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 void 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 } 106 107 /* Local traffic destined to local address. Reinsert the packet to rx 108 * path, similar to loopback handling. 109 */ 110 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 111 struct dst_entry *dst) 112 { 113 int len = skb->len; 114 115 skb_orphan(skb); 116 117 skb_dst_set(skb, dst); 118 skb_dst_force(skb); 119 120 /* set pkt_type to avoid skb hitting packet taps twice - 121 * once on Tx and again in Rx processing 122 */ 123 skb->pkt_type = PACKET_LOOPBACK; 124 125 skb->protocol = eth_type_trans(skb, dev); 126 127 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 128 vrf_rx_stats(dev, len); 129 else 130 this_cpu_inc(dev->dstats->rx_drps); 131 132 return NETDEV_TX_OK; 133 } 134 135 #if IS_ENABLED(CONFIG_IPV6) 136 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 137 struct sk_buff *skb) 138 { 139 int err; 140 141 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 142 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 143 144 if (likely(err == 1)) 145 err = dst_output(net, sk, skb); 146 147 return err; 148 } 149 150 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 151 struct net_device *dev) 152 { 153 const struct ipv6hdr *iph = ipv6_hdr(skb); 154 struct net *net = dev_net(skb->dev); 155 struct flowi6 fl6 = { 156 /* needed to match OIF rule */ 157 .flowi6_oif = dev->ifindex, 158 .flowi6_iif = LOOPBACK_IFINDEX, 159 .daddr = iph->daddr, 160 .saddr = iph->saddr, 161 .flowlabel = ip6_flowinfo(iph), 162 .flowi6_mark = skb->mark, 163 .flowi6_proto = iph->nexthdr, 164 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF, 165 }; 166 int ret = NET_XMIT_DROP; 167 struct dst_entry *dst; 168 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 169 170 dst = ip6_route_output(net, NULL, &fl6); 171 if (dst == dst_null) 172 goto err; 173 174 skb_dst_drop(skb); 175 176 /* if dst.dev is loopback or the VRF device again this is locally 177 * originated traffic destined to a local address. Short circuit 178 * to Rx path using our local dst 179 */ 180 if (dst->dev == net->loopback_dev || dst->dev == dev) { 181 struct net_vrf *vrf = netdev_priv(dev); 182 struct rt6_info *rt6_local; 183 184 /* release looked up dst and use cached local dst */ 185 dst_release(dst); 186 187 rcu_read_lock(); 188 189 rt6_local = rcu_dereference(vrf->rt6_local); 190 if (unlikely(!rt6_local)) { 191 rcu_read_unlock(); 192 goto err; 193 } 194 195 /* Ordering issue: cached local dst is created on newlink 196 * before the IPv6 initialization. Using the local dst 197 * requires rt6i_idev to be set so make sure it is. 198 */ 199 if (unlikely(!rt6_local->rt6i_idev)) { 200 rt6_local->rt6i_idev = in6_dev_get(dev); 201 if (!rt6_local->rt6i_idev) { 202 rcu_read_unlock(); 203 goto err; 204 } 205 } 206 207 dst = &rt6_local->dst; 208 dst_hold(dst); 209 210 rcu_read_unlock(); 211 212 return vrf_local_xmit(skb, dev, &rt6_local->dst); 213 } 214 215 skb_dst_set(skb, dst); 216 217 /* strip the ethernet header added for pass through VRF device */ 218 __skb_pull(skb, skb_network_offset(skb)); 219 220 ret = vrf_ip6_local_out(net, skb->sk, skb); 221 if (unlikely(net_xmit_eval(ret))) 222 dev->stats.tx_errors++; 223 else 224 ret = NET_XMIT_SUCCESS; 225 226 return ret; 227 err: 228 vrf_tx_error(dev, skb); 229 return NET_XMIT_DROP; 230 } 231 #else 232 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 233 struct net_device *dev) 234 { 235 vrf_tx_error(dev, skb); 236 return NET_XMIT_DROP; 237 } 238 #endif 239 240 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 241 static int vrf_ip_local_out(struct net *net, struct sock *sk, 242 struct sk_buff *skb) 243 { 244 int err; 245 246 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 247 skb, NULL, skb_dst(skb)->dev, dst_output); 248 if (likely(err == 1)) 249 err = dst_output(net, sk, skb); 250 251 return err; 252 } 253 254 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 255 struct net_device *vrf_dev) 256 { 257 struct iphdr *ip4h = ip_hdr(skb); 258 int ret = NET_XMIT_DROP; 