[patch 2/5] vfs: O_* bit numbers uniqueness check

[patch 2/5] vfs: O_* bit numbers uniqueness check

[akpm]

From: Wu Fengguang <fengguang.wu@intel.com>

The O_* bit numbers are defined in 20+ arch/*, and can silently overlap. 
Add a compile time check to ensure the uniqueness as suggested by David
Miller.

Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: David Miller <davem@davemloft.net>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Eric Paris <eparis@redhat.com>
Cc: Roland Dreier <rdreier@cisco.com>
Cc: Jamie Lokier <jamie@shareable.org>
Cc: Andreas Schwab <schwab@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 fs/fcntl.c                  |   15 +++++++++++++--
 include/asm-generic/fcntl.h |    4 ++++
 2 files changed, 17 insertions(+), 2 deletions(-)

diff -puN fs/fcntl.c~vfs-o_-bit-numbers-uniqueness-check fs/fcntl.c
--- a/fs/fcntl.c~vfs-o_-bit-numbers-uniqueness-check
+++ a/fs/fcntl.c
@@ -767,11 +767,22 @@ void kill_fasync(struct fasync_struct **
 }
 EXPORT_SYMBOL(kill_fasync);
 
-static int __init fasync_init(void)
+static int __init fcntl_init(void)
 {
+	/* please add new bits here to ensure allocation uniqueness */
+	BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
+		O_RDONLY	| O_WRONLY	| O_RDWR	|
+		O_CREAT		| O_EXCL	| O_NOCTTY	|
+		O_TRUNC		| O_APPEND	| O_NONBLOCK	|
+		__O_SYNC	| O_DSYNC	| FASYNC	|
+		O_DIRECT	| O_LARGEFILE	| O_DIRECTORY	|
+		O_NOFOLLOW	| O_NOATIME	| O_CLOEXEC	|
+		FMODE_EXEC
+		));
+
 	fasync_cache = kmem_cache_create("fasync_cache",
 		sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 	return 0;
 }
 
-module_init(fasync_init)
+module_init(fcntl_init)
diff -puN include/asm-generic/fcntl.h~vfs-o_-bit-numbers-uniqueness-check include/asm-generic/fcntl.h
--- a/include/asm-generic/fcntl.h~vfs-o_-bit-numbers-uniqueness-check
+++ a/include/asm-generic/fcntl.h
@@ -11,6 +11,10 @@
  * -Eric Paris
  */
 
+/*
+ * When introducing new O_* bits, please check its uniqueness in fcntl_init().
+ */
+
 #define O_ACCMODE	00000003
 #define O_RDONLY	00000000
 #define O_WRONLY	00000001
_

Re: [patch 2/5] vfs: O_* bit numbers uniqueness check

[James Bottomley]

On Tue, 2010-07-20 at 15:29 -0700, akpm@linux-foundation.org wrote:
> From: Wu Fengguang <fengguang.wu@intel.com>
> 
> The O_* bit numbers are defined in 20+ arch/*, and can silently overlap. 
> Add a compile time check to ensure the uniqueness as suggested by David
> Miller.

Can we get this reverted or fixed?  It's causing the parisc compiles to
fail.  The reason is O_NONBLOCK on parisc has a dual value:

#define O_NONBLOCK	000200004 /* HPUX has separate NDELAY & NONBLOCK */

The fix would be to take O_NONBLOCK out.

James

---

diff --git a/fs/fcntl.c b/fs/fcntl.c
index 6769fd0..aeb02aa 100644
--- a/fs/fcntl.c
+++ b/fs/fcntl.c
@@ -770,10 +770,11 @@ EXPORT_SYMBOL(kill_fasync);
 static int __init fcntl_init(void)
 {
 	/* please add new bits here to ensure allocation uniqueness */
-	BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
+	BUILD_BUG_ON(18 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
 		O_RDONLY	| O_WRONLY	| O_RDWR	|
 		O_CREAT		| O_EXCL	| O_NOCTTY	|
-		O_TRUNC		| O_APPEND	| O_NONBLOCK	|
+		/* remove O_NONBLOCK it's a two bit define on parisc */
+		O_TRUNC		| O_APPEND	| /* O_NONBLOCK	| */
 		__O_SYNC	| O_DSYNC	| FASYNC	|
 		O_DIRECT	| O_LARGEFILE	| O_DIRECTORY	|
 		O_NOFOLLOW	| O_NOATIME	| O_CLOEXEC	|

Re: [patch 2/5] vfs: O_* bit numbers uniqueness check

[Jamie Lokier]

