drivers/net/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 <net/dst_metadata.h>
  19 #include <net/rtnetlink.h>
  20 #include <net/geneve.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_UDP_PORT         6081
  26
  27 #define GENEVE_N_VID            (1u << 24)
  28 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  29
  30 #define VNI_HASH_BITS           10
  31 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  32
  33 static bool log_ecn_error = true;
  34 module_param(log_ecn_error, bool, 0644);
  35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  36
  37 #define GENEVE_VER 0
  38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  39
  40 /* per-network namespace private data for this module */
  41 struct geneve_net {
  42         struct list_head        geneve_list;
  43         struct list_head        sock_list;
  44 };
  45
  46 static int geneve_net_id;
  47
  48 /* Pseudo network device */
  49 struct geneve_dev {
  50         struct hlist_node  hlist;       /* vni hash table */
  51         struct net         *net;        /* netns for packet i/o */
  52         struct net_device  *dev;        /* netdev for geneve tunnel */
  53         struct geneve_sock *sock;       /* socket used for geneve tunnel */
  54         u8                 vni[3];      /* virtual network ID for tunnel */
  55         u8                 ttl;         /* TTL override */
  56         u8                 tos;         /* TOS override */
  57         struct sockaddr_in remote;      /* IPv4 address for link partner */
  58         struct list_head   next;        /* geneve's per namespace list */
  59         __be16             dst_port;
  60         bool               collect_md;
  61 };
  62
  63 struct geneve_sock {
  64         bool                    collect_md;
  65         struct list_head        list;
  66         struct socket           *sock;
  67         struct rcu_head         rcu;
  68         int                     refcnt;
  69         struct udp_offload      udp_offloads;
  70         struct hlist_head       vni_list[VNI_HASH_SIZE];
  71 };
  72
  73 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  74 {
  75         __u32 vnid;
  76
  77         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  78         return hash_32(vnid, VNI_HASH_BITS);
  79 }
  80
  81 static __be64 vni_to_tunnel_id(const __u8 *vni)
  82 {
  83 #ifdef __BIG_ENDIAN
  84         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  85 #else
  86         return (__force __be64)(((__force u64)vni[0] << 40) |
  87                                 ((__force u64)vni[1] << 48) |
  88                                 ((__force u64)vni[2] << 56));
  89 #endif
  90 }
  91
  92 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
  93                                         __be32 addr, u8 vni[])
  94 {
  95         struct hlist_head *vni_list_head;
  96         struct geneve_dev *geneve;
  97         __u32 hash;
  98
  99         /* Find the device for this VNI */
 100         hash = geneve_net_vni_hash(vni);
 101         vni_list_head = &gs->vni_list[hash];
 102         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 103                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 104                     addr == geneve->remote.sin_addr.s_addr)
 105                         return geneve;
 106         }
 107         return NULL;
 108 }
 109
 110 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 111 {
 112         return (struct genevehdr *)(udp_hdr(skb) + 1);
 113 }
 114
 115 /* geneve receive/decap routine */
 116 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
 117 {
 118         struct genevehdr *gnvh = geneve_hdr(skb);
 119         struct metadata_dst *tun_dst = NULL;
 120         struct geneve_dev *geneve = NULL;
 121         struct pcpu_sw_netstats *stats;
 122         struct iphdr *iph;
 123         u8 *vni;
 124         __be32 addr;
 125         int err;
 126
 127         if (gs->collect_md) {
 128                 static u8 zero_vni[3];
 129
 130                 vni = zero_vni;
 131                 addr = 0;
 132         } else {
 133                 vni = gnvh->vni;
 134                 iph = ip_hdr(skb); /* Still outer IP header... */
 135                 addr = iph->saddr;
 136         }
 137
 138         geneve = geneve_lookup(gs, addr, vni);
 139         if (!geneve)
 140                 goto drop;
 141
 142         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 143                 __be16 flags;
 144                 void *opts;
 145
 146                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 147                         (gnvh->oam ? TUNNEL_OAM : 0) |
 148                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 149
 150                 tun_dst = udp_tun_rx_dst(skb, AF_INET, flags,
 151                                          vni_to_tunnel_id(gnvh->vni),
 152                                          gnvh->opt_len * 4);
 153                 if (!tun_dst)
 154                         goto drop;
 155
 156                 /* Update tunnel dst according to Geneve options. */
 157                 opts = ip_tunnel_info_opts(&tun_dst->u.tun_info,
 158                                            gnvh->opt_len * 4);
 159                 memcpy(opts, gnvh->options, gnvh->opt_len * 4);
 160         } else {
 161                 /* Drop packets w/ critical options,
 162                  * since we don't support any...
 163                  */
 164                 if (gnvh->critical)
 165                         goto drop;
 166         }
 167
 168         skb_reset_mac_header(skb);
 169         skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
 170         skb->protocol = eth_type_trans(skb, geneve->dev);
 171         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 172
 173         if (tun_dst)
 174                 skb_dst_set(skb, &tun_dst->dst);
 175
 176         /* Ignore packet loops (and multicast echo) */
 177         if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
 178                 goto drop;
 179
 180         skb_reset_network_header(skb);
 181
 182         iph = ip_hdr(skb); /* Now inner IP header... */
 183         err = IP_ECN_decapsulate(iph, skb);
 184
 185         if (unlikely(err)) {
 186                 if (log_ecn_error)
 187                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
 188                                              &iph->saddr, iph->tos);
 189                 if (err > 1) {
 190                         ++geneve->dev->stats.rx_frame_errors;
 191                         ++geneve->dev->stats.rx_errors;
 192                         goto drop;
 193                 }
 194         }
 195
 196         stats = this_cpu_ptr(geneve->dev->tstats);
 197         u64_stats_update_begin(&stats->syncp);
 198         stats->rx_packets++;
 199         stats->rx_bytes += skb->len;
 200         u64_stats_update_end(&stats->syncp);
 201
 202         netif_rx(skb);
 203         return;
 204 drop:
 205         /* Consume bad packet */
 206         kfree_skb(skb);
 207 }
 208
 209 /* Setup stats when device is created */
 210 static int geneve_init(struct net_device *dev)
 211 {
 212         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 213         if (!dev->tstats)
 214                 return -ENOMEM;
 215         return 0;
 216 }
 217
 218 static void geneve_uninit(struct net_device *dev)
 219 {
 220         free_percpu(dev->tstats);
 221 }
 222
 223 /* Callback from net/ipv4/udp.c to receive packets */
 224 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 225 {
 226         struct genevehdr *geneveh;
 227         struct geneve_sock *gs;
 228         int opts_len;
 229
 230         /* Need Geneve and inner Ethernet header to be present */
 231         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 232                 goto error;
 233
 234         /* Return packets with reserved bits set */
 235         geneveh = geneve_hdr(skb);
 236         if (unlikely(geneveh->ver != GENEVE_VER))
 237                 goto error;
 238
 239         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 240                 goto error;
 241
 242         opts_len = geneveh->opt_len * 4;
 243         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 244                                  htons(ETH_P_TEB)))
 245                 goto drop;
 246
 247         gs = rcu_dereference_sk_user_data(sk);
 248         if (!gs)
 249                 goto drop;
 250
 251         geneve_rx(gs, skb);
 252         return 0;
 253
 254 drop:
 255         /* Consume bad packet */
 256         kfree_skb(skb);
 257         return 0;
 258
 259 error:
 260         /* Let the UDP layer deal with the skb */
 261         return 1;
 262 }
 263
 264 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 265                                          __be16 port)
 266 {
 267         struct socket *sock;
 268         struct udp_port_cfg udp_conf;
 269         int err;
 270
 271         memset(&udp_conf, 0, sizeof(udp_conf));
 272
 273         if (ipv6) {
 274                 udp_conf.family = AF_INET6;
 275         } else {
 276                 udp_conf.family = AF_INET;
 277                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 278         }
 279
 280         udp_conf.local_udp_port = port;
 281
 282         /* Open UDP socket */
 283         err = udp_sock_create(net, &udp_conf, &sock);
 284         if (err < 0)
 285                 return ERR_PTR(err);
 286
 287         return sock;
 288 }
 289
 290 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
 291 {
 292         struct sock *sk = gs->sock->sk;
 293         sa_family_t sa_family = sk->sk_family;
 294         int err;
 295
 296         if (sa_family == AF_INET) {
 297                 err = udp_add_offload(&gs->udp_offloads);
 298                 if (err)
 299                         pr_warn("geneve: udp_add_offload failed with status %d\n",
 300                                 err);
