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 #include <linux/if_vlan.h>
  20
  21 #include <net/dst_metadata.h>
  22 #include <net/net_namespace.h>
  23 #include <net/netns/generic.h>
  24 #include <net/rtnetlink.h>
  25 #include <net/geneve.h>
  26 #include <net/protocol.h>
  27
  28 #include "gso.h"
  29 #include "vport-netdev.h"
  30 #include "compat.h"
  31
  32 #ifndef HAVE_METADATA_DST
  33 #define GENEVE_NETDEV_VER       "0.6"
  34
  35 #define GENEVE_UDP_PORT         6081
  36
  37 #define GENEVE_N_VID            (1u << 24)
  38 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  39
  40 #define VNI_HASH_BITS           10
  41 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  42
  43 #define GENEVE_VER 0
  44 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  45
  46 /* per-network namespace private data for this module */
  47 struct geneve_net {
  48         struct list_head        geneve_list;
  49         struct list_head        sock_list;
  50 };
  51
  52 static int geneve_net_id;
  53
  54 /* Pseudo network device */
  55 struct geneve_dev {
  56         struct hlist_node  hlist;       /* vni hash table */
  57         struct net         *net;        /* netns for packet i/o */
  58         struct net_device  *dev;        /* netdev for geneve tunnel */
  59         struct geneve_sock *sock;       /* socket used for geneve tunnel */
  60         u8                 vni[3];      /* virtual network ID for tunnel */
  61         u8                 ttl;         /* TTL override */
  62         u8                 tos;         /* TOS override */
  63         struct sockaddr_in remote;      /* IPv4 address for link partner */
  64         struct list_head   next;        /* geneve's per namespace list */
  65         __be16             dst_port;
  66         bool               collect_md;
  67 };
  68
  69 struct geneve_sock {
  70         bool                    collect_md;
  71         struct list_head        list;
  72         struct socket           *sock;
  73         struct rcu_head         rcu;
  74         int                     refcnt;
  75 #ifdef HAVE_UDP_OFFLOAD
  76         struct udp_offload      udp_offloads;
  77 #endif
  78         struct hlist_head       vni_list[VNI_HASH_SIZE];
  79 };
  80
  81 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  82 {
  83         __u32 vnid;
  84
  85         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  86         return hash_32(vnid, VNI_HASH_BITS);
  87 }
  88
  89 static __be64 vni_to_tunnel_id(const __u8 *vni)
  90 {
  91 #ifdef __BIG_ENDIAN
  92         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  93 #else
  94         return (__force __be64)(((__force u64)vni[0] << 40) |
  95                                 ((__force u64)vni[1] << 48) |
  96                                 ((__force u64)vni[2] << 56));
  97 #endif
  98 }
  99
 100 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 101                                         __be32 addr, u8 vni[])
 102 {
 103         struct hlist_head *vni_list_head;
 104         struct geneve_dev *geneve;
 105         __u32 hash;
 106
 107         /* Find the device for this VNI */
 108         hash = geneve_net_vni_hash(vni);
 109         vni_list_head = &gs->vni_list[hash];
 110         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 111                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 112                     addr == geneve->remote.sin_addr.s_addr)
 113                         return geneve;
 114         }
 115         return NULL;
 116 }
 117
 118 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 119 {
 120         return (struct genevehdr *)(udp_hdr(skb) + 1);
 121 }
 122
 123 /* geneve receive/decap routine */
 124 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
 125 {
 126         struct genevehdr *gnvh = geneve_hdr(skb);
 127         struct metadata_dst *tun_dst;
 128         struct geneve_dev *geneve = NULL;
 129         struct pcpu_sw_netstats *stats;
 130         struct iphdr *iph;
 131         u8 *vni;
 132         __be32 addr;
 133         int err;
 134         union {
 135                 struct metadata_dst dst;
 136                 char buf[sizeof(struct metadata_dst) + 256];
 137         } buf;
 138
 139         iph = ip_hdr(skb); /* outer IP header... */
 140
 141         if (gs->collect_md) {
 142                 static u8 zero_vni[3];
 143
 144                 vni = zero_vni;
 145                 addr = 0;
 146         } else {
 147                 vni = gnvh->vni;
 148                 addr = iph->saddr;
 149         }
 150
 151         geneve = geneve_lookup(gs, addr, vni);
 152         if (!geneve)
 153                 goto drop;
 154
 155         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 156                 __be16 flags;
 157
 158                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 159                         (gnvh->oam ? TUNNEL_OAM : 0) |
 160                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 161
 162                 tun_dst = &buf.dst;
 163                 ovs_udp_tun_rx_dst(&tun_dst->u.tun_info, skb, AF_INET, flags,
 164                                    vni_to_tunnel_id(gnvh->vni), gnvh->opt_len * 4);
 165                 /* Update tunnel dst according to Geneve options. */
 166                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 167                                         gnvh->options, gnvh->opt_len * 4);
 168         } else {
 169                 /* Drop packets w/ critical options,
 170                  * since we don't support any...
