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/gro_cells.h>
  20 #include <net/rtnetlink.h>
  21 #include <net/geneve.h>
  22 #include <net/protocol.h>
  23
  24 #define GENEVE_NETDEV_VER       "0.6"
  25
  26 #define GENEVE_UDP_PORT         6081
  27
  28 #define GENEVE_N_VID            (1u << 24)
  29 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  30
  31 #define VNI_HASH_BITS           10
  32 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  33
  34 static bool log_ecn_error = true;
  35 module_param(log_ecn_error, bool, 0644);
  36 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  37
  38 #define GENEVE_VER 0
  39 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  40
  41 /* per-network namespace private data for this module */
  42 struct geneve_net {
  43         struct list_head        geneve_list;
  44         struct list_head        sock_list;
  45 };
  46
  47 static int geneve_net_id;
  48
  49 /* Pseudo network device */
  50 struct geneve_dev {
  51         struct hlist_node  hlist;       /* vni hash table */
  52         struct net         *net;        /* netns for packet i/o */
  53         struct net_device  *dev;        /* netdev for geneve tunnel */
  54         struct geneve_sock *sock;       /* socket used for geneve tunnel */
  55         u8                 vni[3];      /* virtual network ID for tunnel */
  56         u8                 ttl;         /* TTL override */
  57         u8                 tos;         /* TOS override */
  58         struct sockaddr_in remote;      /* IPv4 address for link partner */
  59         struct list_head   next;        /* geneve's per namespace list */
  60         __be16             dst_port;
  61         bool               collect_md;
  62         struct gro_cells   gro_cells;
  63 };
  64
  65 struct geneve_sock {
  66         bool                    collect_md;
  67         struct list_head        list;
  68         struct socket           *sock;
  69         struct rcu_head         rcu;
  70         int                     refcnt;
  71         struct udp_offload      udp_offloads;
  72         struct hlist_head       vni_list[VNI_HASH_SIZE];
  73 };
  74
  75 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  76 {
  77         __u32 vnid;
  78
  79         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  80         return hash_32(vnid, VNI_HASH_BITS);
  81 }
  82
  83 static __be64 vni_to_tunnel_id(const __u8 *vni)
  84 {
  85 #ifdef __BIG_ENDIAN
  86         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
  87 #else
  88         return (__force __be64)(((__force u64)vni[0] << 40) |
  89                                 ((__force u64)vni[1] << 48) |
  90                                 ((__force u64)vni[2] << 56));
  91 #endif
  92 }
  93
  94 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
  95                                         __be32 addr, u8 vni[])
  96 {
  97         struct hlist_head *vni_list_head;
  98         struct geneve_dev *geneve;
  99         __u32 hash;
 100
 101         /* Find the device for this VNI */
 102         hash = geneve_net_vni_hash(vni);
 103         vni_list_head = &gs->vni_list[hash];
 104         hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
 105                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 106                     addr == geneve->remote.sin_addr.s_addr)
 107                         return geneve;
 108         }
 109         return NULL;
 110 }
 111
 112 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 113 {
 114         return (struct genevehdr *)(udp_hdr(skb) + 1);
 115 }
 116
 117 /* geneve receive/decap routine */
 118 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
 119 {
 120         struct genevehdr *gnvh = geneve_hdr(skb);
 121         struct metadata_dst *tun_dst = NULL;
 122         struct geneve_dev *geneve = NULL;
 123         struct pcpu_sw_netstats *stats;
 124         struct iphdr *iph;
 125         u8 *vni;
 126         __be32 addr;
 127         int err;
 128
 129         iph = ip_hdr(skb); /* outer IP header... */
 130
 131         if (gs->collect_md) {
 132                 static u8 zero_vni[3];
 133
 134                 vni = zero_vni;
 135                 addr = 0;
 136         } else {
 137                 vni = gnvh->vni;
 138                 addr = iph->saddr;
 139         }
 140
 141         geneve = geneve_lookup(gs, addr, vni);
 142         if (!geneve)
 143                 goto drop;
 144
 145         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 146                 __be16 flags;
 147
 148                 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
