xfrm: Cache used outbound xfrm states at the policy.

Now that we can have percpu xfrm states, the number of active
states might increase. To get a better lookup performance,
we cache the used xfrm states at the policy for outbound
IPsec traffic.

Signed-off-by: Steffen Klassert <[e-mail]>
Tested-by: Antony Antony <[e-mail]>
Tested-by: Tobias Brunner <[e-mail]>

include/net/xfrm.h

···
 	};
 	struct hlist_node	byspi;
 	struct hlist_node	byseq;
+	struct hlist_node	state_cache;
 
 	refcount_t		refcnt;
 	spinlock_t		lock;
···
  *	@xp_net: network namespace the policy lives in
  *	@bydst: hlist node for SPD hash table or rbtree list
  *	@byidx: hlist node for index hash table
+ *	@state_cache_list: hlist head for policy cached xfrm states
  *	@lock: serialize changes to policy structure members
  *	@refcnt: reference count, freed once it reaches 0
  *	@pos: kernel internal tie-breaker to determine age of policy
···
 	possible_net_t		xp_net;
 	struct hlist_node	bydst;
 	struct hlist_node	byidx;
+
+	struct hlist_head	state_cache_list;
 
 	/* This lock only affects elements except for entry. */
 	rwlock_t		lock;

net/xfrm/xfrm_policy.c

···
 	if (policy) {
 		write_pnet(&policy->xp_net, net);
 		INIT_LIST_HEAD(&policy->walk.all);
+		INIT_HLIST_HEAD(&policy->state_cache_list);
 		INIT_HLIST_NODE(&policy->bydst);
 		INIT_HLIST_NODE(&policy->byidx);
 		rwlock_init(&policy->lock);
···
 
 static void xfrm_policy_kill(struct xfrm_policy *policy)
 {
+	struct net *net = xp_net(policy);
+	struct xfrm_state *x;
+
 	xfrm_dev_policy_delete(policy);
 
 	write_lock_bh(&policy->lock);
···
 
 	if (del_timer(&policy->timer))
 		xfrm_pol_put(policy);
+
+	/* XXX: Flush state cache */
+	spin_lock_bh(&net->xfrm.xfrm_state_lock);
+	hlist_for_each_entry_rcu(x, &policy->state_cache_list, state_cache) {
+		hlist_del_init_rcu(&x->state_cache);
+	}
+	spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 
 	xfrm_pol_put(policy);
 }
···
 		dst_release(dst);
 		dst = dst_orig;
 	}
+
 ok:
 	xfrm_pols_put(pols, drop_pols);
 	if (dst && dst->xfrm &&

net/xfrm/xfrm_state.c

···
 		refcount_set(&x->refcnt, 1);
 		atomic_set(&x->tunnel_users, 0);
 		INIT_LIST_HEAD(&x->km.all);
+		INIT_HLIST_NODE(&x->state_cache);
 		INIT_HLIST_NODE(&x->bydst);
 		INIT_HLIST_NODE(&x->bysrc);
 		INIT_HLIST_NODE(&x->byspi);
···
 
 	if (x->km.state != XFRM_STATE_DEAD) {
 		x->km.state = XFRM_STATE_DEAD;
+
 		spin_lock(&net->xfrm.xfrm_state_lock);
 		list_del(&x->km.all);
 		hlist_del_rcu(&x->bydst);
 		hlist_del_rcu(&x->bysrc);
 		if (x->km.seq)
 			hlist_del_rcu(&x->byseq);
+		if (!hlist_unhashed(&x->state_cache))
+			hlist_del_rcu(&x->state_cache);
 		if (x->id.spi)
 			hlist_del_rcu(&x->byspi);
 		net->xfrm.state_num--;
···
 	unsigned int sequence;
 	struct km_event c;
 	unsigned int pcpu_id;
+	bool cached = false;
 
 	/* We need the cpu id just as a lookup key,
 	 * we don't require it to be stable.
···
 	sequence = read_seqcount_begin(&net->xfrm.xfrm_state_hash_generation);
 
 	rcu_read_lock();
+	hlist_for_each_entry_rcu(x, &pol->state_cache_list, state_cache) {
+		if (x->props.family == encap_family &&
+		    x->props.reqid == tmpl->reqid &&
+		    (mark & x->mark.m) == x->mark.v &&
+		    x->if_id == if_id &&
+		    !(x->props.flags & XFRM_STATE_WILDRECV) &&
+		    xfrm_state_addr_check(x, daddr, saddr, encap_family) &&
+		    tmpl->mode == x->props.mode &&
+		    tmpl->id.proto == x->id.proto &&
+		    (tmpl->id.spi == x->id.spi || !tmpl->id.spi))
+			xfrm_state_look_at(pol, x, fl, encap_family,
+					   &best, &acquire_in_progress, &error);
+	}
+
+	if (best)
+		goto cached;
+
+	hlist_for_each_entry_rcu(x, &pol->state_cache_list, state_cache) {
+		if (x->props.family == encap_family &&
+		    x->props.reqid == tmpl->reqid &&
+		    (mark & x->mark.m) == x->mark.v &&
+		    x->if_id == if_id &&
+		    !(x->props.flags & XFRM_STATE_WILDRECV) &&
+		    xfrm_addr_equal(&x->id.daddr, daddr, encap_family) &&
+		    tmpl->mode == x->props.mode &&
+		    tmpl->id.proto == x->id.proto &&
+		    (tmpl->id.spi == x->id.spi || !tmpl->id.spi))
+			xfrm_state_look_at(pol, x, fl, family,
+					   &best, &acquire_in_progress, &error);
+	}
+
+cached:
+	cached = true;
+	if (best)
+		goto found;
+	else if (error)
+		best = NULL;
+	else if (acquire_in_progress) /* XXX: acquire_in_progress should not happen */
+		WARN_ON(1);
+
 	h = xfrm_dst_hash(net, daddr, saddr, tmpl->reqid, encap_family);
 	hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h, bydst) {
 #ifdef CONFIG_XFRM_OFFLOAD
···
 			XFRM_STATE_INSERT(bysrc, &x->bysrc,
 					  net->xfrm.state_bysrc + h,
 					  x->xso.type);
+			INIT_HLIST_NODE(&x->state_cache);
 			if (x->id.spi) {
 				h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto, encap_family);
 				XFRM_STATE_INSERT(byspi, &x->byspi,
···
 	} else {
 		*err = acquire_in_progress ? -EAGAIN : error;
 	}
+
+	if (x && x->km.state == XFRM_STATE_VALID && !cached &&
+	    (!(pol->flags & XFRM_POLICY_CPU_ACQUIRE) || x->pcpu_num == pcpu_id)) {
+		spin_lock_bh(&net->xfrm.xfrm_state_lock);
+		if (hlist_unhashed(&x->state_cache))
+			hlist_add_head_rcu(&x->state_cache, &pol->state_cache_list);
+		spin_unlock_bh(&net->xfrm.xfrm_state_lock);
+	}
+
 	rcu_read_unlock();
 	if (to_put)
 		xfrm_state_put(to_put);