 171                  */
 172                 tun_dst = NULL;
 173                 if (gnvh->critical)
 174                         goto drop;
 175         }
 176
 177         skb_reset_mac_header(skb);
 178         skb->protocol = eth_type_trans(skb, geneve->dev);
 179         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 180
 181         if (tun_dst)
 182                 ovs_skb_dst_set(skb, &tun_dst->dst);
 183         else
 184                 goto drop;
 185         /* Ignore packet loops (and multicast echo) */
 186         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 187                 goto drop;
 188
 189         skb_reset_network_header(skb);
 190
 191         err = IP_ECN_decapsulate(iph, skb);
 192
 193         if (unlikely(err)) {
 194                 if (err > 1) {
 195                         ++geneve->dev->stats.rx_frame_errors;
 196                         ++geneve->dev->stats.rx_errors;
 197                         goto drop;
 198                 }
 199         }
 200
 201         stats = this_cpu_ptr((struct pcpu_sw_netstats __percpu *)geneve->dev->tstats);
 202         u64_stats_update_begin(&stats->syncp);
 203         stats->rx_packets++;
 204         stats->rx_bytes += skb->len;
 205         u64_stats_update_end(&stats->syncp);
 206         netdev_port_receive(skb, &tun_dst->u.tun_info);
 207         return;
 208 drop:
 209         /* Consume bad packet */
 210         kfree_skb(skb);
 211 }
 212
 213 /* Setup stats when device is created */
 214 static int geneve_init(struct net_device *dev)
 215 {
 216         dev->tstats = (typeof(dev->tstats)) netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 217         if (!dev->tstats)
 218                 return -ENOMEM;
 219
 220         return 0;
 221 }
 222
 223 static void geneve_uninit(struct net_device *dev)
 224 {
 225         free_percpu(dev->tstats);
 226 }
 227
 228 /* Callback from net/ipv4/udp.c to receive packets */
 229 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 230 {
 231         struct genevehdr *geneveh;
 232         struct geneve_sock *gs;
 233         int opts_len;
 234
 235         /* Need Geneve and inner Ethernet header to be present */
 236         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 237                 goto error;
 238
 239         /* Return packets with reserved bits set */
 240         geneveh = geneve_hdr(skb);
 241         if (unlikely(geneveh->ver != GENEVE_VER))
 242                 goto error;
 243
 244         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 245                 goto error;
 246
 247         opts_len = geneveh->opt_len * 4;
 248         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 249                                  htons(ETH_P_TEB), false))
 250                 goto drop;
 251
 252         gs = rcu_dereference_sk_user_data(sk);
 253         if (!gs)
 254                 goto drop;
 255
 256         geneve_rx(gs, skb);
 257         return 0;
 258
 259 drop:
 260         /* Consume bad packet */
 261         kfree_skb(skb);
 262         return 0;
 263
 264 error:
 265         /* Let the UDP layer deal with the skb */
 266         return 1;
 267 }
 268
 269 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 270                                          __be16 port)
 271 {
 272         struct socket *sock;
 273         struct udp_port_cfg udp_conf;
 274         int err;
 275
 276         memset(&udp_conf, 0, sizeof(udp_conf));
 277
 278         if (ipv6) {
 279                 udp_conf.family = AF_INET6;
 280         } else {
 281                 udp_conf.family = AF_INET;
 282                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 283         }
 284
 285         udp_conf.local_udp_port = port;
 286
 287         /* Open UDP socket */
 288         err = udp_sock_create(net, &udp_conf, &sock);
 289         if (err < 0)
 290                 return ERR_PTR(err);
 291
 292         return sock;
 293 }
 294
 295 #ifdef HAVE_UDP_OFFLOAD
 296 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 297 {
 298         struct sock *sk = gs->sock->sk;
 299         sa_family_t sa_family = sk->sk_family;
 300         int err;
 301
 302         if (sa_family == AF_INET) {
 303                 err = udp_add_offload(&gs->udp_offloads);
 304                 if (err)
 305                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 306                                 err);
 307         }
 308 }
 309
 310 static int geneve_hlen(struct genevehdr *gh)
 311 {
 312         return sizeof(*gh) + gh->opt_len * 4;
 313 }
 314
 315 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
 316 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 317                                            struct sk_buff *skb)
 318 #else
 319 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 320                                            struct sk_buff *skb,
