Device classes

A device class turns the raw device API into a reusable, host-recognised interface. The bundled classes (HID, CDC-ACM, …) are written only on the public API - nothing about them is privileged, so your own class is a first-class citizen.

Python - subclass Interface

A class is an Interface subclass: declare endpoints, optionally emit class-specific descriptors, and handle the events you care about (on_out, on_control, …). The HID/CDC/MSC classes follow exactly this shape.

from usbip import Interface, In, Out

class MyClass(Interface):
    bInterfaceClass = 0xFF
    tx = In(0x81, "interrupt", mps=8)

    def on_control(self, setup, data=b""):
        ...                       # class/vendor requests; return bytes or raise Stall

    def notify(self, payload):
        self.tx.write(payload)    # device -> host

Beyond on_out and on_control, an Interface can override on_iso (isochronous streams, producing on IN endpoints and consuming on OUT ones), set_alt (alternate settings), on_reset, and adjust_for_speed (resize endpoints for high speed) - see the class layer API. Users add it the way they add any class - dev.add(YourClass(...)) - which returns the interface itself. A multi-interface Function points its primary property at whichever child carries the data plane, and that is what add() hands back (as the bundled classes do).

C - a usbip_device_class descriptor

In C a class is a usbip_device_class descriptor (build / control / I/O callbacks + per-instance state). Declare it statically, then instantiate it on a device:

static const usbip_device_class my_class = {
    .name = "myclass",
    .bInterfaceClass = 0xFF,
    .build = my_build,
    .control = my_control,
    .state_size = sizeof(struct my_state),
};
usbip_function *fn = usbip_device_add_class(dev, &my_class, &params);

.name is only used in diagnostics. The built-in C classes expose an *_add() helper (hid_add, cdc_acm_add, …) so most code never names the class object at all.

See the ready-made classes in Device Classes and their API.