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 static u8 ipvlan_get_v6_hash(const void *iaddr)
  39 {
  40         const struct in6_addr *ip6_addr = iaddr;
  41
  42         return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
  43                IPVLAN_HASH_MASK;
  44 }
  45
  46 static u8 ipvlan_get_v4_hash(const void *iaddr)
  47 {
  48         const struct in_addr *ip4_addr = iaddr;
  49
  50         return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
  51                IPVLAN_HASH_MASK;
  52 }
  53
  54 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
  55                                                const void *iaddr, bool is_v6)
  56 {
  57         struct ipvl_addr *addr;
  58         u8 hash;
  59
  60         hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
  61                ipvlan_get_v4_hash(iaddr);
  62         hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode) {
  63                 if (is_v6 && addr->atype == IPVL_IPV6 &&
  64                     ipv6_addr_equal(&addr->ip6addr, iaddr))
  65                         return addr;
  66                 else if (!is_v6 && addr->atype == IPVL_IPV4 &&
  67                          addr->ip4addr.s_addr ==
  68                                 ((struct in_addr *)iaddr)->s_addr)
  69                         return addr;
  70         }
  71         return NULL;
  72 }
  73
  74 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
  75 {
  76         struct ipvl_port *port = ipvlan->port;
  77         u8 hash;
  78
  79         hash = (addr->atype == IPVL_IPV6) ?
  80                ipvlan_get_v6_hash(&addr->ip6addr) :
  81                ipvlan_get_v4_hash(&addr->ip4addr);
  82         if (hlist_unhashed(&addr->hlnode))
  83                 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
  84 }
  85
  86 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
  87 {
  88         hlist_del_init_rcu(&addr->hlnode);
  89 }
  90
  91 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
  92                                    const void *iaddr, bool is_v6)
  93 {
  94         struct ipvl_addr *addr;
  95
  96         list_for_each_entry(addr, &ipvlan->addrs, anode) {
  97                 if ((is_v6 && addr->atype == IPVL_IPV6 &&
  98                     ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
  99                     (!is_v6 && addr->atype == IPVL_IPV4 &&
 100                     addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
 101                         return addr;
 102         }
 103         return NULL;
 104 }
 105
 106 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
 107 {
 108         struct ipvl_dev *ipvlan;
 109
 110         ASSERT_RTNL();
 111
 112         list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 113                 if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
 114                         return true;
 115         }
 116         return false;
 117 }
 118
 119 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
 120 {
 121         void *lyr3h = NULL;
 122
 123         switch (skb->protocol) {
 124         case htons(ETH_P_ARP): {
 125                 struct arphdr *arph;
 126
 127                 if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
 128                         return NULL;
 129
 130                 arph = arp_hdr(skb);
 131                 *type = IPVL_ARP;
 132                 lyr3h = arph;
 133                 break;
 134         }
 135         case htons(ETH_P_IP): {
 136                 u32 pktlen;
 137                 struct iphdr *ip4h;
 138
 139                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
 140                         return NULL;
 141
 142                 ip4h = ip_hdr(skb);
 143                 pktlen = ntohs(ip4h->tot_len);
 144                 if (ip4h->ihl < 5 || ip4h->version != 4)
 145                         return NULL;
 146                 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
 147                         return NULL;
 148
 149                 *type = IPVL_IPV4;
 150                 lyr3h = ip4h;
 151                 break;
 152         }
 153         case htons(ETH_P_IPV6): {
 154                 struct ipv6hdr *ip6h;
 155
 156                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
 157                         return NULL;
 158
 159                 ip6h = ipv6_hdr(skb);
 160                 if (ip6h->version != 6)
 161                         return NULL;
 162
 163                 *type = IPVL_IPV6;
 164                 lyr3h = ip6h;
 165                 /* Only Neighbour Solicitation pkts need different treatment */
