X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Freplay-dual.sh;h=07969a1c5543cbfdda188db04621c8922ff9a918;hb=caf5bdffb4eb6e3fb31724a1cb037cecfeb6ae6c;hp=1575ee1d0c8276020ae47ec2ea8b594f7fae4044;hpb=bd8b65394f0a1ffd0c7f8dd0d875c55ba219d9ec;p=fs%2Flustre-release.git

diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index 1575ee1..07969a1 100755
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -1,34 +1,111 @@
 #!/bin/bash
+# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
+# vim:shiftwidth=4:softtabstop=4:tabstop=4:
 
 set -e
 
-# bug number:  6088 10124 10800
-ALWAYS_EXCEPT="8    15c   17    $REPLAY_DUAL_EXCEPT"
-
-#                                                    
-[ "$SLOW" = "no" ] && EXCEPT="$EXCEPT 1 2 3 4 5 14"
+# bug number:  LU-2012 10124
+ALWAYS_EXCEPT="14b     15c   $REPLAY_DUAL_EXCEPT"
 
 SAVE_PWD=$PWD
 PTLDEBUG=${PTLDEBUG:--1}
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 SETUP=${SETUP:-""}
 CLEANUP=${CLEANUP:-""}
 MOUNT_2=${MOUNT_2:-"yes"}
+export MULTIOP=${MULTIOP:-multiop}
 . $LUSTRE/tests/test-framework.sh
 
 init_test_env $@
-
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
+
+remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
 
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
 
 build_test_filter
 
-cleanup_and_setup_lustre
-rm -rf $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
+check_and_setup_lustre
+MOUNTED=$(mounted_lustre_filesystems)
+if ! $(echo $MOUNTED' ' | grep -w -q $MOUNT2' '); then
+    zconf_mount $HOSTNAME $MOUNT2
+    MOUNTED2=yes
+fi
+
+assert_DIR
 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"
+    checkstat $MOUNT1/$tfile-2 && return 1
+    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
@@ -94,10 +171,8 @@ run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
 test_5() {
     # multiclient version of replay_single.sh/test_8
     mcreate $MOUNT1/a
-    multiop $MOUNT2/a o_tSc &
+    multiop_bg_pause $MOUNT2/a o_tSc || return  1
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $MOUNT1/a
     replay_barrier $SINGLEMDS
     kill -USR1 $pid
@@ -112,12 +187,10 @@ run_test 5 "open, unlink |X| close"
 
 test_6() {
     mcreate $MOUNT1/a
-    multiop $MOUNT2/a o_c &
+    multiop_bg_pause $MOUNT2/a o_c || return 1
     pid1=$!
-    multiop $MOUNT1/a o_c &
+    multiop_bg_pause $MOUNT1/a o_c || return 1
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $MOUNT1/a
     replay_barrier $SINGLEMDS
     kill -USR1 $pid1
@@ -147,9 +220,9 @@ test_9() {
     mcreate $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     # drop first reint reply
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
     fail $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
 
     rm $MOUNT1/$tfile-[1,2] || return 1
 
@@ -163,9 +236,9 @@ test_10() {
     munlink $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     # drop first reint reply
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
     fail $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
 
     checkstat $MOUNT1/$tfile-1 && return 1
     checkstat $MOUNT1/$tfile-2 || return 2
@@ -183,12 +256,12 @@ test_11() {
     mcreate $MOUNT2/$tfile-4
     mcreate $MOUNT1/$tfile-5
     # drop all reint replies for a while
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
+    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 sysctl -w lustre.fail_loc=0
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
 
     rm $MOUNT1/$tfile-[1-5] || return 1
 
@@ -199,15 +272,14 @@ run_test 11 "both clients timeout during replay"
 test_12() {
     replay_barrier $SINGLEMDS
 
-    multiop $DIR/$tfile mo_c &
+    multiop_bg_pause $DIR/$tfile mo_c || return 1
     MULTIPID=$!
-    sleep 5
 
-#define OBD_FAIL_LDLM_ENQUEUE            0x302
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
+#define OBD_FAIL_LDLM_ENQUEUE_NET			0x302
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
     facet_failover $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
-    df $MOUNT || return 1
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
+    clients_up || return 1
 
     ls $DIR/$tfile
     kill -USR1 $MULTIPID || return 3
@@ -220,20 +292,19 @@ test_12() {
 run_test 12 "open resend timeout"
 
 test_13() {
-    multiop $DIR/$tfile mo_c &
+    multiop_bg_pause $DIR/$tfile mo_c || return 1
     MULTIPID=$!
-    sleep 5
 
     replay_barrier $SINGLEMDS
 
     kill -USR1 $MULTIPID || return 3
     wait $MULTIPID || return 4
 
-    # drop close 
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
+    # drop close
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
     facet_failover $SINGLEMDS
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
-    df $MOUNT || return 1
+    do_facet $SINGLEMDS lctl set_param fail_loc=0
+    clients_up || return 1
 
     ls $DIR/$tfile
     $CHECKSTAT -t file $DIR/$tfile || return 2
@@ -243,119 +314,71 @@ test_13() {
 }
 run_test 13 "close resend timeout"
 