 301         }
 302 }
 303
 304 static int geneve_hlen(struct genevehdr *gh)
 305 {
 306         return sizeof(*gh) + gh->opt_len * 4;
 307 }
 308
 309 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 310                                            struct sk_buff *skb,
 311                                            struct udp_offload *uoff)
 312 {
 313         struct sk_buff *p, **pp = NULL;
 314         struct genevehdr *gh, *gh2;
 315         unsigned int hlen, gh_len, off_gnv;
 316         const struct packet_offload *ptype;
 317         __be16 type;
 318         int flush = 1;
 319
 320         off_gnv = skb_gro_offset(skb);
 321         hlen = off_gnv + sizeof(*gh);
 322         gh = skb_gro_header_fast(skb, off_gnv);
 323         if (skb_gro_header_hard(skb, hlen)) {
 324                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 325                 if (unlikely(!gh))
 326                         goto out;
 327         }
 328
 329         if (gh->ver != GENEVE_VER || gh->oam)
 330                 goto out;
 331         gh_len = geneve_hlen(gh);
 332
 333         hlen = off_gnv + gh_len;
 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         flush = 0;
 341
 342         for (p = *head; p; p = p->next) {
 343                 if (!NAPI_GRO_CB(p)->same_flow)
 344                         continue;
 345
 346                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 347                 if (gh->opt_len != gh2->opt_len ||
 348                     memcmp(gh, gh2, gh_len)) {
 349                         NAPI_GRO_CB(p)->same_flow = 0;
 350                         continue;
 351                 }
 352         }
 353
 354         type = gh->proto_type;
 355
 356         rcu_read_lock();
 357         ptype = gro_find_receive_by_type(type);
 358         if (!ptype) {
 359                 flush = 1;
 360                 goto out_unlock;
 361         }
 362
 363         skb_gro_pull(skb, gh_len);
 364         skb_gro_postpull_rcsum(skb, gh, gh_len);
 365         pp = ptype->callbacks.gro_receive(head, skb);
 366
 367 out_unlock:
 368         rcu_read_unlock();
 369 out:
 370         NAPI_GRO_CB(skb)->flush |= flush;
 371
 372         return pp;
 373 }
 374
 375 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 376                                struct udp_offload *uoff)
 377 {
 378         struct genevehdr *gh;
 379         struct packet_offload *ptype;
 380         __be16 type;
 381         int gh_len;
 382         int err = -ENOSYS;
 383
 384         udp_tunnel_gro_complete(skb, nhoff);
 385
 386         gh = (struct genevehdr *)(skb->data + nhoff);
 387         gh_len = geneve_hlen(gh);
 388         type = gh->proto_type;
 389
 390         rcu_read_lock();
 391         ptype = gro_find_complete_by_type(type);
 392         if (ptype)
 393                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 394
 395         rcu_read_unlock();
 396         return err;
 397 }
 398
 399 /* Create new listen socket if needed */
 400 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 401                                                 bool ipv6)
 402 {
 403         struct geneve_net *gn = net_generic(net, geneve_net_id);
 404         struct geneve_sock *gs;
 405         struct socket *sock;
 406         struct udp_tunnel_sock_cfg tunnel_cfg;
 407         int h;
 408
 409         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 410         if (!gs)
 411                 return ERR_PTR(-ENOMEM);
 412
 413         sock = geneve_create_sock(net, ipv6, port);
 414         if (IS_ERR(sock)) {
 415                 kfree(gs);
 416                 return ERR_CAST(sock);
 417         }
 418
 419         gs->sock = sock;
 420         gs->refcnt = 1;
 421         for (h = 0; h < VNI_HASH_SIZE; ++h)
 422                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 423
 424         /* Initialize the geneve udp offloads structure */
 425         gs->udp_offloads.port = port;
 426         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 427         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 428         geneve_notify_add_rx_port(gs);
 429
 430         /* Mark socket as an encapsulation socket */
 431         tunnel_cfg.sk_user_data = gs;
 432         tunnel_cfg.encap_type = 1;
 433         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 434         tunnel_cfg.encap_destroy = NULL;
 435         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 436         list_add(&gs->list, &gn->sock_list);
