1 /* 2 * vrf.c: device driver to encapsulate a VRF space 3 * 4 * Copyright (c) 2015 Cumulus Networks. All rights reserved. 5 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com> 6 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com> 7 * 8 * Based on dummy, team and ipvlan drivers 9 * 10 * This program is free software; you can redistribute it and/or modify 11 * it under the terms of the GNU General Public License as published by 12 * the Free Software Foundation; either version 2 of the License, or 13 * (at your option) any later version. 14 */ 15 16 #include <linux/module.h> 17 #include <linux/kernel.h> 18 #include <linux/netdevice.h> 19 #include <linux/etherdevice.h> 20 #include <linux/ip.h> 21 #include <linux/init.h> 22 #include <linux/moduleparam.h> 23 #include <linux/netfilter.h> 24 #include <linux/rtnetlink.h> 25 #include <net/rtnetlink.h> 26 #include <linux/u64_stats_sync.h> 27 #include <linux/hashtable.h> 28 29 #include <linux/inetdevice.h> 30 #include <net/arp.h> 31 #include <net/ip.h> 32 #include <net/ip_fib.h> 33 #include <net/ip6_fib.h> 34 #include <net/ip6_route.h> 35 #include <net/route.h> 36 #include <net/addrconf.h> 37 #include <net/l3mdev.h> 38 #include <net/fib_rules.h> 39 40 #define DRV_NAME "vrf" 41 #define DRV_VERSION "1.0" 42 43 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 44 static bool add_fib_rules = true; 45 46 struct net_vrf { 47 struct rtable __rcu *rth; 48 struct rtable __rcu *rth_local; 49 struct rt6_info __rcu *rt6; 50 struct rt6_info __rcu *rt6_local; 51 u32 tb_id; 52 }; 53 54 struct pcpu_dstats { 55 u64 tx_pkts; 56 u64 tx_bytes; 57 u64 tx_drps; 58 u64 rx_pkts; 59 u64 rx_bytes; 60 u64 rx_drps; 61 struct u64_stats_sync syncp; 62 }; 63 64 static void vrf_rx_stats(struct net_device *dev, int len) 65 { 66 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 67 68 u64_stats_update_begin(&dstats->syncp); 69 dstats->rx_pkts++; 70 dstats->rx_bytes += len; 71 u64_stats_update_end(&dstats->syncp); 72 } 73 74 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 75 { 76 vrf_dev->stats.tx_errors++; 77 kfree_skb(skb); 78 } 79 80 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, 81 struct rtnl_link_stats64 *stats) 82 { 83 int i; 84 85 for_each_possible_cpu(i) { 86 const struct pcpu_dstats *dstats; 87 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 88 unsigned int start; 89 90 dstats = per_cpu_ptr(dev->dstats, i); 91 do { 92 start = u64_stats_fetch_begin_irq(&dstats->syncp); 93 tbytes = dstats->tx_bytes; 94 tpkts = dstats->tx_pkts; 95 tdrops = dstats->tx_drps; 96 rbytes = dstats->rx_bytes; 97 rpkts = dstats->rx_pkts; 98 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 99 stats->tx_bytes += tbytes; 100 stats->tx_packets += tpkts; 101 stats->tx_dropped += tdrops; 102 stats->rx_bytes += rbytes; 103 stats->rx_packets += rpkts; 104 } 105 return stats; 106 } 107 108 /* Local traffic destined to local address. Reinsert the packet to rx 109 * path, similar to loopback handling. 110 */ 111 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 112 struct dst_entry *dst) 113 { 114 int len = skb->len; 115 116 skb_orphan(skb); 117 118 skb_dst_set(skb, dst); 119 skb_dst_force(skb); 120 121 /* set pkt_type to avoid skb hitting packet taps twice - 122 * once on Tx and again in Rx processing 123 */ 124 skb->pkt_type = PACKET_LOOPBACK; 125 126 skb->protocol = eth_type_trans(skb, dev); 127 128 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 129 vrf_rx_stats(dev, len); 130 else 131 this_cpu_inc(dev->dstats->rx_drps); 132 133 return NETDEV_TX_OK; 134 } 135 136 #if IS_ENABLED(CONFIG_IPV6) 137 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 138 struct sk_buff *skb) 139 { 140 int err; 141 142 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 143 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 144 145 if (likely(err == 1)) 146 err = dst_output(net, sk, skb); 147 148 return err; 149 } 150 151 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 152 struct net_device *dev) 153 { 154 const struct ipv6hdr *iph = ipv6_hdr(skb); 155 struct net *net = dev_net(skb->dev); 156 struct flowi6 fl6 = { 157 /* needed to match OIF rule */ 158 .flowi6_oif = dev->ifindex, 159 .flowi6_iif = LOOPBACK_IFINDEX, 160 .daddr = iph->daddr, 161 .saddr = iph->saddr, 162 .flowlabel = ip6_flowinfo(iph), 163 .flowi6_mark = skb->mark, 164 .flowi6_proto = iph->nexthdr, 165 