+test_20() { #16389
+ BEFORE=`date +%s`
+ replay_barrier $SINGLEMDS
+ touch $MOUNT1/a
+ touch $MOUNT2/b
+ umount $MOUNT2
+ fail $SINGLEMDS
+ rm $MOUNT1/a
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
+ TIER1=$((`date +%s` - BEFORE))
+ BEFORE=`date +%s`
+ replay_barrier $SINGLEMDS
+ touch $MOUNT1/a
+ touch $MOUNT2/b
+ umount $MOUNT2
+ fail $SINGLEMDS
+ rm $MOUNT1/a
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
+ TIER2=$((`date +%s` - BEFORE))
+ [ $TIER2 -ge $((TIER1 * 2)) ] && \
+ error "recovery time is growing $TIER2 > $TIER1"
+ return 0
+}
+run_test 20 "recovery time is not increasing"
+
+# commit on sharing tests
+test_21a() {
+ local param_file=$TMP/$tfile-params
+
+ save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
+ do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
+ touch $MOUNT1/$tfile-1
+ mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
+ mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
+ replay_barrier_nosync $SINGLEMDS
+ umount $MOUNT2
+
+ facet_failover $SINGLEMDS
+
+ # all renames are replayed
+ unlink $MOUNT1/$tfile-3 || return 2
+
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
+
+ do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
+ rm -rf $MOUNT1/$tfile-*
+ restore_lustre_params < $param_file
+ rm -f $param_file
+ return 0
+}
+run_test 21a "commit on sharing"
+
+test_21b_sub () {
+ local mds=$1
+ do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
+
+ do_facet $mds sync
+ do_node $CLIENT1 touch $MOUNT1/$tfile-1
+ do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
+ do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
+
+ replay_barrier_nosync $mds
+ shutdown_client $CLIENT2 $MOUNT1
+
+ facet_failover $mds
+
+ # were renames replayed?
+ local rc=0
+ echo UNLINK $MOUNT1/$tfile-3
+ do_node $CLIENT1 unlink $MOUNT1/$tfile-3 || \
+ { echo "unlink $tfile-3 fail!" && rc=1; }
+
+ boot_node $CLIENT2
+ zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail"
+
+ return $rc
+}
+
+test_21b() {
+ [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
+ [ $CLIENTCOUNT -lt 2 ] && \
+ { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
+
+ if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
+ skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
+ return 0
+ fi
+
+
+ zconf_umount_clients $CLIENTS $MOUNT2
+ zconf_mount_clients $CLIENTS $MOUNT1
+
+ local param_file=$TMP/$tfile-params
+
+ local num=$(get_mds_dir $MOUNT1)
+
+ save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
+
+ # COS enabled
+ local COS=1
+ do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+
+ test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
+
+ # COS disabled (should fail)
+ COS=0
+ do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
+
+ # there is still a window when transactions may be written to disk before
+ # the mds device is set R/O. To avoid such a rare test failure, the check
+ # is repeated several times.
+ local n_attempts=1
+ while true; do
+ test_21b_sub mds$num || break;
+ let n_attempts=n_attempts+1
+ [ $n_attempts -gt 3 ] &&
+ error "The test cannot check whether COS works or not: all renames are replied w/o COS"
+ done
+ zconf_mount_clients $CLIENTS $MOUNT2
+ restore_lustre_params < $param_file
+ rm -f $param_file
+ return 0
+}
+run_test 21b "commit on sharing, two clients"
+
+checkstat_22() {
+ checkstat $MOUNT1/$remote_dir || return 1
+ checkstat $MOUNT1/$remote_dir/dir || return 2
+ checkstat $MOUNT1/$remote_dir/$tfile-1 || return 3
+ checkstat $MOUNT1/$remote_dir/dir/$tfile-1 || return 4
+ return 0
+}
+
+create_remote_dir_files_22() {
+ do_node $CLIENT2 mkdir ${MOUNT2}/$remote_dir/dir || return 1
+ do_node $CLIENT1 createmany -o $MOUNT1/$remote_dir/dir/$tfile- 2 ||
+ return 2
+ do_node $CLIENT2 createmany -o $MOUNT2/$remote_dir/$tfile- 2 ||
+ return 3
+ return 0
+}
+
+test_22a () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ ([ $FAILURE_MODE == "HARD" ] &&
+ [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+ skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+ return 0
+
+ local MDTIDX=1
+ local remote_dir=${tdir}/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+ # OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+ CLIENT_PID=$!
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+ fail mds${MDTIDX}
+ wait $CLIENT_PID || error "lfs mkdir failed"
+
+ replay_barrier mds${MDTIDX}
+ create_remote_dir_files_22 || error "Remote creation failed $?"
+ fail mds${MDTIDX}
+
+ checkstat_22 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+ return 0
+}
+run_test 22a "c1 lfs mkdir -i 1 dir1, M0 drop reply & fail, c2 mkdir dir1/dir"
+
+test_22b () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local MDTIDX=1
+ local remote_dir=$tdir/remote_dir
+
+ # OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+ CLIENT_PID=$!
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+ fail mds${MDTIDX},mds$((MDTIDX + 1))
+ wait $CLIENT_PID || error "lfs mkdir failed"
+
+ replay_barrier mds$MDTIDX
+ create_remote_dir_files_22 || error "Remote creation failed $?"
