datapath: Optimize flow key match for non tunnel flows.

Following patch adds start offset for sw_flow-key, so that we can
skip tunneling information in key for non-tunnel flows.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Kyle Mestery <kmestery@cisco.com>
Acked-by: Ben Pfaff <blp@nicira.com>

diff --git a/datapath/datapath.c b/datapath/datapath.c
index 3f963be..65f4dc8 100644 (file)
--- a/datapath/datapath.c
+++ b/datapath/datapath.c
@@ -790,10 +790,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
        if (err)
                goto err_flow_put;
 
-       err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
-                                            &flow->key.phy.in_port,
-                                            &flow->key.tun.tun_key,
-                                            a[OVS_PACKET_ATTR_KEY]);
+       err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
        if (err)
                goto err_flow_put;
 
@@ -801,8 +798,6 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
        if (err)
                goto err_flow_put;
 
-       flow->hash = ovs_flow_hash(&flow->key, key_len);
-
        acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
        err = PTR_ERR(acts);
        if (IS_ERR(acts))
@@ -1073,7 +1068,6 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                        error = PTR_ERR(flow);
                        goto error;
                }
-               flow->key = key;
                clear_stats(flow);
 
                /* Obtain actions. */
@@ -1084,8 +1078,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
                rcu_assign_pointer(flow->sf_acts, acts);
 
                /* Put flow in bucket. */
-               flow->hash = ovs_flow_hash(&key, key_len);
-               ovs_flow_tbl_insert(table, flow);
+               ovs_flow_tbl_insert(table, flow, &key, key_len);
 
                reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
                                                info->snd_seq,

diff --git a/datapath/flow.c b/datapath/flow.c
index 42aff6d..b6efd35 100644 (file)
--- a/datapath/flow.c
+++ b/datapath/flow.c
@@ -365,6 +365,14 @@ struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *la
        return NULL;
 }
 
+static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
+{
+       struct hlist_head *head;
+       head = find_bucket(table, flow->hash);
+       hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
+       table->count++;
+}
+
 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 {
        int old_ver;
@@ -382,7 +390,7 @@ static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new
                head = flex_array_get(old->buckets, i);
 
                hlist_for_each_entry(flow, n, head, hash_node[old_ver])
-                       ovs_flow_tbl_insert(new, flow);
+                       __flow_tbl_insert(new, flow);
        }
        old->keep_flows = true;
 }
@@ -630,7 +638,7 @@ int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 
        key->phy.priority = skb->priority;
        if (OVS_CB(skb)->tun_key)
-               memcpy(&key->tun.tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun.tun_key));
+               memcpy(&key->phy.tun.tun_key, OVS_CB(skb)->tun_key, sizeof(key->phy.tun.tun_key));
        key->phy.in_port = in_port;
 
        skb_reset_mac_header(skb);
@@ -787,9 +795,18 @@ out:
        return error;
 }
 
-u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
+static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 {
-       return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
+       return jhash2((u32 *)((u8 *)key + key_start),
+                     DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
+}
+
+static int flow_key_start(struct sw_flow_key *key)
+{
+       if (key->phy.tun.tun_key.ipv4_dst)
+               return 0;
+       else
+               return offsetof(struct sw_flow_key, phy.priority);
 }
 
 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
@@ -798,28 +815,31 @@ struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
        struct sw_flow *flow;
        struct hlist_node *n;
        struct hlist_head *head;
+       u8 *_key;
+       int key_start;
        u32 hash;
 
-       hash = ovs_flow_hash(key, key_len);
+       key_start = flow_key_start(key);
+       hash = ovs_flow_hash(key, key_start, key_len);
 
+       _key = (u8 *) key + key_start;
        head = find_bucket(table, hash);
        hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 
                if (flow->hash == hash &&
-                   !memcmp(&flow->key, key, key_len)) {
+                   !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
                        return flow;
                }
        }
        return NULL;
 }
 
-void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
+void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
+                        struct sw_flow_key *key, int key_len)
 {
-       struct hlist_head *head;
-
-       head = find_bucket(table, flow->hash);
-       hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
-       table->count++;
+       flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
+       memcpy(&flow->key, key, sizeof(flow->key));
+       __flow_tbl_insert(table, flow);
 }
 
 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
@@ -1040,12 +1060,12 @@ int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
                if (tun_id != tun_key->tun_id)
                        return -EINVAL;
 
-               memcpy(&swkey->tun.tun_key, tun_key, sizeof(swkey->tun.tun_key));
+               memcpy(&swkey->phy.tun.tun_key, tun_key, sizeof(swkey->phy.tun.tun_key));
                attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
                attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
        } else if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
-               swkey->tun.tun_key.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
-               swkey->tun.tun_key.tun_flags |= OVS_FLOW_TNL_F_KEY;
+               swkey->phy.tun.tun_key.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
+               swkey->phy.tun.tun_key.tun_flags |= OVS_FLOW_TNL_F_KEY;
 
                attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
        } else if (attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
@@ -1055,7 +1075,7 @@ int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
                if (!tun_key->ipv4_dst)
                        return -EINVAL;
 
