On 29.07.2021 14:02:34, Thomas Wagner wrote:

Hello Marc,

On 2021-07-29 12:55, Marc Kleine-Budde wrote:

quoted

On 29.07.2021 12:03:56, thomas@the-wagner.de wrote:

quoted

Shouldn't the error only be returned if the
canfd_frame I pass has more than 8 bytes when the interface is not in
FD-mode?

A CAN-2.0 frame with 8 bytes is something different than a CAN-FD frame
with 8 bytes. The kernel uses the length of the frame to decide if it is
a CAN-2.0 or CAN-FD frame. If your CAN controller has switched CAN-FD
off, it cannot send CAN-FD frames, thus you get an error.

Does that make sense?

Sure!

I see how a CAN-2.0 frame with 8 bytes differs from a CAN-FD frame with
8-bytes, but when I receive into a canfd_frame I can't differentiate like that
anymore. In userspace an 8B CAN-2.0 frame and an 8B CAN-FD frame look just
the same, no matter the interface running with FD on or off.

... which is wrong as I just noticed. Paying attention to the actual bytes read
by the socket I can see the 16 vs. 72B that make up a can_frame vs. a
canfd_frame respectively. Even when always writing into a canfd_frame.

With the C language the type information of your read buffer doesn't
leave the scope of your function (write() uses a void * for the buffer).
The length information is checked in the kernel, but only if it's large enough:

https://elixir.bootlin.com/linux/v5.13/source/net/can/raw.c#L850

The same differentiation I must make when sending...

ACK - If you an idea how to improve the documentation, let me know!

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde           |
Embedded Linux                   | https://www.pengutronix.de  |
Vertretung West/Dortmund         | Phone: +49-231-2826-924     |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-5555 |