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF, 166 }; 167 int ret = NET_XMIT_DROP; 168 struct dst_entry *dst; 169 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 170 171 dst = ip6_route_output(net, NULL, &fl6); 172 if (dst == dst_null) 173 goto err; 174 175 skb_dst_drop(skb); 176 177 /* if dst.dev is loopback or the VRF device again this is locally 178 * originated traffic destined to a local address. Short circuit 179 * to Rx path using our local dst 180 */ 181 if (dst->dev == net->loopback_dev || dst->dev == dev) { 182 struct net_vrf *vrf = netdev_priv(dev); 183 struct rt6_info *rt6_local; 184 185 /* release looked up dst and use cached local dst */ 186 dst_release(dst); 187 188 rcu_read_lock(); 189 190 rt6_local = rcu_dereference(vrf->rt6_local); 191 if (unlikely(!rt6_local)) { 192 rcu_read_unlock(); 193 goto err; 194 } 195 196 /* Ordering issue: cached local dst is created on newlink 197 * before the IPv6 initialization. Using the local dst 198 * requires rt6i_idev to be set so make sure it is. 199 */ 200 if (unlikely(!rt6_local->rt6i_idev)) { 201 rt6_local->rt6i_idev = in6_dev_get(dev); 202 if (!rt6_local->rt6i_idev) { 203 rcu_read_unlock(); 204 goto err; 205 } 206 } 207 208 dst = &rt6_local->dst; 209 dst_hold(dst); 210 211 rcu_read_unlock(); 212 213 return vrf_local_xmit(skb, dev, &rt6_local->dst); 214 } 215 216 skb_dst_set(skb, dst); 217 218 /* strip the ethernet header added for pass through VRF device */ 219 __skb_pull(skb, skb_network_offset(skb)); 220 221 ret = vrf_ip6_local_out(net, skb->sk, skb); 222 if (unlikely(net_xmit_eval(ret))) 223 dev->stats.tx_errors++; 224 else 225 ret = NET_XMIT_SUCCESS; 226 227 return ret; 228 err: 229 vrf_tx_error(dev, skb); 230 return NET_XMIT_DROP; 231 } 232 #else 233 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 234 struct net_device *dev) 235 { 236 vrf_tx_error(dev, skb); 237 return NET_XMIT_DROP; 238 } 239 #endif 240 241 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 242 static int vrf_ip_local_out(struct net *net, struct sock *sk, 243 struct sk_buff *skb) 244 { 245 int err; 246 247 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 248 skb, NULL, skb_dst(skb)->dev, dst_output); 249 if (likely(err == 1)) 250 err = dst_output(net, sk, skb); 251 252 return err; 253 } 254 255 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 256 struct net_device *vrf_dev) 257 { 258 struct iphdr *ip4h = ip_hdr(skb); 259 int ret = NET_XMIT_DROP; 260 struct flowi4 fl4 = { 261 /* needed to match OIF rule */ 262 .flowi4_oif = vrf_dev->ifindex, 263 .flowi4_iif = LOOPBACK_IFINDEX, 264 .flowi4_tos = RT_TOS(ip4h->tos), 265 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF, 266 .daddr = ip4h->daddr, 267 }; 268 struct net *net = dev_net(vrf_dev); 269 struct rtable *rt; 270 271 rt = ip_route_output_flow(net, &fl4, NULL); 272 if (IS_ERR(rt)) 273 goto err; 274 275 skb_dst_drop(skb); 276 277 /* if dst.dev is loopback or the VRF device again this is locally 278 * originated traffic destined to a local address. Short circuit 279 * to Rx path using our local dst 280 */ 281 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 282 struct net_vrf *vrf = netdev_priv(vrf_dev); 283 struct rtable *rth_local; 284 struct dst_entry *dst = NULL; 285 286 ip_rt_put(rt); 287 288 rcu_read_lock(); 289 290 rth_local = rcu_dereference(vrf->rth_local); 291 if (likely(rth_local)) { 292 dst = &rth_local->dst; 293 dst_hold(dst); 294 } 295 296 rcu_read_unlock(); 297 298 if (unlikely(!dst)) 299 goto err; 300 301 return vrf_local_xmit(skb, vrf_dev, dst); 302 } 303 304 skb_dst_set(skb, &rt->dst); 305 306 /* strip the ethernet header added for pass through VRF device */ 307 __skb_pull(skb, skb_network_offset(skb)); 308 309 if (!ip4h->saddr) { 310 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 311 RT_SCOPE_LINK); 312 } 313 314 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 315 if (unlikely(net_xmit_eval(ret))) 316 vrf_dev->stats.tx_errors++; 317 else 318 ret = NET_XMIT_SUCCESS; 319 320 out: 321 return ret; 322 err: 323 vrf_tx_error(vrf_dev, skb); 324 goto out; 325 } 326 327 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 328 { 329 switch (skb->protocol) { 330 case htons(ETH_P_IP): 331 return vrf_process_v4_outbound(skb, dev); 332 case htons(ETH_P_IPV6): 333 return vrf_process_v6_outbound(skb, dev); 334 default: 335 vrf_tx_error(dev, skb); 336 return