From mboxrd@z Thu Jan 1 00:00:00 1970 From: Vasiliy Kulikov Subject: Re: [kernel-hardening] Re: [RFC v1] implement SL*B and stack usercopy runtime checks Date: Sun, 3 Jul 2011 23:24:42 +0400 Message-ID: <20110703192442.GA9504@albatros> References: <20110703111028.GA2862@albatros> <20110703185709.GA7414@albatros> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from mail-bw0-f46.google.com ([209.85.214.46]:46561 "EHLO mail-bw0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755964Ab1GCTYr (ORCPT ); Sun, 3 Jul 2011 15:24:47 -0400 Content-Disposition: inline In-Reply-To: Sender: linux-arch-owner@vger.kernel.org List-ID: To: kernel-hardening@lists.openwall.com Cc: Thomas Gleixner , Ingo Molnar , "H. Peter Anvin" , x86@kernel.org, Arnd Bergmann , Christoph Lameter , Pekka Enberg , Matt Mackall , Andrew Morton , linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org, linux-mm@kvack.org On Sun, Jul 03, 2011 at 12:10 -0700, Linus Torvalds wrote: > On Sun, Jul 3, 2011 at 11:57 AM, Vasiliy Kulikov wrote: > >> If you seriously clean it up (that at a minimum includes things like > >> making it configurable using some pretty helper function that just > >> compiles away for all the normal cases, > > > > Hm, it is not as simple as it looks at the first glance - even if the > > object size is known at the compile time (__compiletime_object_size), it > > might be a field of a structure, which crosses the slab object > > boundaries because of an overflow. > > No, I was more talking about having something like > > #ifdef CONFIG_EXPENSIVE_CHECK_USERCOPY > extern int check_user_copy(const void *kptr, unsigned long size); > #else > static inline int check_user_copy(const void *kptr, unsigned long size) > { return 0; } > #endif Sure, will do. This is what I mean by kernel_access_ok() as it is a weak equivalent of access_ok(), check_user_copy() is a bit confusing name IMO. > so that the actual user-copy routines end up being clean and not have > #ifdefs in them or any implementation details like what you check > (stack, slab, page cache - whatever) > > If you can also make it automatically not generate any code for cases > that are somehow obviously safe, then that's an added bonus. OK, then let's stop on "checks for overflows" and remove the check if __compiletime_object_size() says something or length is constant. It should remove most of the checks in fast pathes. > But my concern is that performance is a real issue, and the strict > user-copy checking sounds like mostly a "let's enable this for testing > kernels when chasing some particular issue" feature, the way > DEBUG_PAGEALLOC is. I will measure the perfomance penalty tomorrow. Btw, if the perfomance will be acceptable, what do you think about logging/reacting on the spotted overflows? Thanks, -- Vasiliy Kulikov http://www.openwall.com - bringing security into open computing environments