drop truncate capa after ll_truncate.

[fs/lustre-release.git] / lustre / tests / replay-single.sh

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index b8cbd9b..742c114 100755 (executable)
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -39,13 +39,10 @@ gen_config() {
            --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
        MDS=lmv1
     else
-        add_mds mds1 --dev $MDSDEV --size $MDSSIZE
-        if [ ! -z "$mds1failover_HOST" ]; then
-            add_mdsfailover mds1 --dev $MDSDEV --size $MDSSIZE
-        fi
-       add_lov lov1 mds1 --stripe_sz $STRIPE_BYTES \
+        add_mds $SINGLEMDS --dev $MDSDEV --size $MDSSIZE
+       add_lov lov1 $SINGLEMDS --stripe_sz $STRIPE_BYTES \
            --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
-       MDS=mds1_svc
+       MDS=$SINGLEMDS
     fi
     
     add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
@@ -58,18 +55,18 @@ build_test_filter
 cleanup() {
     # make sure we are using the primary MDS, so the config log will
     # be able to clean up properly.
-    activemds=`facet_active mds1`
-    if [ $activemds != "mds1" ]; then
-        fail mds1
+    activemds=`facet_active $SINGLEMDS`
+    if [ $activemds != "$SINGLEMDS" ]; then
+        fail $SINGLEMDS
     fi
     zconf_umount `hostname` $MOUNT
     for mds in `mds_list`; do
        stop $mds ${FORCE} $MDSLCONFARGS
     done
-    stop_lgssd
-    stop_lsvcgssd
     stop ost2 ${FORCE} --dump cleanup.log
     stop ost ${FORCE} --dump cleanup.log
+    stop_lgssd
+    stop_lsvcgssd
 }
 
 if [ "$ONLY" == "cleanup" ]; then
@@ -85,10 +82,10 @@ setup() {
     gen_config
 
     start_krb5_kdc || exit 1
-    start ost --reformat $OSTLCONFARGS 
-    start ost2 --reformat $OSTLCONFARGS 
     start_lsvcgssd || exit 2
     start_lgssd || exit 3
+    start ost --reformat $OSTLCONFARGS 
+    start ost2 --reformat $OSTLCONFARGS 
     [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
     for mds in `mds_list`; do
        start $mds --reformat $MDSLCONFARGS
@@ -105,8 +102,8 @@ fi
 mkdir -p $DIR
 
 test_0() {
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
 }
 run_test 0 "empty replay"
 
@@ -120,29 +117,33 @@ test_0b() {
 run_test 0b "ensure object created after recover exists. (3284)"
 
 test_0c() {
+    if [ `using_krb5_sec $SECURITY` == 'n' ] ; then
+        echo "Skip 0c in non-gss mode"
+        return 0
+    fi
     # drop gss error notification
-    replay_barrier mds1
-    fail_drop mds1 0x760
+    replay_barrier $SINGLEMDS
+    fail_drop $SINGLEMDS 0x760
 
     # drop gss init request
-    replay_barrier mds1
-    fail_drop mds1 0x780
+    replay_barrier $SINGLEMDS
+    fail_drop $SINGLEMDS 0x780
 }
 run_test 0c "empty replay with gss init failures"
 
 test_1() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mcreate $DIR/$tfile
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile || return 1
     rm $DIR/$tfile
 }
 run_test 1 "simple create"
 
 test_2a() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     touch $DIR/$tfile
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile || return 1
     rm $DIR/$tfile
 }
@@ -150,43 +151,43 @@ run_test 2a "touch"
 
 test_2b() {
     ./mcreate $DIR/$tfile
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     touch $DIR/$tfile
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile || return 1
     rm $DIR/$tfile
 }
 run_test 2b "touch"
 
 test_3a() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mcreate $DIR/$tfile
     o_directory $DIR/$tfile
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile || return 2
     rm $DIR/$tfile
 }
 run_test 3a "replay failed open(O_DIRECTORY)"
 
 test_3b() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
     do_facet mds "sysctl -w lustre.fail_loc=0x80000114"
     touch $DIR/$tfile
     do_facet mds "sysctl -w lustre.fail_loc=0"
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile && return 2
     return 0
 }
 run_test 3b "replay failed open -ENOMEM"
 
 test_3c() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
     do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
     touch $DIR/$tfile
     do_facet mds "sysctl -w lustre.fail_loc=0"
-    fail mds1
+    fail $SINGLEMDS
 