 149                         (gnvh->oam ? TUNNEL_OAM : 0) |
 150                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 151
 152                 tun_dst = udp_tun_rx_dst(skb, AF_INET, flags,
 153                                          vni_to_tunnel_id(gnvh->vni),
 154                                          gnvh->opt_len * 4);
 155                 if (!tun_dst)
 156                         goto drop;
 157                 /* Update tunnel dst according to Geneve options. */
 158                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 159                                         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         err = IP_ECN_decapsulate(iph, skb);
 183
 184         if (unlikely(err)) {
 185                 if (log_ecn_error)
 186                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
 187                                              &iph->saddr, iph->tos);
 188                 if (err > 1) {
 189                         ++geneve->dev->stats.rx_frame_errors;
 190                         ++geneve->dev->stats.rx_errors;
 191                         goto drop;
 192                 }
 193         }
 194
 195         stats = this_cpu_ptr(geneve->dev->tstats);
 196         u64_stats_update_begin(&stats->syncp);
 197         stats->rx_packets++;
 198         stats->rx_bytes += skb->len;
 199         u64_stats_update_end(&stats->syncp);
 200
 201         gro_cells_receive(&geneve->gro_cells, skb);
 202         return;
 203 drop:
 204         /* Consume bad packet */
 205         kfree_skb(skb);
 206 }
 207
 208 /* Setup stats when device is created */
 209 static int geneve_init(struct net_device *dev)
 210 {
 211         struct geneve_dev *geneve = netdev_priv(dev);
 212         int err;
 213
 214         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 215         if (!dev->tstats)
 216                 return -ENOMEM;
 217
 218         err = gro_cells_init(&geneve->gro_cells, dev);
 219         if (err) {
 220                 free_percpu(dev->tstats);
 221                 return err;
 222         }
 223
 224         return 0;
 225 }
 226
 227 static void geneve_uninit(struct net_device *dev)
 228 {
 229         struct geneve_dev *geneve = netdev_priv(dev);
 230
 231         gro_cells_destroy(&geneve->gro_cells);
 232         free_percpu(dev->tstats);
 233 }
 234
 235 /* Callback from net/ipv4/udp.c to receive packets */
 236 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 237 {
 238         struct genevehdr *geneveh;
 239         struct geneve_sock *gs;
 240         int opts_len;
 241
 242         /* Need Geneve and inner Ethernet header to be present */
 243         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 244                 goto error;
 245
 246         /* Return packets with reserved bits set */
 247         geneveh = geneve_hdr(skb);
 248         if (unlikely(geneveh->ver != GENEVE_VER))
 249                 goto error;
 250
 251         if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
 252                 goto error;
 253
 254         opts_len = geneveh->opt_len * 4;
 255         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
 256                                  htons(ETH_P_TEB)))
 257                 goto drop;
 258
 259         gs = rcu_dereference_sk_user_data(sk);
 260         if (!gs)
 261                 goto drop;
 262
 263         geneve_rx(gs, skb);
 264         return 0;
 265
 266 drop:
 267         /* Consume bad packet */
 268         kfree_skb(skb);
 269         return 0;
 270
 271 error:
 272         /* Let the UDP layer deal with the skb */
 273         return 1;
 274 }
 275
 276 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 277                                          __be16 port)
 278 {
 279         struct socket *sock;
 280         struct udp_port_cfg udp_conf;
 281         int err;
 282
 283         memset(&udp_conf, 0, sizeof(udp_conf));
 284
 285         if (ipv6) {
 286                 udp_conf.family = AF_INET6;
 287         } else {
 288                 udp_conf.family = AF_INET;
 289                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 290         }
 291
 292         udp_conf.local_udp_port = port;
 293
 294         /* Open UDP socket */
 295         err = udp_sock_create(net, &udp_conf, &sock);
 296         if (err < 0)
 297                 return ERR_PTR(err);
 298
 299         return sock;
 300 }
 301
 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 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
 322                                            struct sk_buff *skb,
