X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-single.sh;h=3a6299c8c2ffc3369d242b80b7fd3072a0b07794;hp=1d073ec0a0a2c2aa36ed154f60b1709d6bbef1ad;hb=7136981bfaad8b95e15e9d2969b94c933491eb98;hpb=f10a4cfd1146989e3947fc8061f2769002b3f044

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 1d073ec..3a6299c 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -7,6 +7,7 @@ set -e
 # This test needs to be run on the client
 #
 SAVE_PWD=$PWD
+export MULTIOP=${MULTIOP:-multiop}
 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 SETUP=${SETUP:-}
 CLEANUP=${CLEANUP:-}
@@ -20,15 +21,12 @@ GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
 require_dsh_mds || exit 0
 
 # Skip these tests
-# bug number:  17466 18857
-ALWAYS_EXCEPT="61d   33a 33b     $REPLAY_SINGLE_EXCEPT"
-
-if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
-    CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
-    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
+# bug number:  17466 18857      LU1867
+ALWAYS_EXCEPT="61d   33a 33b    89      $REPLAY_SINGLE_EXCEPT"
+
+[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
+# bug number for skipped test:        LU-2342  LU-951
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 20b 70a  73a"
 
 #                                                  63 min  7 min  AT AT AT AT"
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a      44b    65 66 67 68"
@@ -40,10 +38,15 @@ check_and_setup_lustre
 mkdir -p $DIR
 
 assert_DIR
-rm -rf $DIR/[df][0-9]*
+rm -rf $DIR/[df][0-9]* $DIR/f.$TESTSUITE.*
+
+# 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
+fi
 
 test_0a() {	# was test_0
-    sleep 10
     mkdir $DIR/$tfile
     replay_barrier $SINGLEMDS
     fail $SINGLEMDS
@@ -62,74 +65,27 @@ test_0b() {
 }
 run_test 0b "ensure object created after recover exists. (3284)"
 
-seq_set_width()
-{
-    local mds=$1
-    local width=$2
-    lctl set_param -n seq.cli-srv-$mds-mdc-*.width=$width
-}
-
-seq_get_width()
-{
-    local mds=$1
-    lctl get_param -n seq.cli-srv-$mds-mdc-*.width
-}
-
-# This test should pass for single-mds and multi-mds configs.
-# But for different configurations it tests different things.
-#
-# single-mds
-# ----------
-# (1) fld_create replay should happen;
-#
-# (2) fld_create replay should not return -EEXISTS, if it does
-# this means sequence manager recovery code is buggy and allocated
-# same sequence two times after recovery.
-#
-# multi-mds
-# ---------
-# (1) fld_create replay may not happen, because its home MDS is
-# MDS2 which is not involved to revovery;
-#
-# (2) as fld_create does not happen on MDS1, it does not make any
-# problem.
 test_0c() {
-    local label=`mdsdevlabel 1`
-    [ -z "$label" ] && echo "No label for mds1" && return 1
+	replay_barrier $SINGLEMDS
+	mcreate $DIR/$tfile
+	umount $MOUNT
+	facet_failover $SINGLEMDS
+	zconf_mount `hostname` $MOUNT || error "mount fails"
+	client_up || error "post-failover df failed"
+	# file shouldn't exist if replay-barrier works as expected
+	rm $DIR/$tfile && error "File exists and it shouldn't"
+	return 0
+}
+run_test 0c "check replay-barrier"
 
+test_0d() {
     replay_barrier $SINGLEMDS
-    local sw=`seq_get_width $label`
-
-    # make seq manager switch to next sequence each
-    # time as new fid is needed.
-    seq_set_width $label 1
-
-    # make sure that fld has created at least one new
-    # entry on server
-    touch $DIR/$tfile || return 2
-    seq_set_width $label $sw
-
-    # fail $SINGLEMDS and start recovery, replay RPCs, etc.
-    fail $SINGLEMDS
-
-    # wait for recovery finish
-    sleep 10
-    df $MOUNT
-
-    # flush fld cache and dentry cache to make it lookup
-    # created entry instead of revalidating existent one
     umount $MOUNT
-    zconf_mount `hostname` $MOUNT
-
-    # issue lookup which should call fld lookup which
-    # should fail if client did not replay fld create
-    # correctly and server has no fld entry
-    touch $DIR/$tfile || return 3
-    rm $DIR/$tfile || return 4
+    facet_failover $SINGLEMDS
+    zconf_mount `hostname` $MOUNT || error "mount fails"
+    client_up || error "post-failover df failed"
 }
-start_full_debug_logging
-run_test 0c "fld create"
-stop_full_debug_logging
+run_test 0d "expired recovery with no clients"
 
 test_1() {
     replay_barrier $SINGLEMDS
@@ -472,7 +428,7 @@ test_20b() { # bug 10480
         usleep 60                           # give dd a chance to start
     done
 
