* Jean Tourrilhes <20040910201302.GA16556@xxxxxxxxxxxxxxxxxx> 2004-09-10 13:13
> On Fri, Sep 10, 2004 at 10:06:44PM +0200, Thomas Graf wrote:
> > * Jean Tourrilhes <20040910195003.GA13912@xxxxxxxxxxxxxxxxxx> 2004-09-10
> > 12:50
> > > Thomas Graf wrote :
> > > >
> > > > Allows changing of device name via rtnetlink. Last bit needed to do full
> > > > link configuration via rtnetlink.
> > > >
> > > > Signed-off-by: Thomas Graf <tgraf@xxxxxxx>
> > >
> > > This does not work, because you don't return the new name to
> > > user space. If the new name is a pattern, such as "eth%d" or "wlan%d",
> > > you absolutely need to return the new instanciated device name to user
> > > space so that userspace doesn't loose track of the device.
> >
> > The ifindex stays the same, therefore the user space application can
> > simply dump the link list and fetch the new interface name.
>
> It's so simple to return the new name, so why not do it ?
> There is no need to make applications more complex.
Oh, I didn't realize that, can you enlighten me?
> > It would
> > theoretically be possible to provide the new name via an ACK but
> > this would break the RFC.
>
> What do you mean, break the RFC ?
I can think of 2 ways implementing this:
1) Send an ACK and change IFLA_IFNAME in the original message. However,
RFC 3549 specifies it to be the old message so user space
applications can directly reuse it (forward, redirect, log, ...)
2) Implement NETDEV_CHANGE, I was actually about to implement this
but it's not directly related to this change and requires the same
effort by the user space application has refetching the link list.
|