udev-dev: Enumerate evdev devices via kern.evdev.input sysctls

[Defenso-QTH]

When /dev/input/ is not visible to the calling process (for example, a FreeBSD jail without the corresponding devfs ruleset), the existing scandir-based enumeration in udev_dev_enumerate() cannot discover any input devices, even though the kern.evdev.input.* sysctls remain readable and the per-device property handlers already populate from those sysctls.

Add a sysctl tree-walk fallback that runs after the scandir pass. The walk uses CTL_SYSCTL_NEXT to discover the actual children of kern.evdev.input without a fixed upper bound, then resolves each unit's name via CTL_SYSCTL_NAME to extract the unit number for the /dev/input/eventN syspath. Devices already discovered by scandir are deduplicated by udev_list_insert.

This lets libinput (and other libudev consumers) enumerate input devices in environments where the evdev sysctls are accessible but the devnodes are not.

Test setup:

• FreeBSD 15-STABLE
• jail with no devfs exposure of input devices
• seatd exposed through nullfs
• libinput with patch to trust seatd provided by fd even if no device node is found: freebsd/freebsd-ports@90514b8 (under review in libinput: https://gitlab.freedesktop.org/libinput/libinput/-/merge_requests/1471)
• cage compositor and foot terminal

[wulf7]

That results in duplicate entries for input/event devices. Check for /dev/input existence before falling back to sysctls

[Defenso-QTH]

Thanks for looking at this! My understanding of how entries are added to the udev list is that duplicates are automatically removed on insert:

libudev-devd/udev-list.c

Lines 53 to 71 in 0f2c177

	int
	udev_list_insert(struct udev_list *ul, char const *name, char const *value)
	{
	struct udev_list_entry *ule, *old_ule;
	ule = udev_list_entry_alloc(name, value);
	if (!ule)
	return (-1);
	ule->list = ul;
	old_ule = RB_FIND(udev_list, ul, ule);
	if (old_ule != NULL) {
	RB_REMOVE(udev_list, ul, old_ule);
	udev_list_entry_free(old_ule);
	}
	RB_INSERT(udev_list, ul, ule);
	return (0);
	}

Am I not reading that correctly?

[wulf7]

I do not see reasons to do extra device scan pass. Actually you may read security.jail.jailed sysctl value and do extra enumeration attempt if it is not zero.

[Defenso-QTH]

You're right that's enough for my use case. I added that check in 7432dbf.

[wulf7]

Merged. Thank you!