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 void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
  20                             unsigned int len, bool success, bool mcast)
  21 {
  22         if (likely(success)) {
  23                 struct ipvl_pcpu_stats *pcptr;
  24
  25                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
  26                 u64_stats_update_begin(&pcptr->syncp);
  27                 pcptr->rx_pkts++;
  28                 pcptr->rx_bytes += len;
  29                 if (mcast)
  30                         pcptr->rx_mcast++;
  31                 u64_stats_update_end(&pcptr->syncp);
  32         } else {
  33                 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
  34         }
  35 }
  36 EXPORT_SYMBOL_GPL(ipvlan_count_rx);
  37
  38 #if IS_ENABLED(CONFIG_IPV6)
  39 static u8 ipvlan_get_v6_hash(const void *iaddr)
  40 {
  41         const struct in6_addr *ip6_addr = iaddr;
  42
  43         return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
  44                IPVLAN_HASH_MASK;
  45 }
  46 #else
  47 static u8 ipvlan_get_v6_hash(const void *iaddr)
  48 {
  49         return 0;
  50 }
  51 #endif
  52
  53 static u8 ipvlan_get_v4_hash(const void *iaddr)
  54 {
  55         const struct in_addr *ip4_addr = iaddr;
  56
  57         return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
  58                IPVLAN_HASH_MASK;
  59 }
  60
  61 static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
  62 {
  63         if (!is_v6 && addr->atype == IPVL_IPV4) {
  64                 struct in_addr *i4addr = (struct in_addr *)iaddr;
  65
  66                 return addr->ip4addr.s_addr == i4addr->s_addr;
  67 #if IS_ENABLED(CONFIG_IPV6)
  68         } else if (is_v6 && addr->atype == IPVL_IPV6) {
  69                 struct in6_addr *i6addr = (struct in6_addr *)iaddr;
  70
  71                 return ipv6_addr_equal(&addr->ip6addr, i6addr);
  72 #endif
  73         }
  74
  75         return false;
  76 }
  77
  78 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
  79                                                const void *iaddr, bool is_v6)
  80 {
  81         struct ipvl_addr *addr;
  82         u8 hash;
  83
  84         hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
  85                ipvlan_get_v4_hash(iaddr);
  86         hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
  87                 if (addr_equal(is_v6, addr, iaddr))
  88                         return addr;
  89         return NULL;
  90 }
  91
  92 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
  93 {
  94         struct ipvl_port *port = ipvlan->port;
  95         u8 hash;
  96
  97         hash = (addr->atype == IPVL_IPV6) ?
  98                ipvlan_get_v6_hash(&addr->ip6addr) :
  99                ipvlan_get_v4_hash(&addr->ip4addr);
 100         if (hlist_unhashed(&addr->hlnode))
 101                 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
 102 }
 103
 104 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
 105 {
 106         hlist_del_init_rcu(&addr->hlnode);
 107 }
 108
 109 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
 110                                    const void *iaddr, bool is_v6)
 111 {
 112         struct ipvl_addr *addr, *ret = NULL;
 113
 114         rcu_read_lock();
 115         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) {
 116                 if (addr_equal(is_v6, addr, iaddr)) {
 117                         ret = addr;
 118                         break;
 119                 }
 120         }
 121         rcu_read_unlock();
 122         return ret;
 123 }
 124
 125 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
 126 {
 127         struct ipvl_dev *ipvlan;
 128         bool ret = false;
 129
 130         rcu_read_lock();
 131         list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 132                 if (ipvlan_find_addr(ipvlan, iaddr, is_v6)) {
 133                         ret = true;
 134                         break;
 135                 }
 136         }
 137         rcu_read_unlock();
 138         return ret;
 139 }
 140
 141 void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
 142 {
 143         void *lyr3h = NULL;
 144
 145         switch (skb->protocol) {
 146         case htons(ETH_P_ARP): {
 147                 struct arphdr *arph;
 148
 149                 if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
 150                         return NULL;
 151
 152                 arph = arp_hdr(skb);
 153                 *type = IPVL_ARP;
 154                 lyr3h = arph;
 155                 break;
 156         }
 157         case htons(ETH_P_IP): {
 158                 u32 pktlen;
 159                 struct iphdr *ip4h;
 160
 161                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
 162                         return NULL;
 163
 164                 ip4h = ip_hdr(skb);
 165                 pktlen = ntohs(ip4h->tot_len);
 166                 if (ip4h->ihl < 5 || ip4h->version != 4)
 167                         return NULL;
 168                 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
 169                         return NULL;
 170
 171                 *type = IPVL_IPV4;
 172                 lyr3h = ip4h;
 173                 break;
 174         }
 175 #if IS_ENABLED(CONFIG_IPV6)
 176         case htons(ETH_P_IPV6): {
 177                 struct ipv6hdr *ip6h;
 178
 179                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
 180                         return NULL;
 181
 182                 ip6h = ipv6_hdr(skb);
 183                 if (ip6h->version != 6)
 184                         return NULL;
 185
 186                 *type = IPVL_IPV6;
 187                 lyr3h = ip6h;
 188                 /* Only Neighbour Solicitation pkts need different treatment */
 189                 if (ipv6_addr_any(&ip6h->saddr) &&
 190                     ip6h->nexthdr == NEXTHDR_ICMP) {
 191                         struct icmp6hdr *icmph;
 192
 193                         if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
 194                                 return NULL;
 195
 196                         ip6h = ipv6_hdr(skb);
 197                         icmph = (struct icmp6hdr *)(ip6h + 1);
 198
 199                         if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 200                                 /* Need to access the ipv6 address in body */
 201                                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
 202                                                 + sizeof(struct in6_addr))))
 203                                         return NULL;
 204
 205                                 ip6h = ipv6_hdr(skb);
 206                                 icmph = (struct icmp6hdr *)(ip6h + 1);
 207                         }
 208
 209                         *type = IPVL_ICMPV6;
 210                         lyr3h = icmph;
 211                 }
 212                 break;
 213         }
 214 #endif
 215         default:
 216                 return NULL;
 217         }
 218
 219         return lyr3h;
 220 }
 221
 222 unsigned int ipvlan_mac_hash(const unsigned char *addr)
 223 {
 224         u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
 225                                ipvlan_jhash_secret);
 226
 227         return hash & IPVLAN_MAC_FILTER_MASK;
 228 }
 229
 230 void ipvlan_process_multicast(struct work_struct *work)
 231 {
 232         struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
 233         struct ethhdr *ethh;
 234         struct ipvl_dev *ipvlan;
 235         struct sk_buff *skb, *nskb;
 236         struct sk_buff_head list;
 237         unsigned int len;
 238         unsigned int mac_hash;
 239         int ret;
 240         u8 pkt_type;
 241         bool tx_pkt;
 242
 243         __skb_queue_head_init(&list);
 244
 245         spin_lock_bh(&port->backlog.lock);
 246         skb_queue_splice_tail_init(&port->backlog, &list);
 247         spin_unlock_bh(&port->backlog.lock);
 248
 249         while ((skb = __skb_dequeue(&list)) != NULL) {
 250                 struct net_device *dev = skb->dev;
 251                 bool consumed = false;
 252
 253                 ethh = eth_hdr(skb);
 254                 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
 255                 mac_hash = ipvlan_mac_hash(ethh->h_dest);
 256
 257                 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
 258                         pkt_type = PACKET_BROADCAST;
 259                 else
 260                         pkt_type = PACKET_MULTICAST;
 261
 262                 rcu_read_lock();
 263                 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 264                         if (tx_pkt && (ipvlan->dev == skb->dev))
