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 possible). 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 the VRF device again this is locally originated traffic 475 * destined to a local address. Short circuit to Rx path. 476 */ 477 if (dst->dev == dev) 478 return vrf_local_xmit(skb, dev, dst); 479 480 skb_dst_set(skb, dst); 481 482 /* strip the ethernet header added for pass through VRF device */ 483 __skb_pull(skb, skb_network_offset(skb)); 484 485 ret = vrf_ip6_local_out(net, skb->sk, skb); 486 if (unlikely(net_xmit_eval(ret))) 487 dev->stats.tx_errors++; 488 else 489 ret = NET_XMIT_SUCCESS; 490 491 return ret; 492 err: 493 vrf_tx_error(dev, skb); 494 return NET_XMIT_DROP; 495 } 496 #else 497 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 498 struct net_device *dev) 499 { 500 vrf_tx_error(dev, skb); 501 return NET_XMIT_DROP; 502 } 503 #endif 504 505 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 506 static int vrf_ip_local_out(struct net *net, struct sock *sk, 507 struct sk_buff *skb) 508 { 509 int err; 510 511 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 512 skb, NULL, skb_dst(skb)->dev, dst_output); 513 if (likely(err == 1)) 514 err = dst_output(net, sk, skb); 515 516 return err; 517 } 518 519 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 520 struct net_device *vrf_dev) 521 { 522 struct iphdr *ip4h; 523 int ret = NET_XMIT_DROP; 524 struct flowi4 fl4; 525 struct net *net = dev_net(vrf_dev); 526 struct rtable *rt; 527 528 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 529 goto err; 530 531 ip4h = ip_hdr(skb); 532 533 memset(&fl4, 0, sizeof(fl4)); 534 /* needed to match OIF rule */ 535 fl4.flowi4_oif = vrf_dev->ifindex; 536 fl4.flowi4_iif = LOOPBACK_IFINDEX; 537 fl4.flowi4_tos = RT_TOS(ip4h->tos); 538 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF; 539 fl4.flowi4_proto = ip4h->protocol; 540 fl4.daddr = ip4h->daddr; 541 fl4.saddr = ip4h->saddr; 542 543 rt = ip_route_output_flow(net, &fl4, NULL); 544 if (IS_ERR(rt)) 545 goto err; 546 547 skb_dst_drop(skb); 548 549 /* if dst.dev is the VRF device again this is locally originated traffic 550 * destined to a local address. Short circuit to Rx path. 551 */ 552 if (rt->dst.dev == vrf_dev) 553 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 554 555 skb_dst_set(skb, &rt->dst); 556 557 /* strip the ethernet header added for pass through VRF device */ 558 __skb_pull(skb, skb_network_offset(skb)); 559 560 if (!ip4h->saddr) { 561 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 562 RT_SCOPE_LINK); 563 } 564 565 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 566 if (unlikely(net_xmit_eval(ret))) 567 vrf_dev->stats.tx_errors++; 568 else 569 ret = NET_XMIT_SUCCESS; 570 571 out: 572 return ret; 573 err: 574 vrf_tx_error(vrf_dev, skb); 575 goto out; 576 } 577 578 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 579 { 580 switch (skb->protocol) { 581 case htons(ETH_P_IP): 582 return vrf_process_v4_outbound(skb, dev); 583 case htons(ETH_P_IPV6): 584 return vrf_process_v6_outbound(skb, dev); 585 default: 586 vrf_tx_error(dev, skb); 587 return NET_XMIT_DROP; 588 } 589 } 590 591 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 592 { 593 int len = skb->len; 594 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 595 596 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 597 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 598 599 u64_stats_update_begin(&dstats->syncp); 600 dstats->tx_pkts++; 601 dstats->tx_bytes += len; 602 u64_stats_update_end(&dstats->syncp); 603 } else { 604 this_cpu_inc(dev->dstats->tx_drps); 605 } 606 607 return ret; 608 } 609 610 static void vrf_finish_direct(struct sk_buff *skb) 611 { 612 struct net_device *vrf_dev = skb->dev; 613 614 if (!list_empty(&vrf_dev->ptype_all) && 615 likely(skb_headroom(skb) >= ETH_HLEN)) { 616 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 617 618 