 166                 if (ipv6_addr_any(&ip6h->saddr) &&
 167                     ip6h->nexthdr == NEXTHDR_ICMP) {
 168                         *type = IPVL_ICMPV6;
 169                         lyr3h = ip6h + 1;
 170                 }
 171                 break;
 172         }
 173         default:
 174                 return NULL;
 175         }
 176
 177         return lyr3h;
 178 }
 179
 180 unsigned int ipvlan_mac_hash(const unsigned char *addr)
 181 {
 182         u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
 183                                ipvlan_jhash_secret);
 184
 185         return hash & IPVLAN_MAC_FILTER_MASK;
 186 }
 187
 188 void ipvlan_process_multicast(struct work_struct *work)
 189 {
 190         struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
 191         struct ethhdr *ethh;
 192         struct ipvl_dev *ipvlan;
 193         struct sk_buff *skb, *nskb;
 194         struct sk_buff_head list;
 195         unsigned int len;
 196         unsigned int mac_hash;
 197         int ret;
 198         u8 pkt_type;
 199         bool tx_pkt;
 200
 201         __skb_queue_head_init(&list);
 202
 203         spin_lock_bh(&port->backlog.lock);
 204         skb_queue_splice_tail_init(&port->backlog, &list);
 205         spin_unlock_bh(&port->backlog.lock);
 206
 207         while ((skb = __skb_dequeue(&list)) != NULL) {
 208                 struct net_device *dev = skb->dev;
 209                 bool consumed = false;
 210
 211                 ethh = eth_hdr(skb);
 212                 tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
 213                 mac_hash = ipvlan_mac_hash(ethh->h_dest);
 214
 215                 if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
 216                         pkt_type = PACKET_BROADCAST;
 217                 else
 218                         pkt_type = PACKET_MULTICAST;
 219
 220                 rcu_read_lock();
 221                 list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 222                         if (tx_pkt && (ipvlan->dev == skb->dev))
 223                                 continue;
 224                         if (!test_bit(mac_hash, ipvlan->mac_filters))
 225                                 continue;
 226                         if (!(ipvlan->dev->flags & IFF_UP))
 227                                 continue;
 228                         ret = NET_RX_DROP;
 229                         len = skb->len + ETH_HLEN;
 230                         nskb = skb_clone(skb, GFP_ATOMIC);
 231                         local_bh_disable();
 232                         if (nskb) {
 233                                 consumed = true;
 234                                 nskb->pkt_type = pkt_type;
 235                                 nskb->dev = ipvlan->dev;
 236                                 if (tx_pkt)
 237                                         ret = dev_forward_skb(ipvlan->dev, nskb);
 238                                 else
 239                                         ret = netif_rx(nskb);
 240                         }
 241                         ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 242                         local_bh_enable();
 243                 }
 244                 rcu_read_unlock();
 245
 246                 if (tx_pkt) {
 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                         if (consumed)
 253                                 consume_skb(skb);
 254                         else
 255                                 kfree_skb(skb);
 256                 }
 257                 if (dev)
 258                         dev_put(dev);
 259                 cond_resched();
 260         }
 261 }
 262
 263 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 264 {
 265         bool xnet = true;
 266
 267         if (dev)
 268                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 269
 270         skb_scrub_packet(skb, xnet);
 271         if (dev)
 272                 skb->dev = dev;
 273 }
 274
 275 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 276                             bool local)
 277 {
 278         struct ipvl_dev *ipvlan = addr->master;
 279         struct net_device *dev = ipvlan->dev;
 280         unsigned int len;
 281         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 282         bool success = false;
 283         struct sk_buff *skb = *pskb;
 284
 285         len = skb->len + ETH_HLEN;
 286         /* Only packets exchanged between two local slaves need to have
 287          * device-up check as well as skb-share check.
