drivers/net/geneve.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * GENEVE: Generic Network Virtualization Encapsulation
   4  *
   5  * Copyright (c) 2015 Red Hat, Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/ethtool.h>
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/hash.h>
  15 #include <net/ipv6_stubs.h>
  16 #include <net/dst_metadata.h>
  17 #include <net/gro_cells.h>
  18 #include <net/rtnetlink.h>
  19 #include <net/geneve.h>
  20 #include <net/gro.h>
  21 #include <net/protocol.h>
  22
  23 #define GENEVE_NETDEV_VER       "0.6"
  24
  25 #define GENEVE_N_VID            (1u << 24)
  26 #define GENEVE_VID_MASK         (GENEVE_N_VID - 1)
  27
  28 #define VNI_HASH_BITS           10
  29 #define VNI_HASH_SIZE           (1<<VNI_HASH_BITS)
  30
  31 static bool log_ecn_error = true;
  32 module_param(log_ecn_error, bool, 0644);
  33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  34
  35 #define GENEVE_VER 0
  36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
  37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
  38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
  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 unsigned int geneve_net_id;
  47
  48 struct geneve_dev_node {
  49         struct hlist_node hlist;
  50         struct geneve_dev *geneve;
  51 };
  52
  53 struct geneve_config {
  54         struct ip_tunnel_info   info;
  55         bool                    collect_md;
  56         bool                    use_udp6_rx_checksums;
  57         bool                    ttl_inherit;
  58         enum ifla_geneve_df     df;
  59         bool                    inner_proto_inherit;
  60 };
  61
  62 /* Pseudo network device */
  63 struct geneve_dev {
  64         struct geneve_dev_node hlist4;  /* vni hash table for IPv4 socket */
  65 #if IS_ENABLED(CONFIG_IPV6)
  66         struct geneve_dev_node hlist6;  /* vni hash table for IPv6 socket */
  67 #endif
  68         struct net         *net;        /* netns for packet i/o */
  69         struct net_device  *dev;        /* netdev for geneve tunnel */
  70         struct geneve_sock __rcu *sock4;        /* IPv4 socket used for geneve tunnel */
  71 #if IS_ENABLED(CONFIG_IPV6)
  72         struct geneve_sock __rcu *sock6;        /* IPv6 socket used for geneve tunnel */
  73 #endif
  74         struct list_head   next;        /* geneve's per namespace list */
  75         struct gro_cells   gro_cells;
  76         struct geneve_config cfg;
  77 };
  78
  79 struct geneve_sock {
  80         bool                    collect_md;
  81         struct list_head        list;
  82         struct socket           *sock;
  83         struct rcu_head         rcu;
  84         int                     refcnt;
  85         struct hlist_head       vni_list[VNI_HASH_SIZE];
  86 };
  87
  88 static inline __u32 geneve_net_vni_hash(u8 vni[3])
  89 {
  90         __u32 vnid;
  91
  92         vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
  93         return hash_32(vnid, VNI_HASH_BITS);
  94 }
  95
  96 static __be64 vni_to_tunnel_id(const __u8 *vni)
  97 {
  98 #ifdef __BIG_ENDIAN
  99         return (vni[0] << 16) | (vni[1] << 8) | vni[2];
 100 #else
 101         return (__force __be64)(((__force u64)vni[0] << 40) |
 102                                 ((__force u64)vni[1] << 48) |
 103                                 ((__force u64)vni[2] << 56));
 104 #endif
 105 }
 106
 107 /* Convert 64 bit tunnel ID to 24 bit VNI. */
 108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
 109 {
 110 #ifdef __BIG_ENDIAN
 111         vni[0] = (__force __u8)(tun_id >> 16);
 112         vni[1] = (__force __u8)(tun_id >> 8);
 113         vni[2] = (__force __u8)tun_id;
 114 #else
 115         vni[0] = (__force __u8)((__force u64)tun_id >> 40);
 116         vni[1] = (__force __u8)((__force u64)tun_id >> 48);
 117         vni[2] = (__force __u8)((__force u64)tun_id >> 56);
 118 #endif
 119 }
 120
 121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
 122 {
 123         return !memcmp(vni, &tun_id[5], 3);
 124 }
 125
 126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
 127 {
 128         return gs->sock->sk->sk_family;
 129 }
 130
 131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
 132                                         __be32 addr, u8 vni[])
 133 {
 134         struct hlist_head *vni_list_head;
 135         struct geneve_dev_node *node;
 136         __u32 hash;
 137
 138         /* Find the device for this VNI */
 139         hash = geneve_net_vni_hash(vni);
 140         vni_list_head = &gs->vni_list[hash];
 141         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 142                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 143                     addr == node->geneve->cfg.info.key.u.ipv4.dst)
 144                         return node->geneve;
 145         }
 146         return NULL;
 147 }
 148
 149 #if IS_ENABLED(CONFIG_IPV6)
 150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
 151                                          struct in6_addr addr6, u8 vni[])
 152 {
 153         struct hlist_head *vni_list_head;
 154         struct geneve_dev_node *node;
 155         __u32 hash;
 156
 157         /* Find the device for this VNI */
 158         hash = geneve_net_vni_hash(vni);
 159         vni_list_head = &gs->vni_list[hash];
 160         hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
 161                 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
 162                     ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
 163                         return node->geneve;
 164         }
 165         return NULL;
 166 }
 167 #endif
 168
 169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
 170 {
 171         return (struct genevehdr *)(udp_hdr(skb) + 1);
 172 }
 173
 174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
 175                                             struct sk_buff *skb)
 176 {
 177         static u8 zero_vni[3];
 178         u8 *vni;
 179
 180         if (geneve_get_sk_family(gs) == AF_INET) {
 181                 struct iphdr *iph;
 182                 __be32 addr;
 183
 184                 iph = ip_hdr(skb); /* outer IP header... */
 185
 186                 if (gs->collect_md) {
 187                         vni = zero_vni;
 188                         addr = 0;
 189                 } else {
 190                         vni = geneve_hdr(skb)->vni;
 191                         addr = iph->saddr;
 192                 }
 193
 194                 return geneve_lookup(gs, addr, vni);
 195 #if IS_ENABLED(CONFIG_IPV6)
 196         } else if (geneve_get_sk_family(gs) == AF_INET6) {
 197                 static struct in6_addr zero_addr6;
 198                 struct ipv6hdr *ip6h;
 199                 struct in6_addr addr6;
 200
 201                 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
 202
 203                 if (gs->collect_md) {
 204                         vni = zero_vni;
 205                         addr6 = zero_addr6;
 206                 } else {
 207                         vni = geneve_hdr(skb)->vni;
 208                         addr6 = ip6h->saddr;
 209                 }
 210
 211                 return geneve6_lookup(gs, addr6, vni);
 212 #endif
 213         }
 214         return NULL;
 215 }
 216
 217 /* geneve receive/decap routine */
 218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
 219                       struct sk_buff *skb)
 220 {
 221         struct genevehdr *gnvh = geneve_hdr(skb);
 222         struct metadata_dst *tun_dst = NULL;
 223         unsigned int len;
 224         int err = 0;
 225         void *oiph;
 226
 227         if (ip_tunnel_collect_metadata() || gs->collect_md) {
 228                 __be16 flags;
 229
 230                 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
 231                         (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
 232
 233                 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
 234                                          vni_to_tunnel_id(gnvh->vni),
 235                                          gnvh->opt_len * 4);
 236                 if (!tun_dst) {
 237                         geneve->dev->stats.rx_dropped++;
 238                         goto drop;
 239                 }
 240                 /* Update tunnel dst according to Geneve options. */
 241                 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
 242                                         gnvh->options, gnvh->opt_len * 4,
 243                                         TUNNEL_GENEVE_OPT);
 244         } else {
 245                 /* Drop packets w/ critical options,
 246                  * since we don't support any...
