set -e
+#
+# This test needs to be run on the client
+#
+
LUSTRE=${LUSTRE:-`dirname $0`/..}
-LTESTDIR=${LTESTDIR:-$LUSTRE/../ltest}
-PATH=$PATH:$LUSTRE/utils:$LUSTRE/tests
+. $LUSTRE/tests/test-framework.sh
+
+init_test_env $@
+
+. ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
+
+# Skip these tests
+ALWAYS_EXCEPT=""
+
+
+gen_config() {
+ rm -f $XMLCONFIG
+ add_mds mds --dev $MDSDEV --size $MDSSIZE
+ if [ ! -z "$mdsfailover_HOST" ]; then
+ add_mdsfailover mds --dev $MDSDEV --size $MDSSIZE
+ fi
+
+ add_lov lov1 mds --stripe_sz $STRIPE_BYTES\
+ --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
+ add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
+ add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
+ add_client client mds --lov lov1 --path $MOUNT
+}
+
+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 mds`
+ if [ $activemds != "mds" ]; then
+ fail mds
+ fi
+ zconf_umount `hostname` $MOUNT
+ stop mds ${FORCE} $MDSLCONFARGS
+ stop ost2 ${FORCE} --dump cleanup.log
+ stop ost ${FORCE} --dump cleanup.log
+}
+
+if [ "$ONLY" == "cleanup" ]; then
+ sysctl -w portals.debug=0 || true
+ cleanup
+ exit
+fi
+
+SETUP=${SETUP:-"setup"}
+CLEANUP=${CLEANUP:-"cleanup"}
+
+setup() {
+ gen_config
+
+ start ost --reformat $OSTLCONFARGS
+ start ost2 --reformat $OSTLCONFARGS
+ [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
+ start mds $MDSLCONFARGS --reformat
+ zconf_mount `hostname` $MOUNT
+}
+
+$SETUP
+
+if [ "$ONLY" == "setup" ]; then
+ exit 0
+fi
+
+mkdir -p $DIR
+
+test_0() {
+ replay_barrier mds
+ fail mds
+}
+run_test 0 "empty replay"
+
+test_1() {
+ replay_barrier mds
+ mcreate $DIR/$tfile
+ fail mds
+ $CHECKSTAT -t file $DIR/$tfile || return 1
+ rm $DIR/$tfile
+}
+run_test 1 "simple create"
+
+test_2a() {
+ replay_barrier mds
+ touch $DIR/$tfile
+ fail mds
+ $CHECKSTAT -t file $DIR/$tfile || return 1
+ rm $DIR/$tfile
+}
+run_test 2a "touch"
+
+test_2b() {
+ ./mcreate $DIR/$tfile
+ replay_barrier mds
+ touch $DIR/$tfile
+ fail mds
+ $CHECKSTAT -t file $DIR/$tfile || return 1
+ rm $DIR/$tfile
+}
+run_test 2b "touch"
+
+test_3a() {
+ replay_barrier mds
+ mcreate $DIR/$tfile
+ o_directory $DIR/$tfile
+ fail mds
+ $CHECKSTAT -t file $DIR/$tfile || return 2
+ rm $DIR/$tfile
+}
+run_test 3a "replay failed open(O_DIRECTORY)"
+
+test_3b() {
+ replay_barrier mds
+#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 mds
+ $CHECKSTAT -t file $DIR/$tfile && return 2
+ return 0
+}
+run_test 3b "replay failed open -ENOMEM"
+
+test_3c() {
+ replay_barrier mds
+#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 mds
+
+ $CHECKSTAT -t file $DIR/$tfile && return 2
+ return 0
+}
+run_test 3c "replay failed open -ENOMEM"
+
+test_4() {
+ replay_barrier mds
+ for i in `seq 10`; do
+ echo "tag-$i" > $DIR/$tfile-$i
+ done
+ fail mds
+ for i in `seq 10`; do
+ grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
+ done
+}
+run_test 4 "|x| 10 open(O_CREAT)s"
+
+test_4b() {
+ replay_barrier mds
+ rm -rf $DIR/$tfile-*
+ fail mds
+ $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
+}
+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 mds
+ for i in `seq 220`; do
+ echo "tag-$i" > $DIR/$tfile-$i
+ done
+ fail mds
+ for i in `seq 220`; do
+ grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
+ done
+ rm -rf $DIR/$tfile-*
+ sleep 3
+ # waiting for commitment of removal
+}
+run_test 5 "|x| 220 open(O_CREAT)"
+
+
+test_6() {
+ replay_barrier mds
+ mkdir $DIR/$tdir
+ mcreate $DIR/$tdir/$tfile
+ fail mds
+ $CHECKSTAT -t dir $DIR/$tdir || return 1
+ $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
+ sleep 2
+ # waiting for log process thread
+}
+run_test 6 "mkdir + contained create"
+
+test_6b() {
+ replay_barrier mds
+ rm -rf $DIR/$tdir
+ fail mds
+ $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
+}
+run_test 6b "|X| rmdir"
+
+test_7() {
+ mkdir $DIR/$tdir
+ replay_barrier mds
+ mcreate $DIR/$tdir/$tfile
+ fail mds
+ $CHECKSTAT -t dir $DIR/$tdir || return 1
+ $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
+ rm -fr $DIR/$tdir
+}
+run_test 7 "mkdir |X| contained create"
+
+test_8() {
+ replay_barrier mds
+ multiop $DIR/$tfile mo_c &
+ MULTIPID=$!
