datapath/vport-geneve.c

   1 /*
   2  * Copyright (c) 2014 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/version.h>
  22
  23 #include <linux/in.h>
  24 #include <linux/ip.h>
  25 #include <linux/net.h>
  26 #include <linux/rculist.h>
  27 #include <linux/udp.h>
  28
  29 #include <net/geneve.h>
  30 #include <net/icmp.h>
  31 #include <net/ip.h>
  32 #include <net/route.h>
  33 #include <net/udp.h>
  34 #include <net/xfrm.h>
  35
  36 #include "datapath.h"
  37 #include "gso.h"
  38 #include "vport.h"
  39
  40 /*
  41  * Geneve Header:
  42  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  43  *  |Ver|  Opt Len  |O|C|    Rsvd.  |          Protocol Type        |
  44  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  45  *  |        Virtual Network Identifier (VNI)       |    Reserved   |
  46  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  47  *  |                    Variable Length Options                    |
  48  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  49  *
  50  * Option Header:
  51  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  52  *  |          Option Class         |      Type     |R|R|R| Length  |
  53  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  54  *  |                      Variable Option Data                     |
  55  *  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  56  */
  57
  58 struct genevehdr {
  59 #ifdef __LITTLE_ENDIAN_BITFIELD
  60         u8 opt_len:6;
  61         u8 ver:2;
  62         u8 rsvd1:6;
  63         u8 critical:1;
  64         u8 oam:1;
  65 #else
  66         u8 ver:2;
  67         u8 opt_len:6;
  68         u8 oam:1;
  69         u8 critical:1;
  70         u8 rsvd1:6;
  71 #endif
  72         __be16 proto_type;
  73         u8 vni[3];
  74         u8 rsvd2;
  75         struct geneve_opt options[];
  76 };
  77
  78 #define GENEVE_VER 0
  79
  80 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  81
  82 /**
  83  * struct geneve_port - Keeps track of open UDP ports
  84  * @sock: The socket created for this port number.
  85  * @name: vport name.
  86  */
  87 struct geneve_port {
  88         struct socket *sock;
  89         char name[IFNAMSIZ];
  90 };
  91
  92 static LIST_HEAD(geneve_ports);
  93
  94 static inline struct geneve_port *geneve_vport(const struct vport *vport)
  95 {
  96         return vport_priv(vport);
  97 }
  98
  99 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 100 {
 101         return (struct genevehdr *)(udp_hdr(skb) + 1);
 102 }
 103
 104 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 106 {
 107 #ifdef __BIG_ENDIAN
 108         vni[0] = (__force __u8)(tun_id >> 16);
 109         vni[1] = (__force __u8)(tun_id >> 8);
 110         vni[2] = (__force __u8)tun_id;
 111 #else
 112         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 113         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 114         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 115 #endif
 116 }
 117
 118 /* Convert 24 bit VNI to 64 bit tunnel ID. */
 119 static __be64 vni_to_tunnel_id(const __u8 *vni)
 120 {
 121 #ifdef __BIG_ENDIAN
 122         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 123 #else
 124         return (__force __be64)(((__force u64)vni[0] << 40) |
 125                                 ((__force u64)vni[1] << 48) |
 126                                 ((__force u64)vni[2] << 56));
 127 #endif
 128 }
 129
 130 static void geneve_build_header(const struct vport *vport,
 131                               struct sk_buff *skb)
 132 {
 133         struct geneve_port *geneve_port = geneve_vport(vport);
 134         struct net *net = ovs_dp_get_net(vport->dp);
 135         struct udphdr *udph = udp_hdr(skb);
 136         struct genevehdr *geneveh = (struct genevehdr *)(udph + 1);
 137         const struct ovs_tunnel_info *tun_info = OVS_CB(skb)->egress_tun_info;
 138
 139         udph->dest = inet_sport(geneve_port->sock->sk);
 140         udph->source = udp_flow_src_port(net, skb, 0, 0, true);
 141         udph->check = 0;
