datapath: Zero out key when looking up null ports.
[sliver-openvswitch.git] / datapath / tunnel.c
index 020d9d4..26d9014 100644 (file)
 #include "vport-generic.h"
 #include "vport-internal_dev.h"
 
-#ifdef NEED_CACHE_TIMEOUT
-/*
- * On kernels where we can't quickly detect changes in the rest of the system
- * we use an expiration time to invalidate the cache.  A shorter expiration
- * reduces the length of time that we may potentially blackhole packets while
- * a longer time increases performance by reducing the frequency that the
- * cache needs to be rebuilt.  A variety of factors may cause the cache to be
- * invalidated before the expiration time but this is the maximum.  The time
- * is expressed in jiffies.
- */
-#define MAX_CACHE_EXP HZ
-#endif
-
-/*
- * Interval to check for and remove caches that are no longer valid.  Caches
- * are checked for validity before they are used for packet encapsulation and
- * old caches are removed at that time.  However, if no packets are sent through
- * the tunnel then the cache will never be destroyed.  Since it holds
- * references to a number of system objects, the cache will continue to use
- * system resources by not allowing those objects to be destroyed.  The cache
- * cleaner is periodically run to free invalid caches.  It does not
- * significantly affect system performance.  A lower interval will release
- * resources faster but will itself consume resources by requiring more frequent
- * checks.  A longer interval may result in messages being printed to the kernel
- * message buffer about unreleased resources.  The interval is expressed in
- * jiffies.
- */
-#define CACHE_CLEANER_INTERVAL (5 * HZ)
-
-#define CACHE_DATA_ALIGN 16
 #define PORT_TABLE_SIZE  1024
 
 static struct hlist_head *port_table __read_mostly;
-static int port_table_count;
-
-static void cache_cleaner(struct work_struct *work);
-static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
 
 /*
  * These are just used as an optimization: they don't require any kind of
@@ -100,6 +66,7 @@ static unsigned int key_remote_ports __read_mostly;
 static unsigned int key_multicast_ports __read_mostly;
 static unsigned int local_remote_ports __read_mostly;
 static unsigned int remote_ports __read_mostly;
+static unsigned int null_ports __read_mostly;
 static unsigned int multicast_ports __read_mostly;
 
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
@@ -108,60 +75,17 @@ static unsigned int multicast_ports __read_mostly;
 #define rt_dst(rt) (rt->u.dst)
 #endif
 
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
-static struct hh_cache *rt_hh(struct rtable *rt)
-{
-       struct neighbour *neigh = dst_get_neighbour_noref(&rt->dst);
-       if (!neigh || !(neigh->nud_state & NUD_CONNECTED) ||
-                       !neigh->hh.hh_len)
-               return NULL;
-       return &neigh->hh;
-}
-#else
-#define rt_hh(rt) (rt_dst(rt).hh)
-#endif
-
 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
 {
        return vport_from_priv(tnl_vport);
 }
 
-/* This is analogous to rtnl_dereference for the tunnel cache.  It checks that
- * cache_lock is held, so it is only for update side code.
- */
-static struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
-{
-       return rcu_dereference_protected(tnl_vport->cache,
-                                lockdep_is_held(&tnl_vport->cache_lock));
-}
-
-static void schedule_cache_cleaner(void)
-{
-       schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
-}
-
-static void free_cache(struct tnl_cache *cache)
-{
-       if (!cache)
-               return;
-
-       ovs_flow_put(cache->flow);
-       ip_rt_put(cache->rt);
-       kfree(cache);
-}
-
 static void free_config_rcu(struct rcu_head *rcu)
 {
        struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
        kfree(c);
 }
 
-static void free_cache_rcu(struct rcu_head *rcu)
-{
-       struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
-       free_cache(c);
-}
-
 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
  * within an RCU callback.  Fortunately this part doesn't require waiting for
  * an RCU grace period.
@@ -190,18 +114,6 @@ static void assign_config_rcu(struct vport *vport,
        call_rcu(&old_config->rcu, free_config_rcu);
 }
 
-static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
-{
-       struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
-       struct tnl_cache *old_cache;
-
-       old_cache = cache_dereference(tnl_vport);
-       rcu_assign_pointer(tnl_vport->cache, new_cache);
-
-       if (old_cache)
-               call_rcu(&old_cache->rcu, free_cache_rcu);
-}
-
 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 {
        bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
@@ -218,8 +130,10 @@ static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
                        return &key_local_remote_ports;
                else if (is_multicast)
                        return &key_multicast_ports;
-               else
+               else if (mutable->key.daddr)
                        return &key_remote_ports;
+               else
+                       return &null_ports;
        }
 }
 
