Re: [PATCH 1/6] property: add fwnode_property_read_string_index()

[Andy Shevchenko]

On Mon, Mar 21, 2022 at 08:49:21AM +0100, Clément Léger wrote:
> Le Fri, 18 Mar 2022 20:09:37 +0200,
> Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> a écrit :
> > On Fri, Mar 18, 2022 at 05:49:12PM +0100, Clément Léger wrote:
> > > Le Fri, 18 Mar 2022 18:26:00 +0200,
> > > Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> a écrit :  
> > > > On Fri, Mar 18, 2022 at 05:00:47PM +0100, Clément Léger wrote:  

...

> > > > > +	values = kcalloc(nval, sizeof(*values), GFP_KERNEL);
> > > > > +	if (!values)
> > > > > +		return -ENOMEM;
> > > > > +
> > > > > +	ret = fwnode_property_read_string_array(fwnode, propname, values, nval);
> > > > > +	if (ret < 0)
> > > > > +		goto out;
> > > > > +
> > > > > +	*string = values[index];
> > > > > +out:
> > > > > +	kfree(values);    
> > > > 
> > > > Here is UAF (use after free). How is it supposed to work?  
> > > 
> > > values is an array of pointers. I'm only retrieving a pointer out of
> > > it.  
> > 
> > I see, thanks for pointing out.
> > 
> > Nevertheless, I don't like the idea of allocating memory in this case.
> > Can we rather add a new callback that will provide us the necessary
> > property directly?
> > 
> 
> IMHO, it would indeed be better. However,
> fwnode_property_match_string() also allocates memory to do the same
> kind of operation. Would you also like a callback for this one ?

But matching string will need all of them to cover all possible cases.
So, it doesn't rely on the certain index and needs allocation anyway.

-- 
With Best Regards,
Andy Shevchenko