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 619 eth_zero_addr(eth->h_dest); 620 eth->h_proto = skb->protocol; 621 622 rcu_read_lock_bh(); 623 dev_queue_xmit_nit(skb, vrf_dev); 624 rcu_read_unlock_bh(); 625 626 skb_pull(skb, ETH_HLEN); 627 } 628 629 /* reset skb device */ 630 nf_reset_ct(skb); 631 } 632 633 #if IS_ENABLED(CONFIG_IPV6) 634 /* modelled after ip6_finish_output2 */ 635 static int vrf_finish_output6(struct net *net, struct sock *sk, 636 struct sk_buff *skb) 637 { 638 struct dst_entry *dst = skb_dst(skb); 639 struct net_device *dev = dst->dev; 640 const struct in6_addr *nexthop; 641 struct neighbour *neigh; 642 int ret; 643 644 nf_reset_ct(skb); 645 646 skb->protocol = htons(ETH_P_IPV6); 647 skb->dev = dev; 648 649 rcu_read_lock_bh(); 650 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 651 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 652 if (unlikely(!neigh)) 653 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 654 if (!IS_ERR(neigh)) { 655 sock_confirm_neigh(skb, neigh); 656 ret = neigh_output(neigh, skb, false); 657 rcu_read_unlock_bh(); 658 return ret; 659 } 660 rcu_read_unlock_bh(); 661 662 IP6_INC_STATS(dev_net(dst->dev), 663 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 664 kfree_skb(skb); 665 return -EINVAL; 666 } 667 668 /* modelled after ip6_output */ 669 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 670 { 671 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 672 net, sk, skb, NULL, skb_dst(skb)->dev, 673 vrf_finish_output6, 674 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 675 } 676 677 /* set dst on skb to send packet to us via dev_xmit path. Allows 678 * packet to go through device based features such as qdisc, netfilter 679 * hooks and packet sockets with skb->dev set to vrf device. 680 */ 681 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 682 struct sk_buff *skb) 683 { 684 struct net_vrf *vrf = netdev_priv(vrf_dev); 685 struct dst_entry *dst = NULL; 686 struct rt6_info *rt6; 687 688 rcu_read_lock(); 689 690 rt6 = rcu_dereference(vrf->rt6); 691 if (likely(rt6)) { 692 dst = &rt6->dst; 693 dst_hold(dst); 694 } 695 696 rcu_read_unlock(); 697 698 if (unlikely(!dst)) { 699 vrf_tx_error(vrf_dev, skb); 700 return NULL; 701 } 702 703 skb_dst_drop(skb); 704 skb_dst_set(skb, dst); 705 706 return skb; 707 } 708 709 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 710 struct sk_buff *skb) 711 { 712 vrf_finish_direct(skb); 713 714 return vrf_ip6_local_out(net, sk, skb); 715 } 716 717 static int vrf_output6_direct(struct net *net, struct sock *sk, 718 struct sk_buff *skb) 719 { 720 int err = 1; 721 722 skb->protocol = htons(ETH_P_IPV6); 723 724 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 725 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 726 NULL, skb->dev, vrf_output6_direct_finish); 727 728 if (likely(err == 1)) 729 vrf_finish_direct(skb); 730 731 return err; 732 } 733 734 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 735 struct sk_buff *skb) 736 { 737 int err; 738 739 err = vrf_output6_direct(net, sk, skb); 740 if (likely(err == 1)) 741 err = vrf_ip6_local_out(net, sk, skb); 742 743 return err; 744 } 745 746 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 747 struct sock *sk, 748 struct sk_buff *skb) 749 { 750 struct net *net = dev_net(vrf_dev); 751 int err; 752 753 skb->dev = vrf_dev; 754 755 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 756 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 757 758 if (likely(err == 1)) 759 err = vrf_output6_direct(net, sk, skb); 760 761 if (likely(err == 1)) 762 return skb; 763 764 return NULL; 765 } 766 767 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 768 struct sock *sk, 769 struct sk_buff *skb) 770 { 771 /* don't divert link scope packets */ 772 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 773 return skb; 774 775 if (qdisc_tx_is_default(vrf_dev) || 776 