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 static void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
  20                             unsigned int len, bool success, bool mcast)
  21 {
  22         if (!ipvlan)
  23                 return;
  24
  25         if (likely(success)) {
  26                 struct ipvl_pcpu_stats *pcptr;
  27
  28                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
  29                 u64_stats_update_begin(&pcptr->syncp);
  30                 pcptr->rx_pkts++;
  31                 pcptr->rx_bytes += len;
  32                 if (mcast)
  33                         pcptr->rx_mcast++;
  34                 u64_stats_update_end(&pcptr->syncp);
  35         } else {
  36                 this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
  37         }
  38 }
  39
  40 static u8 ipvlan_get_v6_hash(const void *iaddr)
  41 {
  42         const struct in6_addr *ip6_addr = iaddr;
  43
  44         return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
  45                IPVLAN_HASH_MASK;
  46 }
  47
  48 static u8 ipvlan_get_v4_hash(const void *iaddr)
  49 {
  50         const struct in_addr *ip4_addr = iaddr;
  51
  52         return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
  53                IPVLAN_HASH_MASK;
  54 }
  55
  56 struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
  57                                         const void *iaddr, bool is_v6)
  58 {
  59         struct ipvl_addr *addr;
  60         u8 hash;
  61
  62         hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
  63                ipvlan_get_v4_hash(iaddr);
  64         hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode) {
  65                 if (is_v6 && addr->atype == IPVL_IPV6 &&
  66                     ipv6_addr_equal(&addr->ip6addr, iaddr))
  67                         return addr;
  68                 else if (!is_v6 && addr->atype == IPVL_IPV4 &&
  69                          addr->ip4addr.s_addr ==
  70                                 ((struct in_addr *)iaddr)->s_addr)
  71                         return addr;
  72         }
  73         return NULL;
  74 }
  75
  76 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
  77 {
  78         struct ipvl_port *port = ipvlan->port;
  79         u8 hash;
  80
  81         hash = (addr->atype == IPVL_IPV6) ?
  82                ipvlan_get_v6_hash(&addr->ip6addr) :
  83                ipvlan_get_v4_hash(&addr->ip4addr);
  84         if (hlist_unhashed(&addr->hlnode))
  85                 hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
  86 }
  87
  88 void ipvlan_ht_addr_del(struct ipvl_addr *addr, bool sync)
  89 {
  90         hlist_del_init_rcu(&addr->hlnode);
  91         if (sync)
  92                 synchronize_rcu();
  93 }
  94
  95 bool ipvlan_addr_busy(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
  96 {
  97         struct ipvl_port *port = ipvlan->port;
  98         struct ipvl_addr *addr;
  99
 100         list_for_each_entry(addr, &ipvlan->addrs, anode) {
 101                 if ((is_v6 && addr->atype == IPVL_IPV6 &&
 102                     ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
 103                     (!is_v6 && addr->atype == IPVL_IPV4 &&
 104                     addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
 105                         return true;
 106         }
 107
 108         if (ipvlan_ht_addr_lookup(port, iaddr, is_v6))
 109                 return true;
 110
 111         return false;
 112 }
 113
 114 static void *ipvlan_get_L3_hdr(struct sk_buff *skb, int *type)
 115 {
 116         void *lyr3h = NULL;
 117
 118         switch (skb->protocol) {
 119         case htons(ETH_P_ARP): {
 120                 struct arphdr *arph;
 121
 122                 if (unlikely(!pskb_may_pull(skb, sizeof(*arph))))
 123                         return NULL;
