drivers/net/ipvlan/ipvlan_core.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Copyright (c) 2014 Mahesh Bandewar <[email protected]>
   3  */
   4
   5 #include "ipvlan.h"
   6
   7 static u32 ipvlan_jhash_secret __read_mostly;
   8
   9 void ipvlan_init_secret(void)
  10 {
  11         net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
  12 }
  13
  14 void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
  15                             unsigned int len, bool success, bool mcast)
  16 {
  17         if (likely(success)) {
  18                 struct ipvl_pcpu_stats *pcptr;
  19
  20                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
  21                 u64_stats_update_begin(&pcptr->syncp);
  22                 u64_stats_inc(&pcptr->rx_pkts);
  23                 u64_stats_add(&pcptr->rx_bytes, len);
  24                 if (mcast)
  25                         u64_stats_inc(&pcptr->rx_mcast);
  26                 u64_stats_update_end(&pcptr->syncp);
  27         } else {
  28                 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
  29         }
  30 }
  31 EXPORT_SYMBOL_GPL(ipvlan_count_rx);
  32
  33 #if IS_ENABLED(CONFIG_IPV6)
  34 static u8 ipvlan_get_v6_hash(const void *iaddr)
  35 {
  36         const struct in6_addr *ip6_addr = iaddr;
  37
  38         return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
  39                IPVLAN_HASH_MASK;
  40 }
  41 #else
  42 static u8 ipvlan_get_v6_hash(const void *iaddr)
  43 {
  44         return 0;
  45 }
  46 #endif
  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 bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
  57 {
  58         if (!is_v6 && addr->atype == IPVL_IPV4) {
  59                 struct in_addr *i4addr = (struct in_addr *)iaddr;
  60
  61                 return addr->ip4addr.s_addr == i4addr->s_addr;
  62 #if IS_ENABLED(CONFIG_IPV6)
  63         } else if (is_v6 && addr->atype == IPVL_IPV6) {
  64                 struct in6_addr *i6addr = (struct in6_addr *)iaddr;
  65
  66                 return ipv6_addr_equal(&addr->ip6addr, i6addr);
  67 #endif
  68         }
  69
  70         return false;
  71 }
  72
  73 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
  74                                                const void *iaddr, bool is_v6)
  75 {
  76         struct ipvl_addr *addr;
  77         u8 hash;
  78
  79         hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
  80                ipvlan_get_v4_hash(iaddr);
  81         hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
  82                 if (addr_equal(is_v6, addr, iaddr))
  83                         return addr;
  84         return NULL;
  85 }
  86
  87 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
  88 {
  89         struct ipvl_port *port = ipvlan->port;
  90         u8 hash;
  91
  92         hash = (addr->atype == IPVL_IPV6) ?
  93                ipvlan_get_v6_hash(&addr->ip6addr) :
  94                ipvlan_get_v4_hash(&addr->ip4addr);
  95         if (hlist_unhashed(&addr->hlnode))
  96                 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
  97 }
  98
  99 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
 100 {
 101         hlist_del_init_rcu(&addr->hlnode);
 102 }
 103
 104 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
 105                                    const void *iaddr, bool is_v6)
 106 {
 107         struct ipvl_addr *addr, *ret = NULL;
 108
 109         rcu_read_lock();
 110         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) {
 111                 if (addr_equal(is_v6, addr, iaddr)) {
 112                         ret = addr;
 113                         break;
 114                 }
 115         }
 116         rcu_read_unlock();
 117         return ret;
 118 }
 119
 120 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
 121 {
 122         struct ipvl_dev *ipvlan;
 123         bool ret = false;
 124
 125         rcu_read_lock();
 126         list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 127                 if (ipvlan_find_addr(ipvlan, iaddr, is_v6)) {
 128                         ret = true;
 129                         break;
 130                 }
 131         }
 132         rcu_read_unlock();
 133         return ret;
 134 }
 135
 136 void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
 137 {
 138         void *lyr3h = NULL;
 139
 140         switch (skb->protocol) {
 141         case htons(ETH_P_ARP): {