@@ -239,13 +153,9 @@ static void port_table_add_port(struct vport *vport)
        const struct tnl_mutable_config *mutable;
        u32 hash;
 
-       if (port_table_count == 0)
-               schedule_cache_cleaner();
-
        mutable = rtnl_dereference(tnl_vport->mutable);
        hash = port_hash(&mutable->key);
        hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
-       port_table_count++;
 
        (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 }
@@ -271,10 +181,6 @@ static void port_table_remove_port(struct vport *vport)
 
        hlist_del_init_rcu(&tnl_vport->hash_node);
 
-       port_table_count--;
-       if (port_table_count == 0)
-               cancel_delayed_work_sync(&cache_cleaner_wq);
-
        (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 }
 
@@ -364,6 +270,15 @@ struct vport *ovs_tnl_find_port(struct net *net, __be32 saddr, __be32 daddr,
                }
        }
 
+       if (null_ports) {
+               lookup.daddr = 0;
+               lookup.saddr = 0;
+               lookup.in_key = 0;
+               lookup.tunnel_type = tunnel_type;
+               vport = port_table_lookup(&lookup, mutable);
+               if (vport)
+                       return vport;
+       }
        return NULL;
 }
 
@@ -722,7 +637,7 @@ static bool check_mtu(struct sk_buff *skb,
        if (OVS_CB(skb)->tun_key->ipv4_dst) {
                df_inherit = false;
                pmtud = false;
-               frag_off = OVS_CB(skb)->tun_key->tun_flags & OVS_FLOW_TNL_F_DONT_FRAGMENT ?
+               frag_off = OVS_CB(skb)->tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT ?
                                  htons(IP_DF) : 0;
        } else {
                df_inherit = mutable->flags & TNL_F_DF_INHERIT;
@@ -787,284 +702,37 @@ static bool check_mtu(struct sk_buff *skb,
        return true;
 }
 