+ sleep 1
+ fail mds
+ ls $DIR/$tfile
+ $CHECKSTAT -t file $DIR/$tfile || return 1
+ kill -USR1 $MULTIPID || return 2
+ wait $MULTIPID || return 3
+ rm $DIR/$tfile
+}
+run_test 8 "creat open |X| close"
+
+test_9() {
+ replay_barrier mds
+ mcreate $DIR/$tfile
+ local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
+ fail mds
+ local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
+
+ echo " old_inum == $old_inum, new_inum == $new_inum"
+ if [ $old_inum -eq $new_inum ] ;
+ then
+ echo " old_inum and new_inum match"
+ else
+ echo "!!!! old_inum and new_inum NOT match"
+ return 1
+ fi
+ rm $DIR/$tfile
+}
+run_test 9 "|X| create (same inum/gen)"
+
+test_10() {
+ mcreate $DIR/$tfile
+ replay_barrier mds
+ mv $DIR/$tfile $DIR/$tfile-2
+ rm -f $DIR/$tfile
+ fail mds
+ $CHECKSTAT $DIR/$tfile && return 1
+ $CHECKSTAT $DIR/$tfile-2 ||return 2
+ rm $DIR/$tfile-2
+ return 0
+}
+run_test 10 "create |X| rename unlink"
+
+test_11() {
+ mcreate $DIR/$tfile
+ echo "old" > $DIR/$tfile
+ mv $DIR/$tfile $DIR/$tfile-2
+ replay_barrier mds
+ echo "new" > $DIR/$tfile
+ grep new $DIR/$tfile
+ grep old $DIR/$tfile-2
+ fail mds
+ grep new $DIR/$tfile || return 1
+ grep old $DIR/$tfile-2 || return 2
+}
+run_test 11 "create open write rename |X| create-old-name read"
+
+test_12() {
+ mcreate $DIR/$tfile
+ multiop $DIR/$tfile o_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+ replay_barrier mds
+ kill -USR1 $pid
+ wait $pid || return 1
+
+ fail mds
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 12 "open, unlink |X| close"
+
+
+# 1777 - replay open after committed chmod that would make
+# a regular open a failure
+test_13() {
+ mcreate $DIR/$tfile
+ multiop $DIR/$tfile O_wc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ chmod 0 $DIR/$tfile
+ $CHECKSTAT -p 0 $DIR/$tfile
+ replay_barrier mds
+ fail mds
+ kill -USR1 $pid
+ wait $pid || return 1
+
+ $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
+ return 0
+}
+run_test 13 "open chmod 0 |x| write close"
+
+test_14() {
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+ replay_barrier mds
+ kill -USR1 $pid || return 1
+ wait $pid || return 2
+
+ fail mds
+ [ -e $DIR/$tfile ] && return 3
+ return 0
+}
+run_test 14 "open(O_CREAT), unlink |X| close"
+
+test_15() {
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+ replay_barrier mds
+ touch $DIR/g11 || return 1
+ kill -USR1 $pid
+ wait $pid || return 2
-RLUSTRE=${RLUSTRE:-$LUSTRE}
-RPWD=${RPWD:-$PWD}
+ fail mds
+ [ -e $DIR/$tfile ] && return 3
+ touch $DIR/h11 || return 4
+ return 0
+}
+run_test 15 "open(O_CREAT), unlink |X| touch new, close"
-. $LTESTDIR/functional/llite/common/common.sh
-# XXX I wish all this stuff was in some default-config.sh somewhere
-MOUNTPT=${MOUNTPT:-/mnt/lustre}
-MDSDEV=${MDSDEV:-/tmp/mds-`hostname`}
-MDSSIZE=${MDSSIZE:-100000}
-OSTDEV=${OSTDEV:-/tmp/ost-`hostname`}
-OSTSIZE=${OSTSIZE:-100000}
-UPCALL=${UPCALL:-$PWD/replay-single-upcall.sh}
-FSTYPE=${FSTYPE:-ext3}
-TIMEOUT=${TIMEOUT:-5}
+test_16() {
+ replay_barrier mds
+ mcreate $DIR/$tfile
+ munlink $DIR/$tfile
+ mcreate $DIR/$tfile-2
+ fail mds
+ [ -e $DIR/$tfile ] && return 1
