yann@2437: http://bugs.gentoo.org/250342
yann@2437: http://sources.redhat.com/bugzilla/show_bug.cgi?id=9685
yann@2437: 
yann@2437: we cant assume sock_cloexec and pipe2 are bound together as the former defines
yann@2437: are found in glibc only while the latter are a combo of kernel headers and
yann@2437: glibc.  so if we do a runtime detection of SOCK_CLOEXEC, but pipe2() is a stub
yann@2437: inside of glibc, we hit a problem.  for example:
yann@2437: 
yann@2437: #include <grp.h>
yann@2437: #include <stdio.h>
yann@2437: main()
yann@2437: {
yann@2437: 	getgrnam("portage");
yann@2437: 	if (!popen("ls", "r"))
yann@2437: 		perror("popen()");
yann@2437: }
yann@2437: 
yann@2437: getgrnam() will detect that the kernel supports SOCK_CLOEXEC and then set both
yann@2437: __have_sock_cloexec and __have_pipe2 to true.  but if glibc was built against
yann@2437: older kernel headers where __NR_pipe2 does not exist, glibc will have a ENOSYS
yann@2437: stub for it.  so popen() will always fail as glibc assumes pipe2() works.
yann@2437: 
yann@2437: diff -durN glibc-2.12.1.orig/socket/have_sock_cloexec.c glibc-2.12.1/socket/have_sock_cloexec.c
yann@2437: --- glibc-2.12.1.orig/socket/have_sock_cloexec.c	2008-07-25 18:46:23.000000000 +0200
yann@2437: +++ glibc-2.12.1/socket/have_sock_cloexec.c	2009-11-13 00:50:15.000000000 +0100
yann@2437: @@ -16,9 +16,14 @@
yann@2437:     Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
yann@2437:     02111-1307 USA.  */
yann@2437:  
yann@2437: +#include <fcntl.h>
yann@2437:  #include <sys/socket.h>
yann@2437:  #include <kernel-features.h>
yann@2437:  
yann@2437:  #if defined SOCK_CLOEXEC && !defined __ASSUME_SOCK_CLOEXEC
yann@2437:  int __have_sock_cloexec;
yann@2437:  #endif
yann@2437: +
yann@2437: +#if defined O_CLOEXEC && !defined __ASSUME_PIPE2
yann@2437: +int __have_pipe2;
yann@2437: +#endif