[PATCH] lib: parse_area_spec: don't modify *start and *size values if parse failed

[Sascha Hauer]

Hi Peter,

On Tue, Jan 09, 2018 at 05:21:20PM +0300, Peter Mamonov wrote:
> Signed-off-by: Peter Mamonov <pmamonov at gmail.com>
> ---
>  lib/misc.c | 46 ++++++++++++++++++++++++++++++++--------------
>  1 file changed, 32 insertions(+), 14 deletions(-)
> 
> diff --git a/lib/misc.c b/lib/misc.c
> index 62ddd6677..c7d5a0ca5 100644
> --- a/lib/misc.c
> +++ b/lib/misc.c
> @@ -79,38 +79,56 @@ EXPORT_SYMBOL(strtoul_suffix);
>  int parse_area_spec(const char *str, loff_t *start, loff_t *size)
>  {
>  	char *endp;
> -	loff_t end;
> +	loff_t end, _start, _size;
> +	int ret = -1;
>  
>  	if (!isdigit(*str))
>  		return -1;
>  
> -	*start = strtoull_suffix(str, &endp, 0);
> +	_start = strtoull_suffix(str, &endp, 0);
>  
>  	str = endp;
>  
>  	if (!*str) {
>  		/* beginning given, but no size, assume maximum size */
> -		*size = ~0;
> -		return 0;
> +		_size = ~0;
> +		ret = 0;
>  	}
>  
> -	if (*str == '-') {
> +	if (ret && *str == '-') {
>  		/* beginning and end given */
> -		end = strtoull_suffix(str + 1, NULL, 0);
> -		if (end < *start) {
> +		if (!isdigit(*(str + 1)))
> +			return ret;
> +
> +		end = strtoull_suffix(str + 1, &endp, 0);
> +		str = endp;
> +		if (end < _start) {
>  			printf("end < start\n");
> -			return -1;
> +			return ret;
>  		}
> -		*size = end - *start + 1;
> -		return 0;
> +		_size = end - _start + 1;
> +		ret = 0;
>  	}
>  
> -	if (*str == '+') {
> +	if (ret && *str == '+') {
>  		/* beginning and size given */
> -		*size = strtoull_suffix(str + 1, NULL, 0);
> -		return 0;
> +		if (!isdigit(*(str + 1)))
> +			return ret;
> +
> +		_size = strtoull_suffix(str + 1, &endp, 0);
> +		str = endp;
> +		ret = 0;
> +	}
> +
> +	if (!ret && *str)
> +		/* trailing symbols indicate invalid area spec */
> +		ret = -1;

Is this correct? I would assume a whitespace should be fine. We only
do not get trailing whitespaces in here because current users pass in
argv[] elements which are split up at whitespaces. The check would
also deserve a separate patch.

> +
> +	if (!ret) {
> +		*start = _start;
> +		*size = _size;
>  	}
>  
> -	return -1;
> +	return ret;

I find this patch unnecessarily hard to review and also the end result
doesn't look optimal. Could you create a 'success:' label and jump to it
when everything is fine? That would make the additional if(ret) and
if(!ret) checks unnecessary.

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |