Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <1550092329.6911.35.camel@lca.pw>
Subject: Re: [PATCH] slub: untag object before slab end
From:   Qian Cai <cai@lca.pw>
To:     Andrey Konovalov <andreyknvl@google.com>
Cc:     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 Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>
Date:   Wed, 13 Feb 2019 16:12:09 -0500
In-Reply-To: <CAAeHK+w-EWDivYTNiUAeSUVZVGOpUyxbbcC8_nMM1=CcpsJ9Ug@mail.gmail.com>
References: <20190213020550.82453-1-cai@lca.pw>
         <CAAeHK+w-EWDivYTNiUAeSUVZVGOpUyxbbcC8_nMM1=CcpsJ9Ug@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, 2019-02-13 at 11:31 +0100, Andrey Konovalov wrote:
> On Wed, Feb 13, 2019 at 3:06 AM Qian Cai <cai@lca.pw> wrote:
> > 
> > get_freepointer() could return NULL if there is no more free objects in
> > the slab. However, it could return a tagged pointer (like
> > 0x2200000000000000) with KASAN_SW_TAGS which would escape the NULL
> > object checking in check_valid_pointer() and trigger errors below, so
> > untag the object before checking for a NULL object there.
> 
> I think this solution is just masking the issue. get_freepointer()
> shouldn't return tagged NULLs. Apparently when we save a freelist
> pointer, the object where the pointer gets written is tagged
> differently, than this same object when the pointer gets read. I found
> one case where this happens (the last patch out my 5 patch series),
> but apparently there are more.

Well, the problem is that,

__free_slab
  for_each_object(p, s, page_address(page) [1]
    check_object(s, page, p ...)
      get_freepointer(s, p)

[1]: p += s->size

page_address() tags the address using page_kasan_tag(page), so each "p" here has
that tag.

However, at beginning in allocate_slab(), it tags each object with a random tag,
and then calls set_freepointer(s, p, NULL)

As the result, get_freepointer() returns a tagged NULL because it never be able
to obtain the original tag of the object anymore, and this calculation is now
wrong.

return (void *)((unsigned long)ptr ^ s->random ^ ptr_addr);

This also explain why this patch also works, as it unifies the tags.

https://marc.info/?l=linux-mm&m=154955366113951&w=2