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 777 return vrf_ip6_out_direct(vrf_dev, sk, skb); 778 779 return vrf_ip6_out_redirect(vrf_dev, skb); 780 } 781 782 /* holding rtnl */ 783 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 784 { 785 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 786 struct net *net = dev_net(dev); 787 struct dst_entry *dst; 788 789 RCU_INIT_POINTER(vrf->rt6, NULL); 790 synchronize_rcu(); 791 792 /* move dev in dst's to loopback so this VRF device can be deleted 793 * - based on dst_ifdown 794 */ 795 if (rt6) { 796 dst = &rt6->dst; 797 dev_put(dst->dev); 798 dst->dev = net->loopback_dev; 799 dev_hold(dst->dev); 800 dst_release(dst); 801 } 802 } 803 804 static int vrf_rt6_create(struct net_device *dev) 805 { 806 int flags = DST_NOPOLICY | DST_NOXFRM; 807 struct net_vrf *vrf = netdev_priv(dev); 808 struct net *net = dev_net(dev); 809 struct rt6_info *rt6; 810 int rc = -ENOMEM; 811 812 /* IPv6 can be CONFIG enabled and then disabled runtime */ 813 if (!ipv6_mod_enabled()) 814 return 0; 815 816 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 817 if (!vrf->fib6_table) 818 goto out; 819 820 /* create a dst for routing packets out a VRF device */ 821 rt6 = ip6_dst_alloc(net, dev, flags); 822 if (!rt6) 823 goto out; 824 825 rt6->dst.output = vrf_output6; 826 827 rcu_assign_pointer(vrf->rt6, rt6); 828 829 rc = 0; 830 out: 831 return rc; 832 } 833 #else 834 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 835 struct sock *sk, 836 struct sk_buff *skb) 837 { 838 return skb; 839 } 840 841 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 842 { 843 } 844 845 static int vrf_rt6_create(struct net_device *dev) 846 { 847 return 0; 848 } 849 #endif 850 851 /* modelled after ip_finish_output2 */ 852 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 853 { 854 struct dst_entry *dst = skb_dst(skb); 855 struct rtable *rt = (struct rtable *)dst; 856 struct net_device *dev = dst->dev; 857 unsigned int hh_len = LL_RESERVED_SPACE(dev); 858 struct neighbour *neigh; 859 bool is_v6gw = false; 860 int ret = -EINVAL; 861 862 nf_reset_ct(skb); 863 864 /* Be paranoid, rather than too clever. */ 865 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 866 struct sk_buff *skb2; 867 868 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 869 if (!skb2) { 870 ret = -ENOMEM; 871 goto err; 872 } 873 if (skb->sk) 874 skb_set_owner_w(skb2, skb->sk); 875 876 consume_skb(skb); 877 skb = skb2; 878 } 879 880 rcu_read_lock_bh(); 881 882 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 883 if (!IS_ERR(neigh)) { 884 sock_confirm_neigh(skb, neigh); 885 /* if crossing protocols, can not use the cached header */ 886 ret = neigh_output(neigh, skb, is_v6gw); 887 rcu_read_unlock_bh(); 888 return ret; 889 } 890 891 rcu_read_unlock_bh(); 892 err: 893 vrf_tx_error(skb->dev, skb); 894 return ret; 895 } 896 897 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 898 { 899 struct net_device *dev = skb_dst(skb)->dev; 900 901 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 902 903 skb->dev = dev; 904 skb->protocol = htons(ETH_P_IP); 905 906 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 907 net, sk, skb, NULL, dev, 908 vrf_finish_output, 909 !(IPCB(skb)->flags & IPSKB_REROUTED)); 910 } 911 912 /* set dst on skb to send packet to us via dev_xmit path. Allows 913 * packet to go through device based features such as qdisc, netfilter 914 * hooks and packet sockets with skb->dev set to vrf device. 