Re: seccomp selftest fails to build with glibc 2.26

[Kees Cook <keescook@chromium.org>]

On Thu, Sep 7, 2017 at 2:27 PM, Seth Forshee <seth.forshee@canonical.com> wrote:
> Hi Kees,
>
> I'm seeing build failures with your seccomp selftest when using glibc
> 2.26. The first are related to changing macro names from __have_sig*_t
> to __sig*_t_defined. But after defining those there are more conflicting
> definitions. I was able to get it to build with the changes below,
> however it's ugly so I'm hesitant to suggest that it's a fix (and I
> haven't tested with older glibc either).
>
> The full build output is a little lengthy, so I've pasted it at
> http://pastebin.ubuntu.com/25486192/ rather than including it inline.

Thanks for the details! I took a different route for the solution
(basically, we can drop the hack for glibc not having support for
SIGSYS in its siginfo_t), and sent a separate email with the patch.

-Kees

>
> Thanks,
> Seth
>
>
> diff --git a/tools/testing/selftests/seccomp/seccomp_bpf.c b/tools/testing/selftests/seccomp/seccomp_bpf.c
> index 03f1fa49..d234a3e5 100644
> --- a/tools/testing/selftests/seccomp/seccomp_bpf.c
> +++ b/tools/testing/selftests/seccomp/seccomp_bpf.c
> @@ -11,6 +11,13 @@
>  #define __have_sigval_t 1
>  #define __have_sigevent_t 1
>
> +/* These fix errors with glibc 2.26 */
> +#define __siginfo_t_defined 1
> +#define __sigval_t_defined 1
> +#define __sigevent_t_defined 1
> +#define _BITS_SIGINFO_CONSTS_H 1
> +#define _BITS_SIGEVENT_CONSTS_H 1
> +
>  #include <errno.h>
>  #include <linux/filter.h>
>  #include <sys/prctl.h>



-- 
Kees Cook
Pixel Security