datapath/linux/compat/geneve.c

   1 /*
   2  * GENEVE: Generic Network Virtualization Encapsulation
   3  *
   4  * Copyright (c) 2015 Red Hat, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/hash.h>
  18 #include <linux/if_link.h>
  19
  20 #include <net/dst_metadata.h>
  21 #include <net/net_namespace.h>
  22 #include <net/netns/generic.h>
  23 #include <net/rtnetlink.h>
  24 #include <net/geneve.h>
  25 #include <net/protocol.h>
  26
  27 #include "gso.h"
  28 #include "vport-netdev.h"
  29 #include "compat.h"
  30
  31 #ifndef HAVE_METADATA_DST
  32 #define GENEVE_NETDEV_VER       "0.6"
  33
  34 #define GENEVE_UDP_PORT         6081
  35
  36 #define GENEVE_N_VID            (1u << 24)
  37 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  38
  39 #define VNI_HASH_BITS           10
  40 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  41
  42 #define GENEVE_VER 0
  43 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  44
  45 /* per-network namespace private data for this module */
  46 struct geneve_net {
  47         struct list_head        geneve_list;
  48         struct list_head        sock_list;
  49 };
  50
  51 static int geneve_net_id;
  52
  53 /* Pseudo network device */
  54 struct geneve_dev {
  55         struct hlist_node  hlist;       /* vni hash table */
  56         struct net         *net;        /* netns for packet i/o */
  57         struct net_device  *dev;        /* netdev for geneve tunnel */
  58         struct geneve_sock *sock;       /* socket used for geneve tunnel */
  59         u8                 vni[3];      /* virtual network ID for tunnel */
  60         u8                 ttl;         /* TTL override */
  61         u8                 tos;         /* TOS override */
  62         struct sockaddr_in remote;      /* IPv4 address for link partner */
  63         struct list_head   next;        /* geneve's per namespace list */
  64         __be16             dst_port;
  65         bool               collect_md;
  66 };
  67
  68 struct geneve_sock {
  69         bool                    collect_md;
  70         struct list_head        list;
  71         struct socket           *sock;
  72         struct rcu_head         rcu;
  73         int                     refcnt;
  74 #ifdef HAVE_UDP_OFFLOAD
  75         struct udp_offload      udp_offloads;
  76 #endif
  77         struct hlist_head       vni_list[VNI_HASH_SIZE];
  78 };
  79
  80 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  81 {
  82         __u32 vnid;
  83
  84         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  85         return hash_32(vnid, VNI_HASH_BITS);
  86 }
  87
  88 static __be64 vni_to_tunnel_id(const __u8 *vni)
  89 {
  90 #ifdef __BIG_ENDIAN
  91         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  92 #else
  93         return (__force __be64)(((__force u64)vni[0] << 40) |
  94                                 ((__force u64)vni[1] << 48) |
  95                                 ((__force u64)vni[2] << 56));
  96 #endif
  97 }
  98
  99 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 100                                         __be32 addr, u8 vni[])
 101 {
 102         struct hlist_head *vni_list_head;
 103         struct geneve_dev *geneve;
 104         __u32 hash;
 105
 106         /* Find the device for this VNI */
 107         hash = geneve_net_vni_hash(vni);
 108         vni_list_head = &gs->vni_list[hash];
 109         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 110                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 111                     addr == geneve->remote.sin_addr.s_addr)
 112                         return geneve;
 113         }
 114         return NULL;
 115 }
 116
 117 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 118 {
 119         return (struct genevehdr *)(udp_hdr(skb) + 1);
 120 }
 121
 122 /* geneve receive/decap routine */
 123 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
 124 {
 125         struct genevehdr *gnvh = geneve_hdr(skb);
 126         struct metadata_dst *tun_dst;
 127         struct geneve_dev *geneve = NULL;
 128 #ifdef HAVE_DEV_TSTATS
 129         struct pcpu_sw_netstats *stats;
 130 #endif
 131         struct iphdr *iph;
 132         u8 *vni;
 133         __be32 addr;
 134         int err;
 135         union {
 136                 struct metadata_dst dst;
 137                 char buf[sizeof(struct metadata_dst) + 256];
 138         } buf;
 139
 140         iph = ip_hdr(skb); /* outer IP header... */
 141
 142         if (gs->collect_md) {
 143                 static u8 zero_vni[3];
 144
 145                 vni = zero_vni;
 146                 addr = 0;
 147         } else {
 148                 vni = gnvh->vni;
 149                 addr = iph->saddr;
 150         }
 151
 152         geneve = geneve_lookup(gs, addr, vni);
 153         if (!geneve)
 154                 goto drop;
 155
 156         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 157                 __be16 flags;
 158
 159                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 160                         (gnvh->oam ? TUNNEL_OAM : 0) |
 161                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 162
 163                 tun_dst = &buf.dst;
 164                 ovs_udp_tun_rx_dst(&tun_dst->u.tun_info, skb, AF_INET, flags,
 165                                    vni_to_tunnel_id(gnvh->vni), gnvh->opt_len * 4);
 166                 /* Update tunnel dst according to Geneve options. */
 167                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 168                                         gnvh->options, gnvh->opt_len * 4);
 169         } else {
 170                 /* Drop packets w/ critical options,
 171                  * since we don't support any...
