netdev-dpdk: fix mbuf leaks
[cascardo/ovs.git] / datapath / vport-vxlan.c
index 34dd919..c05f5d4 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2013 Nicira, Inc.
+ * Copyright (c) 2015 Nicira, Inc.
  * Copyright (c) 2013 Cisco Systems, Inc.
  *
  * This program is free software; you can redistribute it and/or
  * 02110-1301, USA
  */
 
-#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
-
-#include <linux/version.h>
-
-#include <linux/in.h>
-#include <linux/ip.h>
-#include <linux/net.h>
-#include <linux/rculist.h>
-#include <linux/udp.h>
-
-#include <net/icmp.h>
-#include <net/ip.h>
+#include <linux/kernel.h>
+#include <linux/skbuff.h>
+#include <linux/openvswitch.h>
+#include <linux/module.h>
 #include <net/udp.h>
 #include <net/ip_tunnels.h>
 #include <net/rtnetlink.h>
-#include <net/route.h>
-#include <net/dsfield.h>
-#include <net/inet_ecn.h>
-#include <net/net_namespace.h>
-#include <net/netns/generic.h>
 #include <net/vxlan.h>
 
 #include "datapath.h"
 #include "vport.h"
+#include "vport-netdev.h"
 
-/**
- * struct vxlan_port - Keeps track of open UDP ports
- * @vs: vxlan_sock created for the port.
- * @name: vport name.
- */
-struct vxlan_port {
-       struct vxlan_sock *vs;
-       char name[IFNAMSIZ];
-};
-
-static inline struct vxlan_port *vxlan_vport(const struct vport *vport)
-{
-       return vport_priv(vport);
-}
-
-/* Called with rcu_read_lock and BH disabled. */
-static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb, __be32 vx_vni)
-{
-       struct ovs_tunnel_info tun_info;
-       struct vport *vport = vs->data;
-       struct iphdr *iph;
-       __be64 key;
-
-       /* Save outer tunnel values */
-       iph = ip_hdr(skb);
-       key = cpu_to_be64(ntohl(vx_vni) >> 8);
-       ovs_flow_tun_info_init(&tun_info, iph,
-                              udp_hdr(skb)->source, udp_hdr(skb)->dest,
-                              key, TUNNEL_KEY, NULL, 0);
-
-       ovs_vport_receive(vport, skb, &tun_info);
-}
+static struct vport_ops ovs_vxlan_netdev_vport_ops;
 
 static int vxlan_get_options(const struct vport *vport, struct sk_buff *skb)
 {
-       struct vxlan_port *vxlan_port = vxlan_vport(vport);
-       __be16 dst_port = inet_sport(vxlan_port->vs->sock->sk);
+       struct vxlan_dev *vxlan = netdev_priv(vport->dev);
+       __be16 dst_port = vxlan->cfg.dst_port;
 
        if (nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT, ntohs(dst_port)))
                return -EMSGSIZE;
+
+       if (vxlan->flags & VXLAN_F_GBP) {
+               struct nlattr *exts;
+
+               exts = nla_nest_start(skb, OVS_TUNNEL_ATTR_EXTENSION);
+               if (!exts)
+                       return -EMSGSIZE;
+
+               if (vxlan->flags & VXLAN_F_GBP &&
+                   nla_put_flag(skb, OVS_VXLAN_EXT_GBP))
+                       return -EMSGSIZE;
+
+               nla_nest_end(skb, exts);
+       }
+
        return 0;
 }
 
-static void vxlan_tnl_destroy(struct vport *vport)
+static const struct nla_policy exts_policy[OVS_VXLAN_EXT_MAX + 1] = {
+       [OVS_VXLAN_EXT_GBP]     = { .type = NLA_FLAG, },
+};
+
+static int vxlan_configure_exts(struct vport *vport, struct nlattr *attr,
+                               struct vxlan_config *conf)
 {
-       struct vxlan_port *vxlan_port = vxlan_vport(vport);
+       struct nlattr *exts[OVS_VXLAN_EXT_MAX + 1];
+       int err;
+
+       if (nla_len(attr) < sizeof(struct nlattr))
+               return -EINVAL;
+
+       err = nla_parse_nested(exts, OVS_VXLAN_EXT_MAX, attr, exts_policy);
+       if (err < 0)
+               return err;
 