-               memcpy(&swkey->tun.tun_key, tun_key, sizeof(swkey->tun.tun_key));
+               memcpy(&swkey->phy.tun.tun_key, tun_key, sizeof(swkey->phy.tun.tun_key));
                attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
        }
 
@@ -1194,17 +1214,17 @@ int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
  * get the metadata, that is, the parts of the flow key that cannot be
  * extracted from the packet itself.
  */
-int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
-                                  struct ovs_key_ipv4_tunnel *tun_key,
-                                  const struct nlattr *attr)
+
+int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
 {
+       struct ovs_key_ipv4_tunnel *tun_key = &flow->key.phy.tun.tun_key;
        const struct nlattr *nla;
        int rem;
-       __be64 tun_id;
+       __be64 tun_id = 0;
 
-       *in_port = DP_MAX_PORTS;
-       memset(tun_key, 0, sizeof(*tun_key));
-       *priority = 0;
+       flow->key.phy.in_port = DP_MAX_PORTS;
+       flow->key.phy.priority = 0;
+       memset(tun_key, 0, sizeof(flow->key.phy.tun.tun_key));
 
        nla_for_each_nested(nla, attr, rem) {
                int type = nla_type(nla);
@@ -1215,7 +1235,7 @@ int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
 
                        switch (type) {
                        case OVS_KEY_ATTR_PRIORITY:
-                               *priority = nla_get_u32(nla);
+                               flow->key.phy.priority = nla_get_u32(nla);
                                break;
 
                        case OVS_KEY_ATTR_TUN_ID:
@@ -1253,13 +1273,17 @@ int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
                        case OVS_KEY_ATTR_IN_PORT:
                                if (nla_get_u32(nla) >= DP_MAX_PORTS)
                                        return -EINVAL;
-                               *in_port = nla_get_u32(nla);
+                               flow->key.phy.in_port = nla_get_u32(nla);
                                break;
                        }
                }
        }
        if (rem)
                return -EINVAL;
+
+       flow->hash = ovs_flow_hash(&flow->key,
+                                  flow_key_start(&flow->key), key_len);
+
        return 0;
 }
 
@@ -1272,16 +1296,16 @@ int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
            nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
                goto nla_put_failure;
 
-       if (swkey->tun.tun_key.ipv4_dst) {
+       if (swkey->phy.tun.tun_key.ipv4_dst) {
                struct ovs_key_ipv4_tunnel *tun_key;
                nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4_TUNNEL, sizeof(*tun_key));
                if (!nla)
                        goto nla_put_failure;
                tun_key = nla_data(nla);
-               memcpy(tun_key, &swkey->tun.tun_key, sizeof(*tun_key));
+               memcpy(tun_key, &swkey->phy.tun.tun_key, sizeof(*tun_key));
        }
-       if ((swkey->tun.tun_key.tun_flags & OVS_FLOW_TNL_F_KEY) &&
-           nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->tun.tun_key.tun_id))
+       if ((swkey->phy.tun.tun_key.tun_flags & OVS_FLOW_TNL_F_KEY) &&
+           nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun.tun_key.tun_id))
                goto nla_put_failure;
 
        if (swkey->phy.in_port != DP_MAX_PORTS &&

diff --git a/datapath/flow.h b/datapath/flow.h
index c52e029..f4ef285 100644 (file)
--- a/datapath/flow.h
+++ b/datapath/flow.h
@@ -42,12 +42,12 @@ struct sw_flow_actions {
 
 struct sw_flow_key {
        struct {
+               union {
+                       struct ovs_key_ipv4_tunnel tun_key;  /* Encapsulating tunnel key. */
+               } tun;
                u32     priority;       /* Packet QoS priority. */
                u16     in_port;        /* Input switch port (or DP_MAX_PORTS). */
        } phy;
-       struct {
-               struct ovs_key_ipv4_tunnel tun_key;  /* Encapsulating tunnel key. */
-       } tun;
        struct {
                u8     src[ETH_ALEN];   /* Ethernet source address. */
                u8     dst[ETH_ALEN];   /* Ethernet destination address. */
@@ -170,9 +170,8 @@ u64 ovs_flow_used_time(unsigned long flow_jiffies);
 int ovs_flow_to_nlattrs(const struct sw_flow_key *, struct sk_buff *);
 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
                      const struct nlattr *);
-int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
-                                  struct ovs_key_ipv4_tunnel *tun_key,
-                                  const struct nlattr *);
+int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len,
+                                  const struct nlattr *attr);
 
 #define MAX_ACTIONS_BUFSIZE    (16 * 1024)
 #define TBL_MIN_BUCKETS                1024
@@ -203,9 +202,9 @@ void ovs_flow_tbl_deferred_destroy(struct flow_table *table);
 struct flow_table *ovs_flow_tbl_alloc(int new_size);
 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table);
 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table);
-void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow);
+void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
+                        struct sw_flow_key *key, int key_len);
 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow);
-u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len);
 
 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *idx);
 extern const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1];