libusb0 backend kernel driver detach for macOS with libusb 1.0.25

[mcuee]

One of the major change for the upcoming libusb 1.0.25 release is to add kernel driver detaching for macOS. This change should not affect libusb1 backend of pyusb. However, the current libusb0 has hard coded codes saying the feature is only available for Linux. This may need to be fixed.

libusb/libusb#911
https://github.com/libusb/libusb/blob/master/ChangeLog
2021-05-16: v1.0.25-pre

• Darwin (macOS): Add support for detaching kernel drivers with authorization.

[mcuee]

Take note macOS users will use libusb-compat-0.1 which is just a wrapper of libusb-1.0.
https://github.com/libusb/libusb-compat-0.1/blob/master/libusb/core.c

[jonasmalacofilho]

We don't restrict kernel detachment on Darwin, even on libusb0: we only check that usb_detach_kernel_driver_np() is present and delegate everything to it.

pyusb/usb/backend/libusb0.py

Lines 648 to 652 in 5d7cb25

	@methodtrace(_logger)
	def detach_kernel_driver(self, dev_handle, intf):
	if not hasattr(_lib, 'usb_detach_kernel_driver_np'):
	raise NotImplementedError(self.detach_kernel_driver.__name__)
	_check(_lib.usb_detach_kernel_driver_np(dev_handle, intf))

What I think will be an issue is our is_kernel_driver_active() backend method, because there we do have some custom logic, to handle all the different semantics usb_get_driver_np() may have depending on the implementation.

pyusb/usb/backend/libusb0.py

Lines 620 to 646 in 5d7cb25

	@methodtrace(_logger)
	def is_kernel_driver_active(self, dev_handle, intf):
	if sys.platform == 'linux':
	# based on the implementation of libusb_kernel_driver_active()
	# (see op_kernel_driver_active() in libusb/os/linux_usbfs.c)
	# and the fact that usb_get_driver_np() is a wrapper for
	# IOCTL_USBFS_GETDRIVER
	try:
	driver_name = self.__get_driver_name(dev_handle, intf)
	# 'usbfs' is not considered a [foreign] kernel driver because
	# it is what we use to access the device from userspace
	return driver_name != b'usbfs'
	except USBError as err:
	# ENODATA means that no kernel driver is attached
	if err.backend_error_code == -errno.ENODATA:
	return False
	raise
	elif sys.platform.startswith('freebsd') or sys.platform.startswith('dragonfly'):
	# this is similar to the Linux implementation, but the generic
	# driver is called 'ugen' and usb_get_driver_np() simply returns an
	# empty string is no driver is attached (see comments on PR #366)
	driver_name = self.__get_driver_name(dev_handle, intf)
	# 'ugen' is not considered a [foreign] kernel driver because
	# it is what we use to access the device from userspace
	return driver_name != b'ugen'
	else:
	raise NotImplementedError(self.is_kernel_driver_active.__name__)

Can we assume all Darwin users will be running libusb-compat-0.1? Because if that's the case, the required change is trivial ('dummy' => True, -ENODATA => False).

[mcuee]

Yes we can safe to assume all Darwin users to use libusb-compat-0.1 now (or even since many years ago). Nathan has been with the libusb project since the 0.1 days.