-       vxlan_sock_release(vxlan_port->vs);
+       if (exts[OVS_VXLAN_EXT_GBP])
+               conf->flags |= VXLAN_F_GBP;
 
-       ovs_vport_deferred_free(vport);
+       return 0;
 }
 
 static struct vport *vxlan_tnl_create(const struct vport_parms *parms)
 {
        struct net *net = ovs_dp_get_net(parms->dp);
        struct nlattr *options = parms->options;
-       struct vxlan_port *vxlan_port;
-       struct vxlan_sock *vs;
+       struct net_device *dev;
        struct vport *vport;
        struct nlattr *a;
-       u16 dst_port;
        int err;
+       struct vxlan_config conf = {
+               .no_share = true,
+               .flags = VXLAN_F_COLLECT_METADATA,
+               /* Don't restrict the packets that can be sent by MTU */
+               .mtu = IP_MAX_MTU,
+       };
 
        if (!options) {
                err = -EINVAL;
                goto error;
        }
+
        a = nla_find_nested(options, OVS_TUNNEL_ATTR_DST_PORT);
        if (a && nla_len(a) == sizeof(u16)) {
-               dst_port = nla_get_u16(a);
+               conf.dst_port = htons(nla_get_u16(a));
        } else {
                /* Require destination port from userspace. */
                err = -EINVAL;
                goto error;
        }
 
-       vport = ovs_vport_alloc(sizeof(struct vxlan_port),
-                               &ovs_vxlan_vport_ops, parms);
+       vport = ovs_vport_alloc(0, &ovs_vxlan_netdev_vport_ops, parms);
        if (IS_ERR(vport))
                return vport;
 
-       vxlan_port = vxlan_vport(vport);
-       strncpy(vxlan_port->name, parms->name, IFNAMSIZ);
+       a = nla_find_nested(options, OVS_TUNNEL_ATTR_EXTENSION);
+       if (a) {
+               err = vxlan_configure_exts(vport, a, &conf);
+               if (err) {
+                       ovs_vport_free(vport);
+                       goto error;
+               }
+       }
 
-       vs = vxlan_sock_add(net, htons(dst_port), vxlan_rcv, vport, true, false);
-       if (IS_ERR(vs)) {
+       rtnl_lock();
+       dev = vxlan_dev_create(net, parms->name, NET_NAME_USER, &conf);
+       if (IS_ERR(dev)) {
+               rtnl_unlock();
                ovs_vport_free(vport);
-               return (void *)vs;
+               return ERR_CAST(dev);
        }
-       vxlan_port->vs = vs;
 
+       dev_change_flags(dev, dev->flags | IFF_UP);
+       rtnl_unlock();
        return vport;
-
 error:
        return ERR_PTR(err);
 }
 
-static int vxlan_tnl_send(struct vport *vport, struct sk_buff *skb)
+static struct vport *vxlan_create(const struct vport_parms *parms)
 {
-       struct ovs_key_ipv4_tunnel *tun_key;
-       struct net *net = ovs_dp_get_net(vport->dp);
-       struct vxlan_port *vxlan_port = vxlan_vport(vport);
-       __be16 dst_port = inet_sport(vxlan_port->vs->sock->sk);
-       struct rtable *rt;
-       __be16 src_port;
-       __be32 saddr;
-       __be16 df;
-       int port_min;
-       int port_max;
-       int err;
-
-       if (unlikely(!OVS_CB(skb)->egress_tun_info)) {
-               err = -EINVAL;
-               goto error;
-       }
-
-       tun_key = &OVS_CB(skb)->egress_tun_info->tunnel;
-
-       /* Route lookup */
-       saddr = tun_key->ipv4_src;
-       rt = find_route(ovs_dp_get_net(vport->dp),
-                       &saddr, tun_key->ipv4_dst,
-                       IPPROTO_UDP, tun_key->ipv4_tos,
-                       skb->mark);
-       if (IS_ERR(rt)) {
-               err = PTR_ERR(rt);
-               goto error;
-       }
+       struct vport *vport;
 