 323                                            struct udp_offload *uoff)
 324 {
 325         struct sk_buff *p, **pp = NULL;
 326         struct genevehdr *gh, *gh2;
 327         unsigned int hlen, gh_len, off_gnv;
 328         const struct packet_offload *ptype;
 329         __be16 type;
 330         int flush = 1;
 331
 332         off_gnv = skb_gro_offset(skb);
 333         hlen = off_gnv + sizeof(*gh);
 334         gh = skb_gro_header_fast(skb, off_gnv);
 335         if (skb_gro_header_hard(skb, hlen)) {
 336                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 337                 if (unlikely(!gh))
 338                         goto out;
 339         }
 340
 341         if (gh->ver != GENEVE_VER || gh->oam)
 342                 goto out;
 343         gh_len = geneve_hlen(gh);
 344
 345         hlen = off_gnv + gh_len;
 346         if (skb_gro_header_hard(skb, hlen)) {
 347                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 348                 if (unlikely(!gh))
 349                         goto out;
 350         }
 351
 352         flush = 0;
 353
 354         for (p = *head; p; p = p->next) {
 355                 if (!NAPI_GRO_CB(p)->same_flow)
 356                         continue;
 357
 358                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 359                 if (gh->opt_len != gh2->opt_len ||
 360                     memcmp(gh, gh2, gh_len)) {
 361                         NAPI_GRO_CB(p)->same_flow = 0;
 362                         continue;
 363                 }
 364         }
 365
 366         type = gh->proto_type;
 367
 368         rcu_read_lock();
 369         ptype = gro_find_receive_by_type(type);
 370         if (!ptype) {
 371                 flush = 1;
 372                 goto out_unlock;
 373         }
 374
 375         skb_gro_pull(skb, gh_len);
 376         skb_gro_postpull_rcsum(skb, gh, gh_len);
 377         pp = ptype->callbacks.gro_receive(head, skb);
 378
 379 out_unlock:
 380         rcu_read_unlock();
 381 out:
 382         NAPI_GRO_CB(skb)->flush |= flush;
 383
 384         return pp;
 385 }
 386
 387 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
 388                                struct udp_offload *uoff)
 389 {
 390         struct genevehdr *gh;
 391         struct packet_offload *ptype;
 392         __be16 type;
 393         int gh_len;
 394         int err = -ENOSYS;
 395
 396         udp_tunnel_gro_complete(skb, nhoff);
 397
 398         gh = (struct genevehdr *)(skb->data + nhoff);
 399         gh_len = geneve_hlen(gh);
 400         type = gh->proto_type;
 401
 402         rcu_read_lock();
 403         ptype = gro_find_complete_by_type(type);
 404         if (ptype)
 405                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 406
 407         rcu_read_unlock();
 408         return err;
 409 }
 410
 411 /* Create new listen socket if needed */
 412 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 413                                                 bool ipv6)
 414 {
 415         struct geneve_net *gn = net_generic(net, geneve_net_id);
 416         struct geneve_sock *gs;
 417         struct socket *sock;
 418         struct udp_tunnel_sock_cfg tunnel_cfg;
 419         int h;
 420
 421         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 422         if (!gs)
 423                 return ERR_PTR(-ENOMEM);
 424
 425         sock = geneve_create_sock(net, ipv6, port);
 426         if (IS_ERR(sock)) {
 427                 kfree(gs);
 428                 return ERR_CAST(sock);
 429         }
 430
 431         gs->sock = sock;
 432         gs->refcnt = 1;
 433         for (h = 0; h < VNI_HASH_SIZE; ++h)
 434                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 435
 436         /* Initialize the geneve udp offloads structure */
 437         gs->udp_offloads.port = port;
 438         gs->udp_offloads.callbacks.gro_receive  = geneve_gro_receive;
 439         gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
 440         geneve_notify_add_rx_port(gs);
 441
 442         /* Mark socket as an encapsulation socket */
 443         tunnel_cfg.sk_user_data = gs;
 444         tunnel_cfg.encap_type = 1;
 445         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 446         tunnel_cfg.encap_destroy = NULL;
 447         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 448         list_add(&gs->list, &gn->sock_list);
 449         return gs;
 450 }
 451
 452 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
 453 {
 454         struct sock *sk = gs->sock->sk;
 455         sa_family_t sa_family = sk->sk_family;
 456
 457         if (sa_family == AF_INET)
