X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=891ac23294c38fb48d770bcbee429b749ee99448;hp=e833e5880539452d5d94c1b9c35287b2e21d933d;hb=fba6abdb9818b01d02ac7663e4ac9881258c8ead;hpb=8d1e7317b3ce28a6909e139bee5ad84b787d7c7f

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index e833e58..891ac23 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -2,9 +2,6 @@
 
 set -e
 
-#         bug  5493  LU2034
-ALWAYS_EXCEPT="52    60      $RECOVERY_SMALL_EXCEPT"
-
 export MULTIOP=${MULTIOP:-multiop}
 PTLDEBUG=${PTLDEBUG:--1}
 LUSTRE=${LUSTRE:-`dirname $0`/..}
@@ -13,16 +10,17 @@ init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 init_logging
 
+ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT "
+# bug number for skipped test:
+ALWAYS_EXCEPT+="               "
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+
 require_dsh_mds || exit 0
 
 # also long tests: 19, 21a, 21e, 21f, 23, 27
 
 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
 
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:	      LU-2547
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 24a 24b"
-
 build_test_filter
 
 # Allow us to override the setup if we already have a mounted system by
@@ -175,8 +173,8 @@ test_10b() {
 	local before=$(date +%s)
 	local evict
 
-	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
-		skip "Need MDS version at least 2.6.53" && return
+	[[ "$MDS1_VERSION" -lt $(version_code 2.6.53) ]] &&
+		skip "Need MDS version at least 2.6.53"
 	do_facet client "stat $DIR > /dev/null"  ||
 		error "failed to stat $DIR: $?"
 	drop_bl_callback_once "chmod 0777 $DIR" ||
@@ -215,7 +213,7 @@ test_10c() {
 		  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)
+	conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
 	mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
 
 	drop_bl_callback_once "chmod 0777 ${workdir}" &
@@ -247,19 +245,20 @@ test_10d() {
 	local before=$(date +%s)
 	local evict
 
-	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
-		skip "Need MDS version at least 2.6.90" && return
+	[[ "$MDS1_VERSION" -lt $(version_code 2.6.90) ]] &&
+		skip "Need MDS version at least 2.6.90"
 
 	# 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 > $DIR1/$tfile
+	echo -n hello | dd of=$DIR1/$tfile bs=5
 
 	stat $DIR2/$tfile >& /dev/null
 	$LCTL set_param fail_err=71
@@ -267,8 +266,9 @@ test_10d() {
 
 	client_reconnect
 
-	cmp $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
-	cmp $DIR1/$tfile $TMP/$tfile || error "wrong content found"
+	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 '\-\[\] ' | \
@@ -284,6 +284,45 @@ test_10d() {
 }
 run_test 10d "test failed blocking ast"
 
+test_10e()
+{
+	[[ "$OST1_VERSION" -le $(version_code 2.8.58) ]] &&
+		skip "Need OST version at least 2.8.59"
+	[ $CLIENTCOUNT -lt 2 ] && skip "need two clients"
+	[ $(facet_host client) == $(facet_host ost1) ] &&
+		skip "need ost1 and client on different nodes"
+	local -a clients=(${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
 test_11(){
@@ -308,7 +347,7 @@ run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
 
 #b=2494
 test_12(){
-	$LCTL mark "$HOSTNAME: $MULTIOP $DIR/$tfile OS_c"
+	$LCTL mark $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 ||
@@ -388,7 +427,7 @@ test_16() {
 }
 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
@@ -422,7 +461,93 @@ 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
@@ -519,10 +644,10 @@ 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 $TMP/$tfile
@@ -910,7 +1035,8 @@ 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"
@@ -952,20 +1078,21 @@ 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"
 