 142                 struct arphdr *arph;
 143
 144                 if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
 145                         return NULL;
 146
 147                 arph = arp_hdr(skb);
 148                 *type = IPVL_ARP;
 149                 lyr3h = arph;
 150                 break;
 151         }
 152         case htons(ETH_P_IP): {
 153                 u32 pktlen;
 154                 struct iphdr *ip4h;
 155
 156                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
 157                         return NULL;
 158
 159                 ip4h = ip_hdr(skb);
 160                 pktlen = skb_ip_totlen(skb);
 161                 if (ip4h->ihl < 5 || ip4h->version != 4)
 162                         return NULL;
 163                 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
 164                         return NULL;
 165
 166                 *type = IPVL_IPV4;
 167                 lyr3h = ip4h;
 168                 break;
 169         }
 170 #if IS_ENABLED(CONFIG_IPV6)
 171         case htons(ETH_P_IPV6): {
 172                 struct ipv6hdr *ip6h;
 173
 174                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
 175                         return NULL;
 176
 177                 ip6h = ipv6_hdr(skb);
 178                 if (ip6h->version != 6)
 179                         return NULL;
 180
 181                 *type = IPVL_IPV6;
 182                 lyr3h = ip6h;
 183                 /* Only Neighbour Solicitation pkts need different treatment */
 184                 if (ipv6_addr_any(&ip6h->saddr) &&
 185                     ip6h->nexthdr == NEXTHDR_ICMP) {
 186                         struct icmp6hdr *icmph;
 187
 188                         if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
 189                                 return NULL;
 190
 191                         ip6h = ipv6_hdr(skb);
 192                         icmph = (struct icmp6hdr *)(ip6h + 1);
 193
 194                         if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 195                                 /* Need to access the ipv6 address in body */
 196                                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
 197                                                 + sizeof(struct in6_addr))))
 198                                         return NULL;
 199
 200                                 ip6h = ipv6_hdr(skb);
 201                                 icmph = (struct icmp6hdr *)(ip6h + 1);
 202                         }
 203
 204                         *type = IPVL_ICMPV6;
 205                         lyr3h = icmph;
 206                 }
 207                 break;
 208         }
 209 #endif
 210         default:
 211                 return NULL;
 212         }
 213
 214         return lyr3h;
 215 }
 216
 217 unsigned int ipvlan_mac_hash(const unsigned char *addr)
 218 {
 219         u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
 220                                ipvlan_jhash_secret);
 221
 222         return hash & IPVLAN_MAC_FILTER_MASK;
 223 }
 224
 225 void ipvlan_process_multicast(struct work_struct *work)
 226 {
 227         struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
 228         struct ethhdr *ethh;
 229         struct ipvl_dev *ipvlan;
 230         struct sk_buff *skb, *nskb;
 231         struct sk_buff_head list;
 232         unsigned int len;
 233         unsigned int mac_hash;
 234         int ret;
 235         u8 pkt_type;
 236         bool tx_pkt;
 237
 238         __skb_queue_head_init(&list);
 239
 240         spin_lock_bh(&port->backlog.lock);
 241         skb_queue_splice_tail_init(&port->backlog, &list);
 242         spin_unlock_bh(&port->backlog.lock);
 243
 244         while ((skb = __skb_dequeue(&list)) != NULL) {
 245                 struct net_device *dev = skb->dev;
 246                 bool consumed = false;
 247
 248                 ethh = eth_hdr(skb);
 249                 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
 250                 mac_hash = ipvlan_mac_hash(ethh->h_dest);
 251
 252                 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
 253                         pkt_type = PACKET_BROADCAST;
 254                 else
 255                         pkt_type = PACKET_MULTICAST;
 256
 257                 rcu_read_lock();
 258                 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 259                         if (tx_pkt && (ipvlan->dev == skb->dev))
