Re: [PATCH net-next v2] net/ipv6: silence 'passing zero to ERR_PTR()' warning

[Jakub Kicinski]

On Fri, 14 Apr 2023 09:32:51 +0300 Dan Carpenter wrote:
> Also it can return NULL.
> 
> net/xfrm/xfrm_policy.c
>   3229                  dst = dst_orig;
>   3230          }
>   3231  ok:
>   3232          xfrm_pols_put(pols, drop_pols);
>   3233          if (dst && dst->xfrm &&
>                     ^^^
> "dst" is NULL.

Don't take my word for it, but AFAICT it's impossible to get there with
dst == NULL. I think we can remove this check instead if that's what
makes smatch infer that dst may be NULL.

>   3234              dst->xfrm->props.mode == XFRM_MODE_TUNNEL)
>   3235                  dst->flags |= DST_XFRM_TUNNEL;
>   3236          return dst;
>                 ^^^^^^^^^^^
>   3237  
> 
> So in the original code what happened here was:
> 
> net/ipv6/icmp.c
>    395          dst2 = xfrm_lookup(net, dst2, flowi6_to_flowi(&fl2), sk, XFRM_LOOKUP_ICMP);
>    396          if (!IS_ERR(dst2)) {
> 
> xfrm_lookup() returns NULL.  NULL is not an error pointer.
> 
>    397                  dst_release(dst);
>    398                  dst = dst2;
> 
> We set "dst" to NULL.
> 
>    399          } else {
>    400                  err = PTR_ERR(dst2);
>    401                  if (err == -EPERM) {
>    402                          dst_release(dst);
>    403                          return dst2;
>    404                  } else
>    405                          goto relookup_failed;
>    406          }
>    407  
>    408  relookup_failed:
>    409          if (dst)
>    410                  return dst;
> 
> dst is not NULL so we don't return it.
> 
>    411          return ERR_PTR(err);
> 
> However "err" is not set so we do return NULL and Smatch complains about
> that.
> 
> Returning ERR_PTR(0); is not necessarily a bug, however 80% of the time
> in newly introduced code it is a bug.  Here, returning NULL is correct.
> So this is a false positive, but the code is just wibbly winding and so
> difficult to read.
> 
>    412  }