Re: [PATCH 1/7] netfilter: fix IS_ERR_VALUE usage

[Al Viro]

On Mon, Feb 15, 2016 at 03:35:19PM +0100, Andrzej Hajda wrote:
> IS_ERR_VALUE should be used only with unsigned long type.
> Otherwise it can work incorrectly. To achieve this function
> xt_percpu_counter_alloc is modified to return unsigned long,
> and its result is assigned to temporary variable to perform
> error checking, before assigning to .pcnt field.

	Wrong fix, IMO.  Just have an anon union of u64 pcnt and
struct xt_counters __percpu *pcpu in there.  And make this

> +static inline unsigned long xt_percpu_counter_alloc(void)
>  {
>  	if (nr_cpu_ids > 1) {
>  		void __percpu *res = __alloc_percpu(sizeof(struct xt_counters),
>  						    sizeof(struct xt_counters));
>  
>  		if (res == NULL)
> -			return (u64) -ENOMEM;
> +			return -ENOMEM;
>  
> -		return (u64) (__force unsigned long) res;
> +		return (__force unsigned long) res;
>  	}
>  
>  	return 0;

take struct xt_counters * and return 0 or -ENOMEM.  Storing the result of
allocation in ->pcpu of passed structure.

I mean, look at the callers -

> -	e->counters.pcnt = xt_percpu_counter_alloc();
> -	if (IS_ERR_VALUE(e->counters.pcnt))
> +	pcnt = xt_percpu_counter_alloc();
> +	if (IS_ERR_VALUE(pcnt))
>  		return -ENOMEM;
> +	e->counters.pcnt = pcnt;

should be
	if (xt_percpu_counter_alloc(&e->counters) < 0)
		return -ENOMEM;

and similar for the rest of callers.  Moreover, if you look at the _users_
of that fields, you'll see that a bunch of those actually want to use
->pcpu instead of doing all those casts.

Really, that's the point - IS_ERR_VALUE is a big red flag saying "we need
to figure out what's going on in that place", which does include reading
through the code.  Mechanical "solutions" like that only hide the problem.