 124
 125                 arph = arp_hdr(skb);
 126                 *type = IPVL_ARP;
 127                 lyr3h = arph;
 128                 break;
 129         }
 130         case htons(ETH_P_IP): {
 131                 u32 pktlen;
 132                 struct iphdr *ip4h;
 133
 134                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
 135                         return NULL;
 136
 137                 ip4h = ip_hdr(skb);
 138                 pktlen = ntohs(ip4h->tot_len);
 139                 if (ip4h->ihl < 5 || ip4h->version != 4)
 140                         return NULL;
 141                 if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
 142                         return NULL;
 143
 144                 *type = IPVL_IPV4;
 145                 lyr3h = ip4h;
 146                 break;
 147         }
 148         case htons(ETH_P_IPV6): {
 149                 struct ipv6hdr *ip6h;
 150
 151                 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
 152                         return NULL;
 153
 154                 ip6h = ipv6_hdr(skb);
 155                 if (ip6h->version != 6)
 156                         return NULL;
 157
 158                 *type = IPVL_IPV6;
 159                 lyr3h = ip6h;
 160                 /* Only Neighbour Solicitation pkts need different treatment */
 161                 if (ipv6_addr_any(&ip6h->saddr) &&
 162                     ip6h->nexthdr == NEXTHDR_ICMP) {
 163                         *type = IPVL_ICMPV6;
 164                         lyr3h = ip6h + 1;
 165                 }
 166                 break;
 167         }
 168         default:
 169                 return NULL;
 170         }
 171
 172         return lyr3h;
 173 }
 174
 175 unsigned int ipvlan_mac_hash(const unsigned char *addr)
 176 {
 177         u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
 178                                ipvlan_jhash_secret);
 179
 180         return hash & IPVLAN_MAC_FILTER_MASK;
 181 }
 182
 183 static void ipvlan_multicast_frame(struct ipvl_port *port, struct sk_buff *skb,
 184                                    const struct ipvl_dev *in_dev, bool local)
 185 {
 186         struct ethhdr *eth = eth_hdr(skb);
 187         struct ipvl_dev *ipvlan;
 188         struct sk_buff *nskb;
 189         unsigned int len;
 190         unsigned int mac_hash;
 191         int ret;
 192
 193         if (skb->protocol == htons(ETH_P_PAUSE))
 194                 return;
 195
 196         rcu_read_lock();
 197         list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
 198                 if (local && (ipvlan == in_dev))
 199                         continue;
 200
 201                 mac_hash = ipvlan_mac_hash(eth->h_dest);
 202                 if (!test_bit(mac_hash, ipvlan->mac_filters))
 203                         continue;
 204
 205                 ret = NET_RX_DROP;
 206                 len = skb->len + ETH_HLEN;
 207                 nskb = skb_clone(skb, GFP_ATOMIC);
 208                 if (!nskb)
 209                         goto mcast_acct;
 210
 211                 if (ether_addr_equal(eth->h_dest, ipvlan->phy_dev->broadcast))
 212                         nskb->pkt_type = PACKET_BROADCAST;
 213                 else
 214                         nskb->pkt_type = PACKET_MULTICAST;
 215
 216                 nskb->dev = ipvlan->dev;
 217                 if (local)
 218                         ret = dev_forward_skb(ipvlan->dev, nskb);
 219                 else
 220                         ret = netif_rx(nskb);
 221 mcast_acct:
 222                 ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
 223         }
 224         rcu_read_unlock();
 225
 226         /* Locally generated? ...Forward a copy to the main-device as
 227          * well. On the RX side we'll ignore it (wont give it to any
 228          * of the virtual devices.
