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                 ethh = eth_hdr(skb);
 211                 hlocal = ether_addr_equal(ethh->h_source, port->dev->dev_addr);
 212                 mac_hash = ipvlan_mac_hash(ethh->h_dest);
 213
 214                 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
 215                         pkt_type = PACKET_BROADCAST;
 216                 else
 217                         pkt_type = PACKET_MULTICAST;
 218
 219                 dlocal = false;
 220                 rcu_read_lock();
 221                 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 222                         if (hlocal && (ipvlan->dev == skb->dev)) {
 223                                 dlocal = true;
 224                                 continue;
 225                         }
 226                         if (!test_bit(mac_hash, ipvlan->mac_filters))
 227                                 continue;
 228
 229                         ret = NET_RX_DROP;
 230                         len = skb->len + ETH_HLEN;
 231                         nskb = skb_clone(skb, GFP_ATOMIC);
 232                         if (!nskb)
 233                                 goto acct;
 234
 235                         nskb->pkt_type = pkt_type;
 236                         nskb->dev = ipvlan->dev;
 237                         if (hlocal)
 238                                 ret = dev_forward_skb(ipvlan->dev, nskb);
 239                         else
 240                                 ret = netif_rx(nskb);
 241 acct:
 242                         ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 243                 }
 244                 rcu_read_unlock();
 245
 246                 if (dlocal) {
 247                         /* If the packet originated here, send it out. */
 248                         skb->dev = port->dev;
 249                         skb->pkt_type = pkt_type;
 250                         dev_queue_xmit(skb);
 251                 } else {
 252                         kfree_skb(skb);
 253                 }
 254                 cond_resched();
 255         }
 256 }
 257
 258 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 259 {
 260         bool xnet = true;
 261
 262         if (dev)
 263                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 264
 265         skb_scrub_packet(skb, xnet);
 266         if (dev)
 267                 skb->dev = dev;
 268 }
 269
 270 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 271                             bool local)
 272 {
 273         struct ipvl_dev *ipvlan = addr->master;
 274         struct net_device *dev = ipvlan->dev;
 275         unsigned int len;
 276         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 277         bool success = false;
 278         struct sk_buff *skb = *pskb;
 279
 280         len = skb->len + ETH_HLEN;
 281         /* Only packets exchanged between two local slaves need to have
 282          * device-up check as well as skb-share check.
 283          */
 284         if (local) {
 285                 if (unlikely(!(dev->flags & IFF_UP))) {
 286                         kfree_skb(skb);
 287                         goto out;
 288                 }
 289
 290                 skb = skb_share_check(skb, GFP_ATOMIC);
 291                 if (!skb)
 292                         goto out;
 293
 294                 *pskb = skb;
 295         }
 296         ipvlan_skb_crossing_ns(skb, dev);
 297
 298         if (local) {
 299                 skb->pkt_type = PACKET_HOST;
 300                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 301                         success = true;
 302         } else {
 303                 if (!ether_addr_equal_64bits(eth_hdr(skb)->h_dest,
 304                                              ipvlan->phy_dev->dev_addr))
 305                         skb->pkt_type = PACKET_OTHERHOST;
 306
 307                 ret = RX_HANDLER_ANOTHER;
 308                 success = true;
 309         }
 310
 311 out:
 312         ipvlan_count_rx(ipvlan, len, success, false);
 313         return ret;
 314 }
 315
 316 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
 317                                             void *lyr3h, int addr_type,
 318                                             bool use_dest)
 319 {
 320         struct ipvl_addr *addr = NULL;
 321
 322         if (addr_type == IPVL_IPV6) {
 323                 struct ipv6hdr *ip6h;
 324                 struct in6_addr *i6addr;
 325
 326                 ip6h = (struct ipv6hdr *)lyr3h;
 327                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 328                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 329         } else if (addr_type == IPVL_ICMPV6) {
 330                 struct nd_msg *ndmh;
 331                 struct in6_addr *i6addr;
 332
 333                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 334                  * are handled to avoid DAD issue.