 288          */
 289         if (local) {
 290                 if (unlikely(!(dev->flags & IFF_UP))) {
 291                         kfree_skb(skb);
 292                         goto out;
 293                 }
 294
 295                 skb = skb_share_check(skb, GFP_ATOMIC);
 296                 if (!skb)
 297                         goto out;
 298
 299                 *pskb = skb;
 300         }
 301         ipvlan_skb_crossing_ns(skb, dev);
 302
 303         if (local) {
 304                 skb->pkt_type = PACKET_HOST;
 305                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 306                         success = true;
 307         } else {
 308                 if (!ether_addr_equal_64bits(eth_hdr(skb)->h_dest,
 309                                              ipvlan->phy_dev->dev_addr))
 310                         skb->pkt_type = PACKET_OTHERHOST;
 311
 312                 ret = RX_HANDLER_ANOTHER;
 313                 success = true;
 314         }
 315
 316 out:
 317         ipvlan_count_rx(ipvlan, len, success, false);
 318         return ret;
 319 }
 320
 321 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
 322                                             void *lyr3h, int addr_type,
 323                                             bool use_dest)
 324 {
 325         struct ipvl_addr *addr = NULL;
 326
 327         if (addr_type == IPVL_IPV6) {
 328                 struct ipv6hdr *ip6h;
 329                 struct in6_addr *i6addr;
 330
 331                 ip6h = (struct ipv6hdr *)lyr3h;
 332                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 333                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 334         } else if (addr_type == IPVL_ICMPV6) {
 335                 struct nd_msg *ndmh;
 336                 struct in6_addr *i6addr;
 337
 338                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 339                  * are handled to avoid DAD issue.
 340                  */
 341                 ndmh = (struct nd_msg *)lyr3h;
 342                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 343                         i6addr = &ndmh->target;
 344                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 345                 }
 346         } else if (addr_type == IPVL_IPV4) {
 347                 struct iphdr *ip4h;
 348                 __be32 *i4addr;
 349
 350                 ip4h = (struct iphdr *)lyr3h;
 351                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 352                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 353         } else if (addr_type == IPVL_ARP) {
 354                 struct arphdr *arph;
 355                 unsigned char *arp_ptr;
 356                 __be32 dip;
 357
 358                 arph = (struct arphdr *)lyr3h;
 359                 arp_ptr = (unsigned char *)(arph + 1);
 360                 if (use_dest)
 361                         arp_ptr += (2 * port->dev->addr_len) + 4;
 362                 else
 363                         arp_ptr += port->dev->addr_len;
 364
 365                 memcpy(&dip, arp_ptr, 4);
 366                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 367         }
 368
 369         return addr;
 370 }
 371
 372 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 373 {
 374         const struct iphdr *ip4h = ip_hdr(skb);
 375         struct net_device *dev = skb->dev;
 376         struct net *net = dev_net(dev);
 377         struct rtable *rt;
 378         int err, ret = NET_XMIT_DROP;
 379         struct flowi4 fl4 = {
 380                 .flowi4_oif = dev->ifindex,
 381                 .flowi4_tos = RT_TOS(ip4h->tos),
 382                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 383                 .flowi4_mark = skb->mark,
 384                 .daddr = ip4h->daddr,
 385                 .saddr = ip4h->saddr,
 386         };
 387
 388         rt = ip_route_output_flow(net, &fl4, NULL);
 389         if (IS_ERR(rt))
 390                 goto err;
