[PATCH v2 1/3] landlock: Fix handling of disconnected directories

[Mickaël Salaün]

On Mon, Jul 14, 2025 at 01:39:12PM +0100, Tingmao Wang wrote:
> On 7/11/25 20:19, Mickaël Salaün wrote:
> > [...]
> > @@ -800,6 +802,8 @@ static bool is_access_to_paths_allowed(
> >  		access_masked_parent1 = access_masked_parent2 =
> >  			landlock_union_access_masks(domain).fs;
> >  		is_dom_check = true;
> > +		memcpy(&_layer_masks_parent2_bkp, layer_masks_parent2,
> > +		       sizeof(_layer_masks_parent2_bkp));
> >  	} else {
> >  		if (WARN_ON_ONCE(dentry_child1 || dentry_child2))
> >  			return false;
> > @@ -807,6 +811,8 @@ static bool is_access_to_paths_allowed(
> >  		access_masked_parent1 = access_request_parent1;
> >  		access_masked_parent2 = access_request_parent2;
> >  		is_dom_check = false;
> > +		memcpy(&_layer_masks_parent1_bkp, layer_masks_parent1,
> > +		       sizeof(_layer_masks_parent1_bkp));
> 
> Is this memcpy meant to be in this else branch?  If parent2 is set, we
> will leave _layer_masks_parent1_bkp uninitialized right?

Good catch!  We should always initialize _layer_masks_parent1_bkp.
I'll move both _layer_masks_parent*_bkp initializations just before the
related allowed_parent* initialization.

> 
> >  	}
> >  
> >  	if (unlikely(dentry_child1)) {