-    lfs getstripe $DIR/$tfile || return 1
+    $GETSTRIPE $DIR/$tfile || return 1
     rm -f $DIR/$tfile || return 2       # make it an orphan
     mds_evict_client
     client_up || client_up || true    # reconnect
@@ -799,10 +755,11 @@ test_37() {
 
     replay_barrier $SINGLEMDS
     # clear the dmesg buffer so we only see errors from this recovery
-    dmesg -c >/dev/null
+    do_facet $SINGLEMDS dmesg -c >/dev/null
     fail_abort $SINGLEMDS
     kill -USR1 $pid
-    dmesg | grep  "mds_unlink_orphan.*error .* unlinking orphan" && return 1
+    do_facet $SINGLEMDS dmesg | grep "error .* unlinking .* from PENDING" &&
+    	return 1
     wait $pid || return 3
     sync
     return 0
@@ -881,18 +838,21 @@ run_test 40 "cause recovery in ptlrpc, ensure IO continues"
 # the page, guarnateeing that the unlock from the RPC completion would
 # assert on trying to unlock the unlocked page.
 test_41() {
-    [ $OSTCOUNT -lt 2 ] && \
-	skip_env "skipping test 41: we don't have a second OST to test with" && \
-	return
+    [ $OSTCOUNT -lt 2 ] &&
+        skip_env "skipping test 41: we don't have a second OST to test with" &&
+        return
 
     local f=$MOUNT/$tfile
     # make sure the start of the file is ost1
-    lfs setstripe $f -s $((128 * 1024)) -i 0
+    $SETSTRIPE -S $((128 * 1024)) -i 0 $f
     do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
     cancel_lru_locks osc
     # fail ost2 and read from ost1
-    local osc2dev=`do_facet $SINGLEMDS "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
-    [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
+    local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost2_svc)
+    local osc2dev=$(do_facet $SINGLEMDS "lctl get_param -n devices" | \
+        grep $mdtosc | awk '{print $1}')
+    [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices &&
+        return 4
     do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate || return 1
     do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
     do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
@@ -938,76 +898,95 @@ test_43() { # bug 2530
 run_test 43 "mds osc import failure during recovery; don't LBUG"
 
 test_44a() { # was test_44
-    local at_max_saved=0
-
-    mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
-    [ "$mdcdev" ] || return 2
-    [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
-
-    # adaptive timeouts slow this way down
-    if at_is_enabled; then
-        at_max_saved=$(at_max_get mds)
-        at_max_set 40 mds
-    fi
-
-    for i in `seq 1 10`; do
-        echo "$i of 10 ($(date +%s))"
-        do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
-        #define OBD_FAIL_TGT_CONN_RACE     0x701
-        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
-        # lctl below may fail, it is valid case
-        $LCTL --device $mdcdev recover
-        df $MOUNT
-    done
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
-    [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
-    return 0
+	local at_max_saved=0
+
+	local mdcdev=$($LCTL get_param -n devices |
+		awk "/ ${FSNAME}-MDT0000-mdc-/ {print \$1}")
+	[ "$mdcdev" ] || return 2
+	[ $(echo $mdcdev | wc -w) -eq 1 ] ||
+		{ echo mdcdev=$mdcdev; $LCTL dl; return 3; }
+
+        # adaptive timeouts slow this way down
+	if at_is_enabled; then
+		at_max_saved=$(at_max_get mds)
+		at_max_set 40 mds
+	fi
+
+	for i in `seq 1 10`; do
+		echo "$i of 10 ($(date +%s))"
+		do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
+#define OBD_FAIL_TGT_CONN_RACE     0x701
+		do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
+                # lctl below may fail, it is valid case
+		$LCTL --device $mdcdev recover
+		df $MOUNT
+	done
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+	[ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
+	return 0
 }
 run_test 44a "race in target handle connect"
 
 test_44b() {
-    local mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
-    [ "$mdcdev" ] || return 2
-    [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
-
-    for i in `seq 1 10`; do
-        echo "$i of 10 ($(date +%s))"
-        do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
+	local mdcdev=$($LCTL get_param -n devices |
+		awk "/ ${FSNAME}-MDT0000-mdc-/ {print \$1}")
+	[ "$mdcdev" ] || return 2
+	[ $(echo $mdcdev | wc -w) -eq 1 ] ||
+		{ echo mdcdev=$mdcdev; $LCTL dl; return 3; }
+
+	for i in `seq 1 10`; do
+		echo "$i of 10 ($(date +%s))"
+		do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
         #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
-        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
+		do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
         # lctl below may fail, it is valid case
-        $LCTL --device $mdcdev recover
-        df $MOUNT
-    done
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
-    return 0
+		$LCTL --device $mdcdev recover
+		df $MOUNT
+	done
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+	return 0
 }
 run_test 44b "race in target handle connect"
 
+test_44c() {
+	replay_barrier $SINGLEMDS
+	createmany -m $DIR/$tfile-%d 100 || error "failed to create directories"
+#define OBD_FAIL_TGT_RCVG_FLAG 0x712
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000712"
+	fail_abort $SINGLEMDS
+	unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail abort"
+	fail $SINGLEMDS
+	unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail"
+	return 0
+}
+run_test 44c "race in target handle connect"
+
 # Handle failed close
 test_45() {
-    mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
-    [ "$mdcdev" ] || return 2
-    [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
+	local mdcdev=$($LCTL get_param -n devices |
+		awk "/ ${FSNAME}-MDT0000-mdc-/ {print \$1}")
+	[ "$mdcdev" ] || return 2
+	[ $(echo $mdcdev | wc -w) -eq 1 ] ||
+		{ echo mdcdev=$mdcdev; $LCTL dl; return 3; }
 
-    $LCTL --device $mdcdev recover || return 6 
+	$LCTL --device $mdcdev recover || return 6
 
-    multiop_bg_pause $DIR/$tfile O_c || return 1
-    pid=$!
+	multiop_bg_pause $DIR/$tfile O_c || return 1
+	pid=$!
 
