Re: [PATCH 2/2] mingw: ensure temporary file handles are not inherited by child processes

Eric Wong [off-list ref] writes:

O_CLOEXEC only exists since Linux 2.6.23 and there are likely
still LTS (CentOS 5.x?) and non-Linux systems which do not have
it, as well as machines with could have it defined in userspace
headers but not have it in the kernel.

So I suggest something like the following: (untested)

#define GIT_O_TMP (O_RDWR | O_CREAT | O_EXCL)

#ifndef O_CLOEXEC
#  define O_CLOEXEC 0
#endif
	/* state: -1=unknown; 0=broken; 1=working */
	static int cloexec_state = O_CLOEXEC == 0 ? 0 : -1;
	static int GIT_O_ETMP = (GIT_O_TMP | O_CLOEXEC)

	int fd = open(filename, GIT_O_ETMP, 0666);
...
	/*
	 * This is racy in the presence of threads,
	 * but the best we can do for old *nix:
	 */
#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
	if (fd >= 0 && cloexec_state != 1) {
		int flags = fcntl(fd, F_GETFD);

		if (flags == -1)
			die_errno("F_GETFD failed");
		if (flags & O_CLOEXEC)
			cloexec_state = 1;
		else {
			flags = fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
			if (flags == -1)
				die_errno("F_SETFD failed");
			cloexec_state = 0;

		}
	}
#endif
	...

Our codepaths themselves generally do not care about O_CLOEXEC, so
giving a racy emulation of it is not worth the effort, making the
later half of the above an overkill.  Perhaps the three lines to
define O_CLOEXEC to 0 on older UNIX might be sufficient.

Junio C Hamano [off-list ref] wrote:

Our codepaths themselves generally do not care about O_CLOEXEC, so
giving a racy emulation of it is not worth the effort, making the
later half of the above an overkill.

OK.

Perhaps the three lines to
define O_CLOEXEC to 0 on older UNIX might be sufficient.

I'd be more comfortable keeping the EINVAL check that got
snipped in your reply.  O_CLOEXEC can be defined to non-zero in
new userspace headers, but an older kernel chokes on it with
EINVAL.

I've seen this failure in the past with chroots.