[PATCH] device_cgroup: fix unchecked cgroup parent usage

[PATCH] device_cgroup: fix unchecked cgroup parent usage

[Aristeu Rozanski]

In 4cef7299b4786879a3e113e84084a72b24590c5b the cgroup parent usage is
unchecked. root will not have a parent and trying to use
device.{allow,deny} will cause problems. For some reason my stressing
scripts didn't test the root directory so I didn't catch it on my
regular tests.

Andrew, Tejun, this patch needs to make Linus tree ASAP or a revert for
4cef7299b4786879a3e113e84084a72b24590c5b.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: James Morris <jmorris@namei.org>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Serge Hallyn <serge.hallyn@canonical.com>
Cc: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Aristeu Rozanski <aris@redhat.com>

--- github.orig/security/device_cgroup.c	2012-10-26 17:18:01.739366780 -0400
+++ github/security/device_cgroup.c	2012-10-29 10:03:33.221918003 -0400
@@ -352,6 +352,8 @@
  */
 static inline int may_allow_all(struct dev_cgroup *parent)
 {
+	if (!parent)
+		return 1;
 	return parent->behavior == DEVCG_DEFAULT_ALLOW;
 }
 
@@ -376,11 +378,14 @@
 	int count, rc;
 	struct dev_exception_item ex;
 	struct cgroup *p = devcgroup->css.cgroup;
-	struct dev_cgroup *parent = cgroup_to_devcgroup(p->parent);
+	struct dev_cgroup *parent = NULL;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
 
+	if (p->parent)
+		parent = cgroup_to_devcgroup(p->parent);
+
 	memset(&ex, 0, sizeof(ex));
 	b = buffer;
 
@@ -391,11 +396,14 @@
 			if (!may_allow_all(parent))
 				return -EPERM;
 			dev_exception_clean(devcgroup);
+			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
+			if (!parent)
+				break;
+
 			rc = dev_exceptions_copy(&devcgroup->exceptions,
 						 &parent->exceptions);
 			if (rc)
 				return rc;
-			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
 			break;
 		case DEVCG_DENY:
 			dev_exception_clean(devcgroup);

Re: [PATCH] device_cgroup: fix unchecked cgroup parent usage

[Tejun Heo]

On Wed, Oct 31, 2012 at 12:04:30PM -0400, Aristeu Rozanski wrote:
> In 4cef7299b4786879a3e113e84084a72b24590c5b the cgroup parent usage is
> unchecked. root will not have a parent and trying to use
> device.{allow,deny} will cause problems. For some reason my stressing
> scripts didn't test the root directory so I didn't catch it on my
> regular tests.
> 
> Andrew, Tejun, this patch needs to make Linus tree ASAP or a revert for
> 4cef7299b4786879a3e113e84084a72b24590c5b.

 Acked-by: Tejun Heo <tj@kernel.org>

Andrew, can you please pick this one up?

Thanks.

-- 
tejun

Re: [PATCH] device_cgroup: fix unchecked cgroup parent usage

[Serge Hallyn]

Quoting Aristeu Rozanski (aris@redhat.com):
> In 4cef7299b4786879a3e113e84084a72b24590c5b the cgroup parent usage is
> unchecked. root will not have a parent and trying to use
> device.{allow,deny} will cause problems. For some reason my stressing
> scripts didn't test the root directory so I didn't catch it on my
> regular tests.
> 
> Andrew, Tejun, this patch needs to make Linus tree ASAP or a revert for
> 4cef7299b4786879a3e113e84084a72b24590c5b.
> 
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Tejun Heo <tj@kernel.org>
> Cc: Li Zefan <lizefan@huawei.com>
> Cc: James Morris <jmorris@namei.org>
> Cc: Pavel Emelyanov <xemul@openvz.org>
> Cc: Serge Hallyn <serge.hallyn@canonical.com>

Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>

> Cc: Jiri Slaby <jslaby@suse.cz>
> Signed-off-by: Aristeu Rozanski <aris@redhat.com>
> 
> --- github.orig/security/device_cgroup.c	2012-10-26 17:18:01.739366780 -0400
> +++ github/security/device_cgroup.c	2012-10-29 10:03:33.221918003 -0400
> @@ -352,6 +352,8 @@
>   */
>  static inline int may_allow_all(struct dev_cgroup *parent)
>  {
> +	if (!parent)
> +		return 1;
>  	return parent->behavior == DEVCG_DEFAULT_ALLOW;
>  }
>  
> @@ -376,11 +378,14 @@
>  	int count, rc;
>  	struct dev_exception_item ex;
>  	struct cgroup *p = devcgroup->css.cgroup;
> -	struct dev_cgroup *parent = cgroup_to_devcgroup(p->parent);
> +	struct dev_cgroup *parent = NULL;
>  
>  	if (!capable(CAP_SYS_ADMIN))
>  		return -EPERM;
>  
> +	if (p->parent)
> +		parent = cgroup_to_devcgroup(p->parent);
> +
>  	memset(&ex, 0, sizeof(ex));
>  	b = buffer;
>  
> @@ -391,11 +396,14 @@
>  			if (!may_allow_all(parent))
>  				return -EPERM;
>  			dev_exception_clean(devcgroup);
> +			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
> +			if (!parent)
> +				break;
> +
>  			rc = dev_exceptions_copy(&devcgroup->exceptions,
>  						 &parent->exceptions);
>  			if (rc)
>  				return rc;
> -			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
>  			break;
>  		case DEVCG_DENY:
>  			dev_exception_clean(devcgroup);

Re: [PATCH] device_cgroup: fix unchecked cgroup parent usage

[Andrew Morton]

On Wed, 31 Oct 2012 12:04:30 -0400
Aristeu Rozanski <aris@redhat.com> wrote:

> In 4cef7299b4786879a3e113e84084a72b24590c5b the cgroup parent usage is
> unchecked. root will not have a parent and trying to use
> device.{allow,deny} will cause problems.

>From my reading of the code "problems" means "kernel null pointer
dereference".

> For some reason my stressing
> scripts didn't test the root directory so I didn't catch it on my
> regular tests.
> 
> --- github.orig/security/device_cgroup.c	2012-10-26 17:18:01.739366780 -0400
> +++ github/security/device_cgroup.c	2012-10-29 10:03:33.221918003 -0400
> @@ -352,6 +352,8 @@
>   */
>  static inline int may_allow_all(struct dev_cgroup *parent)

offtopic: this function could quite neatly have a bool return type.

>  {
> +	if (!parent)
> +		return 1;

hm.  Does it need a comment explaining what and why?  I guess not...  just.

>  	return parent->behavior == DEVCG_DEFAULT_ALLOW;
>  }

Re: [PATCH] device_cgroup: fix unchecked cgroup parent usage

[Aristeu Rozanski]

On Wed, Oct 31, 2012 at 04:35:01PM -0700, Andrew Morton wrote:
> On Wed, 31 Oct 2012 12:04:30 -0400
> Aristeu Rozanski <aris@redhat.com> wrote:
> 
> > In 4cef7299b4786879a3e113e84084a72b24590c5b the cgroup parent usage is
> > unchecked. root will not have a parent and trying to use
> > device.{allow,deny} will cause problems.
> 
> From my reading of the code "problems" means "kernel null pointer
> dereference".

yes. you want me to resubmit it fixing it or you want to update it?

> >  static inline int may_allow_all(struct dev_cgroup *parent)
> 
> offtopic: this function could quite neatly have a bool return type.

ok, will do.

> >  {
> > +	if (!parent)
> > +		return 1;
> 
> hm.  Does it need a comment explaining what and why?  I guess not...  just.

it's because the root's parent has to always allows full access to all
devices. I can add a comment when I turn into a bool if you want.

-- 
Aristeu