@@ -1163,21 +1290,22 @@ run_test 52 "failover OST under load"
 # test of open reconstruct
 test_53() {
 	touch $DIR/$tfile
-	drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
+	drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
 		return 2
 }
 run_test 53 "touch: drop rep"
 
 test_54() {
-	zconf_mount `hostname` $MOUNT2
-        touch $DIR/$tfile
-        touch $DIR2/$tfile.1
-        sleep 10
-        cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
-        fail $SINGLEMDS
-        umount $MOUNT2
-        ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
-        [ x"$ERROR" == x ] || error "back in time occured"
+	zconf_mount $(hostname) $MOUNT2
+	touch $DIR/$tfile
+	touch $DIR2/$tfile.1
+	sleep 10
+	cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
+	fail $SINGLEMDS
+	umount $MOUNT2
+	ERROR=$(dmesg | egrep "(test 54|went back in time)" | tail -n1 |
+		grep "went back in time")
+	[ x"$ERROR" == x ] || error "back in time occured"
 }
 run_test 54 "back in time"
 
@@ -1187,10 +1315,17 @@ test_55() {
 
 	mkdir -p $DIR/$tdir
 
+	# This test assumes relatively small max_dirty_mb setting
+	# which we want to walk away from, so just for it we will
+	# temporarily lower the value
+	local max_dirty_mb=$(lctl get_param -n osc.*.max_dirty_mb | head -n 1)
+	lctl set_param -n osc.*.max_dirty_mb=32
+	stack_trap "lctl set_param osc.*.max_dirty_mb=$max_dirty_mb" EXIT
+
 	# Minimum pass speed is 2MBps
 	local ddtimeout=64
 	# LU-2887/LU-3089 - set min pass speed to 500KBps
-	[ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
+	[ "$ost1_FSTYPE" = zfs ] && ddtimeout=256
 
 	# first dd should be finished quickly
 	$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
@@ -1311,63 +1446,47 @@ 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 bz17935 $*
-	else
-		error $*
-	fi
-}
-
 test_60() {
-	MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
-		awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
-
-	NUM_FILES=15000
-	mkdir -p $DIR/$tdir
+	local num_files=${COUNT:-5000}
+	test_mkdir $DIR/$tdir
 
 	# Register (and start) changelog
-	USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
-	echo "Registered as $MDT0 changelog user $USER"
+	changelog_register || error "changelog_register failed"
 
 	# Generate a large number of changelog entries
-	createmany -o $DIR/$tdir/$tfile $NUM_FILES
+	createmany -o $DIR/$tdir/$tfile $num_files
 	sync
 	sleep 5
 
 	# Unlink files in the background
-	unlinkmany $DIR/$tdir/$tfile $NUM_FILES	&
-	CLIENT_PID=$!
+	unlinkmany $DIR/$tdir/$tfile $num_files &
+	local client_pid=$!
 	sleep 1
 
 	# Failover the MDS while unlinks are happening
 	facet_failover $SINGLEMDS
 
 	# Wait for unlinkmany to finish
-	wait $CLIENT_PID
+	wait $client_pid
 
 	# Check if all the create/unlink events were recorded
 	# in the changelog
-	$LFS changelog $MDT0 >> $DIR/$tdir/changelog
-	local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
-	echo "$cl_count unlinks in $MDT0 changelog"
-
-	do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
-	USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
-	    mdd.$MDT0.changelog_users | wc -l) - 2 ))
-	if [ $USERS -eq 0 ]; then
-	    [ $cl_count -eq $NUM_FILES ] || \
-		err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
-	    # Also make sure we can clear large changelogs
-	    cl_count=$($LFS changelog $FSNAME | wc -l)
-	    [ $cl_count -le 2 ] || \
-		error "Changelog not empty: $cl_count entries"
-	else
-	    # If there are other users, there may be other unlinks in the log
-	    [ $cl_count -ge $NUM_FILES ] || \
-		err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
-	    echo "$USERS other changelog users; can't verify clear"
+	local cl_count=$(changelog_dump | grep -c UNLNK)
+	echo "$cl_count unlinks in changelog"
+
+	changelog_deregister || error "changelog_deregister failed"
+	if ! changelog_users $SINGLEMDS | grep -q "^cl"; then
+		[ $cl_count -eq $num_files ] ||
+			error "Recorded $cl_count of $num_files unlinks"
+		# Also make sure we can clear large changelogs
+		cl_count=$(changelog_dump | wc -l)
+		[ $cl_count -le 2 ] ||
+			error "Changelog not empty: $cl_count entries"
+	else # If other users, there may be other unlinks in the log
+		[ $cl_count -ge $num_files ] ||
+			error "Recorded $cl_count of $num_files unlinks"
+		changelog_users $SINGLEMDS
+		echo "other changelog users; can't verify clear"
 	fi
 }
 run_test 60 "Add Changelog entries during MDS failover"
@@ -1411,10 +1530,43 @@ run_test 61 "Verify to not reuse orphan objects - bug 17025"
 #}
 #run_test 62 "Verify connection flags race - bug LU-1716"
 