+ [ -e $DIR/$tfile-2 ] || return 2
+ munlink $DIR/$tfile-2 || return 3
+}
+run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
+
+test_17() {
+ replay_barrier mds
+ multiop $DIR/$tfile O_c &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ fail mds
+ kill -USR1 $pid || return 1
+ wait $pid || return 2
+ $CHECKSTAT -t file $DIR/$tfile || return 3
+ rm $DIR/$tfile
+}
+run_test 17 "|X| open(O_CREAT), |replay| close"
+
+test_18() {
+ replay_barrier mds
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+ touch $DIR/$tfile-2 || return 1
+ kill -USR1 $pid
+ wait $pid || return 2
+
+ fail mds
+ [ -e $DIR/$tfile ] && return 3
+ [ -e $DIR/$tfile-2 ] || return 4
+ # this touch frequently fails
+ touch $DIR/$tfile-3 || return 5
+ munlink $DIR/$tfile-2 || return 6
+ munlink $DIR/$tfile-3 || return 7
+ return 0
+}
+run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
-start() {
- facet=$1
- shift
- lconf --node ${facet}_facet $@ replay-single.xml
+# bug 1855 (a simpler form of test_11 above)
+test_19() {
+ replay_barrier mds
+ mcreate $DIR/$tfile
+ echo "old" > $DIR/$tfile
+ mv $DIR/$tfile $DIR/$tfile-2
+ grep old $DIR/$tfile-2
+ fail mds
+ grep old $DIR/$tfile-2 || return 2
}
+run_test 19 "|X| mcreate, open, write, rename "
+
+test_20() {
+ replay_barrier mds
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+
+ fail mds
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
+
+test_21() {
+ replay_barrier mds
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+ touch $DIR/g11 || return 1
+
+ fail mds
+ kill -USR1 $pid
+ wait $pid || return 2
+ [ -e $DIR/$tfile ] && return 3
+ touch $DIR/h11 || return 4
+ return 0
+}
+run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
+
+test_22() {
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+
+ replay_barrier mds
+ rm -f $DIR/$tfile
+
+ fail mds
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
+
+test_23() {
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+
+ replay_barrier mds
+ rm -f $DIR/$tfile
+ touch $DIR/g11 || return 1
+
+ fail mds
+ kill -USR1 $pid
+ wait $pid || return 2
+ [ -e $DIR/$tfile ] && return 3
+ touch $DIR/h11 || return 4
+ return 0
+}
+run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
+
+test_24() {
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+
+ replay_barrier mds
+ fail mds
+ rm -f $DIR/$tfile
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
+
+test_25() {
+ multiop $DIR/$tfile O_tSc &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+
+ replay_barrier mds
+ fail mds
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
+
+test_26() {
+ replay_barrier mds
+ multiop $DIR/$tfile-1 O_tSc &
+ pid1=$!
+ multiop $DIR/$tfile-2 O_tSc &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile-1
+ rm -f $DIR/$tfile-2
+ kill -USR1 $pid2
+ wait $pid2 || return 1
+
+ fail mds
+ kill -USR1 $pid1
+ wait $pid1 || return 2
+ [ -e $DIR/$tfile-1 ] && return 3
+ [ -e $DIR/$tfile-2 ] && return 4
+ return 0
+}
+run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
+
+test_27() {
+ replay_barrier mds
+ multiop $DIR/$tfile-1 O_tSc &
+ pid1=$!
