X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-lnet.sh;h=485e105d002ffc77b528e1634f0a5394041f5e90;hp=1c09c20a6e4c1075947f322c1ba1b9c6407a5507;hb=3b263dd80ee56efae922e2cfcab375dbe2cb273a;hpb=61f996b07f040376049522c9e8ee4469274607ce;ds=sidebyside

diff --git a/lustre/tests/sanity-lnet.sh b/lustre/tests/sanity-lnet.sh
index 1c09c20..485e105 100755
--- a/lustre/tests/sanity-lnet.sh
+++ b/lustre/tests/sanity-lnet.sh
@@ -31,9 +31,6 @@ init_logging
 
 build_test_filter
 
-export LNETCTL=${LNETCTL:-"$LUSTRE/../lnet/utils/lnetctl"}
-[ ! -f "$LNETCTL" ] &&
-	export LNETCTL=$(which lnetctl 2> /dev/null)
 [[ -z $LNETCTL ]] && skip "Need lnetctl"
 
 restore_mounts=false
@@ -58,7 +55,8 @@ fi
 
 cleanup_testsuite() {
 	trap "" EXIT
-	rm -f $TMP/sanity-dlc*
+	# Cleanup any tmp files created by the sub tests
+	rm -f $TMP/sanity-lnet*
 	cleanup_netns
 	cleanup_lnet
 	if $restore_mounts; then
@@ -849,7 +847,7 @@ peer:
         - nid: 25@gni
 EOF
 	append_global_yaml
-	echo"Add peer with nidrange (gni)"
+	echo "Add peer with nidrange (gni)"
 	compare_peer_add "21@gni" "[22-25]@gni" || error
 	echo "Add peer with nidrange that overlaps primary nid (gni)"
 	compare_peer_add "21@gni" "[21-25]@gni"
@@ -1190,7 +1188,7 @@ test_104() {
 
 	echo "Set < 0;  Should fail"
 	do_lnetctl set response_tracking -1 &&
-		 error "should have failed $?"
+		error "should have failed $?"
 
 	reinit_dlc || return $?
 	cat <<EOF > $tyaml
@@ -1205,7 +1203,7 @@ EOF
 	for ((i = 0; i < 4; i++)); do
 		reinit_dlc || return $?
 		do_lnetctl set response_tracking $i ||
-			 error "should have succeeded $?"
+			error "should have succeeded $?"
 		$LNETCTL global show | grep -q "response_tracking: $i" ||
 			error "Failed to set response_tracking to $i"
 		reinit_dlc || return $?
@@ -1222,7 +1220,7 @@ EOF
 	reinit_dlc || return $?
 	echo "Set > 3; Should fail"
 	do_lnetctl set response_tracking 4 &&
-		 error "should have failed $?"
+		error "should have failed $?"
 
 	reinit_dlc || return $?
 	cat <<EOF > $tyaml
@@ -1413,7 +1411,7 @@ test_204() {
 		$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
 		do_lnetctl discover $($LCTL list_nids | head -n 1) &&
 			error "Should have failed"
-		$LCTL net_drop_del *
+		$LCTL net_drop_del -a
 	done
 
 	lnet_health_post
@@ -1441,7 +1439,7 @@ test_205() {
 		$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
 		do_lnetctl discover $($LCTL list_nids | head -n 1) &&
 			error "Should have failed"
-		$LCTL net_drop_del *
+		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1461,7 +1459,7 @@ test_205() {
 		$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
 		do_lnetctl discover $($LCTL list_nids | head -n 1) &&
 			error "Should have failed"
-		$LCTL net_drop_del *
+		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1475,7 +1473,7 @@ run_test 205 "Check health and resends for multi-rail local failures"
 