 458                 udp_del_offload(&gs->udp_offloads);
 459 }
 460
 461 static void geneve_sock_release(struct geneve_sock *gs)
 462 {
 463         if (--gs->refcnt)
 464                 return;
 465
 466         list_del(&gs->list);
 467         geneve_notify_del_rx_port(gs);
 468         udp_tunnel_sock_release(gs->sock);
 469         kfree_rcu(gs, rcu);
 470 }
 471
 472 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 473                                             __be16 dst_port)
 474 {
 475         struct geneve_sock *gs;
 476
 477         list_for_each_entry(gs, &gn->sock_list, list) {
 478                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 479                     inet_sk(gs->sock->sk)->sk.sk_family == AF_INET) {
 480                         return gs;
 481                 }
 482         }
 483         return NULL;
 484 }
 485
 486 static int geneve_open(struct net_device *dev)
 487 {
 488         struct geneve_dev *geneve = netdev_priv(dev);
 489         struct net *net = geneve->net;
 490         struct geneve_net *gn = net_generic(net, geneve_net_id);
 491         struct geneve_sock *gs;
 492         __u32 hash;
 493
 494         gs = geneve_find_sock(gn, geneve->dst_port);
 495         if (gs) {
 496                 gs->refcnt++;
 497                 goto out;
 498         }
 499
 500         gs = geneve_socket_create(net, geneve->dst_port, false);
 501         if (IS_ERR(gs))
 502                 return PTR_ERR(gs);
 503
 504 out:
 505         gs->collect_md = geneve->collect_md;
 506         geneve->sock = gs;
 507
 508         hash = geneve_net_vni_hash(geneve->vni);
 509         hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
 510         return 0;
 511 }
 512
 513 static int geneve_stop(struct net_device *dev)
 514 {
 515         struct geneve_dev *geneve = netdev_priv(dev);
 516         struct geneve_sock *gs = geneve->sock;
 517
 518         if (!hlist_unhashed(&geneve->hlist))
 519                 hlist_del_rcu(&geneve->hlist);
 520         geneve_sock_release(gs);
 521         return 0;
 522 }
 523
 524 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
 525                             __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
 526                             bool csum)
 527 {
 528         struct genevehdr *gnvh;
 529         int min_headroom;
 530         int err;
 531
 532         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 533                         + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr);
 534         err = skb_cow_head(skb, min_headroom);
 535         if (unlikely(err)) {
 536                 kfree_skb(skb);
 537                 goto free_rt;
 538         }
 539
 540         skb = udp_tunnel_handle_offloads(skb, csum);
 541         if (IS_ERR(skb)) {
 542                 err = PTR_ERR(skb);
 543                 goto free_rt;
 544         }
 545
 546         gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
 547         gnvh->ver = GENEVE_VER;
 548         gnvh->opt_len = opt_len / 4;
 549         gnvh->oam = !!(tun_flags & TUNNEL_OAM);
 550         gnvh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
 551         gnvh->rsvd1 = 0;
 552         memcpy(gnvh->vni, vni, 3);
 553         gnvh->proto_type = htons(ETH_P_TEB);
 554         gnvh->rsvd2 = 0;
 555         memcpy(gnvh->options, opt, opt_len);
 556
 557         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
 558         return 0;
 559
 560 free_rt:
 561         ip_rt_put(rt);
 562         return err;
 563 }
 564
 565 static struct rtable *geneve_get_rt(struct sk_buff *skb,
 566                                     struct net_device *dev,
 567                                     struct flowi4 *fl4,
 568                                     struct ip_tunnel_info *info)
 569 {
 570         struct geneve_dev *geneve = netdev_priv(dev);
 571         struct rtable *rt = NULL;
 572         __u8 tos;
 573
 574         memset(fl4, 0, sizeof(*fl4));
 575         fl4->flowi4_mark = skb->mark;
 576         fl4->flowi4_proto = IPPROTO_UDP;
 577
 578         if (info) {
 579                 fl4->daddr = info->key.u.ipv4.dst;
 580                 fl4->saddr = info->key.u.ipv4.src;
 581                 fl4->flowi4_tos = RT_TOS(info->key.tos);
 582         } else {
 583                 tos = geneve->tos;
 584                 if (tos == 1) {
 585                         const struct iphdr *iip = ip_hdr(skb);
 586
 587                         tos = ip_tunnel_get_dsfield(iip, skb);
 588                 }
 589
 590                 fl4->flowi4_tos = RT_TOS(tos);
 591                 fl4->daddr = geneve->remote.sin_addr.s_addr;