+ multiop $DIR/$tfile-2 O_tSc &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile-1
+ rm -f $DIR/$tfile-2
-stop() {
- facet=$1
- shift
- lconf --node ${facet}_facet $@ -d replay-single.xml
+ fail mds
+ kill -USR1 $pid1
+ wait $pid1 || return 1
+ kill -USR1 $pid2
+ wait $pid2 || return 2
+ [ -e $DIR/$tfile-1 ] && return 3
+ [ -e $DIR/$tfile-2 ] && return 4
+ return 0
}
+run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
-replay_barrier() {
- local dev=$1
+test_28() {
+ multiop $DIR/$tfile-1 O_tSc &
+ pid1=$!
+ multiop $DIR/$tfile-2 O_tSc &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ replay_barrier mds
+ rm -f $DIR/$tfile-1
+ rm -f $DIR/$tfile-2
+ kill -USR1 $pid2
+ wait $pid2 || return 1
+
+ fail mds
+ kill -USR1 $pid1
+ wait $pid1 || return 2
+ [ -e $DIR/$tfile-1 ] && return 3
+ [ -e $DIR/$tfile-2 ] && return 4
+ return 0
+}
+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 &
+ pid1=$!
+ multiop $DIR/$tfile-2 O_tSc &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ replay_barrier mds
+ rm -f $DIR/$tfile-1
+ rm -f $DIR/$tfile-2
+
+ fail mds
+ kill -USR1 $pid1
+ wait $pid1 || return 1
+ kill -USR1 $pid2
+ wait $pid2 || return 2
+ [ -e $DIR/$tfile-1 ] && return 3
+ [ -e $DIR/$tfile-2 ] && return 4
+ return 0
+}
+run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
+
+test_30() {
+ multiop $DIR/$tfile-1 O_tSc &
+ pid1=$!
+ multiop $DIR/$tfile-2 O_tSc &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile-1
+ rm -f $DIR/$tfile-2
+
+ replay_barrier mds
+ fail mds
+ kill -USR1 $pid1
+ wait $pid1 || return 1
+ kill -USR1 $pid2
+ wait $pid2 || return 2
+ [ -e $DIR/$tfile-1 ] && return 3
+ [ -e $DIR/$tfile-2 ] && return 4
+ return 0
+}
+run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
+
+test_31() {
+ multiop $DIR/$tfile-1 O_tSc &
+ pid1=$!
+ multiop $DIR/$tfile-2 O_tSc &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile-1
+
+ replay_barrier mds
+ rm -f $DIR/$tfile-2
+ fail mds
+ kill -USR1 $pid1
+ wait $pid1 || return 1
+ kill -USR1 $pid2
+ wait $pid2 || return 2
+ [ -e $DIR/$tfile-1 ] && return 3
+ [ -e $DIR/$tfile-2 ] && return 4
+ return 0
+}
+run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
+
+# 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 &
+ pid1=$!
+ multiop $DIR/$tfile O_c &
+ pid2=$!
+ # give multiop a chance to open
+ sleep 1
+ mds_evict_client
+ df $MOUNT || df $MOUNT || return 1
+ kill -USR1 $pid1
+ kill -USR1 $pid2
+ sleep 1
+ return 0
+}
+run_test 32 "close() notices client eviction; close() after client eviction"
+
+# Abort recovery before client complete
+test_33() {
+ replay_barrier mds
+ touch $DIR/$tfile
+ fail_abort mds
+ # this file should be gone, because the replay was aborted
+ $CHECKSTAT -t file $DIR/$tfile && return 1
+ return 0
+}
+run_test 33 "abort recovery before client does replay"
+
+test_34() {
+ multiop $DIR/$tfile O_c &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rm -f $DIR/$tfile
+
+ replay_barrier mds
+ fail_abort mds
+ kill -USR1 $pid
+ [ -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
+test_35() {
+ touch $DIR/$tfile
+
+#define OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+ rm -f $DIR/$tfile &
+ sleep 1
sync
- lctl --device %${dev}1 readonly
- lctl --device %${dev}1 notransno
+ sleep 1
+ # give a chance to remove from MDS
+ fail_abort mds
+ $CHECKSTAT -t file $DIR/$tfile && return 1 || true
}
+run_test 35 "test recovery from llog for unlink op"
-fail() {
- stop mds
- start mds
- df $MOUNTPT
+# b=2432 resent cancel after replay uses wrong cookie,
+# so don't resend cancels
+test_36() {
+ replay_barrier mds
+ touch $DIR/$tfile
+ checkstat $DIR/$tfile
+ facet_failover mds
+ cancel_lru_locks MDC
+ if dmesg | grep "unknown lock cookie"; then
+ echo "cancel after replay failed"
+ return 1
+ fi
}
+run_test 36 "don't resend cancel"
+
+# b=2368
+# directory orphans can't be unlinked from PENDING directory
+test_37() {
+ rmdir $DIR/$tfile 2>/dev/null
+ multiop $DIR/$tfile dD_c &
+ pid=$!
