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 static 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                 cond_resched();
 302         }
 303 }
 304
 305 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
 306 {
 307         bool xnet = true;
 308
 309         if (dev)
 310                 xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
 311
 312         skb_scrub_packet(skb, xnet);
 313         if (dev)
 314                 skb->dev = dev;
 315 }
 316
 317 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
 318                             bool local)
 319 {
 320         struct ipvl_dev *ipvlan = addr->master;
 321         struct net_device *dev = ipvlan->dev;
 322         unsigned int len;
 323         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 324         bool success = false;
 325         struct sk_buff *skb = *pskb;
 326
 327         len = skb->len + ETH_HLEN;
 328         /* Only packets exchanged between two local slaves need to have
 329          * device-up check as well as skb-share check.
 330          */
 331         if (local) {
 332                 if (unlikely(!(dev->flags & IFF_UP))) {
 333                         kfree_skb(skb);
 334                         goto out;
 335                 }
 336
 337                 skb = skb_share_check(skb, GFP_ATOMIC);
 338                 if (!skb)
 339                         goto out;
 340
 341                 *pskb = skb;
 342         }
 343
 344         if (local) {
 345                 skb->pkt_type = PACKET_HOST;
 346                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 347                         success = true;
 348         } else {
 349                 skb->dev = dev;
 350                 ret = RX_HANDLER_ANOTHER;
 351                 success = true;
 352         }
 353
 354 out:
 355         ipvlan_count_rx(ipvlan, len, success, false);
 356         return ret;
 357 }
 358
 359 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
 360                                             void *lyr3h, int addr_type,
 361                                             bool use_dest)
 362 {
 363         struct ipvl_addr *addr = NULL;
 364
 365         switch (addr_type) {
 366 #if IS_ENABLED(CONFIG_IPV6)
 367         case IPVL_IPV6: {
 368                 struct ipv6hdr *ip6h;
 369                 struct in6_addr *i6addr;
 370
 371                 ip6h = (struct ipv6hdr *)lyr3h;
 372                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 373                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 374                 break;
 375         }
 376         case IPVL_ICMPV6: {
 377                 struct nd_msg *ndmh;
 378                 struct in6_addr *i6addr;
 379
 380                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 381                  * are handled to avoid DAD issue.
 382                  */
 383                 ndmh = (struct nd_msg *)lyr3h;
 384                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 385                         i6addr = &ndmh->target;
 386                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 387                 }
 388                 break;
 389         }
 390 #endif
 391         case IPVL_IPV4: {
 392                 struct iphdr *ip4h;
 393                 __be32 *i4addr;
 394
 395                 ip4h = (struct iphdr *)lyr3h;
 396                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 397                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 398                 break;
 399         }
 400         case IPVL_ARP: {
 401                 struct arphdr *arph;
 402                 unsigned char *arp_ptr;
 403                 __be32 dip;
 404
 405                 arph = (struct arphdr *)lyr3h;
 406                 arp_ptr = (unsigned char *)(arph + 1);
 407                 if (use_dest)
 408                         arp_ptr += (2 * port->dev->addr_len) + 4;
 409                 else
 410                         arp_ptr += port->dev->addr_len;
 411
 412                 memcpy(&dip, arp_ptr, 4);
 413                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 414                 break;
 415         }
 416         }
 417
 418         return addr;
 419 }
 420
 421 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 422 {
 423         const struct iphdr *ip4h = ip_hdr(skb);
 424         struct net_device *dev = skb->dev;
 425         struct net *net = dev_net(dev);
 426         struct rtable *rt;
 427         int err, ret = NET_XMIT_DROP;
 428         struct flowi4 fl4 = {
 429                 .flowi4_oif = dev->ifindex,
 430                 .flowi4_tos = RT_TOS(ip4h->tos),
 431                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 432                 .flowi4_mark = skb->mark,
 433                 .daddr = ip4h->daddr,
 434                 .saddr = ip4h->saddr,
 435         };
 436
 437         rt = ip_route_output_flow(net, &fl4, NULL);
 438         if (IS_ERR(rt))
 439                 goto err;
 440
 441         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 442                 ip_rt_put(rt);
 443                 goto err;
 444         }
 445         skb_dst_set(skb, &rt->dst);
 446         err = ip_local_out(net, skb->sk, skb);
 447         if (unlikely(net_xmit_eval(err)))
 448                 dev->stats.tx_errors++;
 449         else
 450                 ret = NET_XMIT_SUCCESS;
 451         goto out;
 452 err:
 453         dev->stats.tx_errors++;
 454         kfree_skb(skb);
 455 out:
 456         return ret;
 457 }
 458
 459 #if IS_ENABLED(CONFIG_IPV6)
 460 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 461 {
 462         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 463         struct net_device *dev = skb->dev;
 464         struct net *net = dev_net(dev);
 465         struct dst_entry *dst;
 466         int err, ret = NET_XMIT_DROP;
 467         struct flowi6 fl6 = {
 468                 .flowi6_oif = dev->ifindex,
 469                 .daddr = ip6h->daddr,
 470                 .saddr = ip6h->saddr,
 471                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 472                 .flowlabel = ip6_flowinfo(ip6h),
 473                 .flowi6_mark = skb->mark,
 474                 .flowi6_proto = ip6h->nexthdr,
 475         };
 476
 477         dst = ip6_route_output(net, NULL, &fl6);
 478         if (dst->error) {
 479                 ret = dst->error;
 480                 dst_release(dst);
 481                 goto err;
 482         }
