X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=af5c2097fb9556eabb74d5fcddcd4bb61f1a75dd;hp=ed8ce3aad61d8ca4a26579d779aa3c9846d455a1;hb=8f01f8b51d114b0d2d54a5ab7db3161782e52447;hpb=757a18894de5948b8d8aab053c0b883c64d27782

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index ed8ce3a..af5c209 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -19,6 +19,10 @@ require_dsh_mds || exit 0
 #                                   1  2.5  2.5    4    4          (min)"
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="17  26a  26b    50   51     57"
 
+[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
+# bug number for skipped test:	      LU-2547
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 24a 24b"
+
 build_test_filter
 
 # Allow us to override the setup if we already have a mounted system by
@@ -145,21 +149,92 @@ test_9() {
 run_test 9 "pause bulk on OST (bug 1420)"
 
 #bug 1521
-test_10() {
-	do_facet client mcreate $DIR/$tfile ||
-		{ error "mcreate failed: $?"; return 1; }
-	drop_bl_callback "chmod 0777 $DIR/$tfile"  || echo "evicted as expected"
-	# wait for the mds to evict the client
-	#echo "sleep $(($TIMEOUT*2))"
-	#sleep $(($TIMEOUT*2))
-	do_facet client touch $DIR/$tfile || echo "touch failed, evicted"
-	do_facet client checkstat -v -p 0777 $DIR/$tfile ||
-		{ error "client checkstat failed: $?"; return 3; }
-	do_facet client "munlink $DIR/$tfile"
-	# allow recovery to complete
-	client_up || client_up || sleep $TIMEOUT
+test_10a() {
+	local before=$(date +%s)
+	local evict
+
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_bl_callback "chmod 0777 $DIR" ||
+		error "failed to chmod $DIR: $?"
+
+	# let the client reconnect
+	client_reconnect
+	evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+	[ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "no eviction: $evict before:$before")
+
+	do_facet client checkstat -v -p 0777 $DIR ||
+		error "client checkstat failed: $?"
+}
+run_test 10a "finish request on server after client eviction (bug 1521)"
+
+test_10b() {
+	local before=$(date +%s)
+	local evict
+
+	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
+		skip "Need MDS version at least 2.6.53" && return
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_bl_callback_once "chmod 0777 $DIR" ||
+		error "failed to chmod $DIR: $?"
+
+	# let the client reconnect
+	client_reconnect
+	evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+
+	[ -z "$evict" ] || [[ $evict -le $before ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "eviction happened: $evict before:$before")
+
+	do_facet client checkstat -v -p 0777 $DIR ||
+ 		error "client checkstat failed: $?"
+}
+run_test 10b "re-send BL AST"
+
+test_10d() {
+	local before=$(date +%s)
+	local evict
+
+	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
+		skip "Need MDS version at least 2.6.90" && return
+
+	# sleep 1 is to make sure that BEFORE is not equal to EVICTED below
+	sleep 1
+	rm -f $TMP/$tfile
+	echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
+
+	mount_client $MOUNT2
+
+	cancel_lru_locks osc
+	$LFS setstripe -i 0 -c 1 $DIR1/$tfile
+	echo -n hello > $DIR1/$tfile
+
+	stat $DIR2/$tfile >& /dev/null
+	$LCTL set_param fail_err=71
+	drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
+
+	client_reconnect
+
+	cmp $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
+	cmp $DIR1/$tfile $TMP/$tfile || error "wrong content found"
+
+	evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
+		tr -d '\-\[\] ' | \
+	  awk -F"[ [,]" '/EVICTED$/ { if (mx<$1) {mx=$1;} } END { print mx }')
+
+	[[ $evict -gt $before ]] ||
+		(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
+		    error "no eviction: $evict before:$before")
+
+	rm $TMP/$tfile
+	umount_client $MOUNT2
 }
-run_test 10 "finish request on server after client eviction (bug 1521)"
+run_test 10d "test failed blocking ast"
 
