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 int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb, 258 bool local) 259 { 260 struct ipvl_dev *ipvlan = addr->master; 261 struct net_device *dev = ipvlan->dev; 262 unsigned int len; 263 rx_handler_result_t ret = RX_HANDLER_CONSUMED; 264 bool success = false; 265 struct sk_buff *skb = *pskb; 266 267 len = skb->len + ETH_HLEN; 268 /* Only packets exchanged between two local slaves need to have 269 * device-up check as well as skb-share check. 270 */ 271 if (local) { 272 if (unlikely(!(dev->flags & IFF_UP))) { 273 kfree_skb(skb); 274 goto out; 275 } 276 277 skb = skb_share_check(skb, GFP_ATOMIC); 278 if (!skb) 279 goto out; 280 281 *pskb = skb; 282 } 283 skb->dev = dev; 284 285 if (local) { 286 skb->pkt_type = PACKET_HOST; 287 if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS) 288 success = true; 289 } else { 290 ret = RX_HANDLER_ANOTHER; 291 success = true; 292 } 293 294 out: 295 ipvlan_count_rx(ipvlan, len, success, false); 296 return ret; 297 } 298 299 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, 300 void *lyr3h, int addr_type, 301 bool use_dest) 302 { 303 struct ipvl_addr *addr = NULL; 304 305 if (addr_type == IPVL_IPV6) { 306 struct ipv6hdr *ip6h; 307 struct in6_addr *i6addr; 308 309 ip6h = (struct ipv6hdr *)lyr3h; 310 i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr; 311 addr = ipvlan_ht_addr_lookup(port, i6addr, true); 312 } else if (addr_type == IPVL_ICMPV6) { 313 struct nd_msg *ndmh; 314 struct in6_addr *i6addr; 315 316 /* Make sure that the NeighborSolicitation ICMPv6 packets 317 * are handled to avoid DAD issue. 318 */ 319 ndmh = (struct nd_msg *)lyr3h; 320 if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) { 321 i6addr = &ndmh->target; 322 addr = ipvlan_ht_addr_lookup(port, i6addr, true); 323 } 324 } else if (addr_type == IPVL_IPV4) { 325 struct iphdr *ip4h; 326 __be32 *i4addr; 327 328 ip4h = (struct iphdr *)lyr3h; 329 i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr; 330 addr = ipvlan_ht_addr_lookup(port, i4addr, false); 331 } else if (addr_type == IPVL_ARP) { 332 struct arphdr *arph; 333 unsigned char *arp_ptr; 334 __be32 dip; 335 336 arph = (struct arphdr *)lyr3h; 337 arp_ptr = (unsigned char *)(arph + 1); 338 if (use_dest) 339 arp_ptr += (2 * port->dev->addr_len) + 4; 340 else 341 arp_ptr += port->dev->addr_len; 342 343 memcpy(&dip, arp_ptr, 4); 344 addr = ipvlan_ht_addr_lookup(port, &dip, false); 345 } 346 347 return addr; 348 } 349 350 static int ipvlan_process_v4_outbound(struct sk_buff *skb, bool xnet) 351 { 352 const struct iphdr *ip4h = ip_hdr(skb); 353 struct net_device *dev = skb->dev; 354 struct net *net = dev_net(dev); 355 struct rtable *rt; 356 int err, ret = NET_XMIT_DROP; 357 struct flowi4 fl4 = { 358 .flowi4_oif = dev->ifindex, 359 .flowi4_tos = RT_TOS(ip4h->tos), 360 .flowi4_flags = FLOWI_FLAG_ANYSRC, 361 .daddr = ip4h->daddr, 362 .saddr = ip4h->saddr, 363 }; 364 365 rt = ip_route_output_flow(net, &fl4, NULL); 366 if (IS_ERR(rt)) 367 goto err; 368 369 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 370 ip_rt_put(rt); 371 goto err; 372 } 373 skb_scrub_packet(skb, xnet); 374 skb_dst_set(skb, &rt->dst); 375 err = ip_local_out(net, skb->sk, skb); 376 if (unlikely(net_xmit_eval(err))) 377 dev->stats.tx_errors++; 378 else 379 ret = NET_XMIT_SUCCESS; 380 goto out; 381 err: 382 dev->stats.tx_errors++; 383 kfree_skb(skb); 384 out: 385 return ret; 386 } 387 388 static int ipvlan_process_v6_outbound(struct sk_buff *skb, bool xnet) 389 { 390 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 391 struct net_device *dev = skb->dev; 392 struct net *net = dev_net(dev); 