X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=3f29d3fcb03c8c91da0e8197d19062322c62a51c;hb=0fe560ea5dd4e582f2393f54de106e34a8b2c6ae;hp=4a2b57a86a8c0d5be02b46b30384e87a4c0f954f;hpb=0c6b5334c2355e37c1f2db871ca1c91d7ec5536d;p=fs%2Flustre-release.git

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 4a2b57a..3f29d3f 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -11,36 +11,24 @@ LUSTRE=${LUSTRE:-`dirname $0`/..}
 init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 
+# also long tests: 19, 21a, 21e, 21f, 23, 27
+#                                   1  2.5  2.5    4    4          (min)"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="17  26a  26b    50   51     57"
+
 build_test_filter
 
 # Allow us to override the setup if we already have a mounted system by
 # setting SETUP=" " and CLEANUP=" "
-SETUP=${SETUP:-"setup"}
-CLEANUP=${CLEANUP:-"cleanup"}
-
-setup() {
-    [ "$REFORMAT" ] && formatall
-    setupall
-}
-
-cleanup() {
-	cleanupall || { echo "FAILed to clean up"; exit 20; }
-}
+SETUP=${SETUP:-""}
+CLEANUP=${CLEANUP:-""}
 
-if [ ! -z "$EVAL" ]; then
-    eval "$EVAL"
-    exit $?
-fi
+cleanup_and_setup_lustre
+rm -rf $DIR/[df][0-9]*
 
-if [ "$ONLY" == "cleanup" ]; then
-    sysctl -w lnet.debug=0 || true
-    cleanup
-    exit
-fi
-
-$SETUP
-
-[ "$ONLY" == "setup" ] && exit
+SAMPLE_NAME=recovery-small.junk
+SAMPLE_FILE=$TMP/$SAMPLE_NAME
+# make this big, else test 9 doesn't wait for bulk -- bz 5595
+dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4
 
 test_1() {
     drop_request "mcreate $MOUNT/1"  || return 1
@@ -98,7 +86,7 @@ run_test 8 "touch: drop rep (bug 1423)"
 #bug 1420
 test_9() {
     pause_bulk "cp /etc/profile $MOUNT/$tfile"       || return 1
-    do_facet client "cp /etc/termcap $MOUNT/${tfile}.2"  || return 2
+    do_facet client "cp ${SAMPLE_FILE} $MOUNT/${tfile}.2"  || return 2
     do_facet client "sync"
     do_facet client "rm $MOUNT/$tfile $MOUNT/${tfile}.2" || return 3
 }
@@ -135,12 +123,11 @@ run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
 #b=2494
 test_12(){
     $LCTL mark multiop $MOUNT/$tfile OS_c 
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x115"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
     clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
-    multiop $MOUNT/$tfile OS_c  &
+    multiop_bg_pause $MOUNT/$tfile OS_c || return 1
     PID=$!
 #define OBD_FAIL_MDS_CLOSE_NET           0x115
-    sleep 2
     kill -USR1 $PID
     echo "waiting for multiop $PID"
     wait $PID || return 2
@@ -153,9 +140,9 @@ test_13() {
     mkdir $MOUNT/readdir || return 1
     touch $MOUNT/readdir/newentry || return
 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000104"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
     ls $MOUNT/readdir || return 3
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
     rm -rf $MOUNT/readdir || return 4
 }
 run_test 13 "mdc_readpage restart test (bug 1138)"
@@ -165,30 +152,26 @@ test_14() {
     mkdir $MOUNT/readdir
     touch $MOUNT/readdir/newentry
 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000106"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
     ls $MOUNT/readdir || return 1
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 }
 run_test 14 "mdc_readpage resend test (bug 1138)"
 
 test_15() {
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
     touch $DIR/$tfile && return 1
     return 0
 }
 run_test 15 "failed open (-ENOMEM)"
 
