Re: [PATCH] sched/headers: Fix sched_setattr userspace compilation breakage

[Joel Fernandes]

Hi Linus,

On Thu, May 28, 2020 at 03:21:56PM -0700, Linus Torvalds wrote:
> On Thu, May 28, 2020 at 6:55 AM Joel Fernandes (Google)
> <joel@xxxxxxxxxxxxxxxxx> wrote:
> >
> > On a modern Linux distro, compiling the following program fails:
> >  #include<stdlib.h>
> >  #include<stdint.h>
> >  #include<pthread.h>
> >  #include<linux/sched/types.h>
> 
> You shouldn't include kernel headers in user space - that's the job of
> glibc and friends.

Ah, my bad. Sorry I read the docs now and looks like I got it all backwards.

> 
> > --- a/include/uapi/linux/sched/types.h
> > +++ b/include/uapi/linux/sched/types.h
> > @@ -4,9 +4,11 @@
> >
> >  #include <linux/types.h>
> >
> > +#if defined(__KERNEL__)
> >  struct sched_param {
> >         int sched_priority;
> >  };
> > +#endif
> 
> This makes no sense.
> 
> The point of a 'uapi' header is to export things to user space. Yes,
> they sometimes mix kernel-internal thngs in there (because of how they
> were created by just moving kernel headers to the uapi directory), but
> that ' struct sched_param' is very much part of the very interface
> definition that that file is all about exporting.
> 
> So no, this patch is fundamentally wrong. It negates the whole point
> of having a uapi header at all.

Sorry, I naively assumed that headers in 'include/uapi/' are safe to include
from userspace. I feel terrible.

> The glibc-provided "<sched.h>" should have been where you got all
> these declarations and #defines from, and the point of the uapi file
> was always to help glibc (and other library implementations) get them
> from the kernel.

The problem is <sched.h> still does not get us 'struct sched_attr' even
though the manpage of sched_setattr(2) says including <sched.h> is all that's
needed.

> 
> So why are you including kernel header files and mixing them with
> system-provided stuff?

The include of <sched.h> does not result in availability of the sched_attr
header.

Also, even if glibc included 'include/uapi/linux/sched/types.h' to get struct
sched_attr's definition, we would run into the same issue I reported right?
The 'struct sched_param' is already defined by glibc, and this header
redefines it.

Sorry that this patch is wrong, I'll try to fix it the right way. Thanks for
your help.

thanks,

 - Joel