datapath/vport-vxlan.c

   1 /*
   2  * Copyright (c) 2013 Nicira, Inc.
   3  * Copyright (c) 2013 Cisco Systems, Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of version 2 of the GNU General Public
   7  * License as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  * General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  17  * 02110-1301, USA
  18  */
  19
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/version.h>
  23
  24 #include <linux/in.h>
  25 #include <linux/ip.h>
  26 #include <linux/net.h>
  27 #include <linux/rculist.h>
  28 #include <linux/udp.h>
  29 #include <linux/module.h>
  30
  31 #include <net/icmp.h>
  32 #include <net/ip.h>
  33 #include <net/udp.h>
  34 #include <net/ip_tunnels.h>
  35 #include <net/rtnetlink.h>
  36 #include <net/route.h>
  37 #include <net/dsfield.h>
  38 #include <net/inet_ecn.h>
  39 #include <net/net_namespace.h>
  40 #include <net/netns/generic.h>
  41 #include <net/vxlan.h>
  42
  43 #include "datapath.h"
  44 #include "vport.h"
  45 #include "vport-vxlan.h"
  46
  47 /**
  48  * struct vxlan_port - Keeps track of open UDP ports
  49  * @vs: vxlan_sock created for the port.
  50  * @name: vport name.
  51  */
  52 struct vxlan_port {
  53         struct vxlan_sock *vs;
  54         char name[IFNAMSIZ];
  55         u32 exts; /* VXLAN_F_* in <net/vxlan.h> */
  56 };
  57
  58 static struct vport_ops ovs_vxlan_vport_ops;
  59
  60 static inline struct vxlan_port *vxlan_vport(const struct vport *vport)
  61 {
  62         return vport_priv(vport);
  63 }
  64
  65 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
  66                       struct vxlan_metadata *md)
  67 {
  68         struct ovs_tunnel_info tun_info;
  69         struct vxlan_port *vxlan_port;
  70         struct vport *vport = vs->data;
  71         struct iphdr *iph;
  72         struct ovs_vxlan_opts opts = {
  73                 .gbp = md->gbp,
  74         };
  75         __be64 key;
  76         __be16 flags;
  77
  78         flags = TUNNEL_KEY | (udp_hdr(skb)->check != 0 ? TUNNEL_CSUM : 0);
  79         vxlan_port = vxlan_vport(vport);
  80         if (vxlan_port->exts & VXLAN_F_GBP && md->gbp)
  81                 flags |= TUNNEL_VXLAN_OPT;
  82
  83         /* Save outer tunnel values */
  84         iph = ip_hdr(skb);
  85         key = cpu_to_be64(ntohl(md->vni) >> 8);
  86         ovs_flow_tun_info_init(&tun_info, iph,
  87                                udp_hdr(skb)->source, udp_hdr(skb)->dest,
  88                                key, flags, &opts, sizeof(opts));
  89
  90         ovs_vport_receive(vport, skb, &tun_info);
  91 }
  92
  93 static int vxlan_get_options(const struct vport *vport, struct sk_buff *skb)
  94 {
  95         struct vxlan_port *vxlan_port = vxlan_vport(vport);
  96         __be16 dst_port = inet_sport(vxlan_port->vs->sock->sk);
  97
  98         if (nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT, ntohs(dst_port)))
  99                 return -EMSGSIZE;
 100
 101         if (vxlan_port->exts) {
 102                 struct nlattr *exts;
 103
 104                 exts = nla_nest_start(skb, OVS_TUNNEL_ATTR_EXTENSION);
 105                 if (!exts)
 106                         return -EMSGSIZE;
 107
 108                 if (vxlan_port->exts & VXLAN_F_GBP &&
 109                     nla_put_flag(skb, OVS_VXLAN_EXT_GBP))
 110                         return -EMSGSIZE;
 111
 112                 nla_nest_end(skb, exts);
 113         }
 114
 115         return 0;
 116 }
 117
 118 static void vxlan_tnl_destroy(struct vport *vport)
 119 {
 120         struct vxlan_port *vxlan_port = vxlan_vport(vport);
