Date: Thu, 6 Apr 2017 08:55:14 -0400
From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
To: Tycho Andersen <tycho@docker.com>
Cc: Kees Cook <keescook@chromium.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Alexei Starovoitov <ast@kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Network Development <netdev@vger.kernel.org>,
        "kernel-hardening@lists.openwall.com" 
        <kernel-hardening@lists.openwall.com>,
        =?iso-8859-1?Q?Micka=EBl_Sala=FCn?= <mic@digikod.net>
Subject: Re: [PATCH] ebpf: verify the output of the JIT
Message-ID: <20170406125513.GA56553@ast-mbp.thefacebook.com>
References: <20170404220810.14250-1-tycho@docker.com>
 <CAGXu5jLAXDk1+kYTC60-sJ_SOa5s4+HZ26A0bj67335yAXYhvg@mail.gmail.com>
 <20170405034536.oulhc3ylhlrfgqeg@docker>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20170405034536.oulhc3ylhlrfgqeg@docker>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1562
Lines: 34

On Tue, Apr 04, 2017 at 09:45:36PM -0600, Tycho Andersen wrote:
> Hi Kees,
> 
> On Tue, Apr 04, 2017 at 03:17:57PM -0700, Kees Cook wrote:
> > On Tue, Apr 4, 2017 at 3:08 PM, Tycho Andersen <tycho@docker.com> wrote:
> > > The goal of this patch is to protect the JIT against an attacker with a
> > > write-in-memory primitive. The JIT allocates a buffer which will eventually
> > > be marked +x, so we need to make sure that what was written to this buffer
> > > is what was intended.
> > >
> > > We acheive this by building a hash of the instruction buffer as
> > > instructions are emittted and then comparing that to a hash at the end of
> > > the JIT compile after the buffer has been marked read-only.
> > >
> > > Signed-off-by: Tycho Andersen <tycho@docker.com>
> > > CC: Daniel Borkmann <daniel@iogearbox.net>
> > > CC: Alexei Starovoitov <ast@kernel.org>
> > > CC: Kees Cook <keescook@chromium.org>
> > > CC: Micka?l Sala?n <mic@digikod.net>
> > 
> > Cool! This closes the race condition on producing the JIT vs going
> > read-only. I wonder if it might be possible to make this a more
> > generic interface to the BPF which would be allocate the hash, provide
> > the update callback during emit, and then do the hash check itself at
> > the end of bpf_jit_binary_lock_ro()?
> 
> Yes, probably so. I can look into that for the next version.

Nack.

Please stop wasting yours and our time with buggy code that pretends
to fix a problem that doesn't exist.
This security paranoia around JIT must stop.
Make sure that CONFIG_BPF_JIT is off in your system.