-static void create_tunnel_header(const struct vport *vport,
-                                const struct tnl_mutable_config *mutable,
-                                const struct ovs_key_ipv4_tunnel *tun_key,
-                                const struct rtable *rt, void *header)
-{
-       struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
-       struct iphdr *iph = header;
-
-       iph->version    = 4;
-       iph->ihl        = sizeof(struct iphdr) >> 2;
-       iph->frag_off   = htons(IP_DF);
-       iph->protocol   = tnl_vport->tnl_ops->ipproto;
-       iph->tos        = mutable->tos;
-       iph->daddr      = rt->rt_dst;
-       iph->saddr      = rt->rt_src;
-       iph->ttl        = mutable->ttl;
-       if (!iph->ttl)
-               iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
-
-       tnl_vport->tnl_ops->build_header(vport, mutable, tun_key, iph + 1);
-}
-
-static void *get_cached_header(const struct tnl_cache *cache)
-{
-       return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
-}
-
-#ifdef HAVE_RT_GENID
-static inline int rt_genid(struct net *net)
-{
-       return atomic_read(&net->ipv4.rt_genid);
-}
-#endif
-
-static bool check_cache_valid(const struct tnl_cache *cache,
-                             const struct tnl_mutable_config *mutable)
-{
-       struct hh_cache *hh;
-
-       if (!cache)
-               return false;
-
-       hh = rt_hh(cache->rt);
-       return hh &&
-#ifdef NEED_CACHE_TIMEOUT
-               time_before(jiffies, cache->expiration) &&
-#endif
-#ifdef HAVE_RT_GENID
-               rt_genid(dev_net(rt_dst(cache->rt).dev)) == cache->rt->rt_genid &&
-#endif
-#ifdef HAVE_HH_SEQ
-               hh->hh_lock.sequence == cache->hh_seq &&
-#endif
-               mutable->seq == cache->mutable_seq &&
-               (!ovs_is_internal_dev(rt_dst(cache->rt).dev) ||
-               (cache->flow && !cache->flow->dead));
-}
-
-static void __cache_cleaner(struct tnl_vport *tnl_vport)
-{
-       const struct tnl_mutable_config *mutable =
-                       rcu_dereference(tnl_vport->mutable);
-       const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
-
-       if (cache && !check_cache_valid(cache, mutable) &&
-           spin_trylock_bh(&tnl_vport->cache_lock)) {
-               assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
-               spin_unlock_bh(&tnl_vport->cache_lock);
-       }
-}
-
-static void cache_cleaner(struct work_struct *work)
-{
-       int i;
-
-       schedule_cache_cleaner();
-
-       rcu_read_lock();
-       for (i = 0; i < PORT_TABLE_SIZE; i++) {
-               struct hlist_node *n;
-               struct hlist_head *bucket;
-               struct tnl_vport *tnl_vport;
-
-               bucket = &port_table[i];
-               hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
-                       __cache_cleaner(tnl_vport);
-       }
-       rcu_read_unlock();
-}
-
-static void create_eth_hdr(struct tnl_cache *cache, struct hh_cache *hh)
-{
-       void *cache_data = get_cached_header(cache);
-       int hh_off;
-
-#ifdef HAVE_HH_SEQ
-       unsigned hh_seq;
-
-       do {
-               hh_seq = read_seqbegin(&hh->hh_lock);
-               hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
-               memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
-               cache->hh_len = hh->hh_len;
-       } while (read_seqretry(&hh->hh_lock, hh_seq));
-
-       cache->hh_seq = hh_seq;
-#else
-       read_lock(&hh->hh_lock);
-       hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
-       memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
-       cache->hh_len = hh->hh_len;
-       read_unlock(&hh->hh_lock);
-#endif
-}
-
-static struct tnl_cache *build_cache(struct vport *vport,
-                                    const struct tnl_mutable_config *mutable,
-                                    struct rtable *rt)
-{
-       struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
-       static const struct ovs_key_ipv4_tunnel tun_key;
-       struct tnl_cache *cache;
-       void *cache_data;
-       int cache_len;
-       struct hh_cache *hh;
-       int tunnel_hlen;
-
-       if (!(mutable->flags & TNL_F_HDR_CACHE))
-               return NULL;
-
-       tunnel_hlen = tnl_vport->tnl_ops->hdr_len(mutable, &tun_key);
-       if (tunnel_hlen < 0)
-               return NULL;
-
-       tunnel_hlen += sizeof(struct iphdr);
-
-       /*
-        * If there is no entry in the ARP cache or if this device does not
-        * support hard header caching just fall back to the IP stack.
-        */
-
-       hh = rt_hh(rt);
-       if (!hh)
-               return NULL;
-
-       /*
-        * If lock is contended fall back to directly building the header.
-        * We're not going to help performance by sitting here spinning.
-        */
-       if (!spin_trylock(&tnl_vport->cache_lock))
-               return NULL;
-
-       cache = cache_dereference(tnl_vport);
-       if (check_cache_valid(cache, mutable))
-               goto unlock;
-       else
-               cache = NULL;
-
-       cache_len = LL_RESERVED_SPACE(rt_dst(rt).dev) + tunnel_hlen;
-
-       cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
-                       cache_len, GFP_ATOMIC);
-       if (!cache)
-               goto unlock;
-
-       create_eth_hdr(cache, hh);
-       cache_data = get_cached_header(cache) + cache->hh_len;
-       cache->len = cache->hh_len + tunnel_hlen;
-
-       create_tunnel_header(vport, mutable, &tun_key, rt, cache_data);
-
-       cache->mutable_seq = mutable->seq;
-       cache->rt = rt;
-#ifdef NEED_CACHE_TIMEOUT
-       cache->expiration = jiffies + tnl_vport->cache_exp_interval;
-#endif
-
-       if (ovs_is_internal_dev(rt_dst(rt).dev)) {
-               struct sw_flow_key flow_key;
-               struct vport *dst_vport;
-               struct sk_buff *skb;
-               int err;
-               int flow_key_len;
-               struct sw_flow *flow;
-
-               dst_vport = ovs_internal_dev_get_vport(rt_dst(rt).dev);
-               if (!dst_vport)
-                       goto done;
-
-               skb = alloc_skb(cache->len, GFP_ATOMIC);
-               if (!skb)
-                       goto done;
-
-               __skb_put(skb, cache->len);
-               memcpy(skb->data, get_cached_header(cache), cache->len);
-
-               err = ovs_flow_extract(skb, dst_vport->port_no, &flow_key,
-                                      &flow_key_len);
-
-               consume_skb(skb);
-               if (err)
-                       goto done;
-
-               flow = ovs_flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
-                                          &flow_key, flow_key_len);
-               if (flow) {
-                       cache->flow = flow;
-                       ovs_flow_hold(flow);
-               }
-       }
-
-done:
-       assign_cache_rcu(vport, cache);
-
-unlock:
-       spin_unlock(&tnl_vport->cache_lock);
-
-       return cache;
-}
-
-static struct rtable *__find_route(const struct tnl_mutable_config *mutable,
-                                  __be32 saddr, __be32 daddr, u8 ipproto,
-                                  u8 tos)
+static struct rtable *find_route(struct net *net,
+               __be32 *saddr, __be32 daddr, u8 ipproto,
+               u8 tos)
 {
+       struct rtable *rt;
        /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
         * router expect RT_TOS bits only. */
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
        struct flowi fl = { .nl_u = { .ip4_u = {
                                        .daddr = daddr,
-                                       .saddr = saddr,
+                                       .saddr = *saddr,
                                        .tos   = RT_TOS(tos) } },
                                        .proto = ipproto };
-       struct rtable *rt;
 
