Re: [PATCH 02/82] overflow: Introduce add_would_overflow()

[Rasmus Villemoes]

On 23/01/2024 01.26, Kees Cook wrote:
> For instances where only the overflow needs to be checked (and the sum
> isn't used), provide the new helper add_would_overflow(), which is
> a wrapper for check_add_overflow().
> 
> Cc: "Gustavo A. R. Silva" <gustavoars@xxxxxxxxxx>
> Cc: linux-hardening@xxxxxxxxxxxxxxx
> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> ---
>  include/linux/overflow.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/include/linux/overflow.h b/include/linux/overflow.h
> index 099f2e559aa8..ac088f73e0fd 100644
> --- a/include/linux/overflow.h
> +++ b/include/linux/overflow.h
> @@ -108,6 +108,22 @@ static inline bool __must_check __must_check_overflow(bool overflow)
>  		__builtin_add_overflow(__filter_integral(a), b,		\
>  				       __filter_ptrint(d))))
>  
> +/**
> + * add_would_overflow() - Check if an addition would overflow
> + * @a: first addend
> + * @b: second addend
> + *
> + * Returns true if the sum would overflow.
> + *
> + * To keep a copy of the sum when the addition doesn't overflow, use
> + * check_add_overflow() instead.
> + */
> +#define add_would_overflow(a, b)			\
> +	__must_check_overflow(({			\
> +		size_t __result;			\
> +		check_add_overflow(a, b, &__result);\
> +	}))

Hm, I think this is a bit too ill-defined. Why is the target type
hard-coded as size_t? What if a and b are u64, and we're on a 32 bit
target? Then a+b might not overflow but this helper would claim it did.

But we also cannot just use typeof(a+b) instead of size_t, since that
breaks when a and b are narrower than int (adding two u16 never
overflows since they get promoted to int, but then if assigning the
result to a u16 one truncates...).

Perhaps the target type must be explicit? sum_fits_in_type(T, a, b) ?
IDK, I just don't think size_t is the right thing to use in something
that is otherwise supposed to be type-generic.

Rasmus