From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932121AbdESAo1 (ORCPT ); Thu, 18 May 2017 20:44:27 -0400 Received: from mx2.suse.de ([195.135.220.15]:55797 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1753652AbdESAoR (ORCPT ); Thu, 18 May 2017 20:44:17 -0400 Date: Fri, 19 May 2017 02:44:14 +0200 From: "Luis R. Rodriguez" To: Kees Cook Cc: "Luis R. Rodriguez" , Stephen Smalley , Ingo Molnar , Andy Lutomirski , Michal Hocko , Vlastimil Babka , Andrew Morton , "Eric W. Biederman" , Mateusz Guzik , LKML Subject: Re: next-20170515: WARNING: CPU: 0 PID: 1 at arch/x86/mm/dump_pagetables.c:236 note_page+0x630/0x7e0 Message-ID: <20170519004414.GD8951@wotan.suse.de> References: <20170515220650.GD17314@wotan.suse.de> <20170515221505.GE17314@wotan.suse.de> <20170517164017.GP17314@wotan.suse.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.6.0 (2016-04-01) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, May 17, 2017 at 10:53:06AM -0700, Kees Cook wrote: > On Wed, May 17, 2017 at 9:40 AM, Luis R. Rodriguez wrote: > > Yes, but I had killed that boot session again, so upon my next boot > > I had a different layout, the ASLR gap was much larger: > > > > ---[ Modules ]--- > > 0xffffffffc0000000-0xffffffffc01b0000 1728K pte > > 0xffffffffc01b0000-0xffffffffc01b1000 4K RW GLB x pte > > 0xffffffffc01b1000-0xffffffffc01b2000 4K pte > > 0xffffffffc01b2000-0xffffffffc01c6000 80K ro GLB x pte > > 0xffffffffc01c6000-0xffffffffc01cc000 24K ro GLB NX pte > > 0xffffffffc01cc000-0xffffffffc01d5000 36K RW GLB NX pte > > > > As you can guess if we follow similar pattern the RW hole is the one this boot > > warned about: > > > > [ 1.450483] x86/mm: Found insecure W+X mapping at address ffffffffc01b0000/0xffffffffc01b0000 > > [ 1.451280] ------------[ cut here ]------------ > > [ 1.451721] WARNING: CPU: 1 PID: 1 at arch/x86/mm/dump_pagetables.c:236 note_page+0x630/0x7e0 > > [ 1.452499] Modules linked in: > > [ 1.452791] CPU: 1 PID: 1 Comm: swapper/0 Not tainted 4.12.0-rc1-next-20170515+ #145 > > > > I checked and indeed 0xffffffffc01b2000 is part of a module, it was not the first one > > on the /proc/modules list but then again /proc/modules does not seem to have a specific > > order other than perhaps being pegged into a linked list of modules once they go live, > > and it seems its typically output backwards from when that happened, sorting that > > by address we get: > > Right, sorry, I'd expect it at the bottom of the list in > /proc/modules, but that's fine, it's there. > > > > > root@piggy:~# cat /proc/modules | sort -k 6 | head -3 > > e1000 143360 0 - Live 0xffffffffc01b2000 (E) > > mbcache 16384 1 ext4, Live 0xffffffffc01d6000 (E) > > scsi_mod 217088 4 sg,sr_mod,sd_mod,libata, Live 0xffffffffc01df000 (E) > > > > And this then seems to be the first module loaded: > > > > e1000 143360 0 - Live 0xffffffffc01b2000 (E) > > > > The output of dmesg seems to confirm this as per the list of modules sorted > > as per above. > > > >> Something touched the module gap and left is RW+x... > > > > Lemme try booting with e1000 renamed to e1000.ko.ignore and see how that goes. > > Is it possible a module got loaded before e1000 and then unloaded? > That seems odd, but maybe unload isn't cleaning up? > > >> Are you able to bisect this? > > > > This issue has been present for a while so since I recall this I might be > > able to reduce the number of needed target kernels