 172                  */
 173                 tun_dst = NULL;
 174                 if (gnvh->critical)
 175                         goto drop;
 176         }
 177
 178         skb_reset_mac_header(skb);
 179         skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
 180         skb->protocol = eth_type_trans(skb, geneve->dev);
 181         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 182
 183         if (tun_dst)
 184                 ovs_skb_dst_set(skb, &tun_dst->dst);
 185         else
 186                 goto drop;
 187         /* Ignore packet loops (and multicast echo) */
 188         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 189                 goto drop;
 190
 191         skb_reset_network_header(skb);
 192
 193         err = IP_ECN_decapsulate(iph, skb);
 194
 195         if (unlikely(err)) {
 196                 if (err > 1) {
 197                         ++geneve->dev->stats.rx_frame_errors;
 198                         ++geneve->dev->stats.rx_errors;
 199                         goto drop;
 200                 }
 201         }
 202
 203 #ifdef HAVE_DEV_TSTATS
 204         stats = this_cpu_ptr((struct pcpu_sw_netstats __percpu *)geneve->dev->tstats);
 205         u64_stats_update_begin(&stats->syncp);
 206         stats->rx_packets++;
 207         stats->rx_bytes += skb->len;
 208         u64_stats_update_end(&stats->syncp);
 209 #endif
 210         netdev_port_receive(skb, &tun_dst->u.tun_info);
 211         return;
 212 drop:
 213         /* Consume bad packet */
 214         kfree_skb(skb);
 215 }
 216
 217 #ifdef HAVE_DEV_TSTATS
 218 /* Setup stats when device is created */
 219 static int geneve_init(struct net_device *dev)
 220 {
 221         dev->tstats = (typeof(dev->tstats)) netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 222         if (!dev->tstats)
 223                 return -ENOMEM;
 224
 225         return 0;
 226 }
 227
 228 static void geneve_uninit(struct net_device *dev)
 229 {
 230         free_percpu(dev->tstats);
 231 }
 232 #endif
 233
 234 /* Callback from net/ipv4/udp.c to receive packets */
 235 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 236 {
 237         struct genevehdr *geneveh;
 238         struct geneve_sock *gs;
 239         int opts_len;
 240
 241         /* Need Geneve and inner Ethernet header to be present */
 242         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 243                 goto error;
 244
 245         /* Return packets with reserved bits set */
 246         geneveh = geneve_hdr(skb);
 247         if (unlikely(geneveh->ver != GENEVE_VER))
 248                 goto error;
 249
 250         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 251                 goto error;
 252
 253         opts_len = geneveh->opt_len * 4;
 254         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 255                                  htons(ETH_P_TEB)))
 256                 goto drop;
 257
 258         gs = rcu_dereference_sk_user_data(sk);
 259         if (!gs)
 260                 goto drop;
 261
 262         geneve_rx(gs, skb);
 263         return 0;
 264
 265 drop:
 266         /* Consume bad packet */
 267         kfree_skb(skb);
 268         return 0;
 269
 270 error:
 271         /* Let the UDP layer deal with the skb */
 272         return 1;
 273 }
 274
 275 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 276                                          __be16 port)
 277 {
 278         struct socket *sock;
 279         struct udp_port_cfg udp_conf;
 280         int err;
 281
 282         memset(&udp_conf, 0, sizeof(udp_conf));
 283
 284         if (ipv6) {
 285                 udp_conf.family = AF_INET6;
 286         } else {
 287                 udp_conf.family = AF_INET;
 288                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 289         }
 290
 291         udp_conf.local_udp_port = port;
 292
 293         /* Open UDP socket */
 294         err = udp_sock_create(net, &udp_conf, &sock);
 295         if (err < 0)
 296                 return ERR_PTR(err);
 297
 298         return sock;
 299 }
 300
 301 #ifdef HAVE_UDP_OFFLOAD
 302 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 303 {
 304         struct sock *sk = gs->sock->sk;
 305         sa_family_t sa_family = sk->sk_family;
 306         int err;
 307
 308         if (sa_family == AF_INET) {
 309                 err = udp_add_offload(&gs->udp_offloads);
 310                 if (err)
 311                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 312                                 err);
 313         }
 314 }
 315
 316 static int geneve_hlen(struct genevehdr *gh)
 317 {
 318         return sizeof(*gh) + gh->opt_len * 4;
 319 }
 320
 321 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
 322 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 323                                            struct sk_buff *skb)
 324 #else