259 struct flowi4 fl4 = { 260 /* needed to match OIF rule */ 261 .flowi4_oif = vrf_dev->ifindex, 262 .flowi4_iif = LOOPBACK_IFINDEX, 263 .flowi4_tos = RT_TOS(ip4h->tos), 264 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF, 265 .flowi4_proto = ip4h->protocol, 266 .daddr = ip4h->daddr, 267 .saddr = ip4h->saddr, 268 }; 269 struct net *net = dev_net(vrf_dev); 270 struct rtable *rt; 271 272 rt = ip_route_output_flow(net, &fl4, NULL); 273 if (IS_ERR(rt)) 274 goto err; 275 276 skb_dst_drop(skb); 277 278 /* if dst.dev is loopback or the VRF device again this is locally 279 * originated traffic destined to a local address. Short circuit 280 * to Rx path using our local dst 281 */ 282 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) { 283 struct net_vrf *vrf = netdev_priv(vrf_dev); 284 struct rtable *rth_local; 285 struct dst_entry *dst = NULL; 286 287 ip_rt_put(rt); 288 289 rcu_read_lock(); 290 291 rth_local = rcu_dereference(vrf->rth_local); 292 if (likely(rth_local)) { 293 dst = &rth_local->dst; 294 dst_hold(dst); 295 } 296 297 rcu_read_unlock(); 298 299 if (unlikely(!dst)) 300 goto err; 301 302 return vrf_local_xmit(skb, vrf_dev, dst); 303 } 304 305 skb_dst_set(skb, &rt->dst); 306 307 /* strip the ethernet header added for pass through VRF device */ 308 __skb_pull(skb, skb_network_offset(skb)); 309 310 if (!ip4h->saddr) { 311 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 312 RT_SCOPE_LINK); 313 } 314 315 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 316 if (unlikely(net_xmit_eval(ret))) 317 vrf_dev->stats.tx_errors++; 318 else 319 ret = NET_XMIT_SUCCESS; 320 321 out: 322 return ret; 323 err: 324 vrf_tx_error(vrf_dev, skb); 325 goto out; 326 } 327 328 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 329 { 330 switch (skb->protocol) { 331 case htons(ETH_P_IP): 332 return vrf_process_v4_outbound(skb, dev); 333 case htons(ETH_P_IPV6): 334 return vrf_process_v6_outbound(skb, dev); 335 default: 336 vrf_tx_error(dev, skb); 337 return NET_XMIT_DROP; 338 } 339 } 340 341 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 342 { 343 int len = skb->len; 344 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 345 346 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 347 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 348 349 u64_stats_update_begin(&dstats->syncp); 350 dstats->tx_pkts++; 351 dstats->tx_bytes += len; 352 u64_stats_update_end(&dstats->syncp); 353 } else { 354 this_cpu_inc(dev->dstats->tx_drps); 355 } 356 357 return ret; 358 } 359 360 #if IS_ENABLED(CONFIG_IPV6) 361 /* modelled after ip6_finish_output2 */ 362 static int vrf_finish_output6(struct net *net, struct sock *sk, 363 struct sk_buff *skb) 364 { 365 struct dst_entry *dst = skb_dst(skb); 366 struct net_device *dev = dst->dev; 367 struct neighbour *neigh; 368 struct in6_addr *nexthop; 369 int ret; 370 371 nf_reset(skb); 372 373 skb->protocol = htons(ETH_P_IPV6); 374 skb->dev = dev; 375 376 rcu_read_lock_bh(); 377 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 378 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 379 if (unlikely(!neigh)) 380 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 381 if (!IS_ERR(neigh)) { 382 sock_confirm_neigh(skb, neigh); 383 ret = neigh_output(neigh, skb); 384 rcu_read_unlock_bh(); 385 return ret; 386 } 387 rcu_read_unlock_bh(); 388 389 IP6_INC_STATS(dev_net(dst->dev), 390 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 391 kfree_skb(skb); 392 return -EINVAL; 393 } 394 395 /* modelled after ip6_output */ 396 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 397 { 398 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 399 net, sk, skb, NULL, skb_dst(skb)->dev, 400 vrf_finish_output6, 401 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 402 } 403 404 /* set dst on skb to send packet to us via dev_xmit path. Allows 405 * packet to go through device based features such as qdisc, netfilter 406 * hooks and packet sockets with skb->dev set to vrf device. 