915 */ 916 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 917 struct sk_buff *skb) 918 { 919 struct net_vrf *vrf = netdev_priv(vrf_dev); 920 struct dst_entry *dst = NULL; 921 struct rtable *rth; 922 923 rcu_read_lock(); 924 925 rth = rcu_dereference(vrf->rth); 926 if (likely(rth)) { 927 dst = &rth->dst; 928 dst_hold(dst); 929 } 930 931 rcu_read_unlock(); 932 933 if (unlikely(!dst)) { 934 vrf_tx_error(vrf_dev, skb); 935 return NULL; 936 } 937 938 skb_dst_drop(skb); 939 skb_dst_set(skb, dst); 940 941 return skb; 942 } 943 944 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 945 struct sk_buff *skb) 946 { 947 vrf_finish_direct(skb); 948 949 return vrf_ip_local_out(net, sk, skb); 950 } 951 952 static int vrf_output_direct(struct net *net, struct sock *sk, 953 struct sk_buff *skb) 954 { 955 int err = 1; 956 957 skb->protocol = htons(ETH_P_IP); 958 959 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 960 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 961 NULL, skb->dev, vrf_output_direct_finish); 962 963 if (likely(err == 1)) 964 vrf_finish_direct(skb); 965 966 return err; 967 } 968 969 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 970 struct sk_buff *skb) 971 { 972 int err; 973 974 err = vrf_output_direct(net, sk, skb); 975 if (likely(err == 1)) 976 err = vrf_ip_local_out(net, sk, skb); 977 978 return err; 979 } 980 981 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 982 struct sock *sk, 983 struct sk_buff *skb) 984 { 985 struct net *net = dev_net(vrf_dev); 986 int err; 987 988 skb->dev = vrf_dev; 989 990 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 991 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 992 993 if (likely(err == 1)) 994 err = vrf_output_direct(net, sk, skb); 995 996 if (likely(err == 1)) 997 return skb; 998 999 return NULL; 1000 } 1001 1002 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 1003 struct sock *sk, 1004 struct sk_buff *skb) 1005 { 1006 /* don't divert multicast or local broadcast */ 1007 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 1008 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 1009 return skb; 1010 1011 if (qdisc_tx_is_default(vrf_dev) || 1012 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 1013 return vrf_ip_out_direct(vrf_dev, sk, skb); 1014 1015 return vrf_ip_out_redirect(vrf_dev, skb); 1016 } 1017 1018 /* called with rcu lock held */ 1019 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1020 struct sock *sk, 1021 struct sk_buff *skb, 1022 u16 proto) 1023 { 1024 switch (proto) { 1025 case AF_INET: 1026 return vrf_ip_out(vrf_dev, sk, skb); 1027 case AF_INET6: 1028 return vrf_ip6_out(vrf_dev, sk, skb); 1029 } 1030 1031 return skb; 1032 } 1033 1034 /* holding rtnl */ 1035 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1036 { 1037 struct rtable *rth = rtnl_dereference(vrf->rth); 1038 struct net *net = dev_net(dev); 1039 struct dst_entry *dst; 1040 1041 RCU_INIT_POINTER(vrf->rth, NULL); 1042 synchronize_rcu(); 1043 1044 /* move dev in dst's to loopback so this VRF device can be deleted 1045 * - based on dst_ifdown 1046 */ 1047 if (rth) { 1048 dst = &rth->dst; 1049 dev_put(dst->dev); 1050 dst->dev = net->loopback_dev; 1051 dev_hold(dst->dev); 1052 dst_release(dst); 1053 } 1054 } 1055 1056 static int vrf_rtable_create(struct net_device *dev) 1057 { 1058 struct net_vrf *vrf = netdev_priv(dev); 1059 struct rtable *rth; 1060 1061 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1062 return -ENOMEM; 1063 1064 /* create a dst for routing packets out through a VRF device */ 1065 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1); 1066 if (!rth) 1067 return -ENOMEM; 1068 1069 rth->dst.output = vrf_output; 1070 1071 rcu_assign_pointer(vrf->rth, rth); 1072 1073 return 0; 1074 } 1075 1076 /**************************** device handling ********************/ 1077 1078 /* cycle interface to flush neighbor cache and move routes across tables */ 1079 static void cycle_netdev(struct net_device *dev, 1080 struct netlink_ext_ack *extack) 1081 { 1082 unsigned int flags = dev->flags; 1083 int ret; 1084 1085 if (!netif_running(dev)) 1086 return; 1087 1088 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1089 if (ret >= 0) 1090 ret = dev_change_flags(dev, flags, extack); 1091 1092 if (ret < 0) { 1093 netdev_err(dev, 1094 "Failed to cycle device %s; route tables might be wrong!