Re: [PATCH 4/5] media: v4l2-compat-ioctl32: fix several __user annotations

[Mauro Carvalho Chehab <mchehab@s-opensource.com>]

Em Tue, 17 Apr 2018 15:11:10 +0200
Hans Verkuil <hansverk@cisco.com> escreveu:

> >> Be aware that the unsigned char * cast is actually a bug: it will clamp the
> >> u32 'blocks' value to a u8.
> >>
> >> Regards,
> >>
> >> 	Hans  
> > 
> > What about this approach (code untested)?  
> 
> I prefer the explicit casts. These are special situations and hiding it in
> defines makes it actually harder to follow.

There are just one special case there: USERPTR, where we force a cast
to unsigned long:

drivers/media/v4l2-core/v4l2-compat-ioctl32.c:              put_user((unsigned long)compat_ptr(p), &p64->m.userptr))
...
drivers/media/v4l2-core/v4l2-compat-ioctl32.c:                      put_user((unsigned long)compat_ptr(userptr),
drivers/media/v4l2-core/v4l2-compat-ioctl32.c-                               &p64->m.userptr))

I kept it out of the macros.

IMO, maintaining the code with the type_of() is more error-prune, as
the cast will be done the way it should. The cases where some other
cast like is needed - with is what it was done for USERPTR should
be explicit (and, btw, they should be properly documented why doing
that).

I don't remember anymore why it is casting to unsigned long,
but, based on v4l2-compliance tests, it seems that such cast works.

Thanks,
Mauro