 483         skb_dst_set(skb, dst);
 484         err = ip6_local_out(net, skb->sk, skb);
 485         if (unlikely(net_xmit_eval(err)))
 486                 dev->stats.tx_errors++;
 487         else
 488                 ret = NET_XMIT_SUCCESS;
 489         goto out;
 490 err:
 491         dev->stats.tx_errors++;
 492         kfree_skb(skb);
 493 out:
 494         return ret;
 495 }
 496 #else
 497 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 498 {
 499         return NET_XMIT_DROP;
 500 }
 501 #endif
 502
 503 static int ipvlan_process_outbound(struct sk_buff *skb)
 504 {
 505         struct ethhdr *ethh = eth_hdr(skb);
 506         int ret = NET_XMIT_DROP;
 507
 508         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 509          * will have L2; which need to discarded and processed further
 510          * in the net-ns of the main-device.
 511          */
 512         if (skb_mac_header_was_set(skb)) {
 513                 /* In this mode we dont care about
 514                  * multicast and broadcast traffic */
 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                 if (skb->dev)
 559                         dev_hold(skb->dev);
 560                 __skb_queue_tail(&port->backlog, skb);
 561                 spin_unlock(&port->backlog.lock);
 562                 schedule_work(&port->wq);
 563         } else {
 564                 spin_unlock(&port->backlog.lock);
 565                 atomic_long_inc(&skb->dev->rx_dropped);
 566                 kfree_skb(skb);
 567         }
 568 }
 569
 570 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 571 {
 572         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 573         void *lyr3h;
 574         struct ipvl_addr *addr;
 575         int addr_type;
 576
 577         lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
 578         if (!lyr3h)
 579                 goto out;
 580
 581         if (!ipvlan_is_vepa(ipvlan->port)) {
 582                 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 583                 if (addr) {
 584                         if (ipvlan_is_private(ipvlan->port)) {
 585                                 consume_skb(skb);
 586                                 return NET_XMIT_DROP;
 587                         }
 588                         return ipvlan_rcv_frame(addr, &skb, true);
 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 = 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                                 return ipvlan_rcv_frame(addr, &skb, true);
 615                         }
 616                 }
 617                 skb = skb_share_check(skb, GFP_ATOMIC);
 618                 if (!skb)
 619                         return NET_XMIT_DROP;
 620
 621                 /* Packet definitely does not belong to any of the
 622                  * virtual devices, but the dest is local. So forward
 623                  * the skb for the main-dev. At the RX side we just return
 624                  * RX_PASS for it to be processed further on the stack.
 625                  */
 626                 return dev_forward_skb(ipvlan->phy_dev, skb);
 627
 628         } else if (is_multicast_ether_addr(eth->h_dest)) {
 629                 ipvlan_skb_crossing_ns(skb, NULL);
 630                 ipvlan_multicast_enqueue(ipvlan->port, skb, true);
 631                 return NET_XMIT_SUCCESS;
 632         }
 633
 634         skb->dev = ipvlan->phy_dev;
 635         return dev_queue_xmit(skb);
 636 }
 637
 638 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 639 {
 640         struct ipvl_dev *ipvlan = netdev_priv(dev);
 641         struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
 642
 643         if (!port)
 644                 goto out;
 645
 646         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 647                 goto out;
 648
 649         switch(port->mode) {
 650         case IPVLAN_MODE_L2:
 651                 return ipvlan_xmit_mode_l2(skb, dev);
 652         case IPVLAN_MODE_L3:
 653         case IPVLAN_MODE_L3S:
 654                 return ipvlan_xmit_mode_l3(skb, dev);
 655         }
 656
 657         /* Should not reach here */
 658         WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
 659                           port->mode);
 660 out:
 661         kfree_skb(skb);
 662         return NET_XMIT_DROP;
 663 }
 664
 665 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 666 {
 667         struct ethhdr *eth = eth_hdr(skb);
 668         struct ipvl_addr *addr;
 669         void *lyr3h;
 670         int addr_type;
 671
 672         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 673                 lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 674                 if (!lyr3h)
 675                         return true;
 676
 677                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 678                 if (addr)
 679                         return false;
 680         }
 681
 682         return true;
 683 }
 684
 685 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 686                                                  struct ipvl_port *port)
 687 {
 688         void *lyr3h;
 689         int addr_type;
 690         struct ipvl_addr *addr;
 691         struct sk_buff *skb = *pskb;
 692         rx_handler_result_t ret = RX_HANDLER_PASS;
 693
 694         lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 695         if (!lyr3h)
 696                 goto out;
 697
 698         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 699         if (addr)
 700                 ret = ipvlan_rcv_frame(addr, pskb, false);
 701
 702 out:
 703         return ret;
 704 }
 705
 706 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 707                                                  struct ipvl_port *port)
 708 {
 709         struct sk_buff *skb = *pskb;
 710         struct ethhdr *eth = eth_hdr(skb);
 711         rx_handler_result_t ret = RX_HANDLER_PASS;
 712
 713         if (is_multicast_ether_addr(eth->h_dest)) {
 714                 if (ipvlan_external_frame(skb, port)) {
 715                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 716
 717                         /* External frames are queued for device local
 718                          * distribution, but a copy is given to master
 719                          * straight away to avoid sending duplicates later
 720                          * when work-queue processes this frame. This is
 721                          * achieved by returning RX_HANDLER_PASS.
