Re: [PATCH v3 3/4] media: uvcvideo: Introduce allow_privacy_override module parameter

[Ricardo Ribalda]

Hi Michal

On Thu, 19 Mar 2026 at 12:09, Michal Pecio <michal.pecio@xxxxxxxxx> wrote:
>
> On Thu, 19 Mar 2026 10:56:59 +0100, Ricardo Ribalda wrote:
> > The goal of the deprecation period is exactly this: to trigger a
> > conversation before a permanent block.
>
> Most users will just curse and edit their /etc/modprobe.conf. They may
> post a rant on some distro forum. I suspect no one will monitor this.
>
> > We can then decide if we need a specialized API for their use case or
> > a Kconfig option, rather than leaving the current "anyone can turn
> > off the privacy LED" status quo.
>
> Why not just add the specialized API right away?

We don't know the exact use cases yet, and I do not want to design an
API without understanding the users for it.

At this moment, we have only identified these usecases:

- Disabling the LED to avoid reflections in glasses. (This is
generally a non-issue with modern hardware).
- Baby monitors. (I would argue that physical tape is the correct
solution for a sleep-disturbing light).

>
> I believe users affected by this regression are already known,
> ISTR some negative response to previous iterations of this patch.
>
> Kconfig option sounds crazy, who would want to rebuild the kernel
> for this? Depending on BROKEN is double crazy.

I am not set on the final implementation yet; it is exactly the kind
of topic we should discuss at a media summit.

>
> > The attack vector is that an app with camera access, like your
> > browser, can record you when you don't want to be recorded.
> > The LED will be a signal that something is happening.
> >
> > Imagine that you install a Flatpak for live streaming. Assuming the
> > Flatpak is properly sandboxed, remote code execution is less worrisome
> > than the app spying on you.
>
> Theoretically yes. But also nobody should rely on those LEDs.
> People who care ask HW vendors for physical switches or disconnect
> the camera while not in use. I have seen black tape on laptop lids.

I rely on my LEDs. I know they are wired to the sensor power supply,
so the LED is definitely on when the camera is in use.
I want all users to be able to trust their LEDs like I do.

>
> Are there more owners of affected hardware who want this code than
> those who don't? Maybe it could be a Kconfig option for them :)

I believe the majority of users prefer a system that is "secure by
default." Most people expect that if the LED is off, the camera is
off.

>
> Most of my USB cameras don't even have activity LEDs.
>
> > > So it's not removal of some controversial feature, but 3KB of extra
> > > code in everybody's kernel (I just applied this patch) and a forever
> > > game of whack-a-mole with HW vendors? They will win...
> >
> > Maybe I meassured it wrong. But I can only account for 1.3 KiB
>
> I simply ran stat uvcvideo.ko and calculated difference.
> Could be a matter of different kernel configs.
>
> > I see no need for vendors to hide these features, they simply added
> > them because an OEM thought it was a nice feature to have, or because
> > they left them as hardware debug features.
>
> But how will the kernel know about those random debug backdoors?
> It just seems that whatever is discovered by users and becomes popular
> enough to reach linux-media, will be getting blacklisted and broken.
>

I prefer to say "filtered" rather than "broken." It’s a matter of
perspective: we are filtering out non-standard controls that undermine
user privacy. While we might not catch every debug backdoor
immediately, setting a policy and blocking known overrides is a
significant step and also sends a strong message to vendors.

Best regards!


-- 
Ricardo Ribalda