[PATCH] do not export kernel's NULL #define to userspace

[PATCH] do not export kernel's NULL #define to userspace

[Lubos Lunak]

GCC's NULL is actually __null, which allows detecting some questionable
NULL usage and warn about it. Moreover each platform/compiler should have
its own stddef.h anyway (which is different from linux/stddef.h).
So there's no good reason to leak kernel's NULL to userspace and
override what the compiler provides.

Signed-off-by: Luboš Luňák <l.lunak@suse.cz>
---
 include/linux/stddef.h |    8 ++------
 1 files changed, 2 insertions(+), 6 deletions(-)

diff --git a/include/linux/stddef.h b/include/linux/stddef.h
index 6a40c76..1747b67 100644
--- a/include/linux/stddef.h
+++ b/include/linux/stddef.h
@@ -3,14 +3,10 @@
 
 #include <linux/compiler.h>
 
+#ifdef __KERNEL__
+
 #undef NULL
-#if defined(__cplusplus)
-#define NULL 0
-#else
 #define NULL ((void *)0)
-#endif
-
-#ifdef __KERNEL__
 
 enum {
 	false	= 0,
-- 
1.7.3.4
-- 
 Lubos Lunak
 l.lunak@suse.cz

Re: [PATCH] do not export kernel's NULL #define to userspace

[Arnd Bergmann]

On Wednesday 21 March 2012, Lubos Lunak wrote:
> GCC's NULL is actually __null, which allows detecting some questionable
> NULL usage and warn about it. Moreover each platform/compiler should have
> its own stddef.h anyway (which is different from linux/stddef.h).
> So there's no good reason to leak kernel's NULL to userspace and
> override what the compiler provides.
> 
> Signed-off-by: Luboš Luňák <l.lunak@suse.cz>

Yes, this looks like a bug, but I'm not sure that the solution is
sufficient. Have you checked the other exported header files
for whether they use NULL after including linux/stddef.h?

If so, we might have to replace it with a __KERNEL_NULL constant
or something, like we do for the stuff in linux/types.h, so we
don't accidentally break user applications that rely on the
header files to be self-contained. 

I think there is at least a NULL usage in linux/wireless.h and some
netfilter headers.

	Arnd

Re: [PATCH] do not export kernel's NULL #define to userspace

[Lubos Lunak]

[-- Attachment #1: Type: text/plain, Size: 1378 bytes --]

On Wednesday 21 of March 2012, Arnd Bergmann wrote:
> On Wednesday 21 March 2012, Lubos Lunak wrote:
> > GCC's NULL is actually __null, which allows detecting some questionable
> > NULL usage and warn about it. Moreover each platform/compiler should have
> > its own stddef.h anyway (which is different from linux/stddef.h).
> > So there's no good reason to leak kernel's NULL to userspace and
> > override what the compiler provides.
> >
> > Signed-off-by: Luboš Luňák <l.lunak@suse.cz>
>
> Yes, this looks like a bug, but I'm not sure that the solution is
> sufficient. Have you checked the other exported header files
> for whether they use NULL after including linux/stddef.h?

 I have checked that the kernel builds with this change, and I have checked 
that the whole of LibreOffice compiles after the change. I'm not sure what 
you mean exactly, are you asking me to manually inspect everything under 
include/linux for NULL usage?

> If so, we might have to replace it with a __KERNEL_NULL constant
> or something, like we do for the stuff in linux/types.h, so we
> don't accidentally break user applications that rely on the
> header files to be self-contained.
>
> I think there is at least a NULL usage in linux/wireless.h and some
> netfilter headers.

 I see. How about the attached patch then?

-- 
 Lubos Lunak
 l.lunak@suse.cz

[-- Attachment #2: 0001-do-not-export-kernel-s-NULL-define-to-userspace.patch --]
[-- Type: text/x-diff, Size: 1473 bytes --]

From 9be4118ed1bc6221d4f2a73cad69de33285b0b2a Mon Sep 17 00:00:00 2001
From: Lubos Lunak <l.lunak@suse.cz>
Date: Wed, 21 Mar 2012 14:04:31 +0100
Subject: [PATCH] do not export kernel's NULL #define to userspace
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

GCC's NULL is actually __null, which allows detecting some questionable
NULL usage and warn about it. Moreover each platform/compiler should have
its own stddef.h anyway (which is different from linux/stddef.h).
So there's no good reason to leak kernel's NULL to userspace and
override what the compiler provides. Include <stddef.h> in that
case to make sure linux headers are self-contained in userspace as well.

