Message-ID: <167273840266.530483.6812185939521706359@Monstersaurus>
Date:   Tue, 03 Jan 2023 09:33:22 +0000
From:   Kieran Bingham <kieran.bingham@...asonboard.com>
To:     Laurent Pinchart <laurent.pinchart@...asonboard.com>,
        Sakari Ailus <sakari.ailus@...ux.intel.com>
Cc:     Barnabás Pőcze <pobrn@...tonmail.com>,
        linux-media@...r.kernel.org,
        Mauro Carvalho Chehab <mchehab@...nel.org>,
        Tomi Valkeinen <tomi.valkeinen@...asonboard.com>,
        Hans Verkuil <hverkuil-cisco@...all.nl>,
        Christophe JAILLET <christophe.jaillet@...adoo.fr>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH] media: v4l2-dev: sysfs: Support streaming attribute

Quoting Laurent Pinchart (2023-01-02 13:35:33)
> Hello,
> 
> On Mon, Jan 02, 2023 at 01:14:15PM +0000, Sakari Ailus wrote:
> > On Wed, Dec 28, 2022 at 01:44:38AM +0000, Barnabás Pőcze wrote:
> > > On 2022. december 26., hétfő 10:52, Laurent Pinchart wrote:
> > > > On Fri, Dec 23, 2022 at 11:17:35PM +0000, Kieran Bingham wrote:
> > > > 
> > > > > Provide a streaming attribute to allow userspace to interogate if a device
> > > > > is actively streaming or not.
> > > > > 
> > > > > This will allow desktop notifications to report if a camera or device
> > > > > is active on the system, rather than just 'open' which can occur when
> > > > > configuring the device.
> > > > > 
> > > > > Bug: https://gitlab.freedesktop.org/pipewire/pipewire/-/issues/2669
> > > > > Signed-off-by: Kieran Bingham kieran.bingham@...asonboard.com
> > > > > ---
> > > > > 
> > > > > This is a quick POC to see if such a facility makes sense.
> > > > > I'm weary that not all video devices may have the queues registered on
> > > > > the struct video_device, but this seems like an effective way to be able
> > > > > to determine if a device is actively streaming on a system.
> > > > 
> > > > I can imagine multiple problems, from race conditions to permissions and
> > > > privacy. In order to comment on the fitness of this solution to address
> > > > the problem you're trying to solve, could you describe the actual
> > > > problem ?
> > > 
> > > The issue is explained in the following thread:
> > > https://gitlab.freedesktop.org/pipewire/pipewire/-/issues/2669#note_1697388
> > > 
> > > In short, the user wants to show a "camera-in-use" indicator when the laptop camera
> > > is used. The script that the user previously had only checked if /dev/video0
> > > was open in any process, if it was, the indicator was shown. However, libcamera
> > > - at least at the moment - keeps the file descriptor open as long as the Camera
> > > object exists, which pipewire keeps alive for the entire lifetime of the device,
> > > therefore the "camera-in-use" indicator is always shown.
> > 
> > A sysfs attribute is not a great way to address this.
> > 
> > libcamera certainly has information on whether streaming is ongoing. The
> > information should come from there. Or Pipewire. Dbus perhaps?
> 
> I tend to agree, I think this is best solved in userspace where PipeWire
> can have a centralized view of all cameras in the system, and of their
> users.

I fear that misses the entire point I was trying to make.

Lets say pipewire 'is' available and in use and can be used to capture
video streams for video calls, that's fine. But what happens if a user
runs a gstreamer pipeline without using the pipewire source, or a
suspcious process runs "yavta" and captures an image or stream
discreetly...

Only the kernel has a true centralised view of what devices are in use.

> > Alternatively libcamera could close the video devices while not streaming
> > but that would involve e.g. releasing possible video buffer allocations as
> > well, increasing streaming start latency.

Or is it just that in that case 'lsof' should be sufficient?

The problem I have with that is - just like with the issue when the
Privacy LED comes on during power up/probe - then any time a device is
opened to identify the device and not necessarily use it - the 'camera
in use' notification would get flashed...

> Closing video (and subdev) nodes when the camera is not in use would be
> good I think. It doesn't mean we have to open them when starting
> capture, explicit open/close operation (or similar, maybe introducing a
> capture session object in the libcamera API would also make sense, it
> should be considered as part of the same issue) could help with this.

I'm not talking about libcamera in this thread. It's how does a user
correctly identify when a camera is in use globally in a system.
--
Kieran

Powered by blists - more mailing lists