X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=c311e813b7bbacea365a737082781dd92673468b;hp=9641565916b6a7ce35051ab38d9213261de55b31;hb=44cb3a92d756728195409b368b3e89b1a78ef0ae;hpb=6e704244293aec98b39b8243fb2d13e2a8c9b674

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 9641565..c311e81 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -16,12 +16,8 @@ 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"
 
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:	      LU-2194
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
 
 build_test_filter
 
@@ -149,21 +145,181 @@ 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: $?"
+}
+run_test 10a "finish request on server after client eviction (bug 1521)"
+
+test_10b() {
+	local before=$(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
+
+	remount_client $MOUNT
+	mount_client $MOUNT2
+
+	cancel_lru_locks osc
+	$LFS setstripe -i 0 -c 1 $DIR1/$tfile
+	echo -n hello | dd of=$DIR1/$tfile bs=5
+
+	stat $DIR2/$tfile >& /dev/null
+	$LCTL set_param fail_err=71
+	drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
+
+	client_reconnect
+
+	cancel_lru_locks osc
+	cmp -l $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
+	cmp -l $DIR1/$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 10 "finish request on server after client eviction (bug 1521)"
+run_test 10d "test failed blocking ast"
+
+test_10e()
+{
+	[[ $(lustre_version_code ost1) -le $(version_code 2.8.58) ]] &&
+		skip "Need OST version at least 2.8.59" && return 0
+	[ $CLIENTCOUNT -lt 2 ] && skip "need two clients" && return 0
+	[ $(facet_host client) == $(facet_host ost1) ] &&
+		skip "need ost1 and client on different nodes" && return 0
+	local -a clients=(${CLIENTS//,/ })
+	local client1=${clients[0]}
+	local client2=${clients[1]}
+
+	$LFS setstripe -c 1 -i 0 $DIR/$tfile-1 $DIR/$tfile-2
+	$MULTIOP $DIR/$tfile-1 Ow1048576c
+
+#define OBD_FAIL_LDLM_BL_CALLBACK_NET                   0x305
+	$LCTL set_param fail_loc=0x80000305
+
+#define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
+	do_facet ost1 "$LCTL set_param fail_loc=0x1000030e"
+	# hit OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT twice:
+	# 1. to return ENOTCONN from ldlm_handle_enqueue0
+	# 2. to pause reconnect handling between resend and setting
+	# import to LUSTRE_IMP_FULL state
+	do_facet ost1 "$LCTL set_param fail_val=3"
+
+	# client1 fails ro respond to bl ast
+	do_node $client2 "$MULTIOP $DIR/$tfile-1 Ow1048576c" &
+	MULTIPID=$!
+
+	# ost1 returns error on enqueue, which causes client1 to reconnect
+	do_node $client1 "$MULTIOP $DIR/$tfile-2 Ow1048576c" ||
+		error "multiop failed"
+	wait $MULTIPID
+
+	do_facet ost1 "$LCTL set_param fail_loc=0"
+	do_facet ost1 "$LCTL set_param fail_val=0"
+}
+run_test 10e "re-send BL AST vs reconnect race 2"
 
 #bug 2460
 # wake up a thread waiting for completion after eviction
@@ -177,7 +333,8 @@ 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; }
@@ -264,10 +421,11 @@ test_16() {
 	sleep $TIMEOUT
 	do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
 	start_read_ahead
+	rm -f $TMP/$tfile
 }
 run_test 16 "timeout bulk put, don't evict client (2732)"
 