Signed-off-by: Luboš Luňák <l.lunak@suse.cz>
---
 include/linux/stddef.h |   10 ++++------
 1 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/include/linux/stddef.h b/include/linux/stddef.h
index 6a40c76..1db853b 100644
--- a/include/linux/stddef.h
+++ b/include/linux/stddef.h
@@ -3,14 +3,10 @@
 
 #include <linux/compiler.h>
 
+#ifdef __KERNEL__
+
 #undef NULL
-#if defined(__cplusplus)
-#define NULL 0
-#else
 #define NULL ((void *)0)
-#endif
-
-#ifdef __KERNEL__
 
 enum {
 	false	= 0,
@@ -23,6 +19,8 @@ enum {
 #else
 #define offsetof(TYPE, MEMBER) ((size_t) &((TYPE *)0)->MEMBER)
 #endif
+#else /* __KERNEL__ */
+#include <stddef.h>
 #endif /* __KERNEL__ */
 
 #endif
-- 
1.7.3.4

Re: [PATCH] do not export kernel's NULL #define to userspace

[Arnd Bergmann]

On Wednesday 21 March 2012, Lubos Lunak wrote:
> > If so, we might have to replace it with a __KERNEL_NULL constant
> > or something, like we do for the stuff in linux/types.h, so we
> > don't accidentally break user applications that rely on the
> > header files to be self-contained.
> >
> > I think there is at least a NULL usage in linux/wireless.h and some
> > netfilter headers.
> 
>  I see. How about the attached patch then?

Strictly speaking, you should not include standard headers from kernel
provided headers, and the problems would be similar to those before
your patch: anyone who currently doesn't include <stddef.h> but has
their own definition of NULL will still get a conflict from including
a kernel header that includes <linux/stddef.h>.

	Arnd

Re: [PATCH] do not export kernel's NULL #define to userspace

[Lubos Lunak]

[-- Attachment #1: Type: text/plain, Size: 1444 bytes --]

On Wednesday 21 of March 2012, Arnd Bergmann wrote:
> On Wednesday 21 March 2012, Lubos Lunak wrote:
> > > If so, we might have to replace it with a __KERNEL_NULL constant
> > > or something, like we do for the stuff in linux/types.h, so we
> > > don't accidentally break user applications that rely on the
> > > header files to be self-contained.
> > >
> > > I think there is at least a NULL usage in linux/wireless.h and some
> > > netfilter headers.
> >
> >  I see. How about the attached patch then?
>
> Strictly speaking, you should not include standard headers from kernel
> provided headers, and the problems would be similar to those before
> your patch: anyone who currently doesn't include <stddef.h> but has
> their own definition of NULL will still get a conflict from including
> a kernel header that includes <linux/stddef.h>.

 I guess I should point out that I'm not a kernel developer, I simply want to 
fix the problem that kernel headers redefine NULL to something suboptimal in 
userspace. So I don't know what the requirements on the headers are from the 
kernel side (and I wonder why you need your own NULL in the kernel when it 
comes with the compiler).

 Let me provide one more patch then, which only surrounds the NULL definition 
by #ifndef NULL instead of bluntly doing #undef NULL. That way kernel should 
keep using this NULL definition, while it won't be forced in userspace.

-- 
 Lubos Lunak
 l.lunak@suse.cz

[-- Attachment #2: 0001-do-not-redefine-userspace-s-NULL-define.patch --]
[-- Type: text/x-diff, Size: 1210 bytes --]

From 96e2b6caa5c52daade79635270ce96ce764fcd31 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Lubo=C5=A1=20Lu=C5=88=C3=A1k?= <l.lunak@suse.cz>
Date: Wed, 21 Mar 2012 19:32:02 +0100
Subject: [PATCH] do not redefine userspace's NULL #define
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

GCC's NULL is actually __null, which allows detecting some questionable
NULL usage and warn about it. Moreover each platform/compiler should have
its own stddef.h anyway (which is different from linux/stddef.h).
So there's no good reason to override what the compiler provides.
Keep the #define conditionally, in order to keep the headers self-contained.

Signed-off-by: Luboš Luňák <l.lunak@suse.cz>
---
 include/linux/stddef.h |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/include/linux/stddef.h b/include/linux/stddef.h
index 6a40c76..ce225a9 100644
--- a/include/linux/stddef.h
+++ b/include/linux/stddef.h
@@ -3,12 +3,13 @@
 
 #include <linux/compiler.h>
 
-#undef NULL
+#ifndef NULL
 #if defined(__cplusplus)
 #define NULL 0
 #else
 #define NULL ((void *)0)
 #endif
+#endif
 
 #ifdef __KERNEL__
 
-- 
1.7.7