drivers/net/ipvlan/ipvlan_core.c

   1 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of the GNU General Public License as
   5  * published by the Free Software Foundation; either version 2 of
   6  * the License, or (at your option) any later version.
   7  *
   8  */
   9
  10 #include "ipvlan.h"
  11
  12 static u32 ipvlan_jhash_secret __read_mostly;
  13
  14 void ipvlan_init_secret(void)
  15 {
  16         net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
  17 }
  18
  19 static void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
  20                             unsigned int len, bool success, bool mcast)
  21 {
  22         if (!ipvlan)
  23                 return;
  24
  25         if (likely(success)) {
  26                 struct ipvl_pcpu_stats *pcptr;
  27
  28                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
  29                 u64_stats_update_begin(&pcptr->syncp);
  30                 pcptr->rx_pkts++;
  31                 pcptr->rx_bytes += len;
  32                 if (mcast)
  33                         pcptr->rx_mcast++;
  34                 u64_stats_update_end(&pcptr->syncp);
  35         } else {
  36                 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
  37         }
  38 }
  39
  40 static u8 ipvlan_get_v6_hash(const void *iaddr)
  41 {
  42         const struct in6_addr *ip6_addr = iaddr;
  43
  44         return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
  45                IPVLAN_HASH_MASK;
  46 }
  47
  48 static u8 ipvlan_get_v4_hash(const void *iaddr)
  49 {
  50         const struct in_addr *ip4_addr = iaddr;
  51
  52         return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
  53                IPVLAN_HASH_MASK;
  54 }
  55
  56 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
  57                                                const void *iaddr, bool is_v6)
  58 {
  59         struct ipvl_addr *addr;
  60         u8 hash;
  61
  62         hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
  63                ipvlan_get_v4_hash(iaddr);
  64         hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode) {
  65                 if (is_v6 && addr->atype == IPVL_IPV6 &&
  66                     ipv6_addr_equal(&addr->ip6addr, iaddr))
  67                         return addr;
  68                 else if (!is_v6 && addr->atype == IPVL_IPV4 &&
  69                          addr->ip4addr.s_addr ==
  70                                 ((struct in_addr *)iaddr)->s_addr)
  71                         return addr;
  72         }
  73         return NULL;
  74 }
  75
  76 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
  77 {
  78         struct ipvl_port *port = ipvlan->port;
  79         u8 hash;
  80
  81         hash = (addr->atype == IPVL_IPV6) ?
  82                ipvlan_get_v6_hash(&addr->ip6addr) :
  83                ipvlan_get_v4_hash(&addr->ip4addr);
  84         if (hlist_unhashed(&addr->hlnode))
  85                 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
  86 }
  87
  88 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
  89 {
  90         hlist_del_init_rcu(&addr->hlnode);
  91 }
  92
  93 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
  94                                    const void *iaddr, bool is_v6)
  95 {
  96         struct ipvl_addr *addr;
  97
  98         list_for_each_entry(addr, &ipvlan->addrs, anode) {
  99                 if ((is_v6 && addr->atype == IPVL_IPV6 &&
 100                     ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
 101                     (!is_v6 && addr->atype == IPVL_IPV4 &&
 102                     addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
 103                         return addr;
 104         }
 105         return NULL;
 106 }
 107
 108 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
 109 {
 110         struct ipvl_dev *ipvlan;
 111
 112         ASSERT_RTNL();
 113
 114         list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 115                 if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
 116                         return true;
 117         }
 118         return false;
 119 }
 120
 121 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
 122 {
 123         void *lyr3h = NULL;
 124
 125         switch (skb->protocol) {
 126         case htons(ETH_P_ARP): {
 127                 struct arphdr *arph;
 128
 129                 if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
 130                         return NULL;
 131
 132                 arph = arp_hdr(skb);
 133                 *type = IPVL_ARP;
 134                 lyr3h = arph;
 135                 break;
 136         }
 137         case htons(ETH_P_IP): {
 138                 u32 pktlen;
 139                 struct iphdr *ip4h;
 140
 141                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
 142                         return NULL;
 143
 144                 ip4h = ip_hdr(skb);
 145                 pktlen = ntohs(ip4h->tot_len);
 146                 if (ip4h->ihl < 5 || ip4h->version != 4)
 147                         return NULL;
 148                 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
 149                         return NULL;
 150
 151                 *type = IPVL_IPV4;
 152                 lyr3h = ip4h;
 153                 break;
 154         }
