Re: [PATCH 0/2] Support for set_memory_* outside of module space

[Kees Cook <keescook@chromium.org>]

On Wed, Nov 4, 2015 at 10:51 AM, Laura Abbott <labbott@redhat.com> wrote:
> On 11/03/2015 03:42 PM, Kees Cook wrote:
>>
>> On Tue, Nov 3, 2015 at 1:48 PM, Laura Abbott <labbott@fedoraproject.org>
>> wrote:
>>>
>>>
>>> Hi,
>>>
>>> Based on a recent discussion[1] there is interest in having set_memory_*
>>> work
>>> on kernel memory for security and other use cases. This patch adds the
>>> ability for that to happen behind a kernel option. If this is welcome
>>> enough,
>>> the Kconfig can be dropped. This has been briefly tested but not stress
>>> tested.
>>>
>>> Thanks,
>>> Laura
>>>
>>> [1]
>>> http://lists.infradead.org/pipermail/linux-arm-kernel/2015-October/382079.html
>>>
>>> Laura Abbott (2):
>>>    arm64: Get existing page protections in split_pmd
>>>    arm64: Allow changing of attributes outside of modules
>>>
>>>   arch/arm64/Kconfig.debug | 11 +++++++
>>>   arch/arm64/mm/mm.h       |  3 ++
>>>   arch/arm64/mm/mmu.c      | 11 ++++---
>>>   arch/arm64/mm/pageattr.c | 74
>>> ++++++++++++++++++++++++++++++++++++++++++++----
>>>   4 files changed, 88 insertions(+), 11 deletions(-)
>>
>>
>> This seems like the right thing to have. What's arm64 doing for the
>> equivalent of x86 and arm's set_kernel_text_* functions? x86 and arm
>> call their set_memory_* functions, for example. A quick examination
>> shows mm/mmu.c is just doing it "by hand" in fixup_executable and
>> mark_rodata_ro? Could those functions use the new set_memory_* ones?
>>
>
> It looks like mark_rodata_ro could probably use the set_memory_ro. I'll
> have to test it out. Longer term, the page table setup code should
> probably just be pulled out into a common file.
>
> Do you know the code path in ftrace which would trigger the
> set_kernel_text_*
> If not, I'll go  see if I can figure out if it's implemented yet on arm64.

Looking now, it seems that things like jump label use
aarch64_insn_patch_text, which ultimately call down to
__aarch64_insn_write(), which is using the fixmap via FIX_TEXT_POKE0.

-Kees

-- 
Kees Cook
Chrome OS Security