drivers/net/ipvlan/ipvlan_core.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
   3  */
   4
   5 #include <net/inet_dscp.h>
   6
   7 #include "ipvlan.h"
   8
   9 static u32 ipvlan_jhash_secret __read_mostly;
  10
  11 void ipvlan_init_secret(void)
  12 {
  13         net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
  14 }
  15
  16 void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
  17                             unsigned int len, bool success, bool mcast)
  18 {
  19         if (likely(success)) {
  20                 struct ipvl_pcpu_stats *pcptr;
  21
  22                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
  23                 u64_stats_update_begin(&pcptr->syncp);
  24                 u64_stats_inc(&pcptr->rx_pkts);
  25                 u64_stats_add(&pcptr->rx_bytes, len);
  26                 if (mcast)
  27                         u64_stats_inc(&pcptr->rx_mcast);
  28                 u64_stats_update_end(&pcptr->syncp);
  29         } else {
  30                 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
  31         }
  32 }
  33 EXPORT_SYMBOL_GPL(ipvlan_count_rx);
  34
  35 #if IS_ENABLED(CONFIG_IPV6)
  36 static u8 ipvlan_get_v6_hash(const void *iaddr)
  37 {
  38         const struct in6_addr *ip6_addr = iaddr;
  39
  40         return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
  41                IPVLAN_HASH_MASK;
  42 }
  43 #else
  44 static u8 ipvlan_get_v6_hash(const void *iaddr)
  45 {
  46         return 0;
  47 }
  48 #endif
  49
  50 static u8 ipvlan_get_v4_hash(const void *iaddr)
  51 {
  52         const struct in_addr *ip4_addr = iaddr;
  53
  54         return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
  55                IPVLAN_HASH_MASK;
  56 }
  57
  58 static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
  59 {
  60         if (!is_v6 && addr->atype == IPVL_IPV4) {
  61                 struct in_addr *i4addr = (struct in_addr *)iaddr;
  62
  63                 return addr->ip4addr.s_addr == i4addr->s_addr;
  64 #if IS_ENABLED(CONFIG_IPV6)
  65         } else if (is_v6 && addr->atype == IPVL_IPV6) {
  66                 struct in6_addr *i6addr = (struct in6_addr *)iaddr;
  67
  68                 return ipv6_addr_equal(&addr->ip6addr, i6addr);
  69 #endif
  70         }
  71
  72         return false;
  73 }
  74
  75 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
  76                                                const void *iaddr, bool is_v6)
  77 {
  78         struct ipvl_addr *addr;
  79         u8 hash;
  80
  81         hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
  82                ipvlan_get_v4_hash(iaddr);
  83         hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
  84                 if (addr_equal(is_v6, addr, iaddr))
  85                         return addr;
  86         return NULL;
  87 }
  88
  89 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
  90 {
  91         struct ipvl_port *port = ipvlan->port;
  92         u8 hash;
  93
  94         hash = (addr->atype == IPVL_IPV6) ?
