Re: [IPv6] interface-local multicast escapes the local node

[YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>]

Hannes Frederic Sowa wrote:
> On Sat, Feb 09, 2013 at 11:12:46PM +0900, YOSHIFUJI Hideaki wrote:
>> Hi,
>>
>> Hannes Frederic Sowa wrote:
>>> On Wed, Feb 06, 2013 at 05:54:15PM +0100, Hannes Frederic Sowa wrote:
>>>> On Thu, Feb 07, 2013 at 12:24:14AM +0900, YOSHIFUJI Hideaki wrote:
>>>>> NAK.  I think we should select routes via loopback device here.
>>>>
>>>> Will try your idea, thanks.
>>>
>>> Does this patch look reasonable? Btw. i am pleased to see this kind of
>>> things work out as expected most of the time (addrtype checking etc. all
>>> in place). :)
>>>
>>
>> Well, I rethink of what "interface-local" means.
>>
>> It seems applications will join ff01::/16%eth0 instead of ff01::/16%lo.
>> If so, your original patch seems better.  My bad, sorry.
> 
> I was looking at getpeername et. al. where we should report the scope
> back to the user. A common pattern is:
> 
>                         if (ipv6_addr_type(&sin->sin6_addr) & IPV6_ADDR_LINKLOCAL)
>                                 sin->sin6_scope_id = IP6CB(skb)->iif;
> 
> I propose to introduce something like 'bool ipv6_addr_intf_scoped(in6_addr)'
> and let it check for ll addresses and interface scoped addresses.

Hmm, maybe, we might want to say:

__u32 __ipv6_iface_scope_id(int type, unsigned int iface)
{
	if (type == IPV6_ADDR_ANY ||
	    type & IPV6_ADDR_LOOPBACK ||
	    __ipv6_addr_src_scope(type) > IPV6_ADDR_SCOPE_LINKLOCAL)
		return 0;
	return iface;
}

__u32 ipv6_iface_scope_id(const struct in6_addr *addr, unsigned int iface)
{
	return __ipv6_iface_scope_id(__ipv6_addr_type(addr), iface);
}

And then,
	sin->sin6_scope_id = __ipv6_iface_scope_id(__ipv6_addr_type(&sin->sin6_addr),
						   IP6CB(skb)->iif);
or
	sin->sin6_scope_id = ipv6_iface_scope_id(&sin->sin6_addr, IP6CB(skb)->iif);

--yoshfuji