NET_XMIT_DROP; 337 } 338 } 339 340 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 341 { 342 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 343 344 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 345 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 346 347 u64_stats_update_begin(&dstats->syncp); 348 dstats->tx_pkts++; 349 dstats->tx_bytes += skb->len; 350 u64_stats_update_end(&dstats->syncp); 351 } else { 352 this_cpu_inc(dev->dstats->tx_drps); 353 } 354 355 return ret; 356 } 357 358 #if IS_ENABLED(CONFIG_IPV6) 359 /* modelled after ip6_finish_output2 */ 360 static int vrf_finish_output6(struct net *net, struct sock *sk, 361 struct sk_buff *skb) 362 { 363 struct dst_entry *dst = skb_dst(skb); 364 struct net_device *dev = dst->dev; 365 struct neighbour *neigh; 366 struct in6_addr *nexthop; 367 int ret; 368 369 skb->protocol = htons(ETH_P_IPV6); 370 skb->dev = dev; 371 372 rcu_read_lock_bh(); 373 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 374 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 375 if (unlikely(!neigh)) 376 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 377 if (!IS_ERR(neigh)) { 378 ret = dst_neigh_output(dst, neigh, skb); 379 rcu_read_unlock_bh(); 380 return ret; 381 } 382 rcu_read_unlock_bh(); 383 384 IP6_INC_STATS(dev_net(dst->dev), 385 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 386 kfree_skb(skb); 387 return -EINVAL; 388 } 389 390 /* modelled after ip6_output */ 391 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 392 { 393 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 394 net, sk, skb, NULL, skb_dst(skb)->dev, 395 vrf_finish_output6, 396 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 397 } 398 399 /* set dst on skb to send packet to us via dev_xmit path. Allows 400 * packet to go through device based features such as qdisc, netfilter 401 * hooks and packet sockets with skb->dev set to vrf device. 402 */ 403 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 404 struct sock *sk, 405 struct sk_buff *skb) 406 { 407 struct net_vrf *vrf = netdev_priv(vrf_dev); 408 struct dst_entry *dst = NULL; 409 struct rt6_info *rt6; 410 411 /* don't divert link scope packets */ 412 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 413 return skb; 414 415 rcu_read_lock(); 416 417 rt6 = rcu_dereference(vrf->rt6); 418 if (likely(rt6)) { 419 dst = &rt6->dst; 420 dst_hold(dst); 421 } 422 423 rcu_read_unlock(); 424 425 if (unlikely(!dst)) { 426 vrf_tx_error(vrf_dev, skb); 427 return NULL; 428 } 429 430 skb_dst_drop(skb); 431 skb_dst_set(skb, dst); 432 433 return skb; 434 } 435 436 /* holding rtnl */ 437 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 438 { 439 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 440 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 441 struct net *net = dev_net(dev); 442 struct dst_entry *dst; 443 444 RCU_INIT_POINTER(vrf->rt6, NULL); 445 RCU_INIT_POINTER(vrf->rt6_local, NULL); 446 synchronize_rcu(); 447 448 /* move dev in dst's to loopback so this VRF device can be deleted 449 * - based on dst_ifdown 450 */ 451 if (rt6) { 452 dst = &rt6->dst; 453 dev_put(dst->dev); 454 dst->dev = net->loopback_dev; 455 dev_hold(dst->dev); 456 dst_release(dst); 457 } 458 459 if (rt6_local) { 460 if (rt6_local->rt6i_idev) 461 in6_dev_put(rt6_local->rt6i_idev); 462 463 dst = &rt6_local->dst; 464 dev_put(dst->dev); 465 dst->dev = net->loopback_dev; 466 dev_hold(dst->dev); 467 dst_release(dst); 468 } 469 } 470 471 static int vrf_rt6_create(struct net_device *dev) 472 { 473 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 474 struct net_vrf *vrf = netdev_priv(dev); 475 struct net *net = dev_net(dev); 476 struct fib6_table *rt6i_table; 477 struct rt6_info *rt6, *rt6_local; 478 int rc = -ENOMEM; 479 480 /* IPv6 can be CONFIG enabled and then disabled runtime */ 481 if (!ipv6_mod_enabled()) 482 return 0; 483 484 rt6i_table = fib6_new_table(net, vrf->tb_id); 485 if (!rt6i_table) 486 goto out; 487 488 /* create a dst for routing packets out a VRF device */ 489 rt6 = ip6_dst_alloc(net, dev, flags); 490 if (!rt6) 491 goto out; 492 493 dst_hold(&rt6->dst); 494 495 rt6->rt6i_table = rt6i_table; 496 rt6->dst.output = vrf_output6; 497 498 /* create a dst for local routing - packets sent locally 499 * to local address via the VRF device as a loopback 500 */ 501 rt6_local = ip6_dst_alloc(net, dev, flags); 