-test_17() {
+test_17a() {
     local at_max_saved=0
 
     remote_ost_nodsh && skip "remote OST with nodsh" && return 0
@@ -301,7 +459,93 @@ test_17() {
     [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
     return 0
 }
-run_test 17 "timeout bulk get, don't evict client (2732)"
+run_test 17a "timeout bulk get, don't evict client (2732)"
+
+test_17b() {
+	[ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
+
+	# get one of the clients from client list
+	local rcli=$(echo $RCLIENTS | cut -d ' ' -f 1)
+	local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
+	local ldlm_enqueue_min=$(do_facet ost1 find /sys -name ldlm_enqueue_min)
+	[ -z "$ldlm_enqueue_min" ] &&
+		skip "missing /sys/.../ldlm_enqueue_min" && return 0
+
+	$LFS setstripe -i 0 -c 1 -S 1048576 $DIR/$tfile ||
+		error "setstripe failed"
+	$LFS setstripe -i 0 -c 1 -S 1048576 $DIR/${tfile}2 ||
+		error "setstripe 2 failed"
+
+	save_lustre_params ost1 "at_history" > $p
+	save_lustre_params ost1 "bulk_timeout" >> $p
+	local dev="${FSNAME}-OST0000"
+	save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
+	local ldlm_enqueue_min_save=$(do_facet ost1 cat $ldlm_enqueue_min)
+
+	local new_at_history=15
+
+	do_facet ost1 "$LCTL set_param at_history=$new_at_history"
+	do_facet ost1 "$LCTL set_param bulk_timeout=30"
+	do_facet ost1 "echo 30 > /sys/module/ptlrpc/parameters/ldlm_enqueue_min"
+
+	# brw_size is required to be 4m so that bulk transfer timeout
+	# could be simulated with OBD_FAIL_PTLRPC_CLIENT_BULK_CB3
+	local brw_size=$($LCTL get_param -n \
+	    osc.$FSNAME-OST0000-osc-[^M]*.import |
+	    awk '/max_brw_size/{print $2}')
+	if [ $brw_size -ne 4194304 ]
+	then
+		save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
+		do_facet ost1 "$LCTL set_param obdfilter.$dev.brw_size=4"
+		remount_client $MOUNT
+	fi
+
+	# get service estimate expanded
+	#define OBD_FAIL_OST_BRW_PAUSE_PACK		 0x224
+	do_facet ost1 "$LCTL set_param fail_loc=0x80000224 fail_val=35"
+	echo "delay rpc servicing by 35 seconds"
+	dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=fdatasync
+
+	local estimate
+	estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
+	    awk '/portal 6/ {print $5}')
+	echo "service estimates increased to $estimate"
+
+	# let current worst service estimate to get closer to obliteration
+	sleep $((new_at_history / 3))
+
+	# start i/o and simulate bulk transfer loss
+	#define OBD_FAIL_PTLRPC_CLIENT_BULK_CB3     0x520
+	do_facet ost1 "$LCTL set_param fail_loc=0xa0000520 fail_val=1"
+	dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 conv=fdatasync,notrunc &
+	local writedd=$!
+
+	# start lock conflict handling
+	sleep $((new_at_history / 3))
+	do_node $rcli "dd if=$DIR/$tfile of=/dev/null bs=1M count=1" &
+	local readdd=$!
+
+	# obliterate the worst service estimate
+	sleep $((new_at_history / 3 + 1))
+	dd if=/dev/zero of=$DIR/${tfile}2 bs=1M count=1
+
+	estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
+	    awk '/portal 6/ {print $5}')
+	echo "service estimate dropped to $estimate"
+
+	wait $writedd
+	[[ $? == 0 ]] || error "write failed"
+	wait $readdd
+	[[ $? == 0 ]] || error "read failed"
+
+	restore_lustre_params <$p
+	if [ $brw_size -ne 4194304 ]
+	then
+		remount_client $MOUNT || error "remount_client failed"
+	fi
+	do_facet ost1 "echo $ldlm_enqueue_min_save > $ldlm_enqueue_min"
+}
+run_test 17b "timeout bulk get, dont evict client (3582)"
 
 test_18a() {
     [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
@@ -332,7 +576,7 @@ test_18a() {
     rc=0
     pgcache_empty || rc=2
     $LCTL --device $osc2dev activate
-    rm -f $f
+    rm -f $f $TMP/$tfile
     return $rc
 }
 run_test 18a "manual ost invalidate clears page cache immediately"
@@ -365,7 +609,7 @@ test_18b() {
     # cache after the client reconnects?     
     rc=0
     pgcache_empty || rc=2
-    rm -f $f
+    rm -f $f $TMP/$tfile
     return $rc
 }
 run_test 18b "eviction and reconnect clears page cache (2766)"
@@ -398,13 +642,13 @@ test_18c() {
     do_facet ost1 lctl set_param fail_loc=0x80000225
     # force reconnect
     sleep 1
-    df $MOUNT > /dev/null 2>&1
+    $LFS df $MOUNT > /dev/null 2>&1
     sleep 2
     # my understanding is that there should be nothing in the page
-    # cache after the client reconnects?     
+    # cache after the client reconnects?
     rc=0
     pgcache_empty || rc=2
-    rm -f $f
+    rm -f $f $TMP/$tfile
     return $rc
 }
 run_test 18c "Dropped connect reply after eviction handing (14755)"
@@ -427,8 +671,9 @@ 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;
@@ -456,8 +701,9 @@ 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;
@@ -489,7 +735,7 @@ 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 $BEFORE ]] || error "eviction happened"
 }
@@ -767,7 +1013,8 @@ test_24a() { # bug 11710 details correct fsync() behavior
 	rc=$?
 	lctl set_param fail_loc=0x0
 	client_reconnect
-	[ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
+	[ $rc -eq 0 ] &&
+		error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true
 }
 run_test 24a "fsync error (should return error)"
 
