lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  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:   Thu, 02 Mar 2017 14:40:28 -0800
From:   Joe Perches <joe@...ches.com>
To:     Arnd Bergmann <arnd@...db.de>
Cc:     kasan-dev <kasan-dev@...glegroups.com>,
        Andrey Ryabinin <aryabinin@...tuozzo.com>,
        Alexander Potapenko <glider@...gle.com>,
        Dmitry Vyukov <dvyukov@...gle.com>,
        Networking <netdev@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-media@...r.kernel.org,
        linux-wireless <linux-wireless@...r.kernel.org>,
        kernel-build-reports@...ts.linaro.org,
        "David S . Miller" <davem@...emloft.net>
Subject: Re: [PATCH 24/26] ocfs2: reduce stack size with KASAN

On Thu, 2017-03-02 at 23:22 +0100, Arnd Bergmann wrote:
> On Thu, Mar 2, 2017 at 6:46 PM, Joe Perches <joe@...ches.com> wrote:
> > On Thu, 2017-03-02 at 17:38 +0100, Arnd Bergmann wrote:
> > > The internal logging infrastructure in ocfs2 causes special warning code to be
> > > used with KASAN, which produces rather large stack frames:
> > > fs/ocfs2/super.c: In function 'ocfs2_fill_super':
> > > fs/ocfs2/super.c:1219:1: error: the frame size of 3264 bytes is larger than 3072 bytes [-Werror=frame-larger-than=]
> > 
> > At least by default it doesn't seem to.
> > 
> > gcc 6.2 allyesconfig, CONFIG_KASAN=y
> > with either CONFIG_KASAN_INLINE or CONFIG_KASAN_OUTLINE
> > 
> > gcc doesn't emit a stack warning
> 
> The warning is disabled until patch 26/26. which picks the 3072 default.
> The 3264 number was with gcc-7, which is worse than gcc-6 since it enables
> an extra check.
> 
> > > By simply passing the mask by value instead of reference, we can avoid the
> > > problem completely.
> > 
> > Any idea why that's so?
> 
> With KASAN, every time we inline the function, the compiler has to allocate
> space for another copy of the variable plus a redzone to detect whether
> passing it by reference into another function causes an overflow at runtime.

These logging functions aren't inlined.
You're referring to the stack frame?

> > >  On 64-bit architectures, this is also more efficient,
> > 
> > Efficient true, but the same overall stack no?
> 
> Here is what I see with CONFIG_FRAME_WARN=300 and x86_64-linux-gcc-6.3.1:
> 
> before:
[]
> fs/ocfs2/super.c:1219:1: error: the frame size of 552 bytes is larger
> than 300 bytes [-Werror=frame-larger-than=]
> 
> after:
> fs/ocfs2/super.c: In function 'ocfs2_fill_super':
> fs/ocfs2/super.c:1219:1: error: the frame size of 472 bytes is larger
> than 300 bytes [-Werror=frame-larger-than=]
> 
> and with gcc-7.0.1 (including -fsanitize-address-use-after-scope), before:
[]
> fs/ocfs2/super.c:1219:1: error: the frame size of 3264 bytes is larger
> than 300 bytes [-Werror=frame-larger-than=]
> 
> after:
> fs/ocfs2/super.c: In function 'ocfs2_fill_super':
> fs/ocfs2/super.c:1219:1: error: the frame size of 704 bytes is larger
> than 300 bytes [-Werror=frame-larger-than=]

Still doesn't make sense to me.

None of the logging functions are inlined as they are all
EXPORT_SYMBOL.

This just changes a pointer to a u64, which is the same
size on x86-64 (and is of course larger on x86-32).

Perhaps KASAN has the odd behavior and working around
KASAN's behavior may not be the proper thing to do.

Maybe if CONFIG_KASAN is set, the minimum stack should
be increased via THREAD_SIZE_ORDER or some such.

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux - Powered by OpenVZ