 592         }
 593
 594         rt = ip_route_output_key(geneve->net, fl4);
 595         if (IS_ERR(rt)) {
 596                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 597                 return ERR_PTR(-ENETUNREACH);
 598         }
 599         if (rt->dst.dev == dev) { /* is this necessary? */
 600                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 601                 ip_rt_put(rt);
 602                 return ERR_PTR(-ELOOP);
 603         }
 604         return rt;
 605 }
 606
 607 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 608 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 609 {
 610 #ifdef __BIG_ENDIAN
 611         vni[0] = (__force __u8)(tun_id >> 16);
 612         vni[1] = (__force __u8)(tun_id >> 8);
 613         vni[2] = (__force __u8)tun_id;
 614 #else
 615         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 616         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 617         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 618 #endif
 619 }
 620
 621 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 622 {
 623         struct geneve_dev *geneve = netdev_priv(dev);
 624         struct geneve_sock *gs = geneve->sock;
 625         struct ip_tunnel_info *info = NULL;
 626         struct rtable *rt = NULL;
 627         const struct iphdr *iip; /* interior IP header */
 628         int err = -EINVAL;
 629         struct flowi4 fl4;
 630         __u8 tos, ttl;
 631         __be16 sport;
 632         bool udp_csum;
 633         __be16 df;
 634
 635         if (geneve->collect_md) {
 636                 info = skb_tunnel_info(skb);
 637                 if (unlikely(info && !(info->mode & IP_TUNNEL_INFO_TX))) {
 638                         netdev_dbg(dev, "no tunnel metadata\n");
 639                         goto tx_error;
 640                 }
 641                 if (info && ip_tunnel_info_af(info) != AF_INET)
 642                         goto tx_error;
 643         }
 644
 645         rt = geneve_get_rt(skb, dev, &fl4, info);
 646         if (IS_ERR(rt)) {
 647                 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
 648                 err = PTR_ERR(rt);
 649                 goto tx_error;
 650         }
 651
 652         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 653         skb_reset_mac_header(skb);
 654
 655         iip = ip_hdr(skb);
 656
 657         if (info) {
 658                 const struct ip_tunnel_key *key = &info->key;
 659                 u8 *opts = NULL;
 660                 u8 vni[3];
 661
 662                 tunnel_id_to_vni(key->tun_id, vni);
 663                 if (key->tun_flags & TUNNEL_GENEVE_OPT)
 664                         opts = ip_tunnel_info_opts(info);
 665
 666                 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
 667                 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
 668                                        info->options_len, opts, udp_csum);
 669                 if (unlikely(err))
 670                         goto err;
 671
 672                 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
 673                 ttl = key->ttl;
 674                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 675         } else {
 676                 udp_csum = false;
 677                 err = geneve_build_skb(rt, skb, 0, geneve->vni,
 678                                        0, NULL, udp_csum);
 679                 if (unlikely(err))
 680                         goto err;
 681
 682                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
 683                 ttl = geneve->ttl;
 684                 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
 685                         ttl = 1;
 686                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 687                 df = 0;
 688         }
 689         err = udp_tunnel_xmit_skb(rt, gs->sock->sk, skb, fl4.saddr, fl4.daddr,
 690                                   tos, ttl, df, sport, geneve->dst_port,
 691                                   !net_eq(geneve->net, dev_net(geneve->dev)),
 692                                   !udp_csum);
 693
 694         iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
 695         return NETDEV_TX_OK;
 696
 697 tx_error:
 698         dev_kfree_skb(skb);
 699 err:
 700         if (err == -ELOOP)
 701                 dev->stats.collisions++;
 702         else if (err == -ENETUNREACH)
 703                 dev->stats.tx_carrier_errors++;
 704         else
 705                 dev->stats.tx_errors++;
 706         return NETDEV_TX_OK;
 707 }
 708
 709 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 710 {
