Re: [PATCH v4 4/6] hw_random: fix unregister race.
From: Amos Kong
Date: Fri Dec 05 2014 - 22:52:06 EST
On Wed, Nov 12, 2014 at 02:33:00PM +1030, Rusty Russell wrote:
> Amos Kong <akong@xxxxxxxxxx> writes:
> > From: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
> > The previous patch added one potential problem: we can still be
> > reading from a hwrng when it's unregistered. Add a wait for zero
> > in the hwrng_unregister path.
> > v4: add cleanup_done flag to insure that cleanup is done
> That's a bit weird. The usual pattern would be to hold a reference
> until we're actually finished, but this reference is a bit weird.
> We hold the mutex across cleanup, so we could grab that but we have to
> take care sleeping inside wait_event, otherwise Peter will have to fix
> my code again :)
> AFAICT the wake_woken() stuff isn't merged yet, so your patch will
> have to do for now.
> > @@ -98,6 +99,8 @@ static inline void cleanup_rng(struct kref *kref)
> > if (rng->cleanup)
> > rng->cleanup(rng);
> > + rng->cleanup_done = true;
> > + wake_up_all(&rng_done);
> > }
> > static void set_current_rng(struct hwrng *rng)
> > @@ -536,6 +539,11 @@ void hwrng_unregister(struct hwrng *rng)
> > kthread_stop(hwrng_fill);
> > } else
> > mutex_unlock(&rng_mutex);
> > +
> > + /* Just in case rng is reading right now, wait. */
> > + wait_event(rng_done, rng->cleanup_done &&
> > + atomic_read(&rng->ref.refcount) == 0);
> > +
> The atomic_read() isn't necessary here.
At least, we need it to convert refcount from atomic_t to int.
Otherwise, we will touch this error:
error: invalid operands to binary == (have 'atomic_t' and 'int')
> However, you should probably init cleanup_done in hwrng_register().
> (Probably noone does unregister then register, but let's be clear).
Description: Digital signature