-    # This will cause the CLOSE to fail before even
-    # allocating a reply buffer
-    $LCTL --device $mdcdev deactivate || return 4
+	# This will cause the CLOSE to fail before even
+	# allocating a reply buffer
+	$LCTL --device $mdcdev deactivate || return 4
 
-    # try the close
-    kill -USR1 $pid
-    wait $pid || return 1
+	# try the close
+	kill -USR1 $pid
+	wait $pid || return 1
 
-    $LCTL --device $mdcdev activate || return 5
-    sleep 1
+	$LCTL --device $mdcdev activate || return 5
+	sleep 1
 
-    $CHECKSTAT -t file $DIR/$tfile || return 2
-    return 0
+	$CHECKSTAT -t file $DIR/$tfile || return 2
+	return 0
 }
 run_test 45 "Handle failed close"
 
@@ -1065,7 +1044,9 @@ test_48() {
 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
 
 test_50() {
-    local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
+    local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost1_svc) 
+    local oscdev=$(do_facet $SINGLEMDS "lctl get_param -n devices" | \
+        grep $mdtosc | awk '{print $1}')
     [ "$oscdev" ] || return 1
     do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
     do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
@@ -1419,14 +1400,20 @@ test_58a() {
 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
 
 test_58b() {
+    local orig
+    local new
+
+    large_xattr_enabled &&
+        orig="$(generate_string $(max_xattr_size))" || orig="bar"
+
     mount_client $MOUNT2
     mkdir -p $DIR/$tdir
     touch $DIR/$tdir/$tfile
     replay_barrier $SINGLEMDS
-    setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
+    setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile
     fail $SINGLEMDS
-    VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
-    [ x$VAL = x"bar" ] || return 1
+    new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
+    [[ "$new" = "$orig" ]] || return 1
     rm -f $DIR/$tdir/$tfile
     rmdir $DIR/$tdir
     zconf_umount `hostname` $MOUNT2
@@ -1434,20 +1421,33 @@ test_58b() {
 run_test 58b "test replay of setxattr op"
 
 test_58c() { # bug 16570
-        mount_client $MOUNT2
-        mkdir -p $DIR/$tdir
-        touch $DIR/$tdir/$tfile
-        drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
-                return 1
-        VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
-        [ x$VAL = x"bar" ] || return 2
-        drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
-                return 3
-        VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
-        [ x$VAL = x"bar1" ] || return 4
-        rm -f $DIR/$tdir/$tfile
-        rmdir $DIR/$tdir
-        zconf_umount `hostname` $MOUNT2
+    local orig
+    local orig1
+    local new
+
+    if large_xattr_enabled; then
+        local xattr_size=$(max_xattr_size)
+        orig="$(generate_string $((xattr_size / 2)))"
+        orig1="$(generate_string $xattr_size)"
+    else
+        orig="bar"
+        orig1="bar1"
+    fi
+
+    mount_client $MOUNT2
+    mkdir -p $DIR/$tdir
+    touch $DIR/$tdir/$tfile
+    drop_request "setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile" ||
+        return 1
+    new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
+    [[ "$new" = "$orig" ]] || return 2
+    drop_reint_reply "setfattr -n trusted.foo1 -v $orig1 $DIR/$tdir/$tfile" ||
+        return 3
+    new=$(get_xattr_value trusted.foo1 $MOUNT2/$tdir/$tfile)
+    [[ "$new" = "$orig1" ]] || return 4
+    rm -f $DIR/$tdir/$tfile
+    rmdir $DIR/$tdir
+    zconf_umount $HOSTNAME $MOUNT2
 }
 run_test 58c "resend/reconstruct setxattr op"
 
@@ -1530,13 +1530,13 @@ test_61c() {
 }
 run_test 61c "test race mds llog sync vs llog cleanup"
 
-test_61d() { # bug 16002 # bug 17466
-    shutdown_facet $SINGLEMDS
-#define OBD_FAIL_OBD_LLOG_SETUP        0x605
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
-    start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
-    start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
+test_61d() { # bug 16002 # bug 17466 # bug 22137
+#   OBD_FAIL_OBD_LLOG_SETUP        0x605
+    stop mgs
+    do_facet mgs "lctl set_param fail_loc=0x80000605"
+    start mgs $MGSDEV $MGS_MOUNT_OPTS && error "mgs start should have failed"
+    do_facet mgs "lctl set_param fail_loc=0"
+    start mgs $MGSDEV $MGS_MOUNT_OPTS || error "cannot restart mgs"
 }
 run_test 61d "error in llog_setup should cleanup the llog context correctly"
 
@@ -1564,7 +1564,7 @@ at_cleanup () {
     echo "Cleaning up AT ..."
     if [ -n "$ATOLDBASE" ]; then
         local at_history=$($LCTL get_param -n at_history)
-        do_facet mds "lctl set_param at_history=$at_history" || true
+        do_facet $SINGLEMDS "lctl set_param at_history=$at_history" || true
         do_facet ost1 "lctl set_param at_history=$at_history" || true
     fi
 
@@ -1604,9 +1604,9 @@ at_start()
     done
 
     if [ -z "$ATOLDBASE" ]; then
-	ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
+	ATOLDBASE=$(do_facet $SINGLEMDS "lctl get_param -n at_history")
         # speed up the timebase so we can check decreasing AT
-        do_facet mds "lctl set_param at_history=8" || true
+        do_facet $SINGLEMDS "lctl set_param at_history=8" || true
         do_facet ost1 "lctl set_param at_history=8" || true
 
