[PATCH kobjects] Fix a rare memory leak in kobject_set_name_vargs

[PATCH kobjects] Fix a rare memory leak in kobject_set_name_vargs

[Philippe De Muyter]

Hello Greg,

This is a possible memory leak that I discovered only by accidental code
reading.

--

If kvasprintf fails in kobject_set_name_vargs, the memory used by
the original kobj->name is leaked.  Fix that.  I also avoid useless
memory accesses to kobj->name by using the local variables old_name
and new_name instead.

Signed-off-by: Philippe De Muyter <phdm@macqel.be>

diff -r 373fdd3df333 linux-2.6.x/lib/kobject.c
--- a/linux-2.6.x/lib/kobject.c	Wed Aug 19 23:26:44 2009 +0200
+++ b/linux-2.6.x/lib/kobject.c	Sat Mar 13 13:35:43 2010 +0100
@@ -216,20 +216,22 @@ int kobject_set_name_vargs(struct kobjec
 				  va_list vargs)
 {
 	const char *old_name = kobj->name;
+	char *new_name;
 	char *s;
 
-	if (kobj->name && !fmt)
+	if (old_name && !fmt)
 		return 0;
 
-	kobj->name = kvasprintf(GFP_KERNEL, fmt, vargs);
-	if (!kobj->name)
+	new_name = kvasprintf(GFP_KERNEL, fmt, vargs);
+	if (!new_name)
 		return -ENOMEM;
 
 	/* ewww... some of these buggers have '/' in the name ... */
-	while ((s = strchr(kobj->name, '/')))
+	while ((s = strchr(new_name, '/')))
 		s[0] = '!';
 
 	kfree(old_name);
+	kobj->name = new_name;
 	return 0;
 }
 

Re: [PATCH kobjects] Fix a rare memory leak in kobject_set_name_vargs

[Robert P. J. Day]

On Sat, 13 Mar 2010, Philippe De Muyter wrote:

> Hello Greg,
>
> This is a possible memory leak that I discovered only by accidental code
> reading.
>
> --
>
> If kvasprintf fails in kobject_set_name_vargs, the memory used by
> the original kobj->name is leaked.  Fix that.  I also avoid useless
> memory accesses to kobj->name by using the local variables old_name
> and new_name instead.
>
> Signed-off-by: Philippe De Muyter <phdm@macqel.be>
>
> diff -r 373fdd3df333 linux-2.6.x/lib/kobject.c
> --- a/linux-2.6.x/lib/kobject.c	Wed Aug 19 23:26:44 2009 +0200
> +++ b/linux-2.6.x/lib/kobject.c	Sat Mar 13 13:35:43 2010 +0100
> @@ -216,20 +216,22 @@ int kobject_set_name_vargs(struct kobjec
>  				  va_list vargs)
>  {
>  	const char *old_name = kobj->name;
> +	char *new_name;
>  	char *s;
>
> -	if (kobj->name && !fmt)
> +	if (old_name && !fmt)
>  		return 0;
>
> -	kobj->name = kvasprintf(GFP_KERNEL, fmt, vargs);
> -	if (!kobj->name)
> +	new_name = kvasprintf(GFP_KERNEL, fmt, vargs);
> +	if (!new_name)
>  		return -ENOMEM;
>
>  	/* ewww... some of these buggers have '/' in the name ... */
> -	while ((s = strchr(kobj->name, '/')))
> +	while ((s = strchr(new_name, '/')))
>  		s[0] = '!';
>
>  	kfree(old_name);
> +	kobj->name = new_name;
>  	return 0;
>  }

  the routine kobject_set_name_vargs() is described in
Documentation/kobject.txt as "legacy cruft" to be removed at some
point, so it's not clear there's any value in "fixing" it.

rday
--

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA

            Linux Consulting, Training and Kernel Pedantry.

Web page:                                          http://crashcourse.ca
Twitter:                                       http://twitter.com/rpjday
========================================================================

Re: [PATCH kobjects] Fix a rare memory leak in kobject_set_name_vargs

[Eric Dumazet]