     $CHECKSTAT -t file $DIR/$tfile && return 2
     return 0
@@ -194,11 +195,11 @@ test_3c() {
 run_test 3c "replay failed open -ENOMEM"
 
 test_4() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     for i in `seq 10`; do
         echo "tag-$i" > $DIR/$tfile-$i
     done 
-    fail mds1
+    fail $SINGLEMDS
     for i in `seq 10`; do
       grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
     done 
@@ -206,9 +207,9 @@ test_4() {
 run_test 4 "|x| 10 open(O_CREAT)s"
 
 test_4b() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -rf $DIR/$tfile-*
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
 }
 run_test 4b "|x| rm 10 files"
@@ -216,11 +217,11 @@ run_test 4b "|x| rm 10 files"
 # The idea is to get past the first block of precreated files on both 
 # osts, and then replay.
 test_5() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     for i in `seq 220`; do
         echo "tag-$i" > $DIR/$tfile-$i
     done 
-    fail mds1
+    fail $SINGLEMDS
     for i in `seq 220`; do
       grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
     done 
@@ -232,10 +233,10 @@ run_test 5 "|x| 220 open(O_CREAT)"
 
 
 test_6() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mkdir $DIR/$tdir
     mcreate $DIR/$tdir/$tfile
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t dir $DIR/$tdir || return 1
     $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
     sleep 2
@@ -244,18 +245,18 @@ test_6() {
 run_test 6 "mkdir + contained create"
 
 test_6b() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -rf $DIR/$tdir
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t dir $DIR/$tdir && return 1 || true 
 }
 run_test 6b "|X| rmdir"
 
 test_7() {
     mkdir $DIR/$tdir
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mcreate $DIR/$tdir/$tfile
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t dir $DIR/$tdir || return 1
     $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
     rm -fr $DIR/$tdir
@@ -263,11 +264,11 @@ test_7() {
 run_test 7 "mkdir |X| contained create"
 
 test_8() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile mo_c &
     MULTIPID=$!
     sleep 1
-    fail mds1
+    fail $SINGLEMDS
     ls $DIR/$tfile
     $CHECKSTAT -t file $DIR/$tfile || return 1
     kill -USR1 $MULTIPID || return 2
@@ -277,10 +278,10 @@ test_8() {
 run_test 8 "creat open |X| close"
 
 test_9() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mcreate $DIR/$tfile
     local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
-    fail mds1
+    fail $SINGLEMDS
     local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
 
     echo " old_inum == $old_inum, new_inum == $new_inum"
@@ -297,10 +298,10 @@ run_test 9  "|X| create (same inum/gen)"
 
 test_10() {
     mcreate $DIR/$tfile
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mv $DIR/$tfile $DIR/$tfile-2
     rm -f $DIR/$tfile
-    fail mds1
+    fail $SINGLEMDS
     
     $CHECKSTAT $DIR/$tfile && return 1
     $CHECKSTAT $DIR/$tfile-2 || return 2
@@ -313,11 +314,11 @@ test_11() {
     mcreate $DIR/$tfile
     echo "old" > $DIR/$tfile
     mv $DIR/$tfile $DIR/$tfile-2
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     echo "new" > $DIR/$tfile
     grep new $DIR/$tfile 
     grep old $DIR/$tfile-2
-    fail mds1
+    fail $SINGLEMDS
     grep new $DIR/$tfile || return 1
     grep old $DIR/$tfile-2 || return 2
 }
@@ -330,11 +331,11 @@ test_12() {
     # give multiop a chance to open
     sleep 1
     rm -f $DIR/$tfile
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 1
 