 247                  */
 248                 if (gnvh->critical) {
 249                         geneve->dev->stats.rx_frame_errors++;
 250                         geneve->dev->stats.rx_errors++;
 251                         goto drop;
 252                 }
 253         }
 254
 255         if (tun_dst)
 256                 skb_dst_set(skb, &tun_dst->dst);
 257
 258         if (gnvh->proto_type == htons(ETH_P_TEB)) {
 259                 skb_reset_mac_header(skb);
 260                 skb->protocol = eth_type_trans(skb, geneve->dev);
 261                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 262
 263                 /* Ignore packet loops (and multicast echo) */
 264                 if (ether_addr_equal(eth_hdr(skb)->h_source,
 265                                      geneve->dev->dev_addr)) {
 266                         geneve->dev->stats.rx_errors++;
 267                         goto drop;
 268                 }
 269         } else {
 270                 skb_reset_mac_header(skb);
 271                 skb->dev = geneve->dev;
 272                 skb->pkt_type = PACKET_HOST;
 273         }
 274
 275         oiph = skb_network_header(skb);
 276         skb_reset_network_header(skb);
 277
 278         if (geneve_get_sk_family(gs) == AF_INET)
 279                 err = IP_ECN_decapsulate(oiph, skb);
 280 #if IS_ENABLED(CONFIG_IPV6)
 281         else
 282                 err = IP6_ECN_decapsulate(oiph, skb);
 283 #endif
 284
 285         if (unlikely(err)) {
 286                 if (log_ecn_error) {
 287                         if (geneve_get_sk_family(gs) == AF_INET)
 288                                 net_info_ratelimited("non-ECT from %pI4 "
 289                                                      "with TOS=%#x\n",
 290                                                      &((struct iphdr *)oiph)->saddr,
 291                                                      ((struct iphdr *)oiph)->tos);
 292 #if IS_ENABLED(CONFIG_IPV6)
 293                         else
 294                                 net_info_ratelimited("non-ECT from %pI6\n",
 295                                                      &((struct ipv6hdr *)oiph)->saddr);
 296 #endif
 297                 }
 298                 if (err > 1) {
 299                         ++geneve->dev->stats.rx_frame_errors;
 300                         ++geneve->dev->stats.rx_errors;
 301                         goto drop;
 302                 }
 303         }
 304
 305         len = skb->len;
 306         err = gro_cells_receive(&geneve->gro_cells, skb);
 307         if (likely(err == NET_RX_SUCCESS))
 308                 dev_sw_netstats_rx_add(geneve->dev, len);
 309
 310         return;
 311 drop:
 312         /* Consume bad packet */
 313         kfree_skb(skb);
 314 }
 315
 316 /* Setup stats when device is created */
 317 static int geneve_init(struct net_device *dev)
 318 {
 319         struct geneve_dev *geneve = netdev_priv(dev);
 320         int err;
 321
 322         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 323         if (!dev->tstats)
 324                 return -ENOMEM;
 325
 326         err = gro_cells_init(&geneve->gro_cells, dev);
 327         if (err) {
 328                 free_percpu(dev->tstats);
 329                 return err;
 330         }
 331
 332         err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
 333         if (err) {
 334                 free_percpu(dev->tstats);
 335                 gro_cells_destroy(&geneve->gro_cells);
 336                 return err;
 337         }
 338         return 0;
 339 }
 340
 341 static void geneve_uninit(struct net_device *dev)
 342 {
 343         struct geneve_dev *geneve = netdev_priv(dev);
 344
 345         dst_cache_destroy(&geneve->cfg.info.dst_cache);
 346         gro_cells_destroy(&geneve->gro_cells);
 347         free_percpu(dev->tstats);
 348 }
 349
 350 /* Callback from net/ipv4/udp.c to receive packets */
 351 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 352 {
 353         struct genevehdr *geneveh;
 354         struct geneve_dev *geneve;
 355         struct geneve_sock *gs;
 356         __be16 inner_proto;
 357         int opts_len;
 358
 359         /* Need UDP and Geneve header to be present */
 360         if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
 361                 goto drop;
 362
 363         /* Return packets with reserved bits set */
 364         geneveh = geneve_hdr(skb);
 365         if (unlikely(geneveh->ver != GENEVE_VER))
 366                 goto drop;
 367
 368         inner_proto = geneveh->proto_type;
 369
 370         if (unlikely((inner_proto != htons(ETH_P_TEB) &&
 371                       inner_proto != htons(ETH_P_IP) &&
 372                       inner_proto != htons(ETH_P_IPV6))))
 373                 goto drop;
 374
 375         gs = rcu_dereference_sk_user_data(sk);
 376         if (!gs)
 377                 goto drop;
 378
 379         geneve = geneve_lookup_skb(gs, skb);
 380         if (!geneve)
 381                 goto drop;
 382
 383         if (unlikely((!geneve->cfg.inner_proto_inherit &&
 384                       inner_proto != htons(ETH_P_TEB)))) {
 385                 geneve->dev->stats.rx_dropped++;
 386                 goto drop;
 387         }
 388
 389         opts_len = geneveh->opt_len * 4;
 390         if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
 391                                  !net_eq(geneve->net, dev_net(geneve->dev)))) {
 392                 geneve->dev->stats.rx_dropped++;
 393                 goto drop;
 394         }
 395
 396         geneve_rx(geneve, gs, skb);
 397         return 0;
 398
 399 drop:
 400         /* Consume bad packet */
 401         kfree_skb(skb);
 402         return 0;
 403 }
 404
 405 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
 406 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
 407 {
 408         struct genevehdr *geneveh;
 409         struct geneve_sock *gs;
 410         u8 zero_vni[3] = { 0 };
 411         u8 *vni = zero_vni;
 412
 413         if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
 414                 return -EINVAL;
 415
 416         geneveh = geneve_hdr(skb);
 417         if (geneveh->ver != GENEVE_VER)
 418                 return -EINVAL;
 419
 420         if (geneveh->proto_type != htons(ETH_P_TEB))
 421                 return -EINVAL;
 422
 423         gs = rcu_dereference_sk_user_data(sk);
 424         if (!gs)
 425                 return -ENOENT;
 426
 427         if (geneve_get_sk_family(gs) == AF_INET) {
 428                 struct iphdr *iph = ip_hdr(skb);
 429                 __be32 addr4 = 0;
 430
 431                 if (!gs->collect_md) {
 432                         vni = geneve_hdr(skb)->vni;
 433                         addr4 = iph->daddr;
 434                 }
 435
 436                 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
 437         }
 438
 439 #if IS_ENABLED(CONFIG_IPV6)
 440         if (geneve_get_sk_family(gs) == AF_INET6) {
 441                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 442                 struct in6_addr addr6;
 443
 444                 memset(&addr6, 0, sizeof(struct in6_addr));
 445
 446                 if (!gs->collect_md) {
 447                         vni = geneve_hdr(skb)->vni;
 448                         addr6 = ip6h->daddr;
 449                 }
 450
 451                 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
 452         }
 453 #endif
 454
 455         return -EPFNOSUPPORT;
 456 }
 457
 458 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
 459                                          __be16 port, bool ipv6_rx_csum)
 460 {
 461         struct socket *sock;
 462         struct udp_port_cfg udp_conf;
 463         int err;
 464
 465         memset(&udp_conf, 0, sizeof(udp_conf));
 466
 467         if (ipv6) {
 468                 udp_conf.family = AF_INET6;
 469                 udp_conf.ipv6_v6only = 1;
 470                 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
 471         } else {
 472                 udp_conf.family = AF_INET;
 473                 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
 474         }
 475
 476         udp_conf.local_udp_port = port;
 477
 478         /* Open UDP socket */
 479         err = udp_sock_create(net, &udp_conf, &sock);
 480         if (err < 0)
 481                 return ERR_PTR(err);
 482
 483         udp_allow_gso(sock->sk);
 484         return sock;
 485 }
 486
 487 static int geneve_hlen(struct genevehdr *gh)
 488 {
 489         return sizeof(*gh) + gh->opt_len * 4;
 490 }
 491
 492 static struct sk_buff *geneve_gro_receive(struct sock *sk,
 493                                           struct list_head *head,
 494                                           struct sk_buff *skb)
 495 {
 496         struct sk_buff *pp = NULL;
 497         struct sk_buff *p;
 498         struct genevehdr *gh, *gh2;
 499         unsigned int hlen, gh_len, off_gnv;
 500         const struct packet_offload *ptype;
 501         __be16 type;
 502         int flush = 1;
 503
 504         off_gnv = skb_gro_offset(skb);
 505         hlen = off_gnv + sizeof(*gh);
 506         gh = skb_gro_header_fast(skb, off_gnv);
 507         if (skb_gro_header_hard(skb, hlen)) {
 508                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 509                 if (unlikely(!gh))
 510                         goto out;
 511         }
 512
 513         if (gh->ver != GENEVE_VER || gh->oam)
 514                 goto out;
 515         gh_len = geneve_hlen(gh);
 516
 517         hlen = off_gnv + gh_len;
 518         if (skb_gro_header_hard(skb, hlen)) {