407 */ 408 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 409 struct sock *sk, 410 struct sk_buff *skb) 411 { 412 struct net_vrf *vrf = netdev_priv(vrf_dev); 413 struct dst_entry *dst = NULL; 414 struct rt6_info *rt6; 415 416 /* don't divert link scope packets */ 417 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 418 return skb; 419 420 rcu_read_lock(); 421 422 rt6 = rcu_dereference(vrf->rt6); 423 if (likely(rt6)) { 424 dst = &rt6->dst; 425 dst_hold(dst); 426 } 427 428 rcu_read_unlock(); 429 430 if (unlikely(!dst)) { 431 vrf_tx_error(vrf_dev, skb); 432 return NULL; 433 } 434 435 skb_dst_drop(skb); 436 skb_dst_set(skb, dst); 437 438 return skb; 439 } 440 441 /* holding rtnl */ 442 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 443 { 444 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 445 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local); 446 struct net *net = dev_net(dev); 447 struct dst_entry *dst; 448 449 RCU_INIT_POINTER(vrf->rt6, NULL); 450 RCU_INIT_POINTER(vrf->rt6_local, NULL); 451 synchronize_rcu(); 452 453 /* move dev in dst's to loopback so this VRF device can be deleted 454 * - based on dst_ifdown 455 */ 456 if (rt6) { 457 dst = &rt6->dst; 458 dev_put(dst->dev); 459 dst->dev = net->loopback_dev; 460 dev_hold(dst->dev); 461 dst_release(dst); 462 } 463 464 if (rt6_local) { 465 if (rt6_local->rt6i_idev) 466 in6_dev_put(rt6_local->rt6i_idev); 467 468 dst = &rt6_local->dst; 469 dev_put(dst->dev); 470 dst->dev = net->loopback_dev; 471 dev_hold(dst->dev); 472 dst_release(dst); 473 } 474 } 475 476 static int vrf_rt6_create(struct net_device *dev) 477 { 478 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE; 479 struct net_vrf *vrf = netdev_priv(dev); 480 struct net *net = dev_net(dev); 481 struct fib6_table *rt6i_table; 482 struct rt6_info *rt6, *rt6_local; 483 int rc = -ENOMEM; 484 485 /* IPv6 can be CONFIG enabled and then disabled runtime */ 486 if (!ipv6_mod_enabled()) 487 return 0; 488 489 rt6i_table = fib6_new_table(net, vrf->tb_id); 490 if (!rt6i_table) 491 goto out; 492 493 /* create a dst for routing packets out a VRF device */ 494 rt6 = ip6_dst_alloc(net, dev, flags); 495 if (!rt6) 496 goto out; 497 498 dst_hold(&rt6->dst); 499 500 rt6->rt6i_table = rt6i_table; 501 rt6->dst.output = vrf_output6; 502 503 /* create a dst for local routing - packets sent locally 504 * to local address via the VRF device as a loopback 505 */ 506 rt6_local = ip6_dst_alloc(net, dev, flags); 507 if (!rt6_local) { 508 dst_release(&rt6->dst); 509 goto out; 510 } 511 512 dst_hold(&rt6_local->dst); 513 514 rt6_local->rt6i_idev = in6_dev_get(dev); 515 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL; 516 rt6_local->rt6i_table = rt6i_table; 517 rt6_local->dst.input = ip6_input; 518 519 rcu_assign_pointer(vrf->rt6, rt6); 520 rcu_assign_pointer(vrf->rt6_local, rt6_local); 521 522 rc = 0; 523 out: 524 return rc; 525 } 526 #else 527 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 528 struct sock *sk, 529 struct sk_buff *skb) 530 { 531 return skb; 532 } 533 534 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 535 { 536 } 537 538 static int vrf_rt6_create(struct net_device *dev) 539 { 540 return 0; 541 } 542 #endif 543 544 /* modelled after ip_finish_output2 */ 545 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 546 { 547 struct dst_entry *dst = skb_dst(skb); 548 struct rtable *rt = (struct rtable *)dst; 549 struct net_device *dev = dst->dev; 550 unsigned int hh_len = LL_RESERVED_SPACE(dev); 551 struct neighbour *neigh; 552 u32 nexthop; 553 int ret = -EINVAL; 554 555 nf_reset(skb); 556 557 /* Be paranoid, rather than too clever. */ 558 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 559 struct sk_buff *skb2; 560 561 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 562 if (!skb2) { 563 ret = -ENOMEM; 564 goto err; 565 } 566 if (skb->sk) 567 skb_set_owner_w(skb2, skb->sk); 568 569 consume_skb(skb); 570 skb = skb2; 571 } 572 573 rcu_read_lock_bh(); 574 575 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); 576 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 577 if (unlikely(!neigh)) 578 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 579 if (!IS_ERR(neigh)) { 580 sock_confirm_neigh(skb, neigh); 581 ret = neigh_output(neigh, skb); 582 } 583 584 rcu_read_unlock_bh(); 585 err: 586 if (unlikely(ret < 0)) 587 vrf_tx_error(skb->dev, skb); 588 return ret; 589 } 590 591 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 592 { 593 struct net_device *dev = skb_dst(skb)->dev; 594 595 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 596 597 skb->dev = dev; 598 skb->protocol = htons(ETH_P_IP); 599 600 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 601 net, sk, skb, NULL, dev, 602 vrf_finish_output, 603 !