X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-single.sh;h=1e4e91fa9c57e8eafcd0d29c2a9c1c70e20b289e;hp=78b2c343c2ac287cda8289c0a27ff4dad5ca2a23;hb=8f6d85eebc312b64d8e8a35b0be3ae137a50a45c;hpb=f151a8e900c8fc3aa95377ae9444417975207ce1

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 78b2c34..1e4e91f 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -7,43 +7,54 @@ set -e
 # This test needs to be run on the client
 #
 SAVE_PWD=$PWD
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 SETUP=${SETUP:-}
 CLEANUP=${CLEANUP:-}
 . $LUSTRE/tests/test-framework.sh
 init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
 CHECK_GRANT=${CHECK_GRANT:-"yes"}
 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
 
+require_dsh_mds || exit 0
 
 # Skip these tests
-# bug number: 2766 4176
-ALWAYS_EXCEPT="0b  39   $REPLAY_SINGLE_EXCEPT"
+# bug number:  17466 18857
+ALWAYS_EXCEPT="61d   33a 33b     $REPLAY_SINGLE_EXCEPT"
 
-#                                                       63 min  7 min  AT AT AT AT"
-[ "SLOW" = "no" ] && EXCEPT="$EXCEPT 1 2 3 4 6 6b 12 16 44      44b    65 66 67 68"
+if [ "$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 $CONFIG_EXCEPTIONS"
+fi
+
+#                                                  63 min  7 min  AT AT AT AT"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a      44b    65 66 67 68"
 
 build_test_filter
 
-cleanup_and_setup_lustre
+check_and_setup_lustre
 
 mkdir -p $DIR
 
-rm -rf $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
+assert_DIR
 rm -rf $DIR/[df][0-9]*
 
-test_0() {
+test_0a() {	# was test_0
     sleep 10
     mkdir $DIR/$tfile
     replay_barrier $SINGLEMDS
     fail $SINGLEMDS
     rmdir $DIR/$tfile
 }
-run_test 0 "empty replay"
+run_test 0a "empty replay"
 
 test_0b() {
-    # this test attempts to trigger a race in the precreation code, 
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    # this test attempts to trigger a race in the precreation code,
     # and must run before any other objects are created on the filesystem
     fail ost1
     createmany -o $DIR/$tfile 20 || return 1
@@ -55,15 +66,13 @@ seq_set_width()
 {
     local mds=$1
     local width=$2
-    local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width` 
-    echo $width > $file
+    lctl set_param -n seq.cli-srv-$mds-mdc-*.width=$width
 }
 
 seq_get_width()
 {
     local mds=$1
-    local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width` 
-    cat $file
+    lctl get_param -n seq.cli-srv-$mds-mdc-*.width
 }
 
 # This test should pass for single-mds and multi-mds configs.
@@ -74,15 +83,15 @@ seq_get_width()
 # (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 
+# 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 
+# (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 
+# (2) as fld_create does not happen on MDS1, it does not make any
 # problem.
 test_0c() {
     local label=`mdsdevlabel 1`
@@ -90,35 +99,37 @@ test_0c() {
 
     replay_barrier $SINGLEMDS
     local sw=`seq_get_width $label`
-    
-    # make seq manager switch to next sequence each 
+
+    # 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 
+
+    # 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 
+
+    # 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 
+
+    # 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
 }
+start_full_debug_logging
 run_test 0c "fld create"
+stop_full_debug_logging
 
 test_1() {
     replay_barrier $SINGLEMDS
@@ -149,21 +160,22 @@ test_2b() {
 run_test 2b "touch"
 
 test_3a() {
+    local file=$DIR/$tfile
     replay_barrier $SINGLEMDS
-    mcreate $DIR/$tfile
-    o_directory $DIR/$tfile
+    mcreate $file
+    openfile -f O_DIRECTORY $file
     fail $SINGLEMDS
-    $CHECKSTAT -t file $DIR/$tfile || return 2
-    rm $DIR/$tfile
+    $CHECKSTAT -t file $file || return 2
+    rm $file
 }
 run_test 3a "replay failed open(O_DIRECTORY)"
 
 test_3b() {
     replay_barrier $SINGLEMDS
 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000114"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
     touch $DIR/$tfile
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
     fail $SINGLEMDS
     $CHECKSTAT -t file $DIR/$tfile && return 2
     return 0
@@ -173,9 +185,9 @@ run_test 3b "replay failed open -ENOMEM"
 test_3c() {
     replay_barrier $SINGLEMDS
 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
     touch $DIR/$tfile
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
     fail $SINGLEMDS
 
     $CHECKSTAT -t file $DIR/$tfile && return 2
@@ -183,17 +195,17 @@ test_3c() {
 }
 run_test 3c "replay failed open -ENOMEM"
 
-test_4() {
+test_4a() {	# was test_4
     replay_barrier $SINGLEMDS
     for i in `seq 10`; do
         echo "tag-$i" > $DIR/$tfile-$i
-    done 
+    done
     fail $SINGLEMDS
     for i in `seq 10`; do
       grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
-    done 
+    done
 }
-run_test 4 "|x| 10 open(O_CREAT)s"
+run_test 4a "|x| 10 open(O_CREAT)s"
 
 test_4b() {
     replay_barrier $SINGLEMDS
@@ -203,17 +215,17 @@ test_4b() {
 }
 run_test 4b "|x| rm 10 files"
 
-# The idea is to get past the first block of precreated files on both 
+# The idea is to get past the first block of precreated files on both
 # osts, and then replay.
 test_5() {
     replay_barrier $SINGLEMDS
     for i in `seq 220`; do
         echo "tag-$i" > $DIR/$tfile-$i
-    done 
+    done
     fail $SINGLEMDS
     for i in `seq 220`; do
-      grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
-    done 
+      grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
+    done
     rm -rf $DIR/$tfile-*
     sleep 3
     # waiting for commitment of removal
@@ -221,7 +233,8 @@ test_5() {
 run_test 5 "|x| 220 open(O_CREAT)"
 