 325 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 326                                            struct sk_buff *skb,
 327                                            struct udp_offload *uoff)
 328 #endif
 329 {
 330         struct sk_buff *p, **pp = NULL;
 331         struct genevehdr *gh, *gh2;
 332         unsigned int hlen, gh_len, off_gnv;
 333         const struct packet_offload *ptype;
 334         __be16 type;
 335         int flush = 1;
 336
 337         off_gnv = skb_gro_offset(skb);
 338         hlen = off_gnv + sizeof(*gh);
 339         gh = skb_gro_header_fast(skb, off_gnv);
 340         if (skb_gro_header_hard(skb, hlen)) {
 341                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 342                 if (unlikely(!gh))
 343                         goto out;
 344         }
 345
 346         if (gh->ver != GENEVE_VER || gh->oam)
 347                 goto out;
 348         gh_len = geneve_hlen(gh);
 349
 350         hlen = off_gnv + gh_len;
 351         if (skb_gro_header_hard(skb, hlen)) {
 352                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 353                 if (unlikely(!gh))
 354                         goto out;
 355         }
 356
 357         flush = 0;
 358
 359         for (p = *head; p; p = p->next) {
 360                 if (!NAPI_GRO_CB(p)->same_flow)
 361                         continue;
 362
 363                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 364                 if (gh->opt_len != gh2->opt_len ||
 365                     memcmp(gh, gh2, gh_len)) {
 366                         NAPI_GRO_CB(p)->same_flow = 0;
 367                         continue;
 368                 }
 369         }
 370
 371         type = gh->proto_type;
 372
 373         rcu_read_lock();
 374         ptype = gro_find_receive_by_type(type);
 375         if (!ptype) {
 376                 flush = 1;
 377                 goto out_unlock;
 378         }
 379
 380         skb_gro_pull(skb, gh_len);
 381         skb_gro_postpull_rcsum(skb, gh, gh_len);
 382         pp = ptype->callbacks.gro_receive(head, skb);
 383
 384 out_unlock:
 385         rcu_read_unlock();
 386 out:
 387         NAPI_GRO_CB(skb)->flush |= flush;
 388
 389         return pp;
 390 }
 391
 392 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
 393 static int geneve_gro_complete(struct sk_buff *skb, int nhoff)
 394 #else
 395 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 396                                struct udp_offload *uoff)
 397 #endif
 398 {
 399         struct genevehdr *gh;
 400         struct packet_offload *ptype;
 401         __be16 type;
 402         int gh_len;
 403         int err = -ENOSYS;
 404
 405         udp_tunnel_gro_complete(skb, nhoff);
 406
 407         gh = (struct genevehdr *)(skb->data + nhoff);
 408         gh_len = geneve_hlen(gh);
 409         type = gh->proto_type;
 410
 411         rcu_read_lock();
 412         ptype = gro_find_complete_by_type(type);
 413         if (ptype)
 414                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 415
 416         rcu_read_unlock();
 417         return err;
 418 }
 419 #endif
 420
 421 /* Create new listen socket if needed */
 422 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 423                                                 bool ipv6)
 424 {
 425         struct geneve_net *gn = net_generic(net, geneve_net_id);
 426         struct geneve_sock *gs;
 427         struct socket *sock;
 428         struct udp_tunnel_sock_cfg tunnel_cfg;
 429         int h;
 430
 431         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 432         if (!gs)
 433                 return ERR_PTR(-ENOMEM);
 434
 435         sock = geneve_create_sock(net, ipv6, port);
 436         if (IS_ERR(sock)) {
 437                 kfree(gs);
 438                 return ERR_CAST(sock);
 439         }
 440
 441         gs->sock = sock;
 442         gs->refcnt = 1;
 443         for (h = 0; h < VNI_HASH_SIZE; ++h)
 444                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 445
 446         /* Initialize the geneve udp offloads structure */
 447 #ifdef HAVE_UDP_OFFLOAD
 448         gs->udp_offloads.port = port;
 449         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 450         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 451         geneve_notify_add_rx_port(gs);
 452 #endif
 453         /* Mark socket as an encapsulation socket */
 454         tunnel_cfg.sk_user_data = gs;
 455         tunnel_cfg.encap_type = 1;
 456         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 457         tunnel_cfg.encap_destroy = NULL;
 458         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 459         list_add(&gs->list, &gn->sock_list);
 460         return gs;
 461 }
 462
 463 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 464 {
 465 #ifdef HAVE_UDP_OFFLOAD
 466         struct sock *sk = gs->sock->sk;
 467         sa_family_t sa_family = sk->sk_family;
 468
 469         if (sa_family == AF_INET)
