datapath: Fix unused 'done' label build warning in flow.c.
[sliver-openvswitch.git] / datapath / flow.c
index c37c8e0..a13edeb 100644 (file)
@@ -1,6 +1,6 @@
 /*
  * Distributed under the terms of the GNU GPL version 2.
- * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
+ * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
  *
  * Significant portions of this file may be copied from parts of the Linux
  * kernel, by Linus Torvalds and others.
@@ -8,6 +8,7 @@
 
 #include "flow.h"
 #include "datapath.h"
+#include <asm/uaccess.h>
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
 #include <linux/if_ether.h>
 #include <linux/if_arp.h>
 #include <linux/if_ether.h>
 #include <linux/ip.h>
+#include <linux/ipv6.h>
 #include <linux/tcp.h>
 #include <linux/udp.h>
 #include <linux/icmp.h>
+#include <linux/icmpv6.h>
 #include <net/inet_ecn.h>
 #include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/ndisc.h>
 
-#include "compat.h"
+#include "vlan.h"
 
-struct kmem_cache *flow_cache;
-static unsigned int hash_seed;
+static struct kmem_cache *flow_cache;
+static unsigned int hash_seed __read_mostly;
+
+static int check_header(struct sk_buff *skb, int len)
+{
+       if (unlikely(skb->len < len))
+               return -EINVAL;
+       if (unlikely(!pskb_may_pull(skb, len)))
+               return -ENOMEM;
+       return 0;
+}
 
 static inline bool arphdr_ok(struct sk_buff *skb)
 {
-       int nh_ofs = skb_network_offset(skb);
-       return pskb_may_pull(skb, nh_ofs + sizeof(struct arp_eth_header));
+       return pskb_may_pull(skb, skb_network_offset(skb) +
+                                 sizeof(struct arp_eth_header));
 }
 
-static inline bool iphdr_ok(struct sk_buff *skb)
+static inline int check_iphdr(struct sk_buff *skb)
 {
-       int nh_ofs = skb_network_offset(skb);
-       if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
-               int ip_len = ip_hdrlen(skb);
-               return (ip_len >= sizeof(struct iphdr)
-                       && pskb_may_pull(skb, nh_ofs + ip_len));
-       }
-       return false;
+       unsigned int nh_ofs = skb_network_offset(skb);
+       unsigned int ip_len;
+       int err;
+
+       err = check_header(skb, nh_ofs + sizeof(struct iphdr));
+       if (unlikely(err))
+               return err;
+
+       ip_len = ip_hdrlen(skb);
+       if (unlikely(ip_len < sizeof(struct iphdr) ||
+                    skb->len < nh_ofs + ip_len))
+               return -EINVAL;
+
+       skb_set_transport_header(skb, nh_ofs + ip_len);
+       return 0;
 }
 
 static inline bool tcphdr_ok(struct sk_buff *skb)
 {
        int th_ofs = skb_transport_offset(skb);
-       if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
-               int tcp_len = tcp_hdrlen(skb);
-               return (tcp_len >= sizeof(struct tcphdr)
-                       && skb->len >= th_ofs + tcp_len);
-       }
-       return false;
+       int tcp_len;
+
+       if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
+               return false;
+
+       tcp_len = tcp_hdrlen(skb);
+       if (unlikely(tcp_len < sizeof(struct tcphdr) ||
+                    skb->len < th_ofs + tcp_len))
+               return false;
+
+       return true;
 }
 
 static inline bool udphdr_ok(struct sk_buff *skb)
 {
-       int th_ofs = skb_transport_offset(skb);
-       return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
+       return pskb_may_pull(skb, skb_transport_offset(skb) +
+                                 sizeof(struct udphdr));
 }
 
 static inline bool icmphdr_ok(struct sk_buff *skb)
 {
-       int th_ofs = skb_transport_offset(skb);
-       return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
+       return pskb_may_pull(skb, skb_transport_offset(skb) +
+                                 sizeof(struct icmphdr));
+}
+
+u64 flow_used_time(unsigned long flow_jiffies)
+{
+       struct timespec cur_ts;
+       u64 cur_ms, idle_ms;
+
+       ktime_get_ts(&cur_ts);
+       idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
+       cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
+                cur_ts.tv_nsec / NSEC_PER_MSEC;
+
+       return cur_ms - idle_ms;
+}
+
+#define SW_FLOW_KEY_OFFSET(field)              \
+       offsetof(struct sw_flow_key, field) +   \
+       FIELD_SIZEOF(struct sw_flow_key, field)
+
+static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
+                        int *key_lenp)
+{
+       unsigned int nh_ofs = skb_network_offset(skb);
+       unsigned int nh_len;
+       int payload_ofs;
+       struct ipv6hdr *nh;
+       uint8_t nexthdr;
+       int err;
+
+       *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
+
+       err = check_header(skb, nh_ofs + sizeof(*nh));
+       if (unlikely(err))
+               return err;
+
+       nh = ipv6_hdr(skb);
+       nexthdr = nh->nexthdr;
+       payload_ofs = (u8 *)(nh + 1) - skb->data;
+
+       key->ip.proto = NEXTHDR_NONE;
+       key->ip.tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
+       ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
+       ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
+
+       payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
+       if (unlikely(payload_ofs < 0))
+               return -EINVAL;
+
+       nh_len = payload_ofs - nh_ofs;
+       skb_set_transport_header(skb, nh_ofs + nh_len);
+       key->ip.proto = nexthdr;
+       return nh_len;
+}
+
+static bool icmp6hdr_ok(struct sk_buff *skb)
+{
+       return pskb_may_pull(skb, skb_transport_offset(skb) +
+                                 sizeof(struct icmp6hdr));
 }
 
