X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-dual.sh;h=b8e5094684f9ea022af43f8935d32c94c65b3983;hp=e3f7c5d05a94dc7388262de64525cdd3054ca8ac;hb=9804f872d28315a7ddfe835138bb33e2206ffe52;hpb=680235c4af9e995a64630cb313f44c90f4c2c6e3

diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index e3f7c5d..b8e5094 100755
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -2,8 +2,8 @@
 
 set -e
 
-# bug number:  10124
-ALWAYS_EXCEPT="15c   $REPLAY_DUAL_EXCEPT"
+# bug number:  LU-2012 10124
+ALWAYS_EXCEPT="14b     15c   $REPLAY_DUAL_EXCEPT"
 
 SAVE_PWD=$PWD
 PTLDEBUG=${PTLDEBUG:--1}
@@ -11,15 +11,9 @@ LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 SETUP=${SETUP:-""}
 CLEANUP=${CLEANUP:-""}
 MOUNT_2=${MOUNT_2:-"yes"}
+export MULTIOP=${MULTIOP:-multiop}
 . $LUSTRE/tests/test-framework.sh
 
-if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
-    CONFIG_EXCEPTIONS="17"
-    echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
-    echo "Except the tests: $CONFIG_EXCEPTIONS"
-    ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
-fi
-
 init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 init_logging
@@ -28,11 +22,15 @@ remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
 
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
 
+[[ $(facet_fstype $SINGLEMDS) == zfs ]] &&
+# bug number for skipped test:	      LU-2230
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21b"
+
 build_test_filter
 
 check_and_setup_lustre
 MOUNTED=$(mounted_lustre_filesystems)
-if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
+if ! $(echo $MOUNTED' ' | grep -w -q $MOUNT2' '); then
     zconf_mount $HOSTNAME $MOUNT2
     MOUNTED2=yes
 fi
@@ -42,6 +40,75 @@ rm -rf $DIR/[df][0-9]*
 
 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
 
+# if there is no CLIENT1 defined, some tests can be ran on localhost
+CLIENT1=${CLIENT1:-$HOSTNAME}
+# if CLIENT2 doesn't exist then use CLIENT1 instead
+# All tests should use CLIENT2 with MOUNT2 only therefore it will work if
+# $CLIENT2 == CLIENT1
+# Exception is the test which need two separate nodes
+CLIENT2=${CLIENT2:-$CLIENT1}
+
+# LU-482 Avert LVM and VM inability to flush caches in pre .33 kernels
+if [ $LINUX_VERSION_CODE -lt $(version_code 2.6.33) ]; then
+	sync
+	do_facet $SINGLEMDS "sync; sleep 10; sync; sleep 10; sync"
+fi
+
+LU482_FAILED=$(mktemp -u $TMP/$TESTSUITE.lu482.XXXXXX)
+test_0a() {
+	echo "Check file is LU482_FAILED=$LU482_FAILED"
+	touch $MOUNT2/$tfile-A # force sync FLD/SEQ update before barrier
+	replay_barrier $SINGLEMDS
+#define OBD_FAIL_PTLRPC_FINISH_REPLAY | OBD_FAIL_ONCE
+	touch $MOUNT2/$tfile
+	createmany -o $MOUNT1/$tfile- 50
+	$LCTL set_param fail_loc=0x80000514
+	facet_failover $SINGLEMDS
+	[ -f "$LU482_FAILED" ] && skip "LU-482 failure" && return 0
+	client_up || return 1
+	umount -f $MOUNT2
+	client_up || return 1
+	zconf_mount `hostname` $MOUNT2 || error "mount2 fais"
+	unlinkmany $MOUNT1/$tfile- 50 || return 2
+	rm $MOUNT2/$tfile || return 3
+	rm $MOUNT2/$tfile-A || return 4
+}
+run_test 0a "expired recovery with lost client"
+
+if [ -f "$LU482_FAILED" ]; then
+	log "Found check file $LU482_FAILED, aborting test script"
+	rm -vf "$LU482_FAILED"
+	complete $SECONDS
+	do_nodes $CLIENTS umount -f $MOUNT2 || true
+	do_nodes $CLIENTS umount -f $MOUNT || true
+	# copied from stopall, but avoid the MDS recovery
+    for num in `seq $OSTCOUNT`; do
+        stop ost$num -f
+        rm -f $TMP/ost${num}active
+    done
+    if ! combined_mgs_mds ; then
+        stop mgs
+    fi
+
+	exit_status
+fi
+
+test_0b() {
+    replay_barrier $SINGLEMDS
+    touch $MOUNT2/$tfile
+    touch $MOUNT1/$tfile-2
+    umount $MOUNT2
+    facet_failover $SINGLEMDS
+    umount -f $MOUNT1
+    zconf_mount `hostname` $MOUNT1 || error "mount1 fais"
+    zconf_mount `hostname` $MOUNT2 || error "mount2 fais"
+    # it is uncertain if file-2 exists or not, remove it if it does
+    checkstat $MOUNT1/$tfile-2 && rm $MOUNT1/$tfile-2
+    checkstat $MOUNT2/$tfile && return 2
+    return 0
+}
+run_test 0b "lost client during waiting for next transno"
+
 test_1() {
     touch $MOUNT1/a
     replay_barrier $SINGLEMDS
@@ -185,23 +252,25 @@ test_10() {
 run_test 10 "resending a replayed unlink"
 
