[PATCH 2/2] iproute2: add support for link set

[Marek Behún]

On Sat, 22 Jan 2022 09:38:01 +0000
Rui Salvaterra <rsalvaterra at gmail.com> wrote:

> Hi, Ansuel,
> 
> On Sat, 22 Jan 2022 at 01:08, Ansuel Smith <ansuelsmth at gmail.com> wrote:
> >
> >
> > So.. how should we proceed with this? From what I understand the idea is to merge this ASAP.  
> 
> But not a moment sooner. ;) I'm sure we agree that this patch won't be
> merged upstream in its current form, according to the comments
> received, and the less we diverge from upstream, the better for
> maintenance.
> 
> > Think we have to change this with the DSA specific attribute.  
> 
> Ok, let's step back and take a look at our possibilities. Stephen
> Hemminger suggested auditing all kernel usage of IFLA_LINK and adding
> checks where needed to make sure the current users don't break [1].
> This would certainly work, but that would mean sprinkling error checks
> in possibly quite a number of places [2]. Vladimir Oltean, instead,
> suggested creating a new netlink attribute for this specific purpose
> [3] (let's call it IFLA_CPU, for example). I believe the latter is the
> less intrusive of the options, with the added bonus of not having to
> overload IFLA_LINK with different semantics (something I personally
> dislike). I would also rename "link" to "cpu" in the ip patch
> (avoiding the overload, once again).
> 
> [1] https://lore.kernel.org/netdev/20210411100411.6d16e51d@hermes.local/
> [2] https://elixir.bootlin.com/linux/latest/A/ident/IFLA_LINK
> [3] https://lore.kernel.org/netdev/20210411170939.cxmva5vdcpqu4bmi@skbuf/

Hello guys,

I am pessimistic about this being resolved soon in upstream, so my
suggestion to you for OpenWRT is to do something that can be used now,
for example a sysfs attribute, and create an utility for changing port's
CPU port.

Then if things get solved in upstream, you can just change the
utility's code.

Marek