 #define TCP_FLAGS_OFFSET 13
@@ -81,8 +166,8 @@ void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 {
        u8 tcp_flags = 0;
 
-       if (flow->key.dl_type == htons(ETH_P_IP) &&
-           flow->key.nw_proto == IPPROTO_TCP) {
+       if (flow->key.eth.type == htons(ETH_P_IP) &&
+           flow->key.ip.proto == IPPROTO_TCP) {
                u8 *tcp = (u8 *)tcp_hdr(skb);
                tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
        }
@@ -95,43 +180,57 @@ void flow_used(struct sw_flow *flow, struct sk_buff *skb)
        spin_unlock_bh(&flow->lock);
 }
 
-struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
+struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 {
+       int actions_len = nla_len(actions);
        struct sw_flow_actions *sfa;
 
-       if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
+       /* At least DP_MAX_PORTS actions are required to be able to flood a
+        * packet to every port.  Factor of 2 allows for setting VLAN tags,
+        * etc. */
+       if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
                return ERR_PTR(-EINVAL);
 
-       sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
-                     GFP_KERNEL);
+       sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
        if (!sfa)
                return ERR_PTR(-ENOMEM);
 
-       sfa->n_actions = n_actions;
+       sfa->actions_len = actions_len;
+       memcpy(sfa->actions, nla_data(actions), actions_len);
        return sfa;
 }
 
-
-/* Frees 'flow' immediately. */
-static void flow_free(struct sw_flow *flow)
+struct sw_flow *flow_alloc(void)
 {
-       if (unlikely(!flow))
-               return;
-       kfree(flow->sf_acts);
-       kmem_cache_free(flow_cache, flow);
+       struct sw_flow *flow;
+
+       flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
+       if (!flow)
+               return ERR_PTR(-ENOMEM);
+
+       spin_lock_init(&flow->lock);
+       atomic_set(&flow->refcnt, 1);
+       flow->sf_acts = NULL;
+       flow->dead = false;
+
+       return flow;
 }
 
 void flow_free_tbl(struct tbl_node *node)
 {
        struct sw_flow *flow = flow_cast(node);
-       flow_free(flow);
+
+       flow->dead = true;
+       flow_put(flow);
 }
 
 /* RCU callback used by flow_deferred_free. */
 static void rcu_free_flow_callback(struct rcu_head *rcu)
 {
        struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
-       flow_free(flow);
+
+       flow->dead = true;
+       flow_put(flow);
 }
 
 /* Schedules 'flow' to be freed after the next RCU grace period.
@@ -141,10 +240,26 @@ void flow_deferred_free(struct sw_flow *flow)
        call_rcu(&flow->rcu, rcu_free_flow_callback);
 }
 
+void flow_hold(struct sw_flow *flow)
+{
+       atomic_inc(&flow->refcnt);
+}
+
+void flow_put(struct sw_flow *flow)
+{
+       if (unlikely(!flow))
+               return;
+
+       if (atomic_dec_and_test(&flow->refcnt)) {
+               kfree((struct sf_flow_acts __force *)flow->sf_acts);
+               kmem_cache_free(flow_cache, flow);
+       }
+}
+
 /* RCU callback used by flow_deferred_free_acts. */
 static void rcu_free_acts_callback(struct rcu_head *rcu)
 {
-       struct sw_flow_actions *sf_acts = container_of(rcu, 
+       struct sw_flow_actions *sf_acts = container_of(rcu,
                        struct sw_flow_actions, rcu);
        kfree(sf_acts);
 }
@@ -156,7 +271,7 @@ void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
        call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 }
 
-static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key)
+static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 {
        struct qtag_prefix {
                __be16 eth_type; /* ETH_P_8021Q */
@@ -164,13 +279,15 @@ static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key)
        };
        struct qtag_prefix *qp;
 
