On Sat, May 16, 2020 at 01:55:48PM -0700, David Miller wrote:
> From: Christoph Hellwig <h...@lst.de>
> Date: Fri, 15 May 2020 08:33:24 +0200
>
> > My initial plan was to add a ->tunnel_ctl method to the net_device_ops,
> > and lift the copy_{to,from}_user for SIOCADDTUNNEL, SIOCCHGTUNNEL,
> > SIOCDELTUNNEL and maybe SIOCGETTUNNEL to net/socket.c. But that turned
> > out to have two problems:
> >
> > - first these ioctls names use SIOCDEVPRIVATE range, that can also
> > be implemented by other drivers
> > - the ip_tunnel_parm struture is only used by the ipv4 tunneling
> > drivers (including sit), the "real" ipv6 tunnels use a
> > ip6_tnl_parm or ip6_tnl_parm structure instead
>
> Yes, this is the core of the problem, the user provided data's type
> is unknown until we are very deep in the call chains.
>
> I wonder if there is some clever way to propagate this size value
> "up"?
As far as I can tell the only information vectors is the net_device
structure or its op vector. But even then we have the problem that
other devices use the SIOCDEVPRIVATE range for something else.
I'll look into implenenting the tunnel_ctl method just for kernel
callers (plus maybe a generic helper for the ioctl), and we'll see if
you like that better.