+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"
+
 test_66()
 {
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
-		{ skip "Need MDS version at least 2.7.51"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.51) ]] ||
+		skip "Need MDS version at least 2.7.51"
 
 	local list=$(comma_list $(osts_nodes))
 
@@ -1423,7 +1575,7 @@ test_66()
 
 	# drop 1 reply with UPDATE lock
 	mcreate $DIR/$tfile || error "mcreate failed: $?"
-	drop_ldlm_reply_once "stat $DIR/$tfile" &
+	drop_mdt_ldlm_reply_once "stat $DIR/$tfile" &
 	sleep 2
 
 	# make the re-sent lock to sleep
@@ -1431,8 +1583,9 @@ test_66()
 	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)
+	local mdtname="MDT0000"
+	local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
+	local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
 	$LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
 	sleep 2
 
@@ -1444,38 +1597,24 @@ test_66()
 }
 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
+test_67()
+{
+#define OBD_FAIL_PTLRPC_CONNECT_RACE	 0x531
+	$LCTL set_param fail_loc=0x80000531
 
+	local mdtname="MDT0000"
+	local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{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
-	do_facet ost $LCTL set_param fail_loc=0
 
-	wait $PID1
-	wait $PID2
+	mds_evict_client
+	sleep 1
 
-	umount_client $DIR2
+	client_reconnect
+	wait
 }
-run_test 65 "lock enqueue for destroyed export"
+run_test 67 "connect vs import invalidate race"
 
 check_cli_ir_state()
 {
@@ -1495,6 +1634,11 @@ 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 ] ||
@@ -1616,26 +1760,45 @@ test_100()
 }
 run_test 100 "IR: Make sure normal recovery still works w/o IR"
 
-test_101()
+test_101a()
 {
-        do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
-                { skip "MGS without IR support"; return 0; }
+	do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+		skip "MGS without IR support"
 
-        set_ir_status full
+	set_ir_status full
 
-        local OST1_IMP=$(get_osc_import_name client ost1)
+	local ost1_imp=$(get_osc_import_name client ost1)
 
-        # disable pinger recovery
-        lctl set_param -n osc.$OST1_IMP.pinger_recov=0
+	# disable pinger recovery
+	lctl set_param -n osc.$ost1_imp.pinger_recov=0
+	stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
 
-        fail ost1
+	fail ost1
 
-        target_instance_match ost1 || error "instance mismatch"
-        nidtbl_versions_match || error "version must match"
+	target_instance_match ost1 || error "instance mismatch"
+	nidtbl_versions_match || error "version must match"
+}
+run_test 101a "IR: Make sure IR works w/o normal recovery"
 
