Re: [RFC PATCH] rt2800lib: unconditionally enable MFP

[Stanislaw Gruszka <stf_xl@wp.pl>]

Hi

On Sun, May 24, 2020 at 09:42:51PM +1000, Julian Calaby wrote:
> Hi Stanislaw,
> 
> On Sun, May 24, 2020 at 9:27 PM Stanislaw Gruszka <stf_xl@wp.pl> wrote:
> >
> > On Sun, May 24, 2020 at 10:47:31AM +0100, Rui Salvaterra wrote:
> > > According to Larry [1] (and successfully verified on b43) mac80211
> > > transparently falls back to software for unsupported hardware cyphers. Thus,
> > > there's no reason for not unconditionally enabling MFP. This gives us WPA3
> > > support out of the box, without having to manually disable hardware crypto.
> > >
> > > Tested on an RT2790-based Wi-Fi card.
> > >
> > > [1] https://lore.kernel.org/linux-wireless/8252e6a1-b83c-64eb-2503-2686374216ae@lwfinger.net/
> >
> > AFICT more work need to be done to support MFP by HW encryption properly
> > on rt2x00. See this message and whole thread:
> > https://lore.kernel.org/linux-wireless/977a3cf4-3ec5-4aaa-b3d4-eea2e8593652@nbd.name/
> 
> Am I reading this right: rt2x00 offloads some of the processing to the
> card which interferes with MFP when using software encryption, so
> therefore we need to disable that offload when using software
> encryption with MFP.

Yes.

We offload encryption to HW based on cipher. Modern ciphers like 
GCMP, BIP_GMAC, etc, are not supported by rt2x00 HW. In such case
rt2x00mac_set_key() will return -EOPNOTSUPP and all encryption will
be done by mac80211 - MFP will work just fine.

But MFP can still be used with CCMP cipher, which we offload to HW,
and that would create problems described by Felix.

> So if mac80211 knows that this offload is happening and that we can't
> use hardware crypto for MFP, could it be smart enough to disable the
> offload itself?
> 
> And once mac80211 is smart enough to make those decisions, couldn't we
> just enable MFP by default?

If we will have indicator from mac80211 that MFP is configured, we can
just return -EOPNOTSUPP from rt2x00mac_set_key() for CCMP and that will
make MFP work without specifying nohwcrypt module parameter - software
encryption will be used anyway.

Optimal solution though would be implement similar code like in mt76, so
we will have HW encryption for MFP+CCMP, but this is not trivial, and
I think handling encryption fully in software is ok.

Stanislaw