[PATCH] SYSFS: Fix return values for sysdev_store_{ulong,int}

[PATCH] SYSFS: Fix return values for sysdev_store_{ulong,int}

[Andi Kleen]

SYSFS: Fix return values for sysdev_store_{ulong,int}

Always return the full size instead of the consumed
length of the string in sysdev_store_{ulong,int}

This avoids EINVAL errors in some echo versions.

Signed-off-by: Andi Kleen <ak@linux.intel.com>

Index: linux/drivers/base/sys.c
===================================================================
--- linux.orig/drivers/base/sys.c
+++ linux/drivers/base/sys.c
@@ -488,7 +488,8 @@ ssize_t sysdev_store_ulong(struct sys_de
 	if (end == buf)
 		return -EINVAL;
 	*(unsigned long *)(ea->var) = new;
-	return end - buf;
+	/* Always return full write size even if we didn't consume all */
+	return size;
 }
 EXPORT_SYMBOL_GPL(sysdev_store_ulong);
 
@@ -511,7 +512,8 @@ ssize_t sysdev_store_int(struct sys_devi
 	if (end == buf || new > INT_MAX || new < INT_MIN)
 		return -EINVAL;
 	*(int *)(ea->var) = new;
-	return end - buf;
+	/* Always return full write size even if we didn't consume all */
+	return size;
 }
 EXPORT_SYMBOL_GPL(sysdev_store_int);
 

Re: [PATCH] SYSFS: Fix return values for sysdev_store_{ulong,int}

[Andrew Morton]

On Mon, 13 Oct 2008 12:03:03 +0200
Andi Kleen <andi@firstfloor.org> wrote:

> SYSFS: Fix return values for sysdev_store_{ulong,int}
> 
> Always return the full size instead of the consumed
> length of the string in sysdev_store_{ulong,int}
> 
> This avoids EINVAL errors in some echo versions.

That sounds fairly serious.  Is this a regression or are those versions
of echo really ancient, or...?

Re: [PATCH] SYSFS: Fix return values for sysdev_store_{ulong,int}

[Andi Kleen]

On Thu, Oct 16, 2008 at 01:35:30PM -0700, Andrew Morton wrote:
> On Mon, 13 Oct 2008 12:03:03 +0200
> Andi Kleen <andi@firstfloor.org> wrote:
> 
> > SYSFS: Fix return values for sysdev_store_{ulong,int}
> > 
> > Always return the full size instead of the consumed
> > length of the string in sysdev_store_{ulong,int}
> > 
> > This avoids EINVAL errors in some echo versions.
> 
> That sounds fairly serious.  Is this a regression or are those versions
> of echo really ancient, or...?

It's a regression, but it currently only affects the mce sysfs files,
which are the only ones using it. Also the EINVAL is more a cosmetic
issue because the first write will already change the value successfully,
only the second one will fail but not change anything.

-Andi

-- 
ak@linux.intel.com