(IPCB(skb)->flags & IPSKB_REROUTED)); 604 } 605 606 /* set dst on skb to send packet to us via dev_xmit path. Allows 607 * packet to go through device based features such as qdisc, netfilter 608 * hooks and packet sockets with skb->dev set to vrf device. 609 */ 610 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 611 struct sock *sk, 612 struct sk_buff *skb) 613 { 614 struct net_vrf *vrf = netdev_priv(vrf_dev); 615 struct dst_entry *dst = NULL; 616 struct rtable *rth; 617 618 /* don't divert multicast */ 619 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 620 return skb; 621 622 rcu_read_lock(); 623 624 rth = rcu_dereference(vrf->rth); 625 if (likely(rth)) { 626 dst = &rth->dst; 627 dst_hold(dst); 628 } 629 630 rcu_read_unlock(); 631 632 if (unlikely(!dst)) { 633 vrf_tx_error(vrf_dev, skb); 634 return NULL; 635 } 636 637 skb_dst_drop(skb); 638 skb_dst_set(skb, dst); 639 640 return skb; 641 } 642 643 /* called with rcu lock held */ 644 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 645 struct sock *sk, 646 struct sk_buff *skb, 647 u16 proto) 648 { 649 switch (proto) { 650 case AF_INET: 651 return vrf_ip_out(vrf_dev, sk, skb); 652 case AF_INET6: 653 return vrf_ip6_out(vrf_dev, sk, skb); 654 } 655 656 return skb; 657 } 658 659 /* holding rtnl */ 660 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 661 { 662 struct rtable *rth = rtnl_dereference(vrf->rth); 663 struct rtable *rth_local = rtnl_dereference(vrf->rth_local); 664 struct net *net = dev_net(dev); 665 struct dst_entry *dst; 666 667 RCU_INIT_POINTER(vrf->rth, NULL); 668 RCU_INIT_POINTER(vrf->rth_local, NULL); 669 synchronize_rcu(); 670 671 /* move dev in dst's to loopback so this VRF device can be deleted 672 * - based on dst_ifdown 673 */ 674 if (rth) { 675 dst = &rth->dst; 676 dev_put(dst->dev); 677 dst->dev = net->loopback_dev; 678 dev_hold(dst->dev); 679 dst_release(dst); 680 } 681 682 if (rth_local) { 683 dst = &rth_local->dst; 684 dev_put(dst->dev); 685 dst->dev = net->loopback_dev; 686 dev_hold(dst->dev); 687 dst_release(dst); 688 } 689 } 690 691 static int vrf_rtable_create(struct net_device *dev) 692 { 693 struct net_vrf *vrf = netdev_priv(dev); 694 struct rtable *rth, *rth_local; 695 696 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 697 return -ENOMEM; 698 699 /* create a dst for routing packets out through a VRF device */ 700 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); 701 if (!rth) 702 return -ENOMEM; 703 704 /* create a dst for local ingress routing - packets sent locally 705 * to local address via the VRF device as a loopback 706 */ 707 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0); 708 if (!rth_local) { 709 dst_release(&rth->dst); 710 return -ENOMEM; 711 } 712 713 rth->dst.output = vrf_output; 714 rth->rt_table_id = vrf->tb_id; 715 716 rth_local->rt_table_id = vrf->tb_id; 717 718 rcu_assign_pointer(vrf->rth, rth); 719 rcu_assign_pointer(vrf->rth_local, rth_local); 720 721 return 0; 722 } 723 724 /**************************** device handling ********************/ 725 726 /* cycle interface to flush neighbor cache and move routes across tables */ 727 static void cycle_netdev(struct net_device *dev) 728 { 729 unsigned int flags = dev->flags; 730 int ret; 731 732 if (!netif_running(dev)) 733 return; 734 735 ret = dev_change_flags(dev, flags & ~IFF_UP); 736 if (ret >= 0) 737 ret = dev_change_flags(dev, flags); 738 739 if (ret < 0) { 740 netdev_err(dev, 741 "Failed to cycle device %s; route tables might be wrong!