 155         case htons(ETH_P_IPV6): {
 156                 struct ipv6hdr *ip6h;
 157
 158                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
 159                         return NULL;
 160
 161                 ip6h = ipv6_hdr(skb);
 162                 if (ip6h->version != 6)
 163                         return NULL;
 164
 165                 *type = IPVL_IPV6;
 166                 lyr3h = ip6h;
 167                 /* Only Neighbour Solicitation pkts need different treatment */
 168                 if (ipv6_addr_any(&ip6h->saddr) &&
 169                     ip6h->nexthdr == NEXTHDR_ICMP) {
 170                         *type = IPVL_ICMPV6;
 171                         lyr3h = ip6h + 1;
 172                 }
 173                 break;
 174         }
 175         default:
 176                 return NULL;
 177         }
 178
 179         return lyr3h;
 180 }
 181
 182 unsigned int ipvlan_mac_hash(const unsigned char *addr)
 183 {
 184         u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
 185                                ipvlan_jhash_secret);
 186
 187         return hash & IPVLAN_MAC_FILTER_MASK;
 188 }
 189
 190 void ipvlan_process_multicast(struct work_struct *work)
 191 {
 192         struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
 193         struct ethhdr *ethh;
 194         struct ipvl_dev *ipvlan;
 195         struct sk_buff *skb, *nskb;
 196         struct sk_buff_head list;
 197         unsigned int len;
 198         unsigned int mac_hash;
 199         int ret;
 200         u8 pkt_type;
 201         bool hlocal, dlocal;
 202
 203         __skb_queue_head_init(&list);
 204
 205         spin_lock_bh(&port->backlog.lock);
 206         skb_queue_splice_tail_init(&port->backlog, &list);
 207         spin_unlock_bh(&port->backlog.lock);
 208
 209         while ((skb = __skb_dequeue(&list)) != NULL) {
 210                 struct net_device *dev = skb->dev;
 211                 bool consumed = false;
 212
 213                 ethh = eth_hdr(skb);
 214                 hlocal = ether_addr_equal(ethh->h_source, port->dev->dev_addr);
 215                 mac_hash = ipvlan_mac_hash(ethh->h_dest);
 216
 217                 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
 218                         pkt_type = PACKET_BROADCAST;
 219                 else
 220                         pkt_type = PACKET_MULTICAST;
 221
 222                 dlocal = false;
 223                 rcu_read_lock();
 224                 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 225                         if (hlocal && (ipvlan->dev == dev)) {
 226                                 dlocal = true;
 227                                 continue;
 228                         }
 229                         if (!test_bit(mac_hash, ipvlan->mac_filters))
 230                                 continue;
 231                         if (!(ipvlan->dev->flags & IFF_UP))
 232                                 continue;
 233                         ret = NET_RX_DROP;
 234                         len = skb->len + ETH_HLEN;
 235                         nskb = skb_clone(skb, GFP_ATOMIC);
 236                         local_bh_disable();
 237                         if (nskb) {
 238                                 consumed = true;
 239                                 nskb->pkt_type = pkt_type;
 240                                 nskb->dev = ipvlan->dev;
 241                                 if (hlocal)
 242                                         ret = dev_forward_skb(ipvlan->dev, nskb);
 243                                 else
 244                                         ret = netif_rx(nskb);
 245                         }
 246                         ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 247                         local_bh_enable();
 248                 }
 249                 rcu_read_unlock();
 250
 251                 if (dlocal) {
 252                         /* If the packet originated here, send it out. */
 253                         skb->dev = port->dev;
 254                         skb->pkt_type = pkt_type;
 255                         dev_queue_xmit(skb);
 256                 } else {
 257                         if (consumed)
 258                                 consume_skb(skb);
 259                         else
 260                                 kfree_skb(skb);
 261                 }
 262                 if (dev)
 263                         dev_put(dev);
 264         }
 265 }
 266
 267 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 268 {
 269         bool xnet = true;
 270
 271         if (dev)
 272                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 273
 274         skb_scrub_packet(skb, xnet);
 275         if (dev)
 276                 skb->dev = dev;
 277 }
 278
 279 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 280                             bool local)
 281 {
 282         struct ipvl_dev *ipvlan = addr->master;
 283         struct net_device *dev = ipvlan->dev;
 284         unsigned int len;
 285         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 286         bool success = false;
 287         struct sk_buff *skb = *pskb;
 288
 289         len = skb->len + ETH_HLEN;
 290         /* Only packets exchanged between two local slaves need to have
 291          * device-up check as well as skb-share check.
