From mboxrd@z Thu Jan 1 00:00:00 1970 Reply-To: kernel-hardening@lists.openwall.com References: <1450755641-7856-1-git-send-email-laura@labbott.name> <1450755641-7856-7-git-send-email-laura@labbott.name> <567964F3.2020402@intel.com> <567986E7.50107@intel.com> <56798851.60906@intel.com> <5679943C.1050604@intel.com> <5679A0CB.3060707@labbott.name> From: Dave Hansen Message-ID: <5679A568.9000604@intel.com> Date: Tue, 22 Dec 2015 11:32:56 -0800 MIME-Version: 1.0 In-Reply-To: <5679A0CB.3060707@labbott.name> Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Subject: Re: [kernel-hardening] [RFC][PATCH 6/7] mm: Add Kconfig option for slab sanitization To: Laura Abbott , Christoph Lameter Cc: kernel-hardening@lists.openwall.com, Pekka Enberg , David Rientjes , Joonsoo Kim , Andrew Morton , linux-mm@kvack.org, linux-kernel@vger.kernel.org, Kees Cook List-ID: On 12/22/2015 11:13 AM, Laura Abbott wrote: >> 3. Zero at free, *don't* Zero at alloc (when __GFP_ZERO) >> (what I'm suggesting, possibly less perf impact vs. #2) > > poisoning with non-zero memory makes it easier to determine that the error > came from accessing the sanitized memory vs. some other case. I don't think > the feature would be as strong if the memory was only zeroed vs. some other > data value. How does that scenario work? Your patch description says: > + Use-after-free bugs for structures containing > + pointers can also be detected as dereferencing the sanitized pointer > + will generate an access violation. In the case that we wrote all zeros, we'd be accessing userspace at a known place that we don't generally allow memory to be mapped anyway. Could you elaborate on a scenario where zeros are weaker than a random poison value? In any case (if a poison value is superior to 0's), it's a balance between performance vs. the likelihood of the poisoned value being tripped over. I think the performance impact of this feature is going to be *the* major thing that keeps folks from using it in practice. I'm trying to suggest a way that you _might_ preserve some performance, and get more folks to use it. 1. Keep information from leaking (doesn't matter which value we write) 2. Detect use-after-free bugs (0's are less likely to be detected???) 3. Preserve performance (0's are likely to preserve more performance)