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 rtable *rt;
 126         int min_headroom;
 127         __be16 df;
 128         __be32 saddr;
 129         int err;
 130
 131         /* Route lookup */
 132         saddr = OVS_CB(skb)->tun_key->ipv4_src;
 133         rt = find_route(ovs_dp_get_net(vport->dp),
 134                         &saddr,
 135                         OVS_CB(skb)->tun_key->ipv4_dst,
 136                         IPPROTO_GRE,
 137                         OVS_CB(skb)->tun_key->ipv4_tos,
 138                         skb->mark);
 139         if (IS_ERR(rt)) {
 140                 err = PTR_ERR(rt);
 141                 goto error;
 142         }
 143
 144         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 145                         + tunnel_hlen + sizeof(struct iphdr)
 146                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 147
 148         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 149                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 150                                                 skb_headroom(skb) +
 151                                                 16);
 152                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 153                                         0, GFP_ATOMIC);
 154                 if (unlikely(err))
 155                         goto err_free_rt;
 156         }
 157
 158         if (vlan_tx_tag_present(skb)) {
 159                 if (unlikely(!__vlan_put_tag(skb,
 160                                              skb->vlan_proto,
 161                                              vlan_tx_tag_get(skb)))) {
 162                         err = -ENOMEM;
 163                         goto err_free_rt;
 164                 }
 165                 vlan_set_tci(skb, 0);
 166         }
 167
 168         /* Push Tunnel header. */
 169         skb = __build_header(skb, tunnel_hlen, seq, gre64_flag);
 170         if (unlikely(!skb)) {
 171                 err = 0;
 172                 goto err_free_rt;
 173         }
 174
 175         df = OVS_CB(skb)->tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ?
 176                 htons(IP_DF) : 0;
 177
 178         skb->local_df = 1;
 179
 180         return iptunnel_xmit(rt, skb, saddr,
 181                              OVS_CB(skb)->tun_key->ipv4_dst, IPPROTO_GRE,
 182                              OVS_CB(skb)->tun_key->ipv4_tos,
 183                              OVS_CB(skb)->tun_key->ipv4_ttl, df);
 184 err_free_rt:
 185         ip_rt_put(rt);
 186 error:
 187         return err;
 188 }
 189
 190 static struct gre_cisco_protocol gre_protocol = {
 191         .handler        = gre_rcv,
 192         .priority       = 1,
 193 };
 194
 195 static int gre_ports;
 196 static int gre_init(void)
 197 {
 198         int err;
 199
 200         gre_ports++;
 201         if (gre_ports > 1)
 202                 return 0;
 203
 204         err = gre_cisco_register(&gre_protocol);
 205         if (err)
 206                 pr_warn("cannot register gre protocol handler\n");
 207
 208         return err;
 209 }
 210
 211 static void gre_exit(void)
 212 {
 213         gre_ports--;
 214         if (gre_ports > 0)
 215                 return;
 216
 217         gre_cisco_unregister(&gre_protocol);
 218 }
 219
 220 static const char *gre_get_name(const struct vport *vport)
 221 {
 222         return vport_priv(vport);
 223 }
 224
 225 static struct vport *gre_create(const struct vport_parms *parms)
 226 {
 227         struct net *net = ovs_dp_get_net(parms->dp);
 228         struct ovs_net *ovs_net;
 229         struct vport *vport;
 230         int err;
 231
 232         err = gre_init();
 233         if (err)
 234                 return ERR_PTR(err);
 235
 236         ovs_net = net_generic(net, ovs_net_id);
 237         if (ovsl_dereference(ovs_net->vport_net.gre_vport)) {
 238                 vport = ERR_PTR(-EEXIST);
 239                 goto error;
