The "-rm" version breaks systems by deleting device nodes that are still referenced by the udev database

[foresto]

While I have no doubt that these rules were written with good intentions, 51-these-are-not-joysticks-rm.rules is putting systems in a bad state by deleting device nodes that are still referred to by udev.

The consequence: any software that correctly discovers joysticks by querying udev or reacting to udev events will fail when it tries to open their device nodes, because the node (e.g. /dev/input/js0) doesn't actually exist.

Unfortunately, this rules file has made its way into some widely used packages, such as certain redistributions of Valve's "steam-devices" udev rules collection, thereby causing mysterious failures that shouldn't occur.

Please consider using a different approach.

Suggestion 1:

• Instead of deleting device nodes, unbind the device from the hid driver, using the sysfs facility made for that purpose.
• The generic part of a rule would look roughly like this: ACTION=="bind", SUBSYSTEM=="hid", ATTR{driver/unbind}="%k"
• Docs: https://lwn.net/Articles/143397/
• Example: https://github.com/atar-axis/xpadneo/blob/66d91208/hid-xpadneo/etc-udev-rules.d/60-xpadneo.rules

Suggestion 2:

• Instead of deleting device nodes, deauthorize the interfaces, using the sysfs facility made for that purpose.
• The generic part of a rule would look roughly like this: SUBSYSTEM=="usb", ATTR{authorized}="0"
• Docs: https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-bus-usb

I spent more than a few hours tracing a failure caused by these rules to its source. It was tough to pin down, because most users who have affected devices don't know enough about how these things work to diagnose them on their own, and most people who have the knowledge don't have the affected devices and don't expect custom udev rules to be deleting device nodes.

Thanks for your attention.

[scaronni]

@foresto I'm the packager who added the udev rule in the steam-devices subpackage in Fedora, I was notified by a user.

Your approach in Suggestion 1 seems to be the best, I'll make a Pull Request proposal for that. I don't have one device that falls into this list since a very long time (at least ~8 years) so I won't be able to test, but I will ask the user to test.

[KayJay7]

@scaronni Another possible approach would be creating an /etc/udev/hwdb.d/61-these-are-not-joystick.hwdb file, similarly to how we did here as recommended by a systemd dev.

But that would require the modalias id of all the devices in this repo.

[scaronni]

That's even better. I think we should add all the IDs in this project in the very same way and get rid of the custom udev rule. This repository is very old and the drivers have never been fixed, so putting it there sounds like the best approach.

We can generate the modalias skipping the version id and just using Vendor and Product I guess.

[scaronni]

Do we need anything else beside ID_INPUT_JOYSTICK=?

Example:

id-input:modalias:input:b0003v046bpff10*
  ID_INPUT_JOYSTICK=

id-input:modalias:input:b0003v046dpc30a*
  ID_INPUT_JOYSTICK=
[...]

[scaronni]

If I'm reading correctly, except the ASRock LED Controller all the others are anyway input devices (mouse, keyboard, tablet).