  95                ipvlan_get_v6_hash(&addr->ip6addr) :
  96                ipvlan_get_v4_hash(&addr->ip4addr);
  97         if (hlist_unhashed(&addr->hlnode))
  98                 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
  99 }
 100
 101 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
 102 {
 103         hlist_del_init_rcu(&addr->hlnode);
 104 }
 105
 106 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
 107                                    const void *iaddr, bool is_v6)
 108 {
 109         struct ipvl_addr *addr, *ret = NULL;
 110
 111         rcu_read_lock();
 112         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) {
 113                 if (addr_equal(is_v6, addr, iaddr)) {
 114                         ret = addr;
 115                         break;
 116                 }
 117         }
 118         rcu_read_unlock();
 119         return ret;
 120 }
 121
 122 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
 123 {
 124         struct ipvl_dev *ipvlan;
 125         bool ret = false;
 126
 127         rcu_read_lock();
 128         list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 129                 if (ipvlan_find_addr(ipvlan, iaddr, is_v6)) {
 130                         ret = true;
 131                         break;
 132                 }
 133         }
 134         rcu_read_unlock();
 135         return ret;
 136 }
 137
 138 void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
 139 {
 140         void *lyr3h = NULL;
 141
 142         switch (skb->protocol) {
 143         case htons(ETH_P_ARP): {
 144                 struct arphdr *arph;
 145
 146                 if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
 147                         return NULL;
 148
 149                 arph = arp_hdr(skb);
 150                 *type = IPVL_ARP;
 151                 lyr3h = arph;
 152                 break;
 153         }
 154         case htons(ETH_P_IP): {
 155                 u32 pktlen;
 156                 struct iphdr *ip4h;
 157
 158                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
 159                         return NULL;
 160
 161                 ip4h = ip_hdr(skb);
 162                 pktlen = skb_ip_totlen(skb);
 163                 if (ip4h->ihl < 5 || ip4h->version != 4)
 164                         return NULL;
 165                 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
 166                         return NULL;
 167
 168                 *type = IPVL_IPV4;
 169                 lyr3h = ip4h;
 170                 break;
 171         }
 172 #if IS_ENABLED(CONFIG_IPV6)
 173         case htons(ETH_P_IPV6): {
 174                 struct ipv6hdr *ip6h;
 175
 176                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
 177                         return NULL;
 178
 179                 ip6h = ipv6_hdr(skb);
 180                 if (ip6h->version != 6)
 181                         return NULL;
 182
 183                 *type = IPVL_IPV6;
 184                 lyr3h = ip6h;
 185                 /* Only Neighbour Solicitation pkts need different treatment */
 186                 if (ipv6_addr_any(&ip6h->saddr) &&
 187                     ip6h->nexthdr == NEXTHDR_ICMP) {
 188                         struct icmp6hdr *icmph;
 189
 190                         if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
 191                                 return NULL;
 192
 193                         ip6h = ipv6_hdr(skb);
 194                         icmph = (struct icmp6hdr *)(ip6h + 1);
 195
 196                         if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 197                                 /* Need to access the ipv6 address in body */
 198                                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
 199                                                 + sizeof(struct in6_addr))))
 200                                         return NULL;
 201
 202                                 ip6h = ipv6_hdr(skb);
 203                                 icmph = (struct icmp6hdr *)(ip6h + 1);
 204                         }
 205
 206                         *type = IPVL_ICMPV6;
 207                         lyr3h = icmph;
 208                 }
 209                 break;
 210         }
 211 #endif
 212         default:
 213                 return NULL;
 214         }
 215
 216         return lyr3h;
 217 }
 218
 219 unsigned int ipvlan_mac_hash(const unsigned char *addr)
 220 {
 221         u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
 222                                ipvlan_jhash_secret);
 223
 224         return hash & IPVLAN_MAC_FILTER_MASK;
 225 }
 226
 227 void ipvlan_process_multicast(struct work_struct *work)
 228 {
 229         struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
 230         struct ethhdr *ethh;
 231         struct ipvl_dev *ipvlan;
 232         struct sk_buff *skb, *nskb;
 233         struct sk_buff_head list;
 234         unsigned int len;
 235         unsigned int mac_hash;
 236         int ret;
 237         u8 pkt_type;
 238         bool tx_pkt;
 239
 240         __skb_queue_head_init(&list);
 241
 242         spin_lock_bh(&port->backlog.lock);
 243         skb_queue_splice_tail_init(&port->backlog, &list);