 260                                 continue;
 261                         if (!test_bit(mac_hash, ipvlan->mac_filters))
 262                                 continue;
 263                         if (!(ipvlan->dev->flags & IFF_UP))
 264                                 continue;
 265                         ret = NET_RX_DROP;
 266                         len = skb->len + ETH_HLEN;
 267                         nskb = skb_clone(skb, GFP_ATOMIC);
 268                         local_bh_disable();
 269                         if (nskb) {
 270                                 consumed = true;
 271                                 nskb->pkt_type = pkt_type;
 272                                 nskb->dev = ipvlan->dev;
 273                                 if (tx_pkt)
 274                                         ret = dev_forward_skb(ipvlan->dev, nskb);
 275                                 else
 276                                         ret = netif_rx(nskb);
 277                         }
 278                         ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 279                         local_bh_enable();
 280                 }
 281                 rcu_read_unlock();
 282
 283                 if (tx_pkt) {
 284                         /* If the packet originated here, send it out. */
 285                         skb->dev = port->dev;
 286                         skb->pkt_type = pkt_type;
 287                         dev_queue_xmit(skb);
 288                 } else {
 289                         if (consumed)
 290                                 consume_skb(skb);
 291                         else
 292                                 kfree_skb(skb);
 293                 }
 294                 dev_put(dev);
 295                 cond_resched();
 296         }
 297 }
 298
 299 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 300 {
 301         bool xnet = true;
 302
 303         if (dev)
 304                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 305
 306         skb_scrub_packet(skb, xnet);
 307         if (dev)
 308                 skb->dev = dev;
 309 }
 310
 311 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 312                             bool local)
 313 {
 314         struct ipvl_dev *ipvlan = addr->master;
 315         struct net_device *dev = ipvlan->dev;
 316         unsigned int len;
 317         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 318         bool success = false;
 319         struct sk_buff *skb = *pskb;
 320
 321         len = skb->len + ETH_HLEN;
 322         /* Only packets exchanged between two local slaves need to have
 323          * device-up check as well as skb-share check.
 324          */
 325         if (local) {
 326                 if (unlikely(!(dev->flags & IFF_UP))) {
 327                         kfree_skb(skb);
 328                         goto out;
 329                 }
 330
 331                 skb = skb_share_check(skb, GFP_ATOMIC);
 332                 if (!skb)
 333                         goto out;
 334
 335                 *pskb = skb;
 336         }
 337
 338         if (local) {
 339                 skb->pkt_type = PACKET_HOST;
 340                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 341                         success = true;
 342         } else {
 343                 skb->dev = dev;
 344                 ret = RX_HANDLER_ANOTHER;
 345                 success = true;
 346         }
 347
 348 out:
 349         ipvlan_count_rx(ipvlan, len, success, false);
 350         return ret;
 351 }
 352
 353 struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, void *lyr3h,
 354                                      int addr_type, bool use_dest)
 355 {
 356         struct ipvl_addr *addr = NULL;
 357
 358         switch (addr_type) {
 359 #if IS_ENABLED(CONFIG_IPV6)
 360         case IPVL_IPV6: {
 361                 struct ipv6hdr *ip6h;
 362                 struct in6_addr *i6addr;
 363
 364                 ip6h = (struct ipv6hdr *)lyr3h;
 365                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 366                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 367                 break;
 368         }
 369         case IPVL_ICMPV6: {
 370                 struct nd_msg *ndmh;
 371                 struct in6_addr *i6addr;
 372
 373                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 374                  * are handled to avoid DAD issue.