-       if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
-               return;
+       if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
+                                        sizeof(__be16))))
+               return -ENOMEM;
 
        qp = (struct qtag_prefix *) skb->data;
-       key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
-       key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
+       key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
        __skb_pull(skb, sizeof(struct qtag_prefix));
+
+       return 0;
 }
 
 static __be16 parse_ethertype(struct sk_buff *skb)
@@ -180,7 +297,7 @@ static __be16 parse_ethertype(struct sk_buff *skb)
                u8  ssap;  /* Always 0xAA */
                u8  ctrl;
                u8  oui[3];
-               u16 ethertype;
+               __be16 ethertype;
        };
        struct llc_snap_hdr *llc;
        __be16 proto;
@@ -188,115 +305,221 @@ static __be16 parse_ethertype(struct sk_buff *skb)
        proto = *(__be16 *) skb->data;
        __skb_pull(skb, sizeof(__be16));
 
-       if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF)
+       if (ntohs(proto) >= 1536)
                return proto;
 
-       if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
-               return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
+       if (skb->len < sizeof(struct llc_snap_hdr))
+               return htons(ETH_P_802_2);
+
+       if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
+               return htons(0);
 
        llc = (struct llc_snap_hdr *) skb->data;
        if (llc->dsap != LLC_SAP_SNAP ||
            llc->ssap != LLC_SAP_SNAP ||
            (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
-               return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
+               return htons(ETH_P_802_2);
 
        __skb_pull(skb, sizeof(struct llc_snap_hdr));
        return llc->ethertype;
 }
 
+static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
+                       int *key_lenp, int nh_len)
+{
+       struct icmp6hdr *icmp = icmp6_hdr(skb);
+       int error = 0;
+       int key_len;
+
+       /* The ICMPv6 type and code fields use the 16-bit transport port
+        * fields, so we need to store them in 16-bit network byte order.
+        */
+       key->ipv6.tp.src = htons(icmp->icmp6_type);
+       key->ipv6.tp.dst = htons(icmp->icmp6_code);
+       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+
+       if (icmp->icmp6_code == 0 &&
+           (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
+            icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
+               int icmp_len = skb->len - skb_transport_offset(skb);
+               struct nd_msg *nd;
+               int offset;
+
+               key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
+
+               /* In order to process neighbor discovery options, we need the
+                * entire packet.
+                */
+               if (unlikely(icmp_len < sizeof(*nd)))
+                       goto out;
+               if (unlikely(skb_linearize(skb))) {
+                       error = -ENOMEM;
+                       goto out;
+               }
+
+               nd = (struct nd_msg *)skb_transport_header(skb);
+               ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
+               key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
+
+               icmp_len -= sizeof(*nd);
+               offset = 0;
+               while (icmp_len >= 8) {
+                       struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
+                       int opt_len = nd_opt->nd_opt_len * 8;
+
+                       if (unlikely(!opt_len || opt_len > icmp_len))
+                               goto invalid;
+
+                       /* Store the link layer address if the appropriate
+                        * option is provided.  It is considered an error if
+                        * the same link layer option is specified twice.
+                        */
+                       if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
+                           && opt_len == 8) {
+                               if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
+                                       goto invalid;
+                               memcpy(key->ipv6.nd.sll,
+                                   &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
+                       } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
+                                  && opt_len == 8) {
+                               if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
+                                       goto invalid;
+                               memcpy(key->ipv6.nd.tll,
+                                   &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
+                       }
+
+                       icmp_len -= opt_len;
+                       offset += opt_len;
+               }
+       }
+
+       goto out;
+
+invalid:
+       memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
+       memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
+       memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
+
+out:
+       *key_lenp = key_len;
+       return error;
+}
+
 /**
  * flow_extract - extracts a flow key from an Ethernet frame.
  * @skb: sk_buff that contains the frame, with skb->data pointing to the
  * Ethernet header
  * @in_port: port number on which @skb was received.
  * @key: output flow key
+ * @key_lenp: length of output flow key
+ * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
+ * not contain an IPv4 packet or if it is not a fragment.
  *
  * The caller must ensure that skb->len >= ETH_HLEN.
  *
- * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
- * %false.
+ * Returns 0 if successful, otherwise a negative errno value.
+ *
+ * Initializes @skb header pointers as follows:
+ *
+ *    - skb->mac_header: the Ethernet header.
+ *
+ *    - skb->network_header: just past the Ethernet header, or just past the
+ *      VLAN header, to the first byte of the Ethernet payload.
+ *
+ *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
+ *      on output, then just past the IP header, if one is present and
+ *      of a correct length, otherwise the same as skb->network_header.
+ *      For other key->dl_type values it is left untouched.
  */
-int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
+int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
+                int *key_lenp, bool *is_frag)
 {
+       int error = 0;
+       int key_len = SW_FLOW_KEY_OFFSET(eth);
        struct ethhdr *eth;
 