@@ -786,7 +1033,8 @@ test_24b() {
 
 	dmesg -c > /dev/null
 	mkdir -p $DIR/$tdir
-	lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
+	lfs setstripe $DIR/$tdir -S 0 -i 0 -c 1 ||
+		error "$LFS setstripe failed"
 	cancel_lru_locks osc
 	multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
 		error "mulitop Ow8192_yc failed"
@@ -807,10 +1055,10 @@ test_24b() {
 	lctl set_param fail_loc=0x0
 	client_reconnect
 	[ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
-	error_ignore 5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
+	error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
 		true
 
-	dmesg | grep "dirty page discard:" || \
+	dmesg | grep "dirty page discard:" ||
 		error "no discarded dirty page found!"
 }
 run_test 24b "test dirty page discard due to client eviction"
@@ -828,20 +1076,21 @@ test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
 
 	check_timeout || return 1
 
-	local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
-
-	echo starting with $OST_NEXP OST exports
 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
 	do_facet client lctl set_param fail_loc=0x505
-        # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
-        # But if there's a race to start the evictor from various obds,
-        # the loser might have to wait for the next ping.
-
+	local before=$(date +%s)
 	local rc=0
-	wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
-	rc=$?
+
+	# evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
+	# But if there's a race to start the evictor from various obds,
+	# the loser might have to wait for the next ping.
+	sleep $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
 	do_facet client lctl set_param fail_loc=0x0
-        [ $rc -eq 0 ] || error "client not evicted from OST"
+	do_facet client df > /dev/null
+
+	local oscs=$(lctl dl | awk '/-osc-/ {print $4}')
+	check_clients_evicted $before ${oscs[@]}
+	check_clients_full 10 ${oscs[@]}
 }
 run_test 26a "evict dead exports"
 
@@ -907,7 +1156,8 @@ 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
@@ -923,7 +1173,7 @@ test_29a() { # bug 22273 - error adding new clients
 	# fail abort so client will be new again
 	fail_abort $SINGLEMDS
 	client_up || error "reconnect failed"
-	wait_osc_import_state mds ost FULL
+	wait_osc_import_state $SINGLEMDS ost FULL
 	return 0
 }
 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
@@ -959,7 +1209,8 @@ test_50() {
 	rc=$?
 	echo writemany returned $rc
 	#these may fail because of eviction due to slow AST response.
-	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
+	[ $rc -eq 0 ] ||
+		error_ignore bz13652 "writemany returned rc $rc" || true
 }
 run_test 50 "failover MDS under load"
 
@@ -988,10 +1239,11 @@ test_51() {
 	# and recovery was interrupted
 	sleep $TIMEOUT
         kill -USR1 $CLIENT_PID
-	wait $CLIENT_PID 
+	wait $CLIENT_PID
 	rc=$?
 	echo writemany returned $rc
-	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
+	[ $rc -eq 0 ] ||
+		error_ignore bz13652 "writemany returned rc $rc" || true
 }
 run_test 51 "failover MDS during recovery"
 
