Re: [PATCH RFC] gen_init_cpio: Do fsync() only on regular files

[Dmitry Safonov]

On Tue, Oct 7, 2025 at 2:17 AM David Disseldorp <ddiss@xxxxxxx> wrote:
[..]
> The code change looks fine, but the commit message is a bit verbose IMO.
> Please drop the first and last paragraphs. The reproducer could also be
> simplified to e.g.
> echo | usr/gen_init_cpio /dev/stdin > /dev/null

hehe, OK, let me simplify that in v2.

>
> With that:
> Reviewed-by: David Disseldorp <ddiss@xxxxxxx>
>
> > Fixes: ae18b94099b0 ("gen_init_cpio: support -o <output_file> parameter")
> > Cc: David Disseldorp <ddiss@xxxxxxx>
> > Cc: Nicolas Schier <nsc@xxxxxxxxxx>
> > Signed-off-by: Dmitry Safonov <dima@xxxxxxxxxx>
> > ---
[..]
>
> Another option would be to catch the EINVAL error, e.g.
>
> --- a/usr/gen_init_cpio.c
> +++ b/usr/gen_init_cpio.c
> @@ -112,7 +112,10 @@ static int cpio_trailer(void)
>             push_pad(padlen(offset, 512)) < 0)
>                 return -1;
>
> -       return fsync(outfd);
> +       if (fsync(outfd) < 0 && errno != EINVAL)
> +               return -1;
> +
> +       return 0;
>  }
>
> It may be a little portable than isfdtype(), but I don't feel strongly
> either way.

Yeah, maybe worth avoiding breking compilation on some weird set ups,
going to use your version for v2.

Thanks,
           Dmitry