netfilter: ipset: Prepare the ipset core to use RCU at set level

Replace rwlock_t with spinlock_t in "struct ip_set" and change the locking
accordingly. Convert the comment extension into an rcu-avare object. Also,
simplify the timeout routines.

Signed-off-by: Jozsef Kadlecsik <[e-mail]>

include/linux/netfilter/ipset/ip_set.h

···
 	atomic64_t packets;
 };
 
+struct ip_set_comment_rcu {
+	struct rcu_head rcu;
+	char str[0];
+};
+
 struct ip_set_comment {
-	char *str;
+	struct ip_set_comment_rcu __rcu *c;
 };
 
 struct ip_set_skbinfo {
···
 	/* The name of the set */
 	char name[IPSET_MAXNAMELEN];
 	/* Lock protecting the set data */
-	rwlock_t lock;
+	spinlock_t lock;
 	/* References to the set */
 	u32 ref;
 	/* The core set type */

include/linux/netfilter/ipset/ip_set_comment.h

···
 	return nla_data(tb);
 }
 
+/* Called from uadd only, protected by the set spinlock.
+ * The kadt functions don't use the comment extensions in any way.
+ */
 static inline void
 ip_set_init_comment(struct ip_set_comment *comment,
 		    const struct ip_set_ext *ext)
 {
+	struct ip_set_comment_rcu *c = rcu_dereference_protected(comment->c, 1);
 	size_t len = ext->comment ? strlen(ext->comment) : 0;
 
-	if (unlikely(comment->str)) {
-		kfree(comment->str);
-		comment->str = NULL;
+	if (unlikely(c)) {
+		kfree_rcu(c, rcu);
+		rcu_assign_pointer(comment->c, NULL);
 	}
 	if (!len)
 		return;
 	if (unlikely(len > IPSET_MAX_COMMENT_SIZE))
 		len = IPSET_MAX_COMMENT_SIZE;
-	comment->str = kzalloc(len + 1, GFP_ATOMIC);
-	if (unlikely(!comment->str))
+	c = kzalloc(sizeof(*c) + len + 1, GFP_ATOMIC);
+	if (unlikely(!c))
 		return;
-	strlcpy(comment->str, ext->comment, len + 1);
+	strlcpy(c->str, ext->comment, len + 1);
+	rcu_assign_pointer(comment->c, c);
 }
 
+/* Used only when dumping a set, protected by rcu_read_lock_bh() */
 static inline int
 ip_set_put_comment(struct sk_buff *skb, struct ip_set_comment *comment)
 {
-	if (!comment->str)
+	struct ip_set_comment_rcu *c = rcu_dereference_bh(comment->c);
+
+	if (!c)
 		return 0;
-	return nla_put_string(skb, IPSET_ATTR_COMMENT, comment->str);
+	return nla_put_string(skb, IPSET_ATTR_COMMENT, c->str);
 }
 
+/* Called from uadd/udel, flush or the garbage collectors protected
+ * by the set spinlock.
+ * Called when the set is destroyed and when there can't be any user
+ * of the set data anymore.
+ */
 static inline void
 ip_set_comment_free(struct ip_set_comment *comment)
 {
-	if (unlikely(!comment->str))
+	struct ip_set_comment_rcu *c;
+
+	c = rcu_dereference_protected(comment->c, 1);
+	if (unlikely(!c))
 		return;
-	kfree(comment->str);
-	comment->str = NULL;
+	kfree_rcu(c, rcu);
+	rcu_assign_pointer(comment->c, NULL);
 }
 
 #endif

include/linux/netfilter/ipset/ip_set_timeout.h

···
 }
 
 static inline bool
-ip_set_timeout_test(unsigned long timeout)
+ip_set_timeout_expired(unsigned long *t)
 {
-	return timeout == IPSET_ELEM_PERMANENT ||
-	       time_is_after_jiffies(timeout);
-}
-
-static inline bool
-ip_set_timeout_expired(unsigned long *timeout)
-{
-	return *timeout != IPSET_ELEM_PERMANENT &&
-	       time_is_before_jiffies(*timeout);
+	return *t != IPSET_ELEM_PERMANENT && time_is_before_jiffies(*t);
 }
 
 static inline void
-ip_set_timeout_set(unsigned long *timeout, u32 t)
+ip_set_timeout_set(unsigned long *timeout, u32 value)
 {
-	if (!t) {
+	unsigned long t;
+
+	if (!value) {
 		*timeout = IPSET_ELEM_PERMANENT;
 		return;
 	}
 