-       if (unlikely(ip_route_output_key(port_key_get_net(&mutable->key), &rt, &fl)))
+       if (unlikely(ip_route_output_key(net, &rt, &fl)))
                return ERR_PTR(-EADDRNOTAVAIL);
-
+       *saddr = fl.nl_u.ip4_u.saddr;
        return rt;
 #else
        struct flowi4 fl = { .daddr = daddr,
-                            .saddr = saddr,
+                            .saddr = *saddr,
                             .flowi4_tos = RT_TOS(tos),
                             .flowi4_proto = ipproto };
 
-       return ip_route_output_key(port_key_get_net(&mutable->key), &fl);
+       rt = ip_route_output_key(net, &fl);
+       *saddr = fl.saddr;
+       return rt;
 #endif
 }
 
-static struct rtable *find_route(struct vport *vport,
-                                const struct tnl_mutable_config *mutable,
-                                __be32 saddr, __be32 daddr, u8 tos,
-                                struct tnl_cache **cache)
-{
-       struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
-       struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
-
-       *cache = NULL;
-       tos = RT_TOS(tos);
-
-       if (tos == RT_TOS(mutable->tos) &&
-           check_cache_valid(cur_cache, mutable)) {
-               *cache = cur_cache;
-               return cur_cache->rt;
-       } else {
-               struct rtable *rt;
-
-               rt = __find_route(mutable, saddr, daddr,
-                                 tnl_vport->tnl_ops->ipproto, tos);
-               if (IS_ERR(rt))
-                       return NULL;
-               if (likely(tos == RT_TOS(mutable->tos)))
-                       *cache = build_cache(vport, mutable, rt);
-
-               return rt;
-       }
-}
-
 static bool need_linearize(const struct sk_buff *skb)
 {
        int i;
@@ -1186,8 +854,6 @@ int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
        const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
        enum vport_err_type err = VPORT_E_TX_ERROR;
        struct rtable *rt;
-       struct dst_entry *unattached_dst = NULL;
-       struct tnl_cache *cache;
        struct ovs_key_ipv4_tunnel tun_key;
        int sent_len = 0;
        int tunnel_hlen;
@@ -1245,6 +911,13 @@ int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
                daddr = mutable->key.daddr;
                saddr = mutable->key.saddr;
 
+               if (unlikely(!daddr)) {
+                       /* Trying to sent packet from Null-port without
+                        * tunnel info? Drop this packet. */
+                       err = VPORT_E_TX_DROPPED;
+                       goto error_free;
+               }
+
                /* ToS */
                if (skb->protocol == htons(ETH_P_IP))
                        inner_tos = ip_hdr(skb)->tos;
@@ -1276,11 +949,10 @@ int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
        }
 
        /* Route lookup */
-       rt = find_route(vport, mutable, saddr, daddr, tos, &cache);
-       if (unlikely(!rt))
+       rt = find_route(port_key_get_net(&mutable->key), &saddr, daddr,
+                         tnl_vport->tnl_ops->ipproto, tos);
+       if (IS_ERR(rt))
                goto error_free;
-       if (unlikely(!cache))
-               unattached_dst = &rt_dst(rt);
 
        /* Reset SKB */
        nf_reset(skb);
@@ -1290,24 +962,15 @@ int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
 
        /* Offloading */
        skb = handle_offloads(skb, mutable, rt, tunnel_hlen);
-       if (IS_ERR(skb))
-               goto error;
+       if (IS_ERR(skb)) {
+               skb = NULL;
+               goto err_free_rt;
+       }
 