 437         return gs;
 438 }
 439
 440 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 441 {
 442         struct sock *sk = gs->sock->sk;
 443         sa_family_t sa_family = sk->sk_family;
 444
 445         if (sa_family == AF_INET)
 446                 udp_del_offload(&gs->udp_offloads);
 447 }
 448
 449 static void geneve_sock_release(struct geneve_sock *gs)
 450 {
 451         if (--gs->refcnt)
 452                 return;
 453
 454         list_del(&gs->list);
 455         geneve_notify_del_rx_port(gs);
 456         udp_tunnel_sock_release(gs->sock);
 457         kfree_rcu(gs, rcu);
 458 }
 459
 460 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 461                                             __be16 dst_port)
 462 {
 463         struct geneve_sock *gs;
 464
 465         list_for_each_entry(gs, &gn->sock_list, list) {
 466                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 467                     inet_sk(gs->sock->sk)->sk.sk_family == AF_INET) {
 468                         return gs;
 469                 }
 470         }
 471         return NULL;
 472 }
 473
 474 static int geneve_open(struct net_device *dev)
 475 {
 476         struct geneve_dev *geneve = netdev_priv(dev);
 477         struct net *net = geneve->net;
 478         struct geneve_net *gn = net_generic(net, geneve_net_id);
 479         struct geneve_sock *gs;
 480         __u32 hash;
 481
 482         gs = geneve_find_sock(gn, geneve->dst_port);
 483         if (gs) {
 484                 gs->refcnt++;
 485                 goto out;
 486         }
 487
 488         gs = geneve_socket_create(net, geneve->dst_port, false);
 489         if (IS_ERR(gs))
 490                 return PTR_ERR(gs);
 491
 492 out:
 493         gs->collect_md = geneve->collect_md;
 494         geneve->sock = gs;
 495
 496         hash = geneve_net_vni_hash(geneve->vni);
 497         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 498         return 0;
 499 }
 500
 501 static int geneve_stop(struct net_device *dev)
 502 {
 503         struct geneve_dev *geneve = netdev_priv(dev);
 504         struct geneve_sock *gs = geneve->sock;
 505
 506         if (!hlist_unhashed(&geneve->hlist))
 507                 hlist_del_rcu(&geneve->hlist);
 508         geneve_sock_release(gs);
 509         return 0;
 510 }
 511
 512 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 513                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 514                             bool csum)
 515 {
 516         struct genevehdr *gnvh;
 517         int min_headroom;
 518         int err;
 519
 520         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 521                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 522         err = skb_cow_head(skb, min_headroom);
 523         if (unlikely(err)) {
 524                 kfree_skb(skb);
 525                 goto free_rt;
 526         }
 527
 528         skb = udp_tunnel_handle_offloads(skb, csum);
 529         if (IS_ERR(skb)) {
 530                 err = PTR_ERR(skb);
 531                 goto free_rt;
 532         }
 533
 534         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 535         gnvh->ver = GENEVE_VER;
 536         gnvh->opt_len = opt_len / 4;
 537         gnvh->oam = !!(tun_flags & TUNNEL_OAM);
 538         gnvh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 539         gnvh->rsvd1 = 0;
 540         memcpy(gnvh->vni, vni, 3);
 541         gnvh->proto_type = htons(ETH_P_TEB);
 542         gnvh->rsvd2 = 0;
 543         memcpy(gnvh->options, opt, opt_len);
 544
 545         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 546         return 0;
 547
 548 free_rt:
 549         ip_rt_put(rt);
 550         return err;
 551 }
 552
 553 static struct rtable *geneve_get_rt(struct sk_buff *skb,
 554                                     struct net_device *dev,
 555                                     struct flowi4 *fl4,
 556                                     struct ip_tunnel_info *info)
 557 {
 558         struct geneve_dev *geneve = netdev_priv(dev);
 559         struct rtable *rt = NULL;
 560         __u8 tos;
 561
 562         memset(fl4, 0, sizeof(*fl4));
 563         fl4->flowi4_mark = skb->mark;
 564         fl4->flowi4_proto = IPPROTO_UDP;
 565
 566         if (info) {
 567                 fl4->daddr = info->key.u.ipv4.dst;
 568                 fl4->saddr = info->key.u.ipv4.src;
 569                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 570         } else {