-test_14() {
-    replay_barrier $SINGLEMDS
-    createmany -o $MOUNT1/$tfile- 25
-    createmany -o $MOUNT2/$tfile-2- 1
-    createmany -o $MOUNT1/$tfile-3- 25
-    umount $MOUNT2
+# test 14a removed after 18143 because it shouldn't fail anymore and do the same
+# as test_15a
 
-    facet_failover $SINGLEMDS
-    # expect failover to fail due to missing client 2
-    df $MOUNT && return 1
-    sleep 1
+test_14b() {
+	wait_mds_ost_sync
+	wait_delete_completed
 
-    # first 25 files should have been replayed 
-    unlinkmany $MOUNT1/$tfile- 25 || return 2
+	local BEFOREUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
 
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail" 
-    return 0
-}
-run_test 14 "timeouts waiting for lost client during replay"
+	mkdir -p $MOUNT1/$tdir
+	$SETSTRIPE -i 0 $MOUNT1/$tdir
+	replay_barrier $SINGLEMDS
+	createmany -o $MOUNT1/$tdir/$tfile- 5
 
-test_15() {
-    replay_barrier $SINGLEMDS
-    createmany -o $MOUNT1/$tfile- 25
-    createmany -o $MOUNT2/$tfile-2- 1
-    umount $MOUNT2
+	$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
 
-    facet_failover $SINGLEMDS
-    df $MOUNT || return 1
+	fail $SINGLEMDS
+	wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
 
-    unlinkmany $MOUNT1/$tfile- 25 || return 2
-    [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
+	# 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 fail"
-    return 0
-}
-run_test 15 "timeout waiting for lost client during replay, 1 client completes"
+	zconf_mount $HOSTNAME $MOUNT2 || error "mount $MOUNT2 failed"
+	[ -f $MOUNT2/$tfile-2 ] && error "$MOUNT2/$tfile-2 exists!"
 
-test_15a() {
-    local ost_last_id=""
-    local osc_last_id=""
-    
-    replay_barrier $SINGLEMDS
-    echo "data" > "$MOUNT2/${tfile}-m2"
-
-    umount $MOUNT2
-    facet_failover $SINGLEMDS
-    df $MOUNT || return 1
-    
-    ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
-    mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
-    
-    echo "Ids after MDS<->OST synchonizing"
-    echo "--------------------------------"
-    echo "MDS last_id:"
-    echo $mds_last_id
-    echo "OST last_id:"
-    echo $ost_last_id
-
-    local i=0
-    echo $ost_last_id | while read id; do
-	ost_ids[$i]=$id
-	((i++))
-    done
-    
-    i=0
-    echo $mds_last_id | while read id; do
-	mds_ids[$i]=$id
-	((i++))
-    done
-    
-    local arr_len=${#mds_ids[*]}
-    for ((i=0;i<$arr_len;i++)); do
-	    mds_id=${mds_ids[i]}
-	    ost_id=${ost_ids[i]}
-	    
-	    test $mds_id -ge $ost_id || {
-		echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
-		return 2
-	    }
-    done
+	wait_mds_ost_sync || error "wait_mds_ost_sync failed"
+	wait_delete_completed || error "wait_delete_complete failed"
 
-    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
-    return 0
+	local AFTERUSED=$(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
 }
-#CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
+run_test 14b "delete ost orphans if gap occured in objids due to VBR"
 
-test_15b() {
+test_15a() { # was test_15
     replay_barrier $SINGLEMDS
-    echo "data" > "$MOUNT2/${tfile}-m2"
+    createmany -o $MOUNT1/$tfile- 25
+    createmany -o $MOUNT2/$tfile-2- 1
     umount $MOUNT2
 
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
-    facet_failover $SINGLEMDS
+    fail $SINGLEMDS
+
+    unlinkmany $MOUNT1/$tfile- 25 || return 2
+    [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
 
