netfilter: conntrack: don't cache nlattr_tuple_size result in nla_size

We currently call ->nlattr_tuple_size() once at register time and
cache result in l4proto->nla_size.

nla_size is the only member that is written to, avoiding this would
allow to make l4proto trackers const.

We can use ->nlattr_tuple_size() at run time, and cache result in
the individual trackers instead.

This is an intermediate step, next patch removes nlattr_size()
callback and computes size at compile time, then removes nla_size.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

net/ipv6/netfilter/nf_conntrack_proto_icmpv6.c

@@ -259,9 +259,14 @@ static int icmpv6_nlattr_to_tuple(struct nlattr *tb[],
 	return 0;
 }
 
-static int icmpv6_nlattr_tuple_size(void)
+static unsigned int icmpv6_nlattr_tuple_size(void)
 {
-	return nla_policy_len(icmpv6_nla_policy, CTA_PROTO_MAX + 1);
+	static unsigned int size __read_mostly;
+
+	if (!size)
+		size = nla_policy_len(icmpv6_nla_policy, CTA_PROTO_MAX + 1);
+
+	return size;
 }
 #endif