 244         spin_unlock_bh(&port->backlog.lock);
 245
 246         while ((skb = __skb_dequeue(&list)) != NULL) {
 247                 struct net_device *dev = skb->dev;
 248                 bool consumed = false;
 249
 250                 ethh = eth_hdr(skb);
 251                 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
 252                 mac_hash = ipvlan_mac_hash(ethh->h_dest);
 253
 254                 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
 255                         pkt_type = PACKET_BROADCAST;
 256                 else
 257                         pkt_type = PACKET_MULTICAST;
 258
 259                 rcu_read_lock();
 260                 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 261                         if (tx_pkt && (ipvlan->dev == skb->dev))
 262                                 continue;
 263                         if (!test_bit(mac_hash, ipvlan->mac_filters))
 264                                 continue;
 265                         if (!(ipvlan->dev->flags & IFF_UP))
 266                                 continue;
 267                         ret = NET_RX_DROP;
 268                         len = skb->len + ETH_HLEN;
 269                         nskb = skb_clone(skb, GFP_ATOMIC);
 270                         local_bh_disable();
 271                         if (nskb) {
 272                                 consumed = true;
 273                                 nskb->pkt_type = pkt_type;
 274                                 nskb->dev = ipvlan->dev;
 275                                 if (tx_pkt)
 276                                         ret = dev_forward_skb(ipvlan->dev, nskb);
 277                                 else
 278                                         ret = netif_rx(nskb);
 279                         }
 280                         ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 281                         local_bh_enable();
 282                 }
 283                 rcu_read_unlock();
 284
 285                 if (tx_pkt) {
 286                         /* If the packet originated here, send it out. */
 287                         skb->dev = port->dev;
 288                         skb->pkt_type = pkt_type;
 289                         dev_queue_xmit(skb);
 290                 } else {
 291                         if (consumed)
 292                                 consume_skb(skb);
 293                         else
 294                                 kfree_skb(skb);
 295                 }
 296                 dev_put(dev);
 297                 cond_resched();
 298         }
 299 }
 300
 301 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 302 {
 303         bool xnet = true;
 304
 305         if (dev)
 306                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 307
 308         skb_scrub_packet(skb, xnet);
 309         if (dev)
 310                 skb->dev = dev;
 311 }
 312
 313 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 314                             bool local)
 315 {
 316         struct ipvl_dev *ipvlan = addr->master;
 317         struct net_device *dev = ipvlan->dev;
 318         unsigned int len;
 319         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 320         bool success = false;
 321         struct sk_buff *skb = *pskb;
 322
 323         len = skb->len + ETH_HLEN;
 324         /* Only packets exchanged between two local slaves need to have
 325          * device-up check as well as skb-share check.
 326          */
 327         if (local) {
 328                 if (unlikely(!(dev->flags & IFF_UP))) {
 329                         kfree_skb(skb);
 330                         goto out;
 331                 }
 332
 333                 skb = skb_share_check(skb, GFP_ATOMIC);
 334                 if (!skb)
 335                         goto out;
 336
 337                 *pskb = skb;
 338         }
 339
 340         if (local) {
 341                 skb->pkt_type = PACKET_HOST;
 342                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 343                         success = true;
 344         } else {
 345                 skb->dev = dev;
 346                 ret = RX_HANDLER_ANOTHER;
 347                 success = true;
 348         }
 349
 350 out:
 351         ipvlan_count_rx(ipvlan, len, success, false);
 352         return ret;
 353 }
 354
 355 struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, void *lyr3h,
 356                                      int addr_type, bool use_dest)
 357 {
 358         struct ipvl_addr *addr = NULL;
 359
 360         switch (addr_type) {
 361 #if IS_ENABLED(CONFIG_IPV6)
 362         case IPVL_IPV6: {
 363                 struct ipv6hdr *ip6h;
 364                 struct in6_addr *i6addr;
 365
 366                 ip6h = (struct ipv6hdr *)lyr3h;
 367                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 368                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 369                 break;
 370         }
 371         case IPVL_ICMPV6: {
 372                 struct nd_msg *ndmh;
 373                 struct in6_addr *i6addr;
 374
 375                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 376                  * are handled to avoid DAD issue.