-    df $MOUNT || return 1
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
-    
     zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
     return 0
 }
-#CROW run_test 15b "multiple delayed OST clear orphans"
+run_test 15a "timeout waiting for lost client during replay, 1 client completes"
 
 test_15c() {
     replay_barrier $SINGLEMDS
     for ((i = 0; i < 2000; i++)); do
-	echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
+        echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
     done
-    
     umount $MOUNT2
-    facet_failover $SINGLEMDS
 
-    df $MOUNT || return 1
-    
+    fail $SINGLEMDS
+
     zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
     return 0
 }
@@ -369,8 +392,7 @@ test_16() {
 
     facet_failover $SINGLEMDS
     sleep $TIMEOUT
-    facet_failover $SINGLEMDS
-    df $MOUNT || return 1
+    fail $SINGLEMDS
 
     unlinkmany $MOUNT1/$tfile- 25 || return 2
 
@@ -381,6 +403,8 @@ test_16() {
 run_test 16 "fail MDS during recovery (3571)"
 
 test_17() {
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
 
@@ -390,8 +414,7 @@ test_17() {
 
     facet_failover ost1
     sleep $TIMEOUT
-    facet_failover ost1
-    df $MOUNT || return 1
+    fail ost1
 
     unlinkmany $MOUNT1/$tfile- 25 || return 2
 
@@ -412,16 +435,18 @@ test_18() { # bug 3822 - evicting client with enqueued lock
     statmany -s $MOUNT1/$tdir/f 1 500 &
     OPENPID=$!
     NOW=`date +%s`
-    do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b  # hold enqueue
+    do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b  # hold enqueue
     sleep 1
-#define OBD_FAIL_LDLM_BL_CALLBACK        0x305
-    do_facet client sysctl -w lustre.fail_loc=0x80000305  # drop cb, evict
+#define OBD_FAIL_LDLM_BL_CALLBACK_NET			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"
 }
 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
 
@@ -436,9 +461,406 @@ test_19() { # Bug 10991 - resend of open request does not fail assertion.
 }
 run_test 19 "resend of open request"
 
-equals_msg `basename $0`: test complete, cleaning up
+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 > $TIER1"
+    return 0
+}
+run_test 20 "recovery time is not increasing"
+
+# commit on sharing tests
+test_21a() {
+    local param_file=$TMP/$tfile-params
+
+    save_lustre_params $(facet_active_host $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
+    mv  $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
+    replay_barrier_nosync $SINGLEMDS
+    umount $MOUNT2
+
+    facet_failover $SINGLEMDS
+
+    # all renames are replayed
+    unlink  $MOUNT1/$tfile-3 || return 2
+
+    zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
+
+    do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
+    rm -rf $MOUNT1/$tfile-*
+    restore_lustre_params < $param_file
+    rm -f $param_file
+    return 0
+}
+run_test 21a "commit on sharing"
+
+test_21b_sub () {
+    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
+
+    replay_barrier_nosync $mds
+    shutdown_client $CLIENT2 $MOUNT1
+
+    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; }
+
+    boot_node $CLIENT2
+    zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail" 
+
+    return $rc
+}
+
+test_21b() {
+    [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
+    [ $CLIENTCOUNT -lt 2 ] && \
+        { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
+
+    if [ "$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
+
+
+    zconf_umount_clients $CLIENTS $MOUNT2
+    zconf_mount_clients $CLIENTS $MOUNT1
+
+    local param_file=$TMP/$tfile-params
+
+    local num=$(get_mds_dir $MOUNT1)
+
+    save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
+
+    # COS enabled
+    local COS=1
+    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+
+    test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
+
+    # COS disabled (should fail)
+    COS=0
+    do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+
+    # there is still a window when transactions may be written to disk before
+    # the mds device is set R/O. To avoid such a rare test failure, the check
+    # is repeated several times.
+    local n_attempts=1
+    while true; do
+    test_21b_sub mds$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
+    zconf_mount_clients $CLIENTS $MOUNT2
+    restore_lustre_params < $param_file
+    rm -f $param_file
+    return 0
+}
+run_test 21b "commit on sharing, two clients"
+
+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
+}
+
+test_22a () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=${tdir}/remote_dir
+
+	do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+	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 22a "c1 lfs mkdir -i 1 dir1, M0 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} lctl set_param fail_loc=0x119
+	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 22b "c1 lfs mkdir -i 1 d1, M0 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_MDS_DROP_OBJ_UPDATE       0x188
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x188
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail 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 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_MDS_DROP_OBJ_UPDATE       0x188
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x188
+	do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+	CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) 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_MDS_DROP_OBJ_UPDATE       0x188
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x188
+	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_MDS_DROP_OBJ_UPDATE       0x188
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x188
+	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"
+
+# end commit on sharing tests 
+
+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 || true
-
+exit_status