[PATCH] 8 bytes kernel memory disclosure in AppleTalk getsockname.

[PATCH] 8 bytes kernel memory disclosure in AppleTalk getsockname.

[Clement LECIGNE]

Hi,

In function atalk_getname(), sockaddr_at is returned in userland without
zero'ing the "char sat_zero[8]" field. This bug allows user to display 8
bytes leaked from the kernel stack.

Here is a patch that zero the whole sockaddr_at structure before
processing it. It should fix this bug.

Signed-off-by: Clément Lecigne <clement.lecigne@netasq.com>
--- linux/net/appletalk/ddp.c	2009-08-26 11:35:59.000000000 +0200
+++ linux/net/appletalk/ddp.c	2009-08-26 11:36:30.000000000 +0200
@@ -1241,6 +1241,8 @@ static int atalk_getname(struct socket *
 		if (atalk_autobind(sk) < 0)
 			return -ENOBUFS;
 
+	memset(&sat, 0, sizeof(struct sockaddr_at));
+
 	*uaddr_len = sizeof(struct sockaddr_at);
 
 	if (peer) {

-- 
Clément LECIGNE,
-Only one remote hole in the default install, in more than 10 years!<br>
+Only two remote holes in the default install, in more than 10 years!<br>

Re: [PATCH] 8 bytes kernel memory disclosure in AppleTalk getsockname.

[Tetsuo Handa]

Clement LECIGNE wrote:
> Here is a patch that zero the whole sockaddr_at structure before
> processing it. It should fix this bug.

Same kind of bug at http://lkml.org/lkml/2006/7/25/51 .
I wonder why not zero'ing sockaddr at the caller rather than
zero'ing at individual ->getname(). Doing so will save some lines.

Re: [PATCH] 8 bytes kernel memory disclosure in AppleTalk getsockname.

[Eric Dumazet]

Clement LECIGNE a écrit :
> Hi,
> 
> In function atalk_getname(), sockaddr_at is returned in userland without
> zero'ing the "char sat_zero[8]" field. This bug allows user to display 8
> bytes leaked from the kernel stack.
> 
> Here is a patch that zero the whole sockaddr_at structure before
> processing it. It should fix this bug.
> 
> Signed-off-by: Clément Lecigne <clement.lecigne@netasq.com>
> --- linux/net/appletalk/ddp.c	2009-08-26 11:35:59.000000000 +0200
> +++ linux/net/appletalk/ddp.c	2009-08-26 11:36:30.000000000 +0200
> @@ -1241,6 +1241,8 @@ static int atalk_getname(struct socket *
>  		if (atalk_autobind(sk) < 0)
>  			return -ENOBUFS;
>  
> +	memset(&sat, 0, sizeof(struct sockaddr_at));
> +
>  	*uaddr_len = sizeof(struct sockaddr_at);
>  
>  	if (peer) {
> 
Hi Clement

Well, I submitted same patch some weeks ago and I just checked that
it was already in Linus tree.

author	Eric Dumazet <eric.dumazet@gmail.com>
	Thu, 6 Aug 2009 02:27:43 +0000 (02:27 +0000)
committer	David S. Miller <davem@davemloft.net>
	Thu, 6 Aug 2009 20:08:45 +0000 (13:08 -0700)
commit	3d392475c873c10c10d6d96b94d092a34ebd4791

appletalk: fix atalk_getname() leak

atalk_getname() can leak 8 bytes of kernel memory to user

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>


Dont worry, it'll be included in upcoming 2.6.31 kernel,
and backported to previous ones as well.

Re: [PATCH] 8 bytes kernel memory disclosure in AppleTalk getsockname.

[Eric Dumazet]

Tetsuo Handa a écrit :
> Clement LECIGNE wrote:
>> Here is a patch that zero the whole sockaddr_at structure before
>> processing it. It should fix this bug.
> 
> Same kind of bug at http://lkml.org/lkml/2006/7/25/51 .
> I wonder why not zero'ing sockaddr at the caller rather than
> zero'ing at individual ->getname(). Doing so will save some lines.
> --

It would not help zeroing buffer before calling ->getname(), since
getname() will overwrite it with a memcpy(buffer, &somestruct, sizeof(somestruct));

Re: [PATCH] 8 bytes kernel memory disclosure in AppleTalk getsockname.

[Clement LECIGNE]

Wed, Aug 26, 2009 at 02:35:57PM +0200, Eric Dumazet wrote:
> Clement LECIGNE a écrit :
> > Hi,
> > 
> > In function atalk_getname(), sockaddr_at is returned in userland without
> > zero'ing the "char sat_zero[8]" field. This bug allows user to display 8
> > bytes leaked from the kernel stack.
> > 
> > Here is a patch that zero the whole sockaddr_at structure before
> > processing it. It should fix this bug.
> > 
> > Signed-off-by: Clément Lecigne <clement.lecigne@netasq.com>
> > --- linux/net/appletalk/ddp.c	2009-08-26 11:35:59.000000000 +0200
> > +++ linux/net/appletalk/ddp.c	2009-08-26 11:36:30.000000000 +0200
> > @@ -1241,6 +1241,8 @@ static int atalk_getname(struct socket *
> >  		if (atalk_autobind(sk) < 0)
> >  			return -ENOBUFS;
> >  
> > +	memset(&sat, 0, sizeof(struct sockaddr_at));
> > +
> >  	*uaddr_len = sizeof(struct sockaddr_at);
> >  
> >  	if (peer) {
> > 
> Hi Clement
> 
> Well, I submitted same patch some weeks ago and I just checked that
> it was already in Linus tree.
> 
> author	Eric Dumazet <eric.dumazet@gmail.com>
> 	Thu, 6 Aug 2009 02:27:43 +0000 (02:27 +0000)
> committer	David S. Miller <davem@davemloft.net>
> 	Thu, 6 Aug 2009 20:08:45 +0000 (13:08 -0700)
> commit	3d392475c873c10c10d6d96b94d092a34ebd4791
> 
> appletalk: fix atalk_getname() leak
> 
> atalk_getname() can leak 8 bytes of kernel memory to user
> 
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> Signed-off-by: David S. Miller <davem@davemloft.net>
> 
> 
> Dont worry, it'll be included in upcoming 2.6.31 kernel,
> and backported to previous ones as well.

Hi Eric,

Oups, shame on me, I have not checked Linus tree before submitting the
patch.

Sorry,

-- 
Clément LECIGNE,
-Only one remote hole in the default install, in more than 10 years!<br>
+Only two remote holes in the default install, in more than 10 years!<br>