Re: [PATCH v2] mm: Make ksize() a reporting-only function

[Andrey Konovalov]

On Fri, Nov 18, 2022 at 4:57 AM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>
> With all "silently resizing" callers of ksize() refactored, remove the
> logic in ksize() that would allow it to be used to effectively change
> the size of an allocation (bypassing __alloc_size hints, etc). Users
> wanting this feature need to either use kmalloc_size_roundup() before an
> allocation, or use krealloc() directly.
>
> For kfree_sensitive(), move the unpoisoning logic inline. Replace the
> some of the partially open-coded ksize() in __do_krealloc with ksize()
> now that it doesn't perform unpoisoning.
>
> Adjust the KUnit tests to match the new ksize() behavior.
>
> Cc: Andrey Konovalov <andreyknvl@xxxxxxxxx>
> Cc: Christoph Lameter <cl@xxxxxxxxx>
> Cc: Pekka Enberg <penberg@xxxxxxxxxx>
> Cc: David Rientjes <rientjes@xxxxxxxxxx>
> Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Roman Gushchin <roman.gushchin@xxxxxxxxx>
> Cc: Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx>
> Cc: Andrey Ryabinin <ryabinin.a.a@xxxxxxxxx>
> Cc: Alexander Potapenko <glider@xxxxxxxxxx>
> Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> Cc: Vincenzo Frascino <vincenzo.frascino@xxxxxxx>
> Cc: linux-mm@xxxxxxxxx
> Cc: kasan-dev@xxxxxxxxxxxxxxxx
> Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> ---
> v2:
> - improve kunit test precision (andreyknvl)
> - add Ack (vbabka)
> v1: https://lore.kernel.org/all/20221022180455.never.023-kees@xxxxxxxxxx
> ---
>  mm/kasan/kasan_test.c | 14 +++++++++-----
>  mm/slab_common.c      | 26 ++++++++++----------------
>  2 files changed, 19 insertions(+), 21 deletions(-)
>
> diff --git a/mm/kasan/kasan_test.c b/mm/kasan/kasan_test.c
> index 7502f03c807c..fc4b22916587 100644
> --- a/mm/kasan/kasan_test.c
> +++ b/mm/kasan/kasan_test.c
> @@ -821,7 +821,7 @@ static void kasan_global_oob_left(struct kunit *test)
>         KUNIT_EXPECT_KASAN_FAIL(test, *(volatile char *)p);
>  }
>
> -/* Check that ksize() makes the whole object accessible. */
> +/* Check that ksize() does NOT unpoison whole object. */
>  static void ksize_unpoisons_memory(struct kunit *test)
>  {
>         char *ptr;
> @@ -829,15 +829,19 @@ static void ksize_unpoisons_memory(struct kunit *test)
>
>         ptr = kmalloc(size, GFP_KERNEL);
>         KUNIT_ASSERT_NOT_ERR_OR_NULL(test, ptr);
> +
>         real_size = ksize(ptr);
> +       KUNIT_EXPECT_GT(test, real_size, size);
>
>         OPTIMIZER_HIDE_VAR(ptr);
>
> -       /* This access shouldn't trigger a KASAN report. */
> -       ptr[size] = 'x';
> +       /* These accesses shouldn't trigger a KASAN report. */
> +       ptr[0] = 'x';
> +       ptr[size - 1] = 'x';
>
> -       /* This one must. */
> -       KUNIT_EXPECT_KASAN_FAIL(test, ((volatile char *)ptr)[real_size]);
> +       /* These must trigger a KASAN report. */
> +       KUNIT_EXPECT_KASAN_FAIL(test, ((volatile char *)ptr)[size]);
> +       KUNIT_EXPECT_KASAN_FAIL(test, ((volatile char *)ptr)[real_size - 1]);

Hi Kees,

I just realized there's an issue here with the tag-based modes, as
they align the unpoisoned area to 16 bytes.

One solution would be to change the allocation size to 128 -
KASAN_GRANULE_SIZE - 5, the same way kmalloc_oob_right test does it,
so that the last 16-byte granule won't get unpoisoned for the
tag-based modes. And then check that the ptr[size] access fails only
for the Generic mode.

Thanks!

>
>         kfree(ptr);
>  }
> diff --git a/mm/slab_common.c b/mm/slab_common.c
> index 8276022f0da4..27caa57af070 100644
> --- a/mm/slab_common.c
> +++ b/mm/slab_common.c
> @@ -1335,11 +1335,11 @@ __do_krealloc(const void *p, size_t new_size, gfp_t flags)
>         void *ret;
>         size_t ks;
>
> -       /* Don't use instrumented ksize to allow precise KASAN poisoning. */
> +       /* Check for double-free before calling ksize. */
>         if (likely(!ZERO_OR_NULL_PTR(p))) {
>                 if (!kasan_check_byte(p))
>                         return NULL;
> -               ks = kfence_ksize(p) ?: __ksize(p);
> +               ks = ksize(p);
>         } else
>                 ks = 0;
>
> @@ -1407,21 +1407,21 @@ void kfree_sensitive(const void *p)
>         void *mem = (void *)p;
>
>         ks = ksize(mem);
> -       if (ks)
> +       if (ks) {
> +               kasan_unpoison_range(mem, ks);
>                 memzero_explicit(mem, ks);
> +       }
>         kfree(mem);
>  }
>  EXPORT_SYMBOL(kfree_sensitive);
>
>  size_t ksize(const void *objp)
>  {
> -       size_t size;
> -
>         /*
> -        * We need to first check that the pointer to the object is valid, and
> -        * only then unpoison the memory. The report printed from ksize() is
> -        * more useful, then when it's printed later when the behaviour could
> -        * be undefined due to a potential use-after-free or double-free.
> +        * We need to first check that the pointer to the object is valid.
> +        * The KASAN report printed from ksize() is more useful, then when
> +        * it's printed later when the behaviour could be undefined due to
> +        * a potential use-after-free or double-free.
>          *
>          * We use kasan_check_byte(), which is supported for the hardware
>          * tag-based KASAN mode, unlike kasan_check_read/write().
> @@ -1435,13 +1435,7 @@ size_t ksize(const void *objp)
>         if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
>                 return 0;
>
> -       size = kfence_ksize(objp) ?: __ksize(objp);
> -       /*
> -        * We assume that ksize callers could use whole allocated area,
> -        * so we need to unpoison this area.
> -        */
> -       kasan_unpoison_range(objp, size);
> -       return size;
> +       return kfence_ksize(objp) ?: __ksize(objp);
>  }
>  EXPORT_SYMBOL(ksize);
>
> --
> 2.34.1
>