Message-ID: <d585f719af13d7a7194e7cb734c5a7446954bf01.camel@gmail.com>
Date:   Wed, 10 Aug 2022 09:26:43 -0700
From:   James Prestwood <prestwoj@...il.com>
To:     Johannes Berg <johannes@...solutions.net>,
        Jakub Kicinski <kuba@...nel.org>
Cc:     netdev@...r.kernel.org
Subject: Re: [RFC 1/1] net: move IFF_LIVE_ADDR_CHANGE to public flag

Hi Johannes,

On Tue, 2022-08-09 at 21:04 +0200, Johannes Berg wrote:
> On Thu, 2022-08-04 at 12:49 -0700, James Prestwood wrote:
> > > > 
> > > > The semantics in wireless are also a little stretched because
> > > > normally
> > > > if the flag is not set the netdev will _refuse_ (-EBUSY) to
> > > > change
> > > > the
> > > > address while running, not do some crazy fw reset.
> > > 
> > > Sorry if I wasn't clear, but its not nl80211 doing the fw reset
> > > automatically. The wireless subsystem actually completely disallows
> > > a
> > > MAC change if the device is running, this flag isn't even checked.
> > > This
> > > means userspace has to bring the device down itself, then change
> > > the
> > > MAC.
> > > 
> > > I plan on also modifying mac80211 to first check this flag and
> > > allow
> > > a
> > > live MAC change if possible. But ultimately userspace still needs
> > > to
> > > be
> > > aware of the support.
> > > 
> 
> I'm not sure this is the right approach.
> 
> For the stated purpose (not powering down the NIC), with most mac80211
> drivers the following would work:
> 
>  - add a new virtual interface of any supported type, and bring it up
>  - bring down the other interface, change MAC address, bring it up
> again
>  - remove the interface added in step 1
> 
> though obviously that's not a good way to do it!
> 
> But internally in mac80211, there's a distinction between
> 
>  ->stop() to turn off the NIC, and
>  ->remove_interface() to remove the interface.
> 
> Changing the MAC address should always be possible when the interface
> doesn't exist in the driver (remove_interface), but without stop()ing
> the NIC.
> 
> However, obviously remove_interface() implies that you break the
> connection first, and obviously you cannot change the MAC address
> without breaking the connection (stopping AP, etc.)

> Therefore, the semantics of this flag don't make sense - you cannot
> change the MAC address in a "live" way while there's a connection, and
> at least internally you need not stop the NIC to change it. Since
> ethernet has no concept of a "connection" in the same way, things are
> different there.

There isn't a need for changing the MAC when connected/scanning, as
this doesn't make much sense. I guess "live" can be interpreted
differently, but my interpretation is simply changing the MAC when the
device isn't powered off. IFF_POWERED_ADDR_CHANGE, maybe, is better
suited.

> 
> Not sure how to really solve this - perhaps a wireless-specific way of
> changing the MAC address could be added, though that's quite ugly, or
> we
> might be able to permit changing the MAC address while not active in
> any
> way (connected, scanning etc.) by removing from/re-adding to the driver
> at least as far as mac80211 is concerned.

Ok, so this is how I originally did it in those old patches:

https://lore.kernel.org/linux-wireless/20190913195908.7871-2-prestwoj@gmail.com/

i.e. remove_interface, change the mac, add_interface. 

But before I revive those I want to make sure a flag can be advertised
to userspace e.g. NL80211_EXT_FEATURE_LIVE_ADDRESS_CHANGE. (or
POWERED). Since this was the reason the patches got dropped in the
first place.

> 
> johannes

Powered by blists - more mailing lists