 377                  */
 378                 ndmh = (struct nd_msg *)lyr3h;
 379                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 380                         i6addr = &ndmh->target;
 381                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 382                 }
 383                 break;
 384         }
 385 #endif
 386         case IPVL_IPV4: {
 387                 struct iphdr *ip4h;
 388                 __be32 *i4addr;
 389
 390                 ip4h = (struct iphdr *)lyr3h;
 391                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 392                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 393                 break;
 394         }
 395         case IPVL_ARP: {
 396                 struct arphdr *arph;
 397                 unsigned char *arp_ptr;
 398                 __be32 dip;
 399
 400                 arph = (struct arphdr *)lyr3h;
 401                 arp_ptr = (unsigned char *)(arph + 1);
 402                 if (use_dest)
 403                         arp_ptr += (2 * port->dev->addr_len) + 4;
 404                 else
 405                         arp_ptr += port->dev->addr_len;
 406
 407                 memcpy(&dip, arp_ptr, 4);
 408                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 409                 break;
 410         }
 411         }
 412
 413         return addr;
 414 }
 415
 416 static noinline_for_stack int ipvlan_process_v4_outbound(struct sk_buff *skb)
 417 {
 418         const struct iphdr *ip4h = ip_hdr(skb);
 419         struct net_device *dev = skb->dev;
 420         struct net *net = dev_net(dev);
 421         struct rtable *rt;
 422         int err, ret = NET_XMIT_DROP;
 423         struct flowi4 fl4 = {
 424                 .flowi4_oif = dev->ifindex,
 425                 .flowi4_tos = ip4h->tos & INET_DSCP_MASK,
 426                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 427                 .flowi4_mark = skb->mark,
 428                 .daddr = ip4h->daddr,
 429                 .saddr = ip4h->saddr,
 430         };
 431
 432         rt = ip_route_output_flow(net, &fl4, NULL);
 433         if (IS_ERR(rt))
 434                 goto err;
 435
 436         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 437                 ip_rt_put(rt);
 438                 goto err;
 439         }
 440         skb_dst_set(skb, &rt->dst);
 441
 442         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 443
 444         err = ip_local_out(net, NULL, skb);
 445         if (unlikely(net_xmit_eval(err)))
 446                 DEV_STATS_INC(dev, tx_errors);
 447         else
 448                 ret = NET_XMIT_SUCCESS;
 449         goto out;
 450 err:
 451         DEV_STATS_INC(dev, tx_errors);
 452         kfree_skb(skb);
 453 out:
 454         return ret;
 455 }
 456
 457 #if IS_ENABLED(CONFIG_IPV6)
 458
 459 static noinline_for_stack int
 460 ipvlan_route_v6_outbound(struct net_device *dev, struct sk_buff *skb)
 461 {
 462         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 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         struct dst_entry *dst;
 473         int err;
 474
 475         dst = ip6_route_output(dev_net(dev), NULL, &fl6);
 476         err = dst->error;
 477         if (err) {
 478                 dst_release(dst);
 479                 return err;
 480         }
 481         skb_dst_set(skb, dst);
 482         return 0;
 483 }
 484
 485 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 486 {
 487         struct net_device *dev = skb->dev;
 488         int err, ret = NET_XMIT_DROP;
 489
 490         err = ipvlan_route_v6_outbound(dev, skb);
 491         if (unlikely(err)) {
 492                 DEV_STATS_INC(dev, tx_errors);
 493                 kfree_skb(skb);
 494                 return err;
 495         }
 496
 497         memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
 498
 499         err = ip6_local_out(dev_net(dev), NULL, skb);
 500         if (unlikely(net_xmit_eval(err)))
 501                 DEV_STATS_INC(dev, tx_errors);
 502         else
 503                 ret = NET_XMIT_SUCCESS;
 504         return ret;
 505 }
 506 #else
 507 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 508 {
 509         return NET_XMIT_DROP;
 510 }
 511 #endif
 512
 513 static int ipvlan_process_outbound(struct sk_buff *skb)
 514 {
 515         int ret = NET_XMIT_DROP;
 516
 517         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 518          * will have L2; which need to discarded and processed further
 519          * in the net-ns of the main-device.
