datapath: add skb mark matching and set action
[openvswitch] / datapath / actions.c
index ec9b595c281e86edc69c7cff8ae57d46af60a126..faa6a0023df87d67adab6e09afff9eae7ed54d60 100644 (file)
@@ -28,6 +28,7 @@
 #include <linux/if_arp.h>
 #include <linux/if_vlan.h>
 #include <net/ip.h>
+#include <net/ipv6.h>
 #include <net/checksum.h>
 #include <net/dsfield.h>
 
@@ -37,7 +38,8 @@
 #include "vport.h"
 
 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
-                       const struct nlattr *attr, int len, bool keep_skb);
+                             const struct nlattr *attr, int len,
+                             struct ovs_key_ipv4_tunnel *tun_key, bool keep_skb);
 
 static int make_writable(struct sk_buff *skb, int write_len)
 {
@@ -165,6 +167,54 @@ static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
        *addr = new_addr;
 }
 
+static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
+                                __be32 addr[4], const __be32 new_addr[4])
+{
+       int transport_len = skb->len - skb_transport_offset(skb);
+
+       if (l4_proto == IPPROTO_TCP) {
+               if (likely(transport_len >= sizeof(struct tcphdr)))
+                       inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
+                                                 addr, new_addr, 1);
+       } else if (l4_proto == IPPROTO_UDP) {
+               if (likely(transport_len >= sizeof(struct udphdr))) {
+                       struct udphdr *uh = udp_hdr(skb);
+
+                       if (uh->check ||
+                           get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
+                               inet_proto_csum_replace16(&uh->check, skb,
+                                                         addr, new_addr, 1);
+                               if (!uh->check)
+                                       uh->check = CSUM_MANGLED_0;
+                       }
+               }
+       }
+}
+
+static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
+                         __be32 addr[4], const __be32 new_addr[4],
+                         bool recalculate_csum)
+{
+       if (recalculate_csum)
+               update_ipv6_checksum(skb, l4_proto, addr, new_addr);
+
+       skb_clear_rxhash(skb);
+       memcpy(addr, new_addr, sizeof(__be32[4]));
+}
+
+static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
+{
+       nh->priority = tc >> 4;
+       nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
+}
+
+static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
+{
+       nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
+       nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
+       nh->flow_lbl[2] = fl & 0x000000FF;
+}
+
 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
 {
        csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
@@ -198,6 +248,47 @@ static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
        return 0;
 }
 
+static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
+{
+       struct ipv6hdr *nh;
+       int err;
+       __be32 *saddr;
+       __be32 *daddr;
+
+       err = make_writable(skb, skb_network_offset(skb) +
+                           sizeof(struct ipv6hdr));
+       if (unlikely(err))
+               return err;
+
+       nh = ipv6_hdr(skb);
+       saddr = (__be32 *)&nh->saddr;
+       daddr = (__be32 *)&nh->daddr;
+
+       if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
+               set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
+                             ipv6_key->ipv6_src, true);
+
+       if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
+               unsigned int offset = 0;
+               int flags = OVS_IP6T_FH_F_SKIP_RH;
+               bool recalc_csum = true;
+
+               if (ipv6_ext_hdr(nh->nexthdr))
+                       recalc_csum = ipv6_find_hdr(skb, &offset,
+                                                   NEXTHDR_ROUTING, NULL,
+                                                   &flags) != NEXTHDR_ROUTING;
+
+               set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
+                             ipv6_key->ipv6_dst, recalc_csum);
+       }
+
+       set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
+       set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
+       nh->hop_limit = ipv6_key->ipv6_hlimit;
+
+       return 0;
+}
+
 /* Must follow make_writable() since that can move the skb data. */
 static void set_tp_port(struct sk_buff *skb, __be16 *port,
                         __be16 new_port, __sum16 *check)
@@ -308,7 +399,8 @@ static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 }
 
 static int sample(struct datapath *dp, struct sk_buff *skb,
-                 const struct nlattr *attr)
+                 const struct nlattr *attr,
+                 struct ovs_key_ipv4_tunnel *tun_key)
 {
        const struct nlattr *acts_list = NULL;
        const struct nlattr *a;
@@ -329,11 +421,12 @@ static int sample(struct datapath *dp, struct sk_buff *skb,
        }
 
        return do_execute_actions(dp, skb, nla_data(acts_list),
-                                                nla_len(acts_list), true);
+                                 nla_len(acts_list), tun_key, true);
 }
 
 static int execute_set_action(struct sk_buff *skb,
-                                const struct nlattr *nested_attr)
+                                const struct nlattr *nested_attr,
+                                struct ovs_key_ipv4_tunnel *tun_key)
 {
        int err = 0;
 