 375                  */
 376                 ndmh = (struct nd_msg *)lyr3h;
 377                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 378                         i6addr = &ndmh->target;
 379                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 380                 }
 381                 break;
 382         }
 383 #endif
 384         case IPVL_IPV4: {
 385                 struct iphdr *ip4h;
 386                 __be32 *i4addr;
 387
 388                 ip4h = (struct iphdr *)lyr3h;
 389                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 390                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 391                 break;
 392         }
 393         case IPVL_ARP: {
 394                 struct arphdr *arph;
 395                 unsigned char *arp_ptr;
 396                 __be32 dip;
 397
 398                 arph = (struct arphdr *)lyr3h;
 399                 arp_ptr = (unsigned char *)(arph + 1);
 400                 if (use_dest)
 401                         arp_ptr += (2 * port->dev->addr_len) + 4;
 402                 else
 403                         arp_ptr += port->dev->addr_len;
 404
 405                 memcpy(&dip, arp_ptr, 4);
 406                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 407                 break;
 408         }
 409         }
 410
 411         return addr;
 412 }
 413
 414 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 415 {
 416         const struct iphdr *ip4h = ip_hdr(skb);
 417         struct net_device *dev = skb->dev;
 418         struct net *net = dev_net(dev);
 419         struct rtable *rt;
 420         int err, ret = NET_XMIT_DROP;
 421         struct flowi4 fl4 = {
 422                 .flowi4_oif = dev->ifindex,
 423                 .flowi4_tos = RT_TOS(ip4h->tos),
 424                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 425                 .flowi4_mark = skb->mark,
 426                 .daddr = ip4h->daddr,
 427                 .saddr = ip4h->saddr,
 428         };
 429
 430         rt = ip_route_output_flow(net, &fl4, NULL);
 431         if (IS_ERR(rt))
 432                 goto err;
 433
 434         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 435                 ip_rt_put(rt);
 436                 goto err;
 437         }
 438         skb_dst_set(skb, &rt->dst);
 439
 440         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 441
 442         err = ip_local_out(net, skb->sk, skb);
 443         if (unlikely(net_xmit_eval(err)))
 444                 DEV_STATS_INC(dev, tx_errors);
 445         else
 446                 ret = NET_XMIT_SUCCESS;
 447         goto out;
 448 err:
 449         DEV_STATS_INC(dev, tx_errors);
 450         kfree_skb(skb);
 451 out:
 452         return ret;
 453 }
 454
 455 #if IS_ENABLED(CONFIG_IPV6)
 456 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 457 {
 458         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 459         struct net_device *dev = skb->dev;
 460         struct net *net = dev_net(dev);
 461         struct dst_entry *dst;
 462         int err, ret = NET_XMIT_DROP;
 463         struct flowi6 fl6 = {
 464                 .flowi6_oif = dev->ifindex,
 465                 .daddr = ip6h->daddr,
 466                 .saddr = ip6h->saddr,
 467                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 468                 .flowlabel = ip6_flowinfo(ip6h),
 469                 .flowi6_mark = skb->mark,
 470                 .flowi6_proto = ip6h->nexthdr,
 471         };
 472
 473         dst = ip6_route_output(net, NULL, &fl6);
 474         if (dst->error) {
 475                 ret = dst->error;
 476                 dst_release(dst);
 477                 goto err;
 478         }
 479         skb_dst_set(skb, dst);
 480
 481         memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
 482
 483         err = ip6_local_out(net, skb->sk, skb);
 484         if (unlikely(net_xmit_eval(err)))
 485                 DEV_STATS_INC(dev, tx_errors);
 486         else
 487                 ret = NET_XMIT_SUCCESS;
 488         goto out;
 489 err:
 490         DEV_STATS_INC(dev, tx_errors);
 491         kfree_skb(skb);
 492 out:
 493         return ret;
 494 }
 495 #else
 496 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 497 {
 498         return NET_XMIT_DROP;
 499 }
 500 #endif
 501
 502 static int ipvlan_process_outbound(struct sk_buff *skb)
 503 {
 504         int ret = NET_XMIT_DROP;
 505
 506         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 507          * will have L2; which need to discarded and processed further
 508          * in the net-ns of the main-device.