 test_11() {
-    replay_barrier $SINGLEMDS
-    mcreate $MOUNT1/$tfile-1
-    mcreate $MOUNT2/$tfile-2
-    mcreate $MOUNT1/$tfile-3
-    mcreate $MOUNT2/$tfile-4
-    mcreate $MOUNT1/$tfile-5
-    # drop all reint replies for a while
-    do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
-    # note that with this fail_loc set, facet_failover df will fail
-    facet_failover $SINGLEMDS
-    #sleep for while, let both clients reconnect and timeout
-    sleep $((TIMEOUT * 2))
-    do_facet $SINGLEMDS lctl set_param fail_loc=0
+	replay_barrier $SINGLEMDS
+	mcreate $DIR1/$tfile-1
+	mcreate $DIR2/$tfile-2
+	mcreate $DIR1/$tfile-3
+	mcreate $DIR2/$tfile-4
+	mcreate $DIR1/$tfile-5
+	# drop all reint replies for a while
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0119
+	# note that with this fail_loc set, facet_failover df will fail
+	facet_failover $SINGLEMDS
 
-    rm $MOUNT1/$tfile-[1-5] || return 1
+	local clients=${CLIENTS:-$HOSTNAME}
+	wait_clients_import_state "$clients" $SINGLEMDS FULL
 
-    return 0
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+	rm $DIR1/$tfile-[1-5] || return 1
+
+	return 0
 }
 run_test 11 "both clients timeout during replay"
 
@@ -211,7 +280,7 @@ test_12() {
     multiop_bg_pause $DIR/$tfile mo_c || return 1
     MULTIPID=$!
 
-#define OBD_FAIL_LDLM_ENQUEUE            0x302
+#define OBD_FAIL_LDLM_ENQUEUE_NET			0x302
     do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
     facet_failover $SINGLEMDS
     do_facet $SINGLEMDS lctl set_param fail_loc=0
@@ -254,31 +323,39 @@ run_test 13 "close resend timeout"
 # as test_15a
 
 test_14b() {
-    BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
-    mkdir -p $MOUNT1/$tdir
-    replay_barrier $SINGLEMDS
-    createmany -o $MOUNT1/$tfile- 5
-    echo "data" > $MOUNT2/$tdir/$tfile-2
-    createmany -o $MOUNT1/$tfile-3- 5
-    umount $MOUNT2
+	wait_mds_ost_sync
+	wait_delete_completed
 