James Bottomley wrote:
> On Tue, 2010-07-20 at 15:29 -0700, akpm@linux-foundation.org wrote:
> > From: Wu Fengguang <fengguang.wu@intel.com>
> > 
> > The O_* bit numbers are defined in 20+ arch/*, and can silently overlap. 
> > Add a compile time check to ensure the uniqueness as suggested by David
> > Miller.
> 
> Can we get this reverted or fixed?  It's causing the parisc compiles to
> fail.  The reason is O_NONBLOCK on parisc has a dual value:
> 
> #define O_NONBLOCK	000200004 /* HPUX has separate NDELAY & NONBLOCK */
> 
> The fix would be to take O_NONBLOCK out.

A more thoroughly checking fix would be

BUILD_BUG_ON(18 - 1 /* For O_RDONLY being 0 */
             + HWEIGHT32(O_NONBLOCK) /* Because it's 2 bits on parisc */
             != HWEIGHT32(    .... all the bits ....     ));

Am I allowed to Sign-off handwavy pseudocode? ;-)

-- Jamie

Re: [patch 2/5] vfs: O_* bit numbers uniqueness check

[James Bottomley]

On Fri, 2010-08-27 at 02:27 +0100, Jamie Lokier wrote:
> James Bottomley wrote:
> > On Tue, 2010-07-20 at 15:29 -0700, akpm@linux-foundation.org wrote:
> > > From: Wu Fengguang <fengguang.wu@intel.com>
> > > 
> > > The O_* bit numbers are defined in 20+ arch/*, and can silently overlap. 
> > > Add a compile time check to ensure the uniqueness as suggested by David
> > > Miller.
> > 
> > Can we get this reverted or fixed?  It's causing the parisc compiles to
> > fail.  The reason is O_NONBLOCK on parisc has a dual value:
> > 
> > #define O_NONBLOCK	000200004 /* HPUX has separate NDELAY & NONBLOCK */
> > 
> > The fix would be to take O_NONBLOCK out.
> 
> A more thoroughly checking fix would be
> 
> BUILD_BUG_ON(18 - 1 /* For O_RDONLY being 0 */
>              + HWEIGHT32(O_NONBLOCK) /* Because it's 2 bits on parisc */
>              != HWEIGHT32(    .... all the bits ....     ));

Well, it works, but is it wise?  There are several missing flags in this
expression: O_NDELAY being the most noticeable one.  It's defined to be
the sames as O_NONBLOCK on some platforms and not on others.  Our
O_NONBLOCK problem is essentially the same class.

> Am I allowed to Sign-off handwavy pseudocode? ;-)

Nope ... you'd have to code it as a real patch ...

James

[PATCH] vfs: take O_NONBLOCK out of the O_* uniqueness test

[Wu Fengguang]

I tend to agree with James' approach. How about this patch?
Compile tested on x86.

===
vfs: take O_NONBLOCK out of the O_* uniqueness test
From: James Bottomley <James.Bottomley@suse.de>

O_NONBLOCK on parisc has a dual value:

#define O_NONBLOCK	000200004 /* HPUX has separate NDELAY & NONBLOCK */

It is caught by the O_* bits uniqueness check and leads to a parisc
compile error.  The fix would be to take O_NONBLOCK out.

Cc: Jamie Lokier <jamie@shareable.org>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
---
 fs/fcntl.c |   10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

--- linux-next.orig/fs/fcntl.c	2010-08-20 06:57:11.000000000 +0800
+++ linux-next/fs/fcntl.c	2010-09-04 09:16:09.000000000 +0800
@@ -769,11 +769,15 @@ EXPORT_SYMBOL(kill_fasync);
 
 static int __init fcntl_init(void)
 {
-	/* please add new bits here to ensure allocation uniqueness */
-	BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
+	/*
+	 * Please add new bits here to ensure allocation uniqueness.
+	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
+	 * is defined as O_NONBLOCK on some platforms and not on others.
+	 */
+	BUILD_BUG_ON(18 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
 		O_RDONLY	| O_WRONLY	| O_RDWR	|
 		O_CREAT		| O_EXCL	| O_NOCTTY	|
-		O_TRUNC		| O_APPEND	| O_NONBLOCK	|
+		O_TRUNC		| O_APPEND	| /* O_NONBLOCK	| */
 		__O_SYNC	| O_DSYNC	| FASYNC	|
 		O_DIRECT	| O_LARGEFILE	| O_DIRECTORY	|
 		O_NOFOLLOW	| O_NOATIME	| O_CLOEXEC	|