-       memset(key, 0, sizeof *key);
-       key->tun_id = OVS_CB(skb)->tun_id;
-       key->in_port = in_port;
-       key->dl_vlan = htons(ODP_VLAN_NONE);
-       OVS_CB(skb)->is_frag = false;
-
-       if (!pskb_may_pull(skb, min(skb->len, 64u)))
-               return 0;
+       memset(key, 0, sizeof(*key));
+       key->eth.tun_id = OVS_CB(skb)->tun_id;
+       key->eth.in_port = in_port;
+       *is_frag = false;
 
        skb_reset_mac_header(skb);
 
-       /* Link layer. */
+       /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
+        * header in the linear data area.
+        */
        eth = eth_hdr(skb);
-       memcpy(key->dl_src, eth->h_source, ETH_ALEN);
-       memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
+       memcpy(key->eth.src, eth->h_source, ETH_ALEN);
+       memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 
-       /* dl_type, dl_vlan, dl_vlan_pcp. */
        __skb_pull(skb, 2 * ETH_ALEN);
-       if (eth->h_proto == htons(ETH_P_8021Q))
-               parse_vlan(skb, key);
-       key->dl_type = parse_ethertype(skb);
+
+       if (vlan_tx_tag_present(skb))
+               key->eth.tci = htons(vlan_get_tci(skb));
+       else if (eth->h_proto == htons(ETH_P_8021Q))
+               if (unlikely(parse_vlan(skb, key)))
+                       return -ENOMEM;
+
+       key->eth.type = parse_ethertype(skb);
+       if (unlikely(key->eth.type == htons(0)))
+               return -ENOMEM;
+
        skb_reset_network_header(skb);
-       __skb_push(skb, skb->data - (unsigned char *)eth);
+       __skb_push(skb, skb->data - skb_mac_header(skb));
 
        /* Network layer. */
-       if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
-               struct iphdr *nh = ip_hdr(skb);
-               int th_ofs = skb_network_offset(skb) + nh->ihl * 4;
-               key->nw_src = nh->saddr;
-               key->nw_dst = nh->daddr;
-               key->nw_tos = nh->tos & ~INET_ECN_MASK;
-               key->nw_proto = nh->protocol;
-               skb_set_transport_header(skb, th_ofs);
+       if (key->eth.type == htons(ETH_P_IP)) {
+               struct iphdr *nh;
+
+               key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
+
+               error = check_iphdr(skb);
+               if (unlikely(error)) {
+                       if (error == -EINVAL) {
+                               skb->transport_header = skb->network_header;
+                               error = 0;
+                       }
+                       goto out;
+               }
+
+               nh = ip_hdr(skb);
+               key->ipv4.addr.src = nh->saddr;
+               key->ipv4.addr.dst = nh->daddr;
+               key->ip.tos = nh->tos & ~INET_ECN_MASK;
+               key->ip.proto = nh->protocol;
 
                /* Transport layer. */
-               if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
-                       if (key->nw_proto == IPPROTO_TCP) {
-                               if (tcphdr_ok(skb)) {
-                                       struct tcphdr *tcp = tcp_hdr(skb);
-                                       key->tp_src = tcp->source;
-                                       key->tp_dst = tcp->dest;
-                               } else {
-                                       /* Avoid tricking other code into
-                                        * thinking that this packet has an L4
-                                        * header. */
-                                       key->nw_proto = 0;
-                               }
-                       } else if (key->nw_proto == IPPROTO_UDP) {
-                               if (udphdr_ok(skb)) {
-                                       struct udphdr *udp = udp_hdr(skb);
-                                       key->tp_src = udp->source;
-                                       key->tp_dst = udp->dest;
-                               } else {
-                                       /* Avoid tricking other code into
-                                        * thinking that this packet has an L4
-                                        * header. */
-                                       key->nw_proto = 0;
-                               }
-                       } else if (key->nw_proto == IPPROTO_ICMP) {
-                               if (icmphdr_ok(skb)) {
-                                       struct icmphdr *icmp = icmp_hdr(skb);
-                                       /* The ICMP type and code fields use the 16-bit
-                                        * transport port fields, so we need to store them
-                                        * in 16-bit network byte order. */
-                                       key->tp_src = htons(icmp->type);
-                                       key->tp_dst = htons(icmp->code);
-                               } else {
-                                       /* Avoid tricking other code into
-                                        * thinking that this packet has an L4
-                                        * header. */
-                                       key->nw_proto = 0;
-                               }
+               if ((nh->frag_off & htons(IP_MF | IP_OFFSET)) ||
+                   (skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
+                       *is_frag = true;
+
+               if (key->ip.proto == IPPROTO_TCP) {
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
+                       if (!*is_frag && tcphdr_ok(skb)) {
+                               struct tcphdr *tcp = tcp_hdr(skb);
+                               key->ipv4.tp.src = tcp->source;
+                               key->ipv4.tp.dst = tcp->dest;
+                       }
+               } else if (key->ip.proto == IPPROTO_UDP) {
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
+                       if (!*is_frag && udphdr_ok(skb)) {
+                               struct udphdr *udp = udp_hdr(skb);
+                               key->ipv4.tp.src = udp->source;
+                               key->ipv4.tp.dst = udp->dest;
+                       }
+               } else if (key->ip.proto == IPPROTO_ICMP) {
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
+                       if (!*is_frag && icmphdr_ok(skb)) {
+                               struct icmphdr *icmp = icmp_hdr(skb);
+                               /* The ICMP type and code fields use the 16-bit
+                                * transport port fields, so we need to store them
+                                * in 16-bit network byte order. */
+                               key->ipv4.tp.src = htons(icmp->type);
+                               key->ipv4.tp.dst = htons(icmp->code);
                        }
-               } else {
-                       OVS_CB(skb)->is_frag = true;
                }
-       } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
+
+       } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
                struct arp_eth_header *arp;
 