@@ -1060,57 +1312,58 @@ test_55() {
 
 	mkdir -p $DIR/$tdir
 
+	# Minimum pass speed is 2MBps
+	local ddtimeout=64
+	# LU-2887/LU-3089 - set min pass speed to 500KBps
+	[ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
+
 	# first dd should be finished quickly
 	$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
-	dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4  &
+	dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
 	DDPID=$!
 	count=0
 	echo  "step1: testing ......"
-	while [ true ]; do
-	    if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
-	    count=$[count+1]
-	    if [ $count -gt 64 ]; then
-		error "dd should be finished!"
-	    fi
-	    sleep 1
-	done	
+	while kill -0 $DDPID 2> /dev/null; do
+		let count++
+		if [ $count -gt $ddtimeout ]; then
+			error "dd should be finished!"
+		fi
+		sleep 1
+	done
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
 	$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
 	#define OBD_FAIL_OST_DROP_REQ            0x21d
 	do_facet ost1 lctl set_param fail_loc=0x0000021d
 	# second dd will be never finished
-	dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4  &	
+	dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
 	DDPID=$!
 	count=0
 	echo  "step2: testing ......"
-	while [ $count -le 64 ]; do
-	    dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"	    
-	    if [ -z  $dd_name ]; then 
-                ls -l $DIR/$tdir
-		echo  "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
-		error "dd shouldn't be finished!"
-	    fi
-	    count=$[count+1]
-	    sleep 1
-	done	
+	while [ $count -le $ddtimeout ]; do
+		if ! kill -0 $DDPID 2> /dev/null; then
+			ls -l $DIR/$tdir
+			error "dd shouldn't be finished! (time=$count)"
+		fi
+		let count++
+		sleep 1
+	done
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
 	#Recover fail_loc and dd will finish soon
 	do_facet ost1 lctl set_param fail_loc=0
 	count=0
 	echo  "step3: testing ......"
-	while [ true ]; do
-	    if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
-	    count=$[count+1]
-	    if [ $count -gt 500 ]; then
-		error "dd should be finished!"
-	    fi
-	    sleep 1
-	done	
+	while kill -0 $DDPID 2> /dev/null; do
+		let count++
+		if [ $count -gt $((ddtimeout + 440)) ]; then
+			error "dd should be finished!"
+		fi
+		sleep 1
+	done
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
-        rm -rf $DIR/$tdir
+	rm -rf $DIR/$tdir
 }
 run_test 55 "ost_brw_read/write drops timed-out read/write request"
 
@@ -1118,11 +1371,11 @@ 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 
@@ -1156,10 +1409,10 @@ 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"
 }
@@ -1184,17 +1437,17 @@ test_59() { # bug 10589
 run_test 59 "Read cancel race on client eviction"
 
 err17935 () {
-    # we assume that all md changes are in the MDT0 changelog
-    if [ $MDSCOUNT -gt 1 ]; then
-	error_ignore 17935 $*
-    else
-	error $*
-    fi
+	# we assume that all md changes are in the MDT0 changelog
+	if [ $MDSCOUNT -gt 1 ]; then
+		error_ignore bz17935 $*
+	else
+		error $*
+	fi
 }
 
 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
@@ -1258,12 +1511,14 @@ 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
@@ -1281,6 +1536,72 @@ 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:-$HOSTNAME}
@@ -1299,6 +1620,11 @@ check_target_ir_state()
         local recovery_proc=obdfilter.${!name}.recovery_status
         local st
 
+	while : ; do
+		st=$(do_facet $target "$LCTL get_param -n $recovery_proc |
+			awk '/status:/{ print \\\$2}'")
+		[ x$st = xRECOVERING ] || break
+	done
         st=$(do_facet $target "lctl get_param -n $recovery_proc |
                                awk '/IR:/{ print \\\$2}'")
 	[ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
@@ -1380,8 +1706,8 @@ 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 ])
 }
@@ -1495,7 +1821,7 @@ test_103()
         do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
                 { skip "MGS without IR support"; return 0; }
 
-        combined_mgs_mds && skip "mgs and mds on the same target" && return 0
+	combined_mgs_mds && skip "needs separate mgs and mds" && return 0
 
         # workaround solution to generate config log on the mds
         remount_facet mds1
@@ -1504,10 +1830,11 @@ test_103()
         stop mds1
 
         # We need this test because mds is like a client in IR context.
-        start mds1 $MDSDEV1 || error "MDS should start w/o mgs"
+        start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
+		error "MDS should start w/o mgs"
 
         # start mgs and remount mds w/ ir
-        start mgs $MGSDEV
+        start mgs $(mgsdevname) $MGS_MOUNT_OPTS
         clients_up
 
         # remount client so that fsdb will be created on the MGS
@@ -1560,10 +1887,9 @@ 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)
@@ -1585,8 +1911,7 @@ 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"
 
@@ -1675,13 +2000,38 @@ 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"
@@ -1696,7 +2046,7 @@ 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)
@@ -1712,7 +2062,7 @@ test_110c () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	drop_update_reply $((MDTIDX + 1)) "$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)
@@ -1728,7 +2078,7 @@ 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"
 
