Re: [PATCH 2/2] mm/page_alloc.c: add config option to sanitize freed pages

[Anisse Astier <anisse@astier.eu>]

Hi Andi,

Thinks for taking the time to review this.

On Sun, Apr 26, 2015 at 10:12 PM, Andi Kleen <andi@firstfloor.org> wrote:
> Anisse Astier <anisse@astier.eu> writes:
>> +       If unsure, say N.
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 05fcec9..c71440a 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -803,6 +803,11 @@ static bool free_pages_prepare(struct page *page, unsigned int order)
>>               debug_check_no_obj_freed(page_address(page),
>>                                          PAGE_SIZE << order);
>>       }
>> +
>> +#ifdef CONFIG_SANITIZE_FREED_PAGES
>> +     zero_pages(page, order);
>> +#endif
>
> And not removing the clear on __GFP_ZERO by remembering that?
>
> That means all clears would be done twice.
>
> That patch is far too simple. Clearing is commonly the most
> expensive kernel operation.
>

I thought about this, but if you unconditionally remove the clear on
__GFP_ZERO, you wouldn't be guaranteed to have a zeroed page when
memory is first used (you would protect the kernel from its own info
leaks though); you'd need to clear memory on boot for example.

If you try to remember that a page it's cleared, it means using a page
flag, which is was previously deemed too precious for this kind of
operation.

Regarding the expensive operation, I don't think this is an option
you'd enable on your systems if you care about performance.

Regards,

Anisse

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>