-        lctl set_param -n osc.$OST1_IMP.pinger_recov=1
+test_101b()
+{
+	do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+		skip "MGS without IR support"
+
+	set_ir_status full
+
+	local ost1_imp=$(get_osc_import_name client ost1)
+
+#define OBD_FAIL_OST_PREPARE_DELAY	 0x247
+	do_facet ost1 "$LCTL set_param fail_loc=0x247"
+	# disable pinger recovery
+	$LCTL set_param -n osc.$ost1_imp.pinger_recov=0
+	stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
+
+#OST may return EAGAIN if it is not configured yet
+	fail ost1
 }
-run_test 101 "IR: Make sure IR works w/o normal recovery"
+run_test 101b "IR: Make sure IR works w/o normal recovery and proceed EAGAIN"
 
 test_102()
 {
@@ -1691,7 +1854,7 @@ 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
@@ -1805,8 +1968,8 @@ cleanup_106() {
 }
 
 test_106() { # LU-1789
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.50) ]] ||
-		{ skip "Need MDS version at least 2.3.50"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.3.50) ]] ||
+		skip "Need MDS version at least 2.3.50"
 
 #define OBD_FAIL_MDC_LIGHTWEIGHT         0x805
 	$LCTL set_param fail_loc=0x805
@@ -1825,17 +1988,16 @@ test_106() { # LU-1789
 	touch $DIR2/$tfile || error "failed to create empty file"
 	replay_barrier $SINGLEMDS
 
-	$LCTL set_param debug=console
+	$LCTL set_param debug=ha
 	$LCTL clear
 	facet_failover $SINGLEMDS
 
-	# lightweight connection must be evicted
+	# lightweight goes through LUSTRE_IMP_RECOVER during failover
 	touch -c $DIR2/$tfile || true
 	$LCTL dk $TMP/lustre-log-$TESTNAME.log
-	evicted=`awk '/This client was evicted by .*MDT0000/ {
-				      print;
-		      }' $TMP/lustre-log-$TESTNAME.log`
-	[ -z "$evicted" ] && error "lightweight client not evicted by mds"
+	recovered=$(awk '/MDT0000-mdc-[0-9a-f]*. lwp recover/ { print }' \
+		    $TMP/lustre-log-$TESTNAME.log)
+	[ -z "$recovered" ] && error "lightweight client was not recovered"
 
 	# and all operations performed by lightweight client should be
 	# synchronous, so the file created before mds restart should be there
@@ -1890,21 +2052,21 @@ 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 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.${FSNAME}*.enable_remote_dir=1 \
+			lctl set_param -n mdt.$FSNAME*.enable_remote_dir=1 \
 				2>/dev/null
 	done
 
 	mkdir -p $DIR/$tdir
-	drop_request "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
-					error "lfs mkdir failed"
-	local diridx=$($GETSTRIPE -M $remote_dir)
-	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+	drop_request "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
+		error "lfs mkdir failed"
+	local diridx=$($LFS getstripe -m $remote_dir)
+	[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 }
@@ -1913,14 +2075,14 @@ run_test 110a "create remote directory: drop client req"
 test_110b () {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	local remote_dir=$DIR/$tdir/remote_dir
-	local MDTIDX=1
+	local mdtidx=1
 
 	mkdir -p $DIR/$tdir
-	drop_reint_reply "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
-					error "lfs mkdir failed"
+	drop_reint_reply "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
+		error "lfs mkdir failed"
 
-	diridx=$($GETSTRIPE -M $remote_dir)
-	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+	diridx=$($LFS getstripe -m $remote_dir)
+	[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 }
@@ -1929,14 +2091,14 @@ run_test 110b "create remote directory: drop Master rep"
 test_110c () {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	local remote_dir=$DIR/$tdir/remote_dir
-	local MDTIDX=1
+	local mdtidx=1
 
 	mkdir -p $DIR/$tdir
-	drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
-						error "lfs mkdir failed"
+	drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
+		error "lfs mkdir failed"
 