-READ_AHEAD=`cat $LPROC/llite/*/max_read_ahead_mb | head -n 1`
+READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
 stop_read_ahead() {
-   for f in $LPROC/llite/*/max_read_ahead_mb; do 
-      echo 0 > $f
-   done
+   lctl set_param -n llite.*.max_read_ahead_mb 0
 }
 
 start_read_ahead() {
-   for f in $LPROC/llite/*/max_read_ahead_mb; do 
-      echo $READ_AHEAD > $f
-   done
+   lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
 }
 
 test_16() {
@@ -197,11 +180,11 @@ test_16() {
     stop_read_ahead
 
 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000504"
+    do_facet ost1 "lctl set_param fail_loc=0x80000504"
     cancel_lru_locks osc
     # OST bulk will time out here, client resends
     do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 1
-    do_facet ost1 sysctl -w lustre.fail_loc=0
+    do_facet ost1 lctl set_param fail_loc=0
     # give recovery a chance to finish (shouldn't take long)
     sleep $TIMEOUT
     do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
@@ -212,13 +195,13 @@ run_test 16 "timeout bulk put, don't evict client (2732)"
 test_17() {
     # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
     # OST bulk will time out here, client retries
-    do_facet ost1 sysctl -w lustre.fail_loc=0x80000503
+    do_facet ost1 lctl set_param fail_loc=0x80000503
     # need to ensure we send an RPC
     do_facet client cp /etc/termcap $DIR/$tfile
     sync
 
     sleep $TIMEOUT
-    do_facet ost1 sysctl -w lustre.fail_loc=0
+    do_facet ost1 lctl set_param fail_loc=0
     do_facet client "df $DIR"
     # expect cmp to succeed, client resent bulk
     do_facet client "cmp /etc/termcap $DIR/$tfile" || return 3
@@ -237,11 +220,11 @@ test_18a() {
     pgcache_empty || return 1
 
     # 1 stripe on ost2
-    lfs setstripe $f $((128 * 1024)) 1 1
+    lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1
 
-    do_facet client cp /etc/termcap $f
+    do_facet client cp $SAMPLE_FILE $f
     sync
-    local osc2dev=`grep ${ost2_svc}-osc- $LPROC/devices | egrep -v 'MDT' | awk '{print $1}'`
+    local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
     $LCTL --device $osc2dev deactivate || return 3
     # my understanding is that there should be nothing in the page
     # cache after the client reconnects?     
@@ -262,10 +245,10 @@ test_18b() {
     pgcache_empty || return 1
 
     # shouldn't have to set stripe size of count==1
-    lfs setstripe $f $((128 * 1024)) 0 1
-    lfs setstripe $f2 $((128 * 1024)) 0 1
+    lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
+    lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1
 
-    do_facet client cp /etc/termcap $f
+    do_facet client cp $SAMPLE_FILE $f
     sync
     ost_evict_client
     # allow recovery to complete
@@ -279,6 +262,37 @@ test_18b() {
 }
 run_test 18b "eviction and reconnect clears page cache (2766)"
 
+test_18c() {
+    do_facet client mkdir -p $MOUNT/$tdir
+    f=$MOUNT/$tdir/$tfile
+    f2=$MOUNT/$tdir/${tfile}-2
+
+    cancel_lru_locks osc
+    pgcache_empty || return 1
+
+    # shouldn't have to set stripe size of count==1
+    lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
+    lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1
+
+    do_facet client cp $SAMPLE_FILE $f
+    sync
+    ost_evict_client
+
+    # OBD_FAIL_OST_CONNECT_NET2
+    # lost reply to connect request
+    do_facet ost1 lctl set_param fail_loc=0x80000225
+    # force reconnect
+    df $MOUNT > /dev/null 2>&1
+    sleep 2
+    # my understanding is that there should be nothing in the page
+    # cache after the client reconnects?     
+    rc=0
+    pgcache_empty || rc=2
+    rm -f $f $f2
+    return $rc
+}
+run_test 18c "Dropped connect reply after eviction handing (14755)"
+
 test_19a() {
     f=$MOUNT/$tfile
     do_facet client mcreate $f        || return 1
@@ -307,12 +321,12 @@ run_test 19b "test expired_lock_main on ost (2867)"
 
 test_20a() {	# bug 2983 - ldlm_handle_enqueue cleanup
 	mkdir -p $DIR/$tdir
-	multiop $DIR/$tdir/${tfile} O_wc &
+	lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
+	multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
 	MULTI_PID=$!
-	sleep 1
 	cancel_lru_locks osc
 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
-	do_facet ost1 sysctl -w lustre.fail_loc=0x80000308
+	do_facet ost1 lctl set_param fail_loc=0x80000308
 	kill -USR1 $MULTI_PID
 	wait $MULTI_PID
 	rc=$?
@@ -322,10 +336,10 @@ run_test 20a "ldlm_handle_enqueue error (should return error)"
 