 321                                            struct udp_offload *uoff)
 322 #endif
 323 {
 324         struct sk_buff *p, **pp = NULL;
 325         struct genevehdr *gh, *gh2;
 326         unsigned int hlen, gh_len, off_gnv;
 327         const struct packet_offload *ptype;
 328         __be16 type;
 329         int flush = 1;
 330
 331         off_gnv = skb_gro_offset(skb);
 332         hlen = off_gnv + sizeof(*gh);
 333         gh = skb_gro_header_fast(skb, off_gnv);
 334         if (skb_gro_header_hard(skb, hlen)) {
 335                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 336                 if (unlikely(!gh))
 337                         goto out;
 338         }
 339
 340         if (gh->ver != GENEVE_VER || gh->oam)
 341                 goto out;
 342         gh_len = geneve_hlen(gh);
 343
 344         hlen = off_gnv + gh_len;
 345         if (skb_gro_header_hard(skb, hlen)) {
 346                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 347                 if (unlikely(!gh))
 348                         goto out;
 349         }
 350
 351         flush = 0;
 352
 353         for (p = *head; p; p = p->next) {
 354                 if (!NAPI_GRO_CB(p)->same_flow)
 355                         continue;
 356
 357                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 358                 if (gh->opt_len != gh2->opt_len ||
 359                     memcmp(gh, gh2, gh_len)) {
 360                         NAPI_GRO_CB(p)->same_flow = 0;
 361                         continue;
 362                 }
 363         }
 364
 365         type = gh->proto_type;
 366
 367         rcu_read_lock();
 368         ptype = gro_find_receive_by_type(type);
 369         if (!ptype) {
 370                 flush = 1;
 371                 goto out_unlock;
 372         }
 373
 374         skb_gro_pull(skb, gh_len);
 375         skb_gro_postpull_rcsum(skb, gh, gh_len);
 376         pp = ptype->callbacks.gro_receive(head, skb);
 377
 378 out_unlock:
 379         rcu_read_unlock();
 380 out:
 381         NAPI_GRO_CB(skb)->flush |= flush;
 382
 383         return pp;
 384 }
 385
 386 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
 387 static int geneve_gro_complete(struct sk_buff *skb, int nhoff)
 388 #else
 389 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 390                                struct udp_offload *uoff)
 391 #endif
 392 {
 393         struct genevehdr *gh;
 394         struct packet_offload *ptype;
 395         __be16 type;
 396         int gh_len;
 397         int err = -ENOSYS;
 398
 399         udp_tunnel_gro_complete(skb, nhoff);
 400
 401         gh = (struct genevehdr *)(skb->data + nhoff);
 402         gh_len = geneve_hlen(gh);
 403         type = gh->proto_type;
 404
 405         rcu_read_lock();
 406         ptype = gro_find_complete_by_type(type);
 407         if (ptype)
 408                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 409
 410         rcu_read_unlock();
 411         return err;
 412 }
 413 #endif
 414
 415 /* Create new listen socket if needed */
 416 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 417                                                 bool ipv6)
 418 {
 419         struct geneve_net *gn = net_generic(net, geneve_net_id);
 420         struct geneve_sock *gs;
 421         struct socket *sock;
 422         struct udp_tunnel_sock_cfg tunnel_cfg;
 423         int h;
 424
 425         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 426         if (!gs)
 427                 return ERR_PTR(-ENOMEM);
 428
 429         sock = geneve_create_sock(net, ipv6, port);
 430         if (IS_ERR(sock)) {
 431                 kfree(gs);
 432                 return ERR_CAST(sock);
 433         }
 434
 435         gs->sock = sock;
 436         gs->refcnt = 1;
 437         for (h = 0; h < VNI_HASH_SIZE; ++h)
 438                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 439
 440         /* Initialize the geneve udp offloads structure */
 441 #ifdef HAVE_UDP_OFFLOAD
 442         gs->udp_offloads.port = port;
 443         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 444         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 445         geneve_notify_add_rx_port(gs);
 446 #endif
 447         /* Mark socket as an encapsulation socket */
 448         tunnel_cfg.sk_user_data = gs;
 449         tunnel_cfg.encap_type = 1;
 450         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 451         tunnel_cfg.encap_destroy = NULL;
 452         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 453         list_add(&gs->list, &gn->sock_list);
 454         return gs;
 455 }
 456
 457 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 458 {
 459 #ifdef HAVE_UDP_OFFLOAD
 460         struct sock *sk = gs->sock->sk;
 461         sa_family_t sa_family = sk->sk_family;
 462
 463         if (sa_family == AF_INET)
 464                 udp_del_offload(&gs->udp_offloads);