 265                                 continue;
 266                         if (!test_bit(mac_hash, ipvlan->mac_filters))
 267                                 continue;
 268                         if (!(ipvlan->dev->flags & IFF_UP))
 269                                 continue;
 270                         ret = NET_RX_DROP;
 271                         len = skb->len + ETH_HLEN;
 272                         nskb = skb_clone(skb, GFP_ATOMIC);
 273                         local_bh_disable();
 274                         if (nskb) {
 275                                 consumed = true;
 276                                 nskb->pkt_type = pkt_type;
 277                                 nskb->dev = ipvlan->dev;
 278                                 if (tx_pkt)
 279                                         ret = dev_forward_skb(ipvlan->dev, nskb);
 280                                 else
 281                                         ret = netif_rx(nskb);
 282                         }
 283                         ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 284                         local_bh_enable();
 285                 }
 286                 rcu_read_unlock();
 287
 288                 if (tx_pkt) {
 289                         /* If the packet originated here, send it out. */
 290                         skb->dev = port->dev;
 291                         skb->pkt_type = pkt_type;
 292                         dev_queue_xmit(skb);
 293                 } else {
 294                         if (consumed)
 295                                 consume_skb(skb);
 296                         else
 297                                 kfree_skb(skb);
 298                 }
 299                 if (dev)
 300                         dev_put(dev);
 301         }
 302 }
 303
 304 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 305 {
 306         bool xnet = true;
 307
 308         if (dev)
 309                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 310
 311         skb_scrub_packet(skb, xnet);
 312         if (dev)
 313                 skb->dev = dev;
 314 }
 315
 316 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 317                             bool local)
 318 {
 319         struct ipvl_dev *ipvlan = addr->master;
 320         struct net_device *dev = ipvlan->dev;
 321         unsigned int len;
 322         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 323         bool success = false;
 324         struct sk_buff *skb = *pskb;
 325
 326         len = skb->len + ETH_HLEN;
 327         /* Only packets exchanged between two local slaves need to have
 328          * device-up check as well as skb-share check.
 329          */
 330         if (local) {
 331                 if (unlikely(!(dev->flags & IFF_UP))) {
 332                         kfree_skb(skb);
 333                         goto out;
 334                 }
 335
 336                 skb = skb_share_check(skb, GFP_ATOMIC);
 337                 if (!skb)
 338                         goto out;
 339
 340                 *pskb = skb;
 341         }
 342
 343         if (local) {
 344                 skb->pkt_type = PACKET_HOST;
 345                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 346                         success = true;
 347         } else {
 348                 skb->dev = dev;
 349                 ret = RX_HANDLER_ANOTHER;
 350                 success = true;
 351         }
 352
 353 out:
 354         ipvlan_count_rx(ipvlan, len, success, false);
 355         return ret;
 356 }
 357
 358 struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, void *lyr3h,
 359                                      int addr_type, bool use_dest)
 360 {
 361         struct ipvl_addr *addr = NULL;
 362
 363         switch (addr_type) {
 364 #if IS_ENABLED(CONFIG_IPV6)
 365         case IPVL_IPV6: {
 366                 struct ipv6hdr *ip6h;
 367                 struct in6_addr *i6addr;
 368
 369                 ip6h = (struct ipv6hdr *)lyr3h;
 370                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 371                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 372                 break;
 373         }
 374         case IPVL_ICMPV6: {
 375                 struct nd_msg *ndmh;
 376                 struct in6_addr *i6addr;
 377
 378                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 379                  * are handled to avoid DAD issue.