-	diridx=$($GETSTRIPE -M $remote_dir)
-	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+	diridx=$($GETSTRIPE -m $remote_dir)
+	[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
 }
@@ -1988,31 +2150,37 @@ test_110f () {
 run_test 110f "remove remote directory: drop slave rep"
 
 test_110g () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local remote_dir=$DIR/$tdir/remote_dir
-	local MDTIDX=1
+	[[ "$MDS1_VERSION" -ge $(version_code 2.11.0) ]] ||
+		skip "Need MDS version at least 2.11.0"
 
-	mkdir -p $remote_dir
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
 
-	createmany -o $remote_dir/f 100
+	mkdir -p $DIR/$tdir
+	touch $DIR/$tdir/$tfile
 
-	#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
+	# OBD_FAIL_MDS_REINT_NET_REP	0x119
+	do_facet mds1 $LCTL set_param fail_loc=0x119
+	$LFS migrate -m 1 $DIR/$tdir &
+	migrate_pid=$!
+	sleep 5
+	do_facet mds1 $LCTL set_param fail_loc=0
+	wait $migrate_pid
 
-	for file in $(find $remote_dir); do
-		mdt_index=$($LFS getstripe -M $file)
-		[ $mdt_index == $MDTIDX ] ||
-			error "$file is not on MDT${MDTIDX}"
-	done
+	local mdt_index
+	mdt_index=$($LFS getstripe -m $DIR/$tdir)
+	[ $mdt_index == 1 ] || error "$tdir is not on MDT1"
+	mdt_index=$($LFS getstripe -m $DIR/$tdir/$tfile)
+	[ $mdt_index == 1 ] || error "$tfile is not on MDT1"
 
 	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
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	local src_dir=$DIR/$tdir/source_dir
 	local tgt_dir=$DIR/$tdir/target_dir
 	local MDTIDX=1
@@ -2036,7 +2204,10 @@ test_110h () {
 run_test 110h "drop update reply during cross-MDT file rename"
 
 test_110i () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	local src_dir=$DIR/$tdir/source_dir
 	local tgt_dir=$DIR/$tdir/target_dir
 	local MDTIDX=1
@@ -2063,7 +2234,10 @@ test_110i () {
 run_test 110i "drop update reply during cross-MDT dir rename"
 
 test_110j () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+		skip "Need MDS version at least 2.7.56"
+
 	local remote_dir=$DIR/$tdir/remote_dir
 	local local_dir=$DIR/$tdir/local_dir
 	local MDTIDX=1
@@ -2082,19 +2256,42 @@ test_110j () {
 }
 run_test 110j "drop update reply during cross-MDT ln"
 
+test_110k() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTS"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.12.55) ]] ||
+		skip "Need MDS version at least 2.12.55"
+
+	stop mds2 || error "stop mds2 failed"
+	umount $MOUNT
+
+#define OBD_FAIL_FLD_QUERY_REQ 0x1103
+	do_facet mds2 lctl set_param fail_loc=0x1103
+	start mds2 $(mdsdevname 2) -o abort_recovery ||
+		error "start MDS with abort_recovery should succeed"
+	do_facet mds2 lctl set_param fail_loc=0
+
+	# cleanup
+	stop mds2 || error "cleanup: stop mds2 failed"
+	start mds2 $(mdsdevname 2) || error "cleanup: start mds2 failed"
+	zconf_mount $(hostname) $MOUNT || error "cleanup: mount failed"
+	client_up || error "post-failover df failed"
+}
+run_test 110k "FID_QUERY failed during recovery"
+
 # LU-2844 mdt prepare fail should not cause umount oops
 test_111 ()
 {
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
-		{ skip "Need MDS version at least 2.3.62"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.3.62) ]] ||
+		skip "Need MDS version at least 2.3.62"
 
-	local mdsdev=$(mdsdevname ${SINGLEMDS//mds/})
 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
 	do_facet $SINGLEMDS lctl set_param fail_loc=0x151
 	stop $SINGLEMDS || error "stop MDS failed"
-	start $SINGLEMDS $mdsdev && error "start MDS should fail"
+	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) &&
+		error "start MDS should fail"
 	do_facet $SINGLEMDS lctl set_param fail_loc=0
-	start $SINGLEMDS $mdsdev || error "start MDS failed"
+	start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) ||
+		error "start MDS failed"
 }
 run_test 111 "mdd setup fail should not cause umount oops"
 
