/$tfile &
+	PID=$!
+
+	sleep 2
+	echo "** Evicting and re-connecting client **"
+	mds_evict_client
+
+	client_reconnect
+
+	if [ x"$2" = x"get" ]; then
+		wait $PID2
+	fi
+
+	wait $PID
+	rc=$?
+
+	# check if the return value is allowed
+	[ $rc -eq $3 ] && rc=0
+
+	$LCTL set_param fail_loc=0
+	return $rc
+}
+
+# a lock is taken, unlock vs. cleanup_resource() race for destroying
+# the ORIGINAL lock.
+test_120_destroy()
+{
+	local PID
+
+	flocks_test 5 set write sleep 4 $DIR/$tfile &
+	PID=$!
+	sleep 2
+
+	# let unlock to sleep in CP CB
+	$LCTL set_param fail_loc=$1
+	sleep 4
+
+	# let cleanup to cleep in CP CB
+	mds_evict_client
+
+	client_reconnect
+
+	wait $PID
+	rc=$?
+
+	$LCTL set_param fail_loc=0
+	return $rc
+}
+
+test_120() {
+	flock_is_enabled || { skip "mount w/o flock enabled" && return; }
+	touch $DIR/$tfile
+
+	test_120_reply "CLEANUP" set 5 || error "SET race failed"
+	test_120_reply "CLEANUP" get 5 || error "GET race failed"
+	test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
+
+	test_120_reply "REPLY" set 5 || error "SET race failed"
+	test_120_reply "REPLY" get 5 || error "GET race failed"
+	test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
+
+	# DEADLOCK tests
+	test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
+	test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
+
+	test_120_destroy 0x320 || error "unlock-cleanup race failed"
+}
+run_test 120 "flock race: completion vs. evict"
+
+test_113() {
+	local BEFORE=$(date +%s)
+	local EVICT
+
+	# modify dir so that next revalidate would not obtain UPDATE lock
+	touch $DIR
+
+	# drop 1 reply with UPDATE lock,
+	# resend should not create 2nd lock on server
+	mcreate $DIR/$tfile || error "mcreate failed: $?"
+	drop_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
+
+	# 2 BL AST will be sent to client, both must find the same lock,
+	# race them to not get EINVAL for 2nd BL AST
+	#define OBD_FAIL_LDLM_PAUSE_CANCEL2      0x31f
+	$LCTL set_param fail_loc=0x8000031f
+
+	$LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
+	chmod 0777 $DIR/$tfile || error "chmod failed: $?"
+	$LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
+
+	# let the client reconnect
+	client_reconnect
+	EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+
+	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
+}
+run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
+
+T130_PID=0
+test_130_base() {
+	test_mkdir -p $DIR/$tdir
+
+	# Prevent interference from layout intent RPCs due to
+	# asynchronous writeback. These will be tested in 130c below.
+	do_nodes ${CLIENTS:-$HOSTNAME} sync
+
+	# get only LOOKUP lock on $tdir
+	cancel_lru_locks mdc
+	ls $DIR/$tdir/$tfile 2>/dev/null
+
+	# get getattr by fid on $tdir
+	#
+	# we need to race with unlink, unlink must complete before we will
+	# take a DLM lock, otherwise unlink will wait until getattr will
+	# complete; but later than getattr starts so that getattr found
+	# the object
+#define OBD_FAIL_MDS_INTENT_DELAY		0x160
+	set_nodes_failloc "$(mdts_nodes)" 0x80000160
+	stat $DIR/$tdir &
+	T130_PID=$!
+	sleep 2
+
+	rm -rf $DIR/$tdir
+
+	# drop the reply so that resend happens on an unlinked file.
+#define OBD_FAIL_MDS_LDLM_REPLY_NET	 0x157
+	set_nodes_failloc "$(mdts_nodes)" 0x80000157
+}
+
+test_130a() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	test_130_base
+
+	wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+	return 0
+}
+run_test 130a "enqueue resend on not existing file"
+
+test_130b() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	test_130_base
+	# let the reply to be dropped
+	sleep 10
+
+#define OBD_FAIL_SRV_ENOENT              0x217
+	set_nodes_failloc "$(mdts_nodes)" 0x80000217
+
+	wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+	return 0
+}
+run_test 130b "enqueue resend on a stale inode"
+
+test_130c() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+
+	do_nodes ${CLIENTS:-$HOSTNAME} sync
+	echo XXX > $DIR/$tfile
+
+	cancel_lru_locks mdc
+
+	# Trigger writeback on $tfile.
+	#
+	# we need to race with unlink, unlink must complete before we will
+	# take a DLM lock, otherwise unlink will wait until intent will
+	# complete; but later than intent starts so that intent found
+	# the object
+#define OBD_FAIL_MDS_INTENT_DELAY		0x160
+	set_nodes_failloc "$(mdts_nodes)" 0x80000160
+	sync &
+	T130_PID=$!
+	sleep 2
+
+	rm $DIR/$tfile
+
+	# drop the reply so that resend happens on an unlinked file.
+#define OBD_FAIL_MDS_LDLM_REPLY_NET	 0x157
+	set_nodes_failloc "$(mdts_nodes)" 0x80000157
+
+	# let the reply to be dropped
+	sleep 10
+
+#define OBD_FAIL_SRV_ENOENT              0x217
+	set_nodes_failloc "$(mdts_nodes)" 0x80000217
+
+	wait $T130_PID
+
+	return 0
+}
+run_test 130c "layout intent resend on a stale inode"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status