 142         udph->len = htons(skb->len - skb_transport_offset(skb));
 143
 144         geneveh->ver = GENEVE_VER;
 145         geneveh->opt_len = tun_info->options_len / 4;
 146         geneveh->oam = !!(tun_info->tunnel.tun_flags & TUNNEL_OAM);
 147         geneveh->critical = !!(tun_info->tunnel.tun_flags & TUNNEL_CRIT_OPT);
 148         geneveh->rsvd1 = 0;
 149         geneveh->proto_type = htons(ETH_P_TEB);
 150         tunnel_id_to_vni(tun_info->tunnel.tun_id, geneveh->vni);
 151         geneveh->rsvd2 = 0;
 152
 153         memcpy(geneveh->options, tun_info->options, tun_info->options_len);
 154 }
 155
 156 static int geneve_rcv(struct sock *sk, struct sk_buff *skb)
 157 {
 158         struct geneve_port *geneve_port;
 159         struct genevehdr *geneveh;
 160         int opts_len;
 161         struct ovs_tunnel_info tun_info;
 162         __be64 key;
 163         __be16 flags;
 164
 165 #if LINUX_VERSION_CODE < KERNEL_VERSION(3,16,0)
 166         if (unlikely(udp_lib_checksum_complete(skb)))
 167                 goto error;
 168 #endif
 169
 170         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 171                 goto error;
 172
 173         geneveh = geneve_hdr(skb);
 174
 175         if (unlikely(geneveh->ver != GENEVE_VER))
 176                 goto error;
 177
 178         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 179                 goto error;
 180
 181         geneve_port = rcu_dereference_sk_user_data(sk);
 182         if (unlikely(!geneve_port))
 183                 goto error;
 184
 185         opts_len = geneveh->opt_len * 4;
 186         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 187                                  htons(ETH_P_TEB)))
 188                 goto error;
 189
 190         geneveh = geneve_hdr(skb);
 191
 192         flags = TUNNEL_KEY | TUNNEL_OPTIONS_PRESENT |
 193                 (udp_hdr(skb)->check != 0 ? TUNNEL_CSUM : 0) |
 194                 (geneveh->oam ? TUNNEL_OAM : 0) |
 195                 (geneveh->critical ? TUNNEL_CRIT_OPT : 0);
 196
 197         key = vni_to_tunnel_id(geneveh->vni);
 198         ovs_flow_tun_info_init(&tun_info, ip_hdr(skb),
 199                                 udp_hdr(skb)->source, udp_hdr(skb)->dest,
 200                                 key, flags,
 201                                 geneveh->options, opts_len);
 202
 203         ovs_vport_receive(vport_from_priv(geneve_port), skb, &tun_info);
 204         goto out;
 205
 206 error:
 207         kfree_skb(skb);
 208 out:
 209         return 0;
 210 }
 211
 212 /* Arbitrary value.  Irrelevant as long as it's not 0 since we set the handler. */
 213 #define UDP_ENCAP_GENEVE 1
 214 static int geneve_socket_init(struct geneve_port *geneve_port, struct net *net,
 215                               __be16 dst_port)
 216 {
 217         struct sockaddr_in sin;
 218         int err;
 219
 220         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0,
 221                                &geneve_port->sock);
 222         if (err)
 223                 goto error;
 224
 225         /* release net ref. */
 226         sk_change_net(geneve_port->sock->sk, net);
 227
 228         sin.sin_family = AF_INET;
 229         sin.sin_addr.s_addr = htonl(INADDR_ANY);
 230         sin.sin_port = dst_port;
 231
 232         err = kernel_bind(geneve_port->sock,
 233                           (struct sockaddr *)&sin, sizeof(struct sockaddr_in));
 234         if (err)
 235                 goto error_sock;
 236
 237         rcu_assign_sk_user_data(geneve_port->sock->sk, geneve_port);
 238         udp_sk(geneve_port->sock->sk)->encap_type = UDP_ENCAP_GENEVE;
 239         udp_sk(geneve_port->sock->sk)->encap_rcv = geneve_rcv;
 240
 241         udp_encap_enable();
 242
 243         return 0;
 244
 245 error_sock:
 246         sk_release_kernel(geneve_port->sock->sk);
 247 error:
 248         pr_warn("cannot register geneve protocol handler: %d\n", err);
 249         return err;
 250 }
 251
 252 static int geneve_get_options(const struct vport *vport,
