> Since the kernel already has all of the information needed to build a
> link-layer header, the destination address passed down with
> DL_UNITDATA_REQ is ignored. I can't think of a compelling reason to
> require that the addressed passed down by the application be the same as
> the destination address configured on the tunnel, as that just poses
> unnecessary restrictions on applications (they need to first find out
> what that address is, when there isn't really a need for them to know).
>
> > Minimally, won't 6to4
> > tunnels still require an explicit tunnel destination to be specified?
>
> No, the idea behind automatic tunneling is that the outer destination
> address is automatically generated for each outgoing packet based on the
> inner packet contents (for 6to4, this is based on the inner IPv6
> destination address). For any given IPv6 packet transmitted to a 6to4
> tunnel link, there is only one correct outer IPv4 destination. The
> destination passed down along with the message is therefore ignored.
So, as part of the tunneling putback, should we update the dlpi_send()
manpage to indicate that `daddrp' is ignored for tunnel datalinks? (Seems
like we have some other libdlpi-related work to do, such as the
DL_CURR_DEST_ADDR[1] support we talked about a few weeks back.)
[1] A comment in dld_str.c mentions DL_NOTE_DEST_ADDR instead of
DL_CURR_DEST_ADDR.
--
meem