 470                 udp_del_offload(&gs->udp_offloads);
 471 #endif
 472 }
 473
 474 static void free_gs_rcu(struct rcu_head *rcu)
 475 {
 476         struct geneve_sock *gs = container_of(rcu, struct geneve_sock, rcu);
 477
 478         kfree(gs);
 479 }
 480
 481 static void geneve_sock_release(struct geneve_sock *gs)
 482 {
 483         if (--gs->refcnt)
 484                 return;
 485
 486         list_del(&gs->list);
 487         geneve_notify_del_rx_port(gs);
 488         udp_tunnel_sock_release(gs->sock);
 489         call_rcu(&gs->rcu, free_gs_rcu);
 490 }
 491
 492 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 493                                             __be16 dst_port)
 494 {
 495         struct geneve_sock *gs;
 496
 497         list_for_each_entry(gs, &gn->sock_list, list) {
 498                 if (inet_sport(gs->sock->sk) == dst_port &&
 499                     inet_sk(gs->sock->sk)->sk.sk_family == AF_INET) {
 500                         return gs;
 501                 }
 502         }
 503         return NULL;
 504 }
 505
 506 static int geneve_open(struct net_device *dev)
 507 {
 508         struct geneve_dev *geneve = netdev_priv(dev);
 509         struct net *net = geneve->net;
 510         struct geneve_net *gn = net_generic(net, geneve_net_id);
 511         struct geneve_sock *gs;
 512         __u32 hash;
 513
 514         gs = geneve_find_sock(gn, geneve->dst_port);
 515         if (gs) {
 516                 gs->refcnt++;
 517                 goto out;
 518         }
 519
 520         gs = geneve_socket_create(net, geneve->dst_port, false);
 521         if (IS_ERR(gs))
 522                 return PTR_ERR(gs);
 523
 524 out:
 525         gs->collect_md = geneve->collect_md;
 526         geneve->sock = gs;
 527
 528         hash = geneve_net_vni_hash(geneve->vni);
 529         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 530         return 0;
 531 }
 532
 533 static int geneve_stop(struct net_device *dev)
 534 {
 535         struct geneve_dev *geneve = netdev_priv(dev);
 536         struct geneve_sock *gs = geneve->sock;
 537
 538         if (!hlist_unhashed(&geneve->hlist))
 539                 hlist_del_rcu(&geneve->hlist);
 540         geneve_sock_release(gs);
 541         return 0;
 542 }
 543
 544 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 545                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 546                             bool csum)
 547 {
 548         struct genevehdr *gnvh;
 549         int min_headroom;
 550         int err;
 551
 552         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 553                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr)
 554                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
 555         err = skb_cow_head(skb, min_headroom);
 556         if (unlikely(err)) {
 557                 kfree_skb(skb);
 558                 goto free_rt;
 559         }
 560
 561         skb = vlan_hwaccel_push_inside(skb);
 562         if (!skb) {
 563                 err = -ENOMEM;
 564                 goto free_rt;
 565         }
 566
 567         skb = udp_tunnel_handle_offloads(skb, csum, 0, false);
 568         if (IS_ERR(skb)) {
 569                 err = PTR_ERR(skb);
 570                 goto free_rt;
 571         }
 572         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 573         gnvh->ver = GENEVE_VER;
 574         gnvh->opt_len = opt_len / 4;
 575         gnvh->oam = !!(tun_flags & TUNNEL_OAM);
 576         gnvh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 577         gnvh->rsvd1 = 0;
 578         memcpy(gnvh->vni, vni, 3);
 579         gnvh->proto_type = htons(ETH_P_TEB);
 580         gnvh->rsvd2 = 0;
 581         memcpy(gnvh->options, opt, opt_len);
 582
 583         ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 584         return 0;
 585
 586 free_rt:
 587         ip_rt_put(rt);
 588         return err;
 589 }
 590
 591 static struct rtable *geneve_get_rt(struct sk_buff *skb,
 592                                     struct net_device *dev,
 593                                     struct flowi4 *fl4,
 594                                     struct ip_tunnel_info *info)
 595 {
 596         struct geneve_dev *geneve = netdev_priv(dev);
 597         struct rtable *rt = NULL;
 598         __u8 tos;
 599
 600         memset(fl4, 0, sizeof(*fl4));
 601         fl4->flowi4_mark = skb->mark;
 602         fl4->flowi4_proto = IPPROTO_UDP;
 603
 604         if (info) {
 605                 fl4->daddr = info->key.u.ipv4.dst;
 606                 fl4->saddr = info->key.u.ipv4.src;
 607                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 608         } else {
 609                 tos = geneve->tos;
 610                 if (tos == 1) {
 611                         const struct iphdr *iip = ip_hdr(skb);
 612
 613                         tos = ip_tunnel_get_dsfield(iip, skb);
 614                 }
