X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-lnet.sh;h=7c47dc89254f7af1e6bb079e437571380a052bf3;hb=b90c0100dd93b56f3bfaee037b3bdd077523f43e;hp=64d18166593772c306b8e7c56ac04d98d7e373a6;hpb=8ee85e15412d32fbe60f70c474c0a28ff15b8351;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity-lnet.sh b/lustre/tests/sanity-lnet.sh
index 64d1816..7c47dc8 100755
--- a/lustre/tests/sanity-lnet.sh
+++ b/lustre/tests/sanity-lnet.sh
@@ -1346,6 +1346,8 @@ function lnet_health_post() {
 
 	restore_lnet_params
 
+	$LCTL net_drop_del -a
+
 	do_lnetctl peer set --health 1000 --all
 	do_lnetctl net set --health 1000 --all
 
@@ -1549,7 +1551,6 @@ test_204() {
 		add_health_test_drop_rules ${hstatus}
 		do_lnetctl discover ${RNIDS[0]} &&
 			error "Should have failed"
-		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1574,7 +1575,6 @@ test_205() {
 		add_health_test_drop_rules ${hstatus}
 		do_lnetctl discover ${RNIDS[0]} &&
 			error "Should have failed"
-		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1589,7 +1589,6 @@ test_205() {
 		add_health_test_drop_rules ${hstatus}
 		do_lnetctl discover ${RNIDS[0]} &&
 			error "Should have failed"
-		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1618,7 +1617,6 @@ test_206() {
 		add_health_test_drop_rules ${hstatus}
 		do_lnetctl discover ${RNIDS[0]} &&
 			error "Should have failed"
-		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1648,8 +1646,6 @@ test_207() {
 
 		lnet_health_post
 
-		$LCTL net_drop_del -a
-
 		check_resends || return $?
 		check_no_local_health || return $?
 		check_remote_health || return $?
@@ -1667,8 +1663,6 @@ test_207() {
 
 		lnet_health_post
 
-		$LCTL net_drop_del -a
-
 		check_no_resends || return $?
 		check_no_local_health || return $?
 		check_remote_health || return $?
@@ -1775,7 +1769,6 @@ test_209() {
 
 	do_lnetctl discover ${RNIDS[0]} &&
 		error "Should have failed"
-	$LCTL net_drop_del -a
 
 	lnet_health_post
 
@@ -1795,7 +1788,6 @@ test_209() {
 
 	do_lnetctl discover ${RNIDS[0]} &&
 		error "Should have failed"
-	$LCTL net_drop_del -a
 
 	lnet_health_post
 
@@ -1847,18 +1839,18 @@ check_nid_in_recovq() {
 # If the recovery limit is 10 seconds, then when the 5th enqueue happens
 # we expect the peer NI to have aged out, so it will not actually be
 # queued.
-# If max_recovery_ping_interval is set to 2 then:
+# If max_recovery_ping_interval is set to 4 then:
 #  First enqueue happens at time 0.
-#  2nd at 0 + 2^0 = 1
-#  3rd at 1 + 2^1 = 3
-#  4th at 3 + 2^1 = 5
-#  5th at 5 + 2^1 = 7
-#  6th at 7 + 2^1 = 9
-#  7th at 9 + 2^1 = 11
-# e.g. after 4 seconds we would expect to have seen the 3th enqueue,
+#  2nd at 0 + min(2^0, 4) = 1
+#  3rd at 1 + min(2^1, 4) = 3
+#  4th at 3 + min(2^2, 4) = 7
+#  5th at 7 + min(2^3, 4) = 11
+#  6th at 11 + min(2^4, 4) = 15
+#  7th at 15 + min(2^5, 4) = 19
+# e.g. after 4 seconds we would expect to have seen the 3rd enqueue,
 # (2 pings sent, 3rd about to happen), and the 4th enqueue is yet to happen
-# e.g. after 10 seconds we would expect to have seen the 6th enqueue,
-# (5 pings sent, 6th about to happen), and the 8th enqueue is yet to happen
+# e.g. after 13 seconds we would expect to have seen the 5th enqueue,
+# (4 pings sent, 5th about to happen), and the 6th enqueue is yet to happen
 check_ping_count() {
 	local queue="$1"
 	local expect="$2"
@@ -1947,7 +1939,7 @@ test_210() {
 
 	default=$($LNETCTL global show |
 		  awk '/max_recovery_ping_interval/{print $NF}')
-	do_lnetctl set max_recovery_ping_interval 2 ||
+	do_lnetctl set max_recovery_ping_interval 4 ||
 		error "failed to set max_recovery_ping_interval"
 