 #bug 2460
 # wake up a thread waiting for completion after eviction
@@ -173,7 +248,8 @@ test_11(){
 
 	do_facet client $MULTIOP $DIR/$tfile or  ||
 		{ error "multiop read failed: $?"; return 3; }
-	drop_bl_callback $MULTIOP $DIR/$tfile Ow || echo "evicted as expected"
+	drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
+		echo "evicted as expected"
 
 	do_facet client munlink $DIR/$tfile ||
 		{ error "munlink failed: $?"; return 4; }
@@ -260,6 +336,7 @@ test_16() {
 	sleep $TIMEOUT
 	do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
 	start_read_ahead
+	rm -f $TMP/$tfile
 }
 run_test 16 "timeout bulk put, don't evict client (2732)"
 
@@ -328,7 +405,7 @@ test_18a() {
     rc=0
     pgcache_empty || rc=2
     $LCTL --device $osc2dev activate
-    rm -f $f
+    rm -f $f $TMP/$tfile
     return $rc
 }
 run_test 18a "manual ost invalidate clears page cache immediately"
@@ -361,7 +438,7 @@ test_18b() {
     # cache after the client reconnects?     
     rc=0
     pgcache_empty || rc=2
-    rm -f $f
+    rm -f $f $TMP/$tfile
     return $rc
 }
 run_test 18b "eviction and reconnect clears page cache (2766)"
@@ -400,7 +477,7 @@ test_18c() {
     # cache after the client reconnects?     
     rc=0
     pgcache_empty || rc=2
-    rm -f $f
+    rm -f $f $TMP/$tfile
     return $rc
 }
 run_test 18c "Dropped connect reply after eviction handing (14755)"
@@ -411,6 +488,9 @@ test_19a() {
 
 	mount_client $DIR2 || error "failed to mount $DIR2"
 
+	# cancel cached locks from OST to avoid eviction from it
+	cancel_lru_locks osc
+
 	do_facet client "stat $DIR > /dev/null"  ||
 		error "failed to stat $DIR: $?"
 	drop_ldlm_cancel "chmod 0777 $DIR2" ||
@@ -420,8 +500,9 @@ test_19a() {
 
 	# let the client reconnect
 	client_reconnect
-	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state | \
-	    awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+		awk -F"[ [,]" '/EVICTED ]$/ \
+			{ if (mx<$5) {mx=$5;} } END { print mx }')
 
 	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
 		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
@@ -435,6 +516,9 @@ test_19b() {
 
 	mount_client $DIR2 || error "failed to mount $DIR2: $?"
 
+	# cancel cached locks from MDT to avoid eviction from it
+	cancel_lru_locks mdc
+
 	do_facet client $MULTIOP $DIR/$tfile Ow ||
 		error "failed to run multiop: $?"
 	drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
@@ -446,8 +530,9 @@ test_19b() {
 
 	# let the client reconnect
 	client_reconnect
-	EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state | \
-	    awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
+		awk -F"[ [,]" '/EVICTED ]$/ \
+			{ if (mx < $5) {mx = $5;} } END { print mx }')
 
 	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
 		(do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
@@ -479,7 +564,7 @@ test_19c() {
 	# let the client reconnect
 	sleep 5
 	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
-	   awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
 
 	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
 }
@@ -757,7 +842,8 @@ test_24a() { # bug 11710 details correct fsync() behavior
 	rc=$?
 	lctl set_param fail_loc=0x0
 	client_reconnect
-	[ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
+	[ $rc -eq 0 ] &&
+		error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true
 }
 run_test 24a "fsync error (should return error)"
 
@@ -797,10 +883,10 @@ test_24b() {
 	lctl set_param fail_loc=0x0
 	client_reconnect
 	[ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
-	error_ignore 5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
+	error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
 		true
 
-	dmesg | grep "dirty page discard:" || \
+	dmesg | grep "dirty page discard:" ||
 		error "no discarded dirty page found!"
 }
 run_test 24b "test dirty page discard due to client eviction"
@@ -897,7 +983,8 @@ run_test 27 "fail LOV while using OSC's"
 
 test_28() {      # bug 6086 - error adding new clients
 	do_facet client mcreate $DIR/$tfile       || return 1
-	drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected"
+	drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
+		echo "evicted as expected"
 	#define OBD_FAIL_MDS_CLIENT_ADD 0x12f
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
 	# fail once (evicted), reconnect fail (fail_loc), ok
@@ -913,7 +1000,7 @@ test_29a() { # bug 22273 - error adding new clients
 	# fail abort so client will be new again
 	fail_abort $SINGLEMDS
 	client_up || error "reconnect failed"
-	wait_osc_import_state mds ost FULL
+	wait_osc_import_state $SINGLEMDS ost FULL
 	return 0
 }
 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
@@ -949,7 +1036,8 @@ test_50() {
 	rc=$?
 	echo writemany returned $rc
 	#these may fail because of eviction due to slow AST response.
-	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
+	[ $rc -eq 0 ] ||
+		error_ignore bz13652 "writemany returned rc $rc" || true
 }
 run_test 50 "failover MDS under load"
 
