Re: [PATCH net-next] net: ipconfig: allow IP-Config over DSA devices

2014-06-05 0:01 GMT-07:00 David Miller [off-list ref]:

From: Florian Fainelli <f.fainelli@gmail.com>
Date: Fri, 30 May 2014 11:08:45 -0700

quoted

Let me know your thoughts. If you feel like something like:
netdev_is_upper_dev() or something like that is better.

I'm not happy with this change for several reasons.

First, I don't like the idea that ipconfig works for some stacked
devices and not for others.

Right, although as I described DSA devices are likely to be the only
real kernel-only created devices here, other stacked devices such as
bonds, vlans and tunnels for instance do require either kernel
modifications to be created by the kernel, or user-space.

 I'd rather that we simply accept that
direct devices are the only thing supported.

Would you be willing to revise this position if we can come up with a
generic infrastructure for telling whether a network device depends on
another one and its master device needs to be opened first? Should we
instead force the slave devices in net/dsa/slave.c to open its master
device if it is not UP?

Secondly, the logic in that ipconfig loop is so confusing.  I can't
even figure out what that dsa pointer test is trying to really
accomplish in the second hunk of your patch.

I'd like to defer this for now, sorry.

Thanks for your comments.
-- 
Florian