@@ -1744,7 +2094,7 @@ 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"
@@ -1759,7 +2109,7 @@ 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"
 
@@ -1767,8 +2117,110 @@ test_110f () {
 }
 run_test 110f "remove remote directory: drop slave rep"
 
+test_110g () {
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.6.57) ]] ||
+		{ skip "Need MDS version at least 2.6.57"; return 0; }
+
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $remote_dir
+
+	createmany -o $remote_dir/f 100
+
+	#define OBD_FAIL_MIGRATE_NET_REP		0x1800
+	do_facet mds$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 $remote_dir); do
+		mdt_index=$($LFS getstripe -M $file)
+		[ $mdt_index == $MDTIDX ] ||
+			error "$file is not on MDT${MDTIDX}"
+	done
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+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 () {
+test_111 ()
+{
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
+		{ skip "Need MDS version at least 2.3.62"; return 0; }
+
 	local mdsdev=$(mdsdevname ${SINGLEMDS//mds/})
 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
 	do_facet $SINGLEMDS lctl set_param fail_loc=0x151
@@ -1779,6 +2231,461 @@ test_111 () {
 }
 run_test 111 "mdd setup fail should not cause umount oops"
 
+# LU-793
+test_112a() {
+	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+	do_facet_random_file client $TMP/$tfile 100K ||
+		error_noexit "Create random file $TMP/$tfile"
+
+	pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT ||
+		error_noexit "Can't pause_bulk copy"
+
+	df $DIR
+	# expect cmp to succeed, client resent bulk
+	cmp $TMP/$tfile $DIR/$tfile ||
+		error_noexit "Wrong data has been written"
+	rm $DIR/$tfile ||
+		error_noexit "Can't remove file"
+	rm $TMP/$tfile
+}
+run_test 112a "bulk resend while orignal request is in progress"
+
+test_115_read() {
+	local fail1=$1
+	local fail2=$2
+
+	df $DIR
+	dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
+	cancel_lru_locks osc
+
+	# OST_READ       =  3,
+	$LCTL set_param fail_loc=$fail1 fail_val=3
+	dd of=/dev/null if=$DIR/$tfile bs=4096 count=1 &
+	pid=$!
+	sleep 1
+
+	set_nodes_failloc "$(osts_nodes)" $fail2
+
+	wait $pid || error "dd failed"
+	return 0
+}
+
+test_115_write() {
+	local fail1=$1
+	local fail2=$2
+	local error=$3
+	local fail_val2=${4:-0}
+
+	df $DIR
+	touch $DIR/$tfile
+
+	# OST_WRITE      =  4,
+	$LCTL set_param fail_loc=$fail1 fail_val=4
+	dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=dsync &
+	pid=$!
+	sleep 1
+
+	df $MOUNT
+	set_nodes_failloc "$(osts_nodes)" $fail2 $fail_val2
+
+	wait $pid
+	rc=$?
+	[ $error -eq 0 ] && [ $rc -ne 0 ] && error "dd error ($rc)"
+	[ $error -ne 0 ] && [ $rc -eq 0 ] && error "dd success"
+	return 0
+}
+
+test_115a() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
+	#define OBD_FAIL_PTLRPC_DROP_BULK	 0x51a
+	test_115_read 0x8000051b 0x8000051a
+}
+run_test 115a "read: late REQ MDunlink and no bulk"
+
+test_115b() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
+	#define OBD_FAIL_OST_ENOSPC              0x215
+
+	# pass $OSTCOUNT for the fail_loc to be caught
+	# appropriately by the IO thread
+	test_115_write 0x8000051b 0x80000215 1 $OSTCOUNT
+}
+run_test 115b "write: late REQ MDunlink and no bulk"
+
+test_115c() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
+	#define OBD_FAIL_PTLRPC_DROP_BULK	 0x51a
+	test_115_read 0x8000050f 0x8000051a
+}
+run_test 115c "read: late Reply MDunlink and no bulk"
+
+test_115d() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
+	#define OBD_FAIL_OST_ENOSPC              0x215
+	test_115_write 0x8000050f 0x80000215 0
+}
+run_test 115d "write: late Reply MDunlink and no bulk"
+
+test_115e() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
+	#define OBD_FAIL_OST_ALL_REPLY_NET       0x211
+	test_115_read 0x80000510 0x80000211
+}