 	$LCTL set_param debug=+net
@@ -1962,9 +1954,9 @@ test_210() {
 	check_nid_in_recovq "-l" "1"
 	check_ping_count "ni" "2"
 
-	sleep 6
+	sleep 9
 	check_nid_in_recovq "-l" "1"
-	check_ping_count "ni" "5"
+	check_ping_count "ni" "4"
 
 	$LCTL net_drop_del -a
 
@@ -2045,7 +2037,7 @@ test_211() {
 
 	default=$($LNETCTL global show |
 		  awk '/max_recovery_ping_interval/{print $NF}')
-	do_lnetctl set max_recovery_ping_interval 2 ||
+	do_lnetctl set max_recovery_ping_interval 4 ||
 		error "failed to set max_recovery_ping_interval"
 
 	$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e remote_error
@@ -2060,9 +2052,9 @@ test_211() {
 	check_nid_in_recovq "-p" "1"
 	check_ping_count "peer_ni" "2"
 
-	sleep 6
+	sleep 9
 	check_nid_in_recovq "-p" "1"
-	check_ping_count "peer_ni" "5"
+	check_ping_count "peer_ni" "4"
 
 	$LCTL net_drop_del -a
 
@@ -2501,8 +2493,9 @@ do_net_add() {
 	local node=$1
 	local net=$2
 	local if=$3
+	local opts=$4
 
-	do_rpc_nodes $node "$LNETCTL net add --net $net --if $if" ||
+	do_rpc_nodes $node "$LNETCTL net add --net $net --if $if $opts" ||
 		error "add $net on interface $if on node $node failed rc=$?"
 }
 
@@ -2559,7 +2552,8 @@ LNIDS=()
 LOCAL_NET=${NETTYPE}1
 REMOTE_NET=${NETTYPE}2
 setup_router_test() {
-	local mod_opts="$@"
+	local mod_opts="$1"
+	local rtr_net_opts="$2"
 
 	if [[ ${#RPEER_INTERFACES[@]} -eq 0 ]]; then
 		init_router_test_vars ||
@@ -2579,7 +2573,7 @@ setup_router_test() {
 	do_nodes $all_nodes "$LNETCTL lnet configure" ||
 		error "Failed to initialize DLC"
 
-	do_net_add $ROUTER $LOCAL_NET ${ROUTER_INTERFACES[0]} ||
+	do_net_add $ROUTER $LOCAL_NET ${ROUTER_INTERFACES[0]} $rtr_net_opts ||
 		return $?
 
 	do_net_add $ROUTER $REMOTE_NET ${ROUTER_INTERFACES[0]} ||
@@ -2830,7 +2824,7 @@ run_test 222 "Check avoid_asym_router_failure=1"
 test_223() {
 	local opts="avoid_asym_router_failure=1 lnet_peer_discovery_disabled=1"
 
-	setup_router_test $opts || return $?
+	setup_router_test "$opts" || return $?
 
 	do_aarf_enabled_test || return $?
 
@@ -2899,7 +2893,7 @@ run_test 224 "Check avoid_asym_router_failure=0"
 test_225() {
 	local opts="avoid_asym_router_failure=0 lnet_peer_discovery_disabled=1"
 
-	setup_router_test $opts || return $?
+	setup_router_test "$opts" || return $?
 
 	do_aarf_disabled_test || return $?
 
@@ -2908,6 +2902,86 @@ test_225() {
 }
 run_test 225 "Check avoid_asym_router_failure=0 w/DD disabled"
 