 380                  */
 381                 ndmh = (struct nd_msg *)lyr3h;
 382                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 383                         i6addr = &ndmh->target;
 384                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 385                 }
 386                 break;
 387         }
 388 #endif
 389         case IPVL_IPV4: {
 390                 struct iphdr *ip4h;
 391                 __be32 *i4addr;
 392
 393                 ip4h = (struct iphdr *)lyr3h;
 394                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 395                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 396                 break;
 397         }
 398         case IPVL_ARP: {
 399                 struct arphdr *arph;
 400                 unsigned char *arp_ptr;
 401                 __be32 dip;
 402
 403                 arph = (struct arphdr *)lyr3h;
 404                 arp_ptr = (unsigned char *)(arph + 1);
 405                 if (use_dest)
 406                         arp_ptr += (2 * port->dev->addr_len) + 4;
 407                 else
 408                         arp_ptr += port->dev->addr_len;
 409
 410                 memcpy(&dip, arp_ptr, 4);
 411                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 412                 break;
 413         }
 414         }
 415
 416         return addr;
 417 }
 418
 419 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 420 {
 421         const struct iphdr *ip4h = ip_hdr(skb);
 422         struct net_device *dev = skb->dev;
 423         struct net *net = dev_net(dev);
 424         struct rtable *rt;
 425         int err, ret = NET_XMIT_DROP;
 426         struct flowi4 fl4 = {
 427                 .flowi4_oif = dev->ifindex,
 428                 .flowi4_tos = RT_TOS(ip4h->tos),
 429                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 430                 .flowi4_mark = skb->mark,
 431                 .daddr = ip4h->daddr,
 432                 .saddr = ip4h->saddr,
 433         };
 434
 435         rt = ip_route_output_flow(net, &fl4, NULL);
 436         if (IS_ERR(rt))
 437                 goto err;
 438
 439         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 440                 ip_rt_put(rt);
 441                 goto err;
 442         }
 443         skb_dst_set(skb, &rt->dst);
 444         err = ip_local_out(net, skb->sk, skb);
 445         if (unlikely(net_xmit_eval(err)))
 446                 dev->stats.tx_errors++;
 447         else
 448                 ret = NET_XMIT_SUCCESS;
 449         goto out;
 450 err:
 451         dev->stats.tx_errors++;
 452         kfree_skb(skb);
 453 out:
 454         return ret;
 455 }
 456
 457 #if IS_ENABLED(CONFIG_IPV6)
 458 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 459 {
 460         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 461         struct net_device *dev = skb->dev;
 462         struct net *net = dev_net(dev);
 463         struct dst_entry *dst;
 464         int err, ret = NET_XMIT_DROP;
 465         struct flowi6 fl6 = {
 466                 .flowi6_oif = dev->ifindex,
 467                 .daddr = ip6h->daddr,
 468                 .saddr = ip6h->saddr,
 469                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 470                 .flowlabel = ip6_flowinfo(ip6h),
 471                 .flowi6_mark = skb->mark,
 472                 .flowi6_proto = ip6h->nexthdr,
 473         };
 474
 475         dst = ip6_route_output(net, NULL, &fl6);
 476         if (dst->error) {
 477                 ret = dst->error;
 478                 dst_release(dst);
 479                 goto err;
 480         }
 481         skb_dst_set(skb, dst);
 482         err = ip6_local_out(net, skb->sk, skb);
 483         if (unlikely(net_xmit_eval(err)))
 484                 dev->stats.tx_errors++;
 485         else
 486                 ret = NET_XMIT_SUCCESS;
 487         goto out;
 488 err:
 489         dev->stats.tx_errors++;
 490         kfree_skb(skb);
 491 out:
 492         return ret;
 493 }
 494 #else
 495 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 496 {
 497         return NET_XMIT_DROP;
 498 }
 499 #endif
 500
 501 static int ipvlan_process_outbound(struct sk_buff *skb)
 502 {
 503         struct ethhdr *ethh = eth_hdr(skb);
 504         int ret = NET_XMIT_DROP;
 505
 506         /* In this mode we dont care about multicast and broadcast traffic */
 507         if (is_multicast_ether_addr(ethh->h_dest)) {
 508                 pr_debug_ratelimited("Dropped {multi|broad}cast of type=[%x]\n",
 509                                      ntohs(skb->protocol));
 510                 kfree_skb(skb);
 511                 goto out;
 512         }
 513
 514         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 515          * will have L2; which need to discarded and processed further
 516          * in the net-ns of the main-device.