\n", 742 dev->name); 743 } 744 } 745 746 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 747 { 748 int ret; 749 750 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); 751 if (ret < 0) 752 return ret; 753 754 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 755 cycle_netdev(port_dev); 756 757 return 0; 758 } 759 760 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) 761 { 762 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) 763 return -EINVAL; 764 765 return do_vrf_add_slave(dev, port_dev); 766 } 767 768 /* inverse of do_vrf_add_slave */ 769 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 770 { 771 netdev_upper_dev_unlink(port_dev, dev); 772 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 773 774 cycle_netdev(port_dev); 775 776 return 0; 777 } 778 779 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 780 { 781 return do_vrf_del_slave(dev, port_dev); 782 } 783 784 static void vrf_dev_uninit(struct net_device *dev) 785 { 786 struct net_vrf *vrf = netdev_priv(dev); 787 struct net_device *port_dev; 788 struct list_head *iter; 789 790 vrf_rtable_release(dev, vrf); 791 vrf_rt6_release(dev, vrf); 792 793 netdev_for_each_lower_dev(dev, port_dev, iter) 794 vrf_del_slave(dev, port_dev); 795 796 free_percpu(dev->dstats); 797 dev->dstats = NULL; 798 } 799 800 static int vrf_dev_init(struct net_device *dev) 801 { 802 struct net_vrf *vrf = netdev_priv(dev); 803 804 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 805 if (!dev->dstats) 806 goto out_nomem; 807 808 /* create the default dst which points back to us */ 809 if (vrf_rtable_create(dev) != 0) 810 goto out_stats; 811 812 if (vrf_rt6_create(dev) != 0) 813 goto out_rth; 814 815 dev->flags = IFF_MASTER | IFF_NOARP; 816 817 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 818 dev->mtu = 64 * 1024; 819 820 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 821 dev->operstate = IF_OPER_UP; 822 netdev_lockdep_set_classes(dev); 823 return 0; 824 825 out_rth: 826 vrf_rtable_release(dev, vrf); 827 out_stats: 828 free_percpu(dev->dstats); 829 dev->dstats = NULL; 830 out_nomem: 831 return -ENOMEM; 832 } 833 834 static const struct net_device_ops vrf_netdev_ops = { 835 .ndo_init = vrf_dev_init, 836 .ndo_uninit = vrf_dev_uninit, 837 .ndo_start_xmit = vrf_xmit, 838 .ndo_get_stats64 = vrf_get_stats64, 839 .ndo_add_slave = vrf_add_slave, 840 .ndo_del_slave = vrf_del_slave, 841 }; 842 843 static u32 vrf_fib_table(const struct net_device *dev) 844 { 845 struct net_vrf *vrf = netdev_priv(dev); 846 847 return vrf->tb_id; 848 } 849 850 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 851 { 852 return 0; 853 } 854 855 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 856 struct sk_buff *skb, 857 struct net_device *dev) 858 { 859 struct net *net = dev_net(dev); 860 861 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0) 862 skb = NULL; /* kfree_skb(skb) handled by nf code */ 863 864 return skb; 865 } 866 867 #if IS_ENABLED(CONFIG_IPV6) 868 /* neighbor handling is done with actual device; do not want 869 * to flip skb->dev for those ndisc packets. This really fails 870 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 871 * a start. 872 */ 873 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 874 { 875 const struct ipv6hdr *iph = ipv6_hdr(skb); 876 bool rc = false; 877 878 if (iph->nexthdr == NEXTHDR_ICMP) { 879 const struct icmp6hdr *icmph; 880 struct icmp6hdr _icmph; 881 882 icmph = skb_header_pointer(skb, sizeof(*iph), 883 sizeof(_icmph), &_icmph); 884 if (!icmph) 885 goto out; 886 887 switch (icmph->icmp6_type) { 888 case NDISC_ROUTER_SOLICITATION: 889 case NDISC_ROUTER_ADVERTISEMENT: 890 case NDISC_NEIGHBOUR_SOLICITATION: 891 case NDISC_NEIGHBOUR_ADVERTISEMENT: 892 case NDISC_REDIRECT: 893 rc = true; 894 break; 895 } 896 } 897 898 out: 899 return rc; 900 } 901 902 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 903 const struct net_device *dev, 904 struct flowi6 *fl6, 905 int ifindex, 906 int flags) 907 { 908 struct net_vrf *vrf = netdev_priv(dev); 909 struct fib6_table *table = NULL; 