1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * vrf.c: device driver to encapsulate a VRF space 4 * 5 * Copyright (c) 2015 Cumulus Networks. All rights reserved. 6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com> 7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com> 8 * 9 * Based on dummy, team and ipvlan drivers 10 */ 11 12 #include <linux/ethtool.h> 13 #include <linux/module.h> 14 #include <linux/kernel.h> 15 #include <linux/netdevice.h> 16 #include <linux/etherdevice.h> 17 #include <linux/ip.h> 18 #include <linux/init.h> 19 #include <linux/moduleparam.h> 20 #include <linux/netfilter.h> 21 #include <linux/rtnetlink.h> 22 #include <net/rtnetlink.h> 23 #include <linux/u64_stats_sync.h> 24 #include <linux/hashtable.h> 25 #include <linux/spinlock_types.h> 26 27 #include <linux/inetdevice.h> 28 #include <net/arp.h> 29 #include <net/ip.h> 30 #include <net/ip_fib.h> 31 #include <net/ip6_fib.h> 32 #include <net/ip6_route.h> 33 #include <net/route.h> 34 #include <net/addrconf.h> 35 #include <net/l3mdev.h> 36 #include <net/fib_rules.h> 37 #include <net/netns/generic.h> 38 39 #define DRV_NAME "vrf" 40 #define DRV_VERSION "1.1" 41 42 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 43 44 #define HT_MAP_BITS 4 45 #define HASH_INITVAL ((u32)0xcafef00d) 46 47 struct vrf_map { 48 DECLARE_HASHTABLE(ht, HT_MAP_BITS); 49 spinlock_t vmap_lock; 50 51 /* shared_tables: 52 * count how many distinct tables do not comply with the strict mode 53 * requirement. 54 * shared_tables value must be 0 in order to enable the strict mode. 55 * 56 * example of the evolution of shared_tables: 57 * | time 58 * add vrf0 --> table 100 shared_tables = 0 | t0 59 * add vrf1 --> table 101 shared_tables = 0 | t1 60 * add vrf2 --> table 100 shared_tables = 1 | t2 61 * add vrf3 --> table 100 shared_tables = 1 | t3 62 * add vrf4 --> table 101 shared_tables = 2 v t4 63 * 64 * shared_tables is a "step function" (or "staircase function") 65 * and it is increased by one when the second vrf is associated to a 66 * table. 67 * 68 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1. 69 * 70 * at t3, another dev (vrf3) is bound to the same table 100 but the 71 * value of shared_tables is still 1. 72 * This means that no matter how many new vrfs will register on the 73 * table 100, the shared_tables will not increase (considering only 74 * table 100). 75 * 76 * at t4, vrf4 is bound to table 101, and shared_tables = 2. 77 * 78 * Looking at the value of shared_tables we can immediately know if 79 * the strict_mode can or cannot be enforced. Indeed, strict_mode 80 * can be enforced iff shared_tables = 0. 81 * 82 * Conversely, shared_tables is decreased when a vrf is de-associated 83 * from a table with exactly two associated vrfs. 84 */ 85 u32 shared_tables; 86 87 bool strict_mode; 88 }; 89 90 struct vrf_map_elem { 91 struct hlist_node hnode; 92 struct list_head vrf_list; /* VRFs registered to this table */ 93 94 u32 table_id; 95 int users; 96 int ifindex; 97 }; 98 99 static unsigned int vrf_net_id; 100 101 /* per netns vrf data */ 102 struct netns_vrf { 103 /* protected by rtnl lock */ 104 bool add_fib_rules; 105 106 struct vrf_map vmap; 107 struct ctl_table_header *ctl_hdr; 108 }; 109 110 struct net_vrf { 111 struct rtable __rcu *rth; 112 struct rt6_info __rcu *rt6; 113 #if IS_ENABLED(CONFIG_IPV6) 114 struct fib6_table *fib6_table; 115 #endif 116 u32 tb_id; 117 118 struct list_head me_list; /* entry in vrf_map_elem */ 119 int ifindex; 120 }; 121 122 struct pcpu_dstats { 123 u64 tx_pkts; 124 u64 tx_bytes; 125 u64 tx_drps; 126 u64 rx_pkts; 127 u64 rx_bytes; 128 u64 rx_drps; 129 struct u64_stats_sync syncp; 130 }; 131 132 static void vrf_rx_stats(struct net_device *dev, int len) 133 { 134 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 135 136 u64_stats_update_begin(&dstats->syncp); 137 dstats->rx_pkts++; 138 dstats->rx_bytes += len; 139 u64_stats_update_end(&dstats->syncp); 140 } 141 142 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 143 { 144 vrf_dev->stats.tx_errors++; 145 kfree_skb(skb); 146 } 147 148 static void vrf_get_stats64(struct net_device *dev, 149 struct rtnl_link_stats64 *stats) 150 { 151 int i; 152 153 for_each_possible_cpu(i) { 154 const struct pcpu_dstats *dstats; 155 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 156 unsigned int start; 157 158 dstats = per_cpu_ptr(dev->dstats, i); 159 do { 160 start = u64_stats_fetch_begin_irq(&dstats->syncp); 161 tbytes = dstats->tx_bytes; 162 tpkts = dstats->tx_pkts; 163 tdrops = dstats->tx_drps; 164 rbytes = dstats->rx_bytes; 165 rpkts = dstats->rx_pkts; 166 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); 167 stats->tx_bytes += tbytes; 168 stats->tx_packets += tpkts; 169 stats->tx_dropped += tdrops; 170 stats->rx_bytes += rbytes; 171 stats->rx_packets += rpkts; 172 } 173 } 174 175 static struct vrf_map *netns_vrf_map(struct net *net) 176 { 177 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 178 179 return &nn_vrf->vmap; 180 } 181 182 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 183 { 184 return netns_vrf_map(dev_net(dev)); 185 } 186 187 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 188 { 189 struct list_head *me_head = &me->vrf_list; 190 struct net_vrf *vrf; 191 192 if (list_empty(me_head)) 193 return -ENODEV; 194 195 vrf = list_first_entry(me_head, struct net_vrf, me_list); 196 197 return vrf->ifindex; 198 } 199 200 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 201 { 202 struct vrf_map_elem *me; 203 204 me = kmalloc(sizeof(*me), flags); 205 if (!me) 206 return NULL; 207 208 return me; 209 } 210 211 static void vrf_map_elem_free(struct vrf_map_elem *me) 212 { 213 kfree(me); 214 } 215 216 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 217 int ifindex, int users) 218 { 219 me->table_id = table_id; 220 me->ifindex = ifindex; 221 me->users = users; 222 INIT_LIST_HEAD(&me->vrf_list); 223 } 224 225 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 226 u32 table_id) 227 { 228 struct vrf_map_elem *me; 229 u32 key; 230 231 key = jhash_1word(table_id, HASH_INITVAL); 232 hash_for_each_possible(vmap->ht, me, hnode, key) { 233 if (me->table_id == table_id) 234 return me; 235 } 236 237 return NULL; 238 } 239 240 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 241 { 242 u32 table_id = me->table_id; 243 u32 key; 244 245 key = jhash_1word(table_id, HASH_INITVAL); 246 hash_add(vmap->ht, &me->hnode, key); 247 } 248 249 static void vrf_map_del_elem(struct vrf_map_elem *me) 250 { 251 hash_del(&me->hnode); 252 } 253 254 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 255 { 256 spin_lock(&vmap->vmap_lock); 257 } 258 259 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 260 { 261 spin_unlock(&vmap->vmap_lock); 262 } 263 264 /* called with rtnl lock held */ 265 static int 266 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 267 { 268 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 269 struct net_vrf *vrf = netdev_priv(dev); 270 struct vrf_map_elem *new_me, *me; 271 u32 table_id = vrf->tb_id; 272 bool free_new_me = false; 273 int users; 274 int res; 275 276 /* we pre-allocate elements used in the spin-locked section (so that we 277 * keep the spinlock as short as possibile). 