Re: [PATCH] Avoid null-pointer access in w1/slaves/w1_therm

[David Fries]

On Sun, Mar 01, 2015 at 02:04:53PM +0100, Thorsten Bschorr wrote:
> Hi David,
> 
> thanks for your feedback on my first patch, I wasn't aware of checkpatch.pl.
> 
> Initially, I had just if-ed the usage of family-data, which did not
> look that nice. I was referring to this proof-of-concept workaround in
> my initial bug report.
> 
> The patch I've submitted is different from my proof-of-concept
> workaround. Not unlocking the bus before returning clearly is an
> error, I did not extensively test this patch.
> 
> 
> > or just increment it while sleeping, which is when it's needed, which
> > also looks simpler.
> >
> >                         if (external_power) {
> > +                               int refcnt;
> >                                 mutex_unlock(&dev->bus_mutex);
> >
> > +                               /* prevent the slave from going away */
> > +                               atomic_inc(&sl->refcnt);
> >                                 sleep_rem = msleep_interruptible(tm);
> > +                               refcnt = w1_unref_slave(sl);
> > -                               if (sleep_rem != 0)
> > +                               if (sleep_rem != 0 || !refcnt)
> >                                         return -EINTR;
> >
> >                                 i = mutex_lock_interruptible(&dev->bus_mutex);
> >                                 if (i != 0)
> >                                         return i;
> >                         } else if (!w1_strong_pullup) {
> 
> 
> I like this better than my workaround-patch.
> 
> One thought occurred to me when looking at this proposal: wouldn't it
> be even better to increase sl->refcnt before unlocking the mutex?
> I was asking myself if it is possible that the current thread gets
> suspended between mutex_unlock(&dev->bus_mutex); and
> atomic_inc(&sl->refcnt); thus leaving another thread the change to
> unref the device?
> (I'm not that familiar with linux scheduling, so my assumption might be void.)

You are correct, it would be a race condition if it doesn't increment
the refcnt before unlocking the mutex, and it should get the mutex
before unref.  Here's an updated version, I haven't even tried to
compile it.

What do you think Evgeniy?

			if (external_power) {
				int refcnt;
				/* prevent the slave from going away in sleep */
				atomic_inc(&sl->refcnt);
				mutex_unlock(&dev->bus_mutex);

				sleep_rem = msleep_interruptible(tm);
				if (sleep_rem != 0) {
					w1_unref_slave(sl);
					return -EINTR;
				}

				i = mutex_lock_interruptible(&dev->bus_mutex);
				refcnt = w1_unref_slave(sl);
				if (i != 0) {
					/* failed to lock */
					return i;
				}
				if (!refcnt)
					/* got lock, but slave went away */
					mutex_unlock(&dev->bus_mutex);
					return -EINTR;
				}
			} else if (!w1_strong_pullup) {


-- 
David Fries <david@xxxxxxxxx>    PGP pub CB1EE8F0
http://fries.net/~david/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/