@@ -978,10 +1066,11 @@ test_51() {
 	# and recovery was interrupted
 	sleep $TIMEOUT
         kill -USR1 $CLIENT_PID
-	wait $CLIENT_PID 
+	wait $CLIENT_PID
 	rc=$?
 	echo writemany returned $rc
-	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
+	[ $rc -eq 0 ] ||
+		error_ignore bz13652 "writemany returned rc $rc" || true
 }
 run_test 51 "failover MDS during recovery"
 
@@ -1050,57 +1139,58 @@ test_55() {
 
 	mkdir -p $DIR/$tdir
 
+	# Minimum pass speed is 2MBps
+	local ddtimeout=64
+	# LU-2887/LU-3089 - set min pass speed to 500KBps
+	[ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
+
 	# first dd should be finished quickly
 	$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
-	dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4  &
+	dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
 	DDPID=$!
 	count=0
 	echo  "step1: testing ......"
-	while [ true ]; do
-	    if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
-	    count=$[count+1]
-	    if [ $count -gt 64 ]; then
-		error "dd should be finished!"
-	    fi
-	    sleep 1
-	done	
+	while kill -0 $DDPID 2> /dev/null; do
+		let count++
+		if [ $count -gt $ddtimeout ]; then
+			error "dd should be finished!"
+		fi
+		sleep 1
+	done
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
 	$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
 	#define OBD_FAIL_OST_DROP_REQ            0x21d
 	do_facet ost1 lctl set_param fail_loc=0x0000021d
 	# second dd will be never finished
-	dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4  &	
+	dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
 	DDPID=$!
 	count=0
 	echo  "step2: testing ......"
-	while [ $count -le 64 ]; do
-	    dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"	    
-	    if [ -z  $dd_name ]; then 
-                ls -l $DIR/$tdir
-		echo  "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
-		error "dd shouldn't be finished!"
-	    fi
-	    count=$[count+1]
-	    sleep 1
-	done	
+	while [ $count -le $ddtimeout ]; do
+		if ! kill -0 $DDPID 2> /dev/null; then
+			ls -l $DIR/$tdir
+			error "dd shouldn't be finished! (time=$count)"
+		fi
+		let count++
+		sleep 1
+	done
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
 	#Recover fail_loc and dd will finish soon
 	do_facet ost1 lctl set_param fail_loc=0
 	count=0
 	echo  "step3: testing ......"
-	while [ true ]; do
-	    if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
-	    count=$[count+1]
-	    if [ $count -gt 500 ]; then
-		error "dd should be finished!"
-	    fi
-	    sleep 1
-	done	
+	while kill -0 $DDPID 2> /dev/null; do
+		let count++
+		if [ $count -gt $((ddtimeout + 440)) ]; then
+			error "dd should be finished!"
+		fi
+		sleep 1
+	done
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
-        rm -rf $DIR/$tdir
+	rm -rf $DIR/$tdir
 }
 run_test 55 "ost_brw_read/write drops timed-out read/write request"
 