 	# sleep for a while to cool down, should be > 8s and also allow
@@ -1622,16 +1622,16 @@ test_65a() #bug 3055
     at_start || return 0
     $LCTL dk > /dev/null
     debugsave
-    sysctl -w lnet.debug="+other"
+    $LCTL set_param debug="other"
     # Slow down a request to the current service time, this is critical
     # because previous tests may have caused this value to increase.
     REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
                awk '/portal 12/ {print $5}'`
     REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
 
-    do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
+    do_facet $SINGLEMDS lctl set_param fail_val=$((${REQ_DELAY} * 1000))
 #define OBD_FAIL_PTLRPC_PAUSE_REQ        0x50a
-    do_facet mds sysctl -w lustre.fail_loc=0x8000050a
+    do_facet $SINGLEMDS $LCTL set_param fail_loc=0x8000050a
     createmany -o $DIR/$tfile 10 > /dev/null
     unlinkmany $DIR/$tfile 10 > /dev/null
     # check for log message
@@ -1651,11 +1651,11 @@ test_65b() #bug 3055
     at_start || return 0
     # turn on D_ADAPTTO
     debugsave
-    sysctl -w lnet.debug="other trace"
+    $LCTL set_param debug="other trace"
     $LCTL dk > /dev/null
     # Slow down a request to the current service time, this is critical
     # because previous tests may have caused this value to increase.
-    lfs setstripe $DIR/$tfile --index=0 --count=1
+    $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
     multiop $DIR/$tfile Ow1yc
     REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
                awk '/portal 6/ {print $5}'`
@@ -1663,14 +1663,14 @@ test_65b() #bug 3055
 
     do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
 #define OBD_FAIL_OST_BRW_PAUSE_PACK      0x224
-    do_facet ost1 sysctl -w lustre.fail_loc=0x224
+    do_facet ost1 $LCTL set_param fail_loc=0x224
 
     rm -f $DIR/$tfile
-    lfs setstripe $DIR/$tfile --index=0 --count=1
+    $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
     # force some real bulk transfer
     multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
 
-    do_facet ost1 sysctl -w lustre.fail_loc=0
+    do_facet ost1 $LCTL set_param fail_loc=0
     # check for log message
     $LCTL dk | grep "Early reply #" || error "No early reply"
     debugrestore
@@ -1686,18 +1686,18 @@ test_66a() #bug 3055
     at_start || return 0
     lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
     # adjust 5s at a time so no early reply is sent (within deadline)
-    do_facet mds "sysctl -w lustre.fail_val=5000"
+    do_facet $SINGLEMDS "$LCTL set_param fail_val=5000"
 #define OBD_FAIL_PTLRPC_PAUSE_REQ        0x50a
-    do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
+    do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
     createmany -o $DIR/$tfile 20 > /dev/null
     unlinkmany $DIR/$tfile 20 > /dev/null
     lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
-    do_facet mds "sysctl -w lustre.fail_val=10000"
-    do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
+    do_facet $SINGLEMDS "$LCTL set_param fail_val=10000"
+    do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
     createmany -o $DIR/$tfile 20 > /dev/null
     unlinkmany $DIR/$tfile 20 > /dev/null
     lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "$LCTL set_param fail_loc=0"
     sleep 9
     createmany -o $DIR/$tfile 20 > /dev/null
     unlinkmany $DIR/$tfile 20 > /dev/null
@@ -1711,19 +1711,23 @@ run_test 66a "AT: verify MDT service time adjusts with no early replies"
 
 test_66b() #bug 3055
 {
-    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
-
-    at_start || return 0
-    ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
-    sysctl -w lustre.fail_val=$(($ORIG + 5))
-#define OBD_FAIL_PTLRPC_PAUSE_REP      0x50c
-    sysctl -w lustre.fail_loc=0x50c
-    ls $DIR/$tfile > /dev/null 2>&1
-    sysctl -w lustre.fail_loc=0
-    CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
-    WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
-    echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
-    [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
+	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+	at_start || return 0
+	ORIG=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
+			awk '/network/ {print $4}')
+	$LCTL set_param fail_val=$(($ORIG + 5))
+	#define OBD_FAIL_PTLRPC_PAUSE_REP      0x50c
+	$LCTL set_param fail_loc=0x50c
+	ls $DIR/$tfile > /dev/null 2>&1
+	$LCTL set_param fail_loc=0
+	CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
+				awk '/network/ {print $4}')
+	WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
+				awk '/network/ {print $6}')
+	echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
+	[ $WORST -gt $ORIG ] ||
+		error "Worst $WORST should be worse than orig $ORIG"
 }
 run_test 66b "AT: verify net latency adjusts"
 
@@ -1734,12 +1738,12 @@ test_67a() #bug 3055
     at_start || return 0
     CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
     # sleeping threads may drive values above this
-    do_facet ost1 "sysctl -w lustre.fail_val=400"
+    do_facet ost1 "$LCTL set_param fail_val=400"
 #define OBD_FAIL_PTLRPC_PAUSE_REQ    0x50a
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
+    do_facet ost1 "$LCTL set_param fail_loc=0x50a"
     createmany -o $DIR/$tfile 20 > /dev/null
     unlinkmany $DIR/$tfile 20 > /dev/null
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
+    do_facet ost1 "$LCTL set_param fail_loc=0"
     CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
     ATTEMPTS=$(($CONN2 - $CONN1))
     echo "$ATTEMPTS osc reconnect attempts on gradual slow"
@@ -1756,17 +1760,19 @@ test_67b() #bug 3055
     CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
 