 465 #endif
 466 }
 467
 468 static void free_gs_rcu(struct rcu_head *rcu)
 469 {
 470         struct geneve_sock *gs = container_of(rcu, struct geneve_sock, rcu);
 471
 472         kfree(gs);
 473 }
 474
 475 static void geneve_sock_release(struct geneve_sock *gs)
 476 {
 477         if (--gs->refcnt)
 478                 return;
 479
 480         list_del(&gs->list);
 481         geneve_notify_del_rx_port(gs);
 482         udp_tunnel_sock_release(gs->sock);
 483         call_rcu(&gs->rcu, free_gs_rcu);
 484 }
 485
 486 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 487                                             __be16 dst_port)
 488 {
 489         struct geneve_sock *gs;
 490
 491         list_for_each_entry(gs, &gn->sock_list, list) {
 492                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 493                     inet_sk(gs->sock->sk)->sk.sk_family == AF_INET) {
 494                         return gs;
 495                 }
 496         }
 497         return NULL;
 498 }
 499
 500 static int geneve_open(struct net_device *dev)
 501 {
 502         struct geneve_dev *geneve = netdev_priv(dev);
 503         struct net *net = geneve->net;
 504         struct geneve_net *gn = net_generic(net, geneve_net_id);
 505         struct geneve_sock *gs;
 506         __u32 hash;
 507
 508         gs = geneve_find_sock(gn, geneve->dst_port);
 509         if (gs) {
 510                 gs->refcnt++;
 511                 goto out;
 512         }
 513
 514         gs = geneve_socket_create(net, geneve->dst_port, false);
 515         if (IS_ERR(gs))
 516                 return PTR_ERR(gs);
 517
 518 out:
 519         gs->collect_md = geneve->collect_md;
 520         geneve->sock = gs;
 521
 522         hash = geneve_net_vni_hash(geneve->vni);
 523         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 524         return 0;
 525 }
 526
 527 static int geneve_stop(struct net_device *dev)
 528 {
 529         struct geneve_dev *geneve = netdev_priv(dev);
 530         struct geneve_sock *gs = geneve->sock;
 531
 532         if (!hlist_unhashed(&geneve->hlist))
 533                 hlist_del_rcu(&geneve->hlist);
 534         geneve_sock_release(gs);
 535         return 0;
 536 }
 537
 538 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 539                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 540                             bool csum)
 541 {
 542         struct genevehdr *gnvh;
 543         int min_headroom;
 544         int err;
 545
 546         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 547                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr)
 548                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
 549         err = skb_cow_head(skb, min_headroom);
 550         if (unlikely(err))
 551                 goto free_rt;
 552
 553         skb = vlan_hwaccel_push_inside(skb);
 554         if (!skb) {
 555                 err = -ENOMEM;
 556                 goto free_rt;
 557         }
 558
 559         err = udp_tunnel_handle_offloads(skb, csum, false);
 560         if (err)
 561                 goto free_rt;
 562         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 563         gnvh->ver = GENEVE_VER;
 564         gnvh->opt_len = opt_len / 4;
 565         gnvh->oam = !!(tun_flags & TUNNEL_OAM);
 566         gnvh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 567         gnvh->rsvd1 = 0;
 568         memcpy(gnvh->vni, vni, 3);
 569         gnvh->proto_type = htons(ETH_P_TEB);
 570         gnvh->rsvd2 = 0;
 571         memcpy(gnvh->options, opt, opt_len);
 572
 573         ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 574         return 0;
 575
 576 free_rt:
 577         ip_rt_put(rt);
 578         return err;
 579 }
 580
 581 static struct rtable *geneve_get_rt(struct sk_buff *skb,
 582                                     struct net_device *dev,
 583                                     struct flowi4 *fl4,
 584                                     struct ip_tunnel_info *info)
 585 {
 586         struct geneve_dev *geneve = netdev_priv(dev);
 587         struct rtable *rt = NULL;
 588         __u8 tos;
 589
 590         memset(fl4, 0, sizeof(*fl4));
 591         fl4->flowi4_mark = skb->mark;
 592         fl4->flowi4_proto = IPPROTO_UDP;
 593
 594         if (info) {
 595                 fl4->daddr = info->key.u.ipv4.dst;
 596                 fl4->saddr = info->key.u.ipv4.src;
 597                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 598         } else {
 599                 tos = geneve->tos;
 600                 if (tos == 1) {
 601                         const struct iphdr *iip = ip_hdr(skb);
 602
 603                         tos = ip_tunnel_get_dsfield(iip, skb);
 604                 }
 605
 606                 fl4->flowi4_tos = RT_TOS(tos);
 607                 fl4->daddr = geneve->remote.sin_addr.s_addr;
 608         }
 609
