Re: [PATCH] Input: uinput/uhid - disallow control characters in phys paths

On Mon, Aug 03, 2026 at 06:47:23PM -0700, Dmitry Torokhov wrote:
> On Mon, Aug 03, 2026 at 09:58:35PM +0200, Jiri Kosina wrote:
> > On Fri, 24 Jul 2026, David Rheinsberg wrote:
> > 
> > > > There is no good reason to support those, no physical device will ever
> > > > produce those. Allowing \n in phys previously triggered CVE-2026-50292
> > > > in libinput - there the PHYS udev property value was used as part of
> > > > another udev property value. The linebreak then caused the property
> > > > to be split across two lines, allowing uinput devices to inject
> > > > malicious properties. While the bug is squarely inlibinput's court
> > > > there still isn't a good reason for control characters in uinput/uhid.
> > > >
> > > > Signed-off-by: Peter Hutterer <[email protected]>
> > > > ---
> > > >  drivers/hid/uhid.c          |  1 +
> > > >  drivers/input/misc/uinput.c |  1 +
> > > >  include/linux/input.h       | 15 +++++++++++++++
> > > >  3 files changed, 17 insertions(+)
> > > >
> > > > diff --git a/drivers/hid/uhid.c b/drivers/hid/uhid.c
> > > > index 37b60c3aaf66..baf1fe8290f7 100644
> > > > --- a/drivers/hid/uhid.c
> > > > +++ b/drivers/hid/uhid.c
> > > > @@ -513,16 +513,17 @@ static int uhid_dev_create2(struct uhid_device *uhid,
> > > >  		ret = PTR_ERR(hid);
> > > >  		goto err_free;
> > > >  	}
> > > > 
> > > >  	BUILD_BUG_ON(sizeof(hid->name) != sizeof(ev->u.create2.name));
> > > >  	strscpy(hid->name, ev->u.create2.name, sizeof(hid->name));
> > > >  	BUILD_BUG_ON(sizeof(hid->phys) != sizeof(ev->u.create2.phys));
> > > >  	strscpy(hid->phys, ev->u.create2.phys, sizeof(hid->phys));
> > > > +	input_sanitize_phys(hid->phys);
> > > >  	BUILD_BUG_ON(sizeof(hid->uniq) != sizeof(ev->u.create2.uniq));
> > > >  	strscpy(hid->uniq, ev->u.create2.uniq, sizeof(hid->uniq));
> > > > 
> > > >  	hid->ll_driver = &uhid_hid_driver;
> > > >  	hid->bus = ev->u.create2.bus;
> > > >  	hid->vendor = ev->u.create2.vendor;
> > > >  	hid->product = ev->u.create2.product;
> > > >  	hid->version = ev->u.create2.version;
> > > > diff --git a/drivers/input/misc/uinput.c b/drivers/input/misc/uinput.c
> > > > index d32fa4b508fc..70fe4f3e73bf 100644
> > > > --- a/drivers/input/misc/uinput.c
> > > > +++ b/drivers/input/misc/uinput.c
> > > > @@ -998,16 +998,17 @@ static long uinput_ioctl_handler(struct file 
> > > > *file, unsigned int cmd,
> > > > 
> > > >  		phys = strndup_user(p, 1024);
> > > >  		if (IS_ERR(phys)) {
> > > >  			retval = PTR_ERR(phys);
> > > >  			goto out;
> > > >  		}
> > > > 
> > > >  		kfree(udev->dev->phys);
> > > > +		input_sanitize_phys(phys);
> > > >  		udev->dev->phys = phys;
> > > >  		goto out;
> > > > 
> > > >  	case UI_BEGIN_FF_UPLOAD:
> > > >  		retval = uinput_ff_upload_from_user(p, &ff_up);
> > > >  		if (retval)
> > > >  			goto out;
> > > > 
> > > > diff --git a/include/linux/input.h b/include/linux/input.h
> > > > index 76f7aa226202..6c182f5c783f 100644
> > > > --- a/include/linux/input.h
> > > > +++ b/include/linux/input.h
> > > > @@ -527,16 +527,31 @@ int input_set_keycode(struct input_dev *dev,
> > > > 
> > > >  bool input_match_device_id(const struct input_dev *dev,
> > > >  			   const struct input_device_id *id);
> > > > 
> > > >  void input_enable_softrepeat(struct input_dev *dev, int delay, int period);
> > > > 
> > > >  bool input_device_enabled(struct input_dev *dev);
> > > > 
> > > > +/**
> > > > + * input_sanitize_phys - replace invalid characters in a phys string
> > > > + * @phys: the phys path to sanitize (modified in place)
> > > > + *
> > > > + * Replaces any control characters and non-ASCII characters with '?'.
> > > > + **/
> > > > +static inline void input_sanitize_phys(char *phys)
> > > > +{
> > > > +	char *p;
> > > > +
> > > > +	for (p = phys; *p; p++)
> > > > +		if (*p < 0x20 || *p > 0x7e)
> > > > +			*p = '?';
> > > > +}
> > > > +
> > > 
> > > Reviewed-by: David Rheinsberg <[email protected]>
> > > 
> > > I would also be fine to just reject them in uinput, but I guess this is the less intrusive option.
> > 
> > Thanks for the fix.
> > 
> > Dmitry, can you please Ack the above addition to input.h?
> 
> Jiri, as I mentioned I believe that if we need to sanitize phys we
> should also sanitize other fields, especially given that they can be
> set by userspace.

fwiw it's on my list, just hasn't made it to the top yet

> I also wonder why it needs to be inline? And what is wrong with using
> isprint() here? 

inline - it's only used in two specific instances, if it wasn't for the
uhid/uinput overlap it'd be barely worth a function. Happy to change
though.

isprint - good point, will use that in the next version. 

In regards to sanitizing the name/uniq though: I have a
device here that has the copyright symbol in the device name:
   Microsoft Microsoft® 2.4GHz Transceiver v9.0
That is outside isprint() and IMO we shouldn't mess with device names
more than absolutely necessary wo we should probably strictly reduce
this to control characters only (or maybe just \n)?

Cheers,
  Peter