 test_20b() {	# bug 2986 - ldlm_handle_enqueue error during open
 	mkdir -p $DIR/$tdir
-	touch $DIR/$tdir/${tfile}
+	lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
 	cancel_lru_locks osc
 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
-	do_facet ost1 sysctl -w lustre.fail_loc=0x80000308
+	do_facet ost1 lctl set_param fail_loc=0x80000308
 	dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
 		error "didn't fail open enqueue" || true
 }
@@ -334,21 +348,21 @@ run_test 20b "ldlm_handle_enqueue error (should return error)"
 test_21a() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        close_pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000129"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
        multiop $DIR/$tdir-2/f Oc &
        open_pid=$!
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
        kill -USR1 $close_pid
        cancel_lru_locks mdc
        wait $close_pid || return 1
        wait $open_pid || return 2
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
        $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
@@ -360,14 +374,14 @@ run_test 21a "drop close request while close and open are both in flight"
 test_21b() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        close_pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
        mcreate $DIR/$tdir-2/f &
        open_pid=$!
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        kill -USR1 $close_pid
        cancel_lru_locks mdc
@@ -383,22 +397,22 @@ run_test 21b "drop open request while close and open are both in flight"
 test_21c() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        close_pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
        mcreate $DIR/$tdir-2/f &
        open_pid=$!
        sleep 3
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
        kill -USR1 $close_pid
        cancel_lru_locks mdc
        wait $close_pid || return 1
        wait $open_pid || return 2
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
        $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
@@ -409,19 +423,19 @@ run_test 21c "drop both request while close and open are both in flight"
 test_21d() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000129"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
        multiop $DIR/$tdir-2/f Oc &
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
        kill -USR1 $pid
        cancel_lru_locks mdc
        wait $pid || return 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
        $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
@@ -433,13 +447,13 @@ run_test 21d "drop close reply while close and open are both in flight"
 test_21e() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
        touch $DIR/$tdir-2/f &
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        kill -USR1 $pid
        cancel_lru_locks mdc
@@ -455,19 +469,19 @@ run_test 21e "drop open reply while close and open are both in flight"
 test_21f() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
        touch $DIR/$tdir-2/f &
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
        kill -USR1 $pid
        cancel_lru_locks mdc
        wait $pid || return 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
        $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
@@ -478,19 +492,19 @@ run_test 21f "drop both reply while close and open are both in flight"
 test_21g() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
        touch $DIR/$tdir-2/f &
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
        kill -USR1 $pid
        cancel_lru_locks mdc
        wait $pid || return 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
        $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
@@ -501,20 +515,20 @@ run_test 21g "drop open reply and close request while close and open are both in
 test_21h() {
        mkdir -p $DIR/$tdir-1
        mkdir -p $DIR/$tdir-2
-       multiop $DIR/$tdir-1/f O_c &
+       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
        pid=$!
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
        touch $DIR/$tdir-2/f &
        touch_pid=$!
        sleep 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
        cancel_lru_locks mdc
        kill -USR1 $pid
        wait $pid || return 1
-       do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+       do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
        wait $touch_pid || return 2
 
@@ -529,7 +543,7 @@ test_22() {
     f1=$DIR/${tfile}-1
     f2=$DIR/${tfile}-2
     
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
     multiop $f2 Oc &
     close_pid=$!
 
@@ -537,7 +551,7 @@ test_22() {
     multiop $f1 msu || return 1
 
     cancel_lru_locks mdc
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
 
     wait $close_pid || return 2
     rm -rf $f2 || return 4
@@ -545,7 +559,7 @@ test_22() {
 run_test 22 "drop close request and do mknod"
 
 test_23() { #b=4561
-    multiop $DIR/$tfile O_c &
+    multiop_bg_pause $DIR/$tfile O_c || return 1
     pid=$!
     # give a chance for open
     sleep 5
@@ -562,51 +576,50 @@ run_test 23 "client hang when close a file after mds crash"
 test_24() {	# bug 2248 - eviction fails writeback but app doesn't see it
 	mkdir -p $DIR/$tdir
 	cancel_lru_locks osc
-	multiop $DIR/$tdir/$tfile Owy_wyc &
+	multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
 	MULTI_PID=$!
-	usleep 500
 	ost_evict_client
-	usleep 500
 	kill -USR1 $MULTI_PID
 	wait $MULTI_PID
 	rc=$?
-	sysctl -w lustre.fail_loc=0x0
+	lctl set_param fail_loc=0x0
 	client_reconnect
-	[ $rc -eq 0 ] && error "multiop didn't fail fsync: rc $rc" || true
+	[ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
 }
 run_test 24 "fsync error (should return error)"
 