 335                  */
 336                 ndmh = (struct nd_msg *)lyr3h;
 337                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 338                         i6addr = &ndmh->target;
 339                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 340                 }
 341         } else if (addr_type == IPVL_IPV4) {
 342                 struct iphdr *ip4h;
 343                 __be32 *i4addr;
 344
 345                 ip4h = (struct iphdr *)lyr3h;
 346                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 347                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 348         } else if (addr_type == IPVL_ARP) {
 349                 struct arphdr *arph;
 350                 unsigned char *arp_ptr;
 351                 __be32 dip;
 352
 353                 arph = (struct arphdr *)lyr3h;
 354                 arp_ptr = (unsigned char *)(arph + 1);
 355                 if (use_dest)
 356                         arp_ptr += (2 * port->dev->addr_len) + 4;
 357                 else
 358                         arp_ptr += port->dev->addr_len;
 359
 360                 memcpy(&dip, arp_ptr, 4);
 361                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 362         }
 363
 364         return addr;
 365 }
 366
 367 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 368 {
 369         const struct iphdr *ip4h = ip_hdr(skb);
 370         struct net_device *dev = skb->dev;
 371         struct net *net = dev_net(dev);
 372         struct rtable *rt;
 373         int err, ret = NET_XMIT_DROP;
 374         struct flowi4 fl4 = {
 375                 .flowi4_oif = dev->ifindex,
 376                 .flowi4_tos = RT_TOS(ip4h->tos),
 377                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 378                 .flowi4_mark = skb->mark,
 379                 .daddr = ip4h->daddr,
 380                 .saddr = ip4h->saddr,
 381         };
 382
 383         rt = ip_route_output_flow(net, &fl4, NULL);
 384         if (IS_ERR(rt))
 385                 goto err;
 386
 387         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 388                 ip_rt_put(rt);
 389                 goto err;
 390         }
 391         skb_dst_set(skb, &rt->dst);
 392         err = ip_local_out(net, skb->sk, skb);
 393         if (unlikely(net_xmit_eval(err)))
 394                 dev->stats.tx_errors++;
 395         else
 396                 ret = NET_XMIT_SUCCESS;
 397         goto out;
 398 err:
 399         dev->stats.tx_errors++;
 400         kfree_skb(skb);
 401 out:
 402         return ret;
 403 }
 404
 405 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 406 {
 407         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 408         struct net_device *dev = skb->dev;
 409         struct net *net = dev_net(dev);
 410         struct dst_entry *dst;
 411         int err, ret = NET_XMIT_DROP;
 412         struct flowi6 fl6 = {
 413                 .flowi6_oif = dev->ifindex,
 414                 .daddr = ip6h->daddr,
 415                 .saddr = ip6h->saddr,
 416                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 417                 .flowlabel = ip6_flowinfo(ip6h),
 418                 .flowi6_mark = skb->mark,
 419                 .flowi6_proto = ip6h->nexthdr,
 420         };
 421
 422         dst = ip6_route_output(net, NULL, &fl6);
 423         if (dst->error) {
 424                 ret = dst->error;
 425                 dst_release(dst);
 426                 goto err;
 427         }
 428         skb_dst_set(skb, dst);
 429         err = ip6_local_out(net, skb->sk, skb);
 430         if (unlikely(net_xmit_eval(err)))
 431                 dev->stats.tx_errors++;
 432         else
 433                 ret = NET_XMIT_SUCCESS;
 434         goto out;
 435 err:
 436         dev->stats.tx_errors++;
 437         kfree_skb(skb);
 438 out:
 439         return ret;
 440 }
 441
 442 static int ipvlan_process_outbound(struct sk_buff *skb)
 443 {
 444         int ret = NET_XMIT_DROP;
 445
 446         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 447          * will have L2; which need to discarded and processed further
 448          * in the net-ns of the main-device.
