Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 21 Mar 2022 02:07:19 +0000
From:   Carlos Llamas <cmllamas@google.com>
To:     Miklos Szeredi <miklos@szeredi.hu>
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Alessio Balsini <balsini@android.com>,
        Masahiro Yamada <masahiroy@kernel.org>,
        Arnd Bergmann <arnd@arndb.de>,
        kernel-team <kernel-team@android.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] fuse: fix integer type usage in uapi header
Message-ID: <Yjfd1+k83U+meSbi@google.com>
References: <20220318171405.2728855-1-cmllamas@google.com>
 <CAJfpegsT6BO5P122wrKbni3qFkyHuq_0Qq4ibr05_SOa7gfvcw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAJfpegsT6BO5P122wrKbni3qFkyHuq_0Qq4ibr05_SOa7gfvcw@mail.gmail.com>
Precedence: bulk

On Fri, Mar 18, 2022 at 08:24:55PM +0100, Miklos Szeredi wrote:
> On Fri, 18 Mar 2022 at 18:14, Carlos Llamas <cmllamas@google.com> wrote:
> >
> > Kernel uapi headers are supposed to use __[us]{8,16,32,64} defined by
> > <linux/types.h> instead of 'uint32_t' and similar. This patch changes
> > all the definitions in this header to use the correct type. Previous
> > discussion of this topic can be found here:
> >
> >   https://lkml.org/lkml/2019/6/5/18
> 
> This is effectively a revert of these two commits:
> 
> 4c82456eeb4d ("fuse: fix type definitions in uapi header")
> 7e98d53086d1 ("Synchronize fuse header with one used in library")
> 
> And so we've gone full circle and back to having to modify the header
> to be usable in the cross platform library...
> 
> And also made lots of churn for what reason exactly?

There are currently only two uapi headers making use of C99 types and
one is <linux/fuse.h>. This approach results in different typedefs being
selected when compiling for userspace vs the kernel. Plus only __u32 and
similar types align with the coding style as described in 5(e).

Yet, there is still the cross platform concern you mention. I think the
best way to accommodate this while still conforming with the __u32 types
is to follow something similar to 1a95916f5465 ("drm: Add compatibility
#ifdefs for *BSD"). Basically doing this:

  #if defined(__KERNEL__) || defined(__linux__)
  #include <linux/types.h>
  #else
  #include <stdint.h>
  typedef uint16_t __u16;
  typedef int32_t  __s32;
  typedef uint32_t __u32;
  typedef int64_t  __s64;
  typedef uint64_t __u64;
  #endif

This alternative selects the correct uapi types for both __KERNEL__ and
__linux__ cases which is the main goal of this patch and it's just minor
fixes from 7e98d53086d1 ("Synchronize header with one used in library").

I see there where previous attempts to address similar changes here:
  https://lkml.org/lkml/2013/3/11/620
  https://lkml.org/lkml/2013/4/15/487

So, if you agree with the approach above I'd be happy to send a separate
patch on top to address the *BSD compatibility.

Thanks,
Carlos Llamas