linux-kernel - Re: Questions about commit "ipc/shm: Fix shmat mmap nil-page protection"

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 25 Sep 2017 23:44:38 +0200
From:   Andrea Arcangeli <aarcange@...hat.com>
To:     Joe Lawrence <joe.lawrence@...hat.com>
Cc:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Davidlohr Bueso <dave@...olabs.net>, linux-mm@...ck.org
Subject: Re: Questions about commit "ipc/shm: Fix shmat mmap nil-page
 protection"

Hello,

On Mon, Sep 25, 2017 at 03:38:07PM -0400, Joe Lawrence wrote:
> Hi Davidlohr,
> 
> I was looking into backporting commit 95e91b831f87 ("ipc/shm: Fix shmat
> mmap nil-page protection") to a distro kernel and Andrea brought up some
> interesting questions about that change.
> 
> We saw that a LTP test [1] was added some time ago to reproduce behavior
> matching that of the original report [2].  However, Andrea and I are a
> little confused about that original report and what the upstream commit
> was intended to fix.  A quick summary of our offlist discussion:
> 
> - This is only about privileged users (and no SELinux).
> 
> - We modified the 20170119_shmat_nullpage_poc.c reproducer from [2] to
>   include MAP_FIXED to prove (as root, no SELinux):
> 
>     It is possible to mmap 0
>     It is NOT possible to mmap 1
> 
> - Andrea points out that mmap(1, ...) fails not because of any
>   mmap_min_addr checks, but for alignment reasons.
> 
> - He also wonders about other bogus addr values above 4k, but below
>   mmap_min_addr and whether this change misses those values

Yes, thanks for the accurate summary Joe.

> Is it possible that the original report noticed that shmat allowed
> attach to an address of 1, and it was assumed that somehow mmap_min_addr
> protections were circumvented?  Then commit 95e91b831f87 modified the
> rounding in do_shmat() so that shmat would fail on similar input (but
> for apparently different reasons)?
> 
> I didn't see any discussion when looking up the original commit in the
> list archives, so any explanations or pointers would be very helpful.

We identified only one positive side effect to such change, it is
about the semantics of SHM_REMAP when addr < shmlba (and != 0). Before
the patch SHM_REMAP was erroneously implicit for that virtual
range. However that's not security related either, and there's no
mention of SHM_REMAP in the commit message.

So then we wondered what this CVE is about in the first place, it
looks a invalid CVE for a not existent security issue. The testcase at
least shows no malfunction, mapping addr 0 is fine to succeed with
CAP_SYS_RAWIO.

>From the commit message, testcase and CVE I couldn't get what this
commit is about.

Last but not the least, if there was a security problem in calling
do_mmap_pgoff with addr=0, flags=MAP_FIXED|MAP_SHARED the fix would
better be moved to do_mmap_pgoff, not in ipc/shm.c.

Thanks,
Andrea