 229          */
 230         if (local) {
 231                 nskb = skb_clone(skb, GFP_ATOMIC);
 232                 if (nskb) {
 233                         if (ether_addr_equal(eth->h_dest, port->dev->broadcast))
 234                                 nskb->pkt_type = PACKET_BROADCAST;
 235                         else
 236                                 nskb->pkt_type = PACKET_MULTICAST;
 237
 238                         dev_forward_skb(port->dev, nskb);
 239                 }
 240         }
 241 }
 242
 243 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff *skb,
 244                             bool local)
 245 {
 246         struct ipvl_dev *ipvlan = addr->master;
 247         struct net_device *dev = ipvlan->dev;
 248         unsigned int len;
 249         rx_handler_result_t ret = RX_HANDLER_CONSUMED;
 250         bool success = false;
 251
 252         len = skb->len + ETH_HLEN;
 253         if (unlikely(!(dev->flags & IFF_UP))) {
 254                 kfree_skb(skb);
 255                 goto out;
 256         }
 257
 258         skb = skb_share_check(skb, GFP_ATOMIC);
 259         if (!skb)
 260                 goto out;
 261
 262         skb->dev = dev;
 263         skb->pkt_type = PACKET_HOST;
 264
 265         if (local) {
 266                 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
 267                         success = true;
 268         } else {
 269                 ret = RX_HANDLER_ANOTHER;
 270                 success = true;
 271         }
 272
 273 out:
 274         ipvlan_count_rx(ipvlan, len, success, false);
 275         return ret;
 276 }
 277
 278 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
 279                                             void *lyr3h, int addr_type,
 280                                             bool use_dest)
 281 {
 282         struct ipvl_addr *addr = NULL;
 283
 284         if (addr_type == IPVL_IPV6) {
 285                 struct ipv6hdr *ip6h;
 286                 struct in6_addr *i6addr;
 287
 288                 ip6h = (struct ipv6hdr *)lyr3h;
 289                 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
 290                 addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 291         } else if (addr_type == IPVL_ICMPV6) {
 292                 struct nd_msg *ndmh;
 293                 struct in6_addr *i6addr;
 294
 295                 /* Make sure that the NeighborSolicitation ICMPv6 packets
 296                  * are handled to avoid DAD issue.
 297                  */
 298                 ndmh = (struct nd_msg *)lyr3h;
 299                 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
 300                         i6addr = &ndmh->target;
 301                         addr = ipvlan_ht_addr_lookup(port, i6addr, true);
 302                 }
 303         } else if (addr_type == IPVL_IPV4) {
 304                 struct iphdr *ip4h;
 305                 __be32 *i4addr;
 306
 307                 ip4h = (struct iphdr *)lyr3h;
 308                 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
 309                 addr = ipvlan_ht_addr_lookup(port, i4addr, false);
 310         } else if (addr_type == IPVL_ARP) {
 311                 struct arphdr *arph;
 312                 unsigned char *arp_ptr;
 313                 __be32 dip;
 314
 315                 arph = (struct arphdr *)lyr3h;
 316                 arp_ptr = (unsigned char *)(arph + 1);
 317                 if (use_dest)
 318                         arp_ptr += (2 * port->dev->addr_len) + 4;
 319                 else
 320                         arp_ptr += port->dev->addr_len;
 321
 322                 memcpy(&dip, arp_ptr, 4);
 323                 addr = ipvlan_ht_addr_lookup(port, &dip, false);
 324         }
 325
 326         return addr;
 327 }
 328
 329 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
 330 {
 331         const struct iphdr *ip4h = ip_hdr(skb);
 332         struct net_device *dev = skb->dev;
 333         struct rtable *rt;
 334         int err, ret = NET_XMIT_DROP;
 335         struct flowi4 fl4 = {
 336                 .flowi4_oif = dev->iflink,
 337                 .flowi4_tos = RT_TOS(ip4h->tos),
 338                 .flowi4_flags = FLOWI_FLAG_ANYSRC,
 339                 .daddr = ip4h->daddr,
 340                 .saddr = ip4h->saddr,
 341         };
 342
 343         rt = ip_route_output_flow(dev_net(dev), &fl4, NULL);
 344         if (IS_ERR(rt))
 345                 goto err;
 346
 347         if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
 348                 ip_rt_put(rt);
 349                 goto err;
 350         }
 351         skb_dst_drop(skb);
 352         skb_dst_set(skb, &rt->dst);
 353         err = ip_local_out(skb);
 354         if (unlikely(net_xmit_eval(err)))
 355                 dev->stats.tx_errors++;
