Re: [PATCH] net: rtnetlink: be more strict when setting MAC address

[Phil Sutter <phil@nwl.cc>]

On Mon, Aug 17, 2015 at 02:09:33PM -0700, David Miller wrote:
> From: Phil Sutter <phil@nwl.cc>
> Date: Mon, 17 Aug 2015 23:06:47 +0200
> 
> > Upon evaluation of IFLA_ADDRESS and IFLA_BROADCAST messages, make sure
> > the passed argument length matches dev->addr_len exactly.
> > 
> > This fixes dubious behaviour of 'ip link set eth0 addr <MAC>' where
> > '<MAC>' is too long, e.g. '00:11:22:33:44:55:66:77'. Called like this,
> > 'ip' would return successfully and the kernel sets eth0's MAC address to
> > the leading six octets of the passed argument.
> > 
> > Signed-off-by: Phil Sutter <phil@nwl.cc>
> 
> I don't think this behavior is very "dubious" and making the check more
> strict risks breaking things that have worked for a very long time.
> 
> I'm not applying this, sorry.

No problem. I was already afraid of this breaking something as commit
1840bb1, which introduces validate_linkmsg(), already mentions something
in that direction. Therefore I appreciate your feedback clarifying this
for me, although it is not quite the outcome I had wished for.

Seeing that the kernel checks the minimum length already, shifting the
maximum length check to kernel space felt like a natural choice.

Nevertheless, I think iproute2 should not behave this way. Stephen, do
you think it is reasonable to add a similar logic to iplink_parse() as
done in do_set() (i.e., finding out which is the correct length of LL
address for a given interface and validating input based on this)? If
so, is it appropriate to recycle get_address() function or should the
information come from netlink?

Thanks, Phil