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;
MIME-Version: 1.0
References: <1538943795-30895-1-git-send-email-wang6495@umn.edu>
 <CAPhsuW5vnVjRS7DWFYPSmq5G-dAFxOrHCAOMkK-iaOoirzMazQ@mail.gmail.com> <43898.1539033428@turing-police.cc.vt.edu>
In-Reply-To: <43898.1539033428@turing-police.cc.vt.edu>
From:   Song Liu <liu.song.a23@gmail.com>
Date:   Mon, 8 Oct 2018 17:44:46 -0700
Message-ID: <CAPhsuW5nPGMOEMKa1UFwgquoUeYiUV8i8nVgWfM6McH2Ri1esg@mail.gmail.com>
Subject: Re: [PATCH] bpf: btf: Fix a missing check bug
To:     valdis.kletnieks@vt.edu
Cc:     wang6495@umn.edu, kjlu@umn.edu,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Networking <netdev@vger.kernel.org>,
        open list <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Oct 8, 2018 at 2:17 PM <valdis.kletnieks@vt.edu> wrote:
>
> On Mon, 08 Oct 2018 13:51:09 -0700, Song Liu said:
> > On Sun, Oct 7, 2018 at 1:26 PM Wenwen Wang <wang6495@umn.edu> wrote:
>
> > > same value. Given that the btf data is in the user space, a malicious user
> > > can race to change the data between the two copies. By doing so, the user
> > > can provide malicious data to the kernel and cause undefined behavior.
>
> > These two numbers are copied from same memory location, right? So I
> > think this check is not necessary?
>
> Security researchers call this a TOCTOU bug - Time of Check - Time of Use.
>
> What can happen:
>
> 1) We fetch the value  (say we get 90) from userspace and stash it in hdr_len.
>
> 2) We do some other stuff like check the hdr_len isn't too big, etc..
>
>         meanwhile, on another CPU running another thread of the process...
>                 3) malicious code stuffs a 117 into that field
>
> 4) We fetch the entire header, incliding a now-changed hdr_len  (now 117) and
> stick it in btf->hdr->hdr_len.
>
> 5) Any code that assumes that hdr_len and btf->hdr->hdr_len are the same value
> explodes in interesting ways.

I think I get the security concept here. However, hdr_len here is only used to
copy the whole header into kernel space, and it is not used in other
logic at all.
I cannot image any security flaw with either hdr_len > btf->hdr->hdr_len case or
hdr_len < btf->hdr->hdr_len. Could you please provide more insights on what
would break by malicious user space?

Thanks,
Song