 391
 392         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 393                 ip_rt_put(rt);
 394                 goto err;
 395         }
 396         skb_dst_set(skb, &rt->dst);
 397         err = ip_local_out(net, skb->sk, skb);
 398         if (unlikely(net_xmit_eval(err)))
 399                 dev->stats.tx_errors++;
 400         else
 401                 ret = NET_XMIT_SUCCESS;
 402         goto out;
 403 err:
 404         dev->stats.tx_errors++;
 405         kfree_skb(skb);
 406 out:
 407         return ret;
 408 }
 409
 410 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 411 {
 412         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 413         struct net_device *dev = skb->dev;
 414         struct net *net = dev_net(dev);
 415         struct dst_entry *dst;
 416         int err, ret = NET_XMIT_DROP;
 417         struct flowi6 fl6 = {
 418                 .flowi6_oif = dev->ifindex,
 419                 .daddr = ip6h->daddr,
 420                 .saddr = ip6h->saddr,
 421                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 422                 .flowlabel = ip6_flowinfo(ip6h),
 423                 .flowi6_mark = skb->mark,
 424                 .flowi6_proto = ip6h->nexthdr,
 425         };
 426
 427         dst = ip6_route_output(net, NULL, &fl6);
 428         if (dst->error) {
 429                 ret = dst->error;
 430                 dst_release(dst);
 431                 goto err;
 432         }
 433         skb_dst_set(skb, dst);
 434         err = ip6_local_out(net, skb->sk, skb);
 435         if (unlikely(net_xmit_eval(err)))
 436                 dev->stats.tx_errors++;
 437         else
 438                 ret = NET_XMIT_SUCCESS;
 439         goto out;
 440 err:
 441         dev->stats.tx_errors++;
 442         kfree_skb(skb);
 443 out:
 444         return ret;
 445 }
 446
 447 static int ipvlan_process_outbound(struct sk_buff *skb)
 448 {
 449         struct ethhdr *ethh = eth_hdr(skb);
 450         int ret = NET_XMIT_DROP;
 451
 452         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 453          * will have L2; which need to discarded and processed further
 454          * in the net-ns of the main-device.
 455          */
 456         if (skb_mac_header_was_set(skb)) {
 457                 /* In this mode we dont care about
 458                  * multicast and broadcast traffic */
 459                 if (is_multicast_ether_addr(ethh->h_dest)) {
 460                         pr_debug_ratelimited(
 461                                 "Dropped {multi|broad}cast of type=[%x]\n",
 462                                 ntohs(skb->protocol));
 463                         kfree_skb(skb);
 464                         goto out;
 465                 }
 466
 467                 skb_pull(skb, sizeof(*ethh));
 468                 skb->mac_header = (typeof(skb->mac_header))~0U;
 469                 skb_reset_network_header(skb);
 470         }
 471
 472         if (skb->protocol == htons(ETH_P_IPV6))
 473                 ret = ipvlan_process_v6_outbound(skb);
 474         else if (skb->protocol == htons(ETH_P_IP))
 475                 ret = ipvlan_process_v4_outbound(skb);
 476         else {
 477                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 478                                     ntohs(skb->protocol));
 479                 kfree_skb(skb);
 480         }
 481 out:
 482         return ret;
 483 }
 484
 485 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
 486                                      struct sk_buff *skb, bool tx_pkt)
 487 {
 488         if (skb->protocol == htons(ETH_P_PAUSE)) {
 489                 kfree_skb(skb);
 490                 return;
 491         }
 492
 493         /* Record that the deferred packet is from TX or RX path. By
 494          * looking at mac-addresses on packet will lead to erronus decisions.
 495          * (This would be true for a loopback-mode on master device or a
 496          * hair-pin mode of the switch.)