910 struct rt6_info *rt6; 911 912 rcu_read_lock(); 913 914 /* fib6_table does not have a refcnt and can not be freed */ 915 rt6 = rcu_dereference(vrf->rt6); 916 if (likely(rt6)) 917 table = rt6->rt6i_table; 918 919 rcu_read_unlock(); 920 921 if (!table) 922 return NULL; 923 924 return ip6_pol_route(net, table, ifindex, fl6, flags); 925 } 926 927 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 928 int ifindex) 929 { 930 const struct ipv6hdr *iph = ipv6_hdr(skb); 931 struct flowi6 fl6 = { 932 .daddr = iph->daddr, 933 .saddr = iph->saddr, 934 .flowlabel = ip6_flowinfo(iph), 935 .flowi6_mark = skb->mark, 936 .flowi6_proto = iph->nexthdr, 937 .flowi6_iif = ifindex, 938 }; 939 struct net *net = dev_net(vrf_dev); 940 struct rt6_info *rt6; 941 942 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, 943 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 944 if (unlikely(!rt6)) 945 return; 946 947 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 948 return; 949 950 skb_dst_set(skb, &rt6->dst); 951 } 952 953 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 954 struct sk_buff *skb) 955 { 956 int orig_iif = skb->skb_iif; 957 bool need_strict; 958 959 /* loopback traffic; do not push through packet taps again. 960 * Reset pkt_type for upper layers to process skb 961 */ 962 if (skb->pkt_type == PACKET_LOOPBACK) { 963 skb->dev = vrf_dev; 964 skb->skb_iif = vrf_dev->ifindex; 965 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 966 skb->pkt_type = PACKET_HOST; 967 goto out; 968 } 969 970 /* if packet is NDISC or addressed to multicast or link-local 971 * then keep the ingress interface 972 */ 973 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 974 if (!ipv6_ndisc_frame(skb) && !need_strict) { 975 vrf_rx_stats(vrf_dev, skb->len); 976 skb->dev = vrf_dev; 977 skb->skb_iif = vrf_dev->ifindex; 978 979 skb_push(skb, skb->mac_len); 980 dev_queue_xmit_nit(skb, vrf_dev); 981 skb_pull(skb, skb->mac_len); 982 983 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 984 } 985 986 if (need_strict) 987 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 988 989 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 990 out: 991 return skb; 992 } 993 994 #else 995 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 996 struct sk_buff *skb) 997 { 998 return skb; 999 } 1000 #endif 1001 1002 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1003 struct sk_buff *skb) 1004 { 1005 skb->dev = vrf_dev; 1006 skb->skb_iif = vrf_dev->ifindex; 1007 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1008 1009 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1010 goto out; 1011 1012 /* loopback traffic; do not push through packet taps again. 1013 * Reset pkt_type for upper layers to process skb 1014 */ 1015 if (skb->pkt_type == PACKET_LOOPBACK) { 1016 skb->pkt_type = PACKET_HOST; 1017 goto out; 1018 } 1019 1020 vrf_rx_stats(vrf_dev, skb->len); 1021 1022 skb_push(skb, skb->mac_len); 1023 dev_queue_xmit_nit(skb, vrf_dev); 1024 skb_pull(skb, skb->mac_len); 1025 1026 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1027 out: 1028 return skb; 1029 } 1030 1031 /* called with rcu lock held */ 1032 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1033 struct sk_buff *skb, 1034 u16 proto) 1035 { 1036 switch (proto) { 1037 case AF_INET: 1038 return vrf_ip_rcv(vrf_dev, skb); 1039 case AF_INET6: 1040 return vrf_ip6_rcv(vrf_dev, skb); 1041 } 1042 1043 return skb; 1044 } 1045 1046 #if IS_ENABLED(CONFIG_IPV6) 1047 /* send to link-local or multicast address via interface enslaved to 1048 * VRF device. Force lookup to VRF table without changing flow struct 1049 */ 1050 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1051 struct flowi6 *fl6) 1052 { 1053 struct net *net = dev_net(dev); 1054 int flags = RT6_LOOKUP_F_IFACE; 1055 struct dst_entry *dst = NULL; 1056 struct rt6_info *rt; 1057 1058 /* VRF device does not have a link-local address and 1059 * sending packets to link-local or mcast addresses over 1060 * a VRF device does not make sense 1061 */ 1062 if (fl6->flowi6_oif == dev->ifindex) { 1063 dst = &net->ipv6.ip6_null_entry->dst; 1064 dst_hold(dst); 1065 return dst; 1066 } 1067 1068 if (!ipv6_addr_any(&fl6->saddr)) 1069 flags |= RT6_LOOKUP_F_HAS_SADDR; 1070 1071 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags); 1072 if (rt) 1073 dst = &rt->dst; 1074 1075 return dst; 1076 } 1077 #endif 1078 1079 static const struct l3mdev_ops vrf_l3mdev_ops = { 1080 .l3mdev_fib_table = vrf_fib_table, 1081 .l3mdev_l3_rcv = vrf_l3_rcv, 1082 .l3mdev_l3_out = vrf_l3_out, 1083 #if IS_ENABLED(CONFIG_IPV6) 1084 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1085 #endif 1086 }; 1087 1088 static void vrf_get_drvinfo(struct net_device *dev, 1089 struct ethtool_drvinfo *info) 1090 { 1091 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1092 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1093 } 1094 1095 static const struct ethtool_ops vrf_ethtool_ops = { 1096 .get_drvinfo = vrf_get_drvinfo, 1097 }; 1098 1099 static inline size_t vrf_fib_rule_nl_size(void) 1100 { 1101 size_t sz; 1102 1103 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1104 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1105 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1106 1107 return sz; 1108 } 1109 1110 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1111 { 1112 struct fib_rule_hdr *frh; 1113 struct nlmsghdr *nlh; 1114 struct sk_buff *skb; 1115 int err; 1116 1117 if (family == AF_INET6 && !ipv6_mod_enabled()) 1118 return 0; 1119 1120 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1121 if (!skb) 1122 return -ENOMEM; 1123 1124 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1125 if (!nlh) 1126 goto nla_put_failure; 1127 1128 /* rule only needs to appear once */ 1129 nlh->nlmsg_flags &= NLM_F_EXCL; 1130 1131 frh = nlmsg_data(nlh); 1132 memset(frh, 0, sizeof(*frh)); 1133 frh->family = family; 1134 frh->action = FR_ACT_TO_TBL; 1135 1136 if (nla_put_u32(skb, FRA_L3MDEV, 1)) 1137 goto nla_put_failure; 1138 1139 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1140 goto nla_put_failure; 1141 1142 nlmsg_end(skb, nlh); 1143 1144 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1145 skb->sk = dev_net(dev)->rtnl; 1146 if (add_it) { 1147 err = fib_nl_newrule(skb, nlh); 1148 if (err == -EEXIST) 1149 err = 0; 1150 } else { 1151 err = fib_nl_delrule(skb, nlh); 1152 if (err == -ENOENT) 1153 err = 0; 1154 } 1155 nlmsg_free(skb); 1156 1157 return err; 1158 1159 nla_put_failure: 1160 nlmsg_free(skb); 1161 1162 return -EMSGSIZE; 1163 } 1164 1165 static int vrf_add_fib_rules(const struct net_device *dev) 1166 { 1167 int err; 1168 1169 err = vrf_fib_rule(dev, AF_INET, true); 1170 if (err < 0) 1171 goto out_err; 1172 1173 err = vrf_fib_rule(dev, AF_INET6, true); 1174 if (err < 0) 1175 goto ipv6_err; 1176 1177 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1178 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1179 if (err < 0) 1180 goto ipmr_err; 1181 #endif 1182 1183 return 0; 1184 1185 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1186 ipmr_err: 1187 vrf_fib_rule(dev, AF_INET6, false); 1188 #endif 1189 1190 ipv6_err: 1191 vrf_fib_rule(dev, AF_INET, false); 1192 1193 out_err: 1194 netdev_err(dev, "Failed to add FIB rules.\n"); 1195 return err; 1196 } 1197 1198 static void vrf_setup(struct net_device *dev) 1199 { 1200 ether_setup(dev); 1201 1202 /* Initialize the device structure. */ 1203 dev->netdev_ops = &vrf_netdev_ops; 1204 dev->l3mdev_ops = &vrf_l3mdev_ops; 1205 dev->ethtool_ops = &vrf_ethtool_ops; 1206 dev->destructor = free_netdev; 1207 1208 /* Fill in device structure with ethernet-generic values. */ 1209 eth_hw_addr_random(dev); 1210 1211 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1212 dev->features |= NETIF_F_LLTX; 1213 1214 /* don't allow vrf devices to change network namespaces. */ 1215 dev->features |= NETIF_F_NETNS_LOCAL; 1216 1217 /* does not make sense for a VLAN to be added to a vrf device */ 1218 dev->features |= NETIF_F_VLAN_CHALLENGED; 1219 1220 /* enable offload features */ 1221 dev->features |= NETIF_F_GSO_SOFTWARE; 1222 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM; 