 253                               struct sk_buff *skb)
 254 {
 255         struct geneve_port *geneve_port = geneve_vport(vport);
 256
 257         if (nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT,
 258                         ntohs(inet_sport(geneve_port->sock->sk))))
 259                 return -EMSGSIZE;
 260         return 0;
 261 }
 262
 263 static void geneve_tnl_destroy(struct vport *vport)
 264 {
 265         struct geneve_port *geneve_port = geneve_vport(vport);
 266
 267         /* Release socket */
 268         rcu_assign_sk_user_data(geneve_port->sock->sk, NULL);
 269         sk_release_kernel(geneve_port->sock->sk);
 270
 271         ovs_vport_deferred_free(vport);
 272 }
 273
 274 static struct vport *geneve_tnl_create(const struct vport_parms *parms)
 275 {
 276         struct net *net = ovs_dp_get_net(parms->dp);
 277         struct nlattr *options = parms->options;
 278         struct geneve_port *geneve_port;
 279         struct vport *vport;
 280         struct nlattr *a;
 281         int err;
 282         u16 dst_port;
 283
 284         if (!options) {
 285                 err = -EINVAL;
 286                 goto error;
 287         }
 288
 289         a = nla_find_nested(options, OVS_TUNNEL_ATTR_DST_PORT);
 290         if (a && nla_len(a) == sizeof(u16)) {
 291                 dst_port = nla_get_u16(a);
 292         } else {
 293                 /* Require destination port from userspace. */
 294                 err = -EINVAL;
 295                 goto error;
 296         }
 297
 298         vport = ovs_vport_alloc(sizeof(struct geneve_port),
 299                                 &ovs_geneve_vport_ops, parms);
 300         if (IS_ERR(vport))
 301                 return vport;
 302
 303         geneve_port = geneve_vport(vport);
 304         strncpy(geneve_port->name, parms->name, IFNAMSIZ);
 305
 306         err = geneve_socket_init(geneve_port, net, htons(dst_port));
 307         if (err)
 308                 goto error_free;
 309
 310         return vport;
 311
 312 error_free:
 313         ovs_vport_free(vport);
 314 error:
 315         return ERR_PTR(err);
 316 }
 317
 318 #if LINUX_VERSION_CODE < KERNEL_VERSION(3,12,0)
 319
 320 static void geneve_fix_segment(struct sk_buff *skb)
 321 {
 322         struct udphdr *udph = udp_hdr(skb);
 323
 324         udph->len = htons(skb->len - skb_transport_offset(skb));
 325 }
 326
 327 static int handle_offloads(struct sk_buff *skb)
 328 {
 329         if (skb_is_gso(skb)) {
 330                 if (skb_is_encapsulated(skb))
 331                         return -ENOSYS;
 332                 OVS_GSO_CB(skb)->fix_segment = geneve_fix_segment;
 333         } else if (skb->ip_summed != CHECKSUM_PARTIAL) {
 334                 skb->ip_summed = CHECKSUM_NONE;
 335         }
 336         return 0;
 337 }
 338 #else
 339 static int handle_offloads(struct sk_buff *skb)
 340 {
 341         if (skb_is_gso(skb)) {
 342                 int err;
 343
 344                 if (skb_is_encapsulated(skb))
 345                         return -ENOSYS;
 346
 347                 err = skb_unclone(skb, GFP_ATOMIC);
 348                 if (unlikely(err))
 349                         return err;
 350
 351                 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
 352         } else if (skb->ip_summed != CHECKSUM_PARTIAL)
 353                 skb->ip_summed = CHECKSUM_NONE;
 354
 355         skb->encapsulation = 1;
 356         return 0;
 357 }
 358 #endif
 359
 360 static int geneve_send(struct vport *vport, struct sk_buff *skb)
 361 {
 362         struct ovs_key_ipv4_tunnel *tun_key;
 363         int network_offset = skb_network_offset(skb);
 364         struct rtable *rt;
 365         int min_headroom;
 366         __be32 saddr;
 367         __be16 df;
 368         int sent_len;
 369         int err;
 370
 371         if (unlikely(!OVS_CB(skb)->egress_tun_info))
 372                 return -EINVAL;
 373
 374         tun_key = &OVS_CB(skb)->egress_tun_info->tunnel;
 375
 376         /* Route lookup */
 377         saddr = tun_key->ipv4_src;