502 if (!rt6_local) { 503 dst_release(&rt6->dst); 504 goto out; 505 } 506 507 dst_hold(&rt6_local->dst); 508 509 rt6_local->rt6i_idev = in6_dev_get(dev); 510 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 511 rt6_local->rt6i_table = rt6i_table; 512 rt6_local->dst.input = ip6_input; 513 514 rcu_assign_pointer(vrf->rt6, rt6); 515 rcu_assign_pointer(vrf->rt6_local, rt6_local); 516 517 rc = 0; 518 out: 519 return rc; 520 } 521 #else 522 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 523 struct sock *sk, 524 struct sk_buff *skb) 525 { 526 return skb; 527 } 528 529 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 530 { 531 } 532 533 static int vrf_rt6_create(struct net_device *dev) 534 { 535 return 0; 536 } 537 #endif 538 539 /* modelled after ip_finish_output2 */ 540 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 541 { 542 struct dst_entry *dst = skb_dst(skb); 543 struct rtable *rt = (struct rtable *)dst; 544 struct net_device *dev = dst->dev; 545 unsigned int hh_len = LL_RESERVED_SPACE(dev); 546 struct neighbour *neigh; 547 u32 nexthop; 548 int ret = -EINVAL; 549 550 /* Be paranoid, rather than too clever. */ 551 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 552 struct sk_buff *skb2; 553 554 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 555 if (!skb2) { 556 ret = -ENOMEM; 557 goto err; 558 } 559 if (skb->sk) 560 skb_set_owner_w(skb2, skb->sk); 561 562 consume_skb(skb); 563 skb = skb2; 564 } 565 566 rcu_read_lock_bh(); 567 568 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 569 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 570 if (unlikely(!neigh)) 571 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 572 if (!IS_ERR(neigh)) 573 ret = dst_neigh_output(dst, neigh, skb); 574 575 rcu_read_unlock_bh(); 576 err: 577 if (unlikely(ret < 0)) 578 vrf_tx_error(skb->dev, skb); 579 return ret; 580 } 581 582 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 583 { 584 struct net_device *dev = skb_dst(skb)->dev; 585 586 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 587 588 skb->dev = dev; 589 skb->protocol = htons(ETH_P_IP); 590 591 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 592 net, sk, skb, NULL, dev, 593 vrf_finish_output, 594 !(IPCB(skb)->flags & IPSKB_REROUTED)); 595 } 596 597 /* set dst on skb to send packet to us via dev_xmit path. Allows 598 * packet to go through device based features such as qdisc, netfilter 599 * hooks and packet sockets with skb->dev set to vrf device. 600 */ 601 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 602 struct sock *sk, 603 struct sk_buff *skb) 604 { 605 struct net_vrf *vrf = netdev_priv(vrf_dev); 606 struct dst_entry *dst = NULL; 607 struct rtable *rth; 608 609 /* don't divert multicast */ 610 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 611 return skb; 612 613 rcu_read_lock(); 614 615 rth = rcu_dereference(vrf->rth); 616 if (likely(rth)) { 617 dst = &rth->dst; 618 dst_hold(dst); 619 } 620 621 rcu_read_unlock(); 622 623 if (unlikely(!dst)) { 624 vrf_tx_error(vrf_dev, skb); 625 return NULL; 626 } 627 628 skb_dst_drop(skb); 629 skb_dst_set(skb, dst); 630 631 return skb; 632 } 633 634 /* called with rcu lock held */ 635 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 636 struct sock *sk, 637 struct sk_buff *skb, 638 u16 proto) 639 { 640 switch (proto) { 641 case AF_INET: 642 return vrf_ip_out(vrf_dev, sk, skb); 643 case AF_INET6: 644 return vrf_ip6_out(vrf_dev, sk, skb); 645 } 646 647 return skb; 648 } 649 650 /* holding rtnl */ 651 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 652 { 653 struct rtable *rth = rtnl_dereference(vrf->rth); 654 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 655 struct net *net = dev_net(dev); 656 struct dst_entry *dst; 657 658 RCU_INIT_POINTER(vrf->rth, NULL); 659 RCU_INIT_POINTER(vrf->rth_local, NULL); 660 synchronize_rcu(); 661 662 /* move dev in dst's to loopback so this VRF device can be deleted 663 * - based on dst_ifdown 664 */ 665 if (rth) { 666 dst = &rth->dst; 667 dev_put(dst->dev); 668 dst->dev = net->loopback_dev; 669 dev_hold(dst->dev); 670 dst_release(dst); 671 } 672 673 if (rth_local) { 674 dst = &rth_local->dst; 675 dev_put(dst->dev); 676 dst->dev = net->loopback_dev; 677 dev_hold(dst->dev); 678 dst_release(dst); 679 } 680 } 681 682 static int vrf_rtable_create(struct net_device *dev) 683 { 684 struct net_vrf *vrf = netdev_priv(dev); 685 struct rtable *rth, *rth_local; 686 687 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 688 return -ENOMEM; 689 690 /* create a dst for routing packets out through a VRF device */ 691 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 692 if (!rth) 693 return -ENOMEM; 694 695 /* create a dst for local ingress routing - packets sent locally 696 * to local address via the VRF device as a loopback 697 */ 698 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 699 if (!rth_local) { 700 dst_release(&rth->dst); 701 return -ENOMEM; 702 } 703 704 rth->dst.output = vrf_output; 705 rth->rt_table_id = vrf->tb_id; 706 707 rth_local->rt_table_id = vrf->tb_id; 708 709 rcu_assign_pointer(vrf->rth, rth); 710 rcu_assign_pointer(vrf->rth_local, rth_local); 711 712 return 0; 713 } 714 715 /**************************** device handling ********************/ 716 717 /* cycle interface to flush neighbor cache and move routes across tables */ 718 static void cycle_netdev(struct net_device *dev) 719 { 720 unsigned int flags = dev->flags; 721 int ret; 722 723 if (!netif_running(dev)) 724 return; 725 726 ret = dev_change_flags(dev, flags & ~IFF_UP); 727 if (ret >= 0) 728 ret = dev_change_flags(dev, flags); 729 730 if (ret < 0) { 731 netdev_err(dev, 732 "Failed to cycle device %s; route tables might be wrong!\n", 733 dev->name); 734 } 735 } 736 737 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 738 { 739 int ret; 740 741 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 742 if (ret < 0) 743 return ret; 744 745 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 746 cycle_netdev(port_dev); 747 748 return 0; 749 } 750 751 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 752 { 753 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 754 return -EINVAL; 755 756 return do_vrf_add_slave(dev, port_dev); 757 } 758 759 /* inverse of do_vrf_add_slave */ 760 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 761 { 762 netdev_upper_dev_unlink(port_dev, dev); 763 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 764 765 cycle_netdev(port_dev); 766 767 return 0; 768 } 769 770 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 771 { 772 return do_vrf_del_slave(dev, port_dev); 773 } 774 775 static void vrf_dev_uninit(struct net_device *dev) 776 { 777 struct net_vrf *vrf = netdev_priv(dev); 778 struct net_device *port_dev; 779 struct list_head *iter; 780 781 vrf_rtable_release(dev, vrf); 782 vrf_rt6_release(dev, vrf); 783 784 netdev_for_each_lower_dev(dev, port_dev, iter) 785 vrf_del_slave(dev, port_dev); 786 787 free_percpu(dev->dstats); 788 dev->dstats = NULL; 789 } 790 791 static int vrf_dev_init(struct net_device *dev) 792 { 793 struct net_vrf *vrf = netdev_priv(dev); 794 795 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 796 if (!dev->dstats) 797 goto out_nomem; 798 799 /* create the default dst which points back to us */ 800 if (vrf_rtable_create(dev) != 0) 801 goto out_stats; 802 803 if (vrf_rt6_create(dev) != 0) 804 goto out_rth; 805 806 dev->flags = IFF_MASTER | IFF_NOARP; 807 808 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 809 dev->mtu = 64 * 1024; 810 811 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 812 dev->operstate = IF_OPER_UP; 813 netdev_lockdep_set_classes(dev); 814 return 0; 815 816 out_rth: 817 vrf_rtable_release(dev, vrf); 818 out_stats: 819 free_percpu(dev->dstats); 820 dev->dstats = NULL; 821 out_nomem: 822 return -ENOMEM; 823 } 824 825 static const struct net_device_ops vrf_netdev_ops = { 826 .ndo_init = vrf_dev_init, 827 .ndo_uninit = vrf_dev_uninit, 828 .ndo_start_xmit = vrf_xmit, 829 .ndo_get_stats64 = vrf_get_stats64, 830 .ndo_add_slave = vrf_add_slave, 831 .ndo_del_slave = vrf_del_slave, 832 }; 833 834 static u32 vrf_fib_table(const struct net_device *dev) 835 { 836 struct net_vrf *vrf = netdev_priv(dev); 837 838 return vrf->tb_id; 839 } 840 841 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 842 { 843 return 0; 844 } 845 846 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 847 struct sk_buff *skb, 848 struct net_device *dev) 849 { 850 struct net *net = dev_net(dev); 851 852 nf_reset(skb); 853 854 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 855 skb = NULL; /* kfree_skb(skb) handled by nf code */ 856 857 return skb; 858 } 859 860 #if IS_ENABLED(CONFIG_IPV6) 861 /* neighbor handling is done with actual device; do not want 862 * to flip skb->dev for those ndisc packets. This really fails 863 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 864 * a start. 