278 */ 279 new_me = vrf_map_elem_alloc(GFP_KERNEL); 280 if (!new_me) 281 return -ENOMEM; 282 283 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 284 285 vrf_map_lock(vmap); 286 287 me = vrf_map_lookup_elem(vmap, table_id); 288 if (!me) { 289 me = new_me; 290 vrf_map_add_elem(vmap, me); 291 goto link_vrf; 292 } 293 294 /* we already have an entry in the vrf_map, so it means there is (at 295 * least) a vrf registered on the specific table. 296 */ 297 free_new_me = true; 298 if (vmap->strict_mode) { 299 /* vrfs cannot share the same table */ 300 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 301 res = -EBUSY; 302 goto unlock; 303 } 304 305 link_vrf: 306 users = ++me->users; 307 if (users == 2) 308 ++vmap->shared_tables; 309 310 list_add(&vrf->me_list, &me->vrf_list); 311 312 res = 0; 313 314 unlock: 315 vrf_map_unlock(vmap); 316 317 /* clean-up, if needed */ 318 if (free_new_me) 319 vrf_map_elem_free(new_me); 320 321 return res; 322 } 323 324 /* called with rtnl lock held */ 325 static void vrf_map_unregister_dev(struct net_device *dev) 326 { 327 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 328 struct net_vrf *vrf = netdev_priv(dev); 329 u32 table_id = vrf->tb_id; 330 struct vrf_map_elem *me; 331 int users; 332 333 vrf_map_lock(vmap); 334 335 me = vrf_map_lookup_elem(vmap, table_id); 336 if (!me) 337 goto unlock; 338 339 list_del(&vrf->me_list); 340 341 users = --me->users; 342 if (users == 1) { 343 --vmap->shared_tables; 344 } else if (users == 0) { 345 vrf_map_del_elem(me); 346 347 /* no one will refer to this element anymore */ 348 vrf_map_elem_free(me); 349 } 350 351 unlock: 352 vrf_map_unlock(vmap); 353 } 354 355 /* return the vrf device index associated with the table_id */ 356 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 357 { 358 struct vrf_map *vmap = netns_vrf_map(net); 359 struct vrf_map_elem *me; 360 int ifindex; 361 362 vrf_map_lock(vmap); 363 364 if (!vmap->strict_mode) { 365 ifindex = -EPERM; 366 goto unlock; 367 } 368 369 me = vrf_map_lookup_elem(vmap, table_id); 370 if (!me) { 371 ifindex = -ENODEV; 372 goto unlock; 373 } 374 375 ifindex = vrf_map_elem_get_vrf_ifindex(me); 376 377 unlock: 378 vrf_map_unlock(vmap); 379 380 return ifindex; 381 } 382 383 /* by default VRF devices do not have a qdisc and are expected 384 * to be created with only a single queue. 385 */ 386 static bool qdisc_tx_is_default(const struct net_device *dev) 387 { 388 struct netdev_queue *txq; 389 struct Qdisc *qdisc; 390 391 if (dev->num_tx_queues > 1) 392 return false; 393 394 txq = netdev_get_tx_queue(dev, 0); 395 qdisc = rcu_access_pointer(txq->qdisc); 396 397 return !qdisc->enqueue; 398 } 399 400 /* Local traffic destined to local address. Reinsert the packet to rx 401 * path, similar to loopback handling. 402 */ 403 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 404 struct dst_entry *dst) 405 { 406 int len = skb->len; 407 408 skb_orphan(skb); 409 410 skb_dst_set(skb, dst); 411 412 /* set pkt_type to avoid skb hitting packet taps twice - 413 * once on Tx and again in Rx processing 414 */ 415 skb->pkt_type = PACKET_LOOPBACK; 416 417 skb->protocol = eth_type_trans(skb, dev); 418 419 if (likely(netif_rx(skb) == NET_RX_SUCCESS)) 420 vrf_rx_stats(dev, len); 421 else 422 this_cpu_inc(dev->dstats->rx_drps); 423 424 return NETDEV_TX_OK; 425 } 426 427 #if IS_ENABLED(CONFIG_IPV6) 428 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 429 struct sk_buff *skb) 430 { 431 int err; 432 433 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 434 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 435 436 if (likely(err == 1)) 437 err = dst_output(net, sk, skb); 438 439 return err; 440 } 441 442 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 443 struct net_device *dev) 444 { 445 const struct ipv6hdr *iph; 446 struct net *net = dev_net(skb->dev); 447 struct flowi6 fl6; 448 int ret = NET_XMIT_DROP; 449 struct dst_entry *dst; 450 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 451 452 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 453 goto err; 454 455 iph = ipv6_hdr(skb); 456 457 memset(&fl6, 0, sizeof(fl6)); 458 /* needed to match OIF rule */ 459 fl6.flowi6_oif = dev->ifindex; 460 fl6.flowi6_iif = LOOPBACK_IFINDEX; 461 fl6.daddr = iph->daddr; 462 fl6.saddr = iph->saddr; 463 fl6.flowlabel = ip6_flowinfo(iph); 464 fl6.flowi6_mark = skb->mark; 465 fl6.flowi6_proto = iph->nexthdr; 466 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF; 467 468 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 469 if (IS_ERR(dst) || dst == dst_null) 470 goto err; 471 472 skb_dst_drop(skb); 473 474 /* if dst.dev is loopback or the VRF device again this is locally 475 * originated traffic destined to a local address. Short circuit 476 * to Rx path 477 */ 478 if (dst->dev == dev) 479 return vrf_local_xmit(skb, dev, dst); 480 481 skb_dst_set(skb, dst); 482 483 /* strip the ethernet header added for pass through VRF device */ 484 __skb_pull(skb, skb_network_offset(skb)); 485 486 ret = vrf_ip6_local_out(net, skb->sk, skb); 487 if (unlikely(net_xmit_eval(ret))) 488 dev->stats.tx_errors++; 489 else 490 ret = NET_XMIT_SUCCESS; 491 492 return ret; 493 err: 494 vrf_tx_error(dev, skb); 495 return NET_XMIT_DROP; 496 } 497 #else 498 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 499 struct net_device *dev) 500 { 501 vrf_tx_error(dev, skb); 502 return NET_XMIT_DROP; 503 } 504 #endif 505 506 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 507 static int vrf_ip_local_out(struct net *net, struct sock *sk, 508 struct sk_buff *skb) 509 { 510 int err; 511 512 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 513 skb, NULL, skb_dst(skb)->dev, dst_output); 514 if (likely(err == 1)) 515 err = dst_output(net, sk, skb); 516 517 return err; 518 } 519 520 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 521 struct net_device *vrf_dev) 522 { 523 struct iphdr *ip4h; 524 int ret = NET_XMIT_DROP; 525 struct flowi4 fl4; 