393 struct dst_entry *dst; 394 int err, ret = NET_XMIT_DROP; 395 struct flowi6 fl6 = { 396 .flowi6_iif = dev->ifindex, 397 .daddr = ip6h->daddr, 398 .saddr = ip6h->saddr, 399 .flowi6_flags = FLOWI_FLAG_ANYSRC, 400 .flowlabel = ip6_flowinfo(ip6h), 401 .flowi6_mark = skb->mark, 402 .flowi6_proto = ip6h->nexthdr, 403 }; 404 405 dst = ip6_route_output(net, NULL, &fl6); 406 if (dst->error) { 407 ret = dst->error; 408 dst_release(dst); 409 goto err; 410 } 411 skb_scrub_packet(skb, xnet); 412 skb_dst_set(skb, dst); 413 err = ip6_local_out(net, skb->sk, skb); 414 if (unlikely(net_xmit_eval(err))) 415 dev->stats.tx_errors++; 416 else 417 ret = NET_XMIT_SUCCESS; 418 goto out; 419 err: 420 dev->stats.tx_errors++; 421 kfree_skb(skb); 422 out: 423 return ret; 424 } 425 426 static int ipvlan_process_outbound(struct sk_buff *skb, bool xnet) 427 { 428 struct ethhdr *ethh = eth_hdr(skb); 429 int ret = NET_XMIT_DROP; 430 431 /* In this mode we dont care about multicast and broadcast traffic */ 432 if (is_multicast_ether_addr(ethh->h_dest)) { 433 pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n", 434 ntohs(skb->protocol)); 435 kfree_skb(skb); 436 goto out; 437 } 438 439 /* The ipvlan is a pseudo-L2 device, so the packets that we receive 440 * will have L2; which need to discarded and processed further 441 * in the net-ns of the main-device. 442 */ 443 if (skb_mac_header_was_set(skb)) { 444 skb_pull(skb, sizeof(*ethh)); 445 skb->mac_header = (typeof(skb->mac_header))~0U; 446 skb_reset_network_header(skb); 447 } 448 449 if (skb->protocol == htons(ETH_P_IPV6)) 450 ret = ipvlan_process_v6_outbound(skb, xnet); 451 else if (skb->protocol == htons(ETH_P_IP)) 452 ret = ipvlan_process_v4_outbound(skb, xnet); 453 else { 454 pr_warn_ratelimited("Dropped outbound packet type=%x\n", 455 ntohs(skb->protocol)); 456 kfree_skb(skb); 457 } 458 out: 459 return ret; 460 } 461 462 static void ipvlan_multicast_enqueue(struct ipvl_port *port, 463 struct sk_buff *skb) 464 { 465 if (skb->protocol == htons(ETH_P_PAUSE)) { 466 kfree_skb(skb); 467 return; 468 } 469 470 spin_lock(&port->backlog.lock); 471 if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) { 472 __skb_queue_tail(&port->backlog, skb); 473 spin_unlock(&port->backlog.lock); 474 schedule_work(&port->wq); 475 } else { 476 spin_unlock(&port->backlog.lock); 477 atomic_long_inc(&skb->dev->rx_dropped); 478 kfree_skb(skb); 479 } 480 } 481 482 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev) 483 { 484 const struct ipvl_dev *ipvlan = netdev_priv(dev); 485 void *lyr3h; 486 struct ipvl_addr *addr; 487 int addr_type; 488 bool xnet; 489 490 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type); 491 if (!lyr3h) 492 goto out; 493 494 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true); 495 if (addr) 496 return ipvlan_rcv_frame(addr, &skb, true); 497 498 out: 499 xnet = !net_eq(dev_net(skb->dev), dev_net(ipvlan->phy_dev)); 500 skb->dev = ipvlan->phy_dev; 501 return ipvlan_process_outbound(skb, xnet); 502 } 503 504 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev) 505 { 506 const struct ipvl_dev *ipvlan = netdev_priv(dev); 507 struct ethhdr *eth = eth_hdr(skb); 508 struct ipvl_addr *addr; 509 void *lyr3h; 510 int addr_type; 511 512 if (ether_addr_equal(eth->h_dest, eth->h_source)) { 513 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type); 514 if (lyr3h) { 515 addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true); 516 if (addr) 517 return ipvlan_rcv_frame(addr, &skb, true); 518 } 519 skb = skb_share_check(skb, GFP_ATOMIC); 520 if (!skb) 521 return NET_XMIT_DROP; 522 523 /* Packet definitely does not belong to any of the 524 * virtual devices, but the dest is local. So forward 525 * the skb for the main-dev. At the RX side we just return 526 * RX_PASS for it to be processed further on the stack. 