Initial implementation of sFlow.
[sliver-openvswitch.git] / datapath / actions.c
index 9b82f9e..7c618cc 100644 (file)
 #include "actions.h"
 #include "openvswitch/datapath-protocol.h"
 
-struct sk_buff *
-make_writable(struct sk_buff *skb, gfp_t gfp)
+static struct sk_buff *
+make_writable(struct sk_buff *skb, unsigned min_headroom, gfp_t gfp)
 {
        if (skb_shared(skb) || skb_cloned(skb)) {
-               struct sk_buff *nskb = skb_copy(skb, gfp);
+               struct sk_buff *nskb;
+               unsigned headroom = max(min_headroom, skb_headroom(skb));
+
+               nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), gfp);
                if (nskb) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
+                       /* Before 2.6.24 these fields were not copied when
+                        * doing an skb_copy_expand. */
+                       nskb->ip_summed = skb->ip_summed;
+                       nskb->csum = skb->csum;
+#endif
+#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
+                       /* These fields are copied in skb_clone but not in
+                        * skb_copy or related functions.  We need to manually
+                        * copy them over here. */
+                       nskb->proto_data_valid = skb->proto_data_valid;
+                       nskb->proto_csum_blank = skb->proto_csum_blank;
+#endif
                        kfree_skb(skb);
                        return nskb;
                }
@@ -80,7 +96,7 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
                mask = VLAN_PCP_MASK;
        }
 
-       skb = make_writable(skb, gfp);
+       skb = make_writable(skb, VLAN_HLEN, gfp);
        if (!skb)
                return ERR_PTR(-ENOMEM);
 
@@ -96,7 +112,7 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
                 * when we send the packet out on the wire, and it will fail at
                 * that point because skb_checksum_setup() will not look inside
                 * an 802.1Q header. */
-               skb_checksum_setup(skb);
+               vswitch_skb_checksum_setup(skb);
 
                /* GSO is not implemented for packets with an 802.1Q header, so
                 * we have to do segmentation before we add that header.
@@ -168,7 +184,7 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
 static struct sk_buff *strip_vlan(struct sk_buff *skb,
                                  struct odp_flow_key *key, gfp_t gfp)
 {
-       skb = make_writable(skb, gfp);
+       skb = make_writable(skb, 0, gfp);
        if (skb) {
                vlan_pull_tag(skb);
                key->dl_vlan = htons(ODP_VLAN_NONE);
@@ -180,7 +196,7 @@ static struct sk_buff *set_dl_addr(struct sk_buff *skb,
                                   const struct odp_action_dl_addr *a,
                                   gfp_t gfp)
 {
-       skb = make_writable(skb, gfp);
+       skb = make_writable(skb, 0, gfp);
        if (skb) {
                struct ethhdr *eh = eth_hdr(skb);
                memcpy(a->type == ODPAT_SET_DL_SRC ? eh->h_source : eh->h_dest,
@@ -216,7 +232,7 @@ static struct sk_buff *set_nw_addr(struct sk_buff *skb,
        if (key->dl_type != htons(ETH_P_IP))
                return skb;
 
-       skb = make_writable(skb, gfp);
+       skb = make_writable(skb, 0, gfp);
        if (skb) {
                struct iphdr *nh = ip_hdr(skb);
                u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
@@ -253,14 +269,14 @@ set_tp_port(struct sk_buff *skb, struct odp_flow_key *key,
        else
                return skb;
 
-       skb = make_writable(skb, gfp);
+       skb = make_writable(skb, 0, gfp);
        if (skb) {
                struct udphdr *th = udp_hdr(skb);
                u16 *f = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
                u16 old = *f;
                u16 new = a->tp_port;
-               update_csum((u16*)((u8*)skb->data + check_ofs),
-                           skb, old, new, 1);
+               update_csum((u16*)(skb_transport_header(skb) + check_ofs), 
+                               skb, old, new, 1);
                *f = new;
        }
        return skb;
@@ -307,7 +323,7 @@ do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
        dev = skb->dev = p->dev;
        if (is_dp_dev(dev))
                dp_dev_recv(dev, skb);
-        else
+       else
                dp_xmit_skb(skb);
        return;
 
@@ -350,6 +366,33 @@ output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, gfp_t gfp)
        return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg);
 }
 
+/* Send a copy of this packet up to the sFlow agent, along with extra
+ * information about what happened to it. */
+static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
+                        const union odp_action *a, int n_actions, gfp_t gfp)
+{
+       struct odp_sflow_sample_header *hdr;
+       unsigned int actlen = n_actions * sizeof(union odp_action);
+       unsigned int hdrlen = sizeof(struct odp_sflow_sample_header);
+       struct sk_buff *nskb;
+       int i;
+
+       nskb = skb_copy_expand(skb, actlen + hdrlen, 0, gfp);
+       if (!nskb)
+               return;
+
+       memcpy(__skb_push(nskb, actlen), a, actlen);
+       hdr = (struct odp_sflow_sample_header*)__skb_push(nskb, hdrlen);
+       hdr->n_actions = n_actions;
+       hdr->sample_pool = 0;
+       for_each_possible_cpu (i) {
+               const struct dp_stats_percpu *stats;
+               stats = per_cpu_ptr(dp->stats_percpu, i);
+               hdr->sample_pool += stats->sflow_pool;
+       }
+       dp_output_control(dp, nskb, _ODPL_SFLOW_NR, 0);
+}
+
 /* Execute a list of actions against 'skb'. */
 int execute_actions(struct datapath *dp, struct sk_buff *skb,
                    struct odp_flow_key *key,
@@ -362,6 +405,19 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb,
         * is slightly obscure just to avoid that. */
        int prev_port = -1;
        int err;
+
+       if (dp->sflow_probability) {
+               /* Increment sample pool. */
+               int cpu = get_cpu();
+               per_cpu_ptr(dp->stats_percpu, cpu)->sflow_pool++;
+               put_cpu();
+
+               /* Sample packet. */
+               if (dp->sflow_probability == UINT_MAX ||
+                   net_random() < dp->sflow_probability)
+                       sflow_sample(dp, skb, a, n_actions, gfp);
+       }
+
        for (; n_actions > 0; a++, n_actions--) {
                WARN_ON_ONCE(skb_shared(skb));
                if (prev_port != -1) {