@@ -2118,6 +2315,123 @@ test_112a() {
 }
 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 ] && [ $rc -eq 0 ] && error "dd success"
+	return 0
+}
+
+test_115a() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
+	#define OBD_FAIL_PTLRPC_DROP_BULK	 0x51a
+	test_115_read 0x8000051b 0x8000051a
+}
+run_test 115a "read: late REQ MDunlink and no bulk"
+
+test_115b() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#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 $OSTCOUNT
+}
+run_test 115b "write: late REQ MDunlink and no bulk"
+
+test_115c() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
+	#define OBD_FAIL_PTLRPC_DROP_BULK	 0x51a
+	test_115_read 0x8000050f 0x8000051a
+}
+run_test 115c "read: late Reply MDunlink and no bulk"
+
+test_115d() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
+	#define OBD_FAIL_OST_ENOSPC              0x215
+	test_115_write 0x8000050f 0x80000215 0
+}
+run_test 115d "write: late Reply MDunlink and no bulk"
+
+test_115e() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
+	#define OBD_FAIL_OST_ALL_REPLY_NET       0x211
+	test_115_read 0x80000510 0x80000211
+}
+run_test 115e "read: late Bulk MDunlink and no reply"
+
+test_115f() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK  0x51b
+	#define OBD_FAIL_OST_ALL_REPLY_NET       0x211
+	test_115_read 0x8000051b 0x80000211
+}
+run_test 115f "read: late REQ MDunlink and no reply"
+
+test_115g() {
+	[ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+		skip "need at least 2.8.50 on OST"
+
+	#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
@@ -2232,7 +2546,7 @@ test_113() {
 	# 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: $?"
+	drop_mdt_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
@@ -2284,7 +2598,10 @@ test_130_base() {
 }
 
 test_130a() {
-	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+		skip "Need server version newer than 2.7.1"
+
 	test_130_base
 
 	wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
@@ -2293,7 +2610,10 @@ test_130a() {
 run_test 130a "enqueue resend on not existing file"
 
