Re: [PATCH 1/3] livepatch: introduce shadow variable API

[Petr Mladek]

On Wed 2017-06-14 08:17:44, Josh Poimboeuf wrote:
> On Wed, Jun 14, 2017 at 02:59:13PM +0200, Petr Mladek wrote:
> > > +}
> > > +
> > > +void klp_shadow_detach(void *obj, char *var)
> > > +{
> > > +	unsigned long flags;
> > > +	struct klp_shadow *shadow;
> > > +
> > > +	spin_lock_irqsave(&klp_shadow_lock, flags);
> > > +
> > > +	hash_for_each_possible(klp_shadow_hash, shadow, node,
> > > +			       (unsigned long)obj) {
> > > +		if (shadow->obj == obj && !strcmp(shadow->var, var)) {
> > 
> > Do we need to test "shadow->obj == obj" here? If it is not true,
> > there would be a bug in the hashtable implementation or in
> > klp_shadow_attach().
> > 
> > Well, it might make sense to add a consistency check:
> > 
> >       WARN_ON(shadow->obj != obj);
> > 
> 
> It would make sense if hash_for_each_possible() worked that way, but for
> some reason it doesn't. :-/  It gives you all the hash collisions.

I see. Shame on me. The original code makes perfect sense then.

Best Regards,
Petr