-    fail mds1
+    fail $SINGLEMDS
     [ -e $DIR/$tfile ] && return 2
     return 0
 }
@@ -351,8 +352,8 @@ test_13() {
     sleep 1 
     chmod 0 $DIR/$tfile
     $CHECKSTAT -p 0 $DIR/$tfile
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 1
 
@@ -367,11 +368,11 @@ test_14() {
     # give multiop a chance to open
     sleep 1 
     rm -f $DIR/$tfile
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     kill -USR1 $pid || return 1
     wait $pid || return 2
 
-    fail mds1
+    fail $SINGLEMDS
     [ -e $DIR/$tfile ] && return 3
     return 0
 }
@@ -383,12 +384,12 @@ test_15() {
     # give multiop a chance to open
     sleep 1 
     rm -f $DIR/$tfile
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     touch $DIR/g11 || return 1
     kill -USR1 $pid
     wait $pid || return 2
 
-    fail mds1
+    fail $SINGLEMDS
     [ -e $DIR/$tfile ] && return 3
     touch $DIR/h11 || return 4
     return 0
@@ -397,11 +398,11 @@ run_test 15 "open(O_CREAT), unlink |X|  touch new, close"
 
 
 test_16() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mcreate $DIR/$tfile
     munlink $DIR/$tfile
     mcreate $DIR/$tfile-2
-    fail mds1
+    fail $SINGLEMDS
     [ -e $DIR/$tfile ] && return 1
     [ -e $DIR/$tfile-2 ] || return 2
     munlink $DIR/$tfile-2 || return 3
@@ -409,12 +410,12 @@ test_16() {
 run_test 16 "|X| open(O_CREAT), unlink, touch new,  unlink new"
 
 test_17() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile O_c &
     pid=$!
     # give multiop a chance to open
     sleep 1 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid || return 1
     wait $pid || return 2
     $CHECKSTAT -t file $DIR/$tfile || return 3
@@ -423,7 +424,7 @@ test_17() {
 run_test 17 "|X| open(O_CREAT), |replay| close"
 
 test_18() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile O_tSc &
     pid=$!
     # give multiop a chance to open
@@ -434,7 +435,7 @@ test_18() {
     kill -USR1 $pid
     wait $pid || return 2
 
-    fail mds1
+    fail $SINGLEMDS
     [ -e $DIR/$tfile ] && return 3
     [ -e $DIR/$tfile-2 ] || return 4
     # this touch frequently fails
@@ -447,25 +448,25 @@ run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
 
 # bug 1855 (a simpler form of test_11 above)
 test_19() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     mcreate $DIR/$tfile
     echo "old" > $DIR/$tfile
     mv $DIR/$tfile $DIR/$tfile-2
     grep old $DIR/$tfile-2
-    fail mds1
+    fail $SINGLEMDS
     grep old $DIR/$tfile-2 || return 2
 }
 run_test 19 "|X| mcreate, open, write, rename "
 
 test_20() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile O_tSc &
     pid=$!
     # give multiop a chance to open
     sleep 1 
     rm -f $DIR/$tfile
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 1
     [ -e $DIR/$tfile ] && return 2
@@ -474,7 +475,7 @@ test_20() {
 run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
 
 test_21() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile O_tSc &
     pid=$!
     # give multiop a chance to open
@@ -482,7 +483,7 @@ test_21() {
     rm -f $DIR/$tfile
     touch $DIR/g11 || return 1
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 2
     [ -e $DIR/$tfile ] && return 3
@@ -497,10 +498,10 @@ test_22() {
     # give multiop a chance to open
     sleep 1 
 
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 1
     [ -e $DIR/$tfile ] && return 2
@@ -514,11 +515,11 @@ test_23() {
     # give multiop a chance to open
     sleep 1 
 
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile
     touch $DIR/g11 || return 1
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 2
     [ -e $DIR/$tfile ] && return 3
@@ -533,8 +534,8 @@ test_24() {
     # give multiop a chance to open
     sleep 1 
 
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
     rm -f $DIR/$tfile
     kill -USR1 $pid
     wait $pid || return 1
@@ -550,8 +551,8 @@ test_25() {
     sleep 1 
     rm -f $DIR/$tfile
 
