datapath: Add ref counting for flows.

[sliver-openvswitch.git] / datapath / flow.c

diff --git a/datapath/flow.c b/datapath/flow.c
index eae703d..dfbf769 100644 (file)
--- a/datapath/flow.c
+++ b/datapath/flow.c
@@ -36,8 +36,7 @@ static unsigned int hash_seed;
 
 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 skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
 }
 
 static inline int check_iphdr(struct sk_buff *skb)
@@ -66,7 +65,7 @@ static inline int check_iphdr(struct sk_buff *skb)
 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))) {
+       if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
                int tcp_len = tcp_hdrlen(skb);
                return (tcp_len >= sizeof(struct tcphdr)
                        && skb->len >= th_ofs + tcp_len);
@@ -76,14 +75,12 @@ static inline bool tcphdr_ok(struct sk_buff *skb)
 
 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 skb->len >= 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 skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
 }
 
 #define TCP_FLAGS_OFFSET 13
@@ -111,7 +108,10 @@ struct sw_flow_actions *flow_actions_alloc(size_t n_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 (n_actions > 2 * DP_MAX_PORTS)
                return ERR_PTR(-EINVAL);
 
        sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
@@ -123,27 +123,36 @@ struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
        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->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.
@@ -153,10 +162,26 @@ 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(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);
 }
@@ -227,6 +252,18 @@ static __be16 parse_ethertype(struct sk_buff *skb)
  *
  * 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 on output, then just
+ *      past the IPv4 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.
+ *
  * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
  * %false.
  */
@@ -302,22 +339,12 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
                                        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)) {
@@ -327,11 +354,6 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
                                         * 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;
                                }
                        }
                } else {
@@ -352,14 +374,12 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
                                key->nw_proto = ntohs(arp->ar_op);
                        }
 
-                       if (key->nw_proto == ARPOP_REQUEST 
+                       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));
                        }
                }
-       } else {
-               skb_reset_transport_header(skb);
        }
        return 0;
 }