1223 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1224 1225 dev->hw_features = dev->features; 1226 dev->hw_enc_features = dev->features; 1227 1228 /* default to no qdisc; user can add if desired */ 1229 dev->priv_flags |= IFF_NO_QUEUE; 1230 } 1231 1232 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) 1233 { 1234 if (tb[IFLA_ADDRESS]) { 1235 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1236 return -EINVAL; 1237 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1238 return -EADDRNOTAVAIL; 1239 } 1240 return 0; 1241 } 1242 1243 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1244 { 1245 unregister_netdevice_queue(dev, head); 1246 } 1247 1248 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1249 struct nlattr *tb[], struct nlattr *data[]) 1250 { 1251 struct net_vrf *vrf = netdev_priv(dev); 1252 int err; 1253 1254 if (!data || !data[IFLA_VRF_TABLE]) 1255 return -EINVAL; 1256 1257 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1258 if (vrf->tb_id == RT_TABLE_UNSPEC) 1259 return -EINVAL; 1260 1261 dev->priv_flags |= IFF_L3MDEV_MASTER; 1262 1263 err = register_netdevice(dev); 1264 if (err) 1265 goto out; 1266 1267 if (add_fib_rules) { 1268 err = vrf_add_fib_rules(dev); 1269 if (err) { 1270 unregister_netdevice(dev); 1271 goto out; 1272 } 1273 add_fib_rules = false; 1274 } 1275 1276 out: 1277 return err; 1278 } 1279 1280 static size_t vrf_nl_getsize(const struct net_device *dev) 1281 { 1282 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1283 } 1284 1285 static int vrf_fillinfo(struct sk_buff *skb, 1286 const struct net_device *dev) 1287 { 1288 struct net_vrf *vrf = netdev_priv(dev); 1289 1290 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1291 } 1292 1293 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1294 const struct net_device *slave_dev) 1295 { 1296 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1297 } 1298 1299 static int vrf_fill_slave_info(struct sk_buff *skb, 1300 const struct net_device *vrf_dev, 1301 const struct net_device *slave_dev) 1302 { 1303 struct net_vrf *vrf = netdev_priv(vrf_dev); 1304 1305 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1306 return -EMSGSIZE; 1307 1308 return 0; 1309 } 1310 1311 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1312 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1313 }; 1314 1315 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1316 .kind = DRV_NAME, 1317 .priv_size = sizeof(struct net_vrf), 1318 1319 .get_size = vrf_nl_getsize, 1320 .policy = vrf_nl_policy, 1321 .validate = vrf_validate, 1322 .fill_info = vrf_fillinfo, 1323 1324 .get_slave_size = vrf_get_slave_size, 1325 .fill_slave_info = vrf_fill_slave_info, 1326 1327 .newlink = vrf_newlink, 1328 .dellink = vrf_dellink, 1329 .setup = vrf_setup, 1330 .maxtype = IFLA_VRF_MAX, 1331 }; 1332 1333 static int vrf_device_event(struct notifier_block *unused, 1334 unsigned long event, void *ptr) 1335 { 1336 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1337 1338 /* only care about unregister events to drop slave references */ 1339 if (event == NETDEV_UNREGISTER) { 1340 struct net_device *vrf_dev; 1341 1342 if (!netif_is_l3_slave(dev)) 1343 goto out; 1344 1345 vrf_dev = netdev_master_upper_dev_get(dev); 1346 vrf_del_slave(vrf_dev, dev); 1347 } 1348 out: 1349 return NOTIFY_DONE; 1350 } 1351 1352 static struct notifier_block vrf_notifier_block __read_mostly = { 1353 .notifier_call = vrf_device_event, 1354 }; 1355 1356 static int __init vrf_init_module(void) 1357 { 1358 int rc; 1359 1360 register_netdevice_notifier(&vrf_notifier_block); 1361 1362 rc = rtnl_link_register(&vrf_link_ops); 1363 if (rc < 0) 1364 goto error; 1365 1366 return 0; 1367 1368 error: 1369 unregister_netdevice_notifier(&vrf_notifier_block); 1370 return rc; 1371 } 1372 1373 module_init(vrf_init_module); 1374 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 1375 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 1376 MODULE_LICENSE("GPL"); 1377 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1378 MODULE_VERSION(DRV_VERSION); 1379