 519                 gh = skb_gro_header_slow(skb, hlen, off_gnv);
 520                 if (unlikely(!gh))
 521                         goto out;
 522         }
 523
 524         list_for_each_entry(p, head, list) {
 525                 if (!NAPI_GRO_CB(p)->same_flow)
 526                         continue;
 527
 528                 gh2 = (struct genevehdr *)(p->data + off_gnv);
 529                 if (gh->opt_len != gh2->opt_len ||
 530                     memcmp(gh, gh2, gh_len)) {
 531                         NAPI_GRO_CB(p)->same_flow = 0;
 532                         continue;
 533                 }
 534         }
 535
 536         skb_gro_pull(skb, gh_len);
 537         skb_gro_postpull_rcsum(skb, gh, gh_len);
 538         type = gh->proto_type;
 539         if (likely(type == htons(ETH_P_TEB)))
 540                 return call_gro_receive(eth_gro_receive, head, skb);
 541
 542         ptype = gro_find_receive_by_type(type);
 543         if (!ptype)
 544                 goto out;
 545
 546         pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 547         flush = 0;
 548
 549 out:
 550         skb_gro_flush_final(skb, pp, flush);
 551
 552         return pp;
 553 }
 554
 555 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
 556                                int nhoff)
 557 {
 558         struct genevehdr *gh;
 559         struct packet_offload *ptype;
 560         __be16 type;
 561         int gh_len;
 562         int err = -ENOSYS;
 563
 564         gh = (struct genevehdr *)(skb->data + nhoff);
 565         gh_len = geneve_hlen(gh);
 566         type = gh->proto_type;
 567
 568         /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
 569         if (likely(type == htons(ETH_P_TEB)))
 570                 return eth_gro_complete(skb, nhoff + gh_len);
 571
 572         ptype = gro_find_complete_by_type(type);
 573         if (ptype)
 574                 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
 575
 576         skb_set_inner_mac_header(skb, nhoff + gh_len);
 577
 578         return err;
 579 }
 580
 581 /* Create new listen socket if needed */
 582 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
 583                                                 bool ipv6, bool ipv6_rx_csum)
 584 {
 585         struct geneve_net *gn = net_generic(net, geneve_net_id);
 586         struct geneve_sock *gs;
 587         struct socket *sock;
 588         struct udp_tunnel_sock_cfg tunnel_cfg;
 589         int h;
 590
 591         gs = kzalloc(sizeof(*gs), GFP_KERNEL);
 592         if (!gs)
 593                 return ERR_PTR(-ENOMEM);
 594
 595         sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
 596         if (IS_ERR(sock)) {
 597                 kfree(gs);
 598                 return ERR_CAST(sock);
 599         }
 600
 601         gs->sock = sock;
 602         gs->refcnt = 1;
 603         for (h = 0; h < VNI_HASH_SIZE; ++h)
 604                 INIT_HLIST_HEAD(&gs->vni_list[h]);
 605
 606         /* Initialize the geneve udp offloads structure */
 607         udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 608
 609         /* Mark socket as an encapsulation socket */
 610         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 611         tunnel_cfg.sk_user_data = gs;
 612         tunnel_cfg.encap_type = 1;
 613         tunnel_cfg.gro_receive = geneve_gro_receive;
 614         tunnel_cfg.gro_complete = geneve_gro_complete;
 615         tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
 616         tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
 617         tunnel_cfg.encap_destroy = NULL;
 618         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 619         list_add(&gs->list, &gn->sock_list);
 620         return gs;
 621 }
 622
 623 static void __geneve_sock_release(struct geneve_sock *gs)
 624 {
 625         if (!gs || --gs->refcnt)
 626                 return;
 627
 628         list_del(&gs->list);
 629         udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
 630         udp_tunnel_sock_release(gs->sock);
 631         kfree_rcu(gs, rcu);
 632 }
 633
 634 static void geneve_sock_release(struct geneve_dev *geneve)
 635 {
 636         struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
 637 #if IS_ENABLED(CONFIG_IPV6)
 638         struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
 639
 640         rcu_assign_pointer(geneve->sock6, NULL);
 641 #endif
 642
 643         rcu_assign_pointer(geneve->sock4, NULL);
 644         synchronize_net();
 645
 646         __geneve_sock_release(gs4);
 647 #if IS_ENABLED(CONFIG_IPV6)
 648         __geneve_sock_release(gs6);
 649 #endif
 650 }
 651
 652 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
 653                                             sa_family_t family,
 654                                             __be16 dst_port)
 655 {
 656         struct geneve_sock *gs;
 657
 658         list_for_each_entry(gs, &gn->sock_list, list) {
 659                 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
 660                     geneve_get_sk_family(gs) == family) {
 661                         return gs;
 662                 }
 663         }
 664         return NULL;
 665 }
 666
 667 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
 668 {
 669         struct net *net = geneve->net;
 670         struct geneve_net *gn = net_generic(net, geneve_net_id);
 671         struct geneve_dev_node *node;
 672         struct geneve_sock *gs;
 673         __u8 vni[3];
 674         __u32 hash;
 675
 676         gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
 677         if (gs) {
 678                 gs->refcnt++;
 679                 goto out;
 680         }
 681
 682         gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
 683                                   geneve->cfg.use_udp6_rx_checksums);
 684         if (IS_ERR(gs))
 685                 return PTR_ERR(gs);
 686
 687 out:
 688         gs->collect_md = geneve->cfg.collect_md;
 689 #if IS_ENABLED(CONFIG_IPV6)
 690         if (ipv6) {
 691                 rcu_assign_pointer(geneve->sock6, gs);
 692                 node = &geneve->hlist6;
 693         } else
 694 #endif
 695         {
 696                 rcu_assign_pointer(geneve->sock4, gs);
 697                 node = &geneve->hlist4;
 698         }
 699         node->geneve = geneve;
 700
 701         tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
 702         hash = geneve_net_vni_hash(vni);
 703         hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
 704         return 0;
 705 }
 706
 707 static int geneve_open(struct net_device *dev)
 708 {
 709         struct geneve_dev *geneve = netdev_priv(dev);
 710         bool metadata = geneve->cfg.collect_md;
 711         bool ipv4, ipv6;
 712         int ret = 0;
 713
 714         ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
 715         ipv4 = !ipv6 || metadata;
 716 #if IS_ENABLED(CONFIG_IPV6)
 717         if (ipv6) {
 718                 ret = geneve_sock_add(geneve, true);
 719                 if (ret < 0 && ret != -EAFNOSUPPORT)
 720                         ipv4 = false;
 721         }
 722 #endif
 723         if (ipv4)
 724                 ret = geneve_sock_add(geneve, false);
 725         if (ret < 0)
 726                 geneve_sock_release(geneve);
 727
 728         return ret;
 729 }
 730
 731 static int geneve_stop(struct net_device *dev)
 732 {
 733         struct geneve_dev *geneve = netdev_priv(dev);
 734
 735         hlist_del_init_rcu(&geneve->hlist4.hlist);
 736 #if IS_ENABLED(CONFIG_IPV6)
 737         hlist_del_init_rcu(&geneve->hlist6.hlist);
 738 #endif
 739         geneve_sock_release(geneve);
 740         return 0;
 741 }
 742
 743 static void geneve_build_header(struct genevehdr *geneveh,
 744                                 const struct ip_tunnel_info *info,
 745                                 __be16 inner_proto)
 746 {
 747         geneveh->ver = GENEVE_VER;
 748         geneveh->opt_len = info->options_len / 4;
 749         geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
 750         geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
 751         geneveh->rsvd1 = 0;
 752         tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
 753         geneveh->proto_type = inner_proto;
 754         geneveh->rsvd2 = 0;
 755
 756         if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
 757                 ip_tunnel_info_opts_get(geneveh->options, info);
 758 }
 759
 760 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
 761                             const struct ip_tunnel_info *info,
 762                             bool xnet, int ip_hdr_len,
 763                             bool inner_proto_inherit)
 764 {
 765         bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
 766         struct genevehdr *gnvh;
 767         __be16 inner_proto;
 768         int min_headroom;
 769         int err;
 770
 771         skb_reset_mac_header(skb);
 772         skb_scrub_packet(skb, xnet);
 773
 774         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
 775                        GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
 776         err = skb_cow_head(skb, min_headroom);
 777         if (unlikely(err))
 778                 goto free_dst;
 779
 780         err = udp_tunnel_handle_offloads(skb, udp_sum);
 781         if (err)
 782                 goto free_dst;
 783
 784         gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
 785         inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
 786         geneve_build_header(gnvh, info, inner_proto);