 610         rt = ip_route_output_key(geneve->net, fl4);
 611         if (IS_ERR(rt)) {
 612                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 613                 return ERR_PTR(-ENETUNREACH);
 614         }
 615         if (rt->dst.dev == dev) { /* is this necessary? */
 616                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 617                 ip_rt_put(rt);
 618                 return ERR_PTR(-ELOOP);
 619         }
 620         return rt;
 621 }
 622
 623 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 624 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 625 {
 626 #ifdef __BIG_ENDIAN
 627         vni[0] = (__force __u8)(tun_id >> 16);
 628         vni[1] = (__force __u8)(tun_id >> 8);
 629         vni[2] = (__force __u8)tun_id;
 630 #else
 631         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 632         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 633         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 634 #endif
 635 }
 636
 637 netdev_tx_t rpl_geneve_xmit(struct sk_buff *skb)
 638 {
 639         struct net_device *dev = skb->dev;
 640         struct geneve_dev *geneve = netdev_priv(dev);
 641         struct geneve_sock *gs = geneve->sock;
 642         struct ip_tunnel_info *info = NULL;
 643         struct rtable *rt = NULL;
 644         const struct iphdr *iip; /* interior IP header */
 645         int err = -EINVAL;
 646         struct flowi4 fl4;
 647         __u8 tos, ttl;
 648         __be16 sport;
 649         bool udp_csum;
 650         __be16 df;
 651
 652         if (geneve->collect_md) {
 653                 info = skb_tunnel_info(skb);
 654                 if (unlikely(info && !(info->mode & IP_TUNNEL_INFO_TX))) {
 655                         netdev_dbg(dev, "no tunnel metadata\n");
 656                         goto tx_error;
 657                 }
 658                 if (info && ip_tunnel_info_af(info) != AF_INET)
 659                         goto tx_error;
 660         }
 661
 662         rt = geneve_get_rt(skb, dev, &fl4, info);
 663         if (IS_ERR(rt)) {
 664                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
 665                 err = PTR_ERR(rt);
 666                 goto tx_error;
 667         }
 668
 669         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 670         skb_reset_mac_header(skb);
 671
 672         iip = ip_hdr(skb);
 673
 674         if (info) {
 675                 const struct ip_tunnel_key *key = &info->key;
 676                 u8 *opts = NULL;
 677                 u8 vni[3];
 678
 679                 tunnel_id_to_vni(key->tun_id, vni);
 680                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 681                         opts = ip_tunnel_info_opts(info);
 682
 683                 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
 684                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 685                                        info->options_len, opts, udp_csum);
 686                 if (unlikely(err))
 687                         goto tx_error;
 688
 689                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 690                 ttl = key->ttl;
 691                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 692         } else {
 693                 udp_csum = false;
 694                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 695                                        0, NULL, udp_csum);
 696                 if (unlikely(err))
 697                         goto tx_error;
 698
 699                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 700                 ttl = geneve->ttl;
 701                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 702                         ttl = 1;
 703                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 704                 df = 0;
 705         }
 706         err = udp_tunnel_xmit_skb(rt, gs->sock->sk, skb, fl4.saddr, fl4.daddr,
 707                                   tos, ttl, df, sport, geneve->dst_port,
 708                                   !net_eq(geneve->net, dev_net(geneve->dev)),
 709                                   !udp_csum);
 710
 711         iptunnel_xmit_stats(err, &dev->stats, (struct pcpu_sw_netstats __percpu *) dev->tstats);
 712         return NETDEV_TX_OK;
 713
 714 tx_error:
 715         dev_kfree_skb(skb);
 716
 717         if (err == -ELOOP)
 718                 dev->stats.collisions++;
 719         else if (err == -ENETUNREACH)
 720                 dev->stats.tx_carrier_errors++;
 721         else
 722                 dev->stats.tx_errors++;
 723         return NETDEV_TX_OK;
 724 }
 725 EXPORT_SYMBOL(rpl_geneve_xmit);
 726
 727 static netdev_tx_t geneve_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 728 {
 729         /* Drop All packets coming from networking stack. OVS-CB is
 730          * not initialized for these packets.
