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