 
-test_6() {
+test_6a() {	# was test_6
+    mkdir -p $DIR/$tdir
     replay_barrier $SINGLEMDS
     mcreate $DIR/$tdir/$tfile
     fail $SINGLEMDS
@@ -230,17 +243,19 @@ test_6() {
     sleep 2
     # waiting for log process thread
 }
-run_test 6 "mkdir + contained create"
+run_test 6a "mkdir + contained create"
 
 test_6b() {
+    mkdir -p $DIR/$tdir
     replay_barrier $SINGLEMDS
     rm -rf $DIR/$tdir
     fail $SINGLEMDS
-    $CHECKSTAT -t dir $DIR/$tdir && return 1 || true 
+    $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
 }
 run_test 6b "|X| rmdir"
 
 test_7() {
+    mkdir -p $DIR/$tdir
     replay_barrier $SINGLEMDS
     mcreate $DIR/$tdir/$tfile
     fail $SINGLEMDS
@@ -254,9 +269,8 @@ test_8() {
     # make sure no side-effect from previous test.
     rm -f $DIR/$tfile
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile mo_c &
+    multiop_bg_pause $DIR/$tfile mo_c || return 4
     MULTIPID=$!
-    sleep 1
     fail $SINGLEMDS
     ls $DIR/$tfile
     $CHECKSTAT -t file $DIR/$tfile || return 1
@@ -304,7 +318,7 @@ test_11() {
     mv $DIR/$tfile $DIR/$tfile-2
     replay_barrier $SINGLEMDS
     echo "new" > $DIR/$tfile
-    grep new $DIR/$tfile 
+    grep new $DIR/$tfile
     grep old $DIR/$tfile-2
     fail $SINGLEMDS
     grep new $DIR/$tfile || return 1
@@ -313,11 +327,9 @@ test_11() {
 run_test 11 "create open write rename |X| create-old-name read"
 
 test_12() {
-    mcreate $DIR/$tfile 
-    multiop $DIR/$tfile o_tSc &
+    mcreate $DIR/$tfile
+    multiop_bg_pause $DIR/$tfile o_tSc || return 3
     pid=$!
-    # give multiop a chance to open
-    sleep 1
     rm -f $DIR/$tfile
     replay_barrier $SINGLEMDS
     kill -USR1 $pid
@@ -331,13 +343,11 @@ run_test 12 "open, unlink |X| close"
 
 
 # 1777 - replay open after committed chmod that would make
-#        a regular open a failure    
+#        a regular open a failure
 test_13() {
-    mcreate $DIR/$tfile 
-    multiop $DIR/$tfile O_wc &
+    mcreate $DIR/$tfile
+    multiop_bg_pause $DIR/$tfile O_wc || return 3
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     chmod 0 $DIR/$tfile
     $CHECKSTAT -p 0 $DIR/$tfile
     replay_barrier $SINGLEMDS
@@ -346,15 +356,14 @@ test_13() {
     wait $pid || return 1
 
     $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
+    rm $DIR/$tfile || return 4
     return 0
 }
 run_test 13 "open chmod 0 |x| write close"
 
 test_14() {
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 4
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
     replay_barrier $SINGLEMDS
     kill -USR1 $pid || return 1
@@ -367,10 +376,8 @@ test_14() {
 run_test 14 "open(O_CREAT), unlink |X| close"
 
 test_15() {
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 5
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
     replay_barrier $SINGLEMDS
     touch $DIR/g11 || return 1
@@ -399,10 +406,8 @@ run_test 16 "|X| open(O_CREAT), unlink, touch new,  unlink new"
 
 test_17() {
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile O_c &
+    multiop_bg_pause $DIR/$tfile O_c || return 4
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     fail $SINGLEMDS
     kill -USR1 $pid || return 1
     wait $pid || return 2
@@ -413,10 +418,8 @@ run_test 17 "|X| open(O_CREAT), |replay| close"
 
 test_18() {
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 8
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
     touch $DIR/$tfile-2 || return 1
     echo "pid: $pid will close"
@@ -446,12 +449,10 @@ test_19() {
 }
 run_test 19 "|X| mcreate, open, write, rename "
 
-test_20() {
+test_20a() {	# was test_20
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
 
     fail $SINGLEMDS
@@ -460,7 +461,7 @@ test_20() {
     [ -e $DIR/$tfile ] && return 2
     return 0
 }
-run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
+run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
 
 test_20b() { # bug 10480
     BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
@@ -474,15 +475,11 @@ test_20b() { # bug 10480
     lfs getstripe $DIR/$tfile || return 1
     rm -f $DIR/$tfile || return 2       # make it an orphan
     mds_evict_client
-    df -P $DIR || df -P $DIR || true    # reconnect
+    client_up || client_up || true    # reconnect
 
     fail $SINGLEMDS                            # start orphan recovery
-    df -P $DIR || df -P $DIR || true    # reconnect
-    wait_mds_recovery_done || error "MDS recovery not done"
-
-    # FIXME just because recovery is done doesn't mean we've finished 
-    # orphan cleanup.  Fake it with a sleep for now...
-    sleep 10
+    wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
+    wait_mds_ost_sync || return 3
     AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
     log "before $BEFOREUSED, after $AFTERUSED"
     [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
@@ -492,19 +489,17 @@ test_20b() { # bug 10480
 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
 
 test_20c() { # bug 10480
-    multiop $DIR/$tfile Ow_c &
+    multiop_bg_pause $DIR/$tfile Ow_c || return 1
     pid=$!
-    # give multiop a chance to open
-    sleep 1
 
     ls -la $DIR/$tfile
 
     mds_evict_client
-
-    df -P $DIR || df -P $DIR || true    # reconnect
+    client_up || client_up || true    # reconnect
 
     kill -USR1 $pid
-    test -s $DIR/$tfile || error "File was truncated"
+    wait $pid || return 1
+    [ -s $DIR/$tfile ] || error "File was truncated"
 
     return 0
 }
@@ -512,10 +507,8 @@ run_test 20c "check that client eviction does not affect file content"
 
 test_21() {
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 5
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
     touch $DIR/g11 || return 1
 
