+#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"
+