 292          */
 293         if (local) {
 294                 if (unlikely(!(dev->flags & IFF_UP))) {
 295                         kfree_skb(skb);
 296                         goto out;
 297                 }
 298
 299                 skb = skb_share_check(skb, GFP_ATOMIC);
 300                 if (!skb)
 301                         goto out;
 302
 303                 *pskb = skb;
 304         }
 305         ipvlan_skb_crossing_ns(skb, dev);
 306
 307         if (local) {
 308                 skb->pkt_type = PACKET_HOST;
 309                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 310                         success = true;
 311         } else {
 312                 ret = RX_HANDLER_ANOTHER;
 313                 success = true;
 314         }
 315
 316 out:
 317         ipvlan_count_rx(ipvlan, len, success, false);
 318         return ret;
 319 }
 320
 321 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
 322                                             void *lyr3h, int addr_type,
 323                                             bool use_dest)
 324 {
 325         struct ipvl_addr *addr = NULL;
 326
 327         if (addr_type == IPVL_IPV6) {
 328                 struct ipv6hdr *ip6h;
 329                 struct in6_addr *i6addr;
 330
 331                 ip6h = (struct ipv6hdr *)lyr3h;
 332                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 333                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 334         } else if (addr_type == IPVL_ICMPV6) {
 335                 struct nd_msg *ndmh;
 336                 struct in6_addr *i6addr;
 337
 338                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 339                  * are handled to avoid DAD issue.
 340                  */
 341                 ndmh = (struct nd_msg *)lyr3h;
 342                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 343                         i6addr = &ndmh->target;
 344                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 345                 }
 346         } else if (addr_type == IPVL_IPV4) {
 347                 struct iphdr *ip4h;
 348                 __be32 *i4addr;
 349
 350                 ip4h = (struct iphdr *)lyr3h;
 351                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 352                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 353         } else if (addr_type == IPVL_ARP) {
 354                 struct arphdr *arph;
 355                 unsigned char *arp_ptr;
 356                 __be32 dip;
 357
 358                 arph = (struct arphdr *)lyr3h;
 359                 arp_ptr = (unsigned char *)(arph + 1);
 360                 if (use_dest)
 361                         arp_ptr += (2 * port->dev->addr_len) + 4;
 362                 else
 363                         arp_ptr += port->dev->addr_len;
 364
 365                 memcpy(&dip, arp_ptr, 4);
 366                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 367         }
 368
 369         return addr;
 370 }
 371
 372 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 373 {
 374         const struct iphdr *ip4h = ip_hdr(skb);
 375         struct net_device *dev = skb->dev;
 376         struct net *net = dev_net(dev);
 377         struct rtable *rt;
 378         int err, ret = NET_XMIT_DROP;
 379         struct flowi4 fl4 = {
 380                 .flowi4_oif = dev->ifindex,
 381                 .flowi4_tos = RT_TOS(ip4h->tos),
 382                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 383                 .daddr = ip4h->daddr,
 384                 .saddr = ip4h->saddr,
 385         };
 386
 387         rt = ip_route_output_flow(net, &fl4, NULL);
 388         if (IS_ERR(rt))
 389                 goto err;
 390
 391         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 392                 ip_rt_put(rt);
 393                 goto err;
 394         }
 395         skb_dst_set(skb, &rt->dst);
 396         err = ip_local_out(net, skb->sk, skb);
 397         if (unlikely(net_xmit_eval(err)))
 398                 dev->stats.tx_errors++;
 399         else
 400                 ret = NET_XMIT_SUCCESS;
 401         goto out;
 402 err:
 403         dev->stats.tx_errors++;
 404         kfree_skb(skb);
 405 out:
 406         return ret;
 407 }
 408
 409 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 410 {