-    fail $SINGLEMDS
-    wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
+	local BEFOREUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
 
-    # first 25 files should have been replayed
-    unlinkmany $MOUNT1/$tfile- 5 || return 2
-    unlinkmany $MOUNT1/$tfile-3- 5 || return 3
+	mkdir -p $MOUNT1/$tdir
+	$SETSTRIPE -i 0 $MOUNT1/$tdir
+	replay_barrier $SINGLEMDS
+	createmany -o $MOUNT1/$tdir/$tfile- 5
 
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
+	$SETSTRIPE -i 0 $MOUNT2/$tfile-2
+	dd if=/dev/zero of=$MOUNT2/$tfile-2 bs=1M count=5
+	createmany -o $MOUNT1/$tdir/$tfile-3- 5
+	umount $MOUNT2
 
-    wait_mds_ost_sync || return 4
-    wait_destroy_complete || return 5
+	fail $SINGLEMDS
+	wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
 
-    AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
-    log "before $BEFOREUSED, after $AFTERUSED"
-    [ $AFTERUSED -ne $BEFOREUSED ] && \
-        error "after $AFTERUSED > before $BEFOREUSED" && return 4
-    return 0
+	# first set of files should have been replayed
+	unlinkmany $MOUNT1/$tdir/$tfile- 5 || error "first unlinks failed"
+	unlinkmany $MOUNT1/$tdir/$tfile-3- 5 || error "second unlinks failed"
+
+	zconf_mount $HOSTNAME $MOUNT2 || error "mount $MOUNT2 failed"
+	[ -f $MOUNT2/$tfile-2 ] && error "$MOUNT2/$tfile-2 exists!"
+
+	wait_mds_ost_sync || error "wait_mds_ost_sync failed"
+	wait_delete_completed || error "wait_delete_complete failed"
+
+	local AFTERUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
+	log "before $BEFOREUSED, after $AFTERUSED"
+	# leave some margin for some files/dirs to be modified (OI, llog, etc)
+	[ $AFTERUSED -gt $((BEFOREUSED + 128)) ] &&
+		error "after $AFTERUSED > before $BEFOREUSED" || true
 }
 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
 
@@ -356,25 +433,29 @@ run_test 17 "fail OST during recovery (3571)"
 export NOW=0
 
 test_18() { # bug 3822 - evicting client with enqueued lock
-    #set -vx
-    mkdir -p $MOUNT1/$tdir
-    touch $MOUNT1/$tdir/f0
-#define OBD_FAIL_LDLM_ENQUEUE_BLOCKED    0x30b
-    statmany -s $MOUNT1/$tdir/f 1 500 &
-    OPENPID=$!
-    NOW=`date +%s`
-    do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b  # hold enqueue
-    sleep 1
-#define OBD_FAIL_LDLM_BL_CALLBACK        0x305
-    do_facet client lctl set_param fail_loc=0x80000305  # drop cb, evict
-    cancel_lru_locks mdc
-    usleep 500 # wait to ensure first client is one that will be evicted
-    openfile -f O_RDONLY $MOUNT2/$tdir/f0
-    wait $OPENPID
-    dmesg | grep "entering recovery in server" && \
-        error "client not evicted" || true
-    do_facet client "lctl set_param fail_loc=0"
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+	#set -vx
+	local DLMTRACE=$(do_facet $SINGLEMDS lctl get_param debug)
+	do_facet $SINGLEMDS lctl set_param debug=+dlmtrace
+	mkdir -p $MOUNT1/$tdir || error "mkdir $MOUNT1/$tdir failed"
+	touch $MOUNT1/$tdir/$tfile
+	#define OBD_FAIL_LDLM_ENQUEUE_BLOCKED    0x30b
+	statmany -s $MOUNT1/$tdir/f 1 500 &
+	OPENPID=$!
+	NOW=$(date +%s)
+	do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b  # hold enqueue
+	sleep 1
+	#define OBD_FAIL_LDLM_BL_CALLBACK_NET			0x305
+	do_facet client lctl set_param ldlm.namespaces.*.early_lock_cancel=0
+	do_facet client lctl set_param fail_loc=0x80000305  # drop cb, evict
+	cancel_lru_locks mdc
+	usleep 500 # wait to ensure first client is one that will be evicted
+	openfile -f O_RDONLY $MOUNT2/$tdir/$tfile
+	wait $OPENPID
+	do_facet client lctl set_param ldlm.namespaces.*.early_lock_cancel=1
+	do_facet $SINGLEMDS lctl debug_kernel |
+		grep "not entering recovery" && error "client not evicted"
+	do_facet client "lctl set_param fail_loc=0"
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 }
 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
 