865 */ 866 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 867 { 868 const struct ipv6hdr *iph = ipv6_hdr(skb); 869 bool rc = false; 870 871 if (iph->nexthdr == NEXTHDR_ICMP) { 872 const struct icmp6hdr *icmph; 873 struct icmp6hdr _icmph; 874 875 icmph = skb_header_pointer(skb, sizeof(*iph), 876 sizeof(_icmph), &_icmph); 877 if (!icmph) 878 goto out; 879 880 switch (icmph->icmp6_type) { 881 case NDISC_ROUTER_SOLICITATION: 882 case NDISC_ROUTER_ADVERTISEMENT: 883 case NDISC_NEIGHBOUR_SOLICITATION: 884 case NDISC_NEIGHBOUR_ADVERTISEMENT: 885 case NDISC_REDIRECT: 886 rc = true; 887 break; 888 } 889 } 890 891 out: 892 return rc; 893 } 894 895 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 896 const struct net_device *dev, 897 struct flowi6 *fl6, 898 int ifindex, 899 int flags) 900 { 901 struct net_vrf *vrf = netdev_priv(dev); 902 struct fib6_table *table = NULL; 903 struct rt6_info *rt6; 904 905 rcu_read_lock(); 906 907 /* fib6_table does not have a refcnt and can not be freed */ 908 rt6 = rcu_dereference(vrf->rt6); 909 if (likely(rt6)) 910 table = rt6->rt6i_table; 911 912 rcu_read_unlock(); 913 914 if (!table) 915 return NULL; 916 917 return ip6_pol_route(net, table, ifindex, fl6, flags); 918 } 919 920 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 921 int ifindex) 922 { 923 const struct ipv6hdr *iph = ipv6_hdr(skb); 924 struct flowi6 fl6 = { 925 .daddr = iph->daddr, 926 .saddr = iph->saddr, 927 .flowlabel = ip6_flowinfo(iph), 928 .flowi6_mark = skb->mark, 929 .flowi6_proto = iph->nexthdr, 930 .flowi6_iif = ifindex, 931 }; 932 struct net *net = dev_net(vrf_dev); 933 struct rt6_info *rt6; 934 935 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 936 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 937 if (unlikely(!rt6)) 938 return; 939 940 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 941 return; 942 943 skb_dst_set(skb, &rt6->dst); 944 } 945 946 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 947 struct sk_buff *skb) 948 { 949 int orig_iif = skb->skb_iif; 950 bool need_strict; 951 952 /* loopback traffic; do not push through packet taps again. 953 * Reset pkt_type for upper layers to process skb 954 */ 955 if (skb->pkt_type == PACKET_LOOPBACK) { 956 skb->dev = vrf_dev; 957 skb->skb_iif = vrf_dev->ifindex; 958 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 959 skb->pkt_type = PACKET_HOST; 960 goto out; 961 } 962 963 /* if packet is NDISC or addressed to multicast or link-local 964 * then keep the ingress interface 965 */ 966 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 967 if (!ipv6_ndisc_frame(skb) && !need_strict) { 968 skb->dev = vrf_dev; 969 skb->skb_iif = vrf_dev->ifindex; 970 971 skb_push(skb, skb->mac_len); 972 dev_queue_xmit_nit(skb, vrf_dev); 973 skb_pull(skb, skb->mac_len); 974 975 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 976 } 977 978 if (need_strict) 979 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 980 981 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 982 out: 983 return skb; 984 } 985 986 #else 987 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 988 struct sk_buff *skb) 989 { 990 return skb; 991 } 992 #endif 993 994 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 995 struct sk_buff *skb) 996 { 997 skb->dev = vrf_dev; 998 skb->skb_iif = vrf_dev->ifindex; 999 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1000 1001 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1002 goto out; 1003 1004 /* loopback traffic; do not push through packet taps again. 1005 * Reset pkt_type for upper layers to process skb 1006 */ 1007 if (skb->pkt_type == PACKET_LOOPBACK) { 1008 skb->pkt_type = PACKET_HOST; 1009 goto out; 1010 } 1011 1012 skb_push(skb, skb->mac_len); 1013 dev_queue_xmit_nit(skb, vrf_dev); 1014 skb_pull(skb, skb->mac_len); 1015 1016 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1017 out: 1018 return skb; 1019 } 1020 1021 /* called with rcu lock held */ 1022 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1023 struct sk_buff *skb, 1024 u16 proto) 1025 { 1026 switch (proto) { 1027 case AF_INET: 1028 return vrf_ip_rcv(vrf_dev, skb); 1029 case AF_INET6: 1030 return vrf_ip6_rcv(vrf_dev, skb); 1031 } 1032 1033 return