@@ -529,10 +522,8 @@ test_21() {
 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
 
 test_22() {
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
 
     replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile
@@ -546,10 +537,8 @@ test_22() {
 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
 
 test_23() {
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 5
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
 
     replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile
@@ -565,10 +554,8 @@ test_23() {
 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
 
 test_24() {
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
 
     replay_barrier $SINGLEMDS
     fail $SINGLEMDS
@@ -581,10 +568,8 @@ test_24() {
 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
 
 test_25() {
-    multiop $DIR/$tfile O_tSc &
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
 
     replay_barrier $SINGLEMDS
@@ -598,12 +583,10 @@ run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
 
 test_26() {
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile-1 O_tSc &
+    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
     pid1=$!
-    multiop $DIR/$tfile-2 O_tSc &
+    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
     kill -USR1 $pid2
@@ -620,12 +603,10 @@ run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test m
 
 test_27() {
     replay_barrier $SINGLEMDS
-    multiop $DIR/$tfile-1 O_tSc &
+    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
     pid1=$!
-    multiop $DIR/$tfile-2 O_tSc &
+    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
 
@@ -641,12 +622,10 @@ test_27() {
 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
 
 test_28() {
-    multiop $DIR/$tfile-1 O_tSc &
+    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
     pid1=$!
-    multiop $DIR/$tfile-2 O_tSc &
+    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
@@ -663,12 +642,10 @@ test_28() {
 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
 
 test_29() {
-    multiop $DIR/$tfile-1 O_tSc &
+    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
     pid1=$!
-    multiop $DIR/$tfile-2 O_tSc &
+    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     replay_barrier $SINGLEMDS
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
@@ -685,12 +662,10 @@ test_29() {
 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
 
 test_30() {
-    multiop $DIR/$tfile-1 O_tSc &
+    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
     pid1=$!
-    multiop $DIR/$tfile-2 O_tSc &
+    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile-1
     rm -f $DIR/$tfile-2
 
@@ -707,12 +682,10 @@ test_30() {
 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
 
 test_31() {
-    multiop $DIR/$tfile-1 O_tSc &
+    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
     pid1=$!
-    multiop $DIR/$tfile-2 O_tSc &
+    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
     pid2=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile-1
 
     replay_barrier $SINGLEMDS
@@ -731,35 +704,34 @@ run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_
 # tests for bug 2104; completion without crashing is success.  The close is
 # stale, but we always return 0 for close, so the app never sees it.
 test_32() {
-    multiop $DIR/$tfile O_c &
+    multiop_bg_pause $DIR/$tfile O_c || return 2
     pid1=$!
-    multiop $DIR/$tfile O_c &
+    multiop_bg_pause $DIR/$tfile O_c || return 3
     pid2=$!
-    # give multiop a chance to open
-    sleep 1
     mds_evict_client
-    df $MOUNT || sleep 1 && df $MOUNT || return 1
+    client_up || client_up || return 1
     kill -USR1 $pid1
     kill -USR1 $pid2
-    sleep 1
+    wait $pid1 || return 4
+    wait $pid2 || return 5
     return 0
 }
 run_test 32 "close() notices client eviction; close() after client eviction"
 
 # Abort recovery before client complete
-test_33() {
+test_33a() {	# was test_33
     replay_barrier $SINGLEMDS
     createmany -o $DIR/$tfile-%d 100
     fail_abort $SINGLEMDS
     # this file should be gone, because the replay was aborted
-    $CHECKSTAT -t file $DIR/$tfile-* && return 3 
+    $CHECKSTAT -t file $DIR/$tfile-* && return 3
     unlinkmany $DIR/$tfile-%d 0 100
     return 0
 }
-run_test 33 "abort recovery before client does replay"
+run_test 33a "abort recovery before client does replay"
 
-# Stale FID sequence 
-test_33a() {
+# Stale FID sequence bug 15962
+test_33b() {	# was test_33a
     replay_barrier $SINGLEMDS
     createmany -o $DIR/$tfile-%d 10
     fail_abort $SINGLEMDS
@@ -769,30 +741,29 @@ test_33a() {
     unlinkmany $DIR/$tfile-%d 0 10
     return 0
 }
-run_test 33a "fid shouldn't be reused after abort recovery"
+run_test 33b "fid shouldn't be reused after abort recovery"
 
 test_34() {
-    multiop $DIR/$tfile O_c &
+    multiop_bg_pause $DIR/$tfile O_c || return 2
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rm -f $DIR/$tfile
 
     replay_barrier $SINGLEMDS
     fail_abort $SINGLEMDS
     kill -USR1 $pid
+    wait $pid || return 3
     [ -e $DIR/$tfile ] && return 1
     sync
     return 0
 }
 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
 
-# bug 2278 - generate one orphan on OST, then destroy it during recovery from llog 
+# bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
 test_35() {
     touch $DIR/$tfile
 