                arp = (struct arp_eth_header *)skb_network_header(skb);
@@ -307,33 +530,551 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
                                && arp->ar_pln == 4) {
 
                        /* We only match on the lower 8 bits of the opcode. */
-                       if (ntohs(arp->ar_op) <= 0xff) {
-                               key->nw_proto = ntohs(arp->ar_op);
+                       if (ntohs(arp->ar_op) <= 0xff)
+                               key->ip.proto = ntohs(arp->ar_op);
+
+                       if (key->ip.proto == ARPOP_REQUEST
+                                       || key->ip.proto == ARPOP_REPLY) {
+                               memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
+                               memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
+                               memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
+                               memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
+                               key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
                        }
+               }
+       } else if (key->eth.type == htons(ETH_P_IPV6)) {
+               int nh_len;             /* IPv6 Header + Extensions */
+
+               nh_len = parse_ipv6hdr(skb, key, &key_len);
+               if (unlikely(nh_len < 0)) {
+                       if (nh_len == -EINVAL)
+                               skb->transport_header = skb->network_header;
+                       else
+                               error = nh_len;
+                       goto out;
+               }
 
-                       if (key->nw_proto == ARPOP_REQUEST 
-                                       || key->nw_proto == ARPOP_REPLY) {
-                               memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
-                               memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
+               /* Transport layer. */
+               if (key->ip.proto == NEXTHDR_TCP) {
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+                       if (tcphdr_ok(skb)) {
+                               struct tcphdr *tcp = tcp_hdr(skb);
+                               key->ipv6.tp.src = tcp->source;
+                               key->ipv6.tp.dst = tcp->dest;
+                       }
+               } else if (key->ip.proto == NEXTHDR_UDP) {
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+                       if (udphdr_ok(skb)) {
+                               struct udphdr *udp = udp_hdr(skb);
+                               key->ipv6.tp.src = udp->source;
+                               key->ipv6.tp.dst = udp->dest;
+                       }
+               } else if (key->ip.proto == NEXTHDR_ICMP) {
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+                       if (icmp6hdr_ok(skb)) {
+                               error = parse_icmpv6(skb, key, &key_len, nh_len);
+                               if (error < 0)
+                                       goto out;
                        }
                }
-       } else {
-               skb_reset_transport_header(skb);
        }
-       return 0;
+
+out:
+       *key_lenp = key_len;
+       return error;
 }
 
-u32 flow_hash(const struct odp_flow_key *key)
+u32 flow_hash(const struct sw_flow_key *key, int key_len)
 {
-       return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
+       return jhash2((u32*)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 }
 
