Hi Sascha,
Sorry for being a bit slow here. And even now, I probably don't have enough time to review this whole series today.
But I'll still share some initial thoughts, in case you can help address them before I next look at this:
On Wed, Sep 18, 2024 at 01:10:27PM +0200, Sascha Hauer wrote:
The mwifiex driver tries to derive the MAC addresses of the virtual interfaces from the permanent address by adding the bss_num of the particular interface used. It does so each time the virtual interface is changed from AP to station or the other way round. This means that the devices MAC address changes during a change_virtual_intf call which is pretty unexpected by userspace.
Ack, the "change_virtual_intf" part looks wrong.
Furthermore the driver doesn't use the permanent address to add the bss_num to, but instead the current MAC address increases each time we do a change_virtual_intf.
Fix this by initializing the MAC address once from the permanent MAC address during creation of the virtual interface and never touch it again. This also means that userspace can set a different MAC address which then stays like this forever and is not unexpectedly changed by the driver.
It is not clear how many (if any) MAC addresses after the permanent MAC address are reserved for a device, so set the locally admistered bit for all MAC addresses derived from the permanent address.
I think I'm generally supportive of the direction this changes things, but I'm a bit hesitant about two things: 1. the potential user-visible changes and 2. the linux-stable backport (Cc stable below)
For 1: MAC addresses are important in some contexts, and this might significantly change the addresses that devices get in practice. Such users might not really care about the weird details of when the address incremented; but they *probably* care that a certain sequence of "boot device; run hostapd with <foo> config file" produces the same address.
Also, I'm not sure I know enough of the implications of potential over-use of the locally administered bit. Are there significant downsides to it (aside from the simple fact that it's a different address)?
And I see that you rightly don't know how many addresses are actually reserved, but I have an educated guess that it's not just 1. For one, this driver used to default-create 3 interfaces: 1211c961170c mwifiex: do not create AP and P2P interfaces upon driver loading
and when we stopped doing that, we still kept support for a module parameter for the old way: 0013c7cebed6 mwifiex: module load parameter for interface creation
Perhaps these "initial" interfaces should at least be allowed permanent addresses?
So anyway, I don't really know for sure the right answer, but I want to log my concerns, in case you had more thoughts on backward compatibility.
And given all the uncertainty above, I'm extra hesitant about backporting likely-user-visible changes to stable (#2).
With this patch MWIFIEX_BSS_TYPE_ANY becomes unused, so it's removed.
Signed-off-by: Sascha Hauer s.hauer@pengutronix.de Cc: stable@vger.kernel.org
Regards, Brian