Re: [RFC] pointer poisoning macro

From: Vasily Kulikov
Date: Sat Jun 06 2015 - 08:46:35 EST


Hi,

Any feedback?

On Thu, May 14, 2015 at 14:07 +0300, Vasily Kulikov wrote:
> This is a raw version of the patch inspired by the discussion:
> http://www.openwall.com/lists/oss-security/2015/05/02/6.
>
> The patch tries to achieve two goals:
> 1) Move "overflowed" poison pointers out of the mmap'able memory zone
> 2) Simplify addition of new poison pointers
>
> The current 0x00200200 poison pointer value of LIST_POISON2 might be
> too big for mmap_min_addr values equal or less than 2 MB (common case,
> e.g. Ubuntu uses only 0x10000). There is little point to use such a big
> value given the "poison pointer space" below 2 MB is not yet exhausted.
> Changing it to a smaller value solves the problem for small
> mmap_min_addr setups.
>
> In general, poisoned pointers should meet the following criteria:
>
> 1) poisoned pointer base must be non-mmap'able (already done via
> CONFIG_ILLEGAL_POINTER_VALUE)
> 2) all poisoned pointers (i.e. base+offset) must be non-mmap'able
> 3) a small offset relative to poisoned pointers must be non-mmap'able
> 4) poisoned pointers from different subsystems should be different
>
> (2) can be solved at the compile time by BUILD_BUG_ON().
> (3) and (4) should be solved by a creator of the new poisoned pointer.
>
> At least (2) can be done automatically. I propose a new macro for this
> purpose, POISON_POINTER(). It should check whether the poison pointer
> offset is not too big. E.g. in case of too big offset the compilation
> fails with the following message:
>
> mm/page_alloc.c: In function 'free_pages_prepare':
> mm/page_alloc.c:840:23: error: call to '__compiletime_assert_840' declared with attribute error: BUILD_BUG_ON failed: 0x0111400 >= POISON_AREA_SIZE
>
> There is still an unsolved issue with the macro related to
> static variables initialization. If you uncomment the line just after
> "FIXME" line, you'll see:
>
> kernel/irq/spurious.c:23:8: error: braced-group within expression allowed only inside a function
>
> I'll be happy with the comments to the idea and the implementation.
>
> diff --git a/include/linux/poison.h b/include/linux/poison.h
> index 7b2a7fc..47b43ed 100644
> --- a/include/linux/poison.h
> +++ b/include/linux/poison.h
> @@ -1,40 +1,54 @@
> #ifndef _LINUX_POISON_H
> #define _LINUX_POISON_H
> +#include <linux/bug.h>
>
> /********** include/linux/list.h **********/
>
> /*
> * Architectures might want to move the poison pointer offset
> * into some well-recognized area such as 0xdead000000000000,
> - * that is also not mappable by user-space exploits:
> + * that is also not mappable by user-space exploits,
> + * by adjusting CONFIG_ILLEGAL_POINTER_VALUE:
> */
> #ifdef CONFIG_ILLEGAL_POINTER_VALUE
> # define POISON_POINTER_DELTA _AC(CONFIG_ILLEGAL_POINTER_VALUE, UL)
> #else
> # define POISON_POINTER_DELTA 0
> #endif
> +/*
> + * Poisoned pointers of different subsystems should be different
> + * but must not move far away from POISON_POINTER_DELTA.
> + * Otherwise poisoned pointer might be mmap'able on some architectures.
> + */
> +#define POISON_AREA_SIZE 0x1000
> +#define POISON_POINTER(x) \
> + ({ \
> + BUILD_BUG_ON(x >= POISON_AREA_SIZE); \
> + ((void *)(x) + POISON_POINTER_DELTA);})
>
> /*
> * These are non-NULL pointers that will result in page faults
> * under normal circumstances, used to verify that nobody uses
> * non-initialized list entries.
> */
> -#define LIST_POISON1 ((void *) 0x00100100 + POISON_POINTER_DELTA)
> -#define LIST_POISON2 ((void *) 0x00200200 + POISON_POINTER_DELTA)
> +#define LIST_POISON1 POISON_POINTER(0x0100)
> +#define LIST_POISON2 POISON_POINTER(0x0200)
>
> /********** include/linux/timer.h **********/
> /*
> * Magic number "tsta" to indicate a static timer initializer
> * for the object debugging code.
> */
> -#define TIMER_ENTRY_STATIC ((void *) 0x74737461)
> +#define TIMER_ENTRY_STATIC ((void*)0x0300 + POISON_POINTER_DELTA)
> +// FIXME
> +//#define TIMER_ENTRY_STATIC POISON_POINTER(0x0300)
>
> /********** mm/debug-pagealloc.c **********/
> #define PAGE_POISON 0xaa
>
> /********** mm/page_alloc.c ************/
>
> -#define TAIL_MAPPING ((void *) 0x01014A11 + POISON_POINTER_DELTA)
> +#define TAIL_MAPPING POISON_POINTER(0x400)
>
> /********** mm/slab.c **********/
> /*
>
> --
> Vasily

--
Vasily Kulikov
http://www.openwall.com - bringing security into open computing environments
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/