 497          */
 498         IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
 499
 500         spin_lock(&port->backlog.lock);
 501         if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
 502                 if (skb->dev)
 503                         dev_hold(skb->dev);
 504                 __skb_queue_tail(&port->backlog, skb);
 505                 spin_unlock(&port->backlog.lock);
 506                 schedule_work(&port->wq);
 507         } else {
 508                 spin_unlock(&port->backlog.lock);
 509                 atomic_long_inc(&skb->dev->rx_dropped);
 510                 kfree_skb(skb);
 511         }
 512 }
 513
 514 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 515 {
 516         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 517         void *lyr3h;
 518         struct ipvl_addr *addr;
 519         int addr_type;
 520
 521         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 522         if (!lyr3h)
 523                 goto out;
 524
 525         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 526         if (addr)
 527                 return ipvlan_rcv_frame(addr, &skb, true);
 528
 529 out:
 530         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 531         return ipvlan_process_outbound(skb);
 532 }
 533
 534 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 535 {
 536         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 537         struct ethhdr *eth = eth_hdr(skb);
 538         struct ipvl_addr *addr;
 539         void *lyr3h;
 540         int addr_type;
 541
 542         if (ether_addr_equal(eth->h_dest, eth->h_source)) {
 543                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 544                 if (lyr3h) {
 545                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 546                         if (addr)
 547                                 return ipvlan_rcv_frame(addr, &skb, true);
 548                 }
 549                 skb = skb_share_check(skb, GFP_ATOMIC);
 550                 if (!skb)
 551                         return NET_XMIT_DROP;
 552
 553                 /* Packet definitely does not belong to any of the
 554                  * virtual devices, but the dest is local. So forward
 555                  * the skb for the main-dev. At the RX side we just return
 556                  * RX_PASS for it to be processed further on the stack.
 557                  */
 558                 return dev_forward_skb(ipvlan->phy_dev, skb);
 559
 560         } else if (is_multicast_ether_addr(eth->h_dest)) {
 561                 ipvlan_skb_crossing_ns(skb, NULL);
 562                 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
 563                 return NET_XMIT_SUCCESS;
 564         }
 565
 566         ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
 567         return dev_queue_xmit(skb);
 568 }
 569
 570 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 571 {
 572         struct ipvl_dev *ipvlan = netdev_priv(dev);
 573         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 574
 575         if (!port)
 576                 goto out;
 577
 578         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 579                 goto out;
 580
 581         switch(port->mode) {
 582         case IPVLAN_MODE_L2:
 583                 return ipvlan_xmit_mode_l2(skb, dev);
 584         case IPVLAN_MODE_L3:
 585         case IPVLAN_MODE_L3S:
 586                 return ipvlan_xmit_mode_l3(skb, dev);
 587         }
 588
 589         /* Should not reach here */
 590         WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
 591                           port->mode);
 592 out:
 593         kfree_skb(skb);
 594         return NET_XMIT_DROP;
 595 }
 596
 597 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 598 {
 599         struct ethhdr *eth = eth_hdr(skb);
 600         struct ipvl_addr *addr;
 601         void *lyr3h;
 602         int addr_type;
 603
 604         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 605                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 606                 if (!lyr3h)
 607                         return true;
 608
 609                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 610                 if (addr)
 611                         return false;
 612         }
 613
 614         return true;
 615 }
 616
 617 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 618                                                  struct ipvl_port *port)
 619 {
 620         void *lyr3h;
 621         int addr_type;
 622         struct ipvl_addr *addr;
 623         struct sk_buff *skb = *pskb;
 624         rx_handler_result_t ret = RX_HANDLER_PASS;
 625
 626         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 627         if (!lyr3h)
 628                 goto out;
 629
 630         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 631         if (addr)
 632                 ret = ipvlan_rcv_frame(addr, pskb, false);
 633
 634 out:
 635         return ret;
 636 }
 637
 638 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 639                                                  struct ipvl_port *port)
 640 {
 641         struct sk_buff *skb = *pskb;
 642         struct ethhdr *eth = eth_hdr(skb);
 643         rx_handler_result_t ret = RX_HANDLER_PASS;
 644         void *lyr3h;
 645         int addr_type;
 646
 647         if (is_multicast_ether_addr(eth->h_dest)) {
 648                 if (ipvlan_external_frame(skb, port)) {
 649                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 650
 651                         /* External frames are queued for device local
 652                          * distribution, but a copy is given to master
 653                          * straight away to avoid sending duplicates later
 654                          * when work-queue processes this frame. This is
 655                          * achieved by returning RX_HANDLER_PASS.
