[PATCH] arm64: kernel: Fix incorrect brk randomization

[Kees Cook]

On Wed, May 11, 2016 at 8:29 AM, Will Deacon <will.deacon at arm.com> wrote:
> On Wed, May 11, 2016 at 08:27:14AM -0700, Kees Cook wrote:
>> On Wed, May 11, 2016 at 2:44 AM, Jon Medhurst (Tixy) <tixy at linaro.org> wrote:
>> > On Tue, 2016-05-10 at 10:55 -0700, Kees Cook wrote:
>> >> This fixes two issues with the arm64 brk randomziation. First, the
>> >> STACK_RND_MASK was being used incorrectly. The original code was:
>> >>
>> >>       unsigned long range_end = base + (STACK_RND_MASK << PAGE_SHIFT) + 1;
>> >>
>> >> STACK_RND_MASK is 0x7ff (32-bit) or 0x3ffff (64-bit), with 4K pages where
>> >> PAGE_SHIFT is 12:
>> >>
>> >>       #define STACK_RND_MASK  (test_thread_flag(TIF_32BIT) ? \
>> >>                                               0x7ff >> (PAGE_SHIFT - 12) : \
>> >>                                               0x3ffff >> (PAGE_SHIFT - 12))
>> >>
>> >> This means the resulting offset from base would be 0x7ff0001 or 0x3ffff0001,
>> >> which is wrong since it creates an unaligned end address. It was likely
>> >> intended to be:
>> >>
>> >>       unsigned long range_end = base + ((STACK_RND_MASK + 1) << PAGE_SHIFT)
>> >>
>> >> Which would result in offsets of 0x800000 (32-bit) and 0x40000000 (64-bit).
>> >>
>> >> However, even this corrected 32-bit compat offset (0x00800000) is much
>> >> smaller than native ARM's brk randomization value (0x02000000):
>> >>
>> >>       unsigned long arch_randomize_brk(struct mm_struct *mm)
>> >>       {
>> >>               unsigned long range_end = mm->brk + 0x02000000;
>> >>               return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
>> >>       }
>> >>
>> >> So, instead of basing arm64's brk randomization on mistaken STACK_RND_MASK
>> >> calculations, just use specific corrected values for compat (0x2000000)
>> >> and native arm64 (0x40000000).
>> >>
>> >> Signed-off-by: Kees Cook <keescook at chromium.org>
>> >
>> > There seems to be a helper 'is_compat_task()' that does
>> > 'test_thread_flag(TIF_32BIT)' so could perhaps be used instead, but
>> > that's too nit-picky. This change makes things more consistent with
>>
>> Oh, good call. Yeah, none of the other .c code does direct tests for
>> the TIF_32BIT flag, so I'll use the helper and send a v2. Thanks!
>
> I already applied it with that change and Tixy's reviewed-by. Thanks!

Oh! Perfect, thanks! :)

-Kees

-- 
Kees Cook
Chrome OS & Brillo Security