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
     kill -USR1 $pid
     wait $pid || return 1
     [ -e $DIR/$tfile ] && return 2
@@ -560,7 +561,7 @@ test_25() {
 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
 
 test_26() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile-1 O_tSc &
     pid1=$!
     multiop $DIR/$tfile-2 O_tSc &
@@ -572,7 +573,7 @@ test_26() {
     kill -USR1 $pid2
     wait $pid2 || return 1
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid1
     wait $pid1 || return 2
     [ -e $DIR/$tfile-1 ] && return 3
@@ -582,7 +583,7 @@ test_26() {
 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
 
 test_27() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     multiop $DIR/$tfile-1 O_tSc &
     pid1=$!
     multiop $DIR/$tfile-2 O_tSc &
@@ -592,7 +593,7 @@ test_27() {
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid1
     wait $pid1 || return 1
     kill -USR1 $pid2
@@ -610,13 +611,13 @@ test_28() {
     pid2=$!
     # give multiop a chance to open
     sleep 1 
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
     kill -USR1 $pid2
     wait $pid2 || return 1
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid1
     wait $pid1 || return 2
     [ -e $DIR/$tfile-1 ] && return 3
@@ -632,11 +633,11 @@ test_29() {
     pid2=$!
     # give multiop a chance to open
     sleep 1 
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
 
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid1
     wait $pid1 || return 1
     kill -USR1 $pid2
@@ -657,8 +658,8 @@ test_30() {
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
 
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
     kill -USR1 $pid1
     wait $pid1 || return 1
     kill -USR1 $pid2
@@ -678,9 +679,9 @@ test_31() {
     sleep 1 
     rm -f $DIR/$tfile-1
 
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile-2
-    fail mds1
+    fail $SINGLEMDS
     kill -USR1 $pid1
     wait $pid1 || return 1
     kill -USR1 $pid2
@@ -698,22 +699,24 @@ test_32() {
     pid1=$!
     multiop $DIR/$tfile O_c &
     pid2=$!
-    # give multiop a chance to open
-    sleep 1
+    # give multiop a chance to open.
+    # 1 second is not enough, I increased it to 5, however in ideal word
+    # I should have to wait for open finish in more smart manner. --umka
+    sleep 5
     mds_evict_client
     df $MOUNT || sleep 1 && df $MOUNT || return 1
     kill -USR1 $pid1
     kill -USR1 $pid2
-    sleep 1
+    sleep 5
     return 0
 }
 run_test 32 "close() notices client eviction; close() after client eviction"
 
 # Abort recovery before client complete
 test_33() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     touch $DIR/$tfile
-    fail_abort mds1
+    fail_abort $SINGLEMDS
     # this file should be gone, because the replay was aborted
     $CHECKSTAT -t file $DIR/$tfile && return 1
     return 0
@@ -727,8 +730,8 @@ test_34() {
     sleep 1 
     rm -f $DIR/$tfile
 
-    replay_barrier mds1
-    fail_abort mds1
+    replay_barrier $SINGLEMDS
+    fail_abort $SINGLEMDS
     kill -USR1 $pid
     [ -e $DIR/$tfile ] && return 1
     sync
@@ -747,7 +750,7 @@ test_35() {
     sync
     sleep 1
     # give a chance to remove from MDS
-    fail_abort mds1
+    fail_abort $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile && return 1 || true
 }
 run_test 35 "test recovery from llog for unlink op"
@@ -755,10 +758,10 @@ run_test 35 "test recovery from llog for unlink op"
 # b=2432 resent cancel after replay uses wrong cookie,
 # so don't resend cancels
 test_36() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     touch $DIR/$tfile
     checkstat $DIR/$tfile
-    facet_failover mds1
+    facet_failover $SINGLEMDS
     cancel_lru_locks MDC
     if dmesg | grep "unknown lock cookie"; then 
        echo "cancel after replay failed"
@@ -777,10 +780,10 @@ test_37() {
     sleep 1 
     rmdir $DIR/$tfile
 