@@ -1108,11 +1198,11 @@ test_56() { # b=11277
 #define OBD_FAIL_MDS_RESEND      0x136
         touch $DIR/$tfile
         do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
-        stat $DIR/$tfile
+        stat $DIR/$tfile || error "stat failed"
         do_facet $SINGLEMDS "lctl set_param fail_loc=0"
         rm -f $DIR/$tfile
 }
-run_test 56 "do not allow reconnect to busy exports"
+run_test 56 "do not fail on getattr resend"
 
 test_57_helper() {
         # no oscs means no client or mdt 
@@ -1146,10 +1236,10 @@ test_58() { # bug 11546
         pid=$!
         sleep 1
         lctl set_param fail_loc=0
-        drop_bl_callback rm -f $DIR/$tfile
+        drop_bl_callback_once rm -f $DIR/$tfile
         wait $pid
         # the first 'df' could tigger the eviction caused by
-        # 'drop_bl_callback', and it's normal case.
+        # 'drop_bl_callback_once', and it's normal case.
         # but the next 'df' should return successfully.
         do_facet client "df $DIR" || do_facet client "df $DIR"
 }
@@ -1174,17 +1264,17 @@ test_59() { # bug 10589
 run_test 59 "Read cancel race on client eviction"
 
 err17935 () {
-    # we assume that all md changes are in the MDT0 changelog
-    if [ $MDSCOUNT -gt 1 ]; then
-	error_ignore 17935 $*
-    else
-	error $*
-    fi
+	# we assume that all md changes are in the MDT0 changelog
+	if [ $MDSCOUNT -gt 1 ]; then
+		error_ignore bz17935 $*
+	else
+		error $*
+	fi
 }
 
 test_60() {
-        MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
-	    awk '{gsub(/_UUID/,""); print $1}' | head -1)
+	MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
+		awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
 
 	NUM_FILES=15000
 	mkdir -p $DIR/$tdir
@@ -1248,12 +1338,14 @@ test_61()
 
 	replay_barrier $SINGLEMDS
 	createmany -o $DIR/$tdir/$tfile-%d 10 
-	local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
+	local oid=$(do_facet ost1 "lctl get_param -n \
+		obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
 
 	fail_abort $SINGLEMDS
-	
+
 	touch $DIR/$tdir/$tfile
-	local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
+	local id=$($LFS getstripe $DIR/$tdir/$tfile |
+		awk '$1 == 0 { print $2 }')
 	[ $id -le $oid ] && error "the orphan objid was reused, failed"
 
 	# Cleanup
@@ -1271,6 +1363,72 @@ run_test 61 "Verify to not reuse orphan objects - bug 17025"
 #}
 #run_test 62 "Verify connection flags race - bug LU-1716"
 
+test_66()
+{
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
+		{ skip "Need MDS version at least 2.7.51"; return 0; }
+
+	local list=$(comma_list $(osts_nodes))
+
+	# modify dir so that next revalidate would not obtain UPDATE lock
+	touch $DIR
+
+	# drop 1 reply with UPDATE lock
+	mcreate $DIR/$tfile || error "mcreate failed: $?"
+	drop_ldlm_reply_once "stat $DIR/$tfile" &
+	sleep 2
+
+	# make the re-sent lock to sleep
+#define OBD_FAIL_MDS_RESEND              0x136
+	do_nodes $list $LCTL set_param fail_loc=0x80000136
+
+	#initiate the re-connect & re-send
+	local mdccli=$($LCTL dl | awk '/-mdc-/ {print $4;}')
+	local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+	$LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
+	sleep 2
+
+	#initiate the client eviction while enqueue re-send is in progress
+	mds_evict_client
+
+	client_reconnect
+	wait
+}
+run_test 66 "lock enqueue re-send vs client eviction"
+
+test_65() {
+	mount_client $DIR2
+
+	#grant lock1, export2
+	$SETSTRIPE -i -0 $DIR2/$tfile || return 1
+	$MULTIOP $DIR2/$tfile Ow  || return 2
+
+#define OBD_FAIL_LDLM_BL_EVICT            0x31e
+	do_facet ost $LCTL set_param fail_loc=0x31e
+	#get waiting lock2, export1
+	$MULTIOP $DIR/$tfile Ow &
+	PID1=$!
+	# let enqueue to get asleep
+	sleep 2
+
+	#get lock2 blocked
+	$MULTIOP $DIR2/$tfile Ow &
+	PID2=$!
+	sleep 2
+
+	#evict export1
+	ost_evict_client
+
+	sleep 2
+	do_facet ost $LCTL set_param fail_loc=0
+
+	wait $PID1
+	wait $PID2
+
+	umount_client $DIR2
+}
+run_test 65 "lock enqueue for destroyed export"
+
 check_cli_ir_state()
 {
         local NODE=${1:-$HOSTNAME}
@@ -1370,8 +1528,8 @@ target_instance_match()
 
         local target=${srv}_svc
         local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
-        local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
-                  awk '/instance/{ print $2 }' |head -1)
+	local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
+		awk '/instance/{ print $2 }' | head -n1)
 
         return $([ $si -eq $ci ])
 }
@@ -1494,10 +1652,11 @@ test_103()
         stop mds1
 