 449          */
 450         if (skb_mac_header_was_set(skb)) {
 451                 /* In this mode we dont care about
 452                  * multicast and broadcast traffic */
 453                 struct ethhdr *ethh = eth_hdr(skb);
 454
 455                 if (is_multicast_ether_addr(ethh->h_dest)) {
 456                         pr_debug_ratelimited(
 457                                 "Dropped {multi|broad}cast of type=[%x]\n",
 458                                 ntohs(skb->protocol));
 459                         kfree_skb(skb);
 460                         goto out;
 461                 }
 462
 463                 skb_pull(skb, sizeof(*ethh));
 464                 skb->mac_header = (typeof(skb->mac_header))~0U;
 465                 skb_reset_network_header(skb);
 466         }
 467
 468         if (skb->protocol == htons(ETH_P_IPV6))
 469                 ret = ipvlan_process_v6_outbound(skb);
 470         else if (skb->protocol == htons(ETH_P_IP))
 471                 ret = ipvlan_process_v4_outbound(skb);
 472         else {
 473                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 474                                     ntohs(skb->protocol));
 475                 kfree_skb(skb);
 476         }
 477 out:
 478         return ret;
 479 }
 480
 481 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
 482                                      struct sk_buff *skb)
 483 {
 484         if (skb->protocol == htons(ETH_P_PAUSE)) {
 485                 kfree_skb(skb);
 486                 return;
 487         }
 488
 489         spin_lock(&port->backlog.lock);
 490         if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
 491                 __skb_queue_tail(&port->backlog, skb);
 492                 spin_unlock(&port->backlog.lock);
 493                 schedule_work(&port->wq);
 494         } else {
 495                 spin_unlock(&port->backlog.lock);
 496                 atomic_long_inc(&skb->dev->rx_dropped);
 497                 kfree_skb(skb);
 498         }
 499 }
 500
 501 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 502 {
 503         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 504         void *lyr3h;
 505         struct ipvl_addr *addr;
 506         int addr_type;
 507
 508         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 509         if (!lyr3h)
 510                 goto out;
 511
 512         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 513         if (addr)
 514                 return ipvlan_rcv_frame(addr, &skb, true);
 515
 516 out:
 517         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 518         return ipvlan_process_outbound(skb);
 519 }
 520
 521 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 522 {
 523         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 524         struct ethhdr *eth = skb_eth_hdr(skb);
 525         struct ipvl_addr *addr;
 526         void *lyr3h;
 527         int addr_type;
 528
 529         if (ether_addr_equal(eth->h_dest, eth->h_source)) {
 530                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 531                 if (lyr3h) {
 532                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 533                         if (addr)
 534                                 return ipvlan_rcv_frame(addr, &skb, true);
 535                 }
 536                 skb = skb_share_check(skb, GFP_ATOMIC);
 537                 if (!skb)
 538                         return NET_XMIT_DROP;
 539
 540                 /* Packet definitely does not belong to any of the
 541                  * virtual devices, but the dest is local. So forward
 542                  * the skb for the main-dev. At the RX side we just return
 543                  * RX_PASS for it to be processed further on the stack.
 544                  */
 545                 return dev_forward_skb(ipvlan->phy_dev, skb);
 546
 547         } else if (is_multicast_ether_addr(eth->h_dest)) {
 548                 skb_reset_mac_header(skb);
 549                 ipvlan_skb_crossing_ns(skb, NULL);
 550                 ipvlan_multicast_enqueue(ipvlan->port, skb);
 551                 return NET_XMIT_SUCCESS;
 552         }
 553
 554         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 555         return dev_queue_xmit(skb);
 556 }
 557
 558 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 559 {
 560         struct ipvl_dev *ipvlan = netdev_priv(dev);
 561         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 562
 563         if (!port)
 564                 goto out;
 565
 566         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 567                 goto out;
 568
 569         switch(port->mode) {
 570         case IPVLAN_MODE_L2:
 571                 return ipvlan_xmit_mode_l2(skb, dev);
 572         case IPVLAN_MODE_L3:
 573         case IPVLAN_MODE_L3S:
 574                 return ipvlan_xmit_mode_l3(skb, dev);
 575         }
 576
 577         /* Should not reach here */
 578         WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
 579                           port->mode);
 580 out:
 581         kfree_skb(skb);
 582         return NET_XMIT_DROP;
 583 }
 584
 585 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 586 {
 587         struct ethhdr *eth = eth_hdr(skb);
 588         struct ipvl_addr *addr;
 589         void *lyr3h;
 590         int addr_type;
 591
 592         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 593                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 594                 if (!lyr3h)
 595                         return true;
 596
 597                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 598                 if (addr)
 599                         return false;
 600         }
 601
 602         return true;
 603 }
 604
 605 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 606                                                  struct ipvl_port *port)
 607 {
 608         void *lyr3h;
 609         int addr_type;
 610         struct ipvl_addr *addr;
 611         struct sk_buff *skb = *pskb;
 612         rx_handler_result_t ret = RX_HANDLER_PASS;
 613
 614         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 615         if (!lyr3h)
 616                 goto out;
 617
 618         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 619         if (addr)
 620                 ret = ipvlan_rcv_frame(addr, pskb, false);
 621
 622 out:
 623         return ret;
 624 }
 625
 626 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 627                                                  struct ipvl_port *port)
 628 {
 629         struct sk_buff *skb = *pskb;
 630         struct ethhdr *eth = eth_hdr(skb);
 631         rx_handler_result_t ret = RX_HANDLER_PASS;
 632         void *lyr3h;
 633         int addr_type;
 634
 635         if (is_multicast_ether_addr(eth->h_dest)) {
 636                 if (ipvlan_external_frame(skb, port)) {
 637                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 638
 639                         /* External frames are queued for device local
 640                          * distribution, but a copy is given to master
 641                          * straight away to avoid sending duplicates later
 642                          * when work-queue processes this frame. This is
 643                          * achieved by returning RX_HANDLER_PASS.
