network: Don't send RA with zero router lifetime when restarting radv

While investigating https://github.com/systemd/systemd/issues/16356, I
discovered that networkd stops the radv service before adding or updating
prefixes and then starts it again.  This causes networkd to send an RA with
a router lifetime of zero, causing the routes to flap on systems receiving
the RA for a fraction of a second before radv is started again and proper
RAs are sent.  That has the potential to cause issues with latency-sensitive
traffic like gaming or VoIP.  This patch adds a boolean argument to the
sd_radv_stop() function to control this behavior.  The zero lifetime RA is
still sent whenever radv is actually being stopped, but when it is being
restarted for a prefix update (from networkd-dhcp6.c), the final RA is no
longer sent to avoid the route flapping.

1 files changed, 1 insertions, 1 deletions

diff --git a/src/libsystemd-network/test-ndisc-ra.c b/src/libsystemd-network/test-ndisc-ra.c
index d759ec03a8..7c59418ca7 100644
--- a/src/libsystemd-network/test-ndisc-ra.c
+++ b/src/libsystemd-network/test-ndisc-ra.c
@@ -284,7 +284,7 @@ static int radv_recv(sd_event_source *s, int fd, uint32_t revents, void *userdat
                 return 0;
         }
 
-        assert_se(sd_radv_stop(ra) >= 0);
+        assert_se(sd_radv_stop(ra, true) >= 0);
         test_stopped = true;
 
         return 0;