 #define OBD_FAIL_MDS_REINT_NET_REP       0x119
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
     rm -f $DIR/$tfile &
     sleep 1
     sync
@@ -811,7 +782,7 @@ test_36() {
     checkstat $DIR/$tfile
     facet_failover $SINGLEMDS
     cancel_lru_locks mdc
-    if dmesg | grep "unknown lock cookie"; then 
+    if dmesg | grep "unknown lock cookie"; then
 	echo "cancel after replay failed"
 	return 1
     fi
@@ -822,10 +793,8 @@ run_test 36 "don't resend cancel"
 # directory orphans can't be unlinked from PENDING directory
 test_37() {
     rmdir $DIR/$tfile 2>/dev/null
-    multiop $DIR/$tfile dD_c &
+    multiop_bg_pause $DIR/$tfile dD_c || return 2
     pid=$!
-    # give multiop a chance to open
-    sleep 1 
     rmdir $DIR/$tfile
 
     replay_barrier $SINGLEMDS
@@ -834,10 +803,13 @@ test_37() {
     fail_abort $SINGLEMDS
     kill -USR1 $pid
     dmesg | grep  "mds_unlink_orphan.*error .* unlinking orphan" && return 1
+    wait $pid || return 3
     sync
     return 0
 }
+start_full_debug_logging
 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
+stop_full_debug_logging
 
 test_38() {
     createmany -o $DIR/$tfile-%d 800
@@ -862,12 +834,12 @@ test_39() { # bug 4176
 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
 
 count_ost_writes() {
-    awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
+    lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
 }
 
 #b=2477,2532
 test_40(){
-    $LCTL mark multiop $MOUNT/$tfile OS_c 
+    $LCTL mark multiop $MOUNT/$tfile OS_c
     multiop $MOUNT/$tfile OS_c  &
     PID=$!
     writeme -s $MOUNT/${tfile}-2 &
@@ -875,13 +847,13 @@ test_40(){
     sleep 1
     facet_failover $SINGLEMDS
 #define OBD_FAIL_MDS_CONNECT_NET         0x117
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
     kill -USR1 $PID
     stat1=`count_ost_writes`
     sleep $TIMEOUT
     stat2=`count_ost_writes`
     echo "$stat1, $stat2"
-    if [ $stat1 -lt $stat2 ]; then 
+    if [ $stat1 -lt $stat2 ]; then
        echo "writes continuing during recovery"
        RC=0
     else
@@ -890,7 +862,7 @@ test_40(){
     fi
     echo "waiting for writeme $WRITE_PID"
     kill $WRITE_PID
-    wait $WRITE_PID 
+    wait $WRITE_PID
 
     echo "waiting for multiop $PID"
     wait $PID || return 2
@@ -910,20 +882,20 @@ run_test 40 "cause recovery in ptlrpc, ensure IO continues"
 # assert on trying to unlock the unlocked page.
 test_41() {
     [ $OSTCOUNT -lt 2 ] && \
-	skip "skipping test 41: we don't have a second OST to test with" && \
+	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 $((128 * 1024)) 0 0 
+    lfs setstripe $f -s $((128 * 1024)) -i 0
     do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
     cancel_lru_locks osc
     # fail ost2 and read from ost1
-    local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
-    [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
-    do_facet mds $LCTL --device $osc2dev deactivate || return 1
+    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
+    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 mds $LCTL --device $osc2dev activate || return 2
+    do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
     return 0
 }
 run_test 41 "read from a valid osc while other oscs are invalid"
@@ -935,9 +907,9 @@ test_42() {
     replay_barrier ost1
     unlinkmany $DIR/$tfile-%d 0 400
     debugsave
-    sysctl -w lnet.debug=-1
+    lctl set_param debug=-1
     facet_failover ost1
-    
+
     # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
     #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
     #[ $blocks_after -lt $blocks ] || return 1
@@ -951,57 +923,79 @@ run_test 42 "recovery after ost failure"
 
 # timeout in MDS/OST recovery RPC will LBUG MDS
 test_43() { # bug 2530
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
     replay_barrier $SINGLEMDS
 
     # OBD_FAIL_OST_CREATE_NET 0x204
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
+    do_facet ost1 "lctl set_param fail_loc=0x80000204"
     fail $SINGLEMDS
     sleep 10
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
+    do_facet ost1 "lctl set_param fail_loc=0"
 
     return 0
 }
 run_test 43 "mds osc import failure during recovery; don't LBUG"
 
-test_44() {
-    mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
-    [ "$mdcdev" ] || exit 2
+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
-	#define OBD_FAIL_TGT_CONN_RACE     0x701
-	do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
-	$LCTL --device $mdcdev recover
-	df $MOUNT
+        echo "$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 "sysctl -w lustre.fail_loc=0"
+    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 44 "race in target handle connect"
+run_test 44a "race in target handle connect"
 
 test_44b() {
-    mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
-    [ "$mdcdev" ] || exit 2
+    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
-	#define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
-	do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
-	$LCTL --device $mdcdev recover
-	df $MOUNT
+        echo "$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"
+        # lctl below may fail, it is valid case
+        $LCTL --device $mdcdev recover
+        df $MOUNT
     done
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
     return 0
 }
 run_test 44b "race in target handle connect"
 
 # Handle failed close
 test_45() {
-    mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
-    [ "$mdcdev" ] || exit 2
-    $LCTL --device $mdcdev recover
+    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; }
+
+    $LCTL --device $mdcdev recover || return 6 
 
-    multiop $DIR/$tfile O_c &
+    multiop_bg_pause $DIR/$tfile O_c || return 1
     pid=$!
-    sleep 1
 
-    # This will cause the CLOSE to fail before even 
+    # This will cause the CLOSE to fail before even
     # allocating a reply buffer
     $LCTL --device $mdcdev deactivate || return 4
 
@@ -1029,46 +1023,49 @@ test_46() {
 run_test 46 "Don't leak file handle after open resend (3325)"
 
 test_47() { # bug 2824
-    # create some files to make sure precreate has been done on all 
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    # create some files to make sure precreate has been done on all
     # OSTs. (just in case this test is run independently)
     createmany -o $DIR/$tfile 20  || return 1
 