@@ -390,27 +471,29 @@ test_19() { # Bug 10991 - resend of open request does not fail assertion.
 run_test 19 "resend of open request"
 
 test_20() { #16389
-    BEFORE=`date +%s`
-    replay_barrier $SINGLEMDS
-    touch $MOUNT1/a
-    touch $MOUNT2/b
-    umount $MOUNT2
-    fail $SINGLEMDS
-    rm $MOUNT1/a
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    TIER1=$((`date +%s` - BEFORE))
-    BEFORE=`date +%s`
-    replay_barrier $SINGLEMDS
-    touch $MOUNT1/a
-    touch $MOUNT2/b
-    umount $MOUNT2
-    fail $SINGLEMDS
-    rm $MOUNT1/a
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    TIER2=$((`date +%s` - BEFORE))
-    [ $TIER2 -ge $((TIER1 * 2)) ] && \
-        error "recovery time is growing $TIER2 > $TIER1"
-    return 0
+	local before=$SECONDS
+	replay_barrier $SINGLEMDS
+	touch $DIR1/$tfile.a
+	touch $DIR2/$tfile.b
+	umount $DIR2
+	fail $SINGLEMDS
+	rm $DIR1/$tfile.a
+	zconf_mount $HOSTNAME $DIR2 || error "mount $DIR2 fail"
+	local tier1=$((SECONDS - before))
+
+	before=$SECONDS
+	replay_barrier $SINGLEMDS
+	touch $DIR1/$tfile.a
+	touch $DIR2/$tfile.b
+	umount $DIR2
+	fail $SINGLEMDS
+	rm $DIR1/$tfile.a
+	zconf_mount $HOSTNAME $DIR2 || error "mount $DIR2 fail"
+	local tier2=$((SECONDS - before))
+
+	# timeout is more than 2.25x original timeout
+	((tier2 < tier1 * 9 / 4)) ||
+		error "recovery time $tier2 >= 2.25x original time $tier1"
 }
 run_test 20 "recovery time is not increasing"
 
@@ -418,7 +501,7 @@ run_test 20 "recovery time is not increasing"
 test_21a() {
     local param_file=$TMP/$tfile-params
 
-    save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
+	save_lustre_params $SINGLEMDS "mdt.*.commit_on_sharing" > $param_file
     do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
     touch  $MOUNT1/$tfile-1
     mv  $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
@@ -442,82 +525,398 @@ test_21a() {
 run_test 21a "commit on sharing"
 
 test_21b_sub () {
-    local mds=$1 
-    do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
+	local mds=$1
+	do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
 
-    do_facet $mds sync
-    do_node $CLIENT1 touch  $MOUNT1/$tfile-1
-    do_node $CLIENT2 mv  $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
-    do_node $CLIENT1 mv  $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
+	do_facet $mds sync
+	do_node $CLIENT1 touch $MOUNT1/$tfile-1
+	do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
+	do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
 
-    replay_barrier_nosync $mds
-    shutdown_client $CLIENT2 $MOUNT1
+	replay_barrier_nosync $mds
+	shutdown_client $CLIENT2 $MOUNT1
 