     # exhaust precreations on ost1
-    local OST=$(lfs osts | grep ^0": " | awk '{print $2}' | sed -e 's/_UUID$//')
-    local mdtosc=$(get_mdtosc_proc_path $OST)
-    local last_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_last_id)
-    local next_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_next_id)
+    local OST=$(ostname_from_index 0)
+    local mdtosc=$(get_mdtosc_proc_path mds $OST)
+    local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
+        osc.$mdtosc.prealloc_last_id)
+    local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
+        osc.$mdtosc.prealloc_next_id)
 
     mkdir -p $DIR/$tdir/${OST}
-    lfs setstripe $DIR/$tdir/${OST} -o 0 -c 1 || error "setstripe"
+    $SETSTRIPE -i 0 -c 1 $DIR/$tdir/${OST} || error "$SETSTRIPE"
     echo "Creating to objid $last_id on ost $OST..."
 #define OBD_FAIL_OST_PAUSE_CREATE        0x223
-    do_facet ost1 "sysctl -w lustre.fail_val=20000"
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
+    do_facet ost1 "$LCTL set_param fail_val=20000"
+    do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
     createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
 
     client_reconnect
@@ -1776,9 +1782,9 @@ test_67b() #bug 3055
     ATTEMPTS=$(($CONN2 - $CONN1))
     echo "$ATTEMPTS osc reconnect attempts on instant slow"
     # do it again; should not timeout
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
+    do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
     cp /etc/profile $DIR/$tfile || error "cp failed"
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
+    do_facet ost1 "$LCTL set_param fail_loc=0"
     client_reconnect
     do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
     CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
@@ -1805,15 +1811,15 @@ test_68 () #bug 13813
 
     rm -rf $DIR/$tdir
     mkdir -p $DIR/$tdir
-    lfs setstripe $DIR/$tdir --index=0 --count=1
+    $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tdir
 #define OBD_FAIL_LDLM_PAUSE_CANCEL       0x312
-    sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
-    sysctl -w lustre.fail_loc=0x80000312
+    $LCTL set_param fail_val=$(($TIMEOUT - 1))
+    $LCTL set_param fail_loc=0x80000312
     cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
-    sysctl -w lustre.fail_val=$((TIMEOUT * 5 / 4))
-    sysctl -w lustre.fail_loc=0x80000312
+    $LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
+    $LCTL set_param fail_loc=0x80000312
     cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
-    sysctl -w lustre.fail_loc=0
+    $LCTL set_param fail_loc=0
 
     echo $ENQ_MIN >> $ldlm_enqueue_min
     do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
@@ -1834,7 +1840,7 @@ test_70a () {
 		{ skip "Need two or more clients, have $CLIENTCOUNT" && return; }
 
 	echo "mount clients $CLIENTS ..."
-	zconf_mount_clients $CLIENTS $DIR
+	zconf_mount_clients $CLIENTS $MOUNT
 
 	local clients=${CLIENTS//,/ }
 	echo "Write/read files on $DIR ; clients $CLIENTS ... "
@@ -1855,59 +1861,64 @@ test_70a () {
 }
 run_test 70a "check multi client t-f"
 
-check_dbench_load () {
-	local clients=${1//,/ }
-	local client=
+check_for_process () {
+	local clients=$1
+	shift
+	local prog=$@
 
-	for client in $clients; do
-		if ! do_node $client "ps ax | grep -v grep | awk '{ print $6 }' | grep -q rundbench"; then
-			error_noexit "rundbench load on $client failed!"
-			return 1
-		fi
-	done
-	return 0
+	killall_process $clients "$prog" -0
 }
 
-kill_dbench_load () {
+killall_process () {
 	local clients=${1:-$(hostname)}
-	do_nodes $clients "killall dbench"
+	local name=$2
+	local signal=$3
+	local rc=0
+
+	do_nodes $clients "killall $signal $name"
 }
 
 test_70b () {
 	local clients=${CLIENTS:-$HOSTNAME}
 
-	zconf_mount_clients $clients $DIR
-	
+	zconf_mount_clients $clients $MOUNT
+
 	local duration=300
-	[ "$SLOW" = "no" ] && duration=60
+	[ "$SLOW" = "no" ] && duration=120
+	# set duration to 900 because it takes some time to boot node
+	[ "$FAILURE_MODE" = HARD ] && duration=900
+
+	local start_ts=$(date +%s)
 	local cmd="rundbench 1 -t $duration"
-	local PID=""
+	local pid=""
 	do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
-		PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
+		PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
 		DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
-		LCTL=$LCTL $cmd" &
-	PID=$!
-	log "Started rundbench load PID=$PID ..."
-	ELAPSED=0
-	NUM_FAILOVERS=0
-	START_TS=$(date +%s)
-	CURRENT_TS=$START_TS
-	while [ $ELAPSED -lt $duration ]; do
-		if ! check_dbench_load $clients; then
-			kill_dbench_load $clients
+		MOUNT=$MOUNT DIR=$DIR/$tdir/\\\$(hostname) LCTL=$LCTL $cmd" &
+	pid=$!
+	log "Started rundbench load pid=$pid ..."
+
+	# give rundbench a chance to start, bug 24118
+	sleep 12
+	local elapsed=$(($(date +%s) - start_ts))
+	local num_failovers=0
+	while [ $elapsed -lt $duration ]; do
+		if ! check_for_process $clients dbench; then
+			error_noexit "dbench not running on some of $clients!"
+			killall_process $clients dbench
 			break
 		fi
 		sleep 1
 		replay_barrier $SINGLEMDS
 		sleep 1 # give clients a time to do operations
 		# Increment the number of failovers
-		NUM_FAILOVERS=$((NUM_FAILOVERS+1))
-		log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
+		num_failovers=$((num_failovers+1))
+		log "$TESTNAME fail $SINGLEMDS $num_failovers times"
 		fail $SINGLEMDS
-		CURRENT_TS=$(date +%s)
-		ELAPSED=$((CURRENT_TS - START_TS))
+		elapsed=$(($(date +%s) - start_ts))
 	done
-	wait $PID || error "rundbench load on $CLIENTS failed!"
+
+	wait $pid || error "rundbench load on $clients failed!"
 }
 run_test 70b "mds recovery; $CLIENTCOUNT clients"
 # end multi-client tests
@@ -1918,7 +1929,7 @@ test_73a() {
     rm -f $DIR/$tfile
 