+ fail mds${MDTIDX}
+
+ checkstat_22 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+ return 0
+}
+run_test 22b "c1 lfs mkdir -i 1 d1, M0 drop reply & fail M0/M1, c2 mkdir d1/dir"
+
+test_22c () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ ([ $FAILURE_MODE == "HARD" ] &&
+ [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+ skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+ return 0
+ local MDTIDX=1
+ local remote_dir=${tdir}/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+ # OBD_FAIL_MDS_DROP_OBJ_UPDATE 0x188
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x188
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+ CLIENT_PID=$!
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+ fail mds$((MDTIDX+1))
+ wait $CLIENT_PID || error "lfs mkdir failed"
+
+ replay_barrier mds$MDTIDX
+ create_remote_dir_files_22 || error "Remote creation failed $?"
+ fail mds$MDTIDX
+
+ checkstat_22 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+ return 0
+}
+run_test 22c "c1 lfs mkdir -i 1 d1, M1 drop update & fail M1, c2 mkdir d1/dir"
+
+test_22d () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local MDTIDX=1
+ local remote_dir=${tdir}/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+
+ # OBD_FAIL_MDS_DROP_OBJ_UPDATE 0x188
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x188
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
+ CLIENT_PID=$!
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+ fail mds${MDTIDX},mds$((MDTIDX + 1))
+ wait $CLIENT_PID || error "lfs mkdir failed"
+
+ replay_barrier mds$MDTIDX
+ create_remote_dir_files_22 || error "Remote creation failed $?"
+ fail mds$MDTIDX
+
+ checkstat_22 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+ return 0
+}
+run_test 22d "c1 lfs mkdir -i 1 d1, M1 drop update & fail M0/M1,c2 mkdir d1/dir"
+
+checkstat_23() {
+ checkstat $MOUNT1/$remote_dir || return 1
+ checkstat $MOUNT1/$remote_dir/$tfile-1 || return 2
+ return 0
+}
+
+create_remote_dir_files_23() {
+ do_node $CLIENT2 mkdir ${MOUNT2}/$remote_dir || return 1
+ do_node $CLIENT2 createmany -o $MOUNT2/$remote_dir/$tfile- 2 || return 2
+ return 0
+}
+
+test_23a () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ ([ $FAILURE_MODE == "HARD" ] &&
+ [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+ skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+ return 0
+ local MDTIDX=1
+ local remote_dir=$tdir/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+ error "lfs mkdir failed"
+ # OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+ do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+ local CLIENT_PID=$!
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+ fail mds$((MDTIDX + 1))
+ wait $CLIENT_PID || error "rmdir remote dir failed"
+
+ replay_barrier mds${MDTIDX}
+ create_remote_dir_files_23 || error "Remote creation failed $?"
+ fail mds${MDTIDX}
+
+ checkstat_23 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+ return 0
+}
+run_test 23a "c1 rmdir d1, M1 drop reply and fail, client2 mkdir d1"
+
+test_23b () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local MDTIDX=1
+ local remote_dir=$tdir/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+ error "lfs mkdir failed"
+
+ # OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
+ do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+ local CLIENT_PID=$!
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
+
+ fail mds${MDTIDX},mds$((MDTIDX + 1))
+ wait $CLIENT_PID || error "rmdir remote dir failed"
+
+ replay_barrier mds${MDTIDX}
+ create_remote_dir_files_23 || error "Remote creation failed $?"
+ fail mds${MDTIDX}
+
+ checkstat_23 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || error "rmdir remote_dir failed"
+ return 0
+}
+run_test 23b "c1 rmdir d1, M1 drop reply and fail M0/M1, c2 mkdir d1"
+
+test_23c () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+
+ ([ $FAILURE_MODE == "HARD" ] &&
+ [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+ skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+ return 0
+ local MDTIDX=1
+ local remote_dir=$tdir/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+ error "lfs mkdir failed"
+
+ # OBD_FAIL_MDS_DROP_OBJ_UPDATE 0x188
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x188
+ do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+ CLIENT_PID=$!
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+ fail mds${MDTIDX}
+ wait $CLIENT_PID || error "rmdir remote dir failed"
+
+ replay_barrier mds${MDTIDX}
+ create_remote_dir_files_23 || error "Remote creation failed $?"
+ fail mds${MDTIDX}
+
+ checkstat_23 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || return 6
+ return 0
+}
+run_test 23c "c1 rmdir d1, M0 drop update reply and fail M0, c2 mkdir d1"
+
+test_23d () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local MDTIDX=1
+ local remote_dir=$tdir/remote_dir
+
+ do_node $CLIENT1 mkdir -p $MOUNT1/${tdir}
+ do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir ||
+ error "lfs mkdir failed"
+
+ # OBD_FAIL_MDS_DROP_OBJ_UPDATE 0x188
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x188
+ do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
+ CLIENT_PID=$!
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0
+
+ fail mds${MDTIDX},mds$((MDTIDX + 1))
+ wait $CLIENT_PID || error "rmdir remote dir failed"
+
+ replay_barrier mds${MDTIDX}
+ create_remote_dir_files_23 || error "Remote creation failed $?"
+ fail mds${MDTIDX}
+
+ checkstat_23 || error "check stat failed $?"
+
+ rm -rf $MOUNT1/$tdir || return 6
+ return 0
+}
+run_test 23d "c1 rmdir d1, M0 drop update reply and fail M0/M1, c2 mkdir d1"
+
+# end commit on sharing tests
+
+complete $SECONDS