 711         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 712         struct geneve_dev *geneve = netdev_priv(dev);
 713         struct rtable *rt;
 714         struct flowi4 fl4;
 715
 716         if (ip_tunnel_info_af(info) != AF_INET)
 717                 return -EINVAL;
 718
 719         rt = geneve_get_rt(skb, dev, &fl4, info);
 720         if (IS_ERR(rt))
 721                 return PTR_ERR(rt);
 722
 723         ip_rt_put(rt);
 724         info->key.u.ipv4.src = fl4.saddr;
 725         info->key.tp_src = udp_flow_src_port(geneve->net, skb,
 726                                              1, USHRT_MAX, true);
 727         info->key.tp_dst = geneve->dst_port;
 728         return 0;
 729 }
 730
 731 static const struct net_device_ops geneve_netdev_ops = {
 732         .ndo_init               = geneve_init,
 733         .ndo_uninit             = geneve_uninit,
 734         .ndo_open               = geneve_open,
 735         .ndo_stop               = geneve_stop,
 736         .ndo_start_xmit         = geneve_xmit,
 737         .ndo_get_stats64        = ip_tunnel_get_stats64,
 738         .ndo_change_mtu         = eth_change_mtu,
 739         .ndo_validate_addr      = eth_validate_addr,
 740         .ndo_set_mac_address    = eth_mac_addr,
 741         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
 742 };
 743
 744 static void geneve_get_drvinfo(struct net_device *dev,
 745                                struct ethtool_drvinfo *drvinfo)
 746 {
 747         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
 748         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
 749 }
 750
 751 static const struct ethtool_ops geneve_ethtool_ops = {
 752         .get_drvinfo    = geneve_get_drvinfo,
 753         .get_link       = ethtool_op_get_link,
 754 };
 755
 756 /* Info for udev, that this is a virtual tunnel endpoint */
 757 static struct device_type geneve_type = {
 758         .name = "geneve",
 759 };
 760
 761 /* Initialize the device structure. */
 762 static void geneve_setup(struct net_device *dev)
 763 {
 764         ether_setup(dev);
 765
 766         dev->netdev_ops = &geneve_netdev_ops;
 767         dev->ethtool_ops = &geneve_ethtool_ops;
 768         dev->destructor = free_netdev;
 769
 770         SET_NETDEV_DEVTYPE(dev, &geneve_type);
 771
 772         dev->features    |= NETIF_F_LLTX;
 773         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
 774         dev->features    |= NETIF_F_RXCSUM;
 775         dev->features    |= NETIF_F_GSO_SOFTWARE;
 776
 777         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
 778         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 779
 780         netif_keep_dst(dev);
 781         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
 782         eth_hw_addr_random(dev);
 783 }
 784
 785 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
 786         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
 787         [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 788         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
 789         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
 790         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
 791         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
 792 };
 793
 794 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
 795 {
 796         if (tb[IFLA_ADDRESS]) {
 797                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 798                         return -EINVAL;
 799
 800                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 801                         return -EADDRNOTAVAIL;
 802         }
 803
 804         if (!data)
 805                 return -EINVAL;
 806
 807         if (data[IFLA_GENEVE_ID]) {
 808                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
 809
 810                 if (vni >= GENEVE_VID_MASK)
 811                         return -ERANGE;
 812         }
 813
 814         return 0;
 815 }
 816
 817 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
 818                                           __be16 dst_port,
 819                                           __be32 rem_addr,
 820                                           u8 vni[],
 821                                           bool *tun_on_same_port,
 822                                           bool *tun_collect_md)
 823 {
 824         struct geneve_dev *geneve, *t;
 825
 826         *tun_on_same_port = false;
 827         *tun_collect_md = false;
 828         t = NULL;
 829         list_for_each_entry(geneve, &gn->geneve_list, next) {
