Re: [PATCH bpf v2] bpf: Add nelems overflow check in btf_find_field_one()

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Critical] The patch leaves a critical 32-bit integer overflow in `btf_repeat_fields()` unprotected, leading to a stack buffer overflow.
--

commit d1acecf46d6faa37ff66d19a08184489e682b397
Author: Himanshu Anand <anand.himanshu17@gmail.com>

bpf: Add nelems overflow check in btf_find_field_one()

This commit adds an overflow check when multiplying nested array element
counts in btf_find_field_one(). This aims to make the logic consistent with
__btf_resolve_size() and guards against future validation ordering changes.

> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index a62d785812076..b767a9fcf095d 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -3765,7 +3765,8 @@ static int btf_find_field_one(const struct btf *btf,
>  	 */
>  	for (i = 0; i < MAX_RESOLVE_DEPTH && btf_type_is_array(var_type); i++) {
>  		array = btf_array(var_type);
> -		nelems *= array->nelems;
> +		if (check_mul_overflow(nelems, array->nelems, &nelems))
> +			return -E2BIG;
>  		var_type = btf_type_by_id(btf, array->type);
>  	}
>  	if (i == MAX_RESOLVE_DEPTH)

[Severity: Critical]
This isn't a bug introduced by this patch, but does the bounds check in
btf_repeat_fields() further down in this function still have a 32-bit
integer overflow?

The commit message mentions that btf_array_resolve() prevents the total
array size from overflowing u32, meaning sz * nelems <= U32_MAX. However,
could a crafted BTF union with 11 struct members (field_cnt = 11) and an
element size of 8 bypass the validation if nelems is 390451573?

The array size check would pass since 390451573 * 8 = 3123612584 <= U32_MAX.

When btf_repeat_fields() is called at the end of this function:

    if (nelems > 1) {
        ret = btf_repeat_fields(info, info_cnt, 1, nelems - 1, sz);
        ...
    }

Is the multiplication field_cnt * (repeat_cnt + 1) inside
btf_repeat_fields() evaluated as a 32-bit operation? If so,
11 * 390451573 = 4294967303, which wraps around to 7.

Since 7 <= 11, wouldn't this bypass the bounds check and lead to an
out-of-bounds memcpy on the stack-allocated info_arr[BTF_FIELDS_MAX],
potentially overwriting kernel stack memory?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260523074736.222994-1-anand.himanshu17@gmail.com?part=1