On Tue, Aug 12, 2025 at 11:23:17AM +0200, Paolo Abeni wrote:
On 8/5/25 11:46 AM, Hangbin Liu wrote: @@ -0,0 +1,95 @@
+#!/bin/bash +# SPDX-License-Identifier: GPL-2.0 +# +# Test if a bond interface works with lacp_active=off.
+# shellcheck disable=SC2034 +REQUIRE_MZ=no +NUM_NETIFS=0 +lib_dir=$(dirname "$0") +# shellcheck disable=SC1091 +source "$lib_dir"/../../../net/forwarding/lib.sh
+check_port_state() +{
- local netns=$1
- local port=$2
- local state=$3
- ip -n "${netns}" -d -j link show "$port" | \
jq -e ".[].linkinfo.info_slave_data.ad_actor_oper_port_state_str | index(\"${state}\") != null" > /dev/null+}
+trap cleanup EXIT +setup_ns c_ns s_ns +defer cleanup_all_ns
+# shellcheck disable=SC2154 +ip -n "${c_ns}" link add eth0 type veth peer name eth0 netns "${s_ns}" +ip -n "${c_ns}" link add eth1 type veth peer name eth1 netns "${s_ns}" +ip -n "${s_ns}" link set eth0 up +ip -n "${s_ns}" link set eth1 up +ip -n "${c_ns}" link add bond0 type bond mode 802.3ad lacp_active off lacp_rate fast +ip -n "${c_ns}" link set eth0 master bond0 +ip -n "${c_ns}" link set eth1 master bond0 +ip -n "${c_ns}" link set bond0 up
+# 1. The passive side shouldn't send LACPDU. +RET=0 +client_mac=$(cmd_jq "ip -j -n ${c_ns} link show bond0" ".[].address") +# Wait for the first LACPDU due to state change. +sleep 5 +timeout 62 ip netns exec "${c_ns}" tcpdump --immediate-mode -c 1 -i eth0 \
- -nn -l -vvv ether proto 0x8809 2> /dev/null > /tmp/client_init.out
+grep -q "System $client_mac" /tmp/client_init.out && RET=1 +log_test "802.3ad" "init port pkt lacp_active off"
+# 2. The passive side should not have the 'active' flag. +RET=0 +check_port_state "${c_ns}" "eth0" "active" && RET=1 +log_test "802.3ad" "port state lacp_active off"
+# Set up the switch side with active mode. +ip -n "${s_ns}" link set eth0 down +ip -n "${s_ns}" link set eth1 down +ip -n "${s_ns}" link add bond0 type bond mode 802.3ad lacp_active on lacp_rate fast +ip -n "${s_ns}" link set eth0 master bond0 +ip -n "${s_ns}" link set eth1 master bond0 +ip -n "${s_ns}" link set bond0 up +# Make sure the negotiation finished +sleep 5
Minor nit: I guess the above sleep time depends on some kernel constant, but it's not obvious to me if the minimum time is mandated by the RFC, nor how long is such interval.
We need to make sure the lacp negotiation finished. There is no definition in IEEE standard.
A comment explaining the rationale behind such sleep will help, and possibly a loop with smaller minimal wait and periodic checks for the expected condition up to a significantly higher timeout could make the test both faster on average and more robust.
I will use tc rules to catch pkts and see if we can reduce some time.
Thanks Hangbin