     replay_barrier $SINGLEMDS
-#define OBD_FAIL_LDLM_ENQUEUE       0x302
+#define OBD_FAIL_LDLM_ENQUEUE_NET			0x302
     do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
     fail $SINGLEMDS
     kill -USR1 $pid
@@ -1964,8 +1975,8 @@ run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
 test_74() {
     local clients=${CLIENTS:-$HOSTNAME}
 
-    stop ost1
     zconf_umount_clients $clients $MOUNT
+    stop ost1
     facet_failover $SINGLEMDS
     zconf_mount_clients $clients $MOUNT
     mount_facet ost1
@@ -1976,72 +1987,471 @@ test_74() {
 }
 run_test 74 "Ensure applications don't fail waiting for OST recovery"
 
+remote_dir_check_80() {
+	local MDTIDX=1
+	local diridx=$($GETSTRIPE -M $remote_dir)
+	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+
+	createmany -o $remote_dir/f-%d 20 || error "creation failed"
+	local fileidx=$($GETSTRIPE -M $remote_dir/f-1)
+	[ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
+
+	return 0
+}
+
 test_80a() {
-    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $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
 
-    mkdir -p $DIR/$tdir
-    replay_barrier mds2
-    $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
-    rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
-    fail mds2
-    stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
-    return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1500
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
 }
-run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
+run_test 80a "DNE: create remote dir, drop update rep from MDT1, fail MDT1"
 
 test_80b() {
-    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $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
 
-    mkdir -p $DIR/$tdir
-    replay_barrier mds1
-    $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
-    rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
-    fail mds1
-    stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
-    return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1500
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 80b "DNE: create remote dir, drop update rep from MDT1, fail MDT0"
+
+test_80c() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1500
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
 }
-run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
+run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
+
+test_80d() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1500
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 80d "DNE: create remote dir, drop update rep from MDT1, fail 2 MDTs"
+
+test_80e() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 80e "DNE: create remote dir, drop MDT0 rep, fail MDT0"
+
+test_80f() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 80f "DNE: create remote dir, drop MDT0 rep, fail MDT1"
+
+test_80g() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "remote creation failed"
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 80g "DNE: create remote dir, drop MDT0 rep, fail MDT0, then MDT1"
+
+test_80h() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+	$LFS mkdir -i $MDTIDX $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || return 1
+
+	remote_dir_check_80 || error "remote dir check failed"
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 80h "DNE: create remote dir, drop MDT0 rep, fail 2 MDTs"
 
 test_81a() {
-    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	[ $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
 
-    mkdir -p $DIR/$tdir
-    createmany -o $DIR/$tdir/f 3000 || error "createmany failed"
-    sleep 10
-    $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir failed"
-    $CHECKSTAT -t file $DIR/$tdir/f1002 || error "$CHECKSTAT -t file failed"
-    replay_barrier mds1
-    rm $DIR/$tdir/f1002 || error "rm $DIR/$tdir/f1002 failed"
-    fail mds1
-    stat $DIR/$tdir/f1002
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1500
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 81a "DNE: unlink remote dir, drop MDT0 update rep,  fail MDT1"
+
+test_81b() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1500
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
 }
-run_test 81a "CMD: unlink cross-node file (fail mds with name)"
+run_test 81b "DNE: unlink remote dir, drop MDT0 update reply,  fail MDT0"
+
+test_81c() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
 
-test_82a() {
-    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
 
-    local dir=$DIR/d82a
-    replay_barrier mds2
-    mkdir $dir || error "mkdir $dir failed"
-    log "FAILOVER mds2"
-    fail mds2
-    stat $DIR
-    $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1500
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
 }
-run_test 82a "CMD: mkdir cross-node dir (fail mds with inode)"
+run_test 81c "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0,MDT1"
+
+test_81d() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_UPDATE_OBJ_NET    0x1500
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0x1500
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
 
-test_82b() {
-    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	wait $CLIENT_PID || error "rm remote dir failed"
 