 509          */
 510         if (skb_mac_header_was_set(skb)) {
 511                 /* In this mode we dont care about
 512                  * multicast and broadcast traffic */
 513                 struct ethhdr *ethh = eth_hdr(skb);
 514
 515                 if (is_multicast_ether_addr(ethh->h_dest)) {
 516                         pr_debug_ratelimited(
 517                                 "Dropped {multi|broad}cast of type=[%x]\n",
 518                                 ntohs(skb->protocol));
 519                         kfree_skb(skb);
 520                         goto out;
 521                 }
 522
 523                 skb_pull(skb, sizeof(*ethh));
 524                 skb->mac_header = (typeof(skb->mac_header))~0U;
 525                 skb_reset_network_header(skb);
 526         }
 527
 528         if (skb->protocol == htons(ETH_P_IPV6))
 529                 ret = ipvlan_process_v6_outbound(skb);
 530         else if (skb->protocol == htons(ETH_P_IP))
 531                 ret = ipvlan_process_v4_outbound(skb);
 532         else {
 533                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 534                                     ntohs(skb->protocol));
 535                 kfree_skb(skb);
 536         }
 537 out:
 538         return ret;
 539 }
 540
 541 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
 542                                      struct sk_buff *skb, bool tx_pkt)
 543 {
 544         if (skb->protocol == htons(ETH_P_PAUSE)) {
 545                 kfree_skb(skb);
 546                 return;
 547         }
 548
 549         /* Record that the deferred packet is from TX or RX path. By
 550          * looking at mac-addresses on packet will lead to erronus decisions.
 551          * (This would be true for a loopback-mode on master device or a
 552          * hair-pin mode of the switch.)
 553          */
 554         IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
 555
 556         spin_lock(&port->backlog.lock);
 557         if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
 558                 dev_hold(skb->dev);
 559                 __skb_queue_tail(&port->backlog, skb);
 560                 spin_unlock(&port->backlog.lock);
 561                 schedule_work(&port->wq);
 562         } else {
 563                 spin_unlock(&port->backlog.lock);
 564                 dev_core_stats_rx_dropped_inc(skb->dev);
 565                 kfree_skb(skb);
 566         }
 567 }
 568
 569 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 570 {
 571         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 572         void *lyr3h;
 573         struct ipvl_addr *addr;
 574         int addr_type;
 575
 576         lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 577         if (!lyr3h)
 578                 goto out;
 579
 580         if (!ipvlan_is_vepa(ipvlan->port)) {
 581                 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 582                 if (addr) {
 583                         if (ipvlan_is_private(ipvlan->port)) {
 584                                 consume_skb(skb);
 585                                 return NET_XMIT_DROP;
 586                         }
 587                         ipvlan_rcv_frame(addr, &skb, true);
 588                         return NET_XMIT_SUCCESS;
 589                 }
 590         }
 591 out:
 592         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 593         return ipvlan_process_outbound(skb);
 594 }
 595
 596 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 597 {
 598         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 599         struct ethhdr *eth = skb_eth_hdr(skb);
 600         struct ipvl_addr *addr;
 601         void *lyr3h;
 602         int addr_type;
 603
 604         if (!ipvlan_is_vepa(ipvlan->port) &&
 605             ether_addr_equal(eth->h_dest, eth->h_source)) {
 606                 lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 607                 if (lyr3h) {
 608                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 609                         if (addr) {
 610                                 if (ipvlan_is_private(ipvlan->port)) {
 611                                         consume_skb(skb);
 612                                         return NET_XMIT_DROP;
 613                                 }
 614                                 ipvlan_rcv_frame(addr, &skb, true);
 615                                 return NET_XMIT_SUCCESS;
 616                         }
 617                 }
 618                 skb = skb_share_check(skb, GFP_ATOMIC);
 619                 if (!skb)
 620                         return NET_XMIT_DROP;
 621
 622                 /* Packet definitely does not belong to any of the
 623                  * virtual devices, but the dest is local. So forward
 624                  * the skb for the main-dev. At the RX side we just return
 625                  * RX_PASS for it to be processed further on the stack.