526 struct net *net = dev_net(vrf_dev); 527 struct rtable *rt; 528 529 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 530 goto err; 531 532 ip4h = ip_hdr(skb); 533 534 memset(&fl4, 0, sizeof(fl4)); 535 /* needed to match OIF rule */ 536 fl4.flowi4_oif = vrf_dev->ifindex; 537 fl4.flowi4_iif = LOOPBACK_IFINDEX; 538 fl4.flowi4_tos = RT_TOS(ip4h->tos); 539 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF; 540 fl4.flowi4_proto = ip4h->protocol; 541 fl4.daddr = ip4h->daddr; 542 fl4.saddr = ip4h->saddr; 543 544 rt = ip_route_output_flow(net, &fl4, NULL); 545 if (IS_ERR(rt)) 546 goto err; 547 548 skb_dst_drop(skb); 549 550 /* if dst.dev is loopback or the VRF device again this is locally 551 * originated traffic destined to a local address. Short circuit 552 * to Rx path 553 */ 554 if (rt->dst.dev == vrf_dev) 555 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 556 557 skb_dst_set(skb, &rt->dst); 558 559 /* strip the ethernet header added for pass through VRF device */ 560 __skb_pull(skb, skb_network_offset(skb)); 561 562 if (!ip4h->saddr) { 563 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 564 RT_SCOPE_LINK); 565 } 566 567 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 568 if (unlikely(net_xmit_eval(ret))) 569 vrf_dev->stats.tx_errors++; 570 else 571 ret = NET_XMIT_SUCCESS; 572 573 out: 574 return ret; 575 err: 576 vrf_tx_error(vrf_dev, skb); 577 goto out; 578 } 579 580 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 581 { 582 switch (skb->protocol) { 583 case htons(ETH_P_IP): 584 return vrf_process_v4_outbound(skb, dev); 585 case htons(ETH_P_IPV6): 586 return vrf_process_v6_outbound(skb, dev); 587 default: 588 vrf_tx_error(dev, skb); 589 return NET_XMIT_DROP; 590 } 591 } 592 593 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 594 { 595 int len = skb->len; 596 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 597 598 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 599 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 600 601 u64_stats_update_begin(&dstats->syncp); 602 dstats->tx_pkts++; 603 dstats->tx_bytes += len; 604 u64_stats_update_end(&dstats->syncp); 605 } else { 606 this_cpu_inc(dev->dstats->tx_drps); 607 } 608 609 return ret; 610 } 611 612 static void vrf_finish_direct(struct sk_buff *skb) 613 { 614 struct net_device *vrf_dev = skb->dev; 615 616 if (!list_empty(&vrf_dev->ptype_all) && 617 likely(skb_headroom(skb) >= ETH_HLEN)) { 618 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 619 620 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 621 eth_zero_addr(eth->h_dest); 622 eth->h_proto = skb->protocol; 623 624 rcu_read_lock_bh(); 625 dev_queue_xmit_nit(skb, vrf_dev); 626 rcu_read_unlock_bh(); 627 628 skb_pull(skb, ETH_HLEN); 629 } 630 631 /* reset skb device */ 632 nf_reset_ct(skb); 633 } 634 635 #if IS_ENABLED(CONFIG_IPV6) 636 /* modelled after ip6_finish_output2 */ 637 static int vrf_finish_output6(struct net *net, struct sock *sk, 638 struct sk_buff *skb) 639 { 640 struct dst_entry *dst = skb_dst(skb); 641 struct net_device *dev = dst->dev; 642 const struct in6_addr *nexthop; 643 struct neighbour *neigh; 644 int ret; 645 646 nf_reset_ct(skb); 647 648 skb->protocol = htons(ETH_P_IPV6); 649 skb->dev = dev; 650 651 rcu_read_lock_bh(); 652 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 653 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 654 if (unlikely(!neigh)) 655 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 656 if (!IS_ERR(neigh)) { 657 sock_confirm_neigh(skb, neigh); 658 ret = neigh_output(neigh, skb, false); 659 rcu_read_unlock_bh(); 660 return ret; 661 } 662 rcu_read_unlock_bh(); 663 664 IP6_INC_STATS(dev_net(dst->dev), 665 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 666 kfree_skb(skb); 667 return -EINVAL; 668 } 669 670 /* modelled after ip6_output */ 671 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 672 { 673 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 674 net, sk, skb, NULL, skb_dst(skb)->dev, 675 vrf_finish_output6, 676 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 677 } 678 679 /* set dst on skb to send packet to us via dev_xmit path. Allows 680 * packet to go through device based features such as qdisc, netfilter 681 * hooks and packet sockets with skb->dev set to vrf device. 682 */ 683 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 684 struct sk_buff *skb) 685 { 686 struct net_vrf *vrf = netdev_priv(vrf_dev); 687 struct dst_entry *dst = NULL; 688 struct rt6_info *rt6; 689 690 rcu_read_lock(); 691 692 rt6 = rcu_dereference(vrf->rt6); 693 if (likely(rt6)) { 694 dst = &rt6->dst; 695 dst_hold(dst); 696 } 697 698 rcu_read_unlock(); 699 700 if (unlikely(!dst)) { 701 vrf_tx_error(vrf_dev, skb); 702 return NULL; 703 } 704 705 skb_dst_drop(skb); 706 skb_dst_set(skb, dst); 707 708 return skb; 709 } 710 711 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 712 struct sk_buff *skb) 713 { 714 vrf_finish_direct(skb); 715 716 return vrf_ip6_local_out(net, sk, skb); 717 } 718 719 static int vrf_output6_direct(struct net *net, struct sock *sk, 720 struct sk_buff *skb) 721 { 722 int err = 1; 723 724 skb->protocol = htons(ETH_P_IPV6); 725 726 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 727 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 728 NULL, skb->dev, vrf_output6_direct_finish); 729 730 if (likely(err == 1)) 731 vrf_finish_direct(skb); 732 733 return err; 734 } 735 736 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 737 struct sk_buff *skb) 738 { 739 int err; 740 741 err = vrf_output6_direct(net, sk, skb); 742 if (likely(err == 1)) 743 err = vrf_ip6_local_out(net, sk, skb); 744 745 return err; 746 } 747 748 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 749 struct sock *sk, 750 struct sk_buff *skb) 751 { 752 struct net *net = dev_net(vrf_dev); 753 int err; 754 755 skb->dev = vrf_dev; 756 757 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 758 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 759 760 if (likely(err == 1)) 761 err = vrf_output6_direct(net, sk, skb); 762 763 if (likely(err == 1)) 764 return skb; 765 766 return NULL; 767 } 768 769 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 770 struct sock *sk, 771 struct sk_buff *skb) 772 { 773 /* don't divert link scope packets */ 774 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 775 return skb; 776 777 if (qdisc_tx_is_default(vrf_dev) || 778 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 779 return vrf_ip6_out_direct(vrf_dev, sk, skb); 780 781 return vrf_ip6_out_redirect(vrf_dev, skb); 782 } 783 784 /* holding