-    local dir=$DIR/d82b
-    replay_barrier mds1
-    mkdir $dir || error "mkdir $dir failed"
-    log "FAILOVER mds1"
-    fail mds1
-    stat $DIR
-    $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
 }
-run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
+run_test 81d "DNE: unlink remote dir, drop MDT0 update reply,  fail 2 MDTs"
+
+test_81e() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 81e "DNE: unlink remote dir, drop MDT1 req reply, fail MDT0"
+
+test_81f() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 81f "DNE: unlink remote dir, drop MDT1 req reply, fail MDT1"
+
+test_81g() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	([ $FAILURE_MODE == "HARD" ] &&
+		[ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+		skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+		return 0
+
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX}
+	fail mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 81g "DNE: unlink remote dir, drop req reply, fail M0, then M1"
+
+test_81h() {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local MDTIDX=1
+	local remote_dir=$DIR/$tdir/remote_dir
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	# OBD_FAIL_MDS_REINT_NET_REP       0x119
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+	rmdir $remote_dir &
+	local CLIENT_PID=$!
+	do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+	fail mds${MDTIDX},mds$((MDTIDX + 1))
+
+	wait $CLIENT_PID || error "rm remote dir failed"
+
+	stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
 
 test_83a() {
     mkdir -p $DIR/$tdir
@@ -2074,6 +2484,68 @@ test_84a() {
 }
 run_test 84a "stale open during export disconnect"
 
+test_85a() { #bug 16774
+    lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
+
+    for i in `seq 100`; do
+        echo "tag-$i" > $DIR/$tfile-$i
+        grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
+    done
+
+    lov_id=`lctl dl | grep "clilov"`
+    addr=`echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}'`
+    count=`lctl get_param -n ldlm.namespaces.*MDT0000*$addr.lock_unused_count`
+    echo "before recovery: unused locks count = $count"
+
+    fail $SINGLEMDS
+
+    count2=`lctl get_param -n ldlm.namespaces.*MDT0000*$addr.lock_unused_count`
+    echo "after recovery: unused locks count = $count2"
+
+    if [ $count2 -ge $count ]; then
+        error "unused locks are not canceled"
+    fi
+}
+run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
+
+test_85b() { #bug 16774
+    lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
+
+    do_facet mgs $LCTL pool_new $FSNAME.$TESTNAME || return 1
+    do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 || return 2
+
+    $SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
+
+    for i in `seq 100`; do
+        dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 count=32 >/dev/null 2>&1
+    done
+
+    cancel_lru_locks osc
+
+    for i in `seq 100`; do
+        dd if=$DIR/$tfile-$i of=/dev/null bs=4096 count=32 >/dev/null 2>&1
+    done
+
+    lov_id=`lctl dl | grep "clilov"`
+    addr=`echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}'`
+    count=`lctl get_param -n ldlm.namespaces.*OST0000*$addr.lock_unused_count`
+    echo "before recovery: unused locks count = $count"
+    [ $count != 0 ] || return 3
+
+    fail ost1
+
+    count2=`lctl get_param -n ldlm.namespaces.*OST0000*$addr.lock_unused_count`
+    echo "after recovery: unused locks count = $count2"
+
+    do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 || return 4
+    do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME || return 5
+
+    if [ $count2 -ge $count ]; then
+        error "unused locks are not canceled"
+    fi
+}
+run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
+
 test_86() {
         local clients=${CLIENTS:-$HOSTNAME}
 
@@ -2088,7 +2560,7 @@ test_87() {
     do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
 
     replay_barrier ost1
-    lfs setstripe -i 0 -c 1 $DIR/$tfile
+    $SETSTRIPE -i 0 -c 1 $DIR/$tfile
     dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 || error "Cannot write"
     cksum=`md5sum $DIR/$tfile | awk '{print $1}'`
     cancel_lru_locks osc
@@ -2105,7 +2577,7 @@ test_87b() {
     do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
 
     replay_barrier ost1
-    lfs setstripe -i 0 -c 1 $DIR/$tfile
+    $SETSTRIPE -i 0 -c 1 $DIR/$tfile
     dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 || error "Cannot write"
     sleep 1 # Give it a chance to flush dirty data
     echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
@@ -2120,6 +2592,201 @@ test_87b() {
 }
 run_test 87b "write replay with changed data (checksum resend)"
 
