datapath/vport-gre.c

   1 /*
   2  * Copyright (c) 2007-2012 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 #include <linux/kconfig.h>
  20 #if IS_ENABLED(CONFIG_NET_IPGRE_DEMUX)
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/if.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/ip.h>
  26 #include <linux/if_tunnel.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/in.h>
  29 #include <linux/if_vlan.h>
  30 #include <linux/in.h>
  31 #include <linux/in_route.h>
  32 #include <linux/inetdevice.h>
  33 #include <linux/jhash.h>
  34 #include <linux/list.h>
  35 #include <linux/kernel.h>
  36 #include <linux/workqueue.h>
  37 #include <linux/rculist.h>
  38 #include <net/net_namespace.h>
  39 #include <net/netns/generic.h>
  40 #include <net/route.h>
  41 #include <net/xfrm.h>
  42
  43 #include <net/icmp.h>
  44 #include <net/ip.h>
  45 #include <net/ip_tunnels.h>
  46 #include <net/gre.h>
  47 #include <net/protocol.h>
  48
  49 #include "datapath.h"
  50 #include "vport.h"
  51
  52 /* Returns the least-significant 32 bits of a __be64. */
  53 static __be32 be64_get_low32(__be64 x)
  54 {
  55 #ifdef __BIG_ENDIAN
  56         return (__force __be32)x;
  57 #else
  58         return (__force __be32)((__force u64)x >> 32);
  59 #endif
  60 }
  61
  62 static __be16 filter_tnl_flags(__be16 flags)
  63 {
  64         return flags & (TUNNEL_CSUM | TUNNEL_KEY);
  65 }
  66
  67 static struct sk_buff *__build_header(struct sk_buff *skb,
  68                                       int tunnel_hlen,
  69                                       __be32 seq, __be16 gre64_flag)
  70 {
  71         const struct ovs_key_ipv4_tunnel *tun_key = OVS_CB(skb)->tun_key;
  72         struct tnl_ptk_info tpi;
  73
  74         skb = gre_handle_offloads(skb, !!(tun_key->tun_flags & TUNNEL_CSUM));
  75         if (IS_ERR(skb))
  76                 return NULL;
  77
  78         tpi.flags = filter_tnl_flags(tun_key->tun_flags) | gre64_flag;
  79
  80         tpi.proto = htons(ETH_P_TEB);
  81         tpi.key = be64_get_low32(tun_key->tun_id);
  82         tpi.seq = seq;
  83         gre_build_header(skb, &tpi, tunnel_hlen);
  84
  85         return skb;
  86 }
  87
  88 static __be64 key_to_tunnel_id(__be32 key, __be32 seq)
  89 {
  90 #ifdef __BIG_ENDIAN
  91         return (__force __be64)((__force u64)seq << 32 | (__force u32)key);
  92 #else
  93         return (__force __be64)((__force u64)key << 32 | (__force u32)seq);
  94 #endif
  95 }
  96
  97 /* Called with rcu_read_lock and BH disabled. */
  98 static int gre_rcv(struct sk_buff *skb,
  99                    const struct tnl_ptk_info *tpi)
 100 {
 101         struct ovs_key_ipv4_tunnel tun_key;
 102         struct ovs_net *ovs_net;
 103         struct vport *vport;
 104         __be64 key;
 105
 106         ovs_net = net_generic(dev_net(skb->dev), ovs_net_id);
 107         if ((tpi->flags & TUNNEL_KEY) && (tpi->flags & TUNNEL_SEQ))
 108                 vport = rcu_dereference(ovs_net->vport_net.gre64_vport);
 109         else
 110                 vport = rcu_dereference(ovs_net->vport_net.gre_vport);
 111         if (unlikely(!vport))
 112                 return PACKET_REJECT;
 113
 114         key = key_to_tunnel_id(tpi->key, tpi->seq);
 115         ovs_flow_tun_key_init(&tun_key, ip_hdr(skb), key, filter_tnl_flags(tpi->flags));
 116
 117         ovs_vport_receive(vport, skb, &tun_key);
 118         return PACKET_RCVD;
 119 }
 120
 121 static int __send(struct vport *vport, struct sk_buff *skb,
 122                   int tunnel_hlen,
 123                   __be32 seq, __be16 gre64_flag)
 124 {
 125         struct net *net = ovs_dp_get_net(vport->dp);
 126         struct rtable *rt;
 127         int min_headroom;
 128         __be16 df;
