Return-Path: Received: from mail-it0-f54.google.com ([209.85.214.54]:46931 "EHLO mail-it0-f54.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757063AbdKRFOn (ORCPT ); Sat, 18 Nov 2017 00:14:43 -0500 Received: by mail-it0-f54.google.com with SMTP id r127so6476884itb.5 for ; Fri, 17 Nov 2017 21:14:42 -0800 (PST) MIME-Version: 1.0 In-Reply-To: <242b16cd-056a-1499-2b65-69084dac1d8d@gentoo.org> References: <40ad7c6e-f0d7-959a-bf29-d3e3843f5d31@gentoo.org> <23f7da04-95f7-24e7-ee70-ce40c5b8fee3@gentoo.org> <67939ef3-29c6-762c-7afe-46cc69630d95@gentoo.org> <3d948180-6bd7-c4e9-5ac8-5baef9cc15a7@gentoo.org> <09f2480f-e8e8-645b-6d94-b6ae4ca47806@gentoo.org> <242b16cd-056a-1499-2b65-69084dac1d8d@gentoo.org> From: Kees Cook Date: Fri, 17 Nov 2017 21:14:41 -0800 Message-ID: Subject: Re: [nfsd4] potentially hardware breaking regression in 4.14-rc and 4.13.11 To: Patrick McLean Cc: Linus Torvalds , Emese Revfy , Al Viro , Bruce Fields , "Darrick J. Wong" , Linux Kernel Mailing List , Linux NFS Mailing List , stable , Thorsten Leemhuis , "kernel-hardening@lists.openwall.com" Content-Type: text/plain; charset="UTF-8" Sender: linux-nfs-owner@vger.kernel.org List-ID: On Fri, Nov 17, 2017 at 5:54 PM, Patrick McLean wrote: > On 2017-11-17 04:55 PM, Linus Torvalds wrote: >> On Fri, Nov 17, 2017 at 4:27 PM, Patrick McLean wrote: >>> >>> I am still getting the crash at d9e12200852d, I figured I would >>> double-check the "good" and "bad" kernels before starting a full bisect. >> >> .. but without GCC_PLUGIN_RANDSTRUCT it's solid? > > Yes, without GCC_PLUGIN_RANDSTRUCT it's solid. That's strange. With d9e12200852d the shuffle_seed variables won't ever actually get used. (i.e. I wouldn't expect the seed to change any behavior.) Can you confirm with something like this: diff --git a/scripts/gcc-plugins/randomize_layout_plugin.c b/scripts/gcc-plugins/randomize_layout_plugin.c index cdaac8c66734..aac570a57d7d 100644 --- a/scripts/gcc-plugins/randomize_layout_plugin.c +++ b/scripts/gcc-plugins/randomize_layout_plugin.c @@ -267,12 +267,10 @@ static void shuffle(const_tree type, tree *newtree, unsigned long length) structname = ORIG_TYPE_NAME(type); -#ifdef __DEBUG_PLUGIN fprintf(stderr, "Shuffling struct %s %p\n", (const char *)structname, type); #ifdef __DEBUG_VERBOSE debug_tree((tree)type); #endif -#endif for (i = 0; i < 4; i++) { seed[i] = shuffle_seed[i]; You should see no reports of "Shuffling struct ..." And if it reports nothing, and you're on d9e12200852d, can you confirm that switching to a "good" seed fixes it? (If it _does_, then I suspect a build artifact being left behind or something odd like that.) >> Kees removed even the baseline "randomize pure function pointer >> structures", so at that commit, nothing should be randomized. >> >> But maybe the plugin code itself ends up confusing gcc somehow? >> >> Even when it doesn't actually do that "relayout_struct()" on the >> structure, it always does those TYPE_ATTRIBUTES() games. FWIW, myself doing a build at d9e12200852d with and without GCC_PLUGIN_RANDSTRUCT _appears_ to produce identical objdump output where I did spot-checks. Also, do you have any other plugins enabled? (Can you send your .config?) -Kees -- Kees Cook Pixel Security