        /* MTU */
        if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off, tunnel_hlen))) {
                err = VPORT_E_TX_DROPPED;
-               goto error_free;
-       }
-
-       /*
-        * If we are over the MTU, allow the IP stack to handle fragmentation.
-        * Fragmentation is a slow path anyways.
-        */
-       if (unlikely(skb->len + tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
-                    cache)) {
-               unattached_dst = &rt_dst(rt);
-               dst_hold(unattached_dst);
-               cache = NULL;
+               goto err_free_rt;
        }
 
        /* TTL Fixup. */
@@ -1326,67 +989,34 @@ int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
                if (unlikely(vlan_deaccel_tag(skb)))
                        goto next;
 
-               if (likely(cache)) {
-                       skb_push(skb, cache->len);
-                       memcpy(skb->data, get_cached_header(cache), cache->len);
-                       skb_reset_mac_header(skb);
-                       skb_set_network_header(skb, cache->hh_len);
-
-               } else {
-                       skb_push(skb, tunnel_hlen);
-                       create_tunnel_header(vport, mutable, OVS_CB(skb)->tun_key, rt, skb->data);
-                       skb_reset_network_header(skb);
-
-                       if (next_skb)
-                               skb_dst_set(skb, dst_clone(unattached_dst));
-                       else {
-                               skb_dst_set(skb, unattached_dst);
-                               unattached_dst = NULL;
-                       }
-               }
-               skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
+               skb_push(skb, tunnel_hlen);
+               skb_reset_network_header(skb);
+               skb_set_transport_header(skb, sizeof(struct iphdr));
 
+               if (next_skb)
+                       skb_dst_set(skb, dst_clone(&rt_dst(rt)));
+               else
+                       skb_dst_set(skb, &rt_dst(rt));
+
+               /* Push IP header. */
                iph = ip_hdr(skb);
-               iph->tos = tos;
-               iph->ttl = ttl;
-               iph->frag_off = frag_off;
+               iph->version    = 4;
+               iph->ihl        = sizeof(struct iphdr) >> 2;
+               iph->protocol   = tnl_vport->tnl_ops->ipproto;
+               iph->daddr      = daddr;
+               iph->saddr      = saddr;
+               iph->tos        = tos;
+               iph->ttl        = ttl;
+               iph->frag_off   = frag_off;
                ip_select_ident(iph, &rt_dst(rt), NULL);
 
-               skb = tnl_vport->tnl_ops->update_header(vport, mutable,
+               /* Push Tunnel header. */
+               skb = tnl_vport->tnl_ops->build_header(vport, mutable,
                                                        &rt_dst(rt), skb, tunnel_hlen);
                if (unlikely(!skb))
                        goto next;
 
-               if (likely(cache)) {
-                       int orig_len = skb->len - cache->len;
-                       struct vport *cache_vport;
-
-                       cache_vport = ovs_internal_dev_get_vport(rt_dst(rt).dev);
-                       skb->protocol = htons(ETH_P_IP);
-                       iph = ip_hdr(skb);
-                       iph->tot_len = htons(skb->len - skb_network_offset(skb));
-                       ip_send_check(iph);
-
-                       if (cache_vport) {
-                               if (unlikely(compute_ip_summed(skb, true))) {
-                                       kfree_skb(skb);
-                                       goto next;
-                               }
-
-                               OVS_CB(skb)->flow = cache->flow;
-                               ovs_vport_receive(cache_vport, skb);
-                               sent_len += orig_len;
-                       } else {
-                               int xmit_err;
-
-                               skb->dev = rt_dst(rt).dev;
-                               xmit_err = dev_queue_xmit(skb);
-
-                               if (likely(net_xmit_eval(xmit_err) == 0))
-                                       sent_len += orig_len;
-                       }
-               } else
-                       sent_len += send_frags(skb, tunnel_hlen);
+               sent_len += send_frags(skb, tunnel_hlen);
 
 next:
                skb = next_skb;
@@ -1395,14 +1025,13 @@ next:
        if (unlikely(sent_len == 0))
                ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
 
-       goto out;
+       return sent_len;
 
+err_free_rt:
+       ip_rt_put(rt);
 error_free:
        ovs_tnl_free_linked_skbs(skb);
-error:
        ovs_vport_record_error(vport, err);
-out:
-       dst_release(unattached_dst);
        return sent_len;
 }
 
@@ -1414,6 +1043,7 @@ static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
        [OVS_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
        [OVS_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
        [OVS_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
+       [OVS_TUNNEL_ATTR_DST_PORT] = { .type = NLA_U16 },
 };
 