 520          */
 521         if (skb_mac_header_was_set(skb)) {
 522                 /* In this mode we dont care about
 523                  * multicast and broadcast traffic */
 524                 struct ethhdr *ethh = eth_hdr(skb);
 525
 526                 if (is_multicast_ether_addr(ethh->h_dest)) {
 527                         pr_debug_ratelimited(
 528                                 "Dropped {multi|broad}cast of type=[%x]\n",
 529                                 ntohs(skb->protocol));
 530                         kfree_skb(skb);
 531                         goto out;
 532                 }
 533
 534                 skb_pull(skb, sizeof(*ethh));
 535                 skb->mac_header = (typeof(skb->mac_header))~0U;
 536                 skb_reset_network_header(skb);
 537         }
 538
 539         if (skb->protocol == htons(ETH_P_IPV6))
 540                 ret = ipvlan_process_v6_outbound(skb);
 541         else if (skb->protocol == htons(ETH_P_IP))
 542                 ret = ipvlan_process_v4_outbound(skb);
 543         else {
 544                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 545                                     ntohs(skb->protocol));
 546                 kfree_skb(skb);
 547         }
 548 out:
 549         return ret;
 550 }
 551
 552 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
 553                                      struct sk_buff *skb, bool tx_pkt)
 554 {
 555         if (skb->protocol == htons(ETH_P_PAUSE)) {
 556                 kfree_skb(skb);
 557                 return;
 558         }
 559
 560         /* Record that the deferred packet is from TX or RX path. By
 561          * looking at mac-addresses on packet will lead to erronus decisions.
 562          * (This would be true for a loopback-mode on master device or a
 563          * hair-pin mode of the switch.)
 564          */
 565         IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
 566
 567         spin_lock(&port->backlog.lock);
 568         if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
 569                 dev_hold(skb->dev);
 570                 __skb_queue_tail(&port->backlog, skb);
 571                 spin_unlock(&port->backlog.lock);
 572                 schedule_work(&port->wq);
 573         } else {
 574                 spin_unlock(&port->backlog.lock);
 575                 dev_core_stats_rx_dropped_inc(skb->dev);
 576                 kfree_skb(skb);
 577         }
 578 }
 579
 580 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 581 {
 582         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 583         void *lyr3h;
 584         struct ipvl_addr *addr;
 585         int addr_type;
 586
 587         lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 588         if (!lyr3h)
 589                 goto out;
 590
 591         if (!ipvlan_is_vepa(ipvlan->port)) {
 592                 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 593                 if (addr) {
 594                         if (ipvlan_is_private(ipvlan->port)) {
 595                                 consume_skb(skb);
 596                                 return NET_XMIT_DROP;
 597                         }
 598                         ipvlan_rcv_frame(addr, &skb, true);
 599                         return NET_XMIT_SUCCESS;
 600                 }
 601         }
 602 out:
 603         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 604         return ipvlan_process_outbound(skb);
 605 }
 606
 607 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 608 {
 609         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 610         struct ethhdr *eth = skb_eth_hdr(skb);
 611         struct ipvl_addr *addr;
 612         void *lyr3h;
 613         int addr_type;
 614
 615         if (!ipvlan_is_vepa(ipvlan->port) &&
 616             ether_addr_equal(eth->h_dest, eth->h_source)) {
 617                 lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 618                 if (lyr3h) {
 619                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 620                         if (addr) {
 621                                 if (ipvlan_is_private(ipvlan->port)) {
 622                                         consume_skb(skb);
 623                                         return NET_XMIT_DROP;
 624                                 }
 625                                 ipvlan_rcv_frame(addr, &skb, true);
 626                                 return NET_XMIT_SUCCESS;
 627                         }
 628                 }
 629                 skb = skb_share_check(skb, GFP_ATOMIC);
 630                 if (!skb)
 631                         return NET_XMIT_DROP;
 632
 633                 /* Packet definitely does not belong to any of the
 634                  * virtual devices, but the dest is local. So forward
 635                  * the skb for the main-dev. At the RX side we just return
 636                  * RX_PASS for it to be processed further on the stack.