 517          */
 518         if (skb_mac_header_was_set(skb)) {
 519                 skb_pull(skb, sizeof(*ethh));
 520                 skb->mac_header = (typeof(skb->mac_header))~0U;
 521                 skb_reset_network_header(skb);
 522         }
 523
 524         if (skb->protocol == htons(ETH_P_IPV6))
 525                 ret = ipvlan_process_v6_outbound(skb);
 526         else if (skb->protocol == htons(ETH_P_IP))
 527                 ret = ipvlan_process_v4_outbound(skb);
 528         else {
 529                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 530                                     ntohs(skb->protocol));
 531                 kfree_skb(skb);
 532         }
 533 out:
 534         return ret;
 535 }
 536
 537 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
 538                                      struct sk_buff *skb, bool tx_pkt)
 539 {
 540         if (skb->protocol == htons(ETH_P_PAUSE)) {
 541                 kfree_skb(skb);
 542                 return;
 543         }
 544
 545         /* Record that the deferred packet is from TX or RX path. By
 546          * looking at mac-addresses on packet will lead to erronus decisions.
 547          * (This would be true for a loopback-mode on master device or a
 548          * hair-pin mode of the switch.)
 549          */
 550         IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
 551
 552         spin_lock(&port->backlog.lock);
 553         if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
 554                 if (skb->dev)
 555                         dev_hold(skb->dev);
 556                 __skb_queue_tail(&port->backlog, skb);
 557                 spin_unlock(&port->backlog.lock);
 558                 schedule_work(&port->wq);
 559         } else {
 560                 spin_unlock(&port->backlog.lock);
 561                 atomic_long_inc(&skb->dev->rx_dropped);
 562                 kfree_skb(skb);
 563         }
 564 }
 565
 566 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 567 {
 568         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 569         void *lyr3h;
 570         struct ipvl_addr *addr;
 571         int addr_type;
 572
 573         lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 574         if (!lyr3h)
 575                 goto out;
 576
 577         if (!ipvlan_is_vepa(ipvlan->port)) {
 578                 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 579                 if (addr) {
 580                         if (ipvlan_is_private(ipvlan->port)) {
 581                                 consume_skb(skb);
 582                                 return NET_XMIT_DROP;
 583                         }
 584                         return ipvlan_rcv_frame(addr, &skb, true);
 585                 }
 586         }
 587 out:
 588         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 589         return ipvlan_process_outbound(skb);
 590 }
 591
 592 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 593 {
 594         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 595         struct ethhdr *eth = eth_hdr(skb);
 596         struct ipvl_addr *addr;
 597         void *lyr3h;
 598         int addr_type;
 599
 600         if (!ipvlan_is_vepa(ipvlan->port) &&
 601             ether_addr_equal(eth->h_dest, eth->h_source)) {
 602                 lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 603                 if (lyr3h) {
 604                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 605                         if (addr) {
 606                                 if (ipvlan_is_private(ipvlan->port)) {
 607                                         consume_skb(skb);
 608                                         return NET_XMIT_DROP;
 609                                 }
 610                                 return ipvlan_rcv_frame(addr, &skb, true);
 611                         }
 612                 }
 613                 skb = skb_share_check(skb, GFP_ATOMIC);
 614                 if (!skb)
 615                         return NET_XMIT_DROP;
 616
 617                 /* Packet definitely does not belong to any of the
 618                  * virtual devices, but the dest is local. So forward
 619                  * the skb for the main-dev. At the RX side we just return
 620                  * RX_PASS for it to be processed further on the stack.