527 */ 528 return dev_forward_skb(ipvlan->phy_dev, skb); 529 530 } else if (is_multicast_ether_addr(eth->h_dest)) { 531 ipvlan_multicast_enqueue(ipvlan->port, skb); 532 return NET_XMIT_SUCCESS; 533 } 534 535 skb->dev = ipvlan->phy_dev; 536 return dev_queue_xmit(skb); 537 } 538 539 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev) 540 { 541 struct ipvl_dev *ipvlan = netdev_priv(dev); 542 struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev); 543 544 if (!port) 545 goto out; 546 547 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 548 goto out; 549 550 switch(port->mode) { 551 case IPVLAN_MODE_L2: 552 return ipvlan_xmit_mode_l2(skb, dev); 553 case IPVLAN_MODE_L3: 554 return ipvlan_xmit_mode_l3(skb, dev); 555 } 556 557 /* Should not reach here */ 558 WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n", 559 port->mode); 560 out: 561 kfree_skb(skb); 562 return NET_XMIT_DROP; 563 } 564 565 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port) 566 { 567 struct ethhdr *eth = eth_hdr(skb); 568 struct ipvl_addr *addr; 569 void *lyr3h; 570 int addr_type; 571 572 if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) { 573 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type); 574 if (!lyr3h) 575 return true; 576 577 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false); 578 if (addr) 579 return false; 580 } 581 582 return true; 583 } 584 585 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb, 586 struct ipvl_port *port) 587 { 588 void *lyr3h; 589 int addr_type; 590 struct ipvl_addr *addr; 591 struct sk_buff *skb = *pskb; 592 rx_handler_result_t ret = RX_HANDLER_PASS; 593 594 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type); 595 if (!lyr3h) 596 goto out; 597 598 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true); 599 if (addr) 600 ret = ipvlan_rcv_frame(addr, pskb, false); 601 602 out: 603 return ret; 604 } 605 606 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb, 607 struct ipvl_port *port) 608 { 609 struct sk_buff *skb = *pskb; 610 struct ethhdr *eth = eth_hdr(skb); 611 rx_handler_result_t ret = RX_HANDLER_PASS; 612 void *lyr3h; 613 int addr_type; 614 615 if (is_multicast_ether_addr(eth->h_dest)) { 616 if (ipvlan_external_frame(skb, port)) { 617 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC); 618 619 /* External frames are queued for device local 620 * distribution, but a copy is given to master 621 * straight away to avoid sending duplicates later 622 * when work-queue processes this frame. This is 623 * achieved by returning RX_HANDLER_PASS. 624 */ 625 if (nskb) 626 ipvlan_multicast_enqueue(port, nskb); 627 } 628 } else { 629 struct ipvl_addr *addr; 630 631 lyr3h = ipvlan_get_L3_hdr(skb, &addr_type); 632 if (!lyr3h) 633 return ret; 634 635 addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true); 636 if (addr) 637 ret = ipvlan_rcv_frame(addr, pskb, false); 638 } 639 640 return ret; 641 } 642 643 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb) 644 { 645 struct sk_buff *skb = *pskb; 646 struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev); 647 648 if (!port) 649 return RX_HANDLER_PASS; 650 651 switch (port->mode) { 652 case IPVLAN_MODE_L2: 653 return ipvlan_handle_mode_l2(pskb, port); 654 case IPVLAN_MODE_L3: 655 return ipvlan_handle_mode_l3(pskb, port); 656 } 657 658 /* Should not reach here */ 659 WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n", 660 port->mode); 661 kfree_skb(skb); 662 return RX_HANDLER_CONSUMED; 663 } 664