Re: [PATCH] mm/slub: do not place freelist pointer to middle of object if redzone is on

From: Changbin Du <changbin.du@gmail.com>
To: Kees Cook <keescook@chromium.org>
Cc: Changbin Du <changbin.du@gmail.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Christoph Lameter <cl@linux.com>,
	Pekka Enberg <penberg@kernel.org>,
	David Rientjes <rientjes@google.com>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm/slub: do not place freelist pointer to middle of object if redzone is on
Date: Sun, 26 Apr 2020 08:07:27 +0800	[thread overview]
Message-ID: <20200426000727.u2gfxwfsrvme3a6b@mail.google.com> (raw)
In-Reply-To: <202004251547.0F8E6856B@keescook>

Hi Kees,
On Sat, Apr 25, 2020 at 03:48:31PM -0700, Kees Cook wrote:
> On Sat, Apr 25, 2020 at 05:13:38PM +0800, Changbin Du wrote:
> > The recent kernel fails to boot when slub redzone is turned on. This is
> > caused by commit 3202fa62fb ("slub: relocate freelist pointer to middle of
> > object") which relocates freelist pointer to middle of object. In this
> > case, get_track() gets a wrong address and then the redzone is overwritten.
> 
> Hi! A fix for this is already in -next:
> 
> https://www.ozlabs.org/~akpm/mmotm/broken-out/slub-avoid-redzone-when-choosing-freepointer-location.patch
> 
> the above doesn't disable the mitigation when using redzones, so I
> prefer that to this suggested solution.
>
Glade to see it's been reported. But I am sorry that your patch cannot fix it.

With your fix, I suppose the layout of slub is:
|obj-fp-obj|redzone|track|...

While get_track():
	p = object + s->offset + sizeof(void *);

Then we still get a wrong location. I just tested linux-next and the problem is
still there.

Is the right and left redzone good enough to protect the freepointer? If not,
I will send a patch to fix get_track() along with your patch.

> -- 
> Kees Cook

-- 
Cheers,
Changbin Du