ltp: openat2 build failure with glibc 2.43

[Hemanthkumarmd]

Description

Building LTP against glibc 2.43 fails with the following errors:

lapi/openat2.h:59: error: field 'how' has incomplete type 'struct open_how'
implicit declaration of function 'openat2'
openat202.c:26:46: error: 'RESOLVE_NO_XDEV' undeclared here (not in a function)
openat202.c:27:53: error: 'RESOLVE_NO_MAGICLINKS' undeclared here (not in a function)
openat202.c:28:46: error: 'RESOLVE_NO_SYMLINKS' undeclared here (not in a function)
openat202.c:29:46: error: 'RESOLVE_BENEATH' undeclared here (not in a function)
openat202.c:31:49: error: 'RESOLVE_IN_ROOT' undeclared here (not in a function)

Environment

• LTP version: 20260130
• glibc: 2.43
• GCC: 15.2.0
• Kernel headers: 6.18.0
• OS: OpenEmbedded/Yocto (x86-64)

Steps to Reproduce

This was observed while cross-building LTP with glibc 2.43.

Root Cause

glibc 2.43 added a native openat2() wrapper for the first time
(BZ#31664, commit 0f0a5cd338998f4b603f52f3ce2163df0db7b814
by Adhemerval Zanella, merged 2025-12-09):
https://sourceware.org/git/?p=glibc.git;a=commit;h=0f0a5cd338998f4b603f52f3ce2163df0db7b814

configure.ac uses AC_CHECK_FUNCS_ONCE([openat2]) which detects this
new glibc wrapper and sets HAVE_OPENAT2=1. This causes lapi/openat2.h
to skip its own definitions of struct open_how, RESOLVE_* constants,
and the openat2() wrapper inside #ifndef HAVE_OPENAT2.

However, in glibc 2.43's bits/fcntl-linux.h, both struct open_how
(via bits/openat2.h) and the openat2() prototype are gated inside
#ifdef __USE_GNU. __USE_GNU is only set when _GNU_SOURCE is
explicitly defined. LTP does not define _GNU_SOURCE, so glibc's
definitions are not visible to LTP's test files, leaving struct open_how,
RESOLVE_* constants and openat2() undefined and causing the build
failure.

Workaround Applied

As a temporary workaround, we are undefining HAVE_OPENAT2 in
lapi/openat2.h since LTP always uses tst_syscall(__NR_openat2)
directly and never calls glibc's wrapper.

Possible Fix

One possible approach would be to replace AC_CHECK_FUNCS_ONCE([openat2])
in configure.ac with a check for the kernel syscall number __NR_openat2
instead of the glibc wrapper. However we are not sure if that is the right
approach and would appreciate guidance from the maintainers.

[sshedi]

Try this:

--- a/include/lapi/openat2.h
+++ b/include/lapi/openat2.h
@@ -10,6 +10,9 @@
 #include <sys/syscall.h>
 #include <linux/types.h>
+#include <linux/openat2.h>
+
 #include "lapi/syscalls.h"