Re: [PATCH] cfg80211: add get reg command

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

On Wed, Jan 28, 2009 at 06:50:34AM -0800, Johannes Berg wrote:
> On Wed, 2009-01-28 at 06:44 -0800, Luis R. Rodriguez wrote:
> > On Wed, Jan 28, 2009 at 02:17:21AM -0800, Johannes Berg wrote:
> > > On Tue, 2009-01-27 at 18:52 -0800, Luis R. Rodriguez wrote:
> > > 
> > > > +	mutex_lock(&cfg80211_drv_mutex);
> > > > +	regd = reg_get_current_rd();
> > > > +	mutex_unlock(&cfg80211_drv_mutex);
> > > > +
> > > > +	if (!regd)
> > > > +		goto out;
> > > 
> > > Any reason you need to copy and can't just build the whole thing under
> > > lock?
> > 
> > Glad you asked, we can lock the entire call or we can do the above to
> > copy. I picked a copy to not let a regular user (get reg can be called
> > by anyone) abuse the lock as cfg80211_drv_mutex is used in a lot of places,
> > I guess its silly but other's input on this is appreciated as well. Not sure
> > if the copy was the best choice.
> 
> I don't really see much difference between locking to create a copy and
> locking to create a copy in a different format, do you? 
> Except to create
> a copy needs to allocate memory which could take even longer.

Agreed, that's the thing I was not thinking about those two paths I was
thinking of either using a copy of cfg80211_regdomain for nl80211.c to move
forward without having to keep cfg80211_drv_mutex *or* having
reg_get_current_rd() just return cfg80211_regdomain and have the reg_reg
lock during its entire access.

I am honestly not sure which is the best approach but I picked a copy
since I figure having a user do a kmalloc() is less harmless then contending
the lock more time if he wanted to abuse the system. I also thought about
using a static regd in place just to copy the cfg80211_regdomain but then
that would mean having a large static regd with max num rules and I didn't
like that either. Seemed to me a kmalloc() was in the middle. Let me know
if you guys have better advice.

  Luis