 615
 616                 fl4->flowi4_tos = RT_TOS(tos);
 617                 fl4->daddr = geneve->remote.sin_addr.s_addr;
 618         }
 619
 620         rt = ip_route_output_key(geneve->net, fl4);
 621         if (IS_ERR(rt)) {
 622                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 623                 dev->stats.tx_carrier_errors++;
 624                 return rt;
 625         }
 626         if (rt_dst(rt).dev == dev) { /* is this necessary? */
 627                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 628                 dev->stats.collisions++;
 629                 ip_rt_put(rt);
 630                 return ERR_PTR(-EINVAL);
 631         }
 632         return rt;
 633 }
 634
 635 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 636 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 637 {
 638 #ifdef __BIG_ENDIAN
 639         vni[0] = (__force __u8)(tun_id >> 16);
 640         vni[1] = (__force __u8)(tun_id >> 8);
 641         vni[2] = (__force __u8)tun_id;
 642 #else
 643         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 644         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 645         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 646 #endif
 647 }
 648
 649 netdev_tx_t rpl_geneve_xmit(struct sk_buff *skb)
 650 {
 651         struct net_device *dev = skb->dev;
 652         struct geneve_dev *geneve = netdev_priv(dev);
 653         struct geneve_sock *gs = geneve->sock;
 654         struct ip_tunnel_info *info = NULL;
 655         struct rtable *rt = NULL;
 656         const struct iphdr *iip; /* interior IP header */
 657         struct flowi4 fl4;
 658         __u8 tos, ttl;
 659         __be16 sport;
 660         bool udp_csum;
 661         __be16 df;
 662         int err;
 663
 664         if (geneve->collect_md) {
 665                 info = skb_tunnel_info(skb);
 666                 if (unlikely(info && !(info->mode & IP_TUNNEL_INFO_TX))) {
 667                         netdev_dbg(dev, "no tunnel metadata\n");
 668                         goto tx_error;
 669                 }
 670                 if (info && ip_tunnel_info_af(info) != AF_INET)
 671                         goto tx_error;
 672         }
 673
 674         rt = geneve_get_rt(skb, dev, &fl4, info);
 675         if (IS_ERR(rt)) {
 676                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
 677                 dev->stats.tx_carrier_errors++;
 678                 goto tx_error;
 679         }
 680
 681         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 682         skb_reset_mac_header(skb);
 683
 684         iip = ip_hdr(skb);
 685
 686         if (info) {
 687                 const struct ip_tunnel_key *key = &info->key;
 688                 u8 *opts = NULL;
 689                 u8 vni[3];
 690
 691                 tunnel_id_to_vni(key->tun_id, vni);
 692                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 693                         opts = ip_tunnel_info_opts(info);
 694
 695                 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
 696                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 697                                        info->options_len, opts, udp_csum);
 698                 if (unlikely(err))
 699                         goto err;
 700
 701                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 702                 ttl = key->ttl;
 703                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 704         } else {
 705                 udp_csum = false;
 706                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 707                                        0, NULL, udp_csum);
 708                 if (unlikely(err))
 709                         goto err;
 710
 711                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 712                 ttl = geneve->ttl;
 713                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 714                         ttl = 1;
 715                 ttl = ttl ? : ip4_dst_hoplimit(&rt_dst(rt));
 716                 df = 0;
 717         }
 718         err = udp_tunnel_xmit_skb(rt, gs->sock->sk, skb, fl4.saddr, fl4.daddr,
 719                                   tos, ttl, df, sport, geneve->dst_port,
 720                                   !net_eq(geneve->net, dev_net(geneve->dev)),
 721                                   !udp_csum);
 722
 723         iptunnel_xmit_stats(err, &dev->stats, (struct pcpu_sw_netstats __percpu *) dev->tstats);
 724         return NETDEV_TX_OK;
 725
 726 tx_error:
 727         dev_kfree_skb(skb);
 728 err:
 729         dev->stats.tx_errors++;
 730         return NETDEV_TX_OK;
 731 }
 732 EXPORT_SYMBOL(rpl_geneve_xmit);
 733
 734 static netdev_tx_t geneve_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 735 {
 736         /* Drop All packets coming from networking stack. OVS-CB is
 737          * not initialized for these packets.