-int flow_cmp(const struct tbl_node *node, void *key2_)
+int flow_cmp(const struct tbl_node *node, void *key2_, int len)
 {
-       const struct odp_flow_key *key1 = &flow_cast(node)->key;
-       const struct odp_flow_key *key2 = key2_;
+       const struct sw_flow_key *key1 = &flow_cast(node)->key;
+       const struct sw_flow_key *key2 = key2_;
+
+       return !memcmp(key1, key2, len);
+}
+
+/* The size of the argument for each %ODP_KEY_ATTR_* Netlink attribute.  */
+static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
+       [ODP_KEY_ATTR_TUN_ID] = 8,
+       [ODP_KEY_ATTR_IN_PORT] = 4,
+       [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
+       [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
+       [ODP_KEY_ATTR_ETHERTYPE] = 2,
+       [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
+       [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
+       [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
+       [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
+       [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
+       [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
+       [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
+       [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
+};
+
+/**
+ * flow_from_nlattrs - parses Netlink attributes into a flow key.
+ * @swkey: receives the extracted flow key.
+ * @key_lenp: number of bytes used in @swkey.
+ * @attr: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
+ * sequence.
+ *
+ * This state machine accepts the following forms, with [] for optional
+ * elements and | for alternatives:
+ *
+ * [tun_id] [in_port] ethernet [8021q] [ethertype \
+ *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
+ */
+int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
+                     const struct nlattr *attr)
+{
+       int error = 0;
+       const struct nlattr *nla;
+       u16 prev_type;
+       int rem;
+       int key_len;
+
+       memset(swkey, 0, sizeof(*swkey));
+       swkey->eth.in_port = USHRT_MAX;
+       swkey->eth.type = htons(ETH_P_802_2);
+       key_len = SW_FLOW_KEY_OFFSET(eth);
+
+       prev_type = ODP_KEY_ATTR_UNSPEC;
+       nla_for_each_nested(nla, attr, rem) {
+               const struct odp_key_ethernet *eth_key;
+               const struct odp_key_8021q *q_key;
+               const struct odp_key_ipv4 *ipv4_key;
+               const struct odp_key_ipv6 *ipv6_key;
+               const struct odp_key_tcp *tcp_key;
+               const struct odp_key_udp *udp_key;
+               const struct odp_key_icmp *icmp_key;
+               const struct odp_key_icmpv6 *icmpv6_key;
+               const struct odp_key_arp *arp_key;
+               const struct odp_key_nd *nd_key;
+
+                int type = nla_type(nla);
+
+                if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
+                        goto invalid;
+
+#define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
+               switch (TRANSITION(prev_type, type)) {
+               case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
+                       swkey->eth.tun_id = nla_get_be64(nla);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
+               case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
+                       if (nla_get_u32(nla) >= DP_MAX_PORTS)
+                               goto invalid;
+                       swkey->eth.in_port = nla_get_u32(nla);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_ETHERNET):
+               case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_ETHERNET):
+               case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
+                       eth_key = nla_data(nla);
+                       memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
+                       memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
+                       q_key = nla_data(nla);
+                       /* Only standard 0x8100 VLANs currently supported. */
+                       if (q_key->q_tpid != htons(ETH_P_8021Q))
+                               goto invalid;
+                       if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
+                               goto invalid;
+                       swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
+               case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
+                       swkey->eth.type = nla_get_be16(nla);
+                       if (ntohs(swkey->eth.type) < 1536)
+                               goto invalid;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
+                       if (swkey->eth.type != htons(ETH_P_IP))
+                               goto invalid;
+                       ipv4_key = nla_data(nla);
+                       swkey->ip.proto = ipv4_key->ipv4_proto;
+                       swkey->ip.tos = ipv4_key->ipv4_tos;
+                       swkey->ipv4.addr.src = ipv4_key->ipv4_src;
+                       swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
+                       if (swkey->ip.tos & INET_ECN_MASK)
+                               goto invalid;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
+                       if (swkey->eth.type != htons(ETH_P_IPV6))
+                               goto invalid;
+                       ipv6_key = nla_data(nla);
+                       swkey->ip.proto = ipv6_key->ipv6_proto;
+                       swkey->ip.tos = ipv6_key->ipv6_tos;
+                       memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
+                                       sizeof(swkey->ipv6.addr.src));
+                       memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
+                                       sizeof(swkey->ipv6.addr.dst));
+                       if (swkey->ip.tos & INET_ECN_MASK)
+                               goto invalid;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
+                       if (swkey->ip.proto != IPPROTO_TCP)
+                               goto invalid;
+                       tcp_key = nla_data(nla);
+                       swkey->ipv4.tp.src = tcp_key->tcp_src;
+                       swkey->ipv4.tp.dst = tcp_key->tcp_dst;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+                       if (swkey->ip.proto != IPPROTO_TCP)
+                               goto invalid;
+                       tcp_key = nla_data(nla);
+                       swkey->ipv6.tp.src = tcp_key->tcp_src;
+                       swkey->ipv6.tp.dst = tcp_key->tcp_dst;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
+                       if (swkey->ip.proto != IPPROTO_UDP)
+                               goto invalid;
+                       udp_key = nla_data(nla);