 411         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 412         struct net_device *dev = skb->dev;
 413         struct net *net = dev_net(dev);
 414         struct dst_entry *dst;
 415         int err, ret = NET_XMIT_DROP;
 416         struct flowi6 fl6 = {
 417                 .flowi6_iif = dev->ifindex,
 418                 .daddr = ip6h->daddr,
 419                 .saddr = ip6h->saddr,
 420                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 421                 .flowlabel = ip6_flowinfo(ip6h),
 422                 .flowi6_mark = skb->mark,
 423                 .flowi6_proto = ip6h->nexthdr,
 424         };
 425
 426         dst = ip6_route_output(net, NULL, &fl6);
 427         if (dst->error) {
 428                 ret = dst->error;
 429                 dst_release(dst);
 430                 goto err;
 431         }
 432         skb_dst_set(skb, dst);
 433         err = ip6_local_out(net, skb->sk, skb);
 434         if (unlikely(net_xmit_eval(err)))
 435                 dev->stats.tx_errors++;
 436         else
 437                 ret = NET_XMIT_SUCCESS;
 438         goto out;
 439 err:
 440         dev->stats.tx_errors++;
 441         kfree_skb(skb);
 442 out:
 443         return ret;
 444 }
 445
 446 static int ipvlan_process_outbound(struct sk_buff *skb)
 447 {
 448         struct ethhdr *ethh = eth_hdr(skb);
 449         int ret = NET_XMIT_DROP;
 450
 451         /* In this mode we dont care about multicast and broadcast traffic */
 452         if (is_multicast_ether_addr(ethh->h_dest)) {
 453                 pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
 454                                     ntohs(skb->protocol));
 455                 kfree_skb(skb);
 456                 goto out;
 457         }
 458
 459         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 460          * will have L2; which need to discarded and processed further
 461          * in the net-ns of the main-device.
 462          */
 463         if (skb_mac_header_was_set(skb)) {
 464                 skb_pull(skb, sizeof(*ethh));
 465                 skb->mac_header = (typeof(skb->mac_header))~0U;
 466                 skb_reset_network_header(skb);
 467         }
 468
 469         if (skb->protocol == htons(ETH_P_IPV6))
 470                 ret = ipvlan_process_v6_outbound(skb);
 471         else if (skb->protocol == htons(ETH_P_IP))
 472                 ret = ipvlan_process_v4_outbound(skb);
 473         else {
 474                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 475                                     ntohs(skb->protocol));
 476                 kfree_skb(skb);
 477         }
 478 out:
 479         return ret;
 480 }
 481
 482 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
 483                                      struct sk_buff *skb)
 484 {
 485         if (skb->protocol == htons(ETH_P_PAUSE)) {
 486                 kfree_skb(skb);
 487                 return;
 488         }
 489
 490         spin_lock(&port->backlog.lock);
 491         if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
 492                 if (skb->dev)
 493                         dev_hold(skb->dev);
 494                 __skb_queue_tail(&port->backlog, skb);
 495                 spin_unlock(&port->backlog.lock);
 496                 schedule_work(&port->wq);
 497         } else {
 498                 spin_unlock(&port->backlog.lock);
 499                 atomic_long_inc(&skb->dev->rx_dropped);
 500                 kfree_skb(skb);
 501         }
 502 }
 503
 504 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 505 {
 506         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 507         void *lyr3h;
 508         struct ipvl_addr *addr;
 509         int addr_type;
 510
 511         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 512         if (!lyr3h)
 513                 goto out;
 514
 515         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 516         if (addr)
 517                 return ipvlan_rcv_frame(addr, &skb, true);
 518
 519 out:
 520         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 521         return ipvlan_process_outbound(skb);
 522 }
 523
 524 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 525 {
 526         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 527         struct ethhdr *eth = eth_hdr(skb);
 528         struct ipvl_addr *addr;
 529         void *lyr3h;
 530         int addr_type;