 571                 tos = geneve->tos;
 572                 if (tos == 1) {
 573                         const struct iphdr *iip = ip_hdr(skb);
 574
 575                         tos = ip_tunnel_get_dsfield(iip, skb);
 576                 }
 577
 578                 fl4->flowi4_tos = RT_TOS(tos);
 579                 fl4->daddr = geneve->remote.sin_addr.s_addr;
 580         }
 581
 582         rt = ip_route_output_key(geneve->net, fl4);
 583         if (IS_ERR(rt)) {
 584                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 585                 dev->stats.tx_carrier_errors++;
 586                 return rt;
 587         }
 588         if (rt->dst.dev == dev) { /* is this necessary? */
 589                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 590                 dev->stats.collisions++;
 591                 ip_rt_put(rt);
 592                 return ERR_PTR(-EINVAL);
 593         }
 594         return rt;
 595 }
 596
 597 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 598 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 599 {
 600 #ifdef __BIG_ENDIAN
 601         vni[0] = (__force __u8)(tun_id >> 16);
 602         vni[1] = (__force __u8)(tun_id >> 8);
 603         vni[2] = (__force __u8)tun_id;
 604 #else
 605         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 606         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 607         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 608 #endif
 609 }
 610
 611 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 612 {
 613         struct geneve_dev *geneve = netdev_priv(dev);
 614         struct geneve_sock *gs = geneve->sock;
 615         struct ip_tunnel_info *info = NULL;
 616         struct rtable *rt = NULL;
 617         struct flowi4 fl4;
 618         __u8 tos, ttl;
 619         __be16 sport;
 620         bool udp_csum;
 621         __be16 df;
 622         int err;
 623
 624         if (geneve->collect_md) {
 625                 info = skb_tunnel_info(skb);
 626                 if (unlikely(info && !(info->mode & IP_TUNNEL_INFO_TX))) {
 627                         netdev_dbg(dev, "no tunnel metadata\n");
 628                         goto tx_error;
 629                 }
 630         }
 631
 632         rt = geneve_get_rt(skb, dev, &fl4, info);
 633         if (IS_ERR(rt)) {
 634                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
 635                 dev->stats.tx_carrier_errors++;
 636                 goto tx_error;
 637         }
 638
 639         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 640         skb_reset_mac_header(skb);
 641
 642         if (info) {
 643                 const struct ip_tunnel_key *key = &info->key;
 644                 u8 *opts = NULL;
 645                 u8 vni[3];
 646
 647                 tunnel_id_to_vni(key->tun_id, vni);
 648                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 649                         opts = ip_tunnel_info_opts(info, info->options_len);
 650
 651                 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
 652                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 653                                        info->options_len, opts, udp_csum);
 654                 if (unlikely(err))
 655                         goto err;
 656
 657                 tos = key->tos;
 658                 ttl = key->ttl;
 659                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 660         } else {
 661                 const struct iphdr *iip; /* interior IP header */
 662
 663                 udp_csum = false;
 664                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 665                                        0, NULL, udp_csum);
 666                 if (unlikely(err))
 667                         goto err;
 668
 669                 iip = ip_hdr(skb);
 670                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 671                 ttl = geneve->ttl;
 672                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 673                         ttl = 1;
 674                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 675                 df = 0;
 676         }
 677         err = udp_tunnel_xmit_skb(rt, gs->sock->sk, skb, fl4.saddr, fl4.daddr,
 678                                   tos, ttl, df, sport, geneve->dst_port,
 679                                   !net_eq(geneve->net, dev_net(geneve->dev)),
 680                                   !udp_csum);
 681
