net: nexthop: reduce rcu synchronizations when replacing resilient groups

We can optimize resilient nexthop group replaces by reducing the number of
synchronize_net calls. After commit 1005f19b ("net: nexthop: release
IPv6 per-cpu dsts when replacing a nexthop group") we always do a
synchronize_net because we must ensure no new dsts can be created for the
replaced group's removed nexthops, but we already did that when replacing
resilient groups, so if we always call synchronize_net after any group
type replacement we'll take care of both cases and reduce synchronize_net
calls for resilient groups.
Suggested-by: Ido Schimmel <idosch@idosch.org>
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Tested-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/nexthop.c

@@ -1918,9 +1918,6 @@ static void nh_rt_cache_flush(struct net *net, struct nexthop *nh,
 	if (!replaced_nh->is_group)
 		return;
 
-	/* new dsts must use only the new nexthop group */
-	synchronize_net();
-
 	nhg = rtnl_dereference(replaced_nh->nh_grp);
 	for (i = 0; i < nhg->num_nh; i++) {
 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
@@ -2002,9 +1999,10 @@ static int replace_nexthop_grp(struct net *net, struct nexthop *old,
 
 	rcu_assign_pointer(old->nh_grp, newg);
 
-	if (newg->resilient) {
 	/* Make sure concurrent readers are not using 'oldg' anymore. */
 	synchronize_net();
+
+	if (newg->resilient) {
 		rcu_assign_pointer(oldg->res_table, tmp_table);
 		rcu_assign_pointer(oldg->spare->res_table, tmp_table);
 	}