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