Le samedi 13 mars 2010 à 07:53 -0500, Robert P. J. Day a écrit :
> On Sat, 13 Mar 2010, Philippe De Muyter wrote:
> 
> > Hello Greg,
> >
> > This is a possible memory leak that I discovered only by accidental code
> > reading.
> >
> > --
> >
> > If kvasprintf fails in kobject_set_name_vargs, the memory used by
> > the original kobj->name is leaked.  Fix that.  I also avoid useless
> > memory accesses to kobj->name by using the local variables old_name
> > and new_name instead.
> >
> > Signed-off-by: Philippe De Muyter <phdm@macqel.be>
> >
> > diff -r 373fdd3df333 linux-2.6.x/lib/kobject.c
> > --- a/linux-2.6.x/lib/kobject.c	Wed Aug 19 23:26:44 2009 +0200
> > +++ b/linux-2.6.x/lib/kobject.c	Sat Mar 13 13:35:43 2010 +0100
> > @@ -216,20 +216,22 @@ int kobject_set_name_vargs(struct kobjec
> >  				  va_list vargs)
> >  {
> >  	const char *old_name = kobj->name;
> > +	char *new_name;
> >  	char *s;
> >
> > -	if (kobj->name && !fmt)
> > +	if (old_name && !fmt)
> >  		return 0;
> >
> > -	kobj->name = kvasprintf(GFP_KERNEL, fmt, vargs);
> > -	if (!kobj->name)
> > +	new_name = kvasprintf(GFP_KERNEL, fmt, vargs);
> > +	if (!new_name)
> >  		return -ENOMEM;
> >
> >  	/* ewww... some of these buggers have '/' in the name ... */
> > -	while ((s = strchr(kobj->name, '/')))
> > +	while ((s = strchr(new_name, '/')))
> >  		s[0] = '!';
> >
> >  	kfree(old_name);
> > +	kobj->name = new_name;
> >  	return 0;
> >  }
> 
>   the routine kobject_set_name_vargs() is described in
> Documentation/kobject.txt as "legacy cruft" to be removed at some
> point, so it's not clear there's any value in "fixing" it.
> 

Given I submitted a similar patch two days before, I guess a fix would
be welcome or else we might see one or two attempts per week from
various people.

http://lkml.org/lkml/2010/3/11/438

Legacy or not, this code looks wrong. I caught it while looking for
kmemleaks reports on my dev machine, that were triggered by
CONFIG_NO_BOOTMEM use.

Re: [PATCH kobjects] Fix a rare memory leak in kobject_set_name_vargs

[Robert P. J. Day]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 2699 bytes --]

On Sat, 13 Mar 2010, Eric Dumazet wrote:

> Le samedi 13 mars 2010 à 07:53 -0500, Robert P. J. Day a écrit :
> > On Sat, 13 Mar 2010, Philippe De Muyter wrote:
> >
> > > Hello Greg,
> > >
> > > This is a possible memory leak that I discovered only by accidental code
> > > reading.
> > >
> > > --
> > >
> > > If kvasprintf fails in kobject_set_name_vargs, the memory used by
> > > the original kobj->name is leaked.  Fix that.  I also avoid useless
> > > memory accesses to kobj->name by using the local variables old_name
> > > and new_name instead.
> > >
> > > Signed-off-by: Philippe De Muyter <phdm@macqel.be>
> > >
> > > diff -r 373fdd3df333 linux-2.6.x/lib/kobject.c
> > > --- a/linux-2.6.x/lib/kobject.c	Wed Aug 19 23:26:44 2009 +0200
> > > +++ b/linux-2.6.x/lib/kobject.c	Sat Mar 13 13:35:43 2010 +0100
> > > @@ -216,20 +216,22 @@ int kobject_set_name_vargs(struct kobjec
> > >  				  va_list vargs)
> > >  {
> > >  	const char *old_name = kobj->name;
> > > +	char *new_name;
> > >  	char *s;
> > >
> > > -	if (kobj->name && !fmt)
> > > +	if (old_name && !fmt)
> > >  		return 0;
> > >
> > > -	kobj->name = kvasprintf(GFP_KERNEL, fmt, vargs);
> > > -	if (!kobj->name)
> > > +	new_name = kvasprintf(GFP_KERNEL, fmt, vargs);
> > > +	if (!new_name)
> > >  		return -ENOMEM;
> > >
> > >  	/* ewww... some of these buggers have '/' in the name ... */
> > > -	while ((s = strchr(kobj->name, '/')))
> > > +	while ((s = strchr(new_name, '/')))
> > >  		s[0] = '!';
> > >
> > >  	kfree(old_name);
> > > +	kobj->name = new_name;
> > >  	return 0;
> > >  }
> >
> >   the routine kobject_set_name_vargs() is described in
> > Documentation/kobject.txt as "legacy cruft" to be removed at some
> > point, so it's not clear there's any value in "fixing" it.
> >
>
> Given I submitted a similar patch two days before, I guess a fix
> would be welcome or else we might see one or two attempts per week
> from various people.
>
> http://lkml.org/lkml/2010/3/11/438
>
> Legacy or not, this code looks wrong. I caught it while looking for
> kmemleaks reports on my dev machine, that were triggered by
> CONFIG_NO_BOOTMEM use.

  fair enough, i was just going off of what i read in Doc/kobject.txt.

rday
--

========================================================================
Robert P. J. Day                               Waterloo, Ontario, CANADA

            Linux Consulting, Training and Kernel Pedantry.

Web page:                                          http://crashcourse.ca
Twitter:                                       http://twitter.com/rpjday
========================================================================