-test_26() {      # bug 5921 - evict dead exports by pinger
+test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
 # this test can only run from a client on a separate node.
-	remote_ost || skip "local OST" && return
-	remote_mds || skip "local MDS" && return
-	OST_FILE=$LPROC/obdfilter/${ost1_svc}/num_exports
-        OST_EXP="`do_facet ost1 cat $OST_FILE`"
+	remote_ost || { skip "local OST" && return 0; }
+	remote_mds || { skip "local MDS" && return 0; }
+	OST_FILE=obdfilter.${ost1_svc}.num_exports
+        OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
 	OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
 	echo starting with $OST_NEXP1 OST exports
 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
-	do_facet client sysctl -w lustre.fail_loc=0x505
+	do_facet client lctl set_param fail_loc=0x505
 	# evictor takes up to 2.25x to evict.  But if there's a 
 	# race to start the evictor from various obds, the loser
 	# might have to wait for the next ping.
 	echo Waiting for $(($TIMEOUT * 4)) secs
 	sleep $(($TIMEOUT * 4))
-        OST_EXP="`do_facet ost1 cat $OST_FILE`"
+        OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
 	OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
 	echo ending with $OST_NEXP2 OST exports
-	do_facet client sysctl -w lustre.fail_loc=0x0
+	do_facet client lctl set_param fail_loc=0x0
         [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
 	return 0
 }
-run_test 26 "evict dead exports"
+run_test 26a "evict dead exports"
 
 test_26b() {      # bug 10140 - evict dead exports by pinger
 	client_df
         zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
-	MDS_FILE=$LPROC/mdt/${mds1_svc}/num_exports
-        MDS_NEXP1="`do_facet $SINGLEMDS cat $MDS_FILE | cut -d' ' -f2`"
-	OST_FILE=$LPROC/obdfilter/${ost1_svc}/num_exports
-        OST_NEXP1="`do_facet ost1 cat $OST_FILE | cut -d' ' -f2`"
+        sleep 1 # wait connections being established
+	MDS_FILE=mdt.${mds1_svc}.num_exports
+        MDS_NEXP1="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
+	OST_FILE=obdfilter.${ost1_svc}.num_exports
+        OST_NEXP1="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
 	echo starting with $OST_NEXP1 OST and $MDS_NEXP1 MDS exports
 	zconf_umount `hostname` $MOUNT2 -f
 	# evictor takes up to 2.25x to evict.  But if there's a 
@@ -614,8 +627,8 @@ test_26b() {      # bug 10140 - evict dead exports by pinger
 	# might have to wait for the next ping.
 	echo Waiting for $(($TIMEOUT * 4)) secs
 	sleep $(($TIMEOUT * 4))
-        OST_NEXP2="`do_facet ost1 cat $OST_FILE | cut -d' ' -f2`"
-        MDS_NEXP2="`do_facet $SINGLEMDS cat $MDS_FILE | cut -d' ' -f2`"
+        OST_NEXP2="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
+        MDS_NEXP2="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
 	echo ending with $OST_NEXP2 OST and $MDS_NEXP2 MDS exports
         [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted from OST"
         [ $MDS_NEXP1 -le $MDS_NEXP2 ] && error "client not evicted from MDS"
@@ -632,10 +645,10 @@ test_27() {
 	FAILURE_MODE="SOFT"
 	facet_failover $SINGLEMDS
 #define OBD_FAIL_OSC_SHUTDOWN            0x407
-	sysctl -w lustre.fail_loc=0x80000407
+	lctl set_param fail_loc=0x80000407
 	# need to wait for reconnect
 	echo -n waiting for fail_loc
-	while [ `sysctl -n lustre.fail_loc` -eq -2147482617 ]; do
+	while [ `lctl get_param -n fail_loc` -eq -2147482617 ]; do
 	    sleep 1
 	    echo -n .
 	done
@@ -651,7 +664,7 @@ test_28() {      # bug 6086 - error adding new clients
 	do_facet client mcreate $MOUNT/$tfile       || return 1
 	drop_bl_callback "chmod 0777 $MOUNT/$tfile" ||echo "evicted as expected"
 	#define OBD_FAIL_MDS_CLIENT_ADD 0x12f
-	do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012f"
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
 	# fail once (evicted), reconnect fail (fail_loc), ok
 	df || (sleep 10; df) || (sleep 10; df) || error "reconnect failed"
 	rm -f $MOUNT/$tfile
@@ -680,7 +693,7 @@ test_50() {
 	rc=$?
 	echo writemany returned $rc
 	#these may fail because of eviction due to slow AST response.
-	return $rc
+	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
 }
 run_test 50 "failover MDS under load"
 
