Well, at first I thought so..
On 08/23/2018 11:56 AM, Kyeongdon Kim wrote:
> diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
> index c3bd520..61ad7f1 100644
> --- a/mm/kasan/kasan.c
> +++ b/mm/kasan/kasan.c
> @@ -304,6 +304,29 @@ void *memcpy(void *dest, const void *src, size_t len)
>
> return __memcpy(dest, src, len);
> }
> +#ifdef CONFIG_ARM64
> +/*
> + * Arch arm64 use assembly variant for strcmp/strncmp,
> + * xtensa use inline asm operations and x86_64 use c one,
> + * so now this interceptors only for arm64 kasan.
> + */
> +#undef strcmp
> +int strcmp(const char *cs, const char *ct)
> +{
> + check_memory_region((unsigned long)cs, 1, false, _RET_IP_);
> + check_memory_region((unsigned long)ct, 1, false, _RET_IP_);
> +
Well this is definitely wrong. strcmp() often accesses far more than one byte.
> + return __strcmp(cs, ct);
> +}
> +#undef strncmp
> +int strncmp(const char *cs, const char *ct, size_t len)
> +{
> + check_memory_region((unsigned long)cs, len, false, _RET_IP_);
> + check_memory_region((unsigned long)ct, len, false, _RET_IP_);
This will cause false positives. Both 'cs', and 'ct' could be less than len bytes.
There is no need in these interceptors, just use the C implementations from lib/string.c
like you did in your first patch.
The only thing that was wrong in the first patch is that assembly implementations
were compiled out instead of being declared week.