 787         skb_set_inner_protocol(skb, inner_proto);
 788         return 0;
 789
 790 free_dst:
 791         dst_release(dst);
 792         return err;
 793 }
 794
 795 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
 796                                        struct net_device *dev,
 797                                        struct geneve_sock *gs4,
 798                                        struct flowi4 *fl4,
 799                                        const struct ip_tunnel_info *info,
 800                                        __be16 dport, __be16 sport)
 801 {
 802         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 803         struct geneve_dev *geneve = netdev_priv(dev);
 804         struct dst_cache *dst_cache;
 805         struct rtable *rt = NULL;
 806         __u8 tos;
 807
 808         if (!gs4)
 809                 return ERR_PTR(-EIO);
 810
 811         memset(fl4, 0, sizeof(*fl4));
 812         fl4->flowi4_mark = skb->mark;
 813         fl4->flowi4_proto = IPPROTO_UDP;
 814         fl4->daddr = info->key.u.ipv4.dst;
 815         fl4->saddr = info->key.u.ipv4.src;
 816         fl4->fl4_dport = dport;
 817         fl4->fl4_sport = sport;
 818         fl4->flowi4_flags = info->key.flow_flags;
 819
 820         tos = info->key.tos;
 821         if ((tos == 1) && !geneve->cfg.collect_md) {
 822                 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 823                 use_cache = false;
 824         }
 825         fl4->flowi4_tos = RT_TOS(tos);
 826
 827         dst_cache = (struct dst_cache *)&info->dst_cache;
 828         if (use_cache) {
 829                 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
 830                 if (rt)
 831                         return rt;
 832         }
 833         rt = ip_route_output_key(geneve->net, fl4);
 834         if (IS_ERR(rt)) {
 835                 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
 836                 return ERR_PTR(-ENETUNREACH);
 837         }
 838         if (rt->dst.dev == dev) { /* is this necessary? */
 839                 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
 840                 ip_rt_put(rt);
 841                 return ERR_PTR(-ELOOP);
 842         }
 843         if (use_cache)
 844                 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
 845         return rt;
 846 }
 847
 848 #if IS_ENABLED(CONFIG_IPV6)
 849 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
 850                                            struct net_device *dev,
 851                                            struct geneve_sock *gs6,
 852                                            struct flowi6 *fl6,
 853                                            const struct ip_tunnel_info *info,
 854                                            __be16 dport, __be16 sport)
 855 {
 856         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
 857         struct geneve_dev *geneve = netdev_priv(dev);
 858         struct dst_entry *dst = NULL;
 859         struct dst_cache *dst_cache;
 860         __u8 prio;
 861
 862         if (!gs6)
 863                 return ERR_PTR(-EIO);
 864
 865         memset(fl6, 0, sizeof(*fl6));
 866         fl6->flowi6_mark = skb->mark;
 867         fl6->flowi6_proto = IPPROTO_UDP;
 868         fl6->daddr = info->key.u.ipv6.dst;
 869         fl6->saddr = info->key.u.ipv6.src;
 870         fl6->fl6_dport = dport;
 871         fl6->fl6_sport = sport;
 872
 873         prio = info->key.tos;
 874         if ((prio == 1) && !geneve->cfg.collect_md) {
 875                 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 876                 use_cache = false;
 877         }
 878
 879         fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
 880                                            info->key.label);
 881         dst_cache = (struct dst_cache *)&info->dst_cache;
 882         if (use_cache) {
 883                 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
 884                 if (dst)
 885                         return dst;
 886         }
 887         dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
 888                                               NULL);
 889         if (IS_ERR(dst)) {
 890                 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
 891                 return ERR_PTR(-ENETUNREACH);
 892         }
 893         if (dst->dev == dev) { /* is this necessary? */
 894                 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
 895                 dst_release(dst);
 896                 return ERR_PTR(-ELOOP);
 897         }
 898
 899         if (use_cache)
 900                 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
 901         return dst;
 902 }
 903 #endif
 904
 905 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
 906                            struct geneve_dev *geneve,
 907                            const struct ip_tunnel_info *info)
 908 {
 909         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
 910         struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
 911         const struct ip_tunnel_key *key = &info->key;
 912         struct rtable *rt;
 913         struct flowi4 fl4;
 914         __u8 tos, ttl;
 915         __be16 df = 0;
 916         __be16 sport;
 917         int err;
 918
 919         if (!pskb_inet_may_pull(skb))
 920                 return -EINVAL;
 921
 922         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
 923         rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
 924                               geneve->cfg.info.key.tp_dst, sport);
 925         if (IS_ERR(rt))
 926                 return PTR_ERR(rt);
 927
 928         err = skb_tunnel_check_pmtu(skb, &rt->dst,
 929                                     GENEVE_IPV4_HLEN + info->options_len,
 930                                     netif_is_any_bridge_port(dev));
 931         if (err < 0) {
 932                 dst_release(&rt->dst);
 933                 return err;
 934         } else if (err) {
 935                 struct ip_tunnel_info *info;
 936
 937                 info = skb_tunnel_info(skb);
 938                 if (info) {
 939                         struct ip_tunnel_info *unclone;
 940
 941                         unclone = skb_tunnel_info_unclone(skb);
 942                         if (unlikely(!unclone)) {
 943                                 dst_release(&rt->dst);
 944                                 return -ENOMEM;
 945                         }
 946
 947                         unclone->key.u.ipv4.dst = fl4.saddr;
 948                         unclone->key.u.ipv4.src = fl4.daddr;
 949                 }
 950
 951                 if (!pskb_may_pull(skb, ETH_HLEN)) {
 952                         dst_release(&rt->dst);
 953                         return -EINVAL;
 954                 }
 955
 956                 skb->protocol = eth_type_trans(skb, geneve->dev);
 957                 __netif_rx(skb);
 958                 dst_release(&rt->dst);
 959                 return -EMSGSIZE;
 960         }
 961
 962         if (geneve->cfg.collect_md) {
 963                 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
 964                 ttl = key->ttl;
 965
 966                 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 967         } else {
 968                 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
 969                 if (geneve->cfg.ttl_inherit)
 970                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
 971                 else
 972                         ttl = key->ttl;
 973                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
 974
 975                 if (geneve->cfg.df == GENEVE_DF_SET) {
 976                         df = htons(IP_DF);
 977                 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
 978                         struct ethhdr *eth = eth_hdr(skb);
 979
 980                         if (ntohs(eth->h_proto) == ETH_P_IPV6) {
 981                                 df = htons(IP_DF);
 982                         } else if (ntohs(eth->h_proto) == ETH_P_IP) {
 983                                 struct iphdr *iph = ip_hdr(skb);
 984
 985                                 if (iph->frag_off & htons(IP_DF))
 986                                         df = htons(IP_DF);
 987                         }
 988                 }
 989         }
 990
 991         err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
 992                                geneve->cfg.inner_proto_inherit);
 993         if (unlikely(err))
 994                 return err;
 995
 996         udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
 997                             tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
 998                             !net_eq(geneve->net, dev_net(geneve->dev)),
 999                             !(info->key.tun_flags & TUNNEL_CSUM));
1000         return 0;
1001 }
1002
1003 #if IS_ENABLED(CONFIG_IPV6)
1004 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
1005                             struct geneve_dev *geneve,
1006                             const struct ip_tunnel_info *info)
1007 {
1008         bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1009         struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1010         const struct ip_tunnel_key *key = &info->key;
1011         struct dst_entry *dst = NULL;
1012         struct flowi6 fl6;
1013         __u8 prio, ttl;
1014         __be16 sport;
1015         int err;
1016
1017         if (!pskb_inet_may_pull(skb))
1018                 return -EINVAL;
1019
