Re: [PATCH] kmsan: test: check vmalloc return value before use

[Alexander Potapenko]

On Wed, Sep 2, 2026 at 10:56 AM Liu Jing <liujing@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> In test_init_vmalloc(), the return value of vmalloc() is used directly
> without a NULL check. If vmalloc() fails under memory pressure, the
> subsequent buf[0] = 1 and memset() calls will dereference a NULL
> pointer and cause a kernel panic.

Out of curiosity, have you actually observed vmalloc() failing under
memory pressure during this test?

> Add a NULL check immediately after vmalloc() and skip the test if
> allocation fails.
>
> Signed-off-by: Liu Jing <liujing@xxxxxxxxxxxxxxxxxxxx>
> ---
>  mm/kmsan/kmsan_test.c | 6 ++++++
>  1 file changed, 6 insertion(+), 0 deletion(-)
>
> --- a/mm/kmsan/kmsan_test.c
> +++ b/mm/kmsan/kmsan_test.c
> @@ -352,6 +352,12 @@
>
>         kunit_info(test, "vmalloc buffer can be initialized (no reports)\n");
>         buf = vmalloc(PAGE_SIZE * npages);
> +
> +       if (!buf) {
> +               kunit_skip(test, "vmalloc failed, skipping test\n");
> +               return;
> +       }

While at it, can you add similar checks to other allocation sites in this file?
The problem is not specific to vmalloc()

>         buf[0] = 1;
>         memset(buf, 0xfe, PAGE_SIZE * npages);
>         USE(buf[0]);
>
> --
> 2.43.0
>
>


-- 
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Liana Sebastian
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg