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