rtnl */ 785 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 786 { 787 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 788 struct net *net = dev_net(dev); 789 struct dst_entry *dst; 790 791 RCU_INIT_POINTER(vrf->rt6, NULL); 792 synchronize_rcu(); 793 794 /* move dev in dst's to loopback so this VRF device can be deleted 795 * - based on dst_ifdown 796 */ 797 if (rt6) { 798 dst = &rt6->dst; 799 dev_put(dst->dev); 800 dst->dev = net->loopback_dev; 801 dev_hold(dst->dev); 802 dst_release(dst); 803 } 804 } 805 806 static int vrf_rt6_create(struct net_device *dev) 807 { 808 int flags = DST_NOPOLICY | DST_NOXFRM; 809 struct net_vrf *vrf = netdev_priv(dev); 810 struct net *net = dev_net(dev); 811 struct rt6_info *rt6; 812 int rc = -ENOMEM; 813 814 /* IPv6 can be CONFIG enabled and then disabled runtime */ 815 if (!ipv6_mod_enabled()) 816 return 0; 817 818 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 819 if (!vrf->fib6_table) 820 goto out; 821 822 /* create a dst for routing packets out a VRF device */ 823 rt6 = ip6_dst_alloc(net, dev, flags); 824 if (!rt6) 825 goto out; 826 827 rt6->dst.output = vrf_output6; 828 829 rcu_assign_pointer(vrf->rt6, rt6); 830 831 rc = 0; 832 out: 833 return rc; 834 } 835 #else 836 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 837 struct sock *sk, 838 struct sk_buff *skb) 839 { 840 return skb; 841 } 842 843 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 844 { 845 } 846 847 static int vrf_rt6_create(struct net_device *dev) 848 { 849 return 0; 850 } 851 #endif 852 853 /* modelled after ip_finish_output2 */ 854 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 855 { 856 struct dst_entry *dst = skb_dst(skb); 857 struct rtable *rt = (struct rtable *)dst; 858 struct net_device *dev = dst->dev; 859 unsigned int hh_len = LL_RESERVED_SPACE(dev); 860 struct neighbour *neigh; 861 bool is_v6gw = false; 862 int ret = -EINVAL; 863 864 nf_reset_ct(skb); 865 866 /* Be paranoid, rather than too clever. */ 867 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 868 struct sk_buff *skb2; 869 870 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 871 if (!skb2) { 872 ret = -ENOMEM; 873 goto err; 874 } 875 if (skb->sk) 876 skb_set_owner_w(skb2, skb->sk); 877 878 consume_skb(skb); 879 skb = skb2; 880 } 881 882 rcu_read_lock_bh(); 883 884 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 885 if (!IS_ERR(neigh)) { 886 sock_confirm_neigh(skb, neigh); 887 /* if crossing protocols, can not use the cached header */ 888 ret = neigh_output(neigh, skb, is_v6gw); 889 rcu_read_unlock_bh(); 890 return ret; 891 } 892 893 rcu_read_unlock_bh(); 894 err: 895 vrf_tx_error(skb->dev, skb); 896 return ret; 897 } 898 899 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 900 { 901 struct net_device *dev = skb_dst(skb)->dev; 902 903 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 904 905 skb->dev = dev; 906 skb->protocol = htons(ETH_P_IP); 907 908 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 909 net, sk, skb, NULL, dev, 910 vrf_finish_output, 911 !(IPCB(skb)->flags & IPSKB_REROUTED)); 912 } 913 914 /* set dst on skb to send packet to us via dev_xmit path. Allows 915 * packet to go through device based features such as qdisc, netfilter 916 * hooks and packet sockets with skb->dev set to vrf device. 917 */ 918 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 919 struct sk_buff *skb) 920 { 921 struct net_vrf *vrf = netdev_priv(vrf_dev); 922 struct dst_entry *dst = NULL; 923 struct rtable *rth; 924 925 rcu_read_lock(); 926 927 rth = rcu_dereference(vrf->rth); 928 if (likely(rth)) { 929 dst = &rth->dst; 930 dst_hold(dst); 931 } 932 933 rcu_read_unlock(); 934 935 if (unlikely(!dst)) { 936 vrf_tx_error(vrf_dev, skb); 937 return NULL; 938 } 939 940 skb_dst_drop(skb); 941 skb_dst_set(skb, dst); 942 943 return skb; 944 } 945 946 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 947 struct sk_buff *skb) 948 { 949 vrf_finish_direct(skb); 950 951 return vrf_ip_local_out(net, sk, skb); 952 } 953 954 static int vrf_output_direct(struct net *net, struct sock *sk, 955 struct sk_buff *skb) 956 { 957 int err = 1; 958 959 skb->protocol = htons(ETH_P_IP); 960 961 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 962 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 963 NULL, skb->dev, vrf_output_direct_finish); 964 965 if (likely(err == 1)) 966 vrf_finish_direct(skb); 967 968 return err; 969 } 970 971 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 972 struct sk_buff *skb) 973 { 974 int err; 975 976 err = vrf_output_direct(net, sk, skb); 977 if (likely(err == 1)) 978 err = vrf_ip_local_out(net, sk, skb); 979 980 return err; 981 } 982 983 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 984 struct sock *sk, 985 struct sk_buff *skb) 986 { 987 struct net *net = dev_net(vrf_dev); 988 int err; 989 990 skb->dev = vrf_dev; 991 992 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 993 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 994 995 if (likely(err == 1)) 996 err = vrf_output_direct(net, sk, skb); 997 998 if (likely(err == 1)) 999 return skb; 1000 1001 return NULL; 1002 } 1003 1004 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 1005 struct sock *sk, 1006 struct sk_buff *skb) 1007 { 1008 /* don't divert multicast or local broadcast */ 1009 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 1010 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 1011 return skb; 1012 1013 if (qdisc_tx_is_default(vrf_dev) || 1014 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 1015 return vrf_ip_out_direct(vrf_dev, sk, skb); 1016 1017 return vrf_ip_out_redirect(vrf_dev, skb); 1018 } 1019 1020 /* called with rcu lock held */ 1021 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1022 struct sock *sk, 1023 struct sk_buff *skb, 1024 u16 proto) 1025 { 1026 switch (proto) { 1027 case AF_INET: 1028 return vrf_ip_out(vrf_dev, sk, skb); 1029 case AF_INET6: 1030 return vrf_ip6_out(vrf_dev, sk, skb); 1031 } 1032 1033 return skb; 1034 } 1035 1036 /* holding rtnl */ 1037 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1038 { 1039 struct rtable *rth = rtnl_dereference(vrf->rth); 1040 struct net *net = dev_net(dev); 1041 struct dst_entry *dst; 1042 1043 RCU_INIT_POINTER(vrf->rth, NULL); 1044 synchronize_rcu(); 1045 1046 /* move dev in dst's to loopback so this VRF device can be deleted 1047 * - based on dst_ifdown 1048 */ 1049 if (rth) { 1050 dst = &rth->dst; 1051 dev_put(dst->dev); 1052 dst->dev = net->loopback_dev; 1053 dev_hold(dst->dev); 1054 dst_release(dst); 1055 } 1056 } 1057 1058 static