 731          */
 732
 733         dev_kfree_skb(skb);
 734         dev->stats.tx_dropped++;
 735         return NETDEV_TX_OK;
 736 }
 737
 738 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
 739 {
 740         /* The max_mtu calculation does not take account of GENEVE
 741          * options, to avoid excluding potentially valid
 742          * configurations.
 743          */
 744         int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
 745                       - dev->hard_header_len;
 746
 747         if (new_mtu < 68)
 748                 return -EINVAL;
 749
 750         if (new_mtu > max_mtu) {
 751                 if (strict)
 752                         return -EINVAL;
 753
 754                 new_mtu = max_mtu;
 755         }
 756
 757         dev->mtu = new_mtu;
 758         return 0;
 759 }
 760
 761 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
 762 {
 763         return __geneve_change_mtu(dev, new_mtu, true);
 764 }
 765
 766 int ovs_geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 767 {
 768         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 769         struct geneve_dev *geneve = netdev_priv(dev);
 770         struct rtable *rt;
 771         struct flowi4 fl4;
 772
 773         if (ip_tunnel_info_af(info) != AF_INET)
 774                 return -EINVAL;
 775
 776         rt = geneve_get_rt(skb, dev, &fl4, info);
 777         if (IS_ERR(rt))
 778                 return PTR_ERR(rt);
 779
 780         ip_rt_put(rt);
 781         info->key.u.ipv4.src = fl4.saddr;
 782         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
 783                                              1, USHRT_MAX, true);
 784         info->key.tp_dst = geneve->dst_port;
 785         return 0;
 786 }
 787 EXPORT_SYMBOL_GPL(ovs_geneve_fill_metadata_dst);
 788
 789 static const struct net_device_ops geneve_netdev_ops = {
 790         .ndo_init               = geneve_init,
 791         .ndo_uninit             = geneve_uninit,
 792         .ndo_get_stats64        = ip_tunnel_get_stats64,
 793         .ndo_open               = geneve_open,
 794         .ndo_stop               = geneve_stop,
 795         .ndo_start_xmit         = geneve_dev_xmit,
 796         .ndo_change_mtu         = geneve_change_mtu,
 797         .ndo_validate_addr      = eth_validate_addr,
 798         .ndo_set_mac_address    = eth_mac_addr,
 799 #ifdef HAVE_NDO_FILL_METADATA_DST
 800         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
 801 #endif
 802 };
 803
 804 static void geneve_get_drvinfo(struct net_device *dev,
 805                                struct ethtool_drvinfo *drvinfo)
 806 {
 807         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
 808         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
 809 }
 810
 811 static const struct ethtool_ops geneve_ethtool_ops = {
 812         .get_drvinfo    = geneve_get_drvinfo,
 813         .get_link       = ethtool_op_get_link,
 814 };
 815
 816 /* Info for udev, that this is a virtual tunnel endpoint */
 817 static struct device_type geneve_type = {
 818         .name = "geneve",
 819 };
 820
 821 /* Initialize the device structure. */
 822 static void geneve_setup(struct net_device *dev)
 823 {
 824         ether_setup(dev);
 825
 826         dev->netdev_ops = &geneve_netdev_ops;
 827         dev->ethtool_ops = &geneve_ethtool_ops;
 828         dev->destructor = free_netdev;
 829
 830         SET_NETDEV_DEVTYPE(dev, &geneve_type);
 831
 832         dev->features    |= NETIF_F_LLTX;
 833         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
 834         dev->features    |= NETIF_F_RXCSUM;
 835         dev->features    |= NETIF_F_GSO_SOFTWARE;
 836
 837 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
 838         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
 839         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 840 #endif
 841 #if 0
 842         /* Not required */
 843         netif_keep_dst(dev);
 844 #endif
 845         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
 846         eth_hw_addr_random(dev);
 847 }
 848
 849 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
 850         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
 851         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 852         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
 853         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
 854         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
 855         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
 856 };
 857
 858 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
 859 {
 860         if (tb[IFLA_ADDRESS]) {
 861                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 862                         return -EINVAL;
 863
 864                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 865                         return -EADDRNOTAVAIL;
 866         }
 867
 868         if (!data)
 869                 return -EINVAL;
 870
 871         if (data[IFLA_GENEVE_ID]) {
 872                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
 873
 874                 if (vni >= GENEVE_VID_MASK)
 875                         return -ERANGE;