         # We need this test because mds is like a client in IR context.
-        start mds1 $MDSDEV1 || error "MDS should start w/o mgs"
+        start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
+		error "MDS should start w/o mgs"
 
         # start mgs and remount mds w/ ir
-        start mgs $MGSDEV
+        start mgs $(mgsdevname) $MGS_MOUNT_OPTS
         clients_up
 
         # remount client so that fsdb will be created on the MGS
@@ -1550,10 +1709,9 @@ test_105()
         # get one of the clients from client list
         local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
 
-        local old_MOUNTOPT=$MOUNTOPT
-        MOUNTOPT=${MOUNTOPT},noir
+        local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
         zconf_umount $rcli $MOUNT || error "umount failed"
-        zconf_mount $rcli $MOUNT || error "mount failed"
+        zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
 
         # make sure lustre mount at $rcli disabling IR
         local ir_state=$(check_cli_ir_state $rcli)
@@ -1575,8 +1733,7 @@ test_105()
 	[ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
 		error "IR status on ost1 should be DISABLED"
 
-        # restore it
-        MOUNTOPT=$old_MOUNTOPT
+        # remount with the default MOUNT_OPTS
         zconf_umount $rcli $MOUNT || error "umount failed"
         zconf_mount $rcli $MOUNT || error "mount failed"
 
@@ -1665,13 +1822,38 @@ test_107 () {
 }
 run_test 107 "drop reint reply, then restart MDT"
 
+test_108() {
+	mkdir -p $DIR/$tdir
+	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
+
+	dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
+	local dd_pid=$!
+	sleep 0.1
+
+	ost_evict_client
+
+	wait $dd_pid
+
+	client_up || error "reconnect failed"
+	rm -f $DIR/$tdir/$tfile
+}
+run_test 108 "client eviction don't crash"
+
 test_110a () {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	local remote_dir=$DIR/$tdir/remote_dir
 	local MDTIDX=1
+	local num
+
+	#prepare for 110 test, which need set striped dir on remote MDT.
+	for num in $(seq $MDSCOUNT); do
+		do_facet mds$num \
+			lctl set_param -n mdt.${FSNAME}*.enable_remote_dir=1 \
+				2>/dev/null
+	done
 
 	mkdir -p $DIR/$tdir
-	drop_request "$LFS mkdir -i $MDTIDX $remote_dir" ||
+	drop_request "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
 					error "lfs mkdir failed"
 	local diridx=$($GETSTRIPE -M $remote_dir)
 	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
@@ -1686,7 +1868,7 @@ test_110b () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	drop_reint_reply "$LFS mkdir -i $MDTIDX $remote_dir" ||
+	drop_reint_reply "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
 					error "lfs mkdir failed"
 
 	diridx=$($GETSTRIPE -M $remote_dir)
@@ -1702,7 +1884,7 @@ test_110c () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	drop_update_reply $((MDTIDX + 1)) "$LFS mkdir -i $MDTIDX $remote_dir" ||
+	drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
 						error "lfs mkdir failed"
 
 	diridx=$($GETSTRIPE -M $remote_dir)
@@ -1718,7 +1900,7 @@ test_110d () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+	$LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
 
 	drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
 
@@ -1734,7 +1916,7 @@ test_110e () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	$LFS mkdir -i $MDTIDX $remote_dir  || error "lfs mkdir failed"
+	$LFS mkdir -i $MDTIDX -c2 $remote_dir  || error "lfs mkdir failed"
 	drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
 
 	rm -rf $DIR/$tdir || error "rmdir failed"
@@ -1749,7 +1931,7 @@ test_110f () {
 	local MDTIDX=1
 
 	mkdir -p $DIR/$tdir
-	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+	$LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
 	drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
 					error "rm remote dir failed"
 