 # See lnet/lnet/lib-msg.c:lnet_health_check()
 LNET_REMOTE_RESEND_STATUSES="remote_dropped"
-LNET_REMOTE_NO_RESEND_STATUSES="remote_error remote_timeout network_timeout"
+LNET_REMOTE_NO_RESEND_STATUSES="remote_error remote_timeout"
 test_206() {
 	have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
 	reinit_dlc || return $?
@@ -1493,7 +1491,7 @@ test_206() {
 		$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
 		do_lnetctl discover $($LCTL list_nids | head -n 1) &&
 			error "Should have failed"
-		$LCTL net_drop_del *
+		$LCTL net_drop_del -a
 	done
 
 	lnet_health_post
@@ -1524,7 +1522,7 @@ test_207() {
 		$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
 		do_lnetctl discover $($LCTL list_nids | head -n 1) &&
 			error "Should have failed"
-		$LCTL net_drop_del *
+		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1546,7 +1544,7 @@ test_207() {
 		$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
 		do_lnetctl discover $($LCTL list_nids | head -n 1) &&
 			error "Should have failed"
-		$LCTL net_drop_del *
+		$LCTL net_drop_del -a
 
 		lnet_health_post
 
@@ -1643,6 +1641,353 @@ test_208() {
 }
 run_test 208 "Test various kernel ip2nets configurations"
 
