[ovs-dev] [threaded-learning v2 09/25] ofproto: Remove soon-to-be-invalid optimizations.

[YAMAMOTO Takashi]

> @@ -1788,26 +1815,21 @@ ofproto_delete_flow(struct ofproto *ofproto,
>      struct classifier *cls = &ofproto->tables[0].cls;
>      struct rule *rule;
>  
> +    /* First do a cheap check whether the rule we're looking for has already
> +     * been deleted.  If so, then we're done. */
>      ovs_rwlock_rdlock(&cls->rwlock);
>      rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
>                                                              priority));
>      ovs_rwlock_unlock(&cls->rwlock);
>      if (!rule) {
> -        /* No such rule -> success. */
> -        return true;
> -    } else if (rule->pending) {
> -        /* An operation on the rule is already pending -> failure.
> -         * Caller must retry later if it's important. */
> -        return false;
> -    } else {
> -        /* Initiate deletion -> success. */
> -        ovs_rwlock_wrlock(&cls->rwlock);
> -        ofproto_rule_delete(ofproto, cls, rule);
> -        ovs_rwlock_unlock(&cls->rwlock);
> -
>          return true;
>      }
>  
> +    /* Fall back to a executing a full flow mod.  We can't optimize this at all
> +     * because we didn't take enough locks above to ensure that the flow table
> +     * didn't already change beneath us.  */
> +    return simple_flow_mod(ofproto, target, priority, NULL, 0,
> +                           OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
>  }

this seems wrong to me.  ofproto_delete_flow is used to remove hidden rules.
the new code ends up to use collect_rules_strict which skips hidden rules.

YAMAMOTO Takashi