Re: [PATCH v5 0/2] Remove false-positive VLAs when using max()

From: Kees Cook <keescook@chromium.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>,
	Florian Weimer <fweimer@redhat.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Josh Poimboeuf <jpoimboe@redhat.com>,
	Rasmus Villemoes <linux@rasmusvillemoes.dk>,
	Randy Dunlap <rdunlap@infradead.org>,
	Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>,
	Ingo Molnar <mingo@kernel.org>,
	David Laight <David.Laight@aculab.com>,
	Ian Abbott <abbotti@mev.co.uk>,
	linux-input <linux-input@vger.kernel.org>,
	linux-btrfs <linux-btrfs@vger.kernel.org>,
	Network Development <netdev@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Kernel Hardening <kernel-hardening@lists.openwall.com>
Subject: Re: [PATCH v5 0/2] Remove false-positive VLAs when using max()
Date: Thu, 22 Mar 2018 08:01:25 -0700	[thread overview]
Message-ID: <CAGXu5j+5i+56R0KDLMDA=+_DRW5w9aUGCEo0dq6PZvHPBWkM1g@mail.gmail.com> (raw)
In-Reply-To: <CA+55aFwxk=tUECYQkd4cog08qW4ZT=r2K7FQXzGnc-zuMc7JQA@mail.gmail.com>

On Tue, Mar 20, 2018 at 4:23 PM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
> On Sat, Mar 17, 2018 at 1:07 PM, Kees Cook <keescook@chromium.org> wrote:
>>
>> No luck! :( gcc 4.4 refuses to play along. And, hilariously, not only
>> does it not change the complaint about __builtin_choose_expr(), it
>> also thinks that's a VLA now.
>
> Hmm. So thanks to the diseased mind of Martin Uecker, there's a better
> test for "__is_constant()":
>
>   /* Glory to Martin Uecker <Martin.Uecker@med.uni-goettingen.de> */
>   #define __is_constant(a) \
>         (sizeof(int) == sizeof(*(1 ? ((void*)((a) * 0l)) : (int*)1)))
>
> that is actually *specified* by the C standard to work, and doesn't
> even depend on any gcc extensions.

I feel we risk awakening Cthulhu with this. :)

> The reason is some really subtle pointer conversion rules, where the
> type of the ternary operator will depend on whether one of the
> pointers is NULL or not.
>
> And the definition of NULL, in turn, very much depends on "integer
> constant expression that has the value 0".
>
> Are you willing to do one final try on a generic min/max? Same as my
> last patch, but using the above __is_constant() test instead of
> __builtin_constant_p?

So, this time it's not a catastrophic failure with gcc 4.4. Instead it
fails in 11 distinct places:

$ grep "first argument to ‘__builtin_choose_expr’ not a constant" log
| cut -d: -f1-2
crypto/ablkcipher.c:71
crypto/blkcipher.c:70
crypto/skcipher.c:95
mm/percpu.c:2453
net/ceph/osdmap.c:1545
net/ceph/osdmap.c:1756
net/ceph/osdmap.c:1763
mm/kmemleak.c:1371
mm/kmemleak.c:1403
drivers/infiniband/hw/hfi1/pio_copy.c:421
drivers/infiniband/hw/hfi1/pio_copy.c:547

Seems like it doesn't like void * arguments:

mm/percpu.c:
                void *ptr;
...
                base = min(ptr, base);

mm/kmemleak.c:
static void scan_large_block(void *start, void *end)
...
                next = min(start + MAX_SCAN_SIZE, end);

I'll poke a bit more...

-Kees

-- 
Kees Cook
Pixel Security