 531
 532         if (ether_addr_equal(eth->h_dest, eth->h_source)) {
 533                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 534                 if (lyr3h) {
 535                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 536                         if (addr)
 537                                 return ipvlan_rcv_frame(addr, &skb, true);
 538                 }
 539                 skb = skb_share_check(skb, GFP_ATOMIC);
 540                 if (!skb)
 541                         return NET_XMIT_DROP;
 542
 543                 /* Packet definitely does not belong to any of the
 544                  * virtual devices, but the dest is local. So forward
 545                  * the skb for the main-dev. At the RX side we just return
 546                  * RX_PASS for it to be processed further on the stack.
 547                  */
 548                 return dev_forward_skb(ipvlan->phy_dev, skb);
 549
 550         } else if (is_multicast_ether_addr(eth->h_dest)) {
 551                 ipvlan_skb_crossing_ns(skb, NULL);
 552                 ipvlan_multicast_enqueue(ipvlan->port, skb);
 553                 return NET_XMIT_SUCCESS;
 554         }
 555
 556         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 557         return dev_queue_xmit(skb);
 558 }
 559
 560 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 561 {
 562         struct ipvl_dev *ipvlan = netdev_priv(dev);
 563         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 564
 565         if (!port)
 566                 goto out;
 567
 568         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 569                 goto out;
 570
 571         switch(port->mode) {
 572         case IPVLAN_MODE_L2:
 573                 return ipvlan_xmit_mode_l2(skb, dev);
 574         case IPVLAN_MODE_L3:
 575         case IPVLAN_MODE_L3S:
 576                 return ipvlan_xmit_mode_l3(skb, dev);
 577         }
 578
 579         /* Should not reach here */
 580         WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
 581                           port->mode);
 582 out:
 583         kfree_skb(skb);
 584         return NET_XMIT_DROP;
 585 }
 586
 587 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 588 {
 589         struct ethhdr *eth = eth_hdr(skb);
 590         struct ipvl_addr *addr;
 591         void *lyr3h;
 592         int addr_type;
 593
 594         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 595                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 596                 if (!lyr3h)
 597                         return true;
 598
 599                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 600                 if (addr)
 601                         return false;
 602         }
 603
 604         return true;
 605 }
 606
 607 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 608                                                  struct ipvl_port *port)
 609 {
 610         void *lyr3h;
 611         int addr_type;
 612         struct ipvl_addr *addr;
 613         struct sk_buff *skb = *pskb;
 614         rx_handler_result_t ret = RX_HANDLER_PASS;
 615
 616         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 617         if (!lyr3h)
 618                 goto out;
 619
 620         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 621         if (addr)
 622                 ret = ipvlan_rcv_frame(addr, pskb, false);
 623
 624 out:
 625         return ret;
 626 }
 627
 628 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 629                                                  struct ipvl_port *port)
 630 {
 631         struct sk_buff *skb = *pskb;
 632         struct ethhdr *eth = eth_hdr(skb);
 633         rx_handler_result_t ret = RX_HANDLER_PASS;
 634         void *lyr3h;
 635         int addr_type;
 636
 637         if (is_multicast_ether_addr(eth->h_dest)) {
 638                 if (ipvlan_external_frame(skb, port)) {
 639                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 640
 641                         /* External frames are queued for device local
 642                          * distribution, but a copy is given to master
 643                          * straight away to avoid sending duplicates later
 644                          * when work-queue processes this frame. This is
 645                          * achieved by returning RX_HANDLER_PASS.