     # OBD_FAIL_OST_CREATE_NET 0x204
     fail ost1
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
-    df $MOUNT || return 2
+    do_facet ost1 "lctl set_param fail_loc=0x80000204"
+    client_up || return 2
 
     # let the MDS discover the OST failure, attempt to recover, fail
-    # and recover again.  
+    # and recover again.
     sleep $((3 * TIMEOUT))
 
-    # Without 2824, this createmany would hang 
+    # Without 2824, this createmany would hang
     createmany -o $DIR/$tfile 20 || return 3
     unlinkmany $DIR/$tfile 20 || return 4
 
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
+    do_facet ost1 "lctl set_param fail_loc=0"
     return 0
 }
 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
 
 test_48() {
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+    [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
+
     replay_barrier $SINGLEMDS
     createmany -o $DIR/$tfile 20  || return 1
     # OBD_FAIL_OST_EROFS 0x216
-    fail $SINGLEMDS
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
-    df $MOUNT || return 2
+    facet_failover $SINGLEMDS
+    do_facet ost1 "lctl set_param fail_loc=0x80000216"
+    client_up || return 2
 
     createmany -o $DIR/$tfile 20 20 || return 2
     unlinkmany $DIR/$tfile 40 || return 3
-
-    do_facet ost1 "sysctl -w lustre.fail_loc=0"
     return 0
 }
 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
 
 test_50() {
-    local oscdev=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
+    local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
     [ "$oscdev" ] || return 1
     do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
     do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
@@ -1085,25 +1082,290 @@ test_52() {
     multiop $DIR/$tfile s || return 1
     replay_barrier $SINGLEMDS
 #define OBD_FAIL_LDLM_REPLY              0x30c
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
     fail $SINGLEMDS || return 2
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
 
     $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
 }
 run_test 52 "time out lock replay (3764)"
 
-#b_cray 53 "|X| open request and close reply while two MDC requests in flight"
+# bug 3462 - simultaneous MDC requests
+test_53a() {
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+        # give multiop a change to open
+        sleep 1
+
+        #define OBD_FAIL_MDS_CLOSE_NET 0x115
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+
+        mcreate $DIR/${tdir}-2/f || return 1
+
+        # close should still be here
+        [ -d /proc/$close_pid ] || return 2
+
+        replay_barrier_nodf $SINGLEMDS
+        fail $SINGLEMDS
+        wait $close_pid || return 3
+
+        $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
+        $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53a "|X| close request while two MDC requests in flight"
+
+test_53b() {
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop_bg_pause $DIR/${tdir}-1/f O_c || return 6
+        close_pid=$!
+
+        #define OBD_FAIL_MDS_REINT_NET 0x107
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
+        mcreate $DIR/${tdir}-2/f &
+        open_pid=$!
+        sleep 1
+
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+        wait $close_pid || return 1
+        # open should still be here
+        [ -d /proc/$open_pid ] || return 2
+
+        replay_barrier_nodf $SINGLEMDS
+        fail $SINGLEMDS
+        wait $open_pid || return 3
+
+        $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
+        $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53b "|X| open request while two MDC requests in flight"
+
+test_53c() {
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+
+        #define OBD_FAIL_MDS_REINT_NET 0x107
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
+        mcreate $DIR/${tdir}-2/f &
+        open_pid=$!
+        sleep 1
+
+        #define OBD_FAIL_MDS_CLOSE_NET 0x115
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
+        replay_barrier_nodf $SINGLEMDS
+        fail_nodf $SINGLEMDS
+        wait $open_pid || return 1
+        sleep 2
+        # close should be gone
+        [ -d /proc/$close_pid ] && return 2
+        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
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53c "|X| open request and close request while two MDC requests in flight"
+
+test_53d() {
+        cancel_lru_locks mdc    # cleanup locks from former test cases
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+        # give multiop a chance to open
+        sleep 1
+
+        #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+        mcreate $DIR/${tdir}-2/f || return 1
+
+        # close should still be here
+        [ -d /proc/$close_pid ] || return 2
+        fail $SINGLEMDS
+        wait $close_pid || return 3
+
+        $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
+        $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53d "|X| close reply while two MDC requests in flight"
+
+test_53e() {
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+
+        #define OBD_FAIL_MDS_REINT_NET_REP 0x119
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
+        mcreate $DIR/${tdir}-2/f &
+        open_pid=$!
+        sleep 1
+
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+        wait $close_pid || return 1
+        # open should still be here
+        [ -d /proc/$open_pid ] || return 2
+
+        replay_barrier_nodf $SINGLEMDS
+        fail $SINGLEMDS
+        wait $open_pid || return 3
+
+        $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
+        $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53e "|X| open reply while two MDC requests in flight"
+
+test_53f() {
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+
+        #define OBD_FAIL_MDS_REINT_NET_REP 0x119
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
+        mcreate $DIR/${tdir}-2/f &
+        open_pid=$!
+        sleep 1
+
+        #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
+        replay_barrier_nodf $SINGLEMDS
+        fail_nodf $SINGLEMDS
+        wait $open_pid || return 1
+        sleep 2
+        # close should be gone
+        [ -d /proc/$close_pid ] && return 2
+        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
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53f "|X| open reply and close reply while two MDC requests in flight"
+
+test_53g() {
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+
+        #define OBD_FAIL_MDS_REINT_NET_REP 0x119
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
+        mcreate $DIR/${tdir}-2/f &
+        open_pid=$!
+        sleep 1
+
+        #define OBD_FAIL_MDS_CLOSE_NET 0x115
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
+        replay_barrier_nodf $SINGLEMDS
+        fail_nodf $SINGLEMDS
+        wait $open_pid || return 1
+        sleep 2
+        # close should be gone
+        [ -d /proc/$close_pid ] && return 2
+
+        $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
+        $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53g "|X| drop open reply and close request while close and open are both in flight"
+
+test_53h() {
+        rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
+        mkdir -p $DIR/${tdir}-1
+        mkdir -p $DIR/${tdir}-2
+        multiop $DIR/${tdir}-1/f O_c &
+        close_pid=$!
+
+        #define OBD_FAIL_MDS_REINT_NET 0x107
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
+        mcreate $DIR/${tdir}-2/f &
+        open_pid=$!
+        sleep 1
+
+        #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
+        do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
+        kill -USR1 $close_pid
+        cancel_lru_locks mdc    # force the close
+        sleep 1
+
+        #bz20647: make sure all pids are exists before failover
+        [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+        [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
+        replay_barrier_nodf $SINGLEMDS
+        fail_nodf $SINGLEMDS
+        wait $open_pid || return 1
+        sleep 2
+        # close should be gone
+        [ -d /proc/$close_pid ] && return 2
+        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
+        rm -rf $DIR/${tdir}-*
+}
+run_test 53h "|X| open request and close reply while two MDC requests in flight"
+
 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
 