 356         else
 357                 ret = NET_XMIT_SUCCESS;
 358         goto out;
 359 err:
 360         dev->stats.tx_errors++;
 361         kfree_skb(skb);
 362 out:
 363         return ret;
 364 }
 365
 366 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
 367 {
 368         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 369         struct net_device *dev = skb->dev;
 370         struct dst_entry *dst;
 371         int err, ret = NET_XMIT_DROP;
 372         struct flowi6 fl6 = {
 373                 .flowi6_iif = skb->dev->ifindex,
 374                 .daddr = ip6h->daddr,
 375                 .saddr = ip6h->saddr,
 376                 .flowi6_flags = FLOWI_FLAG_ANYSRC,
 377                 .flowlabel = ip6_flowinfo(ip6h),
 378                 .flowi6_mark = skb->mark,
 379                 .flowi6_proto = ip6h->nexthdr,
 380         };
 381
 382         dst = ip6_route_output(dev_net(dev), NULL, &fl6);
 383         if (dst->error) {
 384                 ret = dst->error;
 385                 dst_release(dst);
 386                 goto err;
 387         }
 388         skb_dst_drop(skb);
 389         skb_dst_set(skb, dst);
 390         err = ip6_local_out(skb);
 391         if (unlikely(net_xmit_eval(err)))
 392                 dev->stats.tx_errors++;
 393         else
 394                 ret = NET_XMIT_SUCCESS;
 395         goto out;
 396 err:
 397         dev->stats.tx_errors++;
 398         kfree_skb(skb);
 399 out:
 400         return ret;
 401 }
 402
 403 static int ipvlan_process_outbound(struct sk_buff *skb,
 404                                    const struct ipvl_dev *ipvlan)
 405 {
 406         struct ethhdr *ethh = eth_hdr(skb);
 407         int ret = NET_XMIT_DROP;
 408
 409         /* In this mode we dont care about multicast and broadcast traffic */
 410         if (is_multicast_ether_addr(ethh->h_dest)) {
 411                 pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
 412                                     ntohs(skb->protocol));
 413                 kfree_skb(skb);
 414                 goto out;
 415         }
 416
 417         /* The ipvlan is a pseudo-L2 device, so the packets that we receive
 418          * will have L2; which need to discarded and processed further
 419          * in the net-ns of the main-device.
 420          */
 421         if (skb_mac_header_was_set(skb)) {
 422                 skb_pull(skb, sizeof(*ethh));
 423                 skb->mac_header = (typeof(skb->mac_header))~0U;
 424                 skb_reset_network_header(skb);
 425         }
 426
 427         if (skb->protocol == htons(ETH_P_IPV6))
 428                 ret = ipvlan_process_v6_outbound(skb);
 429         else if (skb->protocol == htons(ETH_P_IP))
 430                 ret = ipvlan_process_v4_outbound(skb);
 431         else {
 432                 pr_warn_ratelimited("Dropped outbound packet type=%x\n",
 433                                     ntohs(skb->protocol));
 434                 kfree_skb(skb);
 435         }
 436 out:
 437         return ret;
 438 }
 439
 440 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
 441 {
 442         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 443         void *lyr3h;
 444         struct ipvl_addr *addr;
 445         int addr_type;
 446
 447         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 448         if (!lyr3h)
 449                 goto out;
 450
 451         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 452         if (addr)
 453                 return ipvlan_rcv_frame(addr, skb, true);
 454
 455 out:
 456         skb->dev = ipvlan->phy_dev;
 457         return ipvlan_process_outbound(skb, ipvlan);
 458 }
 459
 460 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
 461 {
 462         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 463         struct ethhdr *eth = eth_hdr(skb);
 464         struct ipvl_addr *addr;
 465         void *lyr3h;
 466         int addr_type;
 467
 468         if (ether_addr_equal(eth->h_dest, eth->h_source)) {
 469                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 470                 if (lyr3h) {
 471                         addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
 472                         if (addr)
 473                                 return ipvlan_rcv_frame(addr, skb, true);
 474                 }
 475                 skb = skb_share_check(skb, GFP_ATOMIC);
 476                 if (!skb)
 477                         return NET_XMIT_DROP;
 478
 479                 /* Packet definitely does not belong to any of the
 480                  * virtual devices, but the dest is local. So forward
 481                  * the skb for the main-dev. At the RX side we just return
 482                  * RX_PASS for it to be processed further on the stack.