 682         iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
 683         return NETDEV_TX_OK;
 684
 685 tx_error:
 686         dev_kfree_skb(skb);
 687 err:
 688         dev->stats.tx_errors++;
 689         return NETDEV_TX_OK;
 690 }
 691
 692 static const struct net_device_ops geneve_netdev_ops = {
 693         .ndo_init               = geneve_init,
 694         .ndo_uninit             = geneve_uninit,
 695         .ndo_open               = geneve_open,
 696         .ndo_stop               = geneve_stop,
 697         .ndo_start_xmit         = geneve_xmit,
 698         .ndo_get_stats64        = ip_tunnel_get_stats64,
 699         .ndo_change_mtu         = eth_change_mtu,
 700         .ndo_validate_addr      = eth_validate_addr,
 701         .ndo_set_mac_address    = eth_mac_addr,
 702 };
 703
 704 static void geneve_get_drvinfo(struct net_device *dev,
 705                                struct ethtool_drvinfo *drvinfo)
 706 {
 707         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
 708         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
 709 }
 710
 711 static const struct ethtool_ops geneve_ethtool_ops = {
 712         .get_drvinfo    = geneve_get_drvinfo,
 713         .get_link       = ethtool_op_get_link,
 714 };
 715
 716 /* Info for udev, that this is a virtual tunnel endpoint */
 717 static struct device_type geneve_type = {
 718         .name = "geneve",
 719 };
 720
 721 /* Initialize the device structure. */
 722 static void geneve_setup(struct net_device *dev)
 723 {
 724         ether_setup(dev);
 725
 726         dev->netdev_ops = &geneve_netdev_ops;
 727         dev->ethtool_ops = &geneve_ethtool_ops;
 728         dev->destructor = free_netdev;
 729
 730         SET_NETDEV_DEVTYPE(dev, &geneve_type);
 731
 732         dev->features    |= NETIF_F_LLTX;
 733         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
 734         dev->features    |= NETIF_F_RXCSUM;
 735         dev->features    |= NETIF_F_GSO_SOFTWARE;
 736
 737         dev->vlan_features = dev->features;
 738         dev->features    |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 739
 740         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
 741         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 742         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 743
 744         netif_keep_dst(dev);
 745         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
 746         eth_hw_addr_random(dev);
 747 }
 748
 749 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
 750         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
 751         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 752         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
 753         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
 754         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
 755         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
 756 };
 757
 758 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
 759 {
 760         if (tb[IFLA_ADDRESS]) {
 761                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 762                         return -EINVAL;
 763
 764                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 765                         return -EADDRNOTAVAIL;
 766         }
 767
 768         if (!data)
 769                 return -EINVAL;
 770
 771         if (data[IFLA_GENEVE_ID]) {
 772                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
 773
 774                 if (vni >= GENEVE_VID_MASK)
 775                         return -ERANGE;
 776         }
 777
 778         return 0;
 779 }
 780
 781 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
 782                                           __be16 dst_port,
 783                                           __be32 rem_addr,
 784                                           u8 vni[],
 785                                           bool *tun_on_same_port,
 786                                           bool *tun_collect_md)
 787 {
 788         struct geneve_dev *geneve, *t;
 789
 790         *tun_on_same_port = false;
 791         *tun_collect_md = false;
 792         t = NULL;
 793         list_for_each_entry(geneve, &gn->geneve_list, next) {
 794                 if (geneve->dst_port == dst_port) {
 795                         *tun_collect_md = geneve->collect_md;
 796                         *tun_on_same_port = true;
 797                 }
 798                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 799                     rem_addr == geneve->remote.sin_addr.s_addr &&