 876         }
 877
 878         return 0;
 879 }
 880
 881 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
 882                                           __be16 dst_port,
 883                                           __be32 rem_addr,
 884                                           u8 vni[],
 885                                           bool *tun_on_same_port,
 886                                           bool *tun_collect_md)
 887 {
 888         struct geneve_dev *geneve, *t;
 889
 890         *tun_on_same_port = false;
 891         *tun_collect_md = false;
 892         t = NULL;
 893         list_for_each_entry(geneve, &gn->geneve_list, next) {
 894                 if (geneve->dst_port == dst_port) {
 895                         *tun_collect_md = geneve->collect_md;
 896                         *tun_on_same_port = true;
 897                 }
 898                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 899                     rem_addr == geneve->remote.sin_addr.s_addr &&
 900                     dst_port == geneve->dst_port)
 901                         t = geneve;
 902         }
 903         return t;
 904 }
 905
 906 static int geneve_configure(struct net *net, struct net_device *dev,
 907                             __be32 rem_addr, __u32 vni, __u8 ttl, __u8 tos,
 908                             __be16 dst_port, bool metadata)
 909 {
 910         struct geneve_net *gn = net_generic(net, geneve_net_id);
 911         struct geneve_dev *t, *geneve = netdev_priv(dev);
 912         bool tun_collect_md, tun_on_same_port;
 913         int err;
 914
 915         if (metadata) {
 916                 if (rem_addr || vni || tos || ttl)
 917                         return -EINVAL;
 918         }
 919
 920         geneve->net = net;
 921         geneve->dev = dev;
 922
 923         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
 924         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
 925         geneve->vni[2] =  vni & 0x000000ff;
 926
 927         geneve->remote.sin_addr.s_addr = rem_addr;
 928         if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
 929                 return -EINVAL;
 930
 931         geneve->ttl = ttl;
 932         geneve->tos = tos;
 933         geneve->dst_port = dst_port;
 934         geneve->collect_md = metadata;
 935
 936         t = geneve_find_dev(gn, dst_port, rem_addr, geneve->vni,
 937                             &tun_on_same_port, &tun_collect_md);
 938         if (t)
 939                 return -EBUSY;
 940
 941         if (metadata) {
 942                 if (tun_on_same_port)
 943                         return -EPERM;
 944         } else {
 945                 if (tun_collect_md)
 946                         return -EPERM;
 947         }
 948
 949         err = register_netdevice(dev);
 950         if (err)
 951                 return err;
 952
 953         list_add(&geneve->next, &gn->geneve_list);
 954         return 0;
 955 }
 956
 957 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 958 static int geneve_newlink(struct net_device *dev,
 959                           struct nlattr *tb[], struct nlattr *data[])
 960 {
 961         struct net *net = &init_net;
 962 #else
 963 static int geneve_newlink(struct net *net, struct net_device *dev,
 964                           struct nlattr *tb[], struct nlattr *data[])
 965 {
 966 #endif
 967         __be16 dst_port = htons(GENEVE_UDP_PORT);
 968         __u8 ttl = 0, tos = 0;
 969         bool metadata = false;
 970         __be32 rem_addr;
 971         __u32 vni;
 972
 973         if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
 974                 return -EINVAL;
 975
 976         vni = nla_get_u32(data[IFLA_GENEVE_ID]);
 977         rem_addr = nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
 978
 979         if (data[IFLA_GENEVE_TTL])
 980                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
 981
 982         if (data[IFLA_GENEVE_TOS])
 983                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
 984
 985         if (data[IFLA_GENEVE_PORT])
 986                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
 987
 988         if (data[IFLA_GENEVE_COLLECT_METADATA])
 989                 metadata = true;
 990
 991         return geneve_configure(net, dev, rem_addr, vni,
 992                                 ttl, tos, dst_port, metadata);
 993 }
 994
 995 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 996 static void geneve_dellink(struct net_device *dev)
 997 #else
 998 static void geneve_dellink(struct net_device *dev, struct list_head *head)
 999 #endif
1000 {
1001         struct geneve_dev *geneve = netdev_priv(dev);
1002
1003         list_del(&geneve->next);
1004         unregister_netdevice_queue(dev, head);
1005 }
1006
1007 static size_t geneve_get_size(const struct net_device *dev)
1008 {
1009         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1010                 nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
1011                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1012                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1013                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1014                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1015                 0;