 626                  */
 627                 dev_forward_skb(ipvlan->phy_dev, skb);
 628                 return NET_XMIT_SUCCESS;
 629
 630         } else if (is_multicast_ether_addr(eth->h_dest)) {
 631                 skb_reset_mac_header(skb);
 632                 ipvlan_skb_crossing_ns(skb, NULL);
 633                 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
 634                 return NET_XMIT_SUCCESS;
 635         }
 636
 637         skb->dev = ipvlan->phy_dev;
 638         return dev_queue_xmit(skb);
 639 }
 640
 641 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 642 {
 643         struct ipvl_dev *ipvlan = netdev_priv(dev);
 644         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 645
 646         if (!port)
 647                 goto out;
 648
 649         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 650                 goto out;
 651
 652         switch(port->mode) {
 653         case IPVLAN_MODE_L2:
 654                 return ipvlan_xmit_mode_l2(skb, dev);
 655         case IPVLAN_MODE_L3:
 656 #ifdef CONFIG_IPVLAN_L3S
 657         case IPVLAN_MODE_L3S:
 658 #endif
 659                 return ipvlan_xmit_mode_l3(skb, dev);
 660         }
 661
 662         /* Should not reach here */
 663         WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
 664 out:
 665         kfree_skb(skb);
 666         return NET_XMIT_DROP;
 667 }
 668
 669 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 670 {
 671         struct ethhdr *eth = eth_hdr(skb);
 672         struct ipvl_addr *addr;
 673         void *lyr3h;
 674         int addr_type;
 675
 676         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 677                 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 678                 if (!lyr3h)
 679                         return true;
 680
 681                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 682                 if (addr)
 683                         return false;
 684         }
 685
 686         return true;
 687 }
 688
 689 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 690                                                  struct ipvl_port *port)
 691 {
 692         void *lyr3h;
 693         int addr_type;
 694         struct ipvl_addr *addr;
 695         struct sk_buff *skb = *pskb;
 696         rx_handler_result_t ret = RX_HANDLER_PASS;
 697
 698         lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 699         if (!lyr3h)
 700                 goto out;
 701
 702         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 703         if (addr)
 704                 ret = ipvlan_rcv_frame(addr, pskb, false);
 705
 706 out:
 707         return ret;
 708 }
 709
 710 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 711                                                  struct ipvl_port *port)
 712 {
 713         struct sk_buff *skb = *pskb;
 714         struct ethhdr *eth = eth_hdr(skb);
 715         rx_handler_result_t ret = RX_HANDLER_PASS;
 716
 717         if (is_multicast_ether_addr(eth->h_dest)) {
 718                 if (ipvlan_external_frame(skb, port)) {
 719                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 720
 721                         /* External frames are queued for device local
 722                          * distribution, but a copy is given to master
 723                          * straight away to avoid sending duplicates later
 724                          * when work-queue processes this frame. This is
 725                          * achieved by returning RX_HANDLER_PASS.
 726                          */
 727                         if (nskb) {
 728                                 ipvlan_skb_crossing_ns(nskb, NULL);
 729                                 ipvlan_multicast_enqueue(port, nskb, false);
 730                         }
 731                 }
 732         } else {
 733                 /* Perform like l3 mode for non-multicast packet */
 734                 ret = ipvlan_handle_mode_l3(pskb, port);
 735         }
 736
 737         return ret;
 738 }
 739
 740 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 741 {
 742         struct sk_buff *skb = *pskb;
 743         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 744
 745         if (!port)
 746                 return RX_HANDLER_PASS;
 747
 748         switch (port->mode) {
 749         case IPVLAN_MODE_L2:
 750                 return ipvlan_handle_mode_l2(pskb, port);
 751         case IPVLAN_MODE_L3:
 752                 return ipvlan_handle_mode_l3(pskb, port);
 753 #ifdef CONFIG_IPVLAN_L3S
 754         case IPVLAN_MODE_L3S:
 755                 return RX_HANDLER_PASS;
 756 #endif
 757         }
 758
 759         /* Should not reach here */
 760         WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
 761         kfree_skb(skb);
 762         return RX_HANDLER_CONSUMED;
 763 }