 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
@@ -1428,8 +1058,10 @@ static int tnl_set_config(struct net *net, struct nlattr *options,
        struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
        int err;
 
+       port_key_set_net(&mutable->key, net);
+       mutable->key.tunnel_type = tnl_ops->tunnel_type;
        if (!options)
-               return -EINVAL;
+               goto out;
 
        err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
        if (err)
@@ -1439,9 +1071,8 @@ static int tnl_set_config(struct net *net, struct nlattr *options,
                return -EINVAL;
 
        mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
-
-       port_key_set_net(&mutable->key, net);
        mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
+
        if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
                if (ipv4_is_multicast(mutable->key.daddr))
                        return -EINVAL;
@@ -1458,7 +1089,10 @@ static int tnl_set_config(struct net *net, struct nlattr *options,
        if (a[OVS_TUNNEL_ATTR_TTL])
                mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
 
-       mutable->key.tunnel_type = tnl_ops->tunnel_type;
+       if (a[OVS_TUNNEL_ATTR_DST_PORT])
+               mutable->dst_port =
+                       htons(nla_get_u16(a[OVS_TUNNEL_ATTR_DST_PORT]));
+
        if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
                mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
                mutable->flags |= TNL_F_IN_KEY_MATCH;
@@ -1472,17 +1106,15 @@ static int tnl_set_config(struct net *net, struct nlattr *options,
        else
                mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
 
-       old_vport = port_table_lookup(&mutable->key, &old_mutable);
-       if (old_vport && old_vport != cur_vport)
-               return -EEXIST;
-
        mutable->mlink = 0;
        if (ipv4_is_multicast(mutable->key.daddr)) {
                struct net_device *dev;
                struct rtable *rt;
+               __be32 saddr = mutable->key.saddr;
 
-               rt = __find_route(mutable, mutable->key.saddr, mutable->key.daddr,
-                                 tnl_ops->ipproto, mutable->tos);
+               rt = find_route(port_key_get_net(&mutable->key),
+                            &saddr, mutable->key.daddr,
+                            tnl_ops->ipproto, mutable->tos);
                if (IS_ERR(rt))
                        return -EADDRNOTAVAIL;
                dev = rt_dst(rt).dev;
@@ -1493,6 +1125,11 @@ static int tnl_set_config(struct net *net, struct nlattr *options,
                ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
        }
 
+out:
+       old_vport = port_table_lookup(&mutable->key, &old_mutable);
+       if (old_vport && old_vport != cur_vport)
+               return -EEXIST;
+
        return 0;
 }
 
@@ -1533,13 +1170,6 @@ struct vport *ovs_tnl_create(const struct vport_parms *parms,
        if (err)
                goto error_free_mutable;
 
-       spin_lock_init(&tnl_vport->cache_lock);
-
-#ifdef NEED_CACHE_TIMEOUT
-       tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
-                                      (net_random() % (MAX_CACHE_EXP / 2));
-#endif
-
        rcu_assign_pointer(tnl_vport->mutable, mutable);
 
        port_table_add_port(vport);
@@ -1561,6 +1191,10 @@ int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
        struct tnl_mutable_config *mutable;
        int err;
 
+       old_mutable = rtnl_dereference(tnl_vport->mutable);
+       if (!old_mutable->key.daddr)
+               return -EINVAL;
+
        mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
        if (!mutable) {
                err = -ENOMEM;
@@ -1568,7 +1202,6 @@ int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
        }
 
        /* Copy fields whose values should be retained. */
-       old_mutable = rtnl_dereference(tnl_vport->mutable);
        mutable->seq = old_mutable->seq + 1;
        memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
 
@@ -1615,6 +1248,9 @@ int ovs_tnl_get_options(const struct vport *vport, struct sk_buff *skb)
                goto nla_put_failure;
        if (mutable->ttl && nla_put_u8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl))
                goto nla_put_failure;
+       if (mutable->dst_port && nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT,
+                                            ntohs(mutable->dst_port)))
+               goto nla_put_failure;
 
        return 0;
 
@@ -1627,7 +1263,6 @@ static void free_port_rcu(struct rcu_head *rcu)
        struct tnl_vport *tnl_vport = container_of(rcu,
                                                   struct tnl_vport, rcu);
 
-       free_cache((struct tnl_cache __force *)tnl_vport->cache);
        kfree((struct tnl_mutable __force *)tnl_vport->mutable);
        ovs_vport_free(tnl_vport_to_vport(tnl_vport));
 }