 656                          */
 657                         if (nskb) {
 658                                 ipvlan_skb_crossing_ns(nskb, NULL);
 659                                 ipvlan_multicast_enqueue(port, nskb, false);
 660                         }
 661                 }
 662         } else {
 663                 struct ipvl_addr *addr;
 664
 665                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 666                 if (!lyr3h)
 667                         return ret;
 668
 669                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 670                 if (addr)
 671                         ret = ipvlan_rcv_frame(addr, pskb, false);
 672         }
 673
 674         return ret;
 675 }
 676
 677 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 678 {
 679         struct sk_buff *skb = *pskb;
 680         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 681
 682         if (!port)
 683                 return RX_HANDLER_PASS;
 684
 685         switch (port->mode) {
 686         case IPVLAN_MODE_L2:
 687                 return ipvlan_handle_mode_l2(pskb, port);
 688         case IPVLAN_MODE_L3:
 689                 return ipvlan_handle_mode_l3(pskb, port);
 690         case IPVLAN_MODE_L3S:
 691                 return RX_HANDLER_PASS;
 692         }
 693
 694         /* Should not reach here */
 695         WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
 696                           port->mode);
 697         kfree_skb(skb);
 698         return RX_HANDLER_CONSUMED;
 699 }
 700
 701 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
 702                                             struct net_device *dev)
 703 {
 704         struct ipvl_addr *addr = NULL;
 705         struct ipvl_port *port;
 706         void *lyr3h;
 707         int addr_type;
 708
 709         if (!dev || !netif_is_ipvlan_port(dev))
 710                 goto out;
 711
 712         port = ipvlan_port_get_rcu(dev);
 713         if (!port || port->mode != IPVLAN_MODE_L3S)
 714                 goto out;
 715
 716         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 717         if (!lyr3h)
 718                 goto out;
 719
 720         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 721 out:
 722         return addr;
 723 }
 724
 725 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
 726                               u16 proto)
 727 {
 728         struct ipvl_addr *addr;
 729         struct net_device *sdev;
 730
 731         addr = ipvlan_skb_to_addr(skb, dev);
 732         if (!addr)
 733                 goto out;
 734
 735         sdev = addr->master->dev;
 736         switch (proto) {
 737         case AF_INET:
 738         {
 739                 int err;
 740                 struct iphdr *ip4h = ip_hdr(skb);
 741
 742                 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
 743                                            ip4h->tos, sdev);
 744                 if (unlikely(err))
 745                         goto out;
 746                 break;
 747         }
 748         case AF_INET6:
 749         {
 750                 struct dst_entry *dst;
 751                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 752                 int flags = RT6_LOOKUP_F_HAS_SADDR;
 753                 struct flowi6 fl6 = {
 754                         .flowi6_iif   = sdev->ifindex,
 755                         .daddr        = ip6h->daddr,
 756                         .saddr        = ip6h->saddr,
 757                         .flowlabel    = ip6_flowinfo(ip6h),
 758                         .flowi6_mark  = skb->mark,
 759                         .flowi6_proto = ip6h->nexthdr,
 760                 };
 761
 762                 skb_dst_drop(skb);
 763                 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
 764                 skb_dst_set(skb, dst);
 765                 break;
 766         }
 767         default:
 768                 break;
 769         }
 770
 771 out:
 772         return skb;
 773 }
 774
 775 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
 776                              const struct nf_hook_state *state)
 777 {
 778         struct ipvl_addr *addr;
 779         unsigned int len;
 780
 781         addr = ipvlan_skb_to_addr(skb, skb->dev);
 782         if (!addr)
 783                 goto out;
 784
 785         skb->dev = addr->master->dev;
 786         len = skb->len + ETH_HLEN;
 787         ipvlan_count_rx(addr->master, len, true, false);
 788 out:
 789         return NF_ACCEPT;
 790 }