Lorenzo Colitti [off-list ref] wrote:

quoted

While the former might seem preferable, Pablo pointed out that there
are more xt modules with link-time dependeny issues regarding ipv6,
so lets go for 2).

I had to do this recently for the ping socket as well:

http://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git/commit/?id=6d0bfe22611602f36617bc7aa2ffa1bbb2f54c67

+/* Compatibility glue so we can support IPv6 when it's compiled as a module */
+struct pingv6_ops {
[...]
+ int (*ipv6_chk_addr)(struct net *net, const struct in6_addr *addr,
+     struct net_device *dev, int strict);
+};

Is it a better idea to share these structures and have just one
structure containing all IPv6 dummy functions?

I think so, yes.

If it was in an include
file, it would be easily accessible to most of the tree even when
CONFIG_IPV6={n,m}, and we could have the ipv6 module init (and exit)
code just set all the function pointers. That way, we wouldn't have to
reinvent this particular wheel in multiple places of the code.

FWIW, I agree.  We should avoid having multiple copies of this.