-    facet_failover $mds
+	facet_failover $mds
 
-    # were renames replayed?
-    local rc=0
-    echo UNLINK $MOUNT1/$tfile-3 
-    do_node $CLIENT1 unlink  $MOUNT1/$tfile-3 || \
-        { echo "unlink $tfile-3 fail!" && rc=1; }
+	# were renames replayed?
+	local rc=0
+	echo UNLINK $MOUNT1/$tfile-3
+	do_node $CLIENT1 unlink  $MOUNT1/$tfile-3 ||
+		{ echo "unlink $tfile-3 fail!" && rc=1; }
 
-    boot_node $CLIENT2
-    zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail" 
+	boot_node $CLIENT2
+	zconf_mount_clients $CLIENT2 $MOUNT1 ||
+		error "mount $CLIENT2 $MOUNT1 fail"
 
-    return $rc
+	return $rc
 }
 
 test_21b() {
-    [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
-    [ $CLIENTCOUNT -lt 2 ] && \
-        { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
+	[ -z "$CLIENTS" ] && skip "Need two or more clients" && return
+	[ $CLIENTCOUNT -lt 2 ] &&
+		{ skip "Need 2+ clients, have $CLIENTCOUNT" && return; }
+
+	if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
+		skip "Several MDTs on one MDS with FAILURE_MODE=$FAILURE_MODE"
+		return 0
+	fi
+
+	zconf_umount_clients $CLIENTS $MOUNT2
+	zconf_mount_clients $CLIENTS $MOUNT1
+
+	local param_file=$TMP/$tfile-params
+
+	local mdtidx=$($LFS getstripe -M $MOUNT1)
+	local facet=mds$((mdtidx + 1))
+
+	save_lustre_params $facet "mdt.*.commit_on_sharing" > $param_file
+
+	# COS enabled
+	local COS=1
+	do_facet $facet lctl set_param mdt.*.commit_on_sharing=$COS
+
+	test_21b_sub $facet || error "Not all renames are replayed. COS=$COS"
+
+	# COS disabled (should fail)
+	COS=0
+	do_facet $facet lctl set_param mdt.*.commit_on_sharing=$COS
+
+	# there is still a window when transactions may be written to disk
+	# before the mds device is set R/O. To avoid such a rare test failure,
+	# the check is repeated several times.
+	local n_attempts=1
+	while true; do
+		test_21b_sub $facet || break
+		n_attempts=$((n_attempts + 1))
+		[ $n_attempts -gt 3 ] &&
+			error "can't check if COS works: rename replied w/o COS"
+	done
+	zconf_mount_clients $CLIENTS $MOUNT2
+	restore_lustre_params < $param_file
+	rm -f $param_file
+	return 0
+}
+run_test 21b "commit on sharing, two clients"
 
-    if [ "$FAILURE_MODE" = "HARD" ] &&  mixed_mdt_devs; then
-        skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
-        return 0
-    fi
+checkstat_22() {
+	checkstat $MOUNT1/$remote_dir || return 1
+	checkstat $MOUNT1/$remote_dir/dir || return 2
+	checkstat $MOUNT1/$remote_dir/$tfile-1 || return 3
+	checkstat $MOUNT1/$remote_dir/dir/$tfile-1 || return 4
+	return 0
+}
 
+create_remote_dir_files_22() {
+	do_node $CLIENT2 mkdir ${MOUNT2}/$remote_dir/dir || return 1
+	do_node $CLIENT1 createmany -o $MOUNT1/$remote_dir/dir/$tfile- 2 ||
+							    return 2
+	do_node $CLIENT2 createmany -o $MOUNT2/$remote_dir/$tfile- 2 ||
+							    return 3
+	return 0
+}
 
-    zconf_umount_clients $CLIENTS $MOUNT2
-    zconf_mount_clients $CLIENTS $MOUNT1
+test_22a () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
 
-    local param_file=$TMP/$tfile-params
+	local MDTIDX=1
+	local remote_dir=${tdir}/remote_dir
 
-    local num=$(get_mds_dir $MOUNT1)
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
 
-    save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
 
-    # COS enabled
-    local COS=1
-    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+	fail mds$((MDTIDX + 1))
+	wait $CLIENT_PID || error "lfs mkdir failed"
 
-    test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
+	replay_barrier mds$MDTIDX
+	create_remote_dir_files_22 || error "Remote creation failed $?"
+	fail mds$MDTIDX
 
-    # COS disabled (should fail)
-    COS=0
-    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+	checkstat_22 || error "check stat failed $?"
 