1020         sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1021         dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1022                                 geneve->cfg.info.key.tp_dst, sport);
1023         if (IS_ERR(dst))
1024                 return PTR_ERR(dst);
1025
1026         err = skb_tunnel_check_pmtu(skb, dst,
1027                                     GENEVE_IPV6_HLEN + info->options_len,
1028                                     netif_is_any_bridge_port(dev));
1029         if (err < 0) {
1030                 dst_release(dst);
1031                 return err;
1032         } else if (err) {
1033                 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1034
1035                 if (info) {
1036                         struct ip_tunnel_info *unclone;
1037
1038                         unclone = skb_tunnel_info_unclone(skb);
1039                         if (unlikely(!unclone)) {
1040                                 dst_release(dst);
1041                                 return -ENOMEM;
1042                         }
1043
1044                         unclone->key.u.ipv6.dst = fl6.saddr;
1045                         unclone->key.u.ipv6.src = fl6.daddr;
1046                 }
1047
1048                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1049                         dst_release(dst);
1050                         return -EINVAL;
1051                 }
1052
1053                 skb->protocol = eth_type_trans(skb, geneve->dev);
1054                 __netif_rx(skb);
1055                 dst_release(dst);
1056                 return -EMSGSIZE;
1057         }
1058
1059         if (geneve->cfg.collect_md) {
1060                 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1061                 ttl = key->ttl;
1062         } else {
1063                 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1064                                            ip_hdr(skb), skb);
1065                 if (geneve->cfg.ttl_inherit)
1066                         ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1067                 else
1068                         ttl = key->ttl;
1069                 ttl = ttl ? : ip6_dst_hoplimit(dst);
1070         }
1071         err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1072                                geneve->cfg.inner_proto_inherit);
1073         if (unlikely(err))
1074                 return err;
1075
1076         udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1077                              &fl6.saddr, &fl6.daddr, prio, ttl,
1078                              info->key.label, sport, geneve->cfg.info.key.tp_dst,
1079                              !(info->key.tun_flags & TUNNEL_CSUM));
1080         return 0;
1081 }
1082 #endif
1083
1084 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1085 {
1086         struct geneve_dev *geneve = netdev_priv(dev);
1087         struct ip_tunnel_info *info = NULL;
1088         int err;
1089
1090         if (geneve->cfg.collect_md) {
1091                 info = skb_tunnel_info(skb);
1092                 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1093                         netdev_dbg(dev, "no tunnel metadata\n");
1094                         dev_kfree_skb(skb);
1095                         dev->stats.tx_dropped++;
1096                         return NETDEV_TX_OK;
1097                 }
1098         } else {
1099                 info = &geneve->cfg.info;
1100         }
1101
1102         rcu_read_lock();
1103 #if IS_ENABLED(CONFIG_IPV6)
1104         if (info->mode & IP_TUNNEL_INFO_IPV6)
1105                 err = geneve6_xmit_skb(skb, dev, geneve, info);
1106         else
1107 #endif
1108                 err = geneve_xmit_skb(skb, dev, geneve, info);
1109         rcu_read_unlock();
1110
1111         if (likely(!err))
1112                 return NETDEV_TX_OK;
1113
1114         if (err != -EMSGSIZE)
1115                 dev_kfree_skb(skb);
1116
1117         if (err == -ELOOP)
1118                 dev->stats.collisions++;
1119         else if (err == -ENETUNREACH)
1120                 dev->stats.tx_carrier_errors++;
1121
1122         dev->stats.tx_errors++;
1123         return NETDEV_TX_OK;
1124 }
1125
1126 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1127 {
1128         if (new_mtu > dev->max_mtu)
1129                 new_mtu = dev->max_mtu;
1130         else if (new_mtu < dev->min_mtu)
1131                 new_mtu = dev->min_mtu;
1132
1133         dev->mtu = new_mtu;
1134         return 0;
1135 }
1136
1137 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1138 {
1139         struct ip_tunnel_info *info = skb_tunnel_info(skb);
1140         struct geneve_dev *geneve = netdev_priv(dev);
1141         __be16 sport;
1142
1143         if (ip_tunnel_info_af(info) == AF_INET) {
1144                 struct rtable *rt;
1145                 struct flowi4 fl4;
1146
1147                 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1148                 sport = udp_flow_src_port(geneve->net, skb,
1149                                           1, USHRT_MAX, true);
1150
1151                 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1152                                       geneve->cfg.info.key.tp_dst, sport);
1153                 if (IS_ERR(rt))
1154                         return PTR_ERR(rt);
1155
1156                 ip_rt_put(rt);
1157                 info->key.u.ipv4.src = fl4.saddr;
1158 #if IS_ENABLED(CONFIG_IPV6)
1159         } else if (ip_tunnel_info_af(info) == AF_INET6) {
1160                 struct dst_entry *dst;
1161                 struct flowi6 fl6;
1162
1163                 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1164                 sport = udp_flow_src_port(geneve->net, skb,
1165                                           1, USHRT_MAX, true);
1166
1167                 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1168                                         geneve->cfg.info.key.tp_dst, sport);
1169                 if (IS_ERR(dst))
1170                         return PTR_ERR(dst);
1171
1172                 dst_release(dst);
1173                 info->key.u.ipv6.src = fl6.saddr;
1174 #endif
1175         } else {
1176                 return -EINVAL;
1177         }
1178
1179         info->key.tp_src = sport;
1180         info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1181         return 0;
1182 }
1183
1184 static const struct net_device_ops geneve_netdev_ops = {
1185         .ndo_init               = geneve_init,
1186         .ndo_uninit             = geneve_uninit,
1187         .ndo_open               = geneve_open,
1188         .ndo_stop               = geneve_stop,
1189         .ndo_start_xmit         = geneve_xmit,
1190         .ndo_get_stats64        = dev_get_tstats64,
1191         .ndo_change_mtu         = geneve_change_mtu,
1192         .ndo_validate_addr      = eth_validate_addr,
1193         .ndo_set_mac_address    = eth_mac_addr,
1194         .ndo_fill_metadata_dst  = geneve_fill_metadata_dst,
1195 };
1196
1197 static void geneve_get_drvinfo(struct net_device *dev,
1198                                struct ethtool_drvinfo *drvinfo)
1199 {
1200         strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1201         strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1202 }
1203
1204 static const struct ethtool_ops geneve_ethtool_ops = {
1205         .get_drvinfo    = geneve_get_drvinfo,
1206         .get_link       = ethtool_op_get_link,
1207 };
1208
1209 /* Info for udev, that this is a virtual tunnel endpoint */
1210 static struct device_type geneve_type = {
1211         .name = "geneve",
1212 };
1213
1214 /* Calls the ndo_udp_tunnel_add of the caller in order to
1215  * supply the listening GENEVE udp ports. Callers are expected
1216  * to implement the ndo_udp_tunnel_add.
1217  */
1218 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1219 {
1220         struct net *net = dev_net(dev);
1221         struct geneve_net *gn = net_generic(net, geneve_net_id);
1222         struct geneve_sock *gs;
1223
1224         rcu_read_lock();
1225         list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1226                 if (push) {
1227                         udp_tunnel_push_rx_port(dev, gs->sock,
1228                                                 UDP_TUNNEL_TYPE_GENEVE);
1229                 } else {
1230                         udp_tunnel_drop_rx_port(dev, gs->sock,
1231                                                 UDP_TUNNEL_TYPE_GENEVE);
1232                 }
1233         }
1234         rcu_read_unlock();
1235 }
1236
1237 /* Initialize the device structure. */
1238 static void geneve_setup(struct net_device *dev)
1239 {
1240         ether_setup(dev);
1241
1242         dev->netdev_ops = &geneve_netdev_ops;
1243         dev->ethtool_ops = &geneve_ethtool_ops;
1244         dev->needs_free_netdev = true;
1245
1246         SET_NETDEV_DEVTYPE(dev, &geneve_type);
1247
1248         dev->features    |= NETIF_F_LLTX;
1249         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1250         dev->features    |= NETIF_F_RXCSUM;
1251         dev->features    |= NETIF_F_GSO_SOFTWARE;
1252
1253         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1254         dev->hw_features |= NETIF_F_RXCSUM;
1255         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1256
1257         /* MTU range: 68 - (something less than 65535) */
1258         dev->min_mtu = ETH_MIN_MTU;
1259         /* The max_mtu calculation does not take account of GENEVE
1260          * options, to avoid excluding potentially valid
1261          * configurations. This will be further reduced by IPvX hdr size.
