X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=48de1b9bae14f7e2218abd115369578839414d4b;hp=4b27f7a6a6fdf37ac22185871df6759adde62a8b;hb=630bb7a01c2cdc661ad09644b19b798f48bb66b3;hpb=6c617a3d56e4b84dcc4e76ff0eebcb177b227ac3

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 4b27f7a..48de1b9 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -2,25 +2,29 @@
 
 set -e
 
-#         bug  5493  LU2034
-ALWAYS_EXCEPT="52 $RECOVERY_SMALL_EXCEPT"
-
-export MULTIOP=${MULTIOP:-multiop}
 PTLDEBUG=${PTLDEBUG:--1}
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(dirname $0)/..}
 . $LUSTRE/tests/test-framework.sh
 init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 init_logging
 
-require_dsh_mds || exit 0
-
-# also long tests: 19, 21a, 21e, 21f, 23, 27
+ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT "
+if $SHARED_KEY; then
+	# bug number for skipped test: LU-12896
+	ALWAYS_EXCEPT+="               110k"
+	# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+fi
 
-[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
+selinux_status=$(getenforce)
+if [ "$selinux_status" != "Disabled" ]; then
+	# bug number for skipped test:   LU-12928
+	$SHARED_KEY && ALWAYS_EXCEPT+=" 136"
+fi
 
 build_test_filter
 
+require_dsh_mds || exit 0
+
 # Allow us to override the setup if we already have a mounted system by
 # setting SETUP=" " and CLEANUP=" "
 SETUP=${SETUP:-""}
@@ -171,8 +175,8 @@ test_10b() {
 	local before=$(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" ||
@@ -211,7 +215,7 @@ 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 ${workdir}" &
@@ -243,8 +247,8 @@ test_10d() {
 	local before=$(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
@@ -284,11 +288,11 @@ 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
-	[ $CLIENTCOUNT -lt 2 ] && skip "need two clients" && return 0
+	[[ "$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) == $(facet_host ost1) ] &&
-		skip "need ost1 and client on different nodes" && return 0
+		skip "need ost1 and client on different nodes"
 	local -a clients=(${CLIENTS//,/ })
 	local client1=${clients[0]}
 	local client2=${clients[1]}
@@ -1288,21 +1292,22 @@ 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"
 
@@ -1312,10 +1317,17 @@ 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
@@ -1520,10 +1532,43 @@ 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
+	$LFS setstripe -i -0 $DIR2/$tfile || error "setstripe failed"
+	$MULTIOP $DIR2/$tfile Ow  || error "multiop failed"
+
+#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))
 
@@ -1532,7 +1577,7 @@ 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
@@ -1540,8 +1585,9 @@ 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
 
@@ -1553,38 +1599,24 @@ 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()
 {
@@ -1730,26 +1762,45 @@ 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"
+
+test_101b()
+{
+	do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+		skip "MGS without IR support"
 
-        lctl set_param -n osc.$OST1_IMP.pinger_recov=1
+	set_ir_status full
+
+	local ost1_imp=$(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()
 {
@@ -1883,7 +1934,7 @@ test_105()
 	# Since the client just mounted, its last_rcvd entry is not on disk.
 	# Send an RPC so exp_need_sync forces last_rcvd to commit this export
 	# so the client can reconnect during OST recovery (LU-924, LU-1582)
-	$SETSTRIPE -i 0 $DIR/$tfile
+	$LFS setstripe -i 0 $DIR/$tfile
 	dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
 
         # make sure MGS's state is Partial
@@ -1919,8 +1970,8 @@ 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
@@ -1939,17 +1990,16 @@ 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
@@ -1986,7 +2036,7 @@ run_test 107 "drop reint reply, then restart MDT"
 
 test_108() {
 	mkdir -p $DIR/$tdir
-	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
+	$LFS setstripe -c 1 -i 0 $DIR/$tdir
 
 	dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
 	local dd_pid=$!
@@ -2049,7 +2099,7 @@ test_110c () {
 	drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
 		error "lfs mkdir failed"
 