+                       swkey->ipv4.tp.src = udp_key->udp_src;
+                       swkey->ipv4.tp.dst = udp_key->udp_dst;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+                       if (swkey->ip.proto != IPPROTO_UDP)
+                               goto invalid;
+                       udp_key = nla_data(nla);
+                       swkey->ipv6.tp.src = udp_key->udp_src;
+                       swkey->ipv6.tp.dst = udp_key->udp_dst;
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
+                       if (swkey->ip.proto != IPPROTO_ICMP)
+                               goto invalid;
+                       icmp_key = nla_data(nla);
+                       swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
+                       swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
+                       if (swkey->ip.proto != IPPROTO_ICMPV6)
+                               goto invalid;
+                       icmpv6_key = nla_data(nla);
+                       swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
+                       swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
+                       if (swkey->eth.type != htons(ETH_P_ARP))
+                               goto invalid;
+                       arp_key = nla_data(nla);
+                       swkey->ipv4.addr.src = arp_key->arp_sip;
+                       swkey->ipv4.addr.dst = arp_key->arp_tip;
+                       if (arp_key->arp_op & htons(0xff00))
+                               goto invalid;
+                       swkey->ip.proto = ntohs(arp_key->arp_op);
+                       memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
+                       memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
+                       key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
+                       if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
+                           && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
+                               goto invalid;
+                       nd_key = nla_data(nla);
+                       memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
+                                       sizeof(swkey->ipv6.nd.target));
+                       memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
+                       memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
+                       break;
+
+               default:
+                       goto invalid;
+               }
+
+               prev_type = type;
+       }
+       if (rem)
+               goto invalid;
+
+       switch (prev_type) {
+       case ODP_KEY_ATTR_UNSPEC:
+               goto invalid;
+
+       case ODP_KEY_ATTR_TUN_ID:
+       case ODP_KEY_ATTR_IN_PORT:
+               goto invalid;
+
+       case ODP_KEY_ATTR_ETHERNET:
+       case ODP_KEY_ATTR_8021Q:
+               goto ok;
+
+       case ODP_KEY_ATTR_ETHERTYPE:
+               if (swkey->eth.type == htons(ETH_P_IP) ||
+                   swkey->eth.type == htons(ETH_P_ARP))
+                       goto invalid;
+               goto ok;
+
+       case ODP_KEY_ATTR_IPV4:
+               if (swkey->ip.proto == IPPROTO_TCP ||
+                   swkey->ip.proto == IPPROTO_UDP ||
+                   swkey->ip.proto == IPPROTO_ICMP)
+                       goto invalid;
+               goto ok;
+
+       case ODP_KEY_ATTR_IPV6:
+               if (swkey->ip.proto == IPPROTO_TCP ||
+                   swkey->ip.proto == IPPROTO_UDP ||
+                   swkey->ip.proto == IPPROTO_ICMPV6)
+                       goto invalid;
+               goto ok;
+
+       case ODP_KEY_ATTR_ICMPV6:
+               if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
+                   swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
+                       goto invalid;
+               goto ok;
+
+       case ODP_KEY_ATTR_TCP:
+       case ODP_KEY_ATTR_UDP:
+       case ODP_KEY_ATTR_ICMP:
+       case ODP_KEY_ATTR_ARP:
+       case ODP_KEY_ATTR_ND:
+               goto ok;
+
+       default:
+               WARN_ON_ONCE(1);
+       }
+
+invalid:
+       error = -EINVAL;
+
+ok:
+       WARN_ON_ONCE(!key_len && !error);
+       *key_lenp = key_len;
+       return error;
+}
+
+/**
+ * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
+ * @in_port: receives the extracted input port.
+ * @tun_id: receives the extracted tunnel ID.
+ * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
+ * sequence.
+ *
+ * This parses a series of Netlink attributes that form a flow key, which must
+ * take the same form accepted by flow_from_nlattrs(), but only enough of it to
+ * get the metadata, that is, the parts of the flow key that cannot be
+ * extracted from the packet itself.
+ */
+int flow_metadata_from_nlattrs(u16 *in_port, __be64 *tun_id,
+                              const struct nlattr *attr)
+{
+       const struct nlattr *nla;
+       u16 prev_type;
+       int rem;
+
+       *in_port = USHRT_MAX;
+       *tun_id = 0;
+
+       prev_type = ODP_KEY_ATTR_UNSPEC;
+       nla_for_each_nested(nla, attr, rem) {
+                int type = nla_type(nla);
+
+                if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
+                        return -EINVAL;
+
+               switch (TRANSITION(prev_type, type)) {
+               case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
+                       *tun_id = nla_get_be64(nla);
+                       break;
+
+               case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
+               case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
+                       if (nla_get_u32(nla) >= DP_MAX_PORTS)
+                               return -EINVAL;
+                       *in_port = nla_get_u32(nla);
+                       break;
+
+               default:
+                       return 0;
+               }
+
+               prev_type = type;
+       }
+       if (rem)
+               return -EINVAL;
+
+       return 0;
+}
+
+int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
+{
+       struct odp_key_ethernet *eth_key;
+       struct nlattr *nla;
+
+       /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
+        * to be updated, but will at least raise awareness when new ODP key
+        * types are added. */
+       BUILD_BUG_ON(__ODP_KEY_ATTR_MAX != 14);
+
+       if (swkey->eth.tun_id != cpu_to_be64(0))
+               NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->eth.tun_id);
+
+       if (swkey->eth.in_port != USHRT_MAX)
+               NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->eth.in_port);
+
+       nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
+       if (!nla)
+               goto nla_put_failure;