 240         }
 241
 242         vport = ovs_vport_alloc(IFNAMSIZ, &ovs_gre_vport_ops, parms);
 243         if (IS_ERR(vport))
 244                 goto error;
 245
 246         strncpy(vport_priv(vport), parms->name, IFNAMSIZ);
 247         rcu_assign_pointer(ovs_net->vport_net.gre_vport, vport);
 248         return vport;
 249
 250 error:
 251         gre_exit();
 252         return vport;
 253 }
 254
 255 static void gre_tnl_destroy(struct vport *vport)
 256 {
 257         struct net *net = ovs_dp_get_net(vport->dp);
 258         struct ovs_net *ovs_net;
 259
 260         ovs_net = net_generic(net, ovs_net_id);
 261
 262         rcu_assign_pointer(ovs_net->vport_net.gre_vport, NULL);
 263         ovs_vport_deferred_free(vport);
 264         gre_exit();
 265 }
 266
 267 static int gre_send(struct vport *vport, struct sk_buff *skb)
 268 {
 269         int hlen;
 270
 271         if (unlikely(!OVS_CB(skb)->tun_key))
 272                 return -EINVAL;
 273
 274         hlen = ip_gre_calc_hlen(OVS_CB(skb)->tun_key->tun_flags);
 275
 276         return __send(vport, skb, hlen, 0, 0);
 277 }
 278
 279 const struct vport_ops ovs_gre_vport_ops = {
 280         .type           = OVS_VPORT_TYPE_GRE,
 281         .create         = gre_create,
 282         .destroy        = gre_tnl_destroy,
 283         .get_name       = gre_get_name,
 284         .send           = gre_send,
 285 };
 286
 287 /* GRE64 vport. */
 288 static struct vport *gre64_create(const struct vport_parms *parms)
 289 {
 290         struct net *net = ovs_dp_get_net(parms->dp);
 291         struct ovs_net *ovs_net;
 292         struct vport *vport;
 293         int err;
 294
 295         err = gre_init();
 296         if (err)
 297                 return ERR_PTR(err);
 298
 299         ovs_net = net_generic(net, ovs_net_id);
 300         if (ovsl_dereference(ovs_net->vport_net.gre64_vport)) {
 301                 vport = ERR_PTR(-EEXIST);
 302                 goto error;
 303         }
 304
 305         vport = ovs_vport_alloc(IFNAMSIZ, &ovs_gre64_vport_ops, parms);
 306         if (IS_ERR(vport))
 307                 goto error;
 308
 309         strncpy(vport_priv(vport), parms->name, IFNAMSIZ);
 310         rcu_assign_pointer(ovs_net->vport_net.gre64_vport, vport);
 311         return vport;
 312 error:
 313         gre_exit();
 314         return vport;
 315 }
 316
 317 static void gre64_tnl_destroy(struct vport *vport)
 318 {
 319         struct net *net = ovs_dp_get_net(vport->dp);
 320         struct ovs_net *ovs_net;
 321
 322         ovs_net = net_generic(net, ovs_net_id);
 323
 324         rcu_assign_pointer(ovs_net->vport_net.gre64_vport, NULL);
 325         ovs_vport_deferred_free(vport);
 326         gre_exit();
 327 }
 328
 329 static __be32 be64_get_high32(__be64 x)
 330 {
 331 #ifdef __BIG_ENDIAN
 332         return (__force __be32)((__force u64)x >> 32);
 333 #else
 334         return (__force __be32)x;
 335 #endif
 336 }
 337
 338 static int gre64_send(struct vport *vport, struct sk_buff *skb)
 339 {
 340         int hlen = GRE_HEADER_SECTION +         /* GRE Hdr */
 341                    GRE_HEADER_SECTION +         /* GRE Key */
 342                    GRE_HEADER_SECTION;          /* GRE SEQ */
 343         __be32 seq;
 344
 345         if (unlikely(!OVS_CB(skb)->tun_key))
 346                 return -EINVAL;
 347
 348         if (OVS_CB(skb)->tun_key->tun_flags & TUNNEL_CSUM)
 349                 hlen += GRE_HEADER_SECTION;
 350
 351         seq = be64_get_high32(OVS_CB(skb)->tun_key->tun_id);
 352         return __send(vport, skb, hlen, seq, (TUNNEL_KEY|TUNNEL_SEQ));
 353 }
 354
 355 const struct vport_ops ovs_gre64_vport_ops = {
 356         .type           = OVS_VPORT_TYPE_GRE64,
 357         .create         = gre64_create,
 358         .destroy        = gre64_tnl_destroy,
 359         .get_name       = gre_get_name,
 360         .send           = gre64_send,
 361 };
 362 #endif