 738          */
 739
 740         dev_kfree_skb(skb);
 741         dev->stats.tx_dropped++;
 742         return NETDEV_TX_OK;
 743 }
 744
 745 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
 746 {
 747         /* GENEVE overhead is not fixed, so we can't enforce a more
 748          * precise max MTU.
 749          */
 750         if (new_mtu < 68 || new_mtu > IP_MAX_MTU)
 751                 return -EINVAL;
 752         dev->mtu = new_mtu;
 753         return 0;
 754 }
 755
 756 static const struct net_device_ops geneve_netdev_ops = {
 757 #ifdef HAVE_DEV_TSTATS
 758         .ndo_init               = geneve_init,
 759         .ndo_uninit             = geneve_uninit,
 760         .ndo_get_stats64        = ip_tunnel_get_stats64,
 761 #endif
 762         .ndo_open               = geneve_open,
 763         .ndo_stop               = geneve_stop,
 764         .ndo_start_xmit         = geneve_dev_xmit,
 765         .ndo_change_mtu         = geneve_change_mtu,
 766         .ndo_validate_addr      = eth_validate_addr,
 767         .ndo_set_mac_address    = eth_mac_addr,
 768 };
 769
 770 static void geneve_get_drvinfo(struct net_device *dev,
 771                                struct ethtool_drvinfo *drvinfo)
 772 {
 773         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
 774         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
 775 }
 776
 777 static const struct ethtool_ops geneve_ethtool_ops = {
 778         .get_drvinfo    = geneve_get_drvinfo,
 779         .get_link       = ethtool_op_get_link,
 780 };
 781
 782 /* Info for udev, that this is a virtual tunnel endpoint */
 783 static struct device_type geneve_type = {
 784         .name = "geneve",
 785 };
 786
 787 /* Initialize the device structure. */
 788 static void geneve_setup(struct net_device *dev)
 789 {
 790         ether_setup(dev);
 791
 792         dev->netdev_ops = &geneve_netdev_ops;
 793         dev->ethtool_ops = &geneve_ethtool_ops;
 794         dev->destructor = free_netdev;
 795
 796         SET_NETDEV_DEVTYPE(dev, &geneve_type);
 797
 798         dev->features    |= NETIF_F_LLTX;
 799         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
 800         dev->features    |= NETIF_F_RXCSUM;
 801         dev->features    |= NETIF_F_GSO_SOFTWARE;
 802
 803 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
 804         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
 805         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 806 #endif
 807 #if 0
 808         /* Not required */
 809         netif_keep_dst(dev);
 810 #endif
 811         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
 812         eth_hw_addr_random(dev);
 813 }
 814
 815 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
 816         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
 817         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 818         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
 819         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
 820         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
 821         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
 822 };
 823
 824 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
 825 {
 826         if (tb[IFLA_ADDRESS]) {
 827                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 828                         return -EINVAL;
 829
 830                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 831                         return -EADDRNOTAVAIL;
 832         }
 833
 834         if (!data)
 835                 return -EINVAL;
 836
 837         if (data[IFLA_GENEVE_ID]) {
 838                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
 839
 840                 if (vni >= GENEVE_VID_MASK)
 841                         return -ERANGE;
 842         }
 843
 844         return 0;
 845 }
 846
 847 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
 848                                           __be16 dst_port,
 849                                           __be32 rem_addr,
 850                                           u8 vni[],
 851                                           bool *tun_on_same_port,
 852                                           bool *tun_collect_md)
 853 {
 854         struct geneve_dev *geneve, *t;
 855
 856         *tun_on_same_port = false;
 857         *tun_collect_md = false;
 858         t = NULL;
 859         list_for_each_entry(geneve, &gn->geneve_list, next) {
 860                 if (geneve->dst_port == dst_port) {
 861                         *tun_collect_md = geneve->collect_md;
 862                         *tun_on_same_port = true;
