MP-BGP for IPv4 and IPv6

Hi,

it is specified on MP-BGP that any kind of route (both IPv4 and IPv6)
can be advertised through a TCPv4 connection
and ditto through a TCPv6 connection. But in case we have the following
topology:

________________                                      ________________
|peer 1                     |
|peer2                     |
|144.20.20.1
|-------------------------|144.20.20.2            |
|3ffe:400:100::1        |   TCPv4
|3ffe:400:100::2        |
|dual stack               |                                       |dual
stack               |
|_______________|
|_______________|

An open message will be sent with both IPv4 and IPv6 capabilities.
But when peer 1 receives an IPv6 route from peer 2 and has to advertise
this route to other peers,
it does not have the IPv6 address of peer 2 since the connection was
made in IPv4: so no way to fill up the NH field
of the MP_reach_NLRI in the update message.
The BGP tunnel draft specifies the use of mapped addresses in this case
(::ffff:144.20.20.2) but IPv6 packets will
always be encapsulated in IPv4 to reach peer 2 even if there is an IPv6
path. Moreover, the BGP tunnel solution do not handle
the case of sending IPv4 routes over a TCPv6 connection.

I saw nothing about this issue.
I suggest to give by configuration an IPv6 address of peer 2 when
configuring a TCPv4 connection on peer 1 towards peer 2.
Does it seem a good solution ? Is there another way ?

The best solution would be to only allow advertising IPv4 routes on
TCPv4 connections but that is not what is specified...

Kind Regards,

Christophe.