 #b3761 ASSERTION(hash != 0) failed
 test_55() {
 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
     touch $DIR/$tfile &
     # give touch a chance to run
     sleep 5
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
     rm $DIR/$tfile
     return 0
 }
@@ -1122,43 +1384,80 @@ run_test 56 "don't replay a symlink open request (3440)"
 #recovery one mds-ost setattr from llog
 test_57() {
 #define OBD_FAIL_MDS_OST_SETATTR       0x12c
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
     touch $DIR/$tfile
     replay_barrier $SINGLEMDS
     fail $SINGLEMDS
     sleep 1
     $CHECKSTAT -t file $DIR/$tfile || return 1
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
     rm $DIR/$tfile
 }
 run_test 57 "test recovery from llog for setattr op"
 
 #recovery many mds-ost setattr from llog
-test_58() {
+test_58a() {
+    mkdir -p $DIR/$tdir
 #define OBD_FAIL_MDS_OST_SETATTR       0x12c
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
     createmany -o $DIR/$tdir/$tfile-%d 2500
     replay_barrier $SINGLEMDS
     fail $SINGLEMDS
     sleep 2
     $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
-    do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
     unlinkmany $DIR/$tdir/$tfile-%d 2500
     rmdir $DIR/$tdir
 }
-run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
+run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
+
+test_58b() {
+    mount_client $MOUNT2
+    mkdir -p $DIR/$tdir
+    touch $DIR/$tdir/$tfile
+    replay_barrier $SINGLEMDS
+    setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
+    fail $SINGLEMDS
+    VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
+    [ x$VAL = x"bar" ] || return 1
+    rm -f $DIR/$tdir/$tfile
+    rmdir $DIR/$tdir
+    zconf_umount `hostname` $MOUNT2
+}
+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
+}
+run_test 58c "resend/reconstruct setxattr op"
 
 # log_commit_thread vs filter_destroy race used to lead to import use after free
 # bug 11658
 test_59() {
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    mkdir -p $DIR/$tdir
     createmany -o $DIR/$tdir/$tfile-%d 200
     sync
     unlinkmany $DIR/$tdir/$tfile-%d 200
 #define OBD_FAIL_PTLRPC_DELAY_RECOV       0x507
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
+    do_facet ost1 "lctl set_param fail_loc=0x507"
     fail ost1
     fail $SINGLEMDS
-    do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
+    do_facet ost1 "lctl set_param fail_loc=0x0"
     sleep 20
     rmdir $DIR/$tdir
 }
@@ -1167,16 +1466,583 @@ run_test 59 "test log_commit_thread vs filter_destroy race"
 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
 # bug 12086: should no oops and No ctxt error for this test
 test_60() {
+    mkdir -p $DIR/$tdir
     createmany -o $DIR/$tdir/$tfile-%d 200
     replay_barrier $SINGLEMDS
     unlinkmany $DIR/$tdir/$tfile-%d 0 100
     fail $SINGLEMDS
     unlinkmany $DIR/$tdir/$tfile-%d 100 100
     local no_ctxt=`dmesg | grep "No ctxt"`
-    [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery" 
+    [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
 }
 run_test 60 "test llog post recovery init vs llog unlink"
 
