> > > I'd say for connected and bound sockets, getsockname() would return
the
> > > correct topological source address.
> >
> > But that'd break the getsockname() function, which should return the
local
> > binding of the socket. If the socket is bound to a home address, that
> > address
> > should be returned by the function, even though it might be also
> > (Mobile IP-)bound to a care-of address.
>
> In order to bind() to a address it generally requires that the address
> is locally configured.

Sure. Both a home address and a care-of address can be locally configured
at the same time. (regardless of one being on a logical interface, other
on a physical interface).

> I assumed that the suggestion was that the MN would
> always have a home-address configured as a logical interface or some
pseudo
> interface. It seems a bit odd for a socket API to return a fake address
> while it's bound to a different address.

This part, I didn't get it. A socket can be bound to a home address, or
a care-of address. Either case, socket API should treat them equally.
Getsockname() should return the IP address that local endpoint is
bound to, regardless of the type (i.e., home address vs care-of address,
configured on physical vs. logical interface).

>
> That's why it's clean to direct the kernel to swtich the source address
> selection mechanism  from the default to the desired.
>

OK, this is about changing the address selection policy per application.
This is about controlling the behavior of the kernel, not reading state
from it.

alper

--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to