1262          */
1263         dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1264
1265         netif_keep_dst(dev);
1266         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1267         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1268         eth_hw_addr_random(dev);
1269 }
1270
1271 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1272         [IFLA_GENEVE_UNSPEC]            = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1273         [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
1274         [IFLA_GENEVE_REMOTE]            = { .len = sizeof_field(struct iphdr, daddr) },
1275         [IFLA_GENEVE_REMOTE6]           = { .len = sizeof(struct in6_addr) },
1276         [IFLA_GENEVE_TTL]               = { .type = NLA_U8 },
1277         [IFLA_GENEVE_TOS]               = { .type = NLA_U8 },
1278         [IFLA_GENEVE_LABEL]             = { .type = NLA_U32 },
1279         [IFLA_GENEVE_PORT]              = { .type = NLA_U16 },
1280         [IFLA_GENEVE_COLLECT_METADATA]  = { .type = NLA_FLAG },
1281         [IFLA_GENEVE_UDP_CSUM]          = { .type = NLA_U8 },
1282         [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1283         [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1284         [IFLA_GENEVE_TTL_INHERIT]       = { .type = NLA_U8 },
1285         [IFLA_GENEVE_DF]                = { .type = NLA_U8 },
1286         [IFLA_GENEVE_INNER_PROTO_INHERIT]       = { .type = NLA_FLAG },
1287 };
1288
1289 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1290                            struct netlink_ext_ack *extack)
1291 {
1292         if (tb[IFLA_ADDRESS]) {
1293                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1294                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1295                                             "Provided link layer address is not Ethernet");
1296                         return -EINVAL;
1297                 }
1298
1299                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1300                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1301                                             "Provided Ethernet address is not unicast");
1302                         return -EADDRNOTAVAIL;
1303                 }
1304         }
1305
1306         if (!data) {
1307                 NL_SET_ERR_MSG(extack,
1308                                "Not enough attributes provided to perform the operation");
1309                 return -EINVAL;
1310         }
1311
1312         if (data[IFLA_GENEVE_ID]) {
1313                 __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
1314
1315                 if (vni >= GENEVE_N_VID) {
1316                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1317                                             "Geneve ID must be lower than 16777216");
1318                         return -ERANGE;
1319                 }
1320         }
1321
1322         if (data[IFLA_GENEVE_DF]) {
1323                 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1324
1325                 if (df < 0 || df > GENEVE_DF_MAX) {
1326                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1327                                             "Invalid DF attribute");
1328                         return -EINVAL;
1329                 }
1330         }
1331
1332         return 0;
1333 }
1334
1335 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1336                                           const struct ip_tunnel_info *info,
1337                                           bool *tun_on_same_port,
1338                                           bool *tun_collect_md)
1339 {
1340         struct geneve_dev *geneve, *t = NULL;
1341
1342         *tun_on_same_port = false;
1343         *tun_collect_md = false;
1344         list_for_each_entry(geneve, &gn->geneve_list, next) {
1345                 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1346                         *tun_collect_md = geneve->cfg.collect_md;
1347                         *tun_on_same_port = true;
1348                 }
1349                 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1350                     info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1351                     !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1352                         t = geneve;
1353         }
1354         return t;
1355 }
1356
1357 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1358 {
1359         return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1360                  info->key.ttl || info->key.label || info->key.tp_src ||
1361                  memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1362 }
1363
1364 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1365                                   struct ip_tunnel_info *b)
1366 {
1367         if (ip_tunnel_info_af(a) == AF_INET)
1368                 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1369         else
1370                 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1371 }
1372
1373 static int geneve_configure(struct net *net, struct net_device *dev,
1374                             struct netlink_ext_ack *extack,
1375                             const struct geneve_config *cfg)
1376 {
1377         struct geneve_net *gn = net_generic(net, geneve_net_id);
1378         struct geneve_dev *t, *geneve = netdev_priv(dev);
1379         const struct ip_tunnel_info *info = &cfg->info;
1380         bool tun_collect_md, tun_on_same_port;
1381         int err, encap_len;
1382
1383         if (cfg->collect_md && !is_tnl_info_zero(info)) {
1384                 NL_SET_ERR_MSG(extack,
1385                                "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1386                 return -EINVAL;
1387         }
1388
1389         geneve->net = net;
1390         geneve->dev = dev;
1391
1392         t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1393         if (t)
1394                 return -EBUSY;
1395
1396         /* make enough headroom for basic scenario */
1397         encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1398         if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1399                 encap_len += sizeof(struct iphdr);
1400                 dev->max_mtu -= sizeof(struct iphdr);
1401         } else {
1402                 encap_len += sizeof(struct ipv6hdr);
1403                 dev->max_mtu -= sizeof(struct ipv6hdr);
1404         }
1405         dev->needed_headroom = encap_len + ETH_HLEN;
1406
1407         if (cfg->collect_md) {
1408                 if (tun_on_same_port) {
1409                         NL_SET_ERR_MSG(extack,
1410                                        "There can be only one externally controlled device on a destination port");
1411                         return -EPERM;
1412                 }
1413         } else {
1414                 if (tun_collect_md) {
1415                         NL_SET_ERR_MSG(extack,
1416                                        "There already exists an externally controlled device on this destination port");
1417                         return -EPERM;
1418                 }
1419         }
1420
1421         dst_cache_reset(&geneve->cfg.info.dst_cache);
1422         memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1423
1424         if (geneve->cfg.inner_proto_inherit) {
1425                 dev->header_ops = NULL;
1426                 dev->type = ARPHRD_NONE;
1427                 dev->hard_header_len = 0;
1428                 dev->addr_len = 0;
1429                 dev->flags = IFF_NOARP;
1430         }
1431
1432         err = register_netdevice(dev);
1433         if (err)
1434                 return err;
1435
1436         list_add(&geneve->next, &gn->geneve_list);
1437         return 0;
1438 }
1439
1440 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1441 {
1442         memset(info, 0, sizeof(*info));
1443         info->key.tp_dst = htons(dst_port);
1444 }
1445
1446 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1447                           struct netlink_ext_ack *extack,
1448                           struct geneve_config *cfg, bool changelink)
1449 {
1450         struct ip_tunnel_info *info = &cfg->info;
1451         int attrtype;
1452
1453         if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1454                 NL_SET_ERR_MSG(extack,
1455                                "Cannot specify both IPv4 and IPv6 Remote addresses");
1456                 return -EINVAL;
1457         }
1458
1459         if (data[IFLA_GENEVE_REMOTE]) {
1460                 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1461                         attrtype = IFLA_GENEVE_REMOTE;
1462                         goto change_notsup;
1463                 }
1464
1465                 info->key.u.ipv4.dst =
1466                         nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1467
1468                 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1469                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1470                                             "Remote IPv4 address cannot be Multicast");
1471                         return -EINVAL;
1472                 }
1473         }
1474
1475         if (data[IFLA_GENEVE_REMOTE6]) {
1476 #if IS_ENABLED(CONFIG_IPV6)
1477                 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1478                         attrtype = IFLA_GENEVE_REMOTE6;
1479                         goto change_notsup;
1480                 }
1481
1482                 info->mode = IP_TUNNEL_INFO_IPV6;
1483                 info->key.u.ipv6.dst =
1484                         nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1485
1486                 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1487                     IPV6_ADDR_LINKLOCAL) {
1488                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1489                                             "Remote IPv6 address cannot be link-local");
1490                         return -EINVAL;
1491                 }
1492                 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1493                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1494                                             "Remote IPv6 address cannot be Multicast");
