Re: [PATCH] pinctrl: devicetree: fix refcount leak in pinctrl_dt_to_map()

[Andy Shevchenko]

On Wed, Apr 17, 2024 at 08:49:42PM +0300, Dan Carpenter wrote:
> On Wed, Apr 17, 2024 at 08:12:23PM +0300, Andy Shevchenko wrote:
> > On Wed, Apr 17, 2024 at 06:38:46PM +0300, Dan Carpenter wrote:
> > > On Wed, Apr 17, 2024 at 06:30:59PM +0300, Andy Shevchenko wrote:
> > > > On Mon, Apr 15, 2024 at 06:53:28PM +0800, Zeng Heng wrote:

..

> > > > >  	for (state = 0; ; state++) {
> > > > >  		/* Retrieve the pinctrl-* property */
> > > > >  		propname = kasprintf(GFP_KERNEL, "pinctrl-%d", state);
> > > > > -		if (!propname)
> > > > > -			return -ENOMEM;
> > > > > +		if (!propname) {
> > > > > +			ret = -ENOMEM;
> > > > > +			goto err;
> > > > > +		}
> > > > >  		prop = of_find_property(np, propname, &size);
> > > > >  		kfree(propname);
> > > > >  		if (!prop) {
> > > > >  			if (state == 0) {
> > > > > -				of_node_put(np);
> > > > > -				return -ENODEV;
> > > > > +				ret = -ENODEV;
> > > > > +				goto err;
> > > > 
> > > > Has it been tested? How on earth is this a correct change?
> > > > 
> > > > We iterate over state numbers until we have properties available. This chunk is
> > > > _successful_ exit path, we may not free parsed maps! Am I wrong?
> > > 
> > > In this path state == 0 so we haven't had a successful iteration yet.
> > 
> > Ah, indeed, this is not a status. Okay, makes sense, but calling that free
> > function for the purpose of the putting of_node seems an overkill...
> 
> Sure, that's one way to look at it, but it's suspicious looking when
> there is a direct return which is surrounded by gotos.  As I write this,
> I remember that Smatch has a warning for code like that.
> 
> Probably we should add a comment to say:
> 
> 	/* Return -ENODEV if the property 'pinctrl-0' is not present. */

Good idea, go for it!

-- 
With Best Regards,
Andy Shevchenko