[Buildroot] [PATCH 2/2] util-linux: define mkostemp for uClibc

[Thomas Petazzoni <thomas.petazzoni@free-electrons.com>]

Dear Romain Naour,

On Sun,  6 Apr 2014 11:29:13 +0200, Romain Naour wrote:

> diff --git a/package/util-linux/util-linux-004-define-mkostemp-for-uClibc.patch b/package/util-linux/util-linux-004-define-mkostemp-for-uClibc.patch
> new file mode 100644
> index 0000000..7b49d37
> --- /dev/null
> +++ b/package/util-linux/util-linux-004-define-mkostemp-for-uClibc.patch
> @@ -0,0 +1,29 @@
> +From d339af8fdc4b28175d6986d870f6735a9c2e90a3 Mon Sep 17 00:00:00 2001
> +From: Romain Naour <romain.naour@openwide.fr>
> +Date: Sun, 6 Apr 2014 01:48:58 +0200
> +Subject: [PATCH 1/1] c.h: define mkostemp for uClibc
> +
> +uclibc does not implement mkostemp GNU extension
> +
> +Signed-off-by: Romain Naour <romain.naour@openwide.fr>
> +---
> + include/c.h | 5 +++++
> + 1 file changed, 5 insertions(+)
> +
> +diff --git a/include/c.h b/include/c.h
> +index 4a9bf3d..3ba51c1 100644
> +--- a/include/c.h
> ++++ b/include/c.h
> +@@ -300,4 +300,9 @@ static inline int usleep(useconds_t usec)
> + # define SEEK_HOLE	4
> + #endif
> + 
> ++#ifdef __UCLIBC__
> ++/* uclibc does not implement mkostemp GNU extension */
> ++#define mkostemp(x,y) mkstemp(x)
> ++#endif

Thanks for the patch. However, I believe there are two problems here:

 1/ mkostemp() *can* exist with uClibc: the master branch of uClibc,
    which we use for some architectures (ARC and Xtensa) do have
    mkostemp(). Therefore, there should instead be a test in
    configure.ac like AC_CHECK_FUNCS([mkostemp]) and then make your
    code conditional on #ifndef HAVE_MKOSTEMP.

 2/ The other problem is that in util-linux, mkostemp is used as
    follows:

     fd = mkostemp(n, O_RDWR|O_CREAT|O_EXCL|O_CLOEXEC);

    In your implementation, you discard the flags entirely, by using
    mkstemp(). mkstemp() will pass O_RDWR|O_CREAT|O_EXCL, but not
    O_CLOEXEC, which means that the file descriptor will no longer be
    closed automatically upon exec(). Though, looking at
    http://git.kernel.org/cgit/utils/util-linux/util-linux.git/commit/?id=b1fa3e2234fab95960eaa8499384000f189def13,
    they have switched to using O_CLOEXEC with no specific reason.
    Maybe I'm just worrying too much. But a solution might be to do
    something like:

#ifndef HAVE_MKOSTEMP
static inline int mkostemp(char *template, int flags)
{
	int fd;
	fd = mkstemp(template);
	if (flags & O_CLOEXEC && fd >= 0)
		fcntl(fd, F_SETFD, FD_CLOEXEC);
	return fd;
}
#endif

(Completely untested, not even compiled)

Don't know if we want to worry about this though.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com