Re: MC 7304 ipv4v6

[Bjørn Mork <bjorn mork no>]

Bjørn Mork <bjorn mork no> writes:
> Reinhard Speyerer <rspmn arcor de> writes:
>
> > On Sat, Nov 28, 2015 at 10:58:17PM +0100, Thomas Schäfer wrote:
> > > [...]
> > > I would be very happy if somebody tells me the steps for testing with PDP-
> > > context IPv4v6.
> >
> > Not sure whether this also applies to qmi-cli as I used a different
> > QMI client (which leaves network interface initialization to the
> > application) for a short dualstack test with Linux kernel 3.12.x and a
> > MC7304 but for me IPv6 only worked when explicitly assigning a
> > link-local address with e.g.
> > # ifconfig wwan<n> inet6 add fe80::1:2:3:4/64 up
> > when the network interface was set to raw IP mode instead of simply using
> > # ifconfig wwan<n> inet6 up
> > as no router solicitations were sent out otherwise.
>
> That's a very good point indeed!  I didn't think of the possibility of
> supporting SLAAC over raw-ip interfaces. I only did manual address
> config on IPv6 too. Thanks for verifying that the modem replies to RS
> over raw-ip.
>
> I'll see if I can figure out what it takes to automatically assign a
> link local address for these interfaces. I guess that should be done in
> any case for proper IPv6 support

hmm, "fixing" this seems simple enough: Just add the necessary code to
net/ipv6/addrconf.c for handling ARPHRD_NONE devices.  But I have two
questions:

1) will random addresses be a problem?  We might have to recreate the
 address every time the interface comes up, as we have nowhere to store
 it AFAICS.  So the link-local address will change whenever you toggle
 the device down/up.

2) what about other ARPHRD_NONE devices? This is currently 'tun', 'hso'
 and 'n_gsm'.  Will a default IPv6 link local address be a problem for
 any of these?

The only device type I know how to test is 'tun'.  And to me it looks
like an obvious winner there.  Why shouldn't tun devices do SLAAC and
support other IPv6 link local services by default?  But I don't normally
use such devices, so I know very little about real use cases...

If we can't add such generic ARPHRD_NONE code, then the alternatives I
can see are
 - defined an new ARPHRD_ type with about the same semantics, and add
   ipv6/addrconf code for it
 - do some driver hack - I believe it is possible to manually create an
   IPv6 device and assign an address from the driver.

I don't like either.  So I guess I will propose the ARPHRD_NONE code.


Bjørn