testenv: Don't read interface MAC addresses until we need them

There is a slight delay after the interface is created where the MAC
address can change. This will lead to the wrong neighbour entry being
installed, so defer reading the MAC address until just before we need it.
This seems to be enough to avoid the race.

Fixes #159 (again).

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>

Author: tohojo

testenv/testenv.sh

@@ -240,14 +240,12 @@ setup()
     ip netns add "$NS"
     ip link add dev "$NS" type veth peer name veth0 netns "$NS"
 
-    OUTSIDE_MAC=$(iface_macaddr "$NS")
-    INSIDE_MAC=$(iface_macaddr veth0 "$NS")
-
     set_sysctls $NS
     ip link set dev "$NS" up
     ip addr add dev "$NS" "${OUTSIDE_IP6}/${IP6_PREFIX_SIZE}"
     ethtool -K "$NS" rxvlan off txvlan off
     # Prevent neighbour queries on the link
+    INSIDE_MAC=$(iface_macaddr veth0 "$NS")
     ip neigh add "$INSIDE_IP6" lladdr "$INSIDE_MAC" dev "$NS" nud permanent
 
     set_sysctls veth0 "$NS"
@@ -256,6 +254,7 @@ setup()
     ip -n "$NS" addr add dev veth0 "${INSIDE_IP6}/${IP6_PREFIX_SIZE}"
     ip netns exec "$NS" ethtool -K veth0 rxvlan off txvlan off
     # Prevent neighbour queries on the link
+    OUTSIDE_MAC=$(iface_macaddr "$NS")
     ip -n "$NS" neigh add "$OUTSIDE_IP6" lladdr "$OUTSIDE_MAC" dev veth0 nud permanent
     # Add route for whole test subnet, to make it easier to communicate between
     # namespaces