net/openvswitch/vport-gre.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/if.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/ip.h>
  24 #include <linux/if_tunnel.h>
  25 #include <linux/if_vlan.h>
  26 #include <linux/in.h>
  27 #include <linux/in_route.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/jhash.h>
  30 #include <linux/list.h>
  31 #include <linux/kernel.h>
  32 #include <linux/workqueue.h>
  33 #include <linux/rculist.h>
  34 #include <net/route.h>
  35 #include <net/xfrm.h>
  36
  37 #include <net/icmp.h>
  38 #include <net/ip.h>
  39 #include <net/ip_tunnels.h>
  40 #include <net/gre.h>
  41 #include <net/net_namespace.h>
  42 #include <net/netns/generic.h>
  43 #include <net/protocol.h>
  44
  45 #include "datapath.h"
  46 #include "vport.h"
  47
  48 /* Returns the least-significant 32 bits of a __be64. */
  49 static __be32 be64_get_low32(__be64 x)
  50 {
  51 #ifdef __BIG_ENDIAN
  52         return (__force __be32)x;
  53 #else
  54         return (__force __be32)((__force u64)x >> 32);
  55 #endif
  56 }
  57
  58 static __be16 filter_tnl_flags(__be16 flags)
  59 {
  60         return flags & (TUNNEL_CSUM | TUNNEL_KEY);
  61 }
  62
  63 static struct sk_buff *__build_header(struct sk_buff *skb,
  64                                       int tunnel_hlen)
  65 {
  66         const struct ovs_key_ipv4_tunnel *tun_key = OVS_CB(skb)->tun_key;
  67         struct tnl_ptk_info tpi;
  68
  69         skb = gre_handle_offloads(skb, !!(tun_key->tun_flags & TUNNEL_CSUM));
  70         if (IS_ERR(skb))
  71                 return NULL;
  72
  73         tpi.flags = filter_tnl_flags(tun_key->tun_flags);
  74         tpi.proto = htons(ETH_P_TEB);
  75         tpi.key = be64_get_low32(tun_key->tun_id);
  76         tpi.seq = 0;
  77         gre_build_header(skb, &tpi, tunnel_hlen);
  78
  79         return skb;
  80 }
  81
  82 static __be64 key_to_tunnel_id(__be32 key, __be32 seq)
  83 {
  84 #ifdef __BIG_ENDIAN
  85         return (__force __be64)((__force u64)seq << 32 | (__force u32)key);
  86 #else
  87         return (__force __be64)((__force u64)key << 32 | (__force u32)seq);
  88 #endif
  89 }
  90
  91 /* Called with rcu_read_lock and BH disabled. */
  92 static int gre_rcv(struct sk_buff *skb,
  93                    const struct tnl_ptk_info *tpi)
  94 {
  95         struct ovs_key_ipv4_tunnel tun_key;
  96         struct ovs_net *ovs_net;
  97         struct vport *vport;
  98         __be64 key;
  99
 100         ovs_net = net_generic(dev_net(skb->dev), ovs_net_id);
 101         vport = rcu_dereference(ovs_net->vport_net.gre_vport);
 102         if (unlikely(!vport))
 103                 return PACKET_REJECT;
 104
 105         key = key_to_tunnel_id(tpi->key, tpi->seq);
 106         ovs_flow_tun_key_init(&tun_key, ip_hdr(skb), key,
 107                               filter_tnl_flags(tpi->flags));
 108
 109         ovs_vport_receive(vport, skb, &tun_key);
 110         return PACKET_RCVD;
 111 }
 112
 113 /* Called with rcu_read_lock and BH disabled. */
 114 static int gre_err(struct sk_buff *skb, u32 info,
 115                    const struct tnl_ptk_info *tpi)
 116 {
 117         struct ovs_net *ovs_net;
 118         struct vport *vport;
 119
 120         ovs_net = net_generic(dev_net(skb->dev), ovs_net_id);
 121         vport = rcu_dereference(ovs_net->vport_net.gre_vport);
 122
 123         if (unlikely(!vport))
 124                 return PACKET_REJECT;
 125         else
 126                 return PACKET_RCVD;
 127 }
 128
 129 static int gre_tnl_send(struct vport *vport, struct sk_buff *skb)
 130 {
 131         struct net *net = ovs_dp_get_net(vport->dp);
 132         struct flowi4 fl;
 133         struct rtable *rt;
 134         int min_headroom;
 135         int tunnel_hlen;
 136         __be16 df;
 137         int err;
 138
 139         if (unlikely(!OVS_CB(skb)->tun_key)) {
 140                 err = -EINVAL;
 141                 goto error;
 142         }
 143
 144         /* Route lookup */
 145         memset(&fl, 0, sizeof(fl));
 146         fl.daddr = OVS_CB(skb)->tun_key->ipv4_dst;
 147         fl.saddr = OVS_CB(skb)->tun_key->ipv4_src;
 148         fl.flowi4_tos = RT_TOS(OVS_CB(skb)->tun_key->ipv4_tos);
 149         fl.flowi4_mark = skb->mark;
 150         fl.flowi4_proto = IPPROTO_GRE;
 151
 152         rt = ip_route_output_key(net, &fl);
 153         if (IS_ERR(rt))
 154                 return PTR_ERR(rt);
 155