 644                          */
 645                         if (nskb) {
 646                                 ipvlan_skb_crossing_ns(nskb, NULL);
 647                                 ipvlan_multicast_enqueue(port, nskb);
 648                         }
 649                 }
 650         } else {
 651                 struct ipvl_addr *addr;
 652
 653                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 654                 if (!lyr3h)
 655                         return ret;
 656
 657                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 658                 if (addr)
 659                         ret = ipvlan_rcv_frame(addr, pskb, false);
 660         }
 661
 662         return ret;
 663 }
 664
 665 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 666 {
 667         struct sk_buff *skb = *pskb;
 668         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 669
 670         if (!port)
 671                 return RX_HANDLER_PASS;
 672
 673         switch (port->mode) {
 674         case IPVLAN_MODE_L2:
 675                 return ipvlan_handle_mode_l2(pskb, port);
 676         case IPVLAN_MODE_L3:
 677                 return ipvlan_handle_mode_l3(pskb, port);
 678         case IPVLAN_MODE_L3S:
 679                 return RX_HANDLER_PASS;
 680         }
 681
 682         /* Should not reach here */
 683         WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
 684                           port->mode);
 685         kfree_skb(skb);
 686         return RX_HANDLER_CONSUMED;
 687 }
 688
 689 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
 690                                             struct net_device *dev)
 691 {
 692         struct ipvl_addr *addr = NULL;
 693         struct ipvl_port *port;
 694         void *lyr3h;
 695         int addr_type;
 696
 697         if (!dev || !netif_is_ipvlan_port(dev))
 698                 goto out;
 699
 700         port = ipvlan_port_get_rcu(dev);
 701         if (!port || port->mode != IPVLAN_MODE_L3S)
 702                 goto out;
 703
 704         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 705         if (!lyr3h)
 706                 goto out;
 707
 708         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 709 out:
 710         return addr;
 711 }
 712
 713 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
 714                               u16 proto)
 715 {
 716         struct ipvl_addr *addr;
 717         struct net_device *sdev;
 718
 719         addr = ipvlan_skb_to_addr(skb, dev);
 720         if (!addr)
 721                 goto out;
 722
 723         sdev = addr->master->dev;
 724         switch (proto) {
 725         case AF_INET:
 726         {
 727                 int err;
 728                 struct iphdr *ip4h = ip_hdr(skb);
 729
 730                 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
 731                                            ip4h->tos, sdev);
 732                 if (unlikely(err))
 733                         goto out;
 734                 break;
 735         }
 736         case AF_INET6:
 737         {
 738                 struct dst_entry *dst;
 739                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 740                 int flags = RT6_LOOKUP_F_HAS_SADDR;
 741                 struct flowi6 fl6 = {
 742                         .flowi6_iif   = sdev->ifindex,
 743                         .daddr        = ip6h->daddr,
 744                         .saddr        = ip6h->saddr,
 745                         .flowlabel    = ip6_flowinfo(ip6h),
 746                         .flowi6_mark  = skb->mark,
 747                         .flowi6_proto = ip6h->nexthdr,
 748                 };
 749
 750                 skb_dst_drop(skb);
 751                 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
 752                 skb_dst_set(skb, dst);
 753                 break;
 754         }
 755         default:
 756                 break;
 757         }
 758
 759 out:
 760         return skb;
 761 }
 762
 763 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
 764                              const struct nf_hook_state *state)
 765 {
 766         struct ipvl_addr *addr;
 767         unsigned int len;
 768
 769         addr = ipvlan_skb_to_addr(skb, skb->dev);
 770         if (!addr)
 771                 goto out;
 772
 773         skb->dev = addr->master->dev;
 774         len = skb->len + ETH_HLEN;
 775         ipvlan_count_rx(addr->master, len, true, false);
 776 out:
 777         return NF_ACCEPT;
 778 }