linux-kernel - Re: [PATCH] kasan: fix build for CONFIG_KASAN_HW

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 12 Jul 2021 09:37:17 +0100
From:   Mark Rutland <mark.rutland@....com>
To:     Andrey Konovalov <andreyknvl@...il.com>
Cc:     Sam Tebbs <sam.tebbs@....com>, Robin Murphy <robin.murphy@....com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Linux ARM <linux-arm-kernel@...ts.infradead.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Catalin Marinas <catalin.marinas@....com>,
        Dmitry Vyukov <dvyukov@...gle.com>,
        Alexander Potapenko <glider@...gle.com>,
        Andrey Ryabinin <ryabinin.a.a@...il.com>,
        Will Deacon <will@...nel.org>, Marco Elver <elver@...gle.com>
Subject: Re: [PATCH] kasan: fix build for CONFIG_KASAN_HW_TAGS

On Sat, Jul 10, 2021 at 09:16:14PM +0200, Andrey Konovalov wrote:
> On Thu, Jul 8, 2021 at 4:44 PM Mark Rutland <mark.rutland@....com> wrote:
> >
> > When CONFIG_KASAN_HW_TAGS is selected, <linux/kasan.h> uses _RET_IP_,
> > but doesn't explicitly include <linux/kernel.h> where this is defined.
> >
> > We used to get this via a transitive include, but since commit:
> >
> >   f39650de687e3576 ("kernel.h: split out panic and oops helpers")
> >
> > ... this is no longer the case, and so we get a build failure:
> >
> > |   CC      arch/arm64/mm/kasan_init.o
> > | In file included from arch/arm64/mm/kasan_init.c:10:
> > | ./include/linux/kasan.h: In function 'kasan_slab_free':
> > | ./include/linux/kasan.h:211:39: error: '_RET_IP_' undeclared (first use in this function)
> > |   211 |   return __kasan_slab_free(s, object, _RET_IP_, init);
> > |       |                                       ^~~~~~~~
> > | ./include/linux/kasan.h:211:39: note: each undeclared identifier is reported only once for each function it appears in
> > | ./include/linux/kasan.h: In function 'kasan_kfree_large':
> > | ./include/linux/kasan.h:219:28: error: '_RET_IP_' undeclared (first use in this function)
> > |   219 |   __kasan_kfree_large(ptr, _RET_IP_);
> > |       |                            ^~~~~~~~
> > | ./include/linux/kasan.h: In function 'kasan_slab_free_mempool':
> > | ./include/linux/kasan.h:226:34: error: '_RET_IP_' undeclared (first use in this function)
> > |   226 |   __kasan_slab_free_mempool(ptr, _RET_IP_);
> > |       |                                  ^~~~~~~~
> > | ./include/linux/kasan.h: In function 'kasan_check_byte':
> > | ./include/linux/kasan.h:277:35: error: '_RET_IP_' undeclared (first use in this function)
> > |   277 |   return __kasan_check_byte(addr, _RET_IP_);
> > |       |                                   ^~~~~~~~
> >
> > Fix this by including <linux/kernel.h> explicitly.
> 
> Hi Mark,
> 
> Marco already sent a fix for this. It should be in the mm tree.
> (Although the link to it in the Andrew's notification email doesn't
> work. But they rarely do :)
> 
> > As a heads-up, there are some unrelated runtime issues with
> > CONFIG_KASAN_HW_TAGS and the recent arm64 string routines rework, which
> > I'm looking into now. If you boot-test with this applied, you should
> > expect to see those.
> 
> +Sam, +Robin
> 
> Looks like the new strlen routine is making accesses past the allocated buffer.
> 
> The guilty commit is 325a1de81287 ("arm64: Import updated version of
> Cortex Strings' strlen").

FWIW, I already have a fix for this, I'm just cleaning it up and will
post shortly.

The issue is that the new strlen() will make unaligned 16-byte accesses
within a naturally-aligned 4096-byte window and over-read by up to 15
bytes; we can fiddle with its alignment fixup to always align to 16
bytes when MTE is in use so any over-read is within the same MTE granule
as the final byte of the string.

I've checked the other routines, and AFAICT they never make accesses
which staddle a 16-byte boundary.

Thanks,
Mark.