 830                 if (geneve->dst_port == dst_port) {
 831                         *tun_collect_md = geneve->collect_md;
 832                         *tun_on_same_port = true;
 833                 }
 834                 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
 835                     rem_addr == geneve->remote.sin_addr.s_addr &&
 836                     dst_port == geneve->dst_port)
 837                         t = geneve;
 838         }
 839         return t;
 840 }
 841
 842 static int geneve_configure(struct net *net, struct net_device *dev,
 843                             __be32 rem_addr, __u32 vni, __u8 ttl, __u8 tos,
 844                             __be16 dst_port, bool metadata)
 845 {
 846         struct geneve_net *gn = net_generic(net, geneve_net_id);
 847         struct geneve_dev *t, *geneve = netdev_priv(dev);
 848         bool tun_collect_md, tun_on_same_port;
 849         int err;
 850
 851         if (metadata) {
 852                 if (rem_addr || vni || tos || ttl)
 853                         return -EINVAL;
 854         }
 855
 856         geneve->net = net;
 857         geneve->dev = dev;
 858
 859         geneve->vni[0] = (vni & 0x00ff0000) >> 16;
 860         geneve->vni[1] = (vni & 0x0000ff00) >> 8;
 861         geneve->vni[2] =  vni & 0x000000ff;
 862
 863         geneve->remote.sin_addr.s_addr = rem_addr;
 864         if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
 865                 return -EINVAL;
 866
 867         geneve->ttl = ttl;
 868         geneve->tos = tos;
 869         geneve->dst_port = dst_port;
 870         geneve->collect_md = metadata;
 871
 872         t = geneve_find_dev(gn, dst_port, rem_addr, geneve->vni,
 873                             &tun_on_same_port, &tun_collect_md);
 874         if (t)
 875                 return -EBUSY;
 876
 877         if (metadata) {
 878                 if (tun_on_same_port)
 879                         return -EPERM;
 880         } else {
 881                 if (tun_collect_md)
 882                         return -EPERM;
 883         }
 884
 885         err = register_netdevice(dev);
 886         if (err)
 887                 return err;
 888
 889         list_add(&geneve->next, &gn->geneve_list);
 890         return 0;
 891 }
 892
 893 static int geneve_newlink(struct net *net, struct net_device *dev,
 894                           struct nlattr *tb[], struct nlattr *data[])
 895 {
 896         __be16 dst_port = htons(GENEVE_UDP_PORT);
 897         __u8 ttl = 0, tos = 0;
 898         bool metadata = false;
 899         __be32 rem_addr = 0;
 900         __u32 vni = 0;
 901
 902         if (data[IFLA_GENEVE_ID])
 903                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
 904
 905         if (data[IFLA_GENEVE_REMOTE])
 906                 rem_addr = nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
 907
 908         if (data[IFLA_GENEVE_TTL])
 909                 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
 910
 911         if (data[IFLA_GENEVE_TOS])
 912                 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
 913
 914         if (data[IFLA_GENEVE_PORT])
 915                 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
 916
 917         if (data[IFLA_GENEVE_COLLECT_METADATA])
 918                 metadata = true;
 919
 920         return geneve_configure(net, dev, rem_addr, vni,
 921                                 ttl, tos, dst_port, metadata);
 922 }
 923
 924 static void geneve_dellink(struct net_device *dev, struct list_head *head)
 925 {
 926         struct geneve_dev *geneve = netdev_priv(dev);
 927
 928         list_del(&geneve->next);
 929         unregister_netdevice_queue(dev, head);
 930 }
 931
 932 static size_t geneve_get_size(const struct net_device *dev)
 933 {
 934         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
 935                 nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
 936                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
 937                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
 938                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
 939                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
 940                 0;
 941 }
 942
 943 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
 944 {
 945         struct geneve_dev *geneve = netdev_priv(dev);
 946         __u32 vni;
 947
 948         vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
 949         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
 950                 goto nla_put_failure;
 951
 952         if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