int vrf_rtable_create(struct net_device *dev) 1059 { 1060 struct net_vrf *vrf = netdev_priv(dev); 1061 struct rtable *rth; 1062 1063 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1064 return -ENOMEM; 1065 1066 /* create a dst for routing packets out through a VRF device */ 1067 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1); 1068 if (!rth) 1069 return -ENOMEM; 1070 1071 rth->dst.output = vrf_output; 1072 1073 rcu_assign_pointer(vrf->rth, rth); 1074 1075 return 0; 1076 } 1077 1078 /**************************** device handling ********************/ 1079 1080 /* cycle interface to flush neighbor cache and move routes across tables */ 1081 static void cycle_netdev(struct net_device *dev, 1082 struct netlink_ext_ack *extack) 1083 { 1084 unsigned int flags = dev->flags; 1085 int ret; 1086 1087 if (!netif_running(dev)) 1088 return; 1089 1090 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1091 if (ret >= 0) 1092 ret = dev_change_flags(dev, flags, extack); 1093 1094 if (ret < 0) { 1095 netdev_err(dev, 1096 "Failed to cycle device %s; route tables might be wrong!\n", 1097 dev->name); 1098 } 1099 } 1100 1101 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1102 struct netlink_ext_ack *extack) 1103 { 1104 int ret; 1105 1106 /* do not allow loopback device to be enslaved to a VRF. 1107 * The vrf device acts as the loopback for the vrf. 1108 */ 1109 if (port_dev == dev_net(dev)->loopback_dev) { 1110 NL_SET_ERR_MSG(extack, 1111 "Can not enslave loopback device to a VRF"); 1112 return -EOPNOTSUPP; 1113 } 1114 1115 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1116 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1117 if (ret < 0) 1118 goto err; 1119 1120 cycle_netdev(port_dev, extack); 1121 1122 return 0; 1123 1124 err: 1125 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1126 return ret; 1127 } 1128 1129 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1130 struct netlink_ext_ack *extack) 1131 { 1132 if (netif_is_l3_master(port_dev)) { 1133 NL_SET_ERR_MSG(extack, 1134 "Can not enslave an L3 master device to a VRF"); 1135 return -EINVAL; 1136 } 1137 1138 if (netif_is_l3_slave(port_dev)) 1139 return -EINVAL; 1140 1141 return do_vrf_add_slave(dev, port_dev, extack); 1142 } 1143 1144 /* inverse of do_vrf_add_slave */ 1145 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1146 { 1147 netdev_upper_dev_unlink(port_dev, dev); 1148 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1149 1150 cycle_netdev(port_dev, NULL); 1151 1152 return 0; 1153 } 1154 1155 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1156 { 1157 return do_vrf_del_slave(dev, port_dev); 1158 } 1159 1160 static void vrf_dev_uninit(struct net_device *dev) 1161 { 1162 struct net_vrf *vrf = netdev_priv(dev); 1163 1164 vrf_rtable_release(dev, vrf); 1165 vrf_rt6_release(dev, vrf); 1166 1167 free_percpu(dev->dstats); 1168 dev->dstats = NULL; 1169 } 1170 1171 static int vrf_dev_init(struct net_device *dev) 1172 { 1173 struct net_vrf *vrf = netdev_priv(dev); 1174 1175 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 1176 if (!dev->dstats) 1177 goto out_nomem; 1178 1179 /* create the default dst which points back to us */ 1180 if (vrf_rtable_create(dev) != 0) 1181 goto out_stats; 1182 1183 if (vrf_rt6_create(dev) != 0) 1184 goto out_rth; 1185 1186 dev->flags = IFF_MASTER | IFF_NOARP; 1187 1188 /* MTU is irrelevant for VRF device; set to 64k similar to lo */ 1189 dev->mtu = 64 * 1024; 1190 1191 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1192 dev->operstate = IF_OPER_UP; 1193 netdev_lockdep_set_classes(dev); 1194 return 0; 1195 1196 out_rth: 1197 vrf_rtable_release(dev, vrf); 1198 out_stats: 1199 free_percpu(dev->dstats); 1200 dev->dstats = NULL; 1201 out_nomem: 1202 return -ENOMEM; 1203 } 1204 1205 static const struct net_device_ops vrf_netdev_ops = { 1206 .ndo_init = vrf_dev_init, 1207 .ndo_uninit = vrf_dev_uninit, 1208 .ndo_start_xmit = vrf_xmit, 1209 .ndo_set_mac_address = eth_mac_addr, 1210 .ndo_get_stats64 = vrf_get_stats64, 1211 .ndo_add_slave = vrf_add_slave, 1212 .ndo_del_slave = vrf_del_slave, 1213 }; 1214 1215 static u32 vrf_fib_table(const struct net_device *dev) 1216 { 1217 struct net_vrf *vrf = netdev_priv(dev); 1218 1219 return vrf->tb_id; 1220 } 1221 1222 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1223 { 1224 kfree_skb(skb); 1225 return 0; 1226 } 1227 1228 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1229 struct sk_buff *skb, 1230 struct net_device *dev) 1231 { 1232 struct net *net = dev_net(dev); 1233 1234 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1235 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1236 1237 return skb; 1238 } 1239 1240 #if IS_ENABLED(CONFIG_IPV6) 1241 /* neighbor handling is done with actual device; do not want 1242 * to flip skb->dev for those ndisc packets. This really fails 1243 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1244 * a start. 1245 */ 1246 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1247 { 1248 const struct ipv6hdr *iph = ipv6_hdr(skb); 1249 bool rc = false; 1250 1251 if (iph->nexthdr == NEXTHDR_ICMP) { 1252 const struct icmp6hdr *icmph; 1253 struct icmp6hdr _icmph; 1254 1255 icmph = skb_header_pointer(skb, sizeof(*iph), 1256 sizeof(_icmph), &_icmph); 1257 if (!icmph) 1258 goto out; 1259 1260 switch (icmph->icmp6_type) { 1261 case NDISC_ROUTER_SOLICITATION: 1262 case NDISC_ROUTER_ADVERTISEMENT: 1263 case NDISC_NEIGHBOUR_SOLICITATION: 1264 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1265 case NDISC_REDIRECT: 1266 rc = true; 1267 break; 1268 } 1269 } 1270 1271 out: 1272 return rc; 1273 } 1274 1275 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1276 const struct net_device *dev, 1277 struct flowi6 *fl6, 1278 int ifindex, 1279 const struct sk_buff *skb, 1280 int flags) 1281 { 1282 struct net_vrf *vrf = netdev_priv(dev); 1283 1284 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1285 } 1286 1287 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1288 int ifindex) 1289 { 1290 const struct ipv6hdr *iph = ipv6_hdr(skb); 1291 struct flowi6 fl6 = { 1292 .flowi6_iif = ifindex, 1293 .flowi6_mark = skb->mark, 1294 .flowi6_proto = iph->nexthdr, 1295 .daddr = iph->daddr, 1296 .saddr = iph->saddr, 1297 .flowlabel = ip6_flowinfo(iph), 1298 }; 1299 struct net *net = dev_net(vrf_dev); 1300 struct rt6_info *rt6; 1301 1302 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1303 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1304 if (unlikely(!rt6)) 1305 return; 1306 1307 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1308 return; 1309 1310 skb_dst_set(skb, &rt6->dst); 1311 } 1312 1313 static int vrf_prepare_mac_header(struct sk_buff *skb, 1314 struct net_device *vrf_dev, u16 proto) 1315 { 1316 struct ethhdr *eth; 1317 int err; 1318 1319 /* in general, we do not know if there is enough space in the head of 1320 * the packet for hosting the mac header. 