Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 26 Jan 2022 08:40:49 -0800
From:   Kees Cook <keescook@chromium.org>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     Ariadne Conill <ariadne@dereferenced.org>,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        Eric Biederman <ebiederm@xmission.com>,
        Alexander Viro <viro@zeniv.linux.org.uk>
Subject: Re: [PATCH v2] fs/exec: require argv[0] presence in
 do_execveat_common()
Message-ID: <202201260832.CCC8BB9@keescook>
References: <20220126114447.25776-1-ariadne@dereferenced.org>
 <YfFh6O2JS6MybamT@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YfFh6O2JS6MybamT@casper.infradead.org>
Precedence: bulk

On Wed, Jan 26, 2022 at 02:59:52PM +0000, Matthew Wilcox wrote:
> On Wed, Jan 26, 2022 at 11:44:47AM +0000, Ariadne Conill wrote:
> > Interestingly, Michael Kerrisk opened an issue about this in 2008[1],
> > but there was no consensus to support fixing this issue then.
> > Hopefully now that CVE-2021-4034 shows practical exploitative use
> > of this bug in a shellcode, we can reconsider.
> > 
> > [0]: https://pubs.opengroup.org/onlinepubs/9699919799/functions/exec.html
> > [1]: https://bugzilla.kernel.org/show_bug.cgi?id=8408
> 
> Having now read 8408 ... if ABI change is a concern (and I really doubt
> it is), we could treat calling execve() with a NULL argv as if the
> caller had passed an array of length 1 with the first element set to
> NULL.  Just like we reopen fds 0,1,2 for suid execs if they were closed.

I was having similar thoughts this morning. We can't actually change the
argc, though, because of the various tests (see the debian code search
links) that explicitly tests for argc == 0 in the child. But, the flaw
is not the count, but rather that argv == argp in the argc == 0 case.
(Or that argv NULL-checking iteration begins at argv[1].)

But that would could fix easily by just adding an extra NULL. e.g.:

Currently:

argc = 1
argv = "foo", NULL
envp = "bar=baz", ..., NULL

argc = 0
argv = NULL
envp = "bar=baz", ..., NULL

We could just make the argc = 0 case be:

argc = 0
argv = NULL, NULL
envp = "bar=baz", ..., NULL

We need to be careful with the stack utilization counts, though, so I'm
thinking we could actually make this completely unconditional and just
pad envp by 1 NULL on the user stack:

argv = "what", "ever", NULL
       NULL
envp = "bar=baz", ..., NULL

My only concern there is that there may be some code out there that
depends on envp immediately following the trailing argv NULL, so I think
my preference would be to pad only in the argc == 0 case and correctly
manage the stack utilization.

-- 
Kees Cook