-    # there is still a window when transactions may be written to disk before
-    # the mds device is set R/O. To avoid such a rare test failure, the check
-    # is repeated several times.
-    local n_attempts=1
-    while true; do
-    test_21b_sub mds$num || break;
-    let n_attempts=n_attempts+1
-    [ $n_attempts -gt 3 ] &&
-        error "The test cannot check whether COS works or not: all renames are replied w/o COS"
-    done
-    restore_lustre_params < $param_file
-    rm -f $param_file
-    return 0
+	rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+	return 0
 }
-run_test 21b "commit on sharing, two clients"
+run_test 22a "c1 lfs mkdir -i 1 dir1, M1 drop reply & fail, c2 mkdir dir1/dir"
+
+test_22b () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$tdir/remote_dir
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+	wait $CLIENT_PID || error "lfs mkdir failed"
+
+	replay_barrier mds$MDTIDX
+	create_remote_dir_files_22 || error "Remote creation failed $?"
+	fail mds${MDTIDX}
+
+	checkstat_22 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+	return 0
+}
+run_test 22b "c1 lfs mkdir -i 1 d1, M1 drop reply & fail M0/M1, c2 mkdir d1/dir"
+
+test_22c () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local MDTIDX=1
+	local remote_dir=${tdir}/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+	# OBD_FAIL_UPDATE_OBJ_NET_REP    0x1701
+	do_facet mds$MDTIDX lctl set_param fail_loc=0x1701
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+	do_facet mds$MDTIDX lctl set_param fail_loc=0
+
+	fail mds$MDTIDX
+	wait $CLIENT_PID || error "lfs mkdir failed"
+
+	replay_barrier mds$MDTIDX
+	create_remote_dir_files_22 || error "Remote creation failed $?"
+	fail mds$MDTIDX
+
+	checkstat_22 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+	return 0
+}
+run_test 22c "c1 lfs mkdir -i 1 d1, M1 drop update & fail M1, c2 mkdir d1/dir"
+
+test_22d () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=${tdir}/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+	# OBD_FAIL_UPDATE_OBJ_NET_REP    0x1701
+	do_facet mds$MDTIDX lctl set_param fail_loc=0x1701
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+	do_facet mds$MDTIDX lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+	wait $CLIENT_PID || error "lfs mkdir failed"
+
+	replay_barrier mds$MDTIDX
+	create_remote_dir_files_22 || error "Remote creation failed $?"
+	fail mds$MDTIDX
+
+	checkstat_22 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+	return 0
+}
+run_test 22d "c1 lfs mkdir -i 1 d1, M1 drop update & fail M0/M1,c2 mkdir d1/dir"
+
+checkstat_23() {
+	checkstat $MOUNT1/$remote_dir || return 1
+	checkstat $MOUNT1/$remote_dir/$tfile-1 || return 2
+	return 0
+}
+
+create_remote_dir_files_23() {
+	do_node $CLIENT2 mkdir ${MOUNT2}/$remote_dir || return 1
+	do_node $CLIENT2 createmany -o $MOUNT2/$remote_dir/$tfile- 2 || return 2
+	return 0
+}
+
+test_23a () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local MDTIDX=1
+	local remote_dir=$tdir/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+			error "lfs mkdir failed"
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds$((MDTIDX + 1))
+	wait $CLIENT_PID || error "rmdir remote dir failed"
+
+	replay_barrier mds${MDTIDX}
+	create_remote_dir_files_23 || error "Remote creation failed $?"
+	fail mds${MDTIDX}
+
+	checkstat_23 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+	return 0
+}
+run_test 23a "c1 rmdir d1, M1 drop reply and fail, client2 mkdir d1"
