openvswitch: Eliminate memset() from flow_extract.
authorJesse Gross <jesse@nicira.com>
Fri, 3 Oct 2014 22:35:29 +0000 (15:35 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 6 Oct 2014 04:32:20 +0000 (00:32 -0400)
As new protocols are added, the size of the flow key tends to
increase although few protocols care about all of the fields. In
order to optimize this for hashing and matching, OVS uses a variable
length portion of the key. However, when fields are extracted from
the packet we must still zero out the entire key.

This is no longer necessary now that OVS implements masking. Any
fields (or holes in the structure) which are not part of a given
protocol will be by definition not part of the mask and zeroed out
during lookup. Furthermore, since masking already uses variable
length keys this zeroing operation automatically benefits as well.

In principle, the only thing that needs to be done at this point
is remove the memset() at the beginning of flow. However, some
fields assume that they are initialized to zero, which now must be
done explicitly. In addition, in the event of an error we must also
zero out corresponding fields to signal that there is no valid data
present. These increase the total amount of code but very little of
it is executed in non-error situations.

Removing the memset() reduces the profile of ovs_flow_extract()
from 0.64% to 0.56% when tested with large packets on a 10G link.

Suggested-by: Pravin Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: Andy Zhou <azhou@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/openvswitch/flow.c

index 4010423f2831d4300874150934f2030395afc042..913bdc1a83c06d283089037570179ed0380e516f 100644 (file)
@@ -462,6 +462,7 @@ static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
         * update skb->csum here.
         */
 
+       key->eth.tci = 0;
        if (vlan_tx_tag_present(skb))
                key->eth.tci = htons(skb->vlan_tci);
        else if (eth->h_proto == htons(ETH_P_8021Q))
@@ -482,6 +483,8 @@ static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 
                error = check_iphdr(skb);
                if (unlikely(error)) {
+                       memset(&key->ip, 0, sizeof(key->ip));
+                       memset(&key->ipv4, 0, sizeof(key->ipv4));
                        if (error == -EINVAL) {
                                skb->transport_header = skb->network_header;
                                error = 0;
@@ -503,8 +506,10 @@ static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
                        return 0;
                }
                if (nh->frag_off & htons(IP_MF) ||
-                        skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
+                       skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
                        key->ip.frag = OVS_FRAG_TYPE_FIRST;
+               else
+                       key->ip.frag = OVS_FRAG_TYPE_NONE;
 
                /* Transport layer. */
                if (key->ip.proto == IPPROTO_TCP) {
@@ -513,18 +518,25 @@ static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
                                key->tp.src = tcp->source;
                                key->tp.dst = tcp->dest;
                                key->tp.flags = TCP_FLAGS_BE16(tcp);
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
+
                } else if (key->ip.proto == IPPROTO_UDP) {
                        if (udphdr_ok(skb)) {
                                struct udphdr *udp = udp_hdr(skb);
                                key->tp.src = udp->source;
                                key->tp.dst = udp->dest;
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                } else if (key->ip.proto == IPPROTO_SCTP) {
                        if (sctphdr_ok(skb)) {
                                struct sctphdr *sctp = sctp_hdr(skb);
                                key->tp.src = sctp->source;
                                key->tp.dst = sctp->dest;
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                } else if (key->ip.proto == IPPROTO_ICMP) {
                        if (icmphdr_ok(skb)) {
@@ -534,33 +546,44 @@ static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
                                 * them in 16-bit network byte order. */
                                key->tp.src = htons(icmp->type);
                                key->tp.dst = htons(icmp->code);
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                }
 
-       } else if ((key->eth.type == htons(ETH_P_ARP) ||
-                  key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
+       } else if (key->eth.type == htons(ETH_P_ARP) ||
+                  key->eth.type == htons(ETH_P_RARP)) {
                struct arp_eth_header *arp;
 
                arp = (struct arp_eth_header *)skb_network_header(skb);
 
-               if (arp->ar_hrd == htons(ARPHRD_ETHER)
-                               && arp->ar_pro == htons(ETH_P_IP)
-                               && arp->ar_hln == ETH_ALEN
-                               && arp->ar_pln == 4) {
+               if (arphdr_ok(skb) &&
+                   arp->ar_hrd == htons(ARPHRD_ETHER) &&
+                   arp->ar_pro == htons(ETH_P_IP) &&
+                   arp->ar_hln == ETH_ALEN &&
+                   arp->ar_pln == 4) {
 
                        /* We only match on the lower 8 bits of the opcode. */
                        if (ntohs(arp->ar_op) <= 0xff)
                                key->ip.proto = ntohs(arp->ar_op);
+                       else
+                               key->ip.proto = 0;
+
                        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));
                        ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
                        ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
+               } else {
+                       memset(&key->ip, 0, sizeof(key->ip));
+                       memset(&key->ipv4, 0, sizeof(key->ipv4));
                }
        } else if (key->eth.type == htons(ETH_P_IPV6)) {
                int nh_len;             /* IPv6 Header + Extensions */
 
                nh_len = parse_ipv6hdr(skb, key);
                if (unlikely(nh_len < 0)) {
+                       memset(&key->ip, 0, sizeof(key->ip));
+                       memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
                        if (nh_len == -EINVAL) {
                                skb->transport_header = skb->network_header;
                                error = 0;
@@ -582,24 +605,32 @@ static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
                                key->tp.src = tcp->source;
                                key->tp.dst = tcp->dest;
                                key->tp.flags = TCP_FLAGS_BE16(tcp);
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                } else if (key->ip.proto == NEXTHDR_UDP) {
                        if (udphdr_ok(skb)) {
                                struct udphdr *udp = udp_hdr(skb);
                                key->tp.src = udp->source;
                                key->tp.dst = udp->dest;
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                } else if (key->ip.proto == NEXTHDR_SCTP) {
                        if (sctphdr_ok(skb)) {
                                struct sctphdr *sctp = sctp_hdr(skb);
                                key->tp.src = sctp->source;
                                key->tp.dst = sctp->dest;
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                } else if (key->ip.proto == NEXTHDR_ICMP) {
                        if (icmp6hdr_ok(skb)) {
                                error = parse_icmpv6(skb, key, nh_len);
                                if (error)
                                        return error;
+                       } else {
+                               memset(&key->tp, 0, sizeof(key->tp));
                        }
                }
        }
@@ -615,13 +646,19 @@ int ovs_flow_key_extract(struct ovs_key_ipv4_tunnel *tun_key,
                         struct sk_buff *skb, struct sw_flow_key *key)
 {
        /* Extract metadata from packet. */
-       memset(key, 0, sizeof(*key));
        if (tun_key)
                memcpy(&key->tun_key, tun_key, sizeof(key->tun_key));
+       else
+               memset(&key->tun_key, 0, sizeof(key->tun_key));
 
        key->phy.priority = skb->priority;
        key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
        key->phy.skb_mark = skb->mark;
+       key->ovs_flow_hash = 0;
+       key->recirc_id = 0;
+
+       /* Flags are always used as part of stats */
+       key->tp.flags = 0;
 
        return key_extract(skb, key);
 }
@@ -632,7 +669,6 @@ int ovs_flow_key_extract_userspace(const struct nlattr *attr,
 {
        int err;
 
-       memset(key, 0, sizeof(*key));
        /* Extract metadata from netlink attributes. */
        err = ovs_nla_get_flow_metadata(attr, key);
        if (err)