 121
 122         vxlan_sock_release(vxlan_port->vs);
 123
 124         ovs_vport_deferred_free(vport);
 125 }
 126
 127 static const struct nla_policy exts_policy[OVS_VXLAN_EXT_MAX+1] = {
 128         [OVS_VXLAN_EXT_GBP]     = { .type = NLA_FLAG, },
 129 };
 130
 131 static int vxlan_configure_exts(struct vport *vport, struct nlattr *attr)
 132 {
 133         struct nlattr *exts[OVS_VXLAN_EXT_MAX+1];
 134         struct vxlan_port *vxlan_port;
 135         int err;
 136
 137         if (nla_len(attr) < sizeof(struct nlattr))
 138                 return -EINVAL;
 139
 140         err = nla_parse_nested(exts, OVS_VXLAN_EXT_MAX, attr, exts_policy);
 141         if (err < 0)
 142                 return err;
 143
 144         vxlan_port = vxlan_vport(vport);
 145
 146         if (exts[OVS_VXLAN_EXT_GBP])
 147                 vxlan_port->exts |= VXLAN_F_GBP;
 148
 149         return 0;
 150 }
 151
 152 static struct vport *vxlan_tnl_create(const struct vport_parms *parms)
 153 {
 154         struct net *net = ovs_dp_get_net(parms->dp);
 155         struct nlattr *options = parms->options;
 156         struct vxlan_port *vxlan_port;
 157         struct vxlan_sock *vs;
 158         struct vport *vport;
 159         struct nlattr *a;
 160         u16 dst_port;
 161         int err;
 162
 163         if (!options) {
 164                 err = -EINVAL;
 165                 goto error;
 166         }
 167         a = nla_find_nested(options, OVS_TUNNEL_ATTR_DST_PORT);
 168         if (a && nla_len(a) == sizeof(u16)) {
 169                 dst_port = nla_get_u16(a);
 170         } else {
 171                 /* Require destination port from userspace. */
 172                 err = -EINVAL;
 173                 goto error;
 174         }
 175
 176         vport = ovs_vport_alloc(sizeof(struct vxlan_port),
 177                                 &ovs_vxlan_vport_ops, parms);
 178         if (IS_ERR(vport))
 179                 return vport;
 180
 181         vxlan_port = vxlan_vport(vport);
 182         strncpy(vxlan_port->name, parms->name, IFNAMSIZ);
 183
 184         a = nla_find_nested(options, OVS_TUNNEL_ATTR_EXTENSION);
 185         if (a) {
 186                 err = vxlan_configure_exts(vport, a);
 187                 if (err) {
 188                         ovs_vport_free(vport);
 189                         goto error;
 190                 }
 191         }
 192
 193         vs = vxlan_sock_add(net, htons(dst_port), vxlan_rcv, vport, true,
 194                             vxlan_port->exts);
 195         if (IS_ERR(vs)) {
 196                 ovs_vport_free(vport);
 197                 return (void *)vs;
 198         }
 199         vxlan_port->vs = vs;
 200
 201         return vport;
 202
 203 error:
 204         return ERR_PTR(err);
 205 }
 206
 207 static int vxlan_ext_gbp(struct sk_buff *skb)
 208 {
 209         const struct ovs_tunnel_info *tun_info;
 210         const struct ovs_vxlan_opts *opts;
 211
 212         tun_info = OVS_CB(skb)->egress_tun_info;
 213         opts = tun_info->options;
 214
 215         if (tun_info->tunnel.tun_flags & TUNNEL_VXLAN_OPT &&
 216             tun_info->options_len >= sizeof(*opts))
 217                 return opts->gbp;
 218         else
 219                 return 0;
 220 }
 221
 222 static int vxlan_tnl_send(struct vport *vport, struct sk_buff *skb)
 223 {
 224         struct ovs_key_ipv4_tunnel *tun_key;
 225         struct net *net = ovs_dp_get_net(vport->dp);
 226         struct vxlan_port *vxlan_port = vxlan_vport(vport);
 227         __be16 dst_port = inet_sport(vxlan_port->vs->sock->sk);
 228         struct vxlan_metadata md = {0};
