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;
Date:   Mon, 28 Oct 2019 11:58:34 -0700
From:   Kees Cook <keescook@chromium.org>
To:     Joe Perches <joe@perches.com>
Cc:     Julia Lawall <julia.lawall@lip6.fr>,
        Andrew Morton <akpm@linux-foundation.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Dan Carpenter <error27@gmail.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        zhanglin <zhang.lin16@zte.com.cn>
Subject: Re: [PATCH] kernel: sys.c: Avoid copying possible padding bytes in
 copy_to_user
Message-ID: <201910281153.7B6F79DBD@keescook>
References: <dfa331c00881d61c8ee51577a082d8bebd61805c.camel@perches.com>
 <alpine.DEB.2.21.1910270644590.3186@hadrien>
 <92212e57d45f4410be654183f5dcb1e98d636ef2.camel@perches.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <92212e57d45f4410be654183f5dcb1e98d636ef2.camel@perches.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, Oct 27, 2019 at 03:47:21PM -0700, Joe Perches wrote:
> I think yes as at least it makes it consistent.
> 
> From the link above, as I understand the __user
> gcc extension here
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c61f13eaa1ee17728c41370100d2d45c254ce76f
> 
> gcc does not clear padding from initialized structs
> marked with __user.

It seems to depend on how complete the initialization is and/or how
large the structure is. AFAICT, based on the tests I wrote[1], if
CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF (or ..._ALL) are used, even padding
will get initialized as long as things are in memory. (And the same is
true with Clang under CONFIG_INIT_STACK_ALL.)

> Though that doesn't force the compiler to not
> perform the possible register optimization shown
> in the first document above.

Right. This is the only case where things aren't clear. I haven't been
able to build a test where "store in registers" behavior is tripped.

-Kees

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/lib/test_stackinit.c

-- 
Kees Cook