skb; 1034 } 1035 1036 #if IS_ENABLED(CONFIG_IPV6) 1037 /* send to link-local or multicast address via interface enslaved to 1038 * VRF device. Force lookup to VRF table without changing flow struct 1039 */ 1040 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1041 struct flowi6 *fl6) 1042 { 1043 struct net *net = dev_net(dev); 1044 int flags = RT6_LOOKUP_F_IFACE; 1045 struct dst_entry *dst = NULL; 1046 struct rt6_info *rt; 1047 1048 /* VRF device does not have a link-local address and 1049 * sending packets to link-local or mcast addresses over 1050 * a VRF device does not make sense 1051 */ 1052 if (fl6->flowi6_oif == dev->ifindex) { 1053 dst = &net->ipv6.ip6_null_entry->dst; 1054 dst_hold(dst); 1055 return dst; 1056 } 1057 1058 if (!ipv6_addr_any(&fl6->saddr)) 1059 flags |= RT6_LOOKUP_F_HAS_SADDR; 1060 1061 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1062 if (rt) 1063 dst = &rt->dst; 1064 1065 return dst; 1066 } 1067 #endif 1068 1069 static const struct l3mdev_ops vrf_l3mdev_ops = { 1070 .l3mdev_fib_table = vrf_fib_table, 1071 .l3mdev_l3_rcv = vrf_l3_rcv, 1072 .l3mdev_l3_out = vrf_l3_out, 1073 #if IS_ENABLED(CONFIG_IPV6) 1074 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1075 #endif 1076 }; 1077 1078 static void vrf_get_drvinfo(struct net_device *dev, 1079 struct ethtool_drvinfo *info) 1080 { 1081 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1082 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1083 } 1084 1085 static const struct ethtool_ops vrf_ethtool_ops = { 1086 .get_drvinfo = vrf_get_drvinfo, 1087 }; 1088 1089 static inline size_t vrf_fib_rule_nl_size(void) 1090 { 1091 size_t sz; 1092 1093 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1094 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1095 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1096 1097 return sz; 1098 } 1099 1100 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1101 { 1102 struct fib_rule_hdr *frh; 1103 struct nlmsghdr *nlh; 1104 struct sk_buff *skb; 1105 int err; 1106 1107 if (family == AF_INET6 && !ipv6_mod_enabled()) 1108 return 0; 1109 1110 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1111 if (!skb) 1112 return -ENOMEM; 1113 1114 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1115 if (!nlh) 1116 goto nla_put_failure; 1117 1118 /* rule only needs to appear once */ 1119 nlh->nlmsg_flags &= NLM_F_EXCL; 1120 1121 frh = nlmsg_data(nlh); 1122 memset(frh, 0, sizeof(*frh)); 1123 frh->family = family; 1124 frh->action = FR_ACT_TO_TBL; 1125 1126 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1127 goto nla_put_failure; 1128 1129 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1130 goto nla_put_failure; 1131 1132 nlmsg_end(skb, nlh); 1133 1134 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1135 skb->sk = dev_net(dev)->rtnl; 1136 if (add_it) { 1137 err = fib_nl_newrule(skb, nlh); 1138 if (err == -EEXIST) 1139 err = 0; 1140 } else { 1141 err = fib_nl_delrule(skb, nlh); 1142 if (err == -ENOENT) 1143 err = 0; 1144 } 1145 nlmsg_free(skb); 1146 1147 return err; 1148 1149 nla_put_failure: 1150 nlmsg_free(skb); 1151 1152 return -EMSGSIZE; 1153 } 1154 1155 static int vrf_add_fib_rules(const struct net_device *dev) 1156 { 1157 int err; 1158 1159 err = vrf_fib_rule(dev, AF_INET, true); 1160 if (err < 0) 1161 goto out_err; 1162 1163 err = vrf_fib_rule(dev, AF_INET6, true); 1164 if (err < 0) 1165 goto ipv6_err; 1166 1167 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1168 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1169 if (err < 0) 1170 goto ipmr_err; 1171 #endif 1172 1173 return 0; 1174 1175 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1176 ipmr_err: 1177 vrf_fib_rule(dev, AF_INET6, false); 1178 #endif 1179 1180 ipv6_err: 1181 vrf_fib_rule(dev, AF_INET, false); 1182 1183 out_err: 1184 netdev_err(dev, "Failed to add FIB rules.