@@ -708,7 +721,7 @@ test_51() {
 	wait $CLIENT_PID 
 	rc=$?
 	echo writemany returned $rc
-	return $rc
+	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
 }
 run_test 51 "failover MDS during recovery"
 
@@ -750,7 +763,7 @@ run_test 52 "failover OST under load"
 # test of open reconstruct
 test_53() {
 	touch $DIR/$tfile
-	drop_ldlm_reply "./openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
+	drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
 		return 2
 }
 run_test 53 "touch: drop rep"
@@ -790,7 +803,7 @@ test_55() {
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
         #define OBD_FAIL_OST_DROP_REQ            0x21d
-	do_facet ost sysctl -w lustre.fail_loc=0x0000021d
+	do_facet ost lctl set_param fail_loc=0x0000021d
 	# second dd will be never finished
 	dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4  &	
 	DDPID=$!
@@ -809,7 +822,7 @@ test_55() {
 	echo "(dd_pid=$DDPID, time=$count)successful"
 
 	#Recover fail_loc and dd will finish soon
-	do_facet ost sysctl -w lustre.fail_loc=0
+	do_facet ost lctl set_param fail_loc=0
 	count=0
 	echo  "step3: testing ......"
 	while [ true ]; do
@@ -829,19 +842,17 @@ run_test 55 "ost_brw_read/write drops timed-out read/write request"
 test_56() { # b=11277
 #define OBD_FAIL_MDS_RESEND      0x136
         touch $DIR/$tfile
-        do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000136"
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
         stat $DIR/$tfile
-        do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0"
         rm -f $DIR/$tfile
 }
 run_test 56 "do not allow reconnect to busy exports"
 
 test_57_helper() {
         # no oscs means no client or mdt 
-        while [ -e $LPROC/osc ]; do
-                for f in `find $LPROC -type f`; do 
-                        cat $f > /dev/null 2>&1
-                done
+        while lctl get_param osc.*.* > /dev/null 2>&1; do
+                : # loop until proc file is removed
         done
 }
 
@@ -850,12 +861,12 @@ test_57() { # bug 10866
         pid=$!
         sleep 1
 #define OBD_FAIL_LPROC_REMOVE            0xB00
-        sysctl -w lustre.fail_loc=0x80000B00
+        lctl set_param fail_loc=0x80000B00
         zconf_umount `hostname` $DIR
-        sysctl -w lustre.fail_loc=0x80000B00
+        lctl set_param fail_loc=0x80000B00
         fail_abort $SINGLEMDS
         kill -9 $pid
-        sysctl -w lustre.fail_loc=0
+        lctl set_param fail_loc=0
         mount_client $DIR
         do_facet client "df $DIR"
 }
@@ -865,17 +876,35 @@ test_58() { # bug 11546
 #define OBD_FAIL_MDC_ENQUEUE_PAUSE        0x801
         touch $MOUNT/$tfile
         ls -la $MOUNT/$tfile
-        sysctl -w lustre.fail_loc=0x80000801
+        lctl set_param fail_loc=0x80000801
         cp $MOUNT/$tfile /dev/null &
         pid=$!
         sleep 1
-        sysctl -w lustre.fail_loc=0
+        lctl set_param fail_loc=0
         drop_bl_callback rm -f $MOUNT/$tfile
         wait $pid
         do_facet client "df $DIR"
 }
 run_test 58 "Eviction in the middle of open RPC reply processing"
 
+test_59() { # bug 10589
+	zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
+	echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
+#define OBD_FAIL_LDLM_CANCEL_EVICT_RACE  0x311
+	lctl set_param fail_loc=0x311
+	writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
+	[ $? = 0 ] || error "dd write failed"
+	writes=$(echo $writes | awk  -F '+' '/out/ {print $1}')
+	lctl set_param fail_loc=0
+	sync
+	zconf_umount `hostname` $MOUNT2 -f
+	reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
+	[ $? = 0 ] || error "dd read failed"
+	reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
+	[ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
+}
+run_test 59 "Read cancel race on client eviction"
+
 equals_msg `basename $0`: test complete, cleaning up
-$CLEANUP
+check_and_cleanup_lustre
 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true