Re: [PATCH 1/3] ubsan: Add trap instrumentation option

From: Kees Cook <keescook@chromium.org>
To: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Elena Petrova <lenaptr@google.com>,
	Alexander Potapenko <glider@google.com>,
	Dmitry Vyukov <dvyukov@google.com>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Dan Carpenter <dan.carpenter@oracle.com>,
	"Gustavo A. R. Silva" <gustavo@embeddedor.com>,
	Arnd Bergmann <arnd@arndb.de>,
	Ard Biesheuvel <ard.biesheuvel@linaro.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	kasan-dev@googlegroups.com, linux-kernel@vger.kernel.org,
	kernel-hardening@lists.openwall.com
Subject: Re: [PATCH 1/3] ubsan: Add trap instrumentation option
Date: Thu, 21 Nov 2019 09:57:34 -0800	[thread overview]
Message-ID: <201911210942.3C9F299@keescook> (raw)
In-Reply-To: <35fa415f-1dab-b93d-f565-f0754b886d1b@virtuozzo.com>

On Thu, Nov 21, 2019 at 03:52:52PM +0300, Andrey Ryabinin wrote:
> On 11/20/19 4:06 AM, Kees Cook wrote:
> > +config UBSAN_TRAP
> > +	bool "On Sanitizer warnings, stop the offending kernel thread"

BTW, is there a way (with either GCC or Clang implementations) to
override the trap handler? If I could get the instrumentation to call
an arbitrarily named function, we could build a better version of this
that actually continued without the large increase in image size.

For example, instead of __builtin_trap(), call __ubsan_warning(), which
could be defined as something like:

static __always_inline void __ubsan_warning(void)
{
	WARN_ON_ONCE(1);
}

That would make the warning survivable without the overhead of all the
debugging structures, etc.

-- 
Kees Cook