 129         __be32 saddr;
 130         int err;
 131
 132         forward_ip_summed(skb, true);
 133
 134         /* Route lookup */
 135         saddr = OVS_CB(skb)->tun_key->ipv4_src;
 136         rt = find_route(ovs_dp_get_net(vport->dp),
 137                         &saddr,
 138                         OVS_CB(skb)->tun_key->ipv4_dst,
 139                         IPPROTO_GRE,
 140                         OVS_CB(skb)->tun_key->ipv4_tos,
 141                         skb->mark);
 142         if (IS_ERR(rt)) {
 143                 err = PTR_ERR(rt);
 144                 goto error;
 145         }
 146
 147         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 148                         + tunnel_hlen + sizeof(struct iphdr)
 149                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 150
 151         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 152                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 153                                                 skb_headroom(skb) +
 154                                                 16);
 155                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 156                                         0, GFP_ATOMIC);
 157                 if (unlikely(err))
 158                         goto err_free_rt;
 159         }
 160
 161         if (unlikely(vlan_deaccel_tag(skb))) {
 162                 err = -ENOMEM;
 163                 goto err_free_rt;
 164         }
 165
 166         /* Push Tunnel header. */
 167         skb = __build_header(skb, tunnel_hlen, seq, gre64_flag);
 168         if (unlikely(!skb)) {
 169                 err = 0;
 170                 goto err_free_rt;
 171         }
 172
 173         df = OVS_CB(skb)->tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ?
 174                 htons(IP_DF) : 0;
 175
 176         skb->local_df = 1;
 177
 178         return iptunnel_xmit(net, rt, skb, saddr,
 179                              OVS_CB(skb)->tun_key->ipv4_dst, IPPROTO_GRE,
 180                              OVS_CB(skb)->tun_key->ipv4_tos,
 181                              OVS_CB(skb)->tun_key->ipv4_ttl, df);
 182 err_free_rt:
 183         ip_rt_put(rt);
 184 error:
 185         return err;
 186 }
 187
 188 static struct gre_cisco_protocol gre_protocol = {
 189         .handler        = gre_rcv,
 190         .priority       = 1,
 191 };
 192
 193 static int gre_ports;
 194 static int gre_init(void)
 195 {
 196         int err;
 197
 198         gre_ports++;
 199         if (gre_ports > 1)
 200                 return 0;
 201
 202         err = gre_cisco_register(&gre_protocol);
 203         if (err)
 204                 pr_warn("cannot register gre protocol handler\n");
 205
 206         return err;
 207 }
 208
 209 static void gre_exit(void)
 210 {
 211         gre_ports--;
 212         if (gre_ports > 0)
 213                 return;
 214
 215         gre_cisco_unregister(&gre_protocol);
 216 }
 217
 218 static const char *gre_get_name(const struct vport *vport)
 219 {
 220         return vport_priv(vport);
 221 }
 222
 223 static struct vport *gre_create(const struct vport_parms *parms)
 224 {
 225         struct net *net = ovs_dp_get_net(parms->dp);
 226         struct ovs_net *ovs_net;
 227         struct vport *vport;
 228         int err;
 229
 230         err = gre_init();
 231         if (err)
 232                 return ERR_PTR(err);
 233
 234         ovs_net = net_generic(net, ovs_net_id);
 235         if (ovsl_dereference(ovs_net->vport_net.gre_vport)) {
 236                 vport = ERR_PTR(-EEXIST);
 237                 goto error;
 238         }
 239
 240         vport = ovs_vport_alloc(IFNAMSIZ, &ovs_gre_vport_ops, parms);
 241         if (IS_ERR(vport))
 242                 goto error;
 243