@@ -1757,6 +1939,271 @@ test_110f () {
 }
 run_test 110f "remove remote directory: drop slave rep"
 
+test_110g () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $remote_dir
+
+	createmany -o $remote_dir/f 100
+
+	#define OBD_FAIL_MIGRATE_NET_REP	0x1702
+	do_facet mds$MDTIDX lctl set_param fail_loc=0x1702
+	$LFS migrate -m $MDTIDX $remote_dir || error "migrate failed"
+	do_facet mds$MDTIDX lctl set_param fail_loc=0x0
+
+	for file in $(find $remote_dir); do
+		mdt_index=$($LFS getstripe -M $file)
+		[ $mdt_index == $MDTIDX ] ||
+			error "$file is not on MDT${MDTIDX}"
+	done
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110g "drop reply during migration"
+
+test_110h () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local src_dir=$DIR/$tdir/source_dir
+	local tgt_dir=$DIR/$tdir/target_dir
+	local MDTIDX=1
+
+	mkdir -p $src_dir
+	$LFS mkdir -i $MDTIDX $tgt_dir
+
+	dd if=/etc/hosts of=$src_dir/src_file
+	touch $tgt_dir/tgt_file
+	drop_update_reply $MDTIDX \
+		"mrename $src_dir/src_file $tgt_dir/tgt_file" ||
+		error "mrename failed"
+
+	$CHECKSTAT -t file $src_dir/src_file &&
+				error "src_file present after rename"
+
+	diff /etc/hosts $tgt_dir/tgt_file ||
+			error "file changed after rename"
+
+}
+run_test 110h "drop update reply during cross-MDT file rename"
+
+test_110i () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local src_dir=$DIR/$tdir/source_dir
+	local tgt_dir=$DIR/$tdir/target_dir
+	local MDTIDX=1
+
+	mkdir -p $src_dir
+	$LFS mkdir -i $MDTIDX $tgt_dir
+
+	mkdir $src_dir/src_dir
+	touch $src_dir/src_dir/a
+	mkdir $tgt_dir/tgt_dir
+	drop_update_reply $MDTIDX \
+		"mrename $src_dir/src_dir $tgt_dir/tgt_dir" ||
+		error "mrename failed"
+
+	$CHECKSTAT -t dir $src_dir/src_dir &&
+			error "src_dir present after rename"
+
+	$CHECKSTAT -t dir $tgt_dir/tgt_dir ||
+				error "tgt_dir not present after rename"
+
+	$CHECKSTAT -t file $tgt_dir/tgt_dir/a ||
+				error "a not present after rename"
+}
+run_test 110i "drop update reply during cross-MDT dir rename"
+
+test_110j () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local local_dir=$DIR/$tdir/local_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	mkdir $DIR/$tdir/local_dir
+	$LFS mkdir -i $MDTIDX $remote_dir
+
+	touch $local_dir/local_file
+	drop_update_reply $MDTIDX \
+		"ln $local_dir/local_file $remote_dir/remote_file" ||
+		error "ln failed"
+
+	$CHECKSTAT -t file $remote_dir/remote_file ||
+				error "remote not present after ln"
+}
+run_test 110j "drop update reply during cross-MDT ln"
+
+# LU-2844 mdt prepare fail should not cause umount oops
+test_111 ()
+{
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
+		{ skip "Need MDS version at least 2.3.62"; return 0; }
+
+	local mdsdev=$(mdsdevname ${SINGLEMDS//mds/})
+#define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
+	do_facet $SINGLEMDS lctl set_param fail_loc=0x151
+	stop $SINGLEMDS || error "stop MDS failed"
+	start $SINGLEMDS $mdsdev && error "start MDS should fail"
+	do_facet $SINGLEMDS lctl set_param fail_loc=0
+	start $SINGLEMDS $mdsdev || error "start MDS failed"
+}
+run_test 111 "mdd setup fail should not cause umount oops"
+
+# LU-793
+test_112a() {
+	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+	do_facet_random_file client $TMP/$tfile 100K ||
+		error_noexit "Create random file $TMP/$tfile"
+
+	pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT ||
+		error_noexit "Can't pause_bulk copy"
+
+	df $DIR
+	# expect cmp to succeed, client resent bulk
+	cmp $TMP/$tfile $DIR/$tfile ||