-equals_msg `basename $0`: test complete, cleaning up
+test_88() { #bug 17485
+    mkdir -p $DIR/$tdir
+    mkdir -p $TMP/$tdir
+
+    $SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
+
+    replay_barrier ost1
+    replay_barrier $SINGLEMDS
+
+    # exhaust precreations on ost1
+    local OST=$(ostname_from_index 0)
+    local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $OST)
+    local last_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
+    local next_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
+    echo "before test: last_id = $last_id, next_id = $next_id"
+
+    echo "Creating to objid $last_id on ost $OST..."
+    createmany -o $DIR/$tdir/f-%d $next_id $((last_id - next_id + 2))
+
+    #create some files to use some uncommitted objids
+    last_id=$(($last_id + 1))
+    createmany -o $DIR/$tdir/f-%d $last_id 8
+
+    last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
+    next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
+    echo "before recovery: last_id = $last_id2, next_id = $next_id2" 
+
+    # if test uses shutdown_facet && reboot_facet instead of facet_failover ()
+    # it has to take care about the affected facets, bug20407
+    local affected_mds1=$(affected_facets mds1)
+    local affected_ost1=$(affected_facets ost1)
+
+    shutdown_facet $SINGLEMDS
+    shutdown_facet ost1
+
+    reboot_facet $SINGLEMDS
+    change_active $affected_mds1
+    wait_for_facet $affected_mds1
+    mount_facets $affected_mds1 || error "Restart of mds failed"
+
+    reboot_facet ost1
+    change_active $affected_ost1
+    wait_for_facet $affected_ost1
+    mount_facets $affected_ost1 || error "Restart of ost1 failed"
+
+    clients_up
+
+    last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
+    next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
+    echo "after recovery: last_id = $last_id2, next_id = $next_id2" 
+
+    # create new files, which should use new objids, and ensure the orphan 
+    # cleanup phase for ost1 is completed at the same time
+    for i in `seq 8`; do
+        file_id=$(($last_id + 10 + $i))
+        dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
+    done
+
+    # if the objids were not recreated, then "ls" will failed for -ENOENT
+    ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
+
+    local file_id
+    # write into previously created files
+    for i in `seq 8`; do
+        file_id=$(($last_id + $i))
+        dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
+        cp -f $DIR/$tdir/f-$file_id $TMP/$tdir/
+    done
+
+    # compare the content
+    for i in `seq 8`; do
+        file_id=$(($last_id + $i))
+        cmp $TMP/$tdir/f-$file_id $DIR/$tdir/f-$file_id || error "the content" \
+        "of file is modified!"
+    done
+
+    rm -fr $TMP/$tdir
+}
+run_test 88 "MDS should not assign same objid to different files "
+
+test_89() {
+        cancel_lru_locks osc
+        mkdir -p $DIR/$tdir
+        rm -f $DIR/$tdir/$tfile
+        wait_mds_ost_sync
+	wait_delete_completed
+        BLOCKS1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
+        $SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
+        dd if=/dev/zero bs=1M count=10 of=$DIR/$tdir/$tfile
+        sync
+        stop ost1
+        facet_failover $SINGLEMDS
+        rm $DIR/$tdir/$tfile
+        umount $MOUNT
+        mount_facet ost1
+        zconf_mount $(hostname) $MOUNT
+        client_up || return 1
+        wait_mds_ost_sync
+	wait_delete_completed
+        BLOCKS2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
+	[ $((BLOCKS2 - BLOCKS1)) -le 4  ] || \
+		error $((BLOCKS2 - BLOCKS1)) blocks leaked
+}
+
+run_test 89 "no disk space leak on late ost connection"
+
+cleanup_90 () {
+    local facet=$1
+    trap 0
+    reboot_facet $facet
+    change_active $facet
+    wait_for_facet $facet
+    mount_facet $facet || error "Restart of $facet failed"
+    clients_up
+}
+
+test_90() { # bug 19494
+    local dir=$DIR/$tdir
+    local ostfail=$(get_random_entry $(get_facets OST))
+
+    if [[ $FAILURE_MODE = HARD ]]; then
+        local affected=$(affected_facets $ostfail);
+        if [[ "$affected" != $ostfail ]]; then
+            skip not functional with FAILURE_MODE=$FAILURE_MODE, affected: $affected
+            return 0
+        fi
+    fi
+
+    mkdir -p $dir
+
+    echo "Create the files"
+
+    # file "f${index}" striped over 1 OST
+    # file "all" striped over all OSTs
+
+    $SETSTRIPE -c $OSTCOUNT $dir/all ||
+        error "setstripe failed to create $dir/all"
+
+    for (( i=0; i<$OSTCOUNT; i++ )); do
+        local f=$dir/f$i
+        $SETSTRIPE -i $i -c 1 $f || error "$SETSTRIPE failed to create $f"
+
+        # confirm setstripe actually created the stripe on the requested OST
+        local uuid=$(ostuuid_from_index $i)
+        for file in f$i all; do
+            if [[ $dir/$file != $($LFS find --obd $uuid --name $file $dir) ]]; then
+                $GETSTRIPE $dir/file
+                error wrong stripe: $file, uuid: $uuid
+            fi
+        done
+    done
+
+    # Before failing an OST, get its obd name and index
+    local varsvc=${ostfail}_svc
+    local obd=$(do_facet $ostfail lctl get_param -n obdfilter.${!varsvc}.uuid)
+    local index=${obd:(-6):1}
+
+    echo "Fail $ostfail $obd, display the list of affected files"
+    shutdown_facet $ostfail || return 2
+
+    trap "cleanup_90 $ostfail" EXIT INT
+    echo "General Query: lfs find $dir"
+    local list=$($LFS find $dir)
+    echo "$list"
+    for (( i=0; i<$OSTCOUNT; i++ )); do
+        list_member "$list" $dir/f$i || error_noexit "lfs find $dir: no file f$i"
+    done
+    list_member "$list" $dir/all || error_noexit "lfs find $dir: no file all"
+
+    # focus on the missing OST,
+    # we expect to see only two files affected: "f$(index)" and "all"
+
+    echo "Querying files on shutdown $ostfail: lfs find --obd $obd"
+    list=$($LFS find --obd $obd $dir)
+    echo "$list"
+    for file in all f$index; do
+        list_member "$list" $dir/$file ||
+            error_noexit "lfs find does not report the affected $obd for $file"
+    done
+
+    [[ $(echo $list | wc -w) -eq 2 ]] ||
+        error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
+
+    echo "Check getstripe: $GETSTRIPE -r --obd $obd"
+    list=$($GETSTRIPE -r --obd $obd $dir)
+    echo "$list"
+    for file in all f$index; do
+        echo "$list" | grep $dir/$file ||
+            error_noexit "lfs getsripe does not report the affected $obd for $file"
+    done
+
+    cleanup_90 $ostfail
+}
+run_test 90 "lfs find identifies the missing striped file segments"
+
+complete $SECONDS
 check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
+exit_status