 483                  */
 484                 return dev_forward_skb(ipvlan->phy_dev, skb);
 485
 486         } else if (is_multicast_ether_addr(eth->h_dest)) {
 487                 u8 ip_summed = skb->ip_summed;
 488
 489                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 490                 ipvlan_multicast_frame(ipvlan->port, skb, ipvlan, true);
 491                 skb->ip_summed = ip_summed;
 492         }
 493
 494         skb->dev = ipvlan->phy_dev;
 495         return dev_queue_xmit(skb);
 496 }
 497
 498 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 499 {
 500         struct ipvl_dev *ipvlan = netdev_priv(dev);
 501         struct ipvl_port *port = ipvlan_port_get_rcu(ipvlan->phy_dev);
 502
 503         if (!port)
 504                 goto out;
 505
 506         if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
 507                 goto out;
 508
 509         switch(port->mode) {
 510         case IPVLAN_MODE_L2:
 511                 return ipvlan_xmit_mode_l2(skb, dev);
 512         case IPVLAN_MODE_L3:
 513                 return ipvlan_xmit_mode_l3(skb, dev);
 514         }
 515
 516         /* Should not reach here */
 517         WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
 518                           port->mode);
 519 out:
 520         kfree_skb(skb);
 521         return NET_XMIT_DROP;
 522 }
 523
 524 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
 525 {
 526         struct ethhdr *eth = eth_hdr(skb);
 527         struct ipvl_addr *addr;
 528         void *lyr3h;
 529         int addr_type;
 530
 531         if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
 532                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 533                 if (!lyr3h)
 534                         return true;
 535
 536                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
 537                 if (addr)
 538                         return false;
 539         }
 540
 541         return true;
 542 }
 543
 544 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 545                                                  struct ipvl_port *port)
 546 {
 547         void *lyr3h;
 548         int addr_type;
 549         struct ipvl_addr *addr;
 550         struct sk_buff *skb = *pskb;
 551         rx_handler_result_t ret = RX_HANDLER_PASS;
 552
 553         lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 554         if (!lyr3h)
 555                 goto out;
 556
 557         addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 558         if (addr)
 559                 ret = ipvlan_rcv_frame(addr, skb, false);
 560
 561 out:
 562         return ret;
 563 }
 564
 565 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 566                                                  struct ipvl_port *port)
 567 {
 568         struct sk_buff *skb = *pskb;
 569         struct ethhdr *eth = eth_hdr(skb);
 570         rx_handler_result_t ret = RX_HANDLER_PASS;
 571         void *lyr3h;
 572         int addr_type;
 573
 574         if (is_multicast_ether_addr(eth->h_dest)) {
 575                 if (ipvlan_external_frame(skb, port))
 576                         ipvlan_multicast_frame(port, skb, NULL, false);
 577         } else {
 578                 struct ipvl_addr *addr;
 579
 580                 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type);
 581                 if (!lyr3h)
 582                         return ret;
 583
 584                 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
 585                 if (addr)
 586                         ret = ipvlan_rcv_frame(addr, skb, false);
 587         }
 588
 589         return ret;
 590 }
 591
 592 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
 593 {
 594         struct sk_buff *skb = *pskb;
 595         struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
 596
 597         if (!port)
 598                 return RX_HANDLER_PASS;
 599
 600         switch (port->mode) {
 601         case IPVLAN_MODE_L2:
 602                 return ipvlan_handle_mode_l2(pskb, port);
 603         case IPVLAN_MODE_L3:
 604                 return ipvlan_handle_mode_l3(pskb, port);
 605         }
 606
 607         /* Should not reach here */
 608         WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
 609                           port->mode);
 610         kfree_skb(skb);
 611         return NET_RX_DROP;
 612 }