Initial implementation of sFlow.

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

diff --git a/datapath/datapath.c b/datapath/datapath.c
index b8ef104..f6a02f7 100644 (file)
--- a/datapath/datapath.c
+++ b/datapath/datapath.c
@@ -232,7 +232,7 @@ static int create_dp(int dp_idx, const char __user *devnamep)
        if (!dp->table)
                goto err_free_dp;
 
-       /* Setup our datapath device */
+       /* Set up our datapath device. */
        dp_dev = dp_dev_create(dp, devname, ODPP_LOCAL);
        err = PTR_ERR(dp_dev);
        if (IS_ERR(dp_dev))
@@ -574,7 +574,7 @@ static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 #error
 #endif
 
-#if defined(CONFIG_XEN) && LINUX_VERSION_CODE == KERNEL_VERSION(2,6,18)
+#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
 /* This code is copied verbatim from net/dev/core.c in Xen's
  * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644.  We can't call those functions
  * directly because they aren't exported. */
@@ -715,8 +715,7 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
        int err;
 
        WARN_ON_ONCE(skb_shared(skb));
-       BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR);
-
+       BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
        queue = &dp->queues[queue_no];
        err = -ENOBUFS;
        if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
@@ -782,7 +781,8 @@ static int validate_actions(const struct sw_flow_actions *actions)
                        break;
 
                case ODPAT_SET_VLAN_PCP:
-                       if (a->vlan_pcp.vlan_pcp & ~VLAN_PCP_MASK)
+                       if (a->vlan_pcp.vlan_pcp
+                           & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
                                return -EINVAL;
                        break;
 
@@ -974,13 +974,18 @@ static int put_actions(const struct sw_flow *flow, struct odp_flow __user *ufp)
        return 0;
 }
 
-static int answer_query(struct sw_flow *flow, struct odp_flow __user *ufp)
+static int answer_query(struct sw_flow *flow, u32 query_flags,
+                       struct odp_flow __user *ufp)
 {
        struct odp_flow_stats stats;
        unsigned long int flags;
 
        spin_lock_irqsave(&flow->lock, flags);
        get_stats(flow, &stats);
+
+       if (query_flags & ODPFF_ZERO_TCP_FLAGS) {
+               flow->tcp_flags = 0;
+       }
        spin_unlock_irqrestore(&flow->lock, flags);
 
        if (__copy_to_user(&ufp->stats, &stats, sizeof(struct odp_flow_stats)))
@@ -1015,7 +1020,7 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
         * we get completely accurate stats, but that blows our performance,
         * badly. */
        dp->n_flows--;
-       error = answer_query(flow, ufp);
+       error = answer_query(flow, 0, ufp);
        flow_deferred_free(flow);
 
 error:
@@ -1040,7 +1045,7 @@ static int query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
                if (!flow)
                        error = __put_user(ENOENT, &ufp->stats.error);
                else
-                       error = answer_query(flow, ufp);
+                       error = answer_query(flow, uf.flags, ufp);
                if (error)
                        return -EFAULT;
        }
@@ -1061,7 +1066,7 @@ static int list_flow(struct sw_flow *flow, void *cbdata_)
 
        if (__copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
                return -EFAULT;
-       error = answer_query(flow, ufp);
+       error = answer_query(flow, 0, ufp);
        if (error)
                return error;
 
@@ -1161,9 +1166,9 @@ static int do_execute(struct datapath *dp, const struct odp_execute *executep)
        skb_reset_mac_header(skb);
        eth = eth_hdr(skb);
 
-    /* Normally, setting the skb 'protocol' field would be handled by a
-     * call to eth_type_trans(), but it assumes there's a sending
-     * device, which we may not have. */
+       /* Normally, setting the skb 'protocol' field would be handled by a
+        * call to eth_type_trans(), but it assumes there's a sending
+        * device, which we may not have. */
        if (ntohs(eth->h_proto) >= 1536)
                skb->protocol = eth->h_proto;
        else
@@ -1371,12 +1376,23 @@ get_port_group(struct datapath *dp, struct odp_port_group *upg)
        return 0;
 }
 
+static int get_listen_mask(const struct file *f)
+{
+       return (long)f->private_data;
+}
+
+static void set_listen_mask(struct file *f, int listen_mask)
+{
+       f->private_data = (void*)(long)listen_mask;
+}
+
 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
                           unsigned long argp)
 {
        int dp_idx = iminor(f->f_dentry->d_inode);
        struct datapath *dp;
        int drop_frags, listeners, port_no;
+       unsigned int sflow_probability;
        int err;
 
        /* Handle commands with special locking requirements up front. */
@@ -1426,7 +1442,7 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd,
                break;
 
        case ODP_GET_LISTEN_MASK:
-               err = put_user((int)f->private_data, (int __user *)argp);
+               err = put_user(get_listen_mask(f), (int __user *)argp);
                break;
 
        case ODP_SET_LISTEN_MASK:
@@ -1437,7 +1453,17 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd,
                if (listeners & ~ODPL_ALL)
                        break;
                err = 0;
-               f->private_data = (void*)listeners;
+               set_listen_mask(f, listeners);
+               break;
+
+       case ODP_GET_SFLOW_PROBABILITY:
+               err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
+               break;
+
+       case ODP_SET_SFLOW_PROBABILITY:
+               err = get_user(sflow_probability, (unsigned int __user *)argp);
+               if (!err)
+                       dp->sflow_probability = sflow_probability;
                break;
 
        case ODP_PORT_QUERY:
@@ -1503,7 +1529,7 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
                      loff_t *ppos)
 {
        /* XXX is there sufficient synchronization here? */
-       int listeners = (int) f->private_data;
+       int listeners = get_listen_mask(f);
        int dp_idx = iminor(f->f_dentry->d_inode);
        struct datapath *dp = get_dp(dp_idx);
        struct sk_buff *skb;
@@ -1543,7 +1569,7 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
                }
        }
 success:
-       copy_bytes = min(skb->len, nbytes);
+       copy_bytes = min_t(size_t, skb->len, nbytes);
        iov.iov_base = buf;
        iov.iov_len = copy_bytes;
        retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
@@ -1565,7 +1591,7 @@ static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
        if (dp) {
                mask = 0;
                poll_wait(file, &dp->waitqueue, wait);
-               if (dp_has_packet_of_interest(dp, (int)file->private_data))
+               if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
                        mask |= POLLIN | POLLRDNORM;
        } else {
                mask = POLLIN | POLLRDNORM | POLLHUP;