\n"); 1185 return err; 1186 } 1187 1188 static void vrf_setup(struct net_device *dev) 1189 { 1190 ether_setup(dev); 1191 1192 /* Initialize the device structure. */ 1193 dev->netdev_ops = &vrf_netdev_ops; 1194 dev->l3mdev_ops = &vrf_l3mdev_ops; 1195 dev->ethtool_ops = &vrf_ethtool_ops; 1196 dev->destructor = free_netdev; 1197 1198 /* Fill in device structure with ethernet-generic values. */ 1199 eth_hw_addr_random(dev); 1200 1201 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1202 dev->features |= NETIF_F_LLTX; 1203 1204 /* don't allow vrf devices to change network namespaces. */ 1205 dev->features |= NETIF_F_NETNS_LOCAL; 1206 1207 /* does not make sense for a VLAN to be added to a vrf device */ 1208 dev->features |= NETIF_F_VLAN_CHALLENGED; 1209 1210 /* enable offload features */ 1211 dev->features |= NETIF_F_GSO_SOFTWARE; 1212 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1213 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1214 1215 dev->hw_features = dev->features; 1216 dev->hw_enc_features = dev->features; 1217 1218 /* default to no qdisc; user can add if desired */ 1219 dev->priv_flags |= IFF_NO_QUEUE; 1220 } 1221 1222 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1223 { 1224 if (tb[IFLA_ADDRESS]) { 1225 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1226 return -EINVAL; 1227 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1228 return -EADDRNOTAVAIL; 1229 } 1230 return 0; 1231 } 1232 1233 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1234 { 1235 unregister_netdevice_queue(dev, head); 1236 } 1237 1238 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1239 struct nlattr *tb[], struct nlattr *data[]) 1240 { 1241 struct net_vrf *vrf = netdev_priv(dev); 1242 int err; 1243 1244 if (!data || !data[IFLA_VRF_TABLE]) 1245 return -EINVAL; 1246 1247 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1248 1249 dev->priv_flags |= IFF_L3MDEV_MASTER; 1250 1251 err = register_netdevice(dev); 1252 if (err) 1253 goto out; 1254 1255 if (add_fib_rules) { 1256 err = vrf_add_fib_rules(dev); 1257 if (err) { 1258 unregister_netdevice(dev); 1259 goto out; 1260 } 1261 add_fib_rules = false; 1262 } 1263 1264 out: 1265 return err; 1266 } 1267 1268 static size_t vrf_nl_getsize(const struct net_device *dev) 1269 { 1270 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1271 } 1272 1273 static int vrf_fillinfo(struct sk_buff *skb, 1274 const struct net_device *dev) 1275 { 1276 struct net_vrf *vrf = netdev_priv(dev); 1277 1278 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1279 } 1280 1281 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1282 const struct net_device *slave_dev) 1283 { 1284 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1285 } 1286 1287 static int vrf_fill_slave_info(struct sk_buff *skb, 1288 const struct net_device *vrf_dev, 1289 const struct net_device *slave_dev) 1290 { 1291 struct net_vrf *vrf = netdev_priv(vrf_dev); 1292 1293 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1294 return -EMSGSIZE; 1295 1296 return 0; 1297 } 1298 1299 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1300 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1301 }; 1302 1303 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1304 .kind = DRV_NAME, 1305 .priv_size = sizeof(struct net_vrf), 1306 1307 .get_size = vrf_nl_getsize, 1308 .policy = vrf_nl_policy, 1309 .validate = vrf_validate, 1310 .fill_info = vrf_fillinfo, 1311 1312 .get_slave_size = vrf_get_slave_size, 1313 .fill_slave_info = vrf_fill_slave_info, 1314 1315 .newlink = vrf_newlink, 1316 .dellink = vrf_dellink, 1317 .setup = vrf_setup, 1318 .maxtype = IFLA_VRF_MAX, 1319 }; 1320 1321 static int vrf_device_event(struct notifier_block *unused, 1322 unsigned long event, void *ptr) 1323 { 1324 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1325 1326 /* only care about unregister events to drop slave references */ 1327 if (event == NETDEV_UNREGISTER) { 1328 struct net_device *vrf_dev; 1329 1330 if (!netif_is_l3_slave(dev)) 1331 goto out; 1332 1333 vrf_dev = netdev_master_upper_dev_get(dev); 1334 vrf_del_slave(vrf_dev, dev); 1335 } 1336 out: 1337 return NOTIFY_DONE; 1338 } 1339 1340 static struct notifier_block vrf_notifier_block __read_mostly = { 1341 .notifier_call = vrf_device_event, 1342 }; 1343 1344 static int __init vrf_init_module(void) 1345 { 1346 int rc; 1347 1348 register_netdevice_notifier(&vrf_notifier_block); 1349 1350 rc = rtnl_link_register(&vrf_link_ops); 1351 if (rc < 0) 1352 goto error; 1353 1354 return 0; 1355 1356 error: 1357 unregister_netdevice_notifier(&vrf_notifier_block); 1358 return rc; 1359 } 1360 1361 module_init(vrf_init_module); 1362 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1363 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1364 MODULE_LICENSE("GPL"); 1365 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1366 MODULE_VERSION(DRV_VERSION); 1367