-       df = tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
-       skb->local_df = 1;
+       vport = vxlan_tnl_create(parms);
+       if (IS_ERR(vport))
+               return vport;
 
-       inet_get_local_port_range(net, &port_min, &port_max);
-       src_port = vxlan_src_port(port_min, port_max, skb);
-
-       err = vxlan_xmit_skb(vxlan_port->vs, rt, skb,
-                            saddr, tun_key->ipv4_dst,
-                            tun_key->ipv4_tos,
-                            tun_key->ipv4_ttl, df,
-                            src_port, dst_port,
-                            htonl(be64_to_cpu(tun_key->tun_id) << 8));
-       if (err < 0)
-               ip_rt_put(rt);
-error:
-       return err;
+       return ovs_netdev_link(vport, parms->name);
 }
 
 static int vxlan_get_egress_tun_info(struct vport *vport, struct sk_buff *skb,
-                                    struct ovs_tunnel_info *egress_tun_info)
+                                    struct dp_upcall_info *upcall)
 {
+       struct vxlan_dev *vxlan = netdev_priv(vport->dev);
        struct net *net = ovs_dp_get_net(vport->dp);
-       struct vxlan_port *vxlan_port = vxlan_vport(vport);
-       __be16 dst_port = inet_sport(vxlan_port->vs->sock->sk);
+       __be16 dst_port = vxlan_dev_dst_port(vxlan);
        __be16 src_port;
        int port_min;
        int port_max;
 
        inet_get_local_port_range(net, &port_min, &port_max);
-       src_port = vxlan_src_port(port_min, port_max, skb);
+       src_port = udp_flow_src_port(net, skb, 0, 0, true);
 
-       return ovs_tunnel_get_egress_info(egress_tun_info, net,
-                                         OVS_CB(skb)->egress_tun_info,
-                                         IPPROTO_UDP, skb->mark,
+       return ovs_tunnel_get_egress_info(upcall, net,
+                                         skb, IPPROTO_UDP,
                                          src_port, dst_port);
 }
 
-static const char *vxlan_get_name(const struct vport *vport)
-{
-       struct vxlan_port *vxlan_port = vxlan_vport(vport);
-       return vxlan_port->name;
-}
-
-const struct vport_ops ovs_vxlan_vport_ops = {
+static struct vport_ops ovs_vxlan_netdev_vport_ops = {
        .type                   = OVS_VPORT_TYPE_VXLAN,
-       .create                 = vxlan_tnl_create,
-       .destroy                = vxlan_tnl_destroy,
-       .get_name               = vxlan_get_name,
+       .create                 = vxlan_create,
+       .destroy                = ovs_netdev_tunnel_destroy,
        .get_options            = vxlan_get_options,
-       .send                   = vxlan_tnl_send,
+       .send                   = vxlan_xmit,
        .get_egress_tun_info    = vxlan_get_egress_tun_info,
 };
+
+static int __init ovs_vxlan_tnl_init(void)
+{
+       return ovs_vport_ops_register(&ovs_vxlan_netdev_vport_ops);
+}
+
+static void __exit ovs_vxlan_tnl_exit(void)
+{
+       ovs_vport_ops_unregister(&ovs_vxlan_netdev_vport_ops);
+}
+
+module_init(ovs_vxlan_tnl_init);
+module_exit(ovs_vxlan_tnl_exit);
+
+MODULE_DESCRIPTION("OVS: VXLAN switching port");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("vport-type-4");