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     # clear the dmesg buffer so we only see errors from this recovery
     dmesg -c >/dev/null
-    fail_abort mds1
+    fail_abort $SINGLEMDS
     kill -USR1 $pid
     dmesg | grep  "mds_unlink_orphan.*error .* unlinking orphan" && return 1
     sync
@@ -791,8 +794,8 @@ run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans
 test_38() {
     createmany -o $DIR/$tfile-%d 800
     unlinkmany $DIR/$tfile-%d 0 400
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
     unlinkmany $DIR/$tfile-%d 400 400
     sleep 2
     $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
@@ -801,9 +804,9 @@ run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
 
 test_39() {
     createmany -o $DIR/$tfile-%d 800
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     unlinkmany $DIR/$tfile-%d 0 400
-    fail mds1
+    fail $SINGLEMDS
     unlinkmany $DIR/$tfile-%d 400 400
     sleep 2
     $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
@@ -823,7 +826,7 @@ test_40(){
     writeme -s $MOUNT/${tfile}-2 &
     WRITE_PID=$!
     sleep 1
-    facet_failover mds1
+    facet_failover $SINGLEMDS
 #define OBD_FAIL_MDS_CONNECT_NET         0x117
     do_facet mds "sysctl -w lustre.fail_loc=0x80000117"
     kill -USR1 $PID
@@ -895,11 +898,11 @@ run_test 42 "recovery after ost failure"
 # b=2530
 # timeout in MDS/OST recovery RPC will LBUG MDS
 test_43() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
 
     # OBD_FAIL_OST_CREATE_NET 0x204
     do_facet ost "sysctl -w lustre.fail_loc=0x80000204"
-    facet_failover mds1
+    facet_failover $SINGLEMDS
     df $MOUNT || return 1
     sleep 10
     do_facet ost "sysctl -w lustre.fail_loc=0"
@@ -946,7 +949,7 @@ run_test 45 "Handle failed close"
 test_46() {
     dmesg -c >/dev/null
     drop_reply "touch $DIR/$tfile"
-    fail mds1
+    fail $SINGLEMDS
     # ironically, the previous test, 45, will cause a real forced close,
     # so just look for one for this test
     dmesg | grep -i "force closing client file handle for $tfile" && return 1
@@ -984,13 +987,15 @@ test_48() {
     createmany -o $DIR/${tfile}- 100
     $CHECKSTAT $DIR/${tfile}-99 || return 1
     mds_evict_client
-    df $MOUNT || df $MOUNT || return 2
+    df $MOUNT || echo "first df failed" 
+    sleep 1 
+    df $MOUNT || return 2
     sleep 1
     $CHECKSTAT $DIR/${tfile}-99 || return 3
 
     dmesg -c >/dev/null
-    replay_barrier mds1
-    fail mds1
+    replay_barrier $SINGLEMDS
+    fail $SINGLEMDS
     unlinkmany $DIR/${tfile}- 100 || return 4
     if dmesg | grep "back in time"; then 
        echo "server went back in time!"
@@ -1002,17 +1007,17 @@ run_test 48 "Don't lose transno when client is evicted (2525)"
 
 # b=3550 - replay of unlink
 test_49() {
-    replay_barrier mds1
+    replay_barrier $SINGLEMDS
     createmany -o $DIR/$tfile-%d 400 || return 1
     unlinkmany $DIR/$tfile-%d 0 400 || return 2
-    fail mds1
+    fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
 }
 run_test 49 "re-write records to llog as written during fail"
 
 test_50() {
     local osc_dev=`$LCTL device_list | \
-               awk '(/ost_svc_mds1_svc/){print $4}' `
+               awk '(/ost_svc_$SINGLEMDS_svc/){print $4}' `
     $LCTL --device %$osc_dev recover &&  $LCTL --device %$osc_dev recover
     # give the mds_lov_sync threads a chance to run
     sleep 5
@@ -1021,7 +1026,7 @@ run_test 50 "Double OSC recovery, don't LASSERT (3812)"
 
