Hi Martin, Hi Patrick,

Attached a clean up for current conntrack stats, it applies to current 
bk 2.6 tree. Some comments:

a) I combined ifdef's checkings + macro, so if conntrack stats are 
compiled in the kernel we increase the counters.

 #define CONNTRACK_STAT_INC(count) 
(__get_cpu_var(ip_conntrack_stat).count++)
+#else /* !CONNTRACK STATS */
+#define CONNTRACK_STAT_INC(count)
+#endif

b) Martin, I've killed this to make them use the macro, are you ok with 
this? I know that we call twice smp_proccesor_id now, but it's clean I 
think.

-       /* use per_cpu() to avoid multiple calls to smp_processor_id() */
-       unsigned int cpu = smp_processor_id();

        MUST_BE_READ_LOCKED(&ip_conntrack_lock);
        list_for_each_entry(h, &ip_conntrack_hash[hash], list) {
                if (conntrack_tuple_cmp(h, tuple, ignored_conntrack)) {
-                       per_cpu(ip_conntrack_stat, cpu).found++;
+                       CONNTRACK_STAT_INC(found);
                        return h;

c) This is similar to a)

+#define CT_ADD_COUNTERS(ct, ctinfo, skb) ct_add_counters(ct, ctinfo, skb)
+#else
+#define CT_ADD_COUNTERS(ct, ctinfo, skb)
+#endif

d) You can see that I've play around with ip_conntrack_standalone 
reordering the creation of /proc entries. I did it just to remove 
warnings about unused labels when unsetting conntrack stats.

To conclude, this patch adds up to 8 ifdef's CONFIG_IP_NF_CT_ACCT 
checkings, I must confess that I don't love them, I'll think if there's 
a way to do cleaner.

Wait for your comments.

regards,
Pablo