[CC += libc-alpha]

Hi Alexander,

On Tue, Jan 20, 2026 at 11:10:07AM +0000, Alexander Atanasov wrote:
> While compiling filesystem selftests there is a warning:
> anon_inode_test.c:45:37: warning: null passed to a callee that
>        requires a non-null argument [-Wnonnull]
> 45 |  ASSERT_LT(execveat(fd_context, "", NULL, NULL, AT_EMPTY_PATH), 0);
> 
> Kernel code is okay to be passed NULL for argv, in which
> case it generates argv by itself inside do_execveat_common.
> Man page of execvat lists both argv and envp as _Nullable.

Yup; the manual page uses that to document that Linux accepts NULL.

> But system's unistd.h says different:
> extern int execveat (int __fd, const char *__path, char *const __argv[],
>                      char *const __envp[], int __flags)
>     __THROW __nonnull ((2, 3));

Hmmmm, this would trigger UB for user-space users relying on the fact
that Linux accepts NULL.  Maybe when compiled for Linux, the glibc
headers should remove the [[gnu::nonnull()]] attribute here.


Have a lovely day!
Alex

> (checked ubuntu 24.04 and fedora 43)
> 
> Which is the reason for the compiler warning.
> 
> Fix the warning by passing a proper argv array.
> 
> Fixes: f8ca403ae77cb ("selftests/filesystems: add exec() test for anonymous inodes")
> Cc: Christian Brauner <brauner@kernel.org>
> Cc: Guenter Roeck <linux@roeck-us.net>
> Cc: Michael Kerrisk <mtk.manpages@gmail.com>
> Cc: Alejandro Colomar <alx.manpages@gmail.com>
> Cc: Clint George <clintbgeorge@gmail.com>
> Cc: Ranganath V N <vnranganath.20@gmail.com>
> Signed-off-by: Alexander Atanasov <alex@zazolabs.com>
> ---
>  tools/testing/selftests/filesystems/anon_inode_test.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> Quote from execv's man referenced from execveat's:
> On Linux, argv and envp can be specified as NULL.
> In both cases, this has the same effect as specifying the argument as a pointer 
> to a list containing a  single null pointer.
> ***Do not take advantage of this nonstandard and nonportable misfeature!***
> 
> Listed as possible to pass NULL, then a big fat warning in the man page
> and a header that acctually marks this as invalid usage.
> I failed to find the origin of the above contradiction.
> It looks like POSIX/GNU/Linux missalignment.
> I am putting this here for the record. So may be someone could step in
> and make it right.
> 
> I've Cc-ed people who made fixes but they are just compile fixes, no
> explanation of why the warning exists and in most of them the array is not a proper
> argv.
> 
> diff --git a/tools/testing/selftests/filesystems/anon_inode_test.c b/tools/testing/selftests/filesystems/anon_inode_test.c
> index 94c6c81c2301..191549a4304d 100644
> --- a/tools/testing/selftests/filesystems/anon_inode_test.c
> +++ b/tools/testing/selftests/filesystems/anon_inode_test.c
> @@ -4,6 +4,7 @@
>  
>  #include <fcntl.h>
>  #include <stdio.h>
> +#include <unistd.h>
>  #include <sys/stat.h>
>  
>  #include "kselftest_harness.h"
> @@ -37,12 +38,13 @@ TEST(anon_inode_no_chmod)
>  
>  TEST(anon_inode_no_exec)
>  {
> +	char *const argv[] = { "", NULL };
>  	int fd_context;
>  
>  	fd_context = sys_fsopen("tmpfs", 0);
>  	ASSERT_GE(fd_context, 0);
>  
> -	ASSERT_LT(execveat(fd_context, "", NULL, NULL, AT_EMPTY_PATH), 0);
> +	ASSERT_LT(execveat(fd_context, "", argv, NULL, AT_EMPTY_PATH), 0);
>  	ASSERT_EQ(errno, EACCES);
>  
>  	EXPECT_EQ(close(fd_context), 0);
> -- 
> 2.43.0
> 

-- 
<https://www.alejandro-colomar.es>