+do_rtr_peer_health_test() {
+	local expected="$1"
+
+	do_node $ROUTER "$LNETCTL set routing 1" ||
+		error "Unable to enable routing on $ROUTER"
+
+	do_route_add $HOSTNAME $REMOTE_NET ${ROUTER_NIDS[0]} ||
+		return $?
+
+	do_route_add $RPEER $LOCAL_NET ${ROUTER_NIDS[1]} ||
+		return $?
+
+	check_router_ni_status "up" "up" ||
+		return $?
+
+	check_route_aliveness "$HOSTNAME" "up" ||
+		return $?
+
+	check_route_aliveness "$RPEER" "up" ||
+		return $?
+
+	do_lnetctl ping ${RPEER_NIDS[0]} ||
+		error "Failed to ping ${RPEER_NIDS[0]}"
+
+	do_node $RPEER "$LNETCTL ping ${LNIDS[0]}" ||
+		error "$RPEER failed to ping ${LNIDS[0]}"
+
+	# Stop LNet on local host
+	do_lnetctl lnet unconfigure ||
+		error "Failed to stop LNet rc=$?"
+
+	check_router_ni_status "down" "up" ||
+		return $?
+
+	check_route_aliveness "$RPEER" "up" ||
+		return $?
+
+	# The NI used to send the message to the destination will be the
+	# router's NI on LOCAL_NET, so that's the drop count that will be
+	# incremented
+	local d1=$(do_node $ROUTER $LNETCTL net show -v --net $LOCAL_NET | \
+				   awk '/drop_count:/{print $NF}')
+
+	# Ping from RPEER to local host should be dropped by the router
+	do_node $RPEER "$LCTL ping ${LNIDS[0]}" &&
+		error "$RPEER expected ping to fail"
+
+	local d2=$(do_node $ROUTER $LNETCTL net show -v --net $LOCAL_NET | \
+				   awk '/drop_count:/{print $NF}')
+
+	[[ $((d2 - d1)) -ne $expected ]] &&
+		error "Expected drop count change by $expected: $d1 -> $d2"
+
+	return 0
+}
+
+test_226() {
+	setup_router_test avoid_asym_router_failure=0 --peer-timeout=10 ||
+		return $?
+
+	do_rtr_peer_health_test 1 ||
+		return $?
+
+	cleanup_router_test ||
+		return $?
+}
+run_test 226 "Check router peer health enabled"
+
+test_227() {
+	setup_router_test avoid_asym_router_failure=0 --peer-timeout=0 ||
+		return $?
+
+	do_rtr_peer_health_test 0 ||
+		return $?
+
+	cleanup_router_test ||
+		return $?
+}
+run_test 227 "Check router peer health disabled"
+
 test_230() {
 	# LU-12815
 	echo "Check valid values; Should succeed"
@@ -2958,6 +3032,52 @@ test_230() {
 }
 run_test 230 "Test setting conns-per-peer"
 
+test_231() {
+	reinit_dlc || return $?
+
+	do_lnetctl net add --net tcp --if ${INTERFACES[0]} ||
+		error "Failed to add net"
+
+	$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
+	sed -i 's/peer_timeout: .*$/peer_timeout: 0/' \
+		$TMP/sanity-lnet-$testnum-expected.yaml
+
+	reinit_dlc || return $?
+
+	do_lnetctl import $TMP/sanity-lnet-$testnum-expected.yaml ||
+		error "Failed to import configuration"
+
+	$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
+
+	compare_yaml_files || error "Wrong config after import"
+
+	do_lnetctl net del --net tcp --if ${INTERFACES[0]} ||
+		error "Failed to delete net tcp"
+
+	do_lnetctl net add --net tcp --if ${INTERFACES[0]} --peer-timeout=0 ||
+		error "Failed to add net with peer-timeout=0"
+
+	$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
+
+	compare_yaml_files || error "Wrong config after lnetctl net add"
+
+	reinit_dlc || return $?
+
+	# lnet/include/lnet/lib-lnet.h defines DEFAULT_PEER_TIMEOUT 180
+	sed -i 's/peer_timeout: .*$/peer_timeout: 180/' \
+		$TMP/sanity-lnet-$testnum-expected.yaml
+
+	sed -i '/^.*peer_timeout:.*$/d' $TMP/sanity-lnet-$testnum-actual.yaml
+
+	do_lnetctl import $TMP/sanity-lnet-$testnum-actual.yaml ||
+		error "Failed to import config without peer_timeout"
+
+	$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-actual.yaml
+
+	compare_yaml_files
+}
+run_test 231 "Check DLC handling of peer_timeout parameter"
+
 ### Test that linux route is added for each ni
 test_250() {
 	reinit_dlc || return $?
@@ -3002,6 +3122,20 @@ test_300() {
 }
 run_test 300 "packaged LNet UAPI headers can be compiled"
 
+# LU-16081 lnet: Memory leak on adding existing interface
+
+test_301() {
+	reinit_dlc || return $?
+	do_lnetctl net add --net tcp --if ${INTERFACES[0]} ||
+		error "Failed to add net"
+	do_lnetctl net add --net tcp --if ${INTERFACES[0]} &&
+		error "add net should have failed"
+	do_lnetctl net del --net tcp --if ${INTERFACES[0]} ||
+		error "Failed to del net"
+	unload_modules
+}
+run_test 301 "Check for dynamic adds of same/wrong interface (memory leak)"
+
 complete $SECONDS
 
 cleanup_testsuite