 # bug 3462 - simultaneous MDC requests
 test_51a() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/${tdir}-1
     mkdir -p $DIR/${tdir}-2
     touch $DIR/${tdir}-2/f
@@ -1030,12 +1035,12 @@ test_51a() {
     # give multiop a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
     kill -USR1 $pid
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
     $CHECKSTAT -t file $DIR/${tdir}-2/f || return 1
 
-    fail mds
+    fail $SINGLEMDS
 
     wait $pid || return 2
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
@@ -1044,23 +1049,26 @@ test_51a() {
 run_test 51a "|X| close request while two MDC requests in flight"
 
 test_51b() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/$tdir-1
     mkdir -p $DIR/$tdir-2
     multiop $DIR/$tdir-1/f O_c &
     pid=$!
+
     # give multiop a chance to open
-    sleep 1
+    # 1 second seems to be not enough, we met already such a cases
+    # --umka
+    sleep 5
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
     touch $DIR/${tdir}-2/f &
     usleep 500
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
     kill -USR1 $pid
     wait $pid || return 1
                                                                                                                              
-    fail mds
+    fail $SINGLEMDS
 
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 2
     $CHECKSTAT -t file $DIR/${tdir}-2/f || return 3
@@ -1069,7 +1077,7 @@ test_51b() {
 run_test 51b "|X| open request while two MDC requests in flight"
 
 test_51c() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/${tdir}-1
     mkdir -p $DIR/${tdir}-2
     multiop $DIR/${tdir}-1/f O_c &
@@ -1077,15 +1085,15 @@ test_51c() {
     # give multiop a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
     touch $DIR/${tdir}-2/f &
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
     kill -USR1 $pid
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    fail mds
+    fail $SINGLEMDS
 
     wait $pid || return 1
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 2
@@ -1095,7 +1103,7 @@ test_51c() {
 run_test 51c "|X| open request and close request while two MDC requests in flight"
 
 test_51d() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/${tdir}-1
     mkdir -p $DIR/${tdir}-2
     touch $DIR/${tdir}-2/f
@@ -1104,12 +1112,12 @@ test_51d() {
     # give multiop a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122"
     kill -USR1 $pid
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
     #$CHECKSTAT -t file $DIR/${tdir}-2/f || return 1
 
-    fail mds
+    fail $SINGLEMDS
 
     wait $pid || return 2
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
@@ -1118,7 +1126,7 @@ test_51d() {
 run_test 51d "|X| close reply while two MDC requests in flight"
 
 test_51e() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/$tdir-1
     mkdir -p $DIR/$tdir-2
     multiop $DIR/$tdir-1/f O_c &
@@ -1126,15 +1134,15 @@ test_51e() {
     # give multiop a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
     touch $DIR/${tdir}-2/f &
     usleep 500
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
     kill -USR1 $pid
     wait $pid || return 1
 
-    fail mds
+    fail $SINGLEMDS
 
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 2
     $CHECKSTAT -t file $DIR/${tdir}-2/f || return 3
@@ -1143,7 +1151,7 @@ test_51e() {
 run_test 51e "|X| open reply while two MDC requests in flight"
 
 test_51f() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/${tdir}-1
     mkdir -p $DIR/${tdir}-2
     multiop $DIR/${tdir}-1/f O_c &
@@ -1151,15 +1159,15 @@ test_51f() {
     # give multiop a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
     touch $DIR/${tdir}-2/f &
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122"
     kill -USR1 $pid
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    fail mds
+    fail $SINGLEMDS
 
     wait $pid || return 1
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 2
@@ -1169,7 +1177,7 @@ test_51f() {
 run_test 51f "|X| open reply and close reply while two MDC requests in flight"
 
 test_51g() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/${tdir}-1
     mkdir -p $DIR/${tdir}-2
     multiop $DIR/${tdir}-1/f O_c &
@@ -1177,15 +1185,15 @@ test_51g() {
     # give multiop a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
     touch $DIR/${tdir}-2/f &
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
     kill -USR1 $pid
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    fail mds
+    fail $SINGLEMDS
 