+		error_noexit "Wrong data has been written"
+	rm $DIR/$tfile ||
+		error_noexit "Can't remove file"
+	rm $TMP/$tfile
+}
+run_test 112a "bulk resend while orignal request is in progress"
+
+# parameters: fail_loc CMD RC
+test_120_reply() {
+	local PID
+	local PID2
+	local rc=5
+	local fail
+
+	#define OBD_FAIL_LDLM_CP_CB_WAIT2	0x320
+	#define OBD_FAIL_LDLM_CP_CB_WAIT3	0x321
+	#define OBD_FAIL_LDLM_CP_CB_WAIT4	0x322
+	#define OBD_FAIL_LDLM_CP_CB_WAIT5	0x323
+
+	echo
+	echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
+	[ "$1" = "CLEANUP" ] &&
+		fail=0x80000320 && echo ", $1 cp first"
+	[ "$1" = "REPLY" ] &&
+		fail=0x80000321 && echo ", $1 cp first"
+	[ "$1" = "DEADLOCK CLEANUP" ] &&
+		fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
+	[ "$1" = "DEADLOCK REPLY" ] &&
+		fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
+
+	if [ x"$2" = x"get" ]; then
+		#for TEST lock, take a conflict in advance
+		# sleep longer than evictor to not confuse fail_loc: 2+2+4
+		echo "** Taking conflict **"
+		flocks_test 5 set read sleep 10 $DIR/$tfile &
+		PID2=$!
+
+		sleep 2
+	fi
+
+	$LCTL set_param fail_loc=$fail
+
+	flocks_test 5 $2 write $DIR/$tfile &
+	PID=$!
+
+	sleep 2
+	echo "** Evicting and re-connecting client **"
+	mds_evict_client
+
+	client_reconnect
+
+	if [ x"$2" = x"get" ]; then
+		wait $PID2
+	fi
+
+	wait $PID
+	rc=$?
+
+	# check if the return value is allowed
+	[ $rc -eq $3 ] && rc=0
+
+	$LCTL set_param fail_loc=0
+	return $rc
+}
+
+# a lock is taken, unlock vs. cleanup_resource() race for destroying
+# the ORIGINAL lock.
+test_120_destroy()
+{
+	local PID
+
+	flocks_test 5 set write sleep 4 $DIR/$tfile &
+	PID=$!
+	sleep 2
+
+	# let unlock to sleep in CP CB
+	$LCTL set_param fail_loc=$1
+	sleep 4
+
+	# let cleanup to cleep in CP CB
+	mds_evict_client
+
+	client_reconnect
+
+	wait $PID
+	rc=$?
+
+	$LCTL set_param fail_loc=0
+	return $rc
+}
+
+test_120() {
+	flock_is_enabled || { skip "mount w/o flock enabled" && return; }
+	touch $DIR/$tfile
+
+	test_120_reply "CLEANUP" set 5 || error "SET race failed"
+	test_120_reply "CLEANUP" get 5 || error "GET race failed"
+	test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
+
+	test_120_reply "REPLY" set 5 || error "SET race failed"
+	test_120_reply "REPLY" get 5 || error "GET race failed"
+	test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
+
+	# DEADLOCK tests
+	test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
+	test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
+
+	test_120_destroy 0x320 || error "unlock-cleanup race failed"
+}
+run_test 120 "flock race: completion vs. evict"
+
+test_113() {
+	local BEFORE=$(date +%s)
+	local EVICT
+
+	# modify dir so that next revalidate would not obtain UPDATE lock
+	touch $DIR
+
+	# drop 1 reply with UPDATE lock,
+	# resend should not create 2nd lock on server
+	mcreate $DIR/$tfile || error "mcreate failed: $?"
+	drop_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
+
+	# 2 BL AST will be sent to client, both must find the same lock,
+	# race them to not get EINVAL for 2nd BL AST
+	#define OBD_FAIL_LDLM_PAUSE_CANCEL2      0x31f
+	$LCTL set_param fail_loc=0x8000031f
+
+	$LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
+	chmod 0777 $DIR/$tfile || error "chmod failed: $?"
+	$LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
+
+	# let the client reconnect
+	client_reconnect
+	EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+
+	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
+}
+run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status