[PATCH] kasan: infer the requested size by scanning shadow memory

[Andrey Konovalov]

On Tue, Jan 3, 2023 at 8:56 AM Kuan-Ying Lee <Kuan-Ying.Lee at mediatek.com> wrote:
>
> We scan the shadow memory to infer the requested size instead of
> printing cache->object_size directly.
>
> This patch will fix the confusing generic kasan report like below. [1]
> Report shows "cache kmalloc-192 of size 192", but user
> actually kmalloc(184).
>
> ==================================================================
> BUG: KASAN: slab-out-of-bounds in _find_next_bit+0x143/0x160 lib/find_bit.c:109
> Read of size 8 at addr ffff8880175766b8 by task kworker/1:1/26
> ...
> The buggy address belongs to the object at ffff888017576600
>  which belongs to the cache kmalloc-192 of size 192
> The buggy address is located 184 bytes inside of
>  192-byte region [ffff888017576600, ffff8880175766c0)
> ...
> Memory state around the buggy address:
>  ffff888017576580: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
>  ffff888017576600: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
> >ffff888017576680: 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc fc
>                                         ^
>  ffff888017576700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>  ffff888017576780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
> ==================================================================
>
> After this patch, report will show "cache kmalloc-192 of size 184".

I think this introduces more confusion. kmalloc-192 cache doesn't have
the size of 184.

Let's leave the first two lines as is, and instead change the second
two lines to:

The buggy address is located 0 bytes to the right of
 requested 184-byte region [ffff888017576600, ffff8880175766c0)

This specifically points out an out-of-bounds access.

Note the added "requested". Alternatively, we could say "allocated".

> --- a/mm/kasan/kasan.h
> +++ b/mm/kasan/kasan.h
> @@ -340,8 +340,13 @@ static inline void kasan_print_address_stack_frame(const void *addr) { }
>
>  #ifdef CONFIG_KASAN_GENERIC
>  void kasan_print_aux_stacks(struct kmem_cache *cache, const void *object);
> +int kasan_get_alloc_size(void *object_addr, struct kmem_cache *cache);
>  #else
>  static inline void kasan_print_aux_stacks(struct kmem_cache *cache, const void *object) { }
> +static inline int kasan_get_alloc_size(void *object_addr, struct kmem_cache *cache)
> +{
> +       return cache->object_size;

Please implement similar shadow/tag walking for the tag-based modes.
Even though we can only deduce the requested size with the granularity
of 16 bytes, it still makes sense.

It makes sense to also use the word "allocated" instead of "requested"
for these modes, as the size is not deduced precisely.

> --- a/mm/kasan/report.c
> +++ b/mm/kasan/report.c
> @@ -236,12 +236,13 @@ static void describe_object_addr(const void *addr, struct kmem_cache *cache,
>  {
>         unsigned long access_addr = (unsigned long)addr;
>         unsigned long object_addr = (unsigned long)object;
> +       int real_size = kasan_get_alloc_size((void *)object_addr, cache);

Please add another field to the mode-specific section of the
kasan_report_info structure, fill it in complete_report_info, and use
it here. See kasan_find_first_bad_addr as a reference.

Thanks for working on this!