 800                     dst_port == geneve->dst_port)
 801                         t = geneve;
 802         }
 803         return t;
 804 }
 805
 806 static int geneve_configure(struct net *net, struct net_device *dev,
 807                             __be32 rem_addr, __u32 vni, __u8 ttl, __u8 tos,
 808                             __u16 dst_port, bool metadata)
 809 {
 810         struct geneve_net *gn = net_generic(net, geneve_net_id);
 811         struct geneve_dev *t, *geneve = netdev_priv(dev);
 812         bool tun_collect_md, tun_on_same_port;
 813         int err;
 814
 815         if (metadata) {
 816                 if (rem_addr || vni || tos || ttl)
 817                         return -EINVAL;
 818         }
 819
 820         geneve->net = net;
 821         geneve->dev = dev;
 822
 823         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
 824         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
 825         geneve->vni[2] =  vni & 0x000000ff;
 826
 827         geneve->remote.sin_addr.s_addr = rem_addr;
 828         if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
 829                 return -EINVAL;
 830
 831         geneve->ttl = ttl;
 832         geneve->tos = tos;
 833         geneve->dst_port = htons(dst_port);
 834         geneve->collect_md = metadata;
 835
 836         t = geneve_find_dev(gn, htons(dst_port), rem_addr, geneve->vni,
 837                             &tun_on_same_port, &tun_collect_md);
 838         if (t)
 839                 return -EBUSY;
 840
 841         if (metadata) {
 842                 if (tun_on_same_port)
 843                         return -EPERM;
 844         } else {
 845                 if (tun_collect_md)
 846                         return -EPERM;
 847         }
 848
 849         err = register_netdevice(dev);
 850         if (err)
 851                 return err;
 852
 853         list_add(&geneve->next, &gn->geneve_list);
 854         return 0;
 855 }
 856
 857 static int geneve_newlink(struct net *net, struct net_device *dev,
 858                           struct nlattr *tb[], struct nlattr *data[])
 859 {
 860         __u16 dst_port = GENEVE_UDP_PORT;
 861         __u8 ttl = 0, tos = 0;
 862         bool metadata = false;
 863         __be32 rem_addr;
 864         __u32 vni;
 865
 866         if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
 867                 return -EINVAL;
 868
 869         vni = nla_get_u32(data[IFLA_GENEVE_ID]);
 870         rem_addr = nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
 871
 872         if (data[IFLA_GENEVE_TTL])
 873                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
 874
 875         if (data[IFLA_GENEVE_TOS])
 876                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
 877
 878         if (data[IFLA_GENEVE_PORT])
 879                 dst_port = nla_get_u16(data[IFLA_GENEVE_PORT]);
 880
 881         if (data[IFLA_GENEVE_COLLECT_METADATA])
 882                 metadata = true;
 883
 884         return geneve_configure(net, dev, rem_addr, vni,
 885                                 ttl, tos, dst_port, metadata);
 886 }
 887
 888 static void geneve_dellink(struct net_device *dev, struct list_head *head)
 889 {
 890         struct geneve_dev *geneve = netdev_priv(dev);
 891
 892         list_del(&geneve->next);
 893         unregister_netdevice_queue(dev, head);
 894 }
 895
 896 static size_t geneve_get_size(const struct net_device *dev)
 897 {
 898         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
 899                 nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
 900                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
 901                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
 902                 nla_total_size(sizeof(__u16)) +  /* IFLA_GENEVE_PORT */
 903                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
 904                 0;
 905 }
 906
 907 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
 908 {
 909         struct geneve_dev *geneve = netdev_priv(dev);
 910         __u32 vni;
 911
 912         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
 913         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
 914                 goto nla_put_failure;
 915
 916         if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
 917                             geneve->remote.sin_addr.s_addr))
 918                 goto nla_put_failure;