 646                          */
 647                         if (nskb) {
 648                                 ipvlan_skb_crossing_ns(nskb, NULL);
 649                                 ipvlan_multicast_enqueue(port, nskb);
 650                         }
 651                 }
 652         } else {
 653                 struct ipvl_addr *addr;
 654
 655                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 656                 if (!lyr3h)
 657                         return ret;
 658
 659                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 660                 if (addr)
 661                         ret = ipvlan_rcv_frame(addr, pskb, false);
 662         }
 663
 664         return ret;
 665 }
 666
 667 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 668 {
 669         struct sk_buff *skb = *pskb;
 670         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 671
 672         if (!port)
 673                 return RX_HANDLER_PASS;
 674
 675         switch (port->mode) {
 676         case IPVLAN_MODE_L2:
 677                 return ipvlan_handle_mode_l2(pskb, port);
 678         case IPVLAN_MODE_L3:
 679                 return ipvlan_handle_mode_l3(pskb, port);
 680         case IPVLAN_MODE_L3S:
 681                 return RX_HANDLER_PASS;
 682         }
 683
 684         /* Should not reach here */
 685         WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
 686                           port->mode);
 687         kfree_skb(skb);
 688         return RX_HANDLER_CONSUMED;
 689 }
 690
 691 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
 692                                             struct net_device *dev)
 693 {
 694         struct ipvl_addr *addr = NULL;
 695         struct ipvl_port *port;
 696         void *lyr3h;
 697         int addr_type;
 698
 699         if (!dev || !netif_is_ipvlan_port(dev))
 700                 goto out;
 701
 702         port = ipvlan_port_get_rcu(dev);
 703         if (!port || port->mode != IPVLAN_MODE_L3S)
 704                 goto out;
 705
 706         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 707         if (!lyr3h)
 708                 goto out;
 709
 710         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 711 out:
 712         return addr;
 713 }
 714
 715 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
 716                               u16 proto)
 717 {
 718         struct ipvl_addr *addr;
 719         struct net_device *sdev;
 720
 721         addr = ipvlan_skb_to_addr(skb, dev);
 722         if (!addr)
 723                 goto out;
 724
 725         sdev = addr->master->dev;
 726         switch (proto) {
 727         case AF_INET:
 728         {
 729                 int err;
 730                 struct iphdr *ip4h = ip_hdr(skb);
 731
 732                 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
 733                                            ip4h->tos, sdev);
 734                 if (unlikely(err))
 735                         goto out;
 736                 break;
 737         }
 738         case AF_INET6:
 739         {
 740                 struct dst_entry *dst;
 741                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 742                 int flags = RT6_LOOKUP_F_HAS_SADDR;
 743                 struct flowi6 fl6 = {
 744                         .flowi6_iif   = sdev->ifindex,
 745                         .daddr        = ip6h->daddr,
 746                         .saddr        = ip6h->saddr,
 747                         .flowlabel    = ip6_flowinfo(ip6h),
 748                         .flowi6_mark  = skb->mark,
 749                         .flowi6_proto = ip6h->nexthdr,
 750                 };
 751
 752                 skb_dst_drop(skb);
 753                 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
 754                 skb_dst_set(skb, dst);
 755                 break;
 756         }
 757         default:
 758                 break;
 759         }
 760
 761 out:
 762         return skb;
 763 }
 764
 765 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
 766                              const struct nf_hook_state *state)
 767 {
 768         struct ipvl_addr *addr;
 769         unsigned int len;
 770
 771         addr = ipvlan_skb_to_addr(skb, skb->dev);
 772         if (!addr)
 773                 goto out;
 774
 775         skb->dev = addr->master->dev;
 776         len = skb->len + ETH_HLEN;
 777         ipvlan_count_rx(addr->master, len, true, false);
 778 out:
 779         return NF_ACCEPT;
 780 }