1016 }
1017
1018 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1019 {
1020         struct geneve_dev *geneve = netdev_priv(dev);
1021         __u32 vni;
1022
1023         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1024         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1025                 goto nla_put_failure;
1026
1027         if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1028                             geneve->remote.sin_addr.s_addr))
1029                 goto nla_put_failure;
1030
1031         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1032             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
1033                 goto nla_put_failure;
1034
1035         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1036                 goto nla_put_failure;
1037
1038         if (geneve->collect_md) {
1039                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1040                         goto nla_put_failure;
1041         }
1042
1043         return 0;
1044
1045 nla_put_failure:
1046         return -EMSGSIZE;
1047 }
1048
1049 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1050         .kind           = "ovs_geneve",
1051         .maxtype        = IFLA_GENEVE_MAX,
1052         .policy         = geneve_policy,
1053         .priv_size      = sizeof(struct geneve_dev),
1054         .setup          = geneve_setup,
1055         .validate       = geneve_validate,
1056         .newlink        = geneve_newlink,
1057         .dellink        = geneve_dellink,
1058         .get_size       = geneve_get_size,
1059         .fill_info      = geneve_fill_info,
1060 };
1061
1062 struct net_device *rpl_geneve_dev_create_fb(struct net *net, const char *name,
1063                                         u8 name_assign_type, u16 dst_port)
1064 {
1065         struct nlattr *tb[IFLA_MAX + 1];
1066         struct net_device *dev;
1067         int err;
1068
1069         memset(tb, 0, sizeof(tb));
1070         dev = rtnl_create_link(net, (char *) name, name_assign_type,
1071                                &geneve_link_ops, tb);
1072         if (IS_ERR(dev))
1073                 return dev;
1074
1075         err = geneve_configure(net, dev, 0, 0, 0, 0, htons(dst_port), true);
1076         if (err)
1077                 goto err;
1078
1079         /* openvswitch users expect packet sizes to be unrestricted,
1080          * so set the largest MTU we can.
1081          */
1082         err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1083         if (err)
1084                 goto err;
1085
1086         return dev;
1087
1088 err:
1089         free_netdev(dev);
1090         return ERR_PTR(err);
1091 }
1092 EXPORT_SYMBOL_GPL(rpl_geneve_dev_create_fb);
1093
1094 static __net_init int geneve_init_net(struct net *net)
1095 {
1096         struct geneve_net *gn = net_generic(net, geneve_net_id);
1097
1098         INIT_LIST_HEAD(&gn->geneve_list);
1099         INIT_LIST_HEAD(&gn->sock_list);
1100         return 0;
1101 }
1102
1103 static void __net_exit geneve_exit_net(struct net *net)
1104 {
1105         struct geneve_net *gn = net_generic(net, geneve_net_id);
1106         struct geneve_dev *geneve, *next;
1107         struct net_device *dev, *aux;
1108         LIST_HEAD(list);
1109
1110         rtnl_lock();
1111
1112         /* gather any geneve devices that were moved into this ns */
1113         for_each_netdev_safe(net, dev, aux)
1114                 if (dev->rtnl_link_ops == &geneve_link_ops)
1115                         unregister_netdevice_queue(dev, &list);
1116
1117         /* now gather any other geneve devices that were created in this ns */
1118         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1119                 /* If geneve->dev is in the same netns, it was already added
1120                  * to the list by the previous loop.
1121                  */
1122                 if (!net_eq(dev_net(geneve->dev), net))
1123                         unregister_netdevice_queue(geneve->dev, &list);
1124         }
1125
1126         /* unregister the devices gathered above */
1127         unregister_netdevice_many(&list);
1128         rtnl_unlock();
1129 }
1130
1131 static struct pernet_operations geneve_net_ops = {
1132         .init = geneve_init_net,
1133         .exit = geneve_exit_net,
1134         .id   = &geneve_net_id,
1135         .size = sizeof(struct geneve_net),
1136 };
1137
1138 int rpl_geneve_init_module(void)
1139 {
1140         int rc;
1141
1142         rc = register_pernet_subsys(&geneve_net_ops);
1143         if (rc)
1144                 goto out1;
1145
1146         rc = rtnl_link_register(&geneve_link_ops);
1147         if (rc)
1148                 goto out2;
1149
1150         pr_info("Geneve tunneling driver\n");
1151         return 0;
1152 out2:
1153         unregister_pernet_subsys(&geneve_net_ops);
1154 out1:
1155         return rc;
1156 }
1157
1158 void rpl_geneve_cleanup_module(void)
1159 {
1160         rtnl_link_unregister(&geneve_link_ops);
1161         unregister_pernet_subsys(&geneve_net_ops);
1162 }
1163 #endif