1321 */ 1322 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1323 if (unlikely(err)) 1324 /* no space in the skb head */ 1325 return -ENOBUFS; 1326 1327 __skb_push(skb, ETH_HLEN); 1328 eth = (struct ethhdr *)skb->data; 1329 1330 skb_reset_mac_header(skb); 1331 1332 /* we set the ethernet destination and the source addresses to the 1333 * address of the VRF device. 1334 */ 1335 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1336 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1337 eth->h_proto = htons(proto); 1338 1339 /* the destination address of the Ethernet frame corresponds to the 1340 * address set on the VRF interface; therefore, the packet is intended 1341 * to be processed locally. 1342 */ 1343 skb->protocol = eth->h_proto; 1344 skb->pkt_type = PACKET_HOST; 1345 1346 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1347 1348 skb_pull_inline(skb, ETH_HLEN); 1349 1350 return 0; 1351 } 1352 1353 /* prepare and add the mac header to the packet if it was not set previously. 1354 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1355 * If the mac header was already set, the original mac header is left 1356 * untouched and the function returns immediately. 1357 */ 1358 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1359 struct net_device *vrf_dev, 1360 u16 proto) 1361 { 1362 if (skb_mac_header_was_set(skb)) 1363 return 0; 1364 1365 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1366 } 1367 1368 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1369 struct sk_buff *skb) 1370 { 1371 int orig_iif = skb->skb_iif; 1372 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1373 bool is_ndisc = ipv6_ndisc_frame(skb); 1374 1375 /* loopback, multicast & non-ND link-local traffic; do not push through 1376 * packet taps again. Reset pkt_type for upper layers to process skb 1377 */ 1378 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1379 skb->dev = vrf_dev; 1380 skb->skb_iif = vrf_dev->ifindex; 1381 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1382 if (skb->pkt_type == PACKET_LOOPBACK) 1383 skb->pkt_type = PACKET_HOST; 1384 goto out; 1385 } 1386 1387 /* if packet is NDISC then keep the ingress interface */ 1388 if (!is_ndisc) { 1389 vrf_rx_stats(vrf_dev, skb->len); 1390 skb->dev = vrf_dev; 1391 skb->skb_iif = vrf_dev->ifindex; 1392 1393 if (!list_empty(&vrf_dev->ptype_all)) { 1394 int err; 1395 1396 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1397 ETH_P_IPV6); 1398 if (likely(!err)) { 1399 skb_push(skb, skb->mac_len); 1400 dev_queue_xmit_nit(skb, vrf_dev); 1401 skb_pull(skb, skb->mac_len); 1402 } 1403 } 1404 1405 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1406 } 1407 1408 if (need_strict) 1409 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1410 1411 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1412 out: 1413 return skb; 1414 } 1415 1416 #else 1417 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1418 struct sk_buff *skb) 1419 { 1420 return skb; 1421 } 1422 #endif 1423 1424 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1425 struct sk_buff *skb) 1426 { 1427 skb->dev = vrf_dev; 1428 skb->skb_iif = vrf_dev->ifindex; 1429 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1430 1431 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1432 goto out; 1433 1434 /* loopback traffic; do not push through packet taps again. 1435 * Reset pkt_type for upper layers to process skb 1436 */ 1437 if (skb->pkt_type == PACKET_LOOPBACK) { 1438 skb->pkt_type = PACKET_HOST; 1439 goto out; 1440 } 1441 1442 vrf_rx_stats(vrf_dev, skb->len); 1443 1444 if (!list_empty(&vrf_dev->ptype_all)) { 1445 int err; 1446 1447 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP); 1448 if (likely(!err)) { 1449 skb_push(skb, skb->mac_len); 1450 dev_queue_xmit_nit(skb, vrf_dev); 1451 skb_pull(skb, skb->mac_len); 1452 } 1453 } 1454 1455 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1456 out: 1457 return skb; 1458 } 1459 1460 /* called with rcu lock held */ 1461 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1462 struct sk_buff *skb, 1463 u16 proto) 1464 { 1465 switch (proto) { 1466 case AF_INET: 1467 return vrf_ip_rcv(vrf_dev, skb); 1468 case AF_INET6: 1469 return vrf_ip6_rcv(vrf_dev, skb); 1470 } 1471 1472 return skb; 1473 } 1474 1475 #if IS_ENABLED(CONFIG_IPV6) 1476 /* send to link-local or multicast address via interface enslaved to 1477 * VRF device. Force lookup to VRF table without changing flow struct 1478 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1479 * is taken on the dst by this function. 1480 */ 1481 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1482 struct flowi6 *fl6) 1483 { 1484 struct net *net = dev_net(dev); 1485 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1486 struct dst_entry *dst = NULL; 1487 struct rt6_info *rt; 1488 1489 /* VRF device does not have a link-local address and 1490 * sending packets to link-local or mcast addresses over 1491 * a VRF device does not make sense 1492 */ 1493 if (fl6->flowi6_oif == dev->ifindex) { 1494 dst = &net->ipv6.ip6_null_entry->dst; 1495 return dst; 1496 } 1497 1498 if (!ipv6_addr_any(&fl6->saddr)) 1499 flags |= RT6_LOOKUP_F_HAS_SADDR; 1500 1501 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1502 if (rt) 1503 dst = &rt->dst; 1504 1505 return dst; 1506 } 1507 #endif 1508 1509 static const struct l3mdev_ops vrf_l3mdev_ops = { 1510 .l3mdev_fib_table = vrf_fib_table, 1511 .l3mdev_l3_rcv = vrf_l3_rcv, 1512 .l3mdev_l3_out = vrf_l3_out, 1513 #if IS_ENABLED(CONFIG_IPV6) 1514 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1515 #endif 1516 }; 1517 1518 static void vrf_get_drvinfo(struct net_device *dev, 1519 struct ethtool_drvinfo *info) 1520 { 1521 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1522 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1523 } 1524 1525 static const struct ethtool_ops vrf_ethtool_ops = { 1526 .get_drvinfo = vrf_get_drvinfo, 1527 }; 1528 1529 static inline size_t vrf_fib_rule_nl_size(void) 1530 { 1531 size_t sz; 1532 1533 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1534 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1535 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1536 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1537 1538 return