 637                  */
 638                 dev_forward_skb(ipvlan->phy_dev, skb);
 639                 return NET_XMIT_SUCCESS;
 640
 641         } else if (is_multicast_ether_addr(eth->h_dest)) {
 642                 skb_reset_mac_header(skb);
 643                 ipvlan_skb_crossing_ns(skb, NULL);
 644                 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
 645                 return NET_XMIT_SUCCESS;
 646         }
 647
 648         skb->dev = ipvlan->phy_dev;
 649         return dev_queue_xmit(skb);
 650 }
 651
 652 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 653 {
 654         struct ipvl_dev *ipvlan = netdev_priv(dev);
 655         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 656
 657         if (!port)
 658                 goto out;
 659
 660         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 661                 goto out;
 662
 663         switch(port->mode) {
 664         case IPVLAN_MODE_L2:
 665                 return ipvlan_xmit_mode_l2(skb, dev);
 666         case IPVLAN_MODE_L3:
 667 #ifdef CONFIG_IPVLAN_L3S
 668         case IPVLAN_MODE_L3S:
 669 #endif
 670                 return ipvlan_xmit_mode_l3(skb, dev);
 671         }
 672
 673         /* Should not reach here */
 674         WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
 675 out:
 676         kfree_skb(skb);
 677         return NET_XMIT_DROP;
 678 }
 679
 680 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 681 {
 682         struct ethhdr *eth = eth_hdr(skb);
 683         struct ipvl_addr *addr;
 684         void *lyr3h;
 685         int addr_type;
 686
 687         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 688                 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 689                 if (!lyr3h)
 690                         return true;
 691
 692                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 693                 if (addr)
 694                         return false;
 695         }
 696
 697         return true;
 698 }
 699
 700 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 701                                                  struct ipvl_port *port)
 702 {
 703         void *lyr3h;
 704         int addr_type;
 705         struct ipvl_addr *addr;
 706         struct sk_buff *skb = *pskb;
 707         rx_handler_result_t ret = RX_HANDLER_PASS;
 708
 709         lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 710         if (!lyr3h)
 711                 goto out;
 712
 713         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 714         if (addr)
 715                 ret = ipvlan_rcv_frame(addr, pskb, false);
 716
 717 out:
 718         return ret;
 719 }
 720
 721 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 722                                                  struct ipvl_port *port)
 723 {
 724         struct sk_buff *skb = *pskb;
 725         struct ethhdr *eth = eth_hdr(skb);
 726         rx_handler_result_t ret = RX_HANDLER_PASS;
 727
 728         if (is_multicast_ether_addr(eth->h_dest)) {
 729                 if (ipvlan_external_frame(skb, port)) {
 730                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 731
 732                         /* External frames are queued for device local
 733                          * distribution, but a copy is given to master
 734                          * straight away to avoid sending duplicates later
 735                          * when work-queue processes this frame. This is
 736                          * achieved by returning RX_HANDLER_PASS.
 737                          */
 738                         if (nskb) {
 739                                 ipvlan_skb_crossing_ns(nskb, NULL);
 740                                 ipvlan_multicast_enqueue(port, nskb, false);
 741                         }
 742                 }
 743         } else {
 744                 /* Perform like l3 mode for non-multicast packet */
 745                 ret = ipvlan_handle_mode_l3(pskb, port);
 746         }
 747
 748         return ret;
 749 }
 750
 751 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 752 {
 753         struct sk_buff *skb = *pskb;
 754         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 755
 756         if (!port)
 757                 return RX_HANDLER_PASS;
 758
 759         switch (port->mode) {
 760         case IPVLAN_MODE_L2:
 761                 return ipvlan_handle_mode_l2(pskb, port);
 762         case IPVLAN_MODE_L3:
 763                 return ipvlan_handle_mode_l3(pskb, port);
 764 #ifdef CONFIG_IPVLAN_L3S
 765         case IPVLAN_MODE_L3S:
 766                 return RX_HANDLER_PASS;
 767 #endif
 768         }
 769
 770         /* Should not reach here */
 771         WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
 772         kfree_skb(skb);
 773         return RX_HANDLER_CONSUMED;
 774 }