+       eth_key = nla_data(nla);
+       memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
+       memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
+
+       if (swkey->eth.tci != htons(0)) {
+               struct odp_key_8021q q_key;
+
+               q_key.q_tpid = htons(ETH_P_8021Q);
+               q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
+               NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
+       }
+
+       if (swkey->eth.type == htons(ETH_P_802_2))
+               return 0;
+
+       NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->eth.type);
+
+       if (swkey->eth.type == htons(ETH_P_IP)) {
+               struct odp_key_ipv4 *ipv4_key;
+
+               nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
+               if (!nla)
+                       goto nla_put_failure;
+               ipv4_key = nla_data(nla);
+               memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
+               ipv4_key->ipv4_src = swkey->ipv4.addr.src;
+               ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
+               ipv4_key->ipv4_proto = swkey->ip.proto;
+               ipv4_key->ipv4_tos = swkey->ip.tos;
+       } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
+               struct odp_key_ipv6 *ipv6_key;
+
+               nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
+               if (!nla)
+                       goto nla_put_failure;
+               ipv6_key = nla_data(nla);
+               memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
+               memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
+                               sizeof(ipv6_key->ipv6_src));
+               memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
+                               sizeof(ipv6_key->ipv6_dst));
+               ipv6_key->ipv6_proto = swkey->ip.proto;
+               ipv6_key->ipv6_tos = swkey->ip.tos;
+       } else if (swkey->eth.type == htons(ETH_P_ARP)) {
+               struct odp_key_arp *arp_key;
+
+               nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
+               if (!nla)
+                       goto nla_put_failure;
+               arp_key = nla_data(nla);
+               memset(arp_key, 0, sizeof(struct odp_key_arp));
+               arp_key->arp_sip = swkey->ipv4.addr.src;
+               arp_key->arp_tip = swkey->ipv4.addr.dst;
+               arp_key->arp_op = htons(swkey->ip.proto);
+               memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
+               memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
+       }
+
+       if (swkey->eth.type == htons(ETH_P_IP) ||
+           swkey->eth.type == htons(ETH_P_IPV6)) {
+
+               if (swkey->ip.proto == IPPROTO_TCP) {
+                       struct odp_key_tcp *tcp_key;
+
+                       nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
+                       if (!nla)
+                               goto nla_put_failure;
+                       tcp_key = nla_data(nla);
+                       if (swkey->eth.type == htons(ETH_P_IP)) {
+                               tcp_key->tcp_src = swkey->ipv4.tp.src;
+                               tcp_key->tcp_dst = swkey->ipv4.tp.dst;
+                       } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
+                               tcp_key->tcp_src = swkey->ipv6.tp.src;
+                               tcp_key->tcp_dst = swkey->ipv6.tp.dst;
+                       }
+               } else if (swkey->ip.proto == IPPROTO_UDP) {
+                       struct odp_key_udp *udp_key;
+
+                       nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
+                       if (!nla)
+                               goto nla_put_failure;
+                       udp_key = nla_data(nla);
+                       if (swkey->eth.type == htons(ETH_P_IP)) {
+                               udp_key->udp_src = swkey->ipv4.tp.src;
+                               udp_key->udp_dst = swkey->ipv4.tp.dst;
+                       } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
+                               udp_key->udp_src = swkey->ipv6.tp.src;
+                               udp_key->udp_dst = swkey->ipv6.tp.dst;
+                       }
+               } else if (swkey->eth.type == htons(ETH_P_IP) &&
+                          swkey->ip.proto == IPPROTO_ICMP) {
+                       struct odp_key_icmp *icmp_key;
+
+                       nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
+                       if (!nla)
+                               goto nla_put_failure;
+                       icmp_key = nla_data(nla);
+                       icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
+                       icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
+               } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
+                          swkey->ip.proto == IPPROTO_ICMPV6) {
+                       struct odp_key_icmpv6 *icmpv6_key;
+
+                       nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6,
+                                               sizeof(*icmpv6_key));
+                       if (!nla)
+                               goto nla_put_failure;
+                       icmpv6_key = nla_data(nla);
+                       icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
+                       icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
+
+                       if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
+                           icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
+                               struct odp_key_nd *nd_key;
+
+                               nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
+                               if (!nla)
+                                       goto nla_put_failure;
+                               nd_key = nla_data(nla);
+                               memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
+                                                       sizeof(nd_key->nd_target));
+                               memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
+                               memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
+                       }
+               }
+       }
+
+       return 0;
 
-       return !memcmp(key1, key2, sizeof(struct odp_flow_key));
+nla_put_failure:
+       return -EMSGSIZE;
 }
 
 /* Initializes the flow module.
@@ -345,7 +1086,7 @@ int flow_init(void)
        if (flow_cache == NULL)
                return -ENOMEM;
 
-       get_random_bytes(&hash_seed, sizeof hash_seed);
+       get_random_bytes(&hash_seed, sizeof(hash_seed));
 
        return 0;
 }