to bisect. Lemme tinker > > a bit and if no clear culprit comes up then will try bisect. > > Okay, thanks! Sorry to report that this issue is present since the feature's addition. So the issue is there since its addition and is still present today. *But* it may also be a configuration issue, given I have booted this guest *without* this issue ... So: git checkout -b WX e1a58320a38dfa72be48a0f1a3a92273663ba6db That boots with the warning. To help debug further I've minimized my modules to only a few: scsi_mod, e1000, libata. I suspect at this point this is not the fault of a particular module but instead just an accounting semantic (>= or <= on an edge) but let's see. I now boot on 4.3.0-rc3 on commit (e1a58320a38df ("x86/mm: Warn on W^X mappings") and I with: [ 0.949435] ------------[ cut here ]------------ [ 0.949992] WARNING: CPU: 2 PID: 1 at arch/x86/mm/dump_pagetables.c:225 note_page+0x635/0x7e0() [ 0.950996] x86/mm: Found insecure W+X mapping at address ffffffffc0000000/0xffffffffc0000000 [ 0.951814] Modules linked in: [ 0.952123] CPU: 2 PID: 1 Comm: swapper/0 Not tainted 4.3.0-rc3-FINAL-TEST-WITH-WX-NOFLOPPY+ #365 [ 0.952929] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.10.2-0-g5f4c7b1-prebuilt.qemu-project.org 04/01/2014 [ 0.954033] 0000000000000000 000000001f722925 ffff88013a5d7d40 ffffffff812ff335 [ 0.954742] ffff88013a5d7d88 ffff88013a5d7d78 ffffffff81079be2 ffff88013a5d7e90 [ 0.955522] 0000000000000000 0000000000000004 0000000000000000 0000000000000000 [ 0.956256] Call Trace: [ 0.956496] [] dump_stack+0x44/0x5f [ 0.956953] [] warn_slowpath_common+0x82/0xc0 [ 0.957519] [] warn_slowpath_fmt+0x5c/0x80 [ 0.958066] [] note_page+0x635/0x7e0 [ 0.958595] [] ptdump_walk_pgd_level_core+0x2eb/0x410 [ 0.959219] [] ptdump_walk_pgd_level_checkwx+0x17/0x20 [ 0.959856] [] mark_rodata_ro+0xed/0x100 [ 0.960372] [] ? rest_init+0x80/0x80 [ 0.960869] [] kernel_init+0x1d/0xe0 [ 0.961358] [] ret_from_fork+0x3f/0x70 [ 0.961900] [] ? rest_init+0x80/0x80 [ 0.962389] ---[ end trace 6125ebcb24c9e3d0 ]--- [ 0.962822] x86/mm: Checked W+X mappings: FAILED, 1 W+X pages found. ---[ High Kernel Mapping ]--- 0xffffffff80000000-0xffffffff81000000 16M pmd 0xffffffff81000000-0xffffffff81600000 6M ro PSE GLB x pmd 0xffffffff81600000-0xffffffff81a00000 4M ro PSE GLB NX pmd 0xffffffff81a00000-0xffffffff81c00000 2M RW GLB NX pte 0xffffffff81c00000-0xffffffff82200000 6M RW PSE GLB NX pmd 0xffffffff82200000-0xffffffff82400000 2M RW GLB NX pte 0xffffffff82400000-0xffffffffc0000000 988M pmd ---[ Modules ]--- 0xffffffffc0000000-0xffffffffc0001000 4K RW GLB x pte 0xffffffffc0001000-0xffffffffc0002000 4K pte 0xffffffffc0002000-0xffffffffc0039000 220K RW GLB x pte root@piggy:~# cat /proc/modules | sort -k 6 | head -3 scsi_mod 221979 4 sg,sd_mod,sr_mod,libata, Live 0xffffffffc0002000 (E) e1000 127757 0 - Live 0xffffffffc004d000 (E) libata 229931 2 ata_generic,ata_piix, Live 0xffffffffc0076000 (E) So that 4K RW seems suspect of getting used for allocation purpose on edge for a particular reason and it also happens to be on the edge of the high kernel mapping. Could it be the boundary semantic issue ? For instance can it be that since 0xffffffffc0002000 is given to the first module by the allocator, scsi_mod, and since that address is *technically* part of two boundaries we get a splat ? 0xffffffffc0001000-0xffffffffc0002000 4K pte 0xffffffffc0002000-0xffffffffc0039000 220K RW GLB x pte Luis