 test_130b() {
-	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+		skip "Need server version newer than 2.7.1"
+
 	test_130_base
 	# let the reply to be dropped
 	sleep 10
@@ -2379,6 +2699,183 @@ test_132() {
 }
 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_MDS_LDLM_REPLY_NET 0x157
+	do_nodes $list $LCTL set_param fail_loc=0x80000157
+	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"
+
+test_134() {
+	[ -z "$CLIENTS" ] && skip "Need two or more clients" && return
+	[ $CLIENTCOUNT -lt 2 ] &&
+		{ skip "Need 2+ clients, have $CLIENTCOUNT" && return; }
+
+	mkdir -p $MOUNT/$tdir/1 $MOUNT/$tdir/2 || error "mkdir failed"
+	touch $MOUNT/$tdir/1/$tfile $MOUNT/$tdir/2/$tfile ||
+		error "touch failed"
+	zconf_umount_clients $CLIENTS $MOUNT
+	zconf_mount_clients $CLIENTS $MOUNT
+
+#define OBD_FAIL_TGT_REPLY_DATA_RACE    0x722
+	# assume commit interval is 5
+	do_facet mds1 "$LCTL set_param fail_loc=0x722 fail_val=5"
+
+	local -a clients=(${CLIENTS//,/ })
+	local client1=${clients[0]}
+	local client2=${clients[1]}
+
+	do_node $client1 rm $MOUNT/$tdir/1/$tfile &
+	rmpid=$!
+	do_node $client2 mv $MOUNT/$tdir/2/$tfile $MOUNT/$tdir/2/${tfile}_2 &
+	mvpid=$!
+	fail mds1
+	wait $rmpid || error "rm failed"
+	wait $mvpid || error "mv failed"
+	return 0
+}
+run_test 134 "race between failover and search for reply data free slot"
+
+test_135() {
+	[ "$MDS1_VERSION" -lt $(version_code 2.12.51) ] &&
+		skip "Need MDS version at least 2.12.51"
+
+	mkdir -p $DIR/$tdir
+	$LFS setstripe -E 1M -L mdt $DIR/$tdir
+	# to have parent dir write lock before open/resend
+	touch $DIR/$tdir/$tfile
+	#define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
+	do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x80000157
+	openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tdir/$tfile ||
+		error "Failed to open DOM file"
+}
+run_test 135 "DOM: open/create resend to return size"
+
+test_136() {
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.12.52) ]] ||
+		skip "Need MDS version at least 2.12.52"
+
+	local mdts=$(comma_list $(mdts_nodes))
+	local MDT0=$(facet_svc $SINGLEMDS)
+
+	local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n)
+	[ -n "$clog" ] || error "changelog_register failed"
+	cl_mask=$(do_facet mds1 $LCTL get_param \
+				mdd.$MDT0.changelog_mask -n)
+	changelog_chmask "ALL"
+
+	# generate some changelog records to accumulate
+	test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed"
+	createmany -m $DIR/$tdir/$tfile 10000 ||
+		error "create $DIR/$tdir/$tfile failed"
+
+	local size1=$(do_facet $SINGLEMDS \
+		      $LCTL get_param -n mdd.$MDT0.changelog_size)
+	echo "Changelog size $size1"
+
+	#define OBD_FAIL_LLOG_PURGE_DELAY		    0x1318
+	do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30
+
+	# launch changelog_deregister in background on MDS
+	do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \
+			> foo.out 2> foo.err < /dev/null &"
+	# give time to reach fail_loc
+	sleep 15
+
+	# fail_loc will make MDS sleep in the middle of changelog_deregister
+	# take this opportunity to abruptly kill MDS
+	FAILURE_MODE_save=$FAILURE_MODE
+	FAILURE_MODE=HARD
+	fail mds1
+	FAILURE_MODE=$FAILURE_MODE_save
+
+	do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0
+
+	local size2=$(do_facet $SINGLEMDS \
+		      $LCTL get_param -n mdd.$MDT0.changelog_size)
+	echo "Changelog size $size2"
+	local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+			mdd.$MDT0.changelog_users | grep $clog")
+	echo "After crash, changelog user $clog2"
+
+	[ -n "$clog2" -o $size2 -lt $size1 ] ||
+		error "changelog record count unchanged"
+
+	do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n
+}
+run_test 136 "changelog_deregister leaving pending records"
+
+test_137() {
+	df $DIR
+	mkdir -p $DIR/d1
+	mkdir -p $DIR/d2
+	dd if=/dev/zero of=$DIR/d1/$tfile bs=4096 count=1
+	dd if=/dev/zero of=$DIR/d2/$tfile bs=4096 count=1
+	cancel_lru_locks osc
+
+	#define OBD_FAIL_PTLRPC_RESEND_RACE	 0x525
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000525"
+
+	# RPC1: any reply is to be delayed to disable last_xid logic
+	ln $DIR/d1/$tfile $DIR/d1/f2 &
+	sleep 1
+
+	# RPC2: setattr1 reply is delayed & resent
+	# original reply comes to client; the resend get asleep
+	chmod 666 $DIR/d2/$tfile
+
+	# RPC3: setattr2 on the same file; run ahead of RPC2 resend
+	chmod 777 $DIR/d2/$tfile
+
+	# RPC2 resend wakes up
+	sleep 5
+	[ $(stat -c "%a" $DIR/d2/$tfile) == 777 ] || error "resend got applied"
+}
+run_test 137 "late resend must be skipped if already applied"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status