 863                 }
 864                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 865                     rem_addr == geneve->remote.sin_addr.s_addr &&
 866                     dst_port == geneve->dst_port)
 867                         t = geneve;
 868         }
 869         return t;
 870 }
 871
 872 static int geneve_configure(struct net *net, struct net_device *dev,
 873                             __be32 rem_addr, __u32 vni, __u8 ttl, __u8 tos,
 874                             __be16 dst_port, bool metadata)
 875 {
 876         struct geneve_net *gn = net_generic(net, geneve_net_id);
 877         struct geneve_dev *t, *geneve = netdev_priv(dev);
 878         bool tun_collect_md, tun_on_same_port;
 879         int err;
 880
 881         if (metadata) {
 882                 if (rem_addr || vni || tos || ttl)
 883                         return -EINVAL;
 884         }
 885
 886         geneve->net = net;
 887         geneve->dev = dev;
 888
 889         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
 890         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
 891         geneve->vni[2] =  vni & 0x000000ff;
 892
 893         geneve->remote.sin_addr.s_addr = rem_addr;
 894         if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
 895                 return -EINVAL;
 896
 897         geneve->ttl = ttl;
 898         geneve->tos = tos;
 899         geneve->dst_port = dst_port;
 900         geneve->collect_md = metadata;
 901
 902         t = geneve_find_dev(gn, dst_port, rem_addr, geneve->vni,
 903                             &tun_on_same_port, &tun_collect_md);
 904         if (t)
 905                 return -EBUSY;
 906
 907         if (metadata) {
 908                 if (tun_on_same_port)
 909                         return -EPERM;
 910         } else {
 911                 if (tun_collect_md)
 912                         return -EPERM;
 913         }
 914
 915         err = register_netdevice(dev);
 916         if (err)
 917                 return err;
 918
 919         list_add(&geneve->next, &gn->geneve_list);
 920         return 0;
 921 }
 922
 923 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 924 static int geneve_newlink(struct net_device *dev,
 925                           struct nlattr *tb[], struct nlattr *data[])
 926 {
 927         struct net *net = &init_net;
 928 #else
 929 static int geneve_newlink(struct net *net, struct net_device *dev,
 930                           struct nlattr *tb[], struct nlattr *data[])
 931 {
 932 #endif
 933         __be16 dst_port = htons(GENEVE_UDP_PORT);
 934         __u8 ttl = 0, tos = 0;
 935         bool metadata = false;
 936         __be32 rem_addr;
 937         __u32 vni;
 938
 939         if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
 940                 return -EINVAL;
 941
 942         vni = nla_get_u32(data[IFLA_GENEVE_ID]);
 943         rem_addr = nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
 944
 945         if (data[IFLA_GENEVE_TTL])
 946                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
 947
 948         if (data[IFLA_GENEVE_TOS])
 949                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
 950
 951         if (data[IFLA_GENEVE_PORT])
 952                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
 953
 954         if (data[IFLA_GENEVE_COLLECT_METADATA])
 955                 metadata = true;
 956
 957         return geneve_configure(net, dev, rem_addr, vni,
 958                                 ttl, tos, dst_port, metadata);
 959 }
 960
 961 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 962 static void geneve_dellink(struct net_device *dev)
 963 #else
 964 static void geneve_dellink(struct net_device *dev, struct list_head *head)
 965 #endif
 966 {
 967         struct geneve_dev *geneve = netdev_priv(dev);
 968
 969         list_del(&geneve->next);
 970         unregister_netdevice_queue(dev, head);
 971 }
 972
 973 static size_t geneve_get_size(const struct net_device *dev)
 974 {
 975         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
 976                 nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
 977                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
 978                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
 979                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
 980                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
 981                 0;
 982 }
 983
 984 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
 985 {
 986         struct geneve_dev *geneve = netdev_priv(dev);
 987         __u32 vni;
 988