 722                          */
 723                         if (nskb) {
 724                                 ipvlan_skb_crossing_ns(nskb, NULL);
 725                                 ipvlan_multicast_enqueue(port, nskb, false);
 726                         }
 727                 }
 728         } else {
 729                 /* Perform like l3 mode for non-multicast packet */
 730                 ret = ipvlan_handle_mode_l3(pskb, port);
 731         }
 732
 733         return ret;
 734 }
 735
 736 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 737 {
 738         struct sk_buff *skb = *pskb;
 739         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 740
 741         if (!port)
 742                 return RX_HANDLER_PASS;
 743
 744         switch (port->mode) {
 745         case IPVLAN_MODE_L2:
 746                 return ipvlan_handle_mode_l2(pskb, port);
 747         case IPVLAN_MODE_L3:
 748                 return ipvlan_handle_mode_l3(pskb, port);
 749         case IPVLAN_MODE_L3S:
 750                 return RX_HANDLER_PASS;
 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 }
 759
 760 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
 761                                             struct net_device *dev)
 762 {
 763         struct ipvl_addr *addr = NULL;
 764         struct ipvl_port *port;
 765         void *lyr3h;
 766         int addr_type;
 767
 768         if (!dev || !netif_is_ipvlan_port(dev))
 769                 goto out;
 770
 771         port = ipvlan_port_get_rcu(dev);
 772         if (!port || port->mode != IPVLAN_MODE_L3S)
 773                 goto out;
 774
 775         lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
 776         if (!lyr3h)
 777                 goto out;
 778
 779         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 780 out:
 781         return addr;
 782 }
 783
 784 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
 785                               u16 proto)
 786 {
 787         struct ipvl_addr *addr;
 788         struct net_device *sdev;
 789
 790         addr = ipvlan_skb_to_addr(skb, dev);
 791         if (!addr)
 792                 goto out;
 793
 794         sdev = addr->master->dev;
 795         switch (proto) {
 796         case AF_INET:
 797         {
 798                 int err;
 799                 struct iphdr *ip4h = ip_hdr(skb);
 800
 801                 err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
 802                                            ip4h->tos, sdev);
 803                 if (unlikely(err))
 804                         goto out;
 805                 break;
 806         }
 807 #if IS_ENABLED(CONFIG_IPV6)
 808         case AF_INET6:
 809         {
 810                 struct dst_entry *dst;
 811                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 812                 int flags = RT6_LOOKUP_F_HAS_SADDR;
 813                 struct flowi6 fl6 = {
 814                         .flowi6_iif   = sdev->ifindex,
 815                         .daddr        = ip6h->daddr,
 816                         .saddr        = ip6h->saddr,
 817                         .flowlabel    = ip6_flowinfo(ip6h),
 818                         .flowi6_mark  = skb->mark,
 819                         .flowi6_proto = ip6h->nexthdr,
 820                 };
 821
 822                 skb_dst_drop(skb);
 823                 dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6,
 824                                              skb, flags);
 825                 skb_dst_set(skb, dst);
 826                 break;
 827         }
 828 #endif
 829         default:
 830                 break;
 831         }
 832
 833 out:
 834         return skb;
 835 }
 836
 837 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
 838                              const struct nf_hook_state *state)
 839 {
 840         struct ipvl_addr *addr;
 841         unsigned int len;
 842
 843         addr = ipvlan_skb_to_addr(skb, skb->dev);
 844         if (!addr)
 845                 goto out;
 846
 847         skb->dev = addr->master->dev;
 848         len = skb->len + ETH_HLEN;
 849         ipvlan_count_rx(addr->master, len, true, false);
 850 out:
 851         return NF_ACCEPT;
 852 }