 244         strncpy(vport_priv(vport), parms->name, IFNAMSIZ);
 245         rcu_assign_pointer(ovs_net->vport_net.gre_vport, vport);
 246         return vport;
 247
 248 error:
 249         gre_exit();
 250         return vport;
 251 }
 252
 253 static void gre_tnl_destroy(struct vport *vport)
 254 {
 255         struct net *net = ovs_dp_get_net(vport->dp);
 256         struct ovs_net *ovs_net;
 257
 258         ovs_net = net_generic(net, ovs_net_id);
 259
 260         rcu_assign_pointer(ovs_net->vport_net.gre_vport, NULL);
 261         ovs_vport_deferred_free(vport);
 262         gre_exit();
 263 }
 264
 265 static int gre_send(struct vport *vport, struct sk_buff *skb)
 266 {
 267         int hlen;
 268
 269         if (unlikely(!OVS_CB(skb)->tun_key))
 270                 return -EINVAL;
 271
 272         hlen = ip_gre_calc_hlen(OVS_CB(skb)->tun_key->tun_flags);
 273
 274         return __send(vport, skb, hlen, 0, 0);
 275 }
 276
 277 const struct vport_ops ovs_gre_vport_ops = {
 278         .type           = OVS_VPORT_TYPE_GRE,
 279         .create         = gre_create,
 280         .destroy        = gre_tnl_destroy,
 281         .get_name       = gre_get_name,
 282         .send           = gre_send,
 283 };
 284
 285 /* GRE64 vport. */
 286 static struct vport *gre64_create(const struct vport_parms *parms)
 287 {
 288         struct net *net = ovs_dp_get_net(parms->dp);
 289         struct ovs_net *ovs_net;
 290         struct vport *vport;
 291         int err;
 292
 293         err = gre_init();
 294         if (err)
 295                 return ERR_PTR(err);
 296
 297         ovs_net = net_generic(net, ovs_net_id);
 298         if (ovsl_dereference(ovs_net->vport_net.gre64_vport)) {
 299                 vport = ERR_PTR(-EEXIST);
 300                 goto error;
 301         }
 302
 303         vport = ovs_vport_alloc(IFNAMSIZ, &ovs_gre64_vport_ops, parms);
 304         if (IS_ERR(vport))
 305                 goto error;
 306
 307         strncpy(vport_priv(vport), parms->name, IFNAMSIZ);
 308         rcu_assign_pointer(ovs_net->vport_net.gre64_vport, vport);
 309         return vport;
 310 error:
 311         gre_exit();
 312         return vport;
 313 }
 314
 315 static void gre64_tnl_destroy(struct vport *vport)
 316 {
 317         struct net *net = ovs_dp_get_net(vport->dp);
 318         struct ovs_net *ovs_net;
 319
 320         ovs_net = net_generic(net, ovs_net_id);
 321
 322         rcu_assign_pointer(ovs_net->vport_net.gre64_vport, NULL);
 323         ovs_vport_deferred_free(vport);
 324         gre_exit();
 325 }
 326
 327 static __be32 be64_get_high32(__be64 x)
 328 {
 329 #ifdef __BIG_ENDIAN
 330         return (__force __be32)((__force u64)x >> 32);
 331 #else
 332         return (__force __be32)x;
 333 #endif
 334 }
 335
 336 static int gre64_send(struct vport *vport, struct sk_buff *skb)
 337 {
 338         int hlen = GRE_HEADER_SECTION +         /* GRE Hdr */
 339                    GRE_HEADER_SECTION +         /* GRE Key */
 340                    GRE_HEADER_SECTION;          /* GRE SEQ */
 341         __be32 seq;
 342
 343         if (unlikely(!OVS_CB(skb)->tun_key))
 344                 return -EINVAL;
 345
 346         if (OVS_CB(skb)->tun_key->tun_flags & TUNNEL_CSUM)
 347                 hlen += GRE_HEADER_SECTION;
 348
 349         seq = be64_get_high32(OVS_CB(skb)->tun_key->tun_id);
 350         return __send(vport, skb, hlen, seq, (TUNNEL_KEY|TUNNEL_SEQ));
 351 }
 352
 353 const struct vport_ops ovs_gre64_vport_ops = {
 354         .type           = OVS_VPORT_TYPE_GRE64,
 355         .create         = gre64_create,
 356         .destroy        = gre64_tnl_destroy,
 357         .get_name       = gre_get_name,
 358         .send           = gre64_send,
 359 };
 360 #endif