+run_test 115e "read: late Bulk MDunlink and no reply"
+
+test_115f() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
+	#define OBD_FAIL_OST_ALL_REPLY_NET       0x211
+	test_115_read 0x8000051b 0x80000211
+}
+run_test 115f "read: late REQ MDunlink and no reply"
+
+test_115g() {
+	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST" && return 0
+
+	#define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
+	test_115_read 0x8000051c 0
+}
+run_test 115g "read: late REQ MDunlink and Reply MDunlink"
+
+# 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 "$(mdts_nodes)" 0x80000157
+}
+
+test_130a() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	local server_version=$(lustre_version_code $SINGLEMDS)
+	[[ $server_version -ge $(version_code 2.7.2) ]] ||
+		{ skip "Need server version newer than 2.7.1"; return 0; }
+
+	test_130_base
+
+	wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+	return 0
+}
+run_test 130a "enqueue resend on not existing file"
+
+test_130b() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	local server_version=$(lustre_version_code $SINGLEMDS)
+	[[ $server_version -ge $(version_code 2.7.2) ]] ||
+		{ skip "Need server version newer than 2.7.1"; return 0; }
+
+	test_130_base
+	# let the reply to be dropped
+	sleep 10
+
+#define OBD_FAIL_SRV_ENOENT              0x217
+	set_nodes_failloc "$(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 $T130_PID
+
+	return 0
+}
+run_test 130c "layout intent resend on a stale inode"
+
+test_132() {
+	local before=$(date +%s)
+	local evict
+
+	mount_client $MOUNT2 || error "mount filed"
+
+	rm -f $DIR/$tfile
+	# get a lock on client so that export would reach the stale list
+	$SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+	dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
+		error "dd failed"
+
+	#define OBD_FAIL_OST_PAUSE_PUNCH         0x236
+	do_facet ost1 $LCTL set_param fail_val=120 fail_loc=0x80000236
+
+	$TRUNCATE $DIR/$tfile 100 &
+
+	sleep 1
+	dd if=/dev/zero of=$DIR2/$tfile bs=4096 count=1 conv=notrunc ||
+		error "dd failed"
+
+	wait
+	umount_client $MOUNT2
+
+	evict=$(do_facet client $LCTL get_param \
+		osc.$FSNAME-OST0000-osc-*/state |
+	    awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
+
+	[ -z "$evict" ] || [[ $evict -le $before ]] ||
+		(do_facet client $LCTL get_param \
+			osc.$FSNAME-OST0000-osc-*/state;
+		    error "eviction happened: $evict before:$before")
+}
+run_test 132 "long punch"
+
+test_131() {
+	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+	rm -f $DIR/$tfile
+	# get a lock on client so that export would reach the stale list
+	$SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+	dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
+
+	# another IO under the same lock
+	#define OBD_FAIL_OSC_DELAY_IO            0x414
+	$LCTL set_param fail_loc=0x80000414
+	$LCTL set_param fail_val=4 fail_loc=0x80000414
+	dd if=/dev/zero of=$DIR/$tfile count=1 conv=notrunc oflag=dsync &
+	local pid=$!
+	sleep 1
+
+	#define OBD_FAIL_LDLM_BL_EVICT           0x31e
+	set_nodes_failloc "$(osts_nodes)" 0x8000031e
+	ost_evict_client
+	client_reconnect
+
+	wait $pid && error "dd succeeded"
+	return 0
+}
+run_test 131 "IO vs evict results to IO under staled lock"
+
+test_133() {
+	local list=$(comma_list $(mdts_nodes))
+
+	local t=$((TIMEOUT * 2))
+	touch $DIR/$tfile
+
+	flock $DIR/$tfile -c "echo bl lock;sleep $t;echo bl flock unlocked" &
+	sleep 1
+	multiop_bg_pause $DIR/$tfile O_jc || return 1
+	PID=$!
+
+	#define OBD_FAIL_LDLM_REPLY              0x30c
+	do_nodes $list $LCTL set_param fail_loc=0x8000030c
+	kill -USR1 $PID
+	echo "waiting for multiop $PID"
+	wait $PID || return 2
+
+	rm -f $DIR/$tfile
+
+	return 0
+}
+run_test 133 "don't fail on flock resend"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status