+test_209() {
+	have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+	reinit_dlc || return $?
+	add_net "tcp" "eth0" || return $?
+
+	do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+		error "failed to discover myself"
+
+	echo "Simulate network_timeout w/SR config"
+	lnet_health_pre
+
+	$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e network_timeout
+	do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+		error "Should have failed"
+	$LCTL net_drop_del -a
+
+	lnet_health_post
+
+	check_no_resends || return $?
+	check_no_local_health || return $?
+	check_no_remote_health || return $?
+
+	reinit_dlc || return $?
+	add_net "tcp" "eth0" || return $?
+	add_net "tcp1" "eth0" || return $?
+
+	do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+		error "failed to discover myself"
+
+	echo "Simulate network_timeout w/MR config"
+	lnet_health_pre
+
+	$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e network_timeout
+	$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e network_timeout
+	do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+		error "Should have failed"
+	$LCTL net_drop_del -a
+
+	lnet_health_post
+
+	check_no_resends || return $?
+	check_local_health || return $?
+	check_remote_health || return $?
+
+	return 0
+}
+run_test 209 "Check health, but not resends, for network timeout"
+
+check_nid_in_recovq() {
+	local recovq=$($LNETCTL debug recovery $1)
+	local expect="$2"
+	local nids=$($LCTL list_nids | xargs echo)
+	local found=false
+	local nid=""
+
+	echo "Check recovery queue"
+	echo "$recovq"
+	if [[ $(grep -c 'nid-'<<<$recovq) -ne $expect ]]; then
+		error "Expect $expect NIDs found: \"$recovq\""
+	fi
+
+	[[ $expect -eq 0 ]] && return 0
+
+	for nid in ${nids}; do
+		grep -q "nid-0: $nid"<<<$recovq &&
+			found=true
+	done
+
+	if ! $found; then
+		error "Didn't find local NIDs in recovery queue: \"$recovq\""
+	fi
+
+	return 0
+}
+
+# First enqueue happens at time 0.
+# 2nd at 0 + 2^0 = 1
+# 3rd at 1 + 2^1 = 3
+# 4th at 3 + 2^2 = 7
+# 5th at 7 + 2^3 = 15
+# e.g. after 10 seconds we would expect to have seen the 4th enqueue,
+# (3 pings sent, 4th about to happen) and the 5th enqueue is yet to
+# happen
+# 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.
+check_ping_count() {
+	local queue="$1"
+	local expect="$2"
+
+	echo "Check ping counts:"
+	local ping_count
+	if [[ $queue == "ni" ]]; then
+		$LNETCTL net show -v 2 | egrep 'nid|health value|ping'
+		ping_count=( $($LNETCTL net show -v 2 |
+				awk '/ping_count/{print $NF}') )
+	elif [[ $queue == "peer_ni" ]]; then
+		$LNETCTL peer show -v 2 | egrep 'nid|health value|ping'
+		ping_count=( $($LNETCTL peer show -v 2 |
+				awk '/ping_count/{print $NF}') )
+	else
+		error "Unrecognized queue \"$queue\""
+		return 1
+	fi
+
+	local count
+	local found=false
+	for count in ${ping_count[@]}; do
+		if [[ $count -eq $expect ]]; then
+			if [[ $expect -ne 0 ]] && $found ; then
+				error "Found more than one interface matching \"$expect\" ping count"
+				return 1
+			else
+				echo "Expect ping count \"$expect\" found \"$count\""
+				found=true;
+			fi
+		elif [[ $count -ne 0 ]]; then
+			error "Found interface with ping count \"$count\" but expect \"$expect\""
+			return 1
+		fi
+	done
+
+	return 0
+}
+
+test_210() {
+	have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+	reinit_dlc || return $?
+	add_net "tcp" "eth0" || return $?
+	add_net "tcp1" "eth0" || return $?
+
+	local prim_nid=$($LCTL list_nids | head -n 1)
+
+	do_lnetctl discover $prim_nid ||
+		error "failed to discover myself"
+
+	# Set recovery limit to 10 seconds.
+	do_lnetctl set recovery_limit 10 ||
+		error "failed to set recovery_limit"
+
+	$LCTL set_param debug=+net
+	# Use local_error so LNet doesn't attempt to resend the discovery ping
+	$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e local_error
+	$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e local_error
+	do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+		error "Expected discovery to fail"
+
+	sleep 5
+	check_nid_in_recovq "-l" 1
+	check_ping_count "ni" "2"
+
+	sleep 5
+
+	check_nid_in_recovq "-l" 1
+	check_ping_count "ni" "3"
+
+	$LCTL net_drop_del -a
+
+	return 0
+}
+run_test 210 "Local NI recovery checks"
+
+test_211() {
+	have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+	reinit_dlc || return $?
+	add_net "tcp" "eth0" || return $?
+	add_net "tcp1" "eth0" || return $?
+
+	local prim_nid=$($LCTL list_nids | head -n 1)
+
+	do_lnetctl discover $prim_nid ||
+		error "failed to discover myself"
+
+	# Set recovery limit to 10 seconds.
+	do_lnetctl set recovery_limit 10 ||
+		error "failed to set recovery_limit"
+
+	$LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e remote_error
+	$LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e remote_error
+
+	# Set health to 0 on one interface. This forces it onto the recovery
+	# queue.
+	$LNETCTL peer set --nid $prim_nid --health 0
+
+	# After 5 seconds, we expect the peer NI to still be in recovery
+	sleep 5
+	check_nid_in_recovq "-p" 1
+	check_ping_count "peer_ni" "2"
+
+	# After 15 seconds, the peer NI should have been fully processed out of
+	# the recovery queue. We'll allow a total of 17 seconds to account for