-	*timeout = msecs_to_jiffies(t * MSEC_PER_SEC) + jiffies;
-	if (*timeout == IPSET_ELEM_PERMANENT)
+	t = msecs_to_jiffies(value * MSEC_PER_SEC) + jiffies;
+	if (t == IPSET_ELEM_PERMANENT)
 		/* Bingo! :-) */
-		(*timeout)--;
+		t--;
+	*timeout = t;
 }
 
 static inline u32

net/netfilter/ipset/ip_set_core.c

···
 		pr_warn("ip_set type %s, family %s with revision min %u already registered!\n",
 			type->name, family_name(type->family),
 			type->revision_min);
-		ret = -EINVAL;
-		goto unlock;
+		ip_set_type_unlock();
+		return -EINVAL;
 	}
 	list_add_rcu(&type->list, &ip_set_type_list);
 	pr_debug("type %s, family %s, revision %u:%u registered.\n",
 		 type->name, family_name(type->family),
 		 type->revision_min, type->revision_max);
-unlock:
 	ip_set_type_unlock();
+
 	return ret;
 }
 EXPORT_SYMBOL_GPL(ip_set_type_register);
···
 		pr_warn("ip_set type %s, family %s with revision min %u not registered\n",
 			type->name, family_name(type->family),
 			type->revision_min);
-		goto unlock;
+		ip_set_type_unlock();
+		return;
 	}
 	list_del_rcu(&type->list);
 	pr_debug("type %s, family %s with revision min %u unregistered.\n",
 		 type->name, family_name(type->family), type->revision_min);
-unlock:
 	ip_set_type_unlock();
 
 	synchronize_rcu();
···
 	    !(opt->family == set->family || set->family == NFPROTO_UNSPEC))
 		return 0;
 
-	read_lock_bh(&set->lock);
+	rcu_read_lock_bh();
 	ret = set->variant->kadt(set, skb, par, IPSET_TEST, opt);
-	read_unlock_bh(&set->lock);
+	rcu_read_unlock_bh();
 
 	if (ret == -EAGAIN) {
 		/* Type requests element to be completed */
 		pr_debug("element must be completed, ADD is triggered\n");
-		write_lock_bh(&set->lock);
+		spin_lock_bh(&set->lock);
 		set->variant->kadt(set, skb, par, IPSET_ADD, opt);
-		write_unlock_bh(&set->lock);
+		spin_unlock_bh(&set->lock);
 		ret = 1;
 	} else {
 		/* --return-nomatch: invert matched element */
···
 	    !(opt->family == set->family || set->family == NFPROTO_UNSPEC))
 		return -IPSET_ERR_TYPE_MISMATCH;
 
-	write_lock_bh(&set->lock);
+	spin_lock_bh(&set->lock);
 	ret = set->variant->kadt(set, skb, par, IPSET_ADD, opt);
-	write_unlock_bh(&set->lock);
+	spin_unlock_bh(&set->lock);
 
 	return ret;
 }
···
 	    !(opt->family == set->family || set->family == NFPROTO_UNSPEC))
 		return -IPSET_ERR_TYPE_MISMATCH;
 
-	write_lock_bh(&set->lock);
+	spin_lock_bh(&set->lock);
 	ret = set->variant->kadt(set, skb, par, IPSET_DEL, opt);
-	write_unlock_bh(&set->lock);
+	spin_unlock_bh(&set->lock);
 
 	return ret;
 }
···
 	set = kzalloc(sizeof(struct ip_set), GFP_KERNEL);
 	if (!set)
 		return -ENOMEM;
-	rwlock_init(&set->lock);
+	spin_lock_init(&set->lock);
 	strlcpy(set->name, name, IPSET_MAXNAMELEN);
 	set->family = family;
 	set->revision = revision;
···
 {
 	pr_debug("set: %s\n",  set->name);
 
-	write_lock_bh(&set->lock);
+	spin_lock_bh(&set->lock);
 	set->variant->flush(set);
-	write_unlock_bh(&set->lock);
+	spin_unlock_bh(&set->lock);
 }
 
 static int
···
 				set->variant->uref(set, cb, true);
 			/* Fall through and add elements */
 		default:
-			read_lock_bh(&set->lock);
+			rcu_read_lock_bh();
 			ret = set->variant->list(set, skb, cb);
-			read_unlock_bh(&set->lock);
+			rcu_read_unlock_bh();
 			if (!cb->args[IPSET_CB_ARG0])
 				/* Set is done, proceed with next one */
 				goto next_set;
···
 	bool eexist = flags & IPSET_FLAG_EXIST, retried = false;
 
 	do {
-		write_lock_bh(&set->lock);
+		spin_lock_bh(&set->lock);
 		ret = set->variant->uadt(set, tb, adt, &lineno, flags, retried);
-		write_unlock_bh(&set->lock);
+		spin_unlock_bh(&set->lock);
 		retried = true;
 	} while (ret == -EAGAIN &&
 		 set->variant->resize &&
···
 			     set->type->adt_policy))
 		return -IPSET_ERR_PROTOCOL;
 
-	read_lock_bh(&set->lock);
+	rcu_read_lock_bh();
 	ret = set->variant->uadt(set, tb, IPSET_TEST, NULL, 0, 0);
-	read_unlock_bh(&set->lock);
+	rcu_read_unlock_bh();
 	/* Userspace can't trigger element to be re-added */
 	if (ret == -EAGAIN)
 		ret = 1;