 621                  */
 622                 return dev_forward_skb(ipvlan->phy_dev, skb);
 623
 624         } else if (is_multicast_ether_addr(eth->h_dest)) {
 625                 ipvlan_skb_crossing_ns(skb, NULL);
 626                 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
 627                 return NET_XMIT_SUCCESS;
 628         }
 629
 630         skb->dev = ipvlan->phy_dev;
 631         return dev_queue_xmit(skb);
 632 }
 633
 634 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 635 {
 636         struct ipvl_dev *ipvlan = netdev_priv(dev);
 637         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 638
 639         if (!port)
 640                 goto out;
 641
 642         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 643                 goto out;
 644
 645         switch(port->mode) {
 646         case IPVLAN_MODE_L2:
 647                 return ipvlan_xmit_mode_l2(skb, dev);
 648         case IPVLAN_MODE_L3:
 649 #ifdef CONFIG_IPVLAN_L3S
 650         case IPVLAN_MODE_L3S:
 651 #endif
 652                 return ipvlan_xmit_mode_l3(skb, dev);
 653         }
 654
 655         /* Should not reach here */
 656         WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
 657                           port->mode);
 658 out:
 659         kfree_skb(skb);
 660         return NET_XMIT_DROP;
 661 }
 662
 663 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 664 {
 665         struct ethhdr *eth = eth_hdr(skb);
 666         struct ipvl_addr *addr;
 667         void *lyr3h;
 668         int addr_type;
 669
 670         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 671                 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 672                 if (!lyr3h)
 673                         return true;
 674
 675                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 676                 if (addr)
 677                         return false;
 678         }
 679
 680         return true;
 681 }
 682
 683 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 684                                                  struct ipvl_port *port)
 685 {
 686         void *lyr3h;
 687         int addr_type;
 688         struct ipvl_addr *addr;
 689         struct sk_buff *skb = *pskb;
 690         rx_handler_result_t ret = RX_HANDLER_PASS;
 691
 692         lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 693         if (!lyr3h)
 694                 goto out;
 695
 696         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 697         if (addr)
 698                 ret = ipvlan_rcv_frame(addr, pskb, false);
 699
 700 out:
 701         return ret;
 702 }
 703
 704 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 705                                                  struct ipvl_port *port)
 706 {
 707         struct sk_buff *skb = *pskb;
 708         struct ethhdr *eth = eth_hdr(skb);
 709         rx_handler_result_t ret = RX_HANDLER_PASS;
 710
 711         if (is_multicast_ether_addr(eth->h_dest)) {
 712                 if (ipvlan_external_frame(skb, port)) {
 713                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 714
 715                         /* External frames are queued for device local
 716                          * distribution, but a copy is given to master
 717                          * straight away to avoid sending duplicates later
 718                          * when work-queue processes this frame. This is
 719                          * achieved by returning RX_HANDLER_PASS.
 720                          */
 721                         if (nskb) {
 722                                 ipvlan_skb_crossing_ns(nskb, NULL);
 723                                 ipvlan_multicast_enqueue(port, nskb, false);
 724                         }
 725                 }
 726         } else {
 727                 /* Perform like l3 mode for non-multicast packet */
 728                 ret = ipvlan_handle_mode_l3(pskb, port);
 729         }
 730
 731         return ret;
 732 }
 733
 734 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 735 {
 736         struct sk_buff *skb = *pskb;
 737         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 738
 739         if (!port)
 740                 return RX_HANDLER_PASS;
 741
 742         switch (port->mode) {
 743         case IPVLAN_MODE_L2:
 744                 return ipvlan_handle_mode_l2(pskb, port);
 745         case IPVLAN_MODE_L3:
 746                 return ipvlan_handle_mode_l3(pskb, port);
 747 #ifdef CONFIG_IPVLAN_L3S
 748         case IPVLAN_MODE_L3S:
 749                 return RX_HANDLER_PASS;
 750 #endif
 751         }
 752
 753         /* Should not reach here */
 754         WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
 755                           port->mode);
 756         kfree_skb(skb);
 757         return RX_HANDLER_CONSUMED;
 758 }