+#test race  llog recovery thread vs llog cleanup
+test_61a() {	# was test_61
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    mkdir -p $DIR/$tdir
+    createmany -o $DIR/$tdir/$tfile-%d 800
+    replay_barrier ost1
+#   OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
+    unlinkmany $DIR/$tdir/$tfile-%d 800
+    set_nodes_failloc "$(osts_nodes)" 0x80000221
+    facet_failover ost1
+    sleep 10
+    fail ost1
+    sleep 30
+    set_nodes_failloc "$(osts_nodes)" 0x0
+
+    $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
+    rmdir $DIR/$tdir
+}
+run_test 61a "test race llog recovery vs llog cleanup"
+
+#test race  mds llog sync vs llog cleanup
+test_61b() {
+#   OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
+    facet_failover $SINGLEMDS
+    sleep 10
+    fail $SINGLEMDS
+    do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
+}
+run_test 61b "test race mds llog sync vs llog cleanup"
+
+#test race  cancel cookie cb vs llog cleanup
+test_61c() {
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+#   OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
+    touch $DIR/$tfile
+    set_nodes_failloc "$(osts_nodes)" 0x80000222
+    rm $DIR/$tfile
+    sleep 10
+    fail ost1
+    set_nodes_failloc "$(osts_nodes)" 0x0
+}
+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"
+}
+run_test 61d "error in llog_setup should cleanup the llog context correctly"
+
+test_62() { # Bug 15756 - don't mis-drop resent replay
+    mkdir -p $DIR/$tdir
+    replay_barrier $SINGLEMDS
+    createmany -o $DIR/$tdir/$tfile- 25
+#define OBD_FAIL_TGT_REPLAY_DROP         0x707
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
+    fail $SINGLEMDS
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+    unlinkmany $DIR/$tdir/$tfile- 25 || return 2
+    return 0
+}
+run_test 62 "don't mis-drop resent replay"
+
+#Adaptive Timeouts (bug 3055)
+AT_MAX_SET=0
+
+at_cleanup () {
+    local var
+    local facet
+    local at_new
+
+    echo "Cleaning up AT ..."
+    if [ -n "$ATOLDBASE" ]; then
+        local at_history=$($LCTL get_param -n at_history)
+        do_facet mds "lctl set_param at_history=$at_history" || true
+        do_facet ost1 "lctl set_param at_history=$at_history" || true
+    fi
+
+    if [ $AT_MAX_SET -ne 0 ]; then
+        for facet in mds client ost; do
+            var=AT_MAX_SAVE_${facet}
+            echo restore AT on $facet to saved value ${!var}
+            at_max_set ${!var} $facet
+            at_new=$(at_max_get $facet)
+            echo Restored AT value on $facet $at_new
+            [ $at_new -eq ${!var} ] || \
+            error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
+        done
+    fi
+}
+
+at_start()
+{
+    local at_max_new=600
+
+    # Save at_max original values
+    local facet
+    if [ $AT_MAX_SET -eq 0 ]; then
+        # Suppose that all osts have the same at_max
+        for facet in mds client ost; do
+            eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
+        done
+    fi
+    local at_max
+    for facet in mds client ost; do
+        at_max=$(at_max_get $facet)
+        if [ $at_max -ne $at_max_new ]; then
+            echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
+            at_max_set $at_max_new $facet
+            AT_MAX_SET=1
+        fi
+    done
+
+    if [ -z "$ATOLDBASE" ]; then
+	ATOLDBASE=$(do_facet mds "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 ost1 "lctl set_param at_history=8" || true
+
+	# sleep for a while to cool down, should be > 8s and also allow
+	# at least one ping to be sent. simply use TIMEOUT to be safe.
+	sleep $TIMEOUT
+    fi
+}
+
+test_65a() #bug 3055
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    at_start || return 0
+    $LCTL dk > /dev/null
+    debugsave
+    sysctl -w lnet.debug="+other"
+    # Slow down a request to the current service time, this is critical
+    # because previous tests may have caused this value to increase.
+    REQ_DELAY=`lctl get_param -n mdc.${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))
+#define OBD_FAIL_PTLRPC_PAUSE_REQ        0x50a
+    do_facet mds sysctl -w lustre.fail_loc=0x8000050a
+    createmany -o $DIR/$tfile 10 > /dev/null
+    unlinkmany $DIR/$tfile 10 > /dev/null
+    # check for log message
+    $LCTL dk | grep "Early reply #" || error "No early reply"
+    debugrestore
+    # client should show REQ_DELAY estimates
+    lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
+    sleep 9
+    lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
+}
+run_test 65a "AT: verify early replies"
+
+test_65b() #bug 3055
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    at_start || return 0
+    # turn on D_ADAPTTO
+    debugsave
+    sysctl -w lnet.debug="other trace"
+    $LCTL dk > /dev/null
+    # Slow down a request to the current service time, this is critical
+    # because previous tests may have caused this value to increase.
+    REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
+               awk '/portal 6/ {print $5}'`
+    REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
+
+    do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
+#define OBD_FAIL_OST_BRW_PAUSE_PACK      0x224
+    do_facet ost1 sysctl -w lustre.fail_loc=0x224
+
+    rm -f $DIR/$tfile
+    lfs setstripe $DIR/$tfile --index=0 --count=1
+    # force some real bulk transfer
+    multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+
+    do_facet ost1 sysctl -w lustre.fail_loc=0
+    # check for log message
+    $LCTL dk | grep "Early reply #" || error "No early reply"
+    debugrestore
+    # client should show REQ_DELAY estimates
+    lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
+}
+run_test 65b "AT: verify early replies on packed reply / bulk"
+
+test_66a() #bug 3055
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    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"
+#define OBD_FAIL_PTLRPC_PAUSE_REQ        0x50a
+    do_facet mds "sysctl -w lustre.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"
+    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"
+    sleep 9
+    createmany -o $DIR/$tfile 20 > /dev/null
+    unlinkmany $DIR/$tfile 20 > /dev/null
+    lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
+    CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
+    WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
+    echo "Current MDT timeout $CUR, worst $WORST"
+    [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
+}
+run_test 66a "AT: verify MDT service time adjusts with no early replies"
+
+test_66b() #bug 3055
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    at_start || return 0
+    ORIG=$(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 $ORIG"
+}
+run_test 66b "AT: verify net latency adjusts"
+
+test_67a() #bug 3055
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    at_start || return 0
+    CONN1=$(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"
+#define OBD_FAIL_PTLRPC_PAUSE_REQ    0x50a
+    do_facet ost1 "sysctl -w lustre.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"
+    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"
+    [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
+    return 0
+}
+run_test 67a "AT: verify slow request processing doesn't induce reconnects"
+
+test_67b() #bug 3055
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    at_start || return 0