1495                         return -EINVAL;
1496                 }
1497                 info->key.tun_flags |= TUNNEL_CSUM;
1498                 cfg->use_udp6_rx_checksums = true;
1499 #else
1500                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1501                                     "IPv6 support not enabled in the kernel");
1502                 return -EPFNOSUPPORT;
1503 #endif
1504         }
1505
1506         if (data[IFLA_GENEVE_ID]) {
1507                 __u32 vni;
1508                 __u8 tvni[3];
1509                 __be64 tunid;
1510
1511                 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1512                 tvni[0] = (vni & 0x00ff0000) >> 16;
1513                 tvni[1] = (vni & 0x0000ff00) >> 8;
1514                 tvni[2] =  vni & 0x000000ff;
1515
1516                 tunid = vni_to_tunnel_id(tvni);
1517                 if (changelink && (tunid != info->key.tun_id)) {
1518                         attrtype = IFLA_GENEVE_ID;
1519                         goto change_notsup;
1520                 }
1521                 info->key.tun_id = tunid;
1522         }
1523
1524         if (data[IFLA_GENEVE_TTL_INHERIT]) {
1525                 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1526                         cfg->ttl_inherit = true;
1527                 else
1528                         cfg->ttl_inherit = false;
1529         } else if (data[IFLA_GENEVE_TTL]) {
1530                 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1531                 cfg->ttl_inherit = false;
1532         }
1533
1534         if (data[IFLA_GENEVE_TOS])
1535                 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1536
1537         if (data[IFLA_GENEVE_DF])
1538                 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1539
1540         if (data[IFLA_GENEVE_LABEL]) {
1541                 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1542                                   IPV6_FLOWLABEL_MASK;
1543                 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1544                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1545                                             "Label attribute only applies for IPv6 Geneve devices");
1546                         return -EINVAL;
1547                 }
1548         }
1549
1550         if (data[IFLA_GENEVE_PORT]) {
1551                 if (changelink) {
1552                         attrtype = IFLA_GENEVE_PORT;
1553                         goto change_notsup;
1554                 }
1555                 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1556         }
1557
1558         if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1559                 if (changelink) {
1560                         attrtype = IFLA_GENEVE_COLLECT_METADATA;
1561                         goto change_notsup;
1562                 }
1563                 cfg->collect_md = true;
1564         }
1565
1566         if (data[IFLA_GENEVE_UDP_CSUM]) {
1567                 if (changelink) {
1568                         attrtype = IFLA_GENEVE_UDP_CSUM;
1569                         goto change_notsup;
1570                 }
1571                 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1572                         info->key.tun_flags |= TUNNEL_CSUM;
1573         }
1574
1575         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1576 #if IS_ENABLED(CONFIG_IPV6)
1577                 if (changelink) {
1578                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1579                         goto change_notsup;
1580                 }
1581                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1582                         info->key.tun_flags &= ~TUNNEL_CSUM;
1583 #else
1584                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1585                                     "IPv6 support not enabled in the kernel");
1586                 return -EPFNOSUPPORT;
1587 #endif
1588         }
1589
1590         if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1591 #if IS_ENABLED(CONFIG_IPV6)
1592                 if (changelink) {
1593                         attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1594                         goto change_notsup;
1595                 }
1596                 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1597                         cfg->use_udp6_rx_checksums = false;
1598 #else
1599                 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1600                                     "IPv6 support not enabled in the kernel");
1601                 return -EPFNOSUPPORT;
1602 #endif
1603         }
1604
1605         if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1606                 if (changelink) {
1607                         attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1608                         goto change_notsup;
1609                 }
1610                 cfg->inner_proto_inherit = true;
1611         }
1612
1613         return 0;
1614 change_notsup:
1615         NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1616                             "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1617         return -EOPNOTSUPP;
1618 }
1619
1620 static void geneve_link_config(struct net_device *dev,
1621                                struct ip_tunnel_info *info, struct nlattr *tb[])
1622 {
1623         struct geneve_dev *geneve = netdev_priv(dev);
1624         int ldev_mtu = 0;
1625
1626         if (tb[IFLA_MTU]) {
1627                 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1628                 return;
1629         }
1630
1631         switch (ip_tunnel_info_af(info)) {
1632         case AF_INET: {
1633                 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1634                 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1635
1636                 if (!IS_ERR(rt) && rt->dst.dev) {
1637                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1638                         ip_rt_put(rt);
1639                 }
1640                 break;
1641         }
1642 #if IS_ENABLED(CONFIG_IPV6)
1643         case AF_INET6: {
1644                 struct rt6_info *rt;
1645
1646                 if (!__in6_dev_get(dev))
1647                         break;
1648
1649                 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1650                                 NULL, 0);
1651
1652                 if (rt && rt->dst.dev)
1653                         ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1654                 ip6_rt_put(rt);
1655                 break;
1656         }
1657 #endif
1658         }
1659
1660         if (ldev_mtu <= 0)
1661                 return;
1662
1663         geneve_change_mtu(dev, ldev_mtu - info->options_len);
1664 }
1665
1666 static int geneve_newlink(struct net *net, struct net_device *dev,
1667                           struct nlattr *tb[], struct nlattr *data[],
1668                           struct netlink_ext_ack *extack)
1669 {
1670         struct geneve_config cfg = {
1671                 .df = GENEVE_DF_UNSET,
1672                 .use_udp6_rx_checksums = false,
1673                 .ttl_inherit = false,
1674                 .collect_md = false,
1675         };
1676         int err;
1677
1678         init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1679         err = geneve_nl2info(tb, data, extack, &cfg, false);
1680         if (err)
1681                 return err;
1682
1683         err = geneve_configure(net, dev, extack, &cfg);
1684         if (err)
1685                 return err;
1686
1687         geneve_link_config(dev, &cfg.info, tb);
1688
1689         return 0;
1690 }
1691
1692 /* Quiesces the geneve device data path for both TX and RX.
1693  *
1694  * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1695  * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1696  * to complete for the existing set of in-flight packets to be transmitted,
1697  * then we would have quiesced the transmit data path. All the future packets
1698  * will get dropped until we unquiesce the data path.
1699  *
1700  * On receive geneve dereference the geneve_sock stashed in the socket. So,
1701  * if we set that to NULL under RCU and wait for synchronize_net() to
1702  * complete, then we would have quiesced the receive data path.
1703  */
1704 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1705                            struct geneve_sock **gs6)
1706 {
1707         *gs4 = rtnl_dereference(geneve->sock4);
1708         rcu_assign_pointer(geneve->sock4, NULL);
1709         if (*gs4)
1710                 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1711 #if IS_ENABLED(CONFIG_IPV6)
1712         *gs6 = rtnl_dereference(geneve->sock6);
1713         rcu_assign_pointer(geneve->sock6, NULL);
1714         if (*gs6)
1715                 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1716 #else
1717         *gs6 = NULL;
1718 #endif
1719         synchronize_net();
1720 }
1721
1722 /* Resumes the geneve device data path for both TX and RX. */
1723 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1724                              struct geneve_sock __maybe_unused *gs6)
1725 {
1726         rcu_assign_pointer(geneve->sock4, gs4);
1727         if (gs4)
1728                 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1729 #if IS_ENABLED(CONFIG_IPV6)
1730         rcu_assign_pointer(geneve->sock6, gs6);
1731         if (gs6)
1732                 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1733 #endif
1734         synchronize_net();
1735 }
1736
1737 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1738                              struct nlattr *data[],
1739                              struct netlink_ext_ack *extack)
1740 {
1741         struct geneve_dev *geneve = netdev_priv(dev);
1742         struct geneve_sock *gs4, *gs6;
1743         struct geneve_config cfg;
1744         int err;
1745
1746         /* If the geneve device is configured for metadata (or externally
1747          * controlled, for example, OVS), then nothing can be changed.