     wait $pid || return 1
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 2
@@ -1195,23 +1203,23 @@ test_51g() {
 run_test 51g "|X| open reply and close request while two MDC requests in flight"
 
 test_51h() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     mkdir -p $DIR/${tdir}-1
     mkdir -p $DIR/${tdir}-2
     multiop $DIR/${tdir}-1/f O_c &
     pid=$!
-    # give multiop a chance to open
+    # give multio:wp a chance to open
     sleep 1
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
     touch $DIR/${tdir}-2/f &
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000122"
     kill -USR1 $pid
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    fail mds
+    fail $SINGLEMDS
 
     wait $pid || return 1
     $CHECKSTAT -t file $DIR/${tdir}-1/f || return 2
@@ -1226,37 +1234,37 @@ test_52() {
     cancel_lru_locks MDC
 
     multiop $DIR/$tfile s
-    replay_barrier mds1
-    do_facet mds1 "sysctl -w lustre.fail_loc=0x8000030c"
-    fail mds1
-    do_facet mds1 "sysctl -w lustre.fail_loc=0x0"
+    replay_barrier $SINGLEMDS
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
+    fail $SINGLEMDS
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
 
     $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
 }
 run_test 52 "time out lock replay (3764)"
 
 test_53() {
-    replay_barrier_nodf mds
+    replay_barrier_nodf $SINGLEMDS
     f1=$DIR/${tfile}-1
     cat <<EOF > $f1
 #!/bin/sh
 true
 EOF
     chmod +x $f1
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
     $f1 || return 1
-    do_facet mds "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
 
-    fail mds
+    fail $SINGLEMDS
     rm -f $f1
 }
 run_test 53 "|X| open request and close reply while two MDC requests in flight"
 
 test_54() {
-    replay_barrier mds
+    replay_barrier $SINGLEMDS
     createmany -o $DIR/$tfile 20
     unlinkmany $DIR/$tfile 20
-    fail mds
+    fail $SINGLEMDS
 }
 run_test 54 "|X| open request and close reply while two MDC requests in flight"
 
@@ -1264,28 +1272,46 @@ run_test 54 "|X| open request and close reply while two MDC requests in flight"
 test_55() {
     sysctl -w portals.debug=-1 portals.debug_mb=25
     ln -s foo $DIR/$tfile
-    replay_barrier mds
+    replay_barrier $SINGLEMDS
     #drop_reply "cat $DIR/$tfile"
-    fail mds
+    fail $SINGLEMDS
     sleep 10
-    lctl dk /r/tmp/debug
 }
 run_test 55 "don't replay a symlink open request (3440)"
 
 #b3761 ASSERTION(hash != 0) failed
 test_56() {
 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
-    do_facet mds "sysctl -w lustre.fail_loc=0x8000012b"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
     touch $DIR/$tfile
     pid=$!
     # give a chance for touch to run
     sleep 5
-    do_facet mds "sysctl -w lustre.fail_loc=0x0"
+    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
     wait $pid || return 1
     rm $DIR/$tfile
     return 0
 }
-run_test 56 "let MDS_CHECK_RESENT return the original return code instead of 0
+run_test 56 "let MDS_CHECK_RESENT return the original return code instead of 0"
+
+#b7312 LASSERT(!IS_ERR(parent)) in reconstruct_open()
+test_57() {
+    mkdir $DIR/$tdir || return 1
+    touch $DIR/$tdir/$tfile || return 2
+    multiop $DIR/$tdir/$tfile o_ &
+    MULTIPID=$!
+    sleep 1
+    rm -f $DIR/$tdir/$tfile || return 3
+    rm -rf $DIR/$tdir || return 4
+    # drop first reint reply
+    sysctl -w lustre.fail_loc=0x0000030c
+    facet_failover $SINGLEMDS
+    df $MOUNT || return 1
+    kill -USR1 $MULTIPID || return 5
+    wait $MULTIPID || return 6
+    sysctl -w lustre.fail_loc=0
+}
+run_test 57 "open orphan in reconstruct_open()"
 
 equals_msg test complete, cleaning up
 $CLEANUP