+ # give multiop a chance to open
+ sleep 1
+ rmdir $DIR/$tfile
-do_lmc() {
- lmc -m replay-single.xml $@
+ replay_barrier mds
+ # clear the dmesg buffer so we only see errors from this recovery
+ dmesg -c >/dev/null
+ fail_abort mds
+ kill -USR1 $pid
+ dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
+ sync
+ return 0
}
+run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
-add_facet() {
- local facet=$1
- shift
- do_lmc --add node --node ${facet}_facet $@ --timeout $TIMEOUT
- do_lmc --add net --node ${facet}_facet --nid localhost --nettype tcp
+test_38() {
+ createmany -o $DIR/$tfile-%d 800
+ unlinkmany $DIR/$tfile-%d 0 400
+ replay_barrier mds
+ fail mds
+ unlinkmany $DIR/$tfile-%d 400 400
+ sleep 2
+ $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
+run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
+
+test_39() {
+ createmany -o $DIR/$tfile-%d 800
+ replay_barrier mds
+ unlinkmany $DIR/$tfile-%d 0 400
+ fail mds
+ unlinkmany $DIR/$tfile-%d 400 400
+ sleep 2
+ $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
+}
+run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
+
+count_ost_writes() {
+ cat /proc/fs/lustre/osc/*/stats |
+ awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
+}
+
+#b=2477,2532
+test_40(){
+ $LCTL mark multiop $MOUNT/$tfile OS_c
+ multiop $MOUNT/$tfile OS_c &
+ PID=$!
+ writeme -s $MOUNT/${tfile}-2 &
+ WRITE_PID=$!
+ sleep 1
+ facet_failover mds
+#define OBD_FAIL_MDS_CONNECT_NET 0x117
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000117"
+ kill -USR1 $PID
+ stat1=`count_ost_writes`
+ sleep $TIMEOUT
+ stat2=`count_ost_writes`
+ echo "$stat1, $stat2"
+ if [ $stat1 -lt $stat2 ]; then
+ echo "writes continuing during recovery"
+ RC=0
+ else
+ echo "writes not continuing during recovery, bug 2477"
+ RC=4
+ fi
+ echo "waiting for writeme $WRITE_PID"
+ kill $WRITE_PID
+ wait $WRITE_PID
+
+ echo "waiting for multiop $PID"
+ wait $PID || return 2
+ do_facet client munlink $MOUNT/$tfile || return 3
+ do_facet client munlink $MOUNT/${tfile}-2 || return 3
+ return $RC
+}
+run_test 40 "cause recovery in ptlrpc, ensure IO continues"
+
+equals_msg test complete, cleaning up
+$CLEANUP
-gen_config() {
- rm -f replay-single.xml
- add_facet mds
- add_facet ost
- add_facet client --lustre_upcall $UPCALL
- do_lmc --add mds --node mds_facet --mds mds1 --dev $MDSDEV --size $MDSSIZE
- do_lmc --add ost --node ost_facet --ost ost1 --dev $OSTDEV --size $OSTSIZE
- do_lmc --add mtpt --node client_facet --path $MOUNTPT --mds mds1 --ost ost1
-}
-
-gen_config
-start mds
-start ost
-start client
-
-touch $MOUNTPT/lustre-works
-replay_barrier mds
-touch $MOUNTPT/lustre-does-not-work
-
-stop client
-stop ost
-stop mds