+
+test_23b () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$tdir/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+			error "lfs mkdir failed"
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+	wait $CLIENT_PID || error "rmdir remote dir failed"
+
+	replay_barrier mds${MDTIDX}
+	create_remote_dir_files_23 || error "Remote creation failed $?"
+	fail mds${MDTIDX}
+
+	checkstat_23 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+	return 0
+}
+run_test 23b "c1 rmdir d1, M1 drop reply and fail M0/M1, c2 mkdir d1"
+
+test_23c () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local MDTIDX=1
+	local remote_dir=$tdir/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+			error "lfs mkdir failed"
+
+	# OBD_FAIL_UPDATE_OBJ_NET_REP    0x1701
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
+	do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+	wait $CLIENT_PID || error "rmdir remote dir failed"
+
+	replay_barrier mds${MDTIDX}
+	create_remote_dir_files_23 || error "Remote creation failed $?"
+	fail mds${MDTIDX}
+
+	checkstat_23 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || return 6
+	return 0
+}
+run_test 23c "c1 rmdir d1, M0 drop update reply and fail M0, c2 mkdir d1"
+
+test_23d () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$tdir/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+			error "lfs mkdir failed"
+
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1701
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
+	do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+	wait $CLIENT_PID || error "rmdir remote dir failed"
+
+	replay_barrier mds${MDTIDX}
+	create_remote_dir_files_23 || error "Remote creation failed $?"
+	fail mds${MDTIDX}
+
+	checkstat_23 || error "check stat failed $?"
+
+	rm -rf $MOUNT1/$tdir || return 6
+	return 0
+}
+run_test 23d "c1 rmdir d1, M0 drop update reply and fail M0/M1, c2 mkdir d1"
+
+test_24 () {
+	[[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.5.2) ]] ||
+		{ skip "Need MDS version newer than 2.5.2"; return 0; }
+
+	touch $MOUNT/$tfile
+	stat $MOUNT/$tfile >&/dev/null
+# OBD_FAIL_MDS_REINT_NET_REP
+	do_facet $SINGLEMDS $LCTL set_param fail_loc=0x119
+	$TRUNCATE $MOUNT/$tfile 100 &
+	PID=$!
+	sleep 1
+	do_facet $SINGLEMDS lctl set_param fail_loc=0
+	# sync to release rep-ack lock quickly
+	do_nodes $(comma_list $(mdts_nodes)) \
+	    "lctl set_param -n osd*.*MDT*.force_sync 1"
+	rm $MOUNT2/$tfile
+	wait
+}
+run_test 24 "reconstruct on non-existing object"
 
 # end commit on sharing tests 
 
-equals_msg `basename $0`: test complete, cleaning up
+test_25() {
+	cancel_lru_locks osc
+
+	$SETSTRIPE -i 0 -c 1 $DIR/$tfile
+
+	# get lock for the 1st client
+	dd if=/dev/zero of=$DIR/$tfile count=1 >/dev/null ||
+		error "failed to write data"
+
+	# get waiting locks for the 2nd client
+	drop_ldlm_cancel "multiop $DIR2/$tfile Ow512" &
+	sleep 1
+
+#define OBD_FAIL_OST_LDLM_REPLY_NET      0x213
+	# failover, replay and resend replayed waiting locks
+	do_facet ost1 lctl set_param fail_loc=0x80000213
+	fail ost1
+
+	# multiop does not finish because CP AST is skipped;
+	# it is ok to kill it in the test, because CP AST is already re-sent
+	# and it does not hung forever in real life
+	killall multiop
+	wait
+}
+run_test 25 "replay|resend"
+
+complete $SECONDS
 SLEEP=$((`date +%s` - $NOW))
 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
 check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
+exit_status