 953                             geneve->remote.sin_addr.s_addr))
 954                 goto nla_put_failure;
 955
 956         if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
 957             nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
 958                 goto nla_put_failure;
 959
 960         if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
 961                 goto nla_put_failure;
 962
 963         if (geneve->collect_md) {
 964                 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
 965                         goto nla_put_failure;
 966         }
 967
 968         return 0;
 969
 970 nla_put_failure:
 971         return -EMSGSIZE;
 972 }
 973
 974 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
 975         .kind           = "geneve",
 976         .maxtype        = IFLA_GENEVE_MAX,
 977         .policy         = geneve_policy,
 978         .priv_size      = sizeof(struct geneve_dev),
 979         .setup          = geneve_setup,
 980         .validate       = geneve_validate,
 981         .newlink        = geneve_newlink,
 982         .dellink        = geneve_dellink,
 983         .get_size       = geneve_get_size,
 984         .fill_info      = geneve_fill_info,
 985 };
 986
 987 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
 988                                         u8 name_assign_type, u16 dst_port)
 989 {
 990         struct nlattr *tb[IFLA_MAX + 1];
 991         struct net_device *dev;
 992         int err;
 993
 994         memset(tb, 0, sizeof(tb));
 995         dev = rtnl_create_link(net, name, name_assign_type,
 996                                &geneve_link_ops, tb);
 997         if (IS_ERR(dev))
 998                 return dev;
 999
1000         err = geneve_configure(net, dev, 0, 0, 0, 0, htons(dst_port), true);
1001         if (err) {
1002                 free_netdev(dev);
1003                 return ERR_PTR(err);
1004         }
1005         return dev;
1006 }
1007 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1008
1009 static __net_init int geneve_init_net(struct net *net)
1010 {
1011         struct geneve_net *gn = net_generic(net, geneve_net_id);
1012
1013         INIT_LIST_HEAD(&gn->geneve_list);
1014         INIT_LIST_HEAD(&gn->sock_list);
1015         return 0;
1016 }
1017
1018 static void __net_exit geneve_exit_net(struct net *net)
1019 {
1020         struct geneve_net *gn = net_generic(net, geneve_net_id);
1021         struct geneve_dev *geneve, *next;
1022         struct net_device *dev, *aux;
1023         LIST_HEAD(list);
1024
1025         rtnl_lock();
1026
1027         /* gather any geneve devices that were moved into this ns */
1028         for_each_netdev_safe(net, dev, aux)
1029                 if (dev->rtnl_link_ops == &geneve_link_ops)
1030                         unregister_netdevice_queue(dev, &list);
1031
1032         /* now gather any other geneve devices that were created in this ns */
1033         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1034                 /* If geneve->dev is in the same netns, it was already added
1035                  * to the list by the previous loop.
1036                  */
1037                 if (!net_eq(dev_net(geneve->dev), net))
1038                         unregister_netdevice_queue(geneve->dev, &list);
1039         }
1040
1041         /* unregister the devices gathered above */
1042         unregister_netdevice_many(&list);
1043         rtnl_unlock();
1044 }
1045
1046 static struct pernet_operations geneve_net_ops = {
1047         .init = geneve_init_net,
1048         .exit = geneve_exit_net,
1049         .id   = &geneve_net_id,
1050         .size = sizeof(struct geneve_net),
1051 };
1052
1053 static int __init geneve_init_module(void)
1054 {
1055         int rc;
1056
1057         rc = register_pernet_subsys(&geneve_net_ops);
1058         if (rc)
1059                 goto out1;
1060
1061         rc = rtnl_link_register(&geneve_link_ops);
1062         if (rc)
1063                 goto out2;
1064
1065         return 0;
1066 out2:
1067         unregister_pernet_subsys(&geneve_net_ops);
1068 out1:
1069         return rc;
1070 }
1071 late_initcall(geneve_init_module);
1072
1073 static void __exit geneve_cleanup_module(void)
1074 {
1075         rtnl_link_unregister(&geneve_link_ops);
1076         unregister_pernet_subsys(&geneve_net_ops);
1077 }
1078 module_exit(geneve_cleanup_module);
1079
1080 MODULE_LICENSE("GPL");
1081 MODULE_VERSION(GENEVE_NETDEV_VER);
1082 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1083 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1084 MODULE_ALIAS_RTNL_LINK("geneve");