sz; 1539 } 1540 1541 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1542 { 1543 struct fib_rule_hdr *frh; 1544 struct nlmsghdr *nlh; 1545 struct sk_buff *skb; 1546 int err; 1547 1548 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1549 !ipv6_mod_enabled()) 1550 return 0; 1551 1552 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1553 if (!skb) 1554 return -ENOMEM; 1555 1556 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1557 if (!nlh) 1558 goto nla_put_failure; 1559 1560 /* rule only needs to appear once */ 1561 nlh->nlmsg_flags |= NLM_F_EXCL; 1562 1563 frh = nlmsg_data(nlh); 1564 memset(frh, 0, sizeof(*frh)); 1565 frh->family = family; 1566 frh->action = FR_ACT_TO_TBL; 1567 1568 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1569 goto nla_put_failure; 1570 1571 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1572 goto nla_put_failure; 1573 1574 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1575 goto nla_put_failure; 1576 1577 nlmsg_end(skb, nlh); 1578 1579 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1580 skb->sk = dev_net(dev)->rtnl; 1581 if (add_it) { 1582 err = fib_nl_newrule(skb, nlh, NULL); 1583 if (err == -EEXIST) 1584 err = 0; 1585 } else { 1586 err = fib_nl_delrule(skb, nlh, NULL); 1587 if (err == -ENOENT) 1588 err = 0; 1589 } 1590 nlmsg_free(skb); 1591 1592 return err; 1593 1594 nla_put_failure: 1595 nlmsg_free(skb); 1596 1597 return -EMSGSIZE; 1598 } 1599 1600 static int vrf_add_fib_rules(const struct net_device *dev) 1601 { 1602 int err; 1603 1604 err = vrf_fib_rule(dev, AF_INET, true); 1605 if (err < 0) 1606 goto out_err; 1607 1608 err = vrf_fib_rule(dev, AF_INET6, true); 1609 if (err < 0) 1610 goto ipv6_err; 1611 1612 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1613 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1614 if (err < 0) 1615 goto ipmr_err; 1616 #endif 1617 1618 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1619 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1620 if (err < 0) 1621 goto ip6mr_err; 1622 #endif 1623 1624 return 0; 1625 1626 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1627 ip6mr_err: 1628 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1629 #endif 1630 1631 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1632 ipmr_err: 1633 vrf_fib_rule(dev, AF_INET6, false); 1634 #endif 1635 1636 ipv6_err: 1637 vrf_fib_rule(dev, AF_INET, false); 1638 1639 out_err: 1640 netdev_err(dev, "Failed to add FIB rules.\n"); 1641 return err; 1642 } 1643 1644 static void vrf_setup(struct net_device *dev) 1645 { 1646 ether_setup(dev); 1647 1648 /* Initialize the device structure. */ 1649 dev->netdev_ops = &vrf_netdev_ops; 1650 dev->l3mdev_ops = &vrf_l3mdev_ops; 1651 dev->ethtool_ops = &vrf_ethtool_ops; 1652 dev->needs_free_netdev = true; 1653 1654 /* Fill in device structure with ethernet-generic values. */ 1655 eth_hw_addr_random(dev); 1656 1657 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1658 dev->features |= NETIF_F_LLTX; 1659 1660 /* don't allow vrf devices to change network namespaces. */ 1661 dev->features |= NETIF_F_NETNS_LOCAL; 1662 1663 /* does not make sense for a VLAN to be added to a vrf device */ 1664 dev->features |= NETIF_F_VLAN_CHALLENGED; 1665 1666 /* enable offload features */ 1667 dev->features |= NETIF_F_GSO_SOFTWARE; 1668 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1669 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1670 1671 dev->hw_features = dev->features; 1672 dev->hw_enc_features = dev->features; 1673 1674 /* default to no qdisc; user can add if desired */ 1675 dev->priv_flags |= IFF_NO_QUEUE; 1676 dev->priv_flags |= IFF_NO_RX_HANDLER; 1677 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1678 1679 /* VRF devices do not care about MTU, but if the MTU is set 1680 * too low then the ipv4 and ipv6 protocols are disabled 1681 * which breaks networking. 1682 */ 1683 dev->min_mtu = IPV6_MIN_MTU; 1684 dev->max_mtu = ETH_MAX_MTU; 1685 } 1686 1687 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1688 struct netlink_ext_ack *extack) 1689 { 1690 if (tb[IFLA_ADDRESS]) { 1691 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1692 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1693 return -EINVAL; 1694 } 1695 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1696 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1697 return -EADDRNOTAVAIL; 1698 } 1699 } 1700 return 0; 1701 } 1702 1703 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1704 { 1705 struct net_device *port_dev; 1706 struct list_head *iter; 1707 1708 netdev_for_each_lower_dev(dev, port_dev, iter) 1709 vrf_del_slave(dev, port_dev); 1710 1711 vrf_map_unregister_dev(dev); 1712 1713 unregister_netdevice_queue(dev, head); 1714 } 1715 1716 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1717 struct nlattr *tb[], struct nlattr *data[], 1718 struct netlink_ext_ack *extack) 1719 { 1720 struct net_vrf *vrf = netdev_priv(dev); 1721 struct netns_vrf *nn_vrf; 1722 bool *add_fib_rules; 1723 struct net *net; 1724 int err; 1725 1726 if (!data || !data[IFLA_VRF_TABLE]) { 1727 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1728 return -EINVAL; 1729 } 1730 1731 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1732 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1733 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1734 "Invalid VRF table id"); 1735 return -EINVAL; 1736 } 1737 1738 dev->priv_flags |= IFF_L3MDEV_MASTER; 1739 1740 err = register_netdevice(dev); 1741 if (err) 1742 goto out; 1743 1744 /* mapping between table_id and vrf; 1745 * note: such binding could not be done in the dev init function 1746 * because dev->ifindex id is not available yet. 1747 */ 1748 vrf->ifindex = dev->ifindex; 1749 1750 err = vrf_map_register_dev(dev, extack); 1751 if (err) { 1752 unregister_netdevice(dev); 1753 goto out; 1754 } 1755 1756 net = dev_net(dev); 1757 nn_vrf = net_generic(net, vrf_net_id); 1758 1759 add_fib_rules = &nn_vrf->add_fib_rules; 1760 if (*add_fib_rules) { 1761 err = vrf_add_fib_rules(dev); 1762 if (err) { 1763 vrf_map_unregister_dev(dev); 1764 unregister_netdevice(dev); 1765 goto out; 1766 } 1767 *add_fib_rules = false; 1768 } 1769 1770 out: 1771 return err; 1772 } 1773 1774 static size_t vrf_nl_getsize(const struct net_device *dev) 1775 { 1776 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1777 } 1778 1779 static int vrf_fillinfo(struct sk_buff *skb, 1780 const struct net_device *dev) 1781 { 1782 struct net_vrf *vrf = netdev_priv(dev); 1783 1784 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1785 } 1786 