Re: [PATCH v3] leds: Introduce userspace leds driver

[David Lechner]

On 09/15/2016 09:54 AM, Jacek Anaszewski wrote:
> Hi Pavel,
>
> On 09/15/2016 03:08 PM, Pavel Machek wrote:
> > Hi!
> >
> > > > +    if (copy_from_user(&udev->user_dev, buffer,
> > > > +               sizeof(struct uleds_user_dev))) {
> > > > +        ret = -EFAULT;
> > > > +        goto out;
> > > > +    }
> > > > +
> > > > +    if (!udev->user_dev.name[0]) {
> > > > +        ret = -EINVAL;
> > > > +        goto out;
> > > > +    }
> > > > +
> > > > +    ret = led_classdev_register(NULL, &udev->led_cdev);
> > > > +    if (ret < 0)
> > > > +        goto out;
> >
> > No sanity checking on the name -> probably a security hole. Do not
> > push this upstream before this is fixed.

If this is a serious security issue, then you should also raise an issue 
with input maintainers because this is the extent of sanity checking for 
uinput device names as well.

I must confess that I am no security expert, so unless you can give 
specific examples of what potential threats are, I will not be able to 
guess what I need to do to fix it.

After some digging around the kernel, I don't see many instances of 
validating device node names. The best I have found so far comes from 
create_entry() in binfmt_misc.c

	if (!e->name[0] ||
	    !strcmp(e->name, ".") ||
	    !strcmp(e->name, "..") ||
	    strchr(e->name, '/'))
		goto einval;

Would something like this be a sufficient sanity check? I suppose we 
could also check for non-printing characters, but I don't think ignoring 
them would be a security issue.


> Thanks for catching this.
>
> David, please check if the LED name sticks to the LED class
> device naming convention.
>
> And one thing that caught my eye only now - please use
> devm_led_classdev_register().
>
> For now I'm dropping the patch.