+    CONN1=$(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)
+
+    mkdir -p $DIR/$tdir/${OST}
+    lfs setstripe $DIR/$tdir/${OST} -o 0 -c 1 || 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"
+    createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
+
+    client_reconnect
+    do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
+    log "phase 2"
+    CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
+    ATTEMPTS=$(($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"
+    cp /etc/profile $DIR/$tfile || error "cp failed"
+    do_facet ost1 "sysctl -w lustre.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+=$2} END {print total}')
+    ATTEMPTS=$(($CONN3 - $CONN2))
+    echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
+    [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
+    return 0
+}
+run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
+
+test_68 () #bug 13813
+{
+    remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+    at_start || return 0
+    local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
+    [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
+    local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
+    [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
+    local ENQ_MIN=$(cat $ldlm_enqueue_min)
+    local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
+    echo $TIMEOUT >> $ldlm_enqueue_min
+    do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
+
+    rm -rf $DIR/$tdir
+    mkdir -p $DIR/$tdir
+    lfs setstripe $DIR/$tdir --index=0 --count=1
+#define OBD_FAIL_LDLM_PAUSE_CANCEL       0x312
+    sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
+    sysctl -w lustre.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
+    cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
+    sysctl -w lustre.fail_loc=0
+
+    echo $ENQ_MIN >> $ldlm_enqueue_min
+    do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
+    rm -rf $DIR/$tdir
+    return 0
+}
+run_test 68 "AT: verify slowing locks"
+
+at_cleanup
+# end of AT tests includes above lines
+
+
+# start multi-client tests
+test_70a () {
+	[ -z "$CLIENTS" ] && \
+		{ skip "Need two or more clients." && return; }
+	[ $CLIENTCOUNT -lt 2 ] && \
+		{ skip "Need two or more clients, have $CLIENTCOUNT" && return; }
+
+	echo "mount clients $CLIENTS ..."
+	zconf_mount_clients $CLIENTS $DIR
+
+	local clients=${CLIENTS//,/ }
+	echo "Write/read files on $DIR ; clients $CLIENTS ... "
+	for CLIENT in $clients; do
+		do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
+			of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
+				error "dd failed on $CLIENT"
+	done
+
+	local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
+	for C in ${CLIENTS//,/ }; do
+		do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
+			error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
+		prev_client=$C
+	done
+	
+	ls $DIR
+}
+run_test 70a "check multi client t-f"
+
+test_70b () {
+	local clients=${CLIENTS:-$HOSTNAME}
+
+	zconf_mount_clients $clients $DIR
+	
+	local duration=300
+	[ "$SLOW" = "no" ] && duration=60
+	local cmd="rundbench 1 -t $duration"
+	local PID=""
+	do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
+		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
+		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"
+		fail $SINGLEMDS
+		CURRENT_TS=$(date +%s)
+		ELAPSED=$((CURRENT_TS - START_TS))
+	done
+	wait $PID || error "rundbench load on $CLIENTS failed!"
+}
+run_test 70b "mds recovery; $CLIENTCOUNT clients"
+# end multi-client tests
+
+test_73a() {
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
+    pid=$!
+    rm -f $DIR/$tfile
+
+    replay_barrier $SINGLEMDS
+#define OBD_FAIL_LDLM_ENQUEUE       0x302
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
+    fail $SINGLEMDS
+    kill -USR1 $pid
+    wait $pid || return 1
+    [ -e $DIR/$tfile ] && return 2
+    return 0
+}
+run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
+
+test_73b() {
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
+    pid=$!
+    rm -f $DIR/$tfile
+
+    replay_barrier $SINGLEMDS
+#define OBD_FAIL_LDLM_REPLY       0x30c
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
+    fail $SINGLEMDS
+    kill -USR1 $pid
+    wait $pid || return 1
+    [ -e $DIR/$tfile ] && return 2
+    return 0
+}
+run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
+
+test_73c() {
+    multiop_bg_pause $DIR/$tfile O_tSc || return 3
+    pid=$!
+    rm -f $DIR/$tfile
+
+    replay_barrier $SINGLEMDS
+#define OBD_FAIL_TGT_LAST_REPLAY       0x710
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
+    fail $SINGLEMDS
+    kill -USR1 $pid
+    wait $pid || return 1
+    [ -e $DIR/$tfile ] && return 2
+    return 0
+}
+run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
+
+# bug 18554
+test_74() {
+    local clients=${CLIENTS:-$HOSTNAME}
+
+    stop ost1
+    zconf_umount_clients $clients $MOUNT
+    facet_failover $SINGLEMDS
+    zconf_mount_clients $clients $MOUNT
+    mount_facet ost1
+    touch $DIR/$tfile || return 1
+    rm $DIR/$tfile || return 2
+    clients_up || error "client evicted: $?"
+    return 0
+}
+run_test 74 "Ensure applications don't fail waiting for OST recovery"
+
+test_80a() {
+    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && 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
+}
+run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
+
+test_80b() {
+    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && 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
+}
+run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
+
+test_81a() {
+    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && 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
+}
+run_test 81a "CMD: unlink cross-node file (fail mds with name)"
+
+test_82a() {
+    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+
+    local dir=$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"
+}
+run_test 82a "CMD: mkdir cross-node dir (fail mds with inode)"
+
+test_82b() {
+    [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+
+    local dir=$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"
+}
+run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
+
+test_83a() {
+    mkdir -p $DIR/$tdir
+    createmany -o $DIR/$tdir/$tfile- 10 || return 1
+#define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD       0x140
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
+    unlinkmany $DIR/$tdir/$tfile- 10 || return 2
+}
+run_test 83a "fail log_add during unlink recovery"
+
+test_83b() {
+    mkdir -p $DIR/$tdir
+    createmany -o $DIR/$tdir/$tfile- 10 || return 1
+    replay_barrier $SINGLEMDS
+    unlinkmany $DIR/$tdir/$tfile- 10 || return 2
+#define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD       0x140
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
+    fail $SINGLEMDS
+}
+run_test 83b "fail log_add during unlink recovery"
+
+test_84a() {
+#define OBD_FAIL_MDS_OPEN_WAIT_CREATE  0x144
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
+    createmany -o $DIR/$tfile- 1 &
+    PID=$!
+    mds_evict_client
+    wait $PID
+    client_up || client_up || true    # reconnect
+}
+run_test 84a "stale open during export disconnect"
+
 equals_msg `basename $0`: test complete, cleaning up
 check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true
+[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true