-	diridx=$($GETSTRIPE -m $remote_dir)
+	diridx=$($LFS getstripe -m $remote_dir)
 	[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
@@ -2102,37 +2152,36 @@ test_110f () {
 run_test 110f "remove remote directory: drop slave rep"
 
 test_110g () {
-	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.6.57) ]] ||
-		{ skip "Need MDS version at least 2.6.57"; return 0; }
+	[[ "$MDS1_VERSION" -ge $(version_code 2.11.0) ]] ||
+		skip "Need MDS version at least 2.11.0"
 
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local remote_dir=$DIR/$tdir/remote_dir
-	local mdtidx=1
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
 
-	mkdir -p $remote_dir
-
-	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 () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local server_version=$(lustre_version_code mds1)
-	[[ $server_version -ge $(version_code 2.7.56) ]] ||
-		{ skip "Need MDS version at least 2.7.56"; 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
@@ -2157,10 +2206,9 @@ test_110h () {
 run_test 110h "drop update reply during cross-MDT file rename"
 
 test_110i () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local server_version=$(lustre_version_code mds1)
-	[[ $server_version -ge $(version_code 2.7.56) ]] ||
-		{ skip "Need MDS version at least 2.7.56"; 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
@@ -2188,10 +2236,9 @@ test_110i () {
 run_test 110i "drop update reply during cross-MDT dir rename"
 
 test_110j () {
-	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
-	local server_version=$(lustre_version_code mds1)
-	[[ $server_version -ge $(version_code 2.7.56) ]] ||
-		{ skip "Need MDS version at least 2.7.56"; 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
@@ -2211,11 +2258,33 @@ 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"
 
 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
 	do_facet $SINGLEMDS lctl set_param fail_loc=0x151
@@ -2294,8 +2363,8 @@ test_115_write() {
 }
 
 test_115a() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2304,8 +2373,8 @@ test_115a() {
 run_test 115a "read: late REQ MDunlink and no bulk"
 
 test_115b() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2317,8 +2386,8 @@ test_115b() {
 run_test 115b "write: late REQ MDunlink and no bulk"
 
 test_115c() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2327,8 +2396,8 @@ test_115c() {
 run_test 115c "read: late Reply MDunlink and no bulk"
 
 test_115d() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2337,8 +2406,8 @@ test_115d() {
 run_test 115d "write: late Reply MDunlink and no bulk"
 
 test_115e() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2347,8 +2416,8 @@ test_115e() {
 run_test 115e "read: late Bulk MDunlink and no reply"
 
 test_115f() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2357,8 +2426,8 @@ test_115f() {
 run_test 115f "read: late REQ MDunlink and no reply"
 
 test_115g() {
-	[ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
-		skip "need at least 2.8.50 on OST" && return 0
+	[ "$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
@@ -2479,7 +2548,7 @@ 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
@@ -2531,10 +2600,9 @@ test_130_base() {
 }
 
 test_130a() {
-	remote_mds_nodsh && skip "remote MDS with nodsh" && return
-	local server_version=$(lustre_version_code $SINGLEMDS)
-	[[ $server_version -ge $(version_code 2.7.2) ]] ||
-		{ skip "Need server version newer than 2.7.1"; return 0; }
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+		skip "Need server version newer than 2.7.1"
 
 	test_130_base
 
@@ -2544,10 +2612,9 @@ test_130a() {
 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 $(version_code 2.7.2) ]] ||
-		{ skip "Need server version newer than 2.7.1"; return 0; }
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ "$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
@@ -2607,7 +2674,7 @@ test_132() {
 
 	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"
+	$LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
 	dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
 		error "dd failed"
 
@@ -2639,7 +2706,7 @@ test_131() {
 
 	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"
+	$LFS 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
@@ -2671,8 +2738,8 @@ test_133() {
 	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
+	#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
@@ -2683,6 +2750,134 @@ test_133() {
 }
 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