+	# differences in sleeping for whole seconds vs. the more accurate time
+	# keeping that is done in the recovery code.
+	sleep 12
+	check_nid_in_recovq "-p" 0
+	check_ping_count "peer_ni" "4"
+
+	$LCTL net_drop_del -a
+
+	# Set health to force it back onto the recovery queue. Set to 500 means
+	# in 5 seconds it should be back at maximum value. We'll wait a couple
+	# more seconds than that to be safe.
+	# NB: we need to increase the recovery limit so the peer NI is
+	# eligible again
+	do_lnetctl set recovery_limit 50 ||
+		error "failed to set recovery_limit"
+
+	$LNETCTL peer set --nid $prim_nid --health 500
+
+	sleep 7
+
+	check_nid_in_recovq "-p" 0
+	check_ping_count "peer_ni" "0"
+
+	return 0
+}
+run_test 211 "Remote NI recovery checks"
+
+test_212() {
+	local rnodes=$(remote_nodes_list)
+	[[ -z $rnodes ]] && skip "Need at least 1 remote node"
+
+	cleanup_lnet || error "Failed to cleanup before test execution"
+
+	# Loading modules should configure LNet with the appropriate
+	# test-framework configuration
+	load_modules || error "Failed to load modules"
+
+	local my_nid=$($LCTL list_nids | head -n 1)
+	[[ -z $my_nid ]] &&
+		error "Failed to get primary NID for local host $HOSTNAME"
+
+	local rnode=$(awk '{print $1}' <<<$rnodes)
+	local rnodenids=$(do_node $rnode $LCTL list_nids | xargs echo)
+	local rloaded=false
+
+	if [[ -z $rnodenids ]]; then
+		do_rpc_nodes $rnode load_modules_local
+		rloaded=true
+		rnodenids=$(do_node $rnode $LCTL list_nids | xargs echo)
+	fi
+
+	local rnodepnid=$(awk '{print $1}' <<< $rnodenids)
+
+	[[ -z $rnodepnid ]] &&
+		error "Failed to get primary NID for remote host $rnode"
+
+	log "Initial discovery"
+	do_lnetctl discover --force $rnodepnid ||
+		error "Failed to discover $rnodepnid"
+
+	do_node $rnode "$LNETCTL discover --force $my_nid" ||
+		error "$rnode failed to discover $my_nid"
+
+	log "Fail local discover ping to set LNET_PEER_REDISCOVER flag"
+	$LCTL net_drop_add -s "*@$NETTYPE" -d "*@$NETTYPE" -r 1 -e local_error
+	do_lnetctl discover --force $rnodepnid &&
+		error "Discovery should have failed"
+	$LCTL net_drop_del -a
+
+	local nid
+	for nid in $rnodenids; do
+		# We need GET (PING) delay just long enough so we can trigger
+		# discovery on the remote peer
+		$LCTL net_delay_add -s "*@$NETTYPE" -d $nid -r 1 -m GET -l 3
+		$LCTL net_drop_add -s "*@$NETTYPE" -d $nid -r 1 -m GET -e local_error
+		# We need PUT (PUSH) delay just long enough so we can process
+		# the PING failure
+		$LCTL net_delay_add -s "*@$NETTYPE" -d $nid -r 1 -m PUT -l 6
+	done
+
+	log "Force $HOSTNAME to discover $rnodepnid (in background)"
+	# We want to get a PING sent that we know will eventually fail.
+	# The delay rules we added will ensure the ping is not sent until
+	# the PUSH is also in flight (see below), and the drop rule ensures that
+	# when the PING is eventually sent it will error out
+	do_lnetctl discover --force $rnodepnid &
+	local pid1=$!
+
+	# We want a discovery PUSH from rnode to put rnode back on our
+	# discovery queue. This should cause us to try and send a PUSH to rnode
+	# while the PING is still outstanding.
+	log "Force $rnode to discover $my_nid"
+	do_node $rnode $LNETCTL discover --force $my_nid
+
+	# At this point we'll have both PING_SENT and PUSH_SENT set for the
+	# rnode peer. Wait for the PING to error out which should terminate the
+	# discovery process that we backgrounded.
+	log "Wait for $pid1"
+	wait $pid1
+	log "Finished wait on $pid1"
+
+	# The PING send failure clears the PING_SENT flag and puts the peer back
+	# on the discovery queue. When discovery thread processes the peer it
+	# will mistakenly clear the PUSH_SENT flag (and set PUSH_FAILED).
+	# Discovery will then complete for this peer even though we have an
+	# outstanding PUSH.
+	# When PUSH is actually unlinked it will be forced back onto the
+	# discovery queue, but we no longer have a ref on the peer. When
+	# discovery completes again, we'll trip the ASSERT in
+	# lnet_destroy_peer_locked()
+
+	# Delete the delay rules to send the PUSH
+	$LCTL net_delay_del -a
+	# Delete the drop rules
+	$LCTL net_drop_del -a
+
+	unload_modules ||
+		error "Failed to unload modules"
+	if $rloaded; then
+		do_rpc_nodes $rnode unload_modules_local ||
+			error "Failed to unload modules on $rnode"
+	fi
+
+	return 0
+}
+run_test 212 "Check discovery refcount loss bug (LU-14627)"
+
+test_213() {
+	have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+	cleanup_netns || error "Failed to cleanup netns before test execution"
+	cleanup_lnet || error "Failed to unload modules before test execution"
+
+	setup_fakeif || error "Failed to add fake IF"
+	have_interface "$FAKE_IF" ||
+		error "Expect $FAKE_IF configured but not found"
+
+	reinit_dlc || return $?
+
+	add_net "tcp" "eth0" || return $?
+	add_net "tcp" "$FAKE_IF" || return $?
+
+	local nid1=$(lctl list_nids | head -n 1)
+	local nid2=$(lctl list_nids | tail --lines 1)
+
+	[[ $(lctl which_nid $nid1 $nid2) == $nid1 ]] ||
+		error "Expect nid1 \"$nid1\" to be preferred"
+
+	[[ $(lctl which_nid $nid2 $nid1) == $nid2 ]] ||
+		error "Expect nid2 \"$nid2\" to be preferred"
+
+	return 0
+}
+run_test 213 "Check LNetDist calculation for multiple local NIDs"
+
 test_300() {
 	# LU-13274
 	local header