 156         tunnel_hlen = ip_gre_calc_hlen(OVS_CB(skb)->tun_key->tun_flags);
 157
 158         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 159                         + tunnel_hlen + sizeof(struct iphdr)
 160                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 161         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 162                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 163                                                 skb_headroom(skb) +
 164                                                 16);
 165                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 166                                         0, GFP_ATOMIC);
 167                 if (unlikely(err))
 168                         goto err_free_rt;
 169         }
 170
 171         if (vlan_tx_tag_present(skb)) {
 172                 if (unlikely(!__vlan_put_tag(skb,
 173                                              skb->vlan_proto,
 174                                              vlan_tx_tag_get(skb)))) {
 175                         err = -ENOMEM;
 176                         goto err_free_rt;
 177                 }
 178                 skb->vlan_tci = 0;
 179         }
 180
 181         /* Push Tunnel header. */
 182         skb = __build_header(skb, tunnel_hlen);
 183         if (unlikely(!skb)) {
 184                 err = 0;
 185                 goto err_free_rt;
 186         }
 187
 188         df = OVS_CB(skb)->tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ?
 189                 htons(IP_DF) : 0;
 190
 191         skb->ignore_df = 1;
 192
 193         return iptunnel_xmit(skb->sk, rt, skb, fl.saddr,
 194                              OVS_CB(skb)->tun_key->ipv4_dst, IPPROTO_GRE,
 195                              OVS_CB(skb)->tun_key->ipv4_tos,
 196                              OVS_CB(skb)->tun_key->ipv4_ttl, df, false);
 197 err_free_rt:
 198         ip_rt_put(rt);
 199 error:
 200         return err;
 201 }
 202
 203 static struct gre_cisco_protocol gre_protocol = {
 204         .handler        = gre_rcv,
 205         .err_handler    = gre_err,
 206         .priority       = 1,
 207 };
 208
 209 static int gre_ports;
 210 static int gre_init(void)
 211 {
 212         int err;
 213
 214         gre_ports++;
 215         if (gre_ports > 1)
 216                 return 0;
 217
 218         err = gre_cisco_register(&gre_protocol);
 219         if (err)
 220                 pr_warn("cannot register gre protocol handler\n");
 221
 222         return err;
 223 }
 224
 225 static void gre_exit(void)
 226 {
 227         gre_ports--;
 228         if (gre_ports > 0)
 229                 return;
 230
 231         gre_cisco_unregister(&gre_protocol);
 232 }
 233
 234 static const char *gre_get_name(const struct vport *vport)
 235 {
 236         return vport_priv(vport);
 237 }
 238
 239 static struct vport *gre_create(const struct vport_parms *parms)
 240 {
 241         struct net *net = ovs_dp_get_net(parms->dp);
 242         struct ovs_net *ovs_net;
 243         struct vport *vport;
 244         int err;
 245
 246         err = gre_init();
 247         if (err)
 248                 return ERR_PTR(err);
 249
 250         ovs_net = net_generic(net, ovs_net_id);
 251         if (ovsl_dereference(ovs_net->vport_net.gre_vport)) {
 252                 vport = ERR_PTR(-EEXIST);
 253                 goto error;
 254         }
 255
 256         vport = ovs_vport_alloc(IFNAMSIZ, &ovs_gre_vport_ops, parms);
 257         if (IS_ERR(vport))
 258                 goto error;
 259
 260         strncpy(vport_priv(vport), parms->name, IFNAMSIZ);
 261         rcu_assign_pointer(ovs_net->vport_net.gre_vport, vport);
 262         return vport;
 263
 264 error:
 265         gre_exit();
 266         return vport;
 267 }
 268
 269 static void gre_tnl_destroy(struct vport *vport)
 270 {
 271         struct net *net = ovs_dp_get_net(vport->dp);
 272         struct ovs_net *ovs_net;
 273
 274         ovs_net = net_generic(net, ovs_net_id);
 275
 276         RCU_INIT_POINTER(ovs_net->vport_net.gre_vport, NULL);
 277         ovs_vport_deferred_free(vport);
 278         gre_exit();
 279 }
 280
 281 const struct vport_ops ovs_gre_vport_ops = {
 282         .type           = OVS_VPORT_TYPE_GRE,
 283         .create         = gre_create,
 284         .destroy        = gre_tnl_destroy,
 285         .get_name       = gre_get_name,
 286         .send           = gre_tnl_send,
 287 };