 989         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
 990         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
 991                 goto nla_put_failure;
 992
 993         if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
 994                             geneve->remote.sin_addr.s_addr))
 995                 goto nla_put_failure;
 996
 997         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
 998             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
 999                 goto nla_put_failure;
1000
1001         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1002                 goto nla_put_failure;
1003
1004         if (geneve->collect_md) {
1005                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1006                         goto nla_put_failure;
1007         }
1008
1009         return 0;
1010
1011 nla_put_failure:
1012         return -EMSGSIZE;
1013 }
1014
1015 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1016         .kind           = "ovs_geneve",
1017         .maxtype        = IFLA_GENEVE_MAX,
1018         .policy         = geneve_policy,
1019         .priv_size      = sizeof(struct geneve_dev),
1020         .setup          = geneve_setup,
1021         .validate       = geneve_validate,
1022         .newlink        = geneve_newlink,
1023         .dellink        = geneve_dellink,
1024         .get_size       = geneve_get_size,
1025         .fill_info      = geneve_fill_info,
1026 };
1027
1028 struct net_device *rpl_geneve_dev_create_fb(struct net *net, const char *name,
1029                                         u8 name_assign_type, u16 dst_port)
1030 {
1031         struct nlattr *tb[IFLA_MAX + 1];
1032         struct net_device *dev;
1033         int err;
1034
1035         memset(tb, 0, sizeof(tb));
1036         dev = rtnl_create_link(net, (char *) name, name_assign_type,
1037                                &geneve_link_ops, tb);
1038         if (IS_ERR(dev))
1039                 return dev;
1040
1041         err = geneve_configure(net, dev, 0, 0, 0, 0, htons(dst_port), true);
1042         if (err) {
1043                 free_netdev(dev);
1044                 return ERR_PTR(err);
1045         }
1046         return dev;
1047 }
1048 EXPORT_SYMBOL_GPL(rpl_geneve_dev_create_fb);
1049
1050 static __net_init int geneve_init_net(struct net *net)
1051 {
1052         struct geneve_net *gn = net_generic(net, geneve_net_id);
1053
1054         INIT_LIST_HEAD(&gn->geneve_list);
1055         INIT_LIST_HEAD(&gn->sock_list);
1056         return 0;
1057 }
1058
1059 static void __net_exit geneve_exit_net(struct net *net)
1060 {
1061         struct geneve_net *gn = net_generic(net, geneve_net_id);
1062         struct geneve_dev *geneve, *next;
1063         struct net_device *dev, *aux;
1064         LIST_HEAD(list);
1065
1066         rtnl_lock();
1067
1068         /* gather any geneve devices that were moved into this ns */
1069         for_each_netdev_safe(net, dev, aux)
1070                 if (dev->rtnl_link_ops == &geneve_link_ops)
1071                         unregister_netdevice_queue(dev, &list);
1072
1073         /* now gather any other geneve devices that were created in this ns */
1074         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1075                 /* If geneve->dev is in the same netns, it was already added
1076                  * to the list by the previous loop.
1077                  */
1078                 if (!net_eq(dev_net(geneve->dev), net))
1079                         unregister_netdevice_queue(geneve->dev, &list);
1080         }
1081
1082         /* unregister the devices gathered above */
1083         unregister_netdevice_many(&list);
1084         rtnl_unlock();
1085 }
1086
1087 static struct pernet_operations geneve_net_ops = {
1088         .init = geneve_init_net,
1089         .exit = geneve_exit_net,
1090         .id   = &geneve_net_id,
1091         .size = sizeof(struct geneve_net),
1092 };
1093
1094 DEFINE_COMPAT_PNET_REG_FUNC(device)
1095 int rpl_geneve_init_module(void)
1096 {
1097         int rc;
1098
1099         rc = register_pernet_subsys(&geneve_net_ops);
1100         if (rc)
1101                 goto out1;
1102
1103         rc = rtnl_link_register(&geneve_link_ops);
1104         if (rc)
1105                 goto out2;
1106
1107         pr_info("Geneve tunneling driver\n");
1108         return 0;
1109 out2:
1110         unregister_pernet_subsys(&geneve_net_ops);
1111 out1:
1112         return rc;
1113 }
1114
1115 void rpl_geneve_cleanup_module(void)
1116 {
1117         rtnl_link_unregister(&geneve_link_ops);
1118         unregister_pernet_subsys(&geneve_net_ops);
1119 }
1120 #endif