 229         struct rtable *rt;
 230         __be16 src_port;
 231         __be32 saddr;
 232         __be16 df;
 233         int err;
 234         u32 vxflags;
 235
 236         if (unlikely(!OVS_CB(skb)->egress_tun_info)) {
 237                 err = -EINVAL;
 238                 goto error;
 239         }
 240
 241         tun_key = &OVS_CB(skb)->egress_tun_info->tunnel;
 242
 243         /* Route lookup */
 244         saddr = tun_key->ipv4_src;
 245         rt = find_route(ovs_dp_get_net(vport->dp),
 246                         &saddr, tun_key->ipv4_dst,
 247                         IPPROTO_UDP, tun_key->ipv4_tos,
 248                         skb->mark);
 249         if (IS_ERR(rt)) {
 250                 err = PTR_ERR(rt);
 251                 goto error;
 252         }
 253
 254         df = tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 255         skb->ignore_df = 1;
 256
 257         src_port = udp_flow_src_port(net, skb, 0, 0, true);
 258         md.vni = htonl(be64_to_cpu(tun_key->tun_id) << 8);
 259         md.gbp = vxlan_ext_gbp(skb);
 260         vxflags = vxlan_port->exts |
 261                       (tun_key->tun_flags & TUNNEL_CSUM ? VXLAN_F_UDP_CSUM : 0);
 262
 263         err = vxlan_xmit_skb(vxlan_port->vs, rt, skb,
 264                              saddr, tun_key->ipv4_dst,
 265                              tun_key->ipv4_tos,
 266                              tun_key->ipv4_ttl, df,
 267                              src_port, dst_port,
 268                              &md, false, vxflags);
 269         if (err < 0)
 270                 ip_rt_put(rt);
 271         return err;
 272 error:
 273         kfree_skb(skb);
 274         return err;
 275 }
 276
 277 static int vxlan_get_egress_tun_info(struct vport *vport, struct sk_buff *skb,
 278                                      struct ovs_tunnel_info *egress_tun_info)
 279 {
 280         struct net *net = ovs_dp_get_net(vport->dp);
 281         struct vxlan_port *vxlan_port = vxlan_vport(vport);
 282         __be16 dst_port = inet_sport(vxlan_port->vs->sock->sk);
 283         __be16 src_port;
 284
 285         src_port = udp_flow_src_port(net, skb, 0, 0, true);
 286
 287         return ovs_tunnel_get_egress_info(egress_tun_info, net,
 288                                           OVS_CB(skb)->egress_tun_info,
 289                                           IPPROTO_UDP, skb->mark,
 290                                           src_port, dst_port);
 291 }
 292
 293 static const char *vxlan_get_name(const struct vport *vport)
 294 {
 295         struct vxlan_port *vxlan_port = vxlan_vport(vport);
 296         return vxlan_port->name;
 297 }
 298
 299 static struct vport_ops ovs_vxlan_vport_ops = {
 300         .type                   = OVS_VPORT_TYPE_VXLAN,
 301         .create                 = vxlan_tnl_create,
 302         .destroy                = vxlan_tnl_destroy,
 303         .get_name               = vxlan_get_name,
 304         .get_options            = vxlan_get_options,
 305         .send                   = vxlan_tnl_send,
 306         .get_egress_tun_info    = vxlan_get_egress_tun_info,
 307         .owner                  = THIS_MODULE,
 308 };
 309
 310 static int __init ovs_vxlan_tnl_init(void)
 311 {
 312         return ovs_vport_ops_register(&ovs_vxlan_vport_ops);
 313 }
 314
 315 static void __exit ovs_vxlan_tnl_exit(void)
 316 {
 317         ovs_vport_ops_unregister(&ovs_vxlan_vport_ops);
 318 }
 319
 320 module_init(ovs_vxlan_tnl_init);
 321 module_exit(ovs_vxlan_tnl_exit);
 322
 323 MODULE_DESCRIPTION("OVS: VXLAN switching port");
 324 MODULE_LICENSE("GPL");
 325 MODULE_ALIAS("vport-type-4");