@@ -342,8 +435,28 @@ static int execute_set_action(struct sk_buff *skb,
                skb->priority = nla_get_u32(nested_attr);
                break;
 
+       case OVS_KEY_ATTR_SKB_MARK:
+               skb_set_mark(skb, nla_get_u32(nested_attr));
+               break;
+
        case OVS_KEY_ATTR_TUN_ID:
-               OVS_CB(skb)->tun_id = nla_get_be64(nested_attr);
+               /* If we're only using the TUN_ID action, store the value in a
+                * temporary instance of struct ovs_key_ipv4_tunnel on the stack.
+                * If both IPV4_TUNNEL and TUN_ID are being used together we
+                * can't write into the IPV4_TUNNEL action, so make a copy and
+                * write into that version.
+                */
+               if (!OVS_CB(skb)->tun_key)
+                       memset(tun_key, 0, sizeof(*tun_key));
+               else if (OVS_CB(skb)->tun_key != tun_key)
+                       memcpy(tun_key, OVS_CB(skb)->tun_key, sizeof(*tun_key));
+               OVS_CB(skb)->tun_key = tun_key;
+
+               OVS_CB(skb)->tun_key->tun_id = nla_get_be64(nested_attr);
+               break;
+
+       case OVS_KEY_ATTR_IPV4_TUNNEL:
+               OVS_CB(skb)->tun_key = nla_data(nested_attr);
                break;
 
        case OVS_KEY_ATTR_ETHERNET:
@@ -354,6 +467,10 @@ static int execute_set_action(struct sk_buff *skb,
                err = set_ipv4(skb, nla_data(nested_attr));
                break;
 
+       case OVS_KEY_ATTR_IPV6:
+               err = set_ipv6(skb, nla_data(nested_attr));
+               break;
+
        case OVS_KEY_ATTR_TCP:
                err = set_tcp(skb, nla_data(nested_attr));
                break;
@@ -368,7 +485,8 @@ static int execute_set_action(struct sk_buff *skb,
 
 /* Execute a list of actions against 'skb'. */
 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
-                       const struct nlattr *attr, int len, bool keep_skb)
+                       const struct nlattr *attr, int len,
+                       struct ovs_key_ipv4_tunnel *tun_key, bool keep_skb)
 {
        /* Every output action needs a separate clone of 'skb', but the common
         * case is just a single output action, so that doing a clone and
@@ -407,11 +525,11 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
                        break;
 
                case OVS_ACTION_ATTR_SET:
-                       err = execute_set_action(skb, nla_data(a));
+                       err = execute_set_action(skb, nla_data(a), tun_key);
                        break;
 
                case OVS_ACTION_ATTR_SAMPLE:
-                       err = sample(dp, skb, a);
+                       err = sample(dp, skb, a, tun_key);
                        break;
                }
 
@@ -458,6 +576,7 @@ int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
        struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
        struct loop_counter *loop;
        int error;
+       struct ovs_key_ipv4_tunnel tun_key;
 
        /* Check whether we've looped too much. */
        loop = &__get_cpu_var(loop_counters);
@@ -469,9 +588,9 @@ int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
                goto out_loop;
        }
 
-       OVS_CB(skb)->tun_id = 0;
+       OVS_CB(skb)->tun_key = NULL;
        error = do_execute_actions(dp, skb, acts->actions,
-                                        acts->actions_len, false);
+                                        acts->actions_len, &tun_key, false);
 
        /* Check whether sub-actions looped too much. */
        if (unlikely(loop->looping))