Re: [PATCH] usbmon: expose the usbmon structures and constants as an UAPI header.

["Diego Elio Pettenò" <flameeyes@flameeyes.com>]

On Mon, Jul 6, 2020, at 11:30, Greg KH wrote:
> You can't change someone's copyright lines, sorry.

Then I hope Paolo can send a follow-up patch updating his email address — email.it doesn't seem to exist anymore/doesn't seem to have his address anymore.

> Put the copyright notices back in here.

Done.

> And you see the kbuild error for this, please fix that up.

This should be fixed now by dropping the conditional, which AIUI is the right thing to do in this case? I couldn't find any documentation referencing what to do with CONFIG_COMPAT for headers, but I assume that anything optionally available should be exposed in UAPI and worst case fail at runtime.

> As you are making this a "real" uapi .h file, you have to use the "real"
> data types as well.  That means using __u32 and not u32.  Same for all
> others.

Done, and in the documentation.

> Does this cross the user/kernel boundry?  Ick, no wonder we need compat
> fixups :(

I fear so. The usbmon interface is fairly long-living, and I don't think it matches most of the best practices of today.

I keep telling myself I should spend more time trying to figure out what I can improve overall, because I fear this is better served by having a good thought on the design, and be replaced. But that's a topic for another thread.

Although if someone is interested in whiteboarding an alternative design, I'd be happy to start such a thread or have an offline chat. My current castle-in-the-sky would involve integrating BPF so that capture filters can be applied — and I think that the modern way to stream data from the kernel to userspace would be netlink?