 919
 920         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
 921             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
 922                 goto nla_put_failure;
 923
 924         if (nla_put_u16(skb, IFLA_GENEVE_PORT, ntohs(geneve->dst_port)))
 925                 goto nla_put_failure;
 926
 927         if (geneve->collect_md) {
 928                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
 929                         goto nla_put_failure;
 930         }
 931
 932         return 0;
 933
 934 nla_put_failure:
 935         return -EMSGSIZE;
 936 }
 937
 938 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
 939         .kind           = "geneve",
 940         .maxtype        = IFLA_GENEVE_MAX,
 941         .policy         = geneve_policy,
 942         .priv_size      = sizeof(struct geneve_dev),
 943         .setup          = geneve_setup,
 944         .validate       = geneve_validate,
 945         .newlink        = geneve_newlink,
 946         .dellink        = geneve_dellink,
 947         .get_size       = geneve_get_size,
 948         .fill_info      = geneve_fill_info,
 949 };
 950
 951 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
 952                                         u8 name_assign_type, u16 dst_port)
 953 {
 954         struct nlattr *tb[IFLA_MAX + 1];
 955         struct net_device *dev;
 956         int err;
 957
 958         memset(tb, 0, sizeof(tb));
 959         dev = rtnl_create_link(net, name, name_assign_type,
 960                                &geneve_link_ops, tb);
 961         if (IS_ERR(dev))
 962                 return dev;
 963
 964         err = geneve_configure(net, dev, 0, 0, 0, 0, dst_port, true);
 965         if (err) {
 966                 free_netdev(dev);
 967                 return ERR_PTR(err);
 968         }
 969         return dev;
 970 }
 971 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
 972
 973 static __net_init int geneve_init_net(struct net *net)
 974 {
 975         struct geneve_net *gn = net_generic(net, geneve_net_id);
 976
 977         INIT_LIST_HEAD(&gn->geneve_list);
 978         INIT_LIST_HEAD(&gn->sock_list);
 979         return 0;
 980 }
 981
 982 static void __net_exit geneve_exit_net(struct net *net)
 983 {
 984         struct geneve_net *gn = net_generic(net, geneve_net_id);
 985         struct geneve_dev *geneve, *next;
 986         struct net_device *dev, *aux;
 987         LIST_HEAD(list);
 988
 989         rtnl_lock();
 990
 991         /* gather any geneve devices that were moved into this ns */
 992         for_each_netdev_safe(net, dev, aux)
 993                 if (dev->rtnl_link_ops == &geneve_link_ops)
 994                         unregister_netdevice_queue(dev, &list);
 995
 996         /* now gather any other geneve devices that were created in this ns */
 997         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
 998                 /* If geneve->dev is in the same netns, it was already added
 999                  * to the list by the previous loop.
1000                  */
1001                 if (!net_eq(dev_net(geneve->dev), net))
1002                         unregister_netdevice_queue(geneve->dev, &list);
1003         }
1004
1005         /* unregister the devices gathered above */
1006         unregister_netdevice_many(&list);
1007         rtnl_unlock();
1008 }
1009
1010 static struct pernet_operations geneve_net_ops = {
1011         .init = geneve_init_net,
1012         .exit = geneve_exit_net,
1013         .id   = &geneve_net_id,
1014         .size = sizeof(struct geneve_net),
1015 };
1016
1017 static int __init geneve_init_module(void)
1018 {
1019         int rc;
1020
1021         rc = register_pernet_subsys(&geneve_net_ops);
1022         if (rc)
1023                 goto out1;
1024
1025         rc = rtnl_link_register(&geneve_link_ops);
1026         if (rc)
1027                 goto out2;
1028
1029         return 0;
1030 out2:
1031         unregister_pernet_subsys(&geneve_net_ops);
1032 out1:
1033         return rc;
1034 }
1035 late_initcall(geneve_init_module);
1036
1037 static void __exit geneve_cleanup_module(void)
1038 {
1039         rtnl_link_unregister(&geneve_link_ops);
1040         unregister_pernet_subsys(&geneve_net_ops);
1041 }
1042 module_exit(geneve_cleanup_module);
1043
1044 MODULE_LICENSE("GPL");
1045 MODULE_VERSION(GENEVE_NETDEV_VER);
1046 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1047 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1048 MODULE_ALIAS_RTNL_LINK("geneve");