1748          */
1749         if (geneve->cfg.collect_md)
1750                 return -EOPNOTSUPP;
1751
1752         /* Start with the existing info. */
1753         memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1754         err = geneve_nl2info(tb, data, extack, &cfg, true);
1755         if (err)
1756                 return err;
1757
1758         if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1759                 dst_cache_reset(&cfg.info.dst_cache);
1760                 geneve_link_config(dev, &cfg.info, tb);
1761         }
1762
1763         geneve_quiesce(geneve, &gs4, &gs6);
1764         memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1765         geneve_unquiesce(geneve, gs4, gs6);
1766
1767         return 0;
1768 }
1769
1770 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1771 {
1772         struct geneve_dev *geneve = netdev_priv(dev);
1773
1774         list_del(&geneve->next);
1775         unregister_netdevice_queue(dev, head);
1776 }
1777
1778 static size_t geneve_get_size(const struct net_device *dev)
1779 {
1780         return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
1781                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1782                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TTL */
1783                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_TOS */
1784                 nla_total_size(sizeof(__u8)) +  /* IFLA_GENEVE_DF */
1785                 nla_total_size(sizeof(__be32)) +  /* IFLA_GENEVE_LABEL */
1786                 nla_total_size(sizeof(__be16)) +  /* IFLA_GENEVE_PORT */
1787                 nla_total_size(0) +      /* IFLA_GENEVE_COLLECT_METADATA */
1788                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1789                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1790                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1791                 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1792                 nla_total_size(0) +      /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1793                 0;
1794 }
1795
1796 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1797 {
1798         struct geneve_dev *geneve = netdev_priv(dev);
1799         struct ip_tunnel_info *info = &geneve->cfg.info;
1800         bool ttl_inherit = geneve->cfg.ttl_inherit;
1801         bool metadata = geneve->cfg.collect_md;
1802         __u8 tmp_vni[3];
1803         __u32 vni;
1804
1805         tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1806         vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1807         if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1808                 goto nla_put_failure;
1809
1810         if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1811                 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1812                                     info->key.u.ipv4.dst))
1813                         goto nla_put_failure;
1814                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1815                                !!(info->key.tun_flags & TUNNEL_CSUM)))
1816                         goto nla_put_failure;
1817
1818 #if IS_ENABLED(CONFIG_IPV6)
1819         } else if (!metadata) {
1820                 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1821                                      &info->key.u.ipv6.dst))
1822                         goto nla_put_failure;
1823                 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1824                                !(info->key.tun_flags & TUNNEL_CSUM)))
1825                         goto nla_put_failure;
1826 #endif
1827         }
1828
1829         if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1830             nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1831             nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1832                 goto nla_put_failure;
1833
1834         if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1835                 goto nla_put_failure;
1836
1837         if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1838                 goto nla_put_failure;
1839
1840         if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1841                 goto nla_put_failure;
1842
1843 #if IS_ENABLED(CONFIG_IPV6)
1844         if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1845                        !geneve->cfg.use_udp6_rx_checksums))
1846                 goto nla_put_failure;
1847 #endif
1848
1849         if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1850                 goto nla_put_failure;
1851
1852         if (geneve->cfg.inner_proto_inherit &&
1853             nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1854                 goto nla_put_failure;
1855
1856         return 0;
1857
1858 nla_put_failure:
1859         return -EMSGSIZE;
1860 }
1861
1862 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1863         .kind           = "geneve",
1864         .maxtype        = IFLA_GENEVE_MAX,
1865         .policy         = geneve_policy,
1866         .priv_size      = sizeof(struct geneve_dev),
1867         .setup          = geneve_setup,
1868         .validate       = geneve_validate,
1869         .newlink        = geneve_newlink,
1870         .changelink     = geneve_changelink,
1871         .dellink        = geneve_dellink,
1872         .get_size       = geneve_get_size,
1873         .fill_info      = geneve_fill_info,
1874 };
1875
1876 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1877                                         u8 name_assign_type, u16 dst_port)
1878 {
1879         struct nlattr *tb[IFLA_MAX + 1];
1880         struct net_device *dev;
1881         LIST_HEAD(list_kill);
1882         int err;
1883         struct geneve_config cfg = {
1884                 .df = GENEVE_DF_UNSET,
1885                 .use_udp6_rx_checksums = true,
1886                 .ttl_inherit = false,
1887                 .collect_md = true,
1888         };
1889
1890         memset(tb, 0, sizeof(tb));
1891         dev = rtnl_create_link(net, name, name_assign_type,
1892                                &geneve_link_ops, tb, NULL);
1893         if (IS_ERR(dev))
1894                 return dev;
1895
1896         init_tnl_info(&cfg.info, dst_port);
1897         err = geneve_configure(net, dev, NULL, &cfg);
1898         if (err) {
1899                 free_netdev(dev);
1900                 return ERR_PTR(err);
1901         }
1902
1903         /* openvswitch users expect packet sizes to be unrestricted,
1904          * so set the largest MTU we can.
1905          */
1906         err = geneve_change_mtu(dev, IP_MAX_MTU);
1907         if (err)
1908                 goto err;
1909
1910         err = rtnl_configure_link(dev, NULL);
1911         if (err < 0)
1912                 goto err;
1913
1914         return dev;
1915 err:
1916         geneve_dellink(dev, &list_kill);
1917         unregister_netdevice_many(&list_kill);
1918         return ERR_PTR(err);
1919 }
1920 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1921
1922 static int geneve_netdevice_event(struct notifier_block *unused,
1923                                   unsigned long event, void *ptr)
1924 {
1925         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1926
1927         if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1928                 geneve_offload_rx_ports(dev, true);
1929         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1930                 geneve_offload_rx_ports(dev, false);
1931
1932         return NOTIFY_DONE;
1933 }
1934
1935 static struct notifier_block geneve_notifier_block __read_mostly = {
1936         .notifier_call = geneve_netdevice_event,
1937 };
1938
1939 static __net_init int geneve_init_net(struct net *net)
1940 {
1941         struct geneve_net *gn = net_generic(net, geneve_net_id);
1942
1943         INIT_LIST_HEAD(&gn->geneve_list);
1944         INIT_LIST_HEAD(&gn->sock_list);
1945         return 0;
1946 }
1947
1948 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1949 {
1950         struct geneve_net *gn = net_generic(net, geneve_net_id);
1951         struct geneve_dev *geneve, *next;
1952         struct net_device *dev, *aux;
1953
1954         /* gather any geneve devices that were moved into this ns */
1955         for_each_netdev_safe(net, dev, aux)
1956                 if (dev->rtnl_link_ops == &geneve_link_ops)
1957                         unregister_netdevice_queue(dev, head);
1958
1959         /* now gather any other geneve devices that were created in this ns */
1960         list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1961                 /* If geneve->dev is in the same netns, it was already added
1962                  * to the list by the previous loop.
1963                  */
1964                 if (!net_eq(dev_net(geneve->dev), net))
1965                         unregister_netdevice_queue(geneve->dev, head);
1966         }
1967 }
1968
1969 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1970 {
1971         struct net *net;
1972         LIST_HEAD(list);
1973
1974         rtnl_lock();
1975         list_for_each_entry(net, net_list, exit_list)
1976                 geneve_destroy_tunnels(net, &list);
1977
1978         /* unregister the devices gathered above */
1979         unregister_netdevice_many(&list);
1980         rtnl_unlock();
1981
1982         list_for_each_entry(net, net_list, exit_list) {
1983                 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1984
1985                 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1986         }
1987 }
1988
1989 static struct pernet_operations geneve_net_ops = {
1990         .init = geneve_init_net,
1991         .exit_batch = geneve_exit_batch_net,
1992         .id   = &geneve_net_id,
1993         .size = sizeof(struct geneve_net),
1994 };
1995
1996 static int __init geneve_init_module(void)
1997 {
1998         int rc;
1999
2000         rc = register_pernet_subsys(&geneve_net_ops);
2001         if (rc)
2002                 goto out1;
2003
2004         rc = register_netdevice_notifier(&geneve_notifier_block);
2005         if (rc)
2006                 goto out2;
2007
2008         rc = rtnl_link_register(&geneve_link_ops);
2009         if (rc)
2010                 goto out3;
2011
2012         return 0;
2013 out3:
2014         unregister_netdevice_notifier(&geneve_notifier_block);
2015 out2:
2016         unregister_pernet_subsys(&geneve_net_ops);
2017 out1:
2018         return rc;
2019 }
2020 late_initcall(geneve_init_module);
2021
2022 static void __exit geneve_cleanup_module(void)
2023 {
2024         rtnl_link_unregister(&geneve_link_ops);
2025         unregister_netdevice_notifier(&geneve_notifier_block);
2026         unregister_pernet_subsys(&geneve_net_ops);
2027 }
2028 module_exit(geneve_cleanup_module);
2029
2030 MODULE_LICENSE("GPL");
2031 MODULE_VERSION(GENEVE_NETDEV_VER);
2032 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2033 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2034 MODULE_ALIAS_RTNL_LINK("geneve");