Re: [PATCH v3 1/2] cfg80211: fix regulatory restore upon user hints

["Luis R. Rodriguez" <lrodriguez@atheros.com>]

On Fri, Apr 1, 2011 at 2:49 PM, Johannes Berg <johannes@sipsolutions.net> wrote:
> On Fri, 2011-04-01 at 14:47 -0700, Luis R. Rodriguez wrote:
>> On Fri, Apr 1, 2011 at 1:28 PM, Johannes Berg <johannes@sipsolutions.net> wrote:
>> > On Fri, 2011-04-01 at 13:10 -0700, Luis R. Rodriguez wrote:
>> >>
>> >> The order should not matter except that we want the queue to be
>> >> cleared before processing core hints when doing restoration, otherwise
>> >> the next user hint in the queue can be bogus and it will prevent a
>> >> restore.
>> >
>> > I'm just thinking this temporary clearing could cause us to reject a
>> > reply from CRDA that's coming in at the same time that is due to a user
>> > request.
>>
>> Ah, I see, hmm well I tested this with:
>>
>> iw reg set US; iw reg set XA; iw reg set XB; iw reg set CR; iw dev
>> wlan0 disconnect; iw reg set FR;
>
> was CRDA running though?

Yes

> If so, it will probably reply right away

Except for the cases that are bogus: XA, XB

> -- I  don't think you can hit the race easily.

I'm just trying to see the entry points on the code for such a race,
apart from the points I mentioned I cannot see any other case. Can
you?

>> and things were still being processed in the order sent. There is a
>> small race between the unlock of the reg_mutex on
>> restore_regulatory_settings() down until where we lock the
>> regulatory_hint_core() and regulatory_hint_user(). The chances of a
>> user regulatory hint coming in between is very low, I could not do it.
>> Then there is also a small race of getting a user reg hint on
>> restore_regulatory_settings() in between the regulatory_hint_user()
>> and the lock of the reg_mutex again for processing old regulatory
>> hints, but the worst that can happen in that case is we squeeze in a
>> user regulatory hint in before the other older regulatory hints, and
>> this is fine.
>>
>> I cannot see any other cases here where we'd block a request from userspace.
>
> No not block a request. I'm just thinking that this taking things off
> the list temporarily might erroneously discard a crda response.

Ah, but the stuff in the queue is stuff which we have not yet sent out
to userspace. CRDA and the kernel process regulatory requests
atomically.

  Luis