 378         rt = find_route(ovs_dp_get_net(vport->dp),
 379                         &saddr, tun_key->ipv4_dst,
 380                         IPPROTO_UDP, tun_key->ipv4_tos,
 381                         skb->mark);
 382         if (IS_ERR(rt)) {
 383                 err = PTR_ERR(rt);
 384                 goto error;
 385         }
 386
 387         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 388                         + GENEVE_BASE_HLEN
 389                         + OVS_CB(skb)->egress_tun_info->options_len
 390                         + sizeof(struct iphdr)
 391                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 392
 393         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 394                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 395                                                 skb_headroom(skb) +
 396                                                 16);
 397
 398                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 399                                         0, GFP_ATOMIC);
 400                 if (unlikely(err))
 401                         goto err_free_rt;
 402         }
 403
 404         if (vlan_tx_tag_present(skb)) {
 405                 if (unlikely(!__vlan_put_tag(skb,
 406                                              skb->vlan_proto,
 407                                              vlan_tx_tag_get(skb)))) {
 408                         err = -ENOMEM;
 409                         goto err_free_rt;
 410                 }
 411                 vlan_set_tci(skb, 0);
 412         }
 413
 414         skb_reset_inner_headers(skb);
 415
 416         __skb_push(skb, GENEVE_BASE_HLEN +
 417                         OVS_CB(skb)->egress_tun_info->options_len);
 418         skb_reset_transport_header(skb);
 419
 420         geneve_build_header(vport, skb);
 421
 422         /* Offloading */
 423         err = handle_offloads(skb);
 424         if (err)
 425                 goto err_free_rt;
 426
 427         df = tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 428
 429         sent_len = iptunnel_xmit(skb->sk, rt, skb,
 430                              saddr, tun_key->ipv4_dst,
 431                              IPPROTO_UDP, tun_key->ipv4_tos,
 432                              tun_key->ipv4_ttl,
 433                              df, false);
 434
 435         return sent_len > 0 ? sent_len + network_offset : sent_len;
 436
 437 err_free_rt:
 438         ip_rt_put(rt);
 439 error:
 440         return err;
 441 }
 442
 443 static const char *geneve_get_name(const struct vport *vport)
 444 {
 445         struct geneve_port *geneve_port = geneve_vport(vport);
 446         return geneve_port->name;
 447 }
 448
 449 static int geneve_get_egress_tun_info(struct vport *vport, struct sk_buff *skb,
 450                                       struct ovs_tunnel_info *egress_tun_info)
 451 {
 452         struct geneve_port *geneve_port = geneve_vport(vport);
 453         struct net *net = ovs_dp_get_net(vport->dp);
 454
 455         /*
 456          * Get tp_src and tp_dst, refert to geneve_build_header().
 457          */
 458         return ovs_tunnel_get_egress_info(egress_tun_info,
 459                                           ovs_dp_get_net(vport->dp),
 460                                           OVS_CB(skb)->egress_tun_info,
 461                                           IPPROTO_UDP, skb->mark,
 462                                           udp_flow_src_port(net, skb, 0, 0, true),
 463                                           inet_sport(geneve_port->sock->sk));
 464
 465 }
 466
 467 const struct vport_ops ovs_geneve_vport_ops = {
 468         .type                   = OVS_VPORT_TYPE_GENEVE,
 469         .create                 = geneve_tnl_create,
 470         .destroy                = geneve_tnl_destroy,
 471         .get_name               = geneve_get_name,
 472         .get_options            = geneve_get_options,
 473         .send                   = geneve_send,
 474         .get_egress_tun_info    = geneve_get_egress_tun_info,
 475 };