1787 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1788 const struct net_device *slave_dev) 1789 { 1790 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1791 } 1792 1793 static int vrf_fill_slave_info(struct sk_buff *skb, 1794 const struct net_device *vrf_dev, 1795 const struct net_device *slave_dev) 1796 { 1797 struct net_vrf *vrf = netdev_priv(vrf_dev); 1798 1799 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1800 return -EMSGSIZE; 1801 1802 return 0; 1803 } 1804 1805 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1806 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1807 }; 1808 1809 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1810 .kind = DRV_NAME, 1811 .priv_size = sizeof(struct net_vrf), 1812 1813 .get_size = vrf_nl_getsize, 1814 .policy = vrf_nl_policy, 1815 .validate = vrf_validate, 1816 .fill_info = vrf_fillinfo, 1817 1818 .get_slave_size = vrf_get_slave_size, 1819 .fill_slave_info = vrf_fill_slave_info, 1820 1821 .newlink = vrf_newlink, 1822 .dellink = vrf_dellink, 1823 .setup = vrf_setup, 1824 .maxtype = IFLA_VRF_MAX, 1825 }; 1826 1827 static int vrf_device_event(struct notifier_block *unused, 1828 unsigned long event, void *ptr) 1829 { 1830 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1831 1832 /* only care about unregister events to drop slave references */ 1833 if (event == NETDEV_UNREGISTER) { 1834 struct net_device *vrf_dev; 1835 1836 if (!netif_is_l3_slave(dev)) 1837 goto out; 1838 1839 vrf_dev = netdev_master_upper_dev_get(dev); 1840 vrf_del_slave(vrf_dev, dev); 1841 } 1842 out: 1843 return NOTIFY_DONE; 1844 } 1845 1846 static struct notifier_block vrf_notifier_block __read_mostly = { 1847 .notifier_call = vrf_device_event, 1848 }; 1849 1850 static int vrf_map_init(struct vrf_map *vmap) 1851 { 1852 spin_lock_init(&vmap->vmap_lock); 1853 hash_init(vmap->ht); 1854 1855 vmap->strict_mode = false; 1856 1857 return 0; 1858 } 1859 1860 #ifdef CONFIG_SYSCTL 1861 static bool vrf_strict_mode(struct vrf_map *vmap) 1862 { 1863 bool strict_mode; 1864 1865 vrf_map_lock(vmap); 1866 strict_mode = vmap->strict_mode; 1867 vrf_map_unlock(vmap); 1868 1869 return strict_mode; 1870 } 1871 1872 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1873 { 1874 bool *cur_mode; 1875 int res = 0; 1876 1877 vrf_map_lock(vmap); 1878 1879 cur_mode = &vmap->strict_mode; 1880 if (*cur_mode == new_mode) 1881 goto unlock; 1882 1883 if (*cur_mode) { 1884 /* disable strict mode */ 1885 *cur_mode = false; 1886 } else { 1887 if (vmap->shared_tables) { 1888 /* we cannot allow strict_mode because there are some 1889 * vrfs that share one or more tables. 1890 */ 1891 res = -EBUSY; 1892 goto unlock; 1893 } 1894 1895 /* no tables are shared among vrfs, so we can go back 1896 * to 1:1 association between a vrf with its table. 1897 */ 1898 *cur_mode = true; 1899 } 1900 1901 unlock: 1902 vrf_map_unlock(vmap); 1903 1904 return res; 1905 } 1906 1907 static int vrf_shared_table_handler(struct ctl_table *table, int write, 1908 void *buffer, size_t *lenp, loff_t *ppos) 1909 { 1910 struct net *net = (struct net *)table->extra1; 1911 struct vrf_map *vmap = netns_vrf_map(net); 1912 int proc_strict_mode = 0; 1913 struct ctl_table tmp = { 1914 .procname = table->procname, 1915 .data = &proc_strict_mode, 1916 .maxlen = sizeof(int), 1917 .mode = table->mode, 1918 .extra1 = SYSCTL_ZERO, 1919 .extra2 = SYSCTL_ONE, 1920 }; 1921 int ret; 1922 1923 if (!write) 1924 proc_strict_mode = vrf_strict_mode(vmap); 1925 1926 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1927 1928 if (write && ret == 0) 1929 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1930 1931 return ret; 1932 } 1933 1934 static const struct ctl_table vrf_table[] = { 1935 { 1936 .procname = "strict_mode", 1937 .data = NULL, 1938 .maxlen = sizeof(int), 1939 .mode = 0644, 1940 .proc_handler = vrf_shared_table_handler, 1941 /* set by the vrf_netns_init */ 1942 .extra1 = NULL, 1943 }, 1944 { }, 1945 }; 1946 1947 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1948 { 1949 struct ctl_table *table; 1950 1951 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1952 if (!table) 1953 return -ENOMEM; 1954 1955 /* init the extra1 parameter with the reference to current netns */ 1956 table[0].extra1 = net; 1957 1958 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table); 1959 if (!nn_vrf->ctl_hdr) { 1960 kfree(table); 1961 return -ENOMEM; 1962 } 1963 1964 return 0; 1965 } 1966 1967 static void vrf_netns_exit_sysctl(struct net *net) 1968 { 1969 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1970 struct ctl_table *table; 1971 1972 table = nn_vrf->ctl_hdr->ctl_table_arg; 1973 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1974 kfree(table); 1975 } 1976 #else 1977 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1978 { 1979 return 0; 1980 } 1981 1982 static void vrf_netns_exit_sysctl(struct net *net) 1983 { 1984 } 1985 #endif 1986 1987 /* Initialize per network namespace state */ 1988 static int __net_init vrf_netns_init(struct net *net) 1989 { 1990 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1991 1992 nn_vrf->add_fib_rules = true; 1993 vrf_map_init(&nn_vrf->vmap); 1994 1995 return vrf_netns_init_sysctl(net, nn_vrf); 1996 } 1997 1998 static void __net_exit vrf_netns_exit(struct net *net) 1999 { 2000 vrf_netns_exit_sysctl(net); 2001 } 2002 2003 static struct pernet_operations vrf_net_ops __net_initdata = { 2004 .init = vrf_netns_init, 2005 .exit = vrf_netns_exit, 2006 .id = &vrf_net_id, 2007 .size = sizeof(struct netns_vrf), 2008 }; 2009 2010 static int __init vrf_init_module(void) 2011 { 2012 int rc; 2013 2014 register_netdevice_notifier(&vrf_notifier_block); 2015 2016 rc = register_pernet_subsys(&vrf_net_ops); 2017 if (rc < 0) 2018 goto error; 2019 2020 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2021 vrf_ifindex_lookup_by_table_id); 2022 if (rc < 0) 2023 goto unreg_pernet; 2024 2025 rc = rtnl_link_register(&vrf_link_ops); 2026 if (rc < 0) 2027 goto table_lookup_unreg; 2028 2029 return 0; 2030 2031 table_lookup_unreg: 2032 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2033 vrf_ifindex_lookup_by_table_id); 2034 2035 unreg_pernet: 2036 unregister_pernet_subsys(&vrf_net_ops); 2037 2038 error: 2039 unregister_netdevice_notifier(&vrf_notifier_block); 2040 return rc; 2041 } 2042 2043 module_init(vrf_init_module); 2044 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2045 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2046 MODULE_LICENSE("GPL"); 2047 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2048 MODULE_VERSION(DRV_VERSION); 2049