\n", 1095 dev->name); 1096 } 1097 } 1098 1099 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1100 struct netlink_ext_ack *extack) 1101 { 1102 int ret; 1103 1104 /* do not allow loopback device to be enslaved to a VRF. 1105 * The vrf device acts as the loopback for the vrf. 1106 */ 1107 if (port_dev == dev_net(dev)->loopback_dev) { 1108 NL_SET_ERR_MSG(extack, 1109 "Can not enslave loopback device to a VRF"); 1110 return -EOPNOTSUPP; 1111 } 1112 1113 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1114 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1115 if (ret < 0) 1116 goto err; 1117 1118 cycle_netdev(port_dev, extack); 1119 1120 return 0; 1121 1122 err: 1123 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1124 return ret; 1125 } 1126 1127 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1128 struct netlink_ext_ack *extack) 1129 { 1130 if (netif_is_l3_master(port_dev)) { 1131 NL_SET_ERR_MSG(extack, 1132 "Can not enslave an L3 master device to a VRF"); 1133 return -EINVAL; 1134 } 1135 1136 if (netif_is_l3_slave(port_dev)) 1137 return -EINVAL; 1138 1139 return do_vrf_add_slave(dev, port_dev, extack); 1140 } 1141 1142 /* inverse of do_vrf_add_slave */ 1143 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1144 { 1145 netdev_upper_dev_unlink(port_dev, dev); 1146 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1147 1148 cycle_netdev(port_dev, NULL); 1149 1150 return 0; 1151 } 1152 1153 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1154 { 1155 return do_vrf_del_slave(dev, port_dev); 1156 } 1157 1158 static void vrf_dev_uninit(struct net_device *dev) 1159 { 1160 struct net_vrf *vrf = netdev_priv(dev); 1161 1162 vrf_rtable_release(dev, vrf); 1163 vrf_rt6_release(dev, vrf); 1164 1165 free_percpu(dev->dstats); 1166 dev->dstats = NULL; 1167 } 1168 1169 static int vrf_dev_init(struct net_device *dev) 1170 { 1171 struct net_vrf *vrf = netdev_priv(dev); 1172 1173 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 1174 if (!dev->dstats) 1175 goto out_nomem; 1176 1177 /* create the default dst which points back to us */ 1178 if (vrf_rtable_create(dev) != 0) 1179 goto out_stats; 1180 1181 if (vrf_rt6_create(dev) != 0) 1182 goto out_rth; 1183 1184 dev->flags = IFF_MASTER | IFF_NOARP; 1185 1186 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1187 dev->operstate = IF_OPER_UP; 1188 netdev_lockdep_set_classes(dev); 1189 return 0; 1190 1191 out_rth: 1192 vrf_rtable_release(dev, vrf); 1193 out_stats: 1194 free_percpu(dev->dstats); 1195 dev->dstats = NULL; 1196 out_nomem: 1197 return -ENOMEM; 1198 } 1199 1200 static const struct net_device_ops vrf_netdev_ops = { 1201 .ndo_init = vrf_dev_init, 1202 .ndo_uninit = vrf_dev_uninit, 1203 .ndo_start_xmit = vrf_xmit, 1204 .ndo_set_mac_address = eth_mac_addr, 1205 .ndo_get_stats64 = vrf_get_stats64, 1206 .ndo_add_slave = vrf_add_slave, 1207 .ndo_del_slave = vrf_del_slave, 1208 }; 1209 1210 static u32 vrf_fib_table(const struct net_device *dev) 1211 { 1212 struct net_vrf *vrf = netdev_priv(dev); 1213 1214 return vrf->tb_id; 1215 } 1216 1217 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1218 { 1219 kfree_skb(skb); 1220 return 0; 1221 } 1222 1223 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1224 struct sk_buff *skb, 1225 struct net_device *dev) 1226 { 1227 struct net *net = dev_net(dev); 1228 1229 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1230 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1231 1232 return skb; 1233 } 1234 1235 static int vrf_prepare_mac_header(struct sk_buff *skb, 1236 struct net_device *vrf_dev, u16 proto) 1237 { 1238 struct ethhdr *eth; 1239 int err; 1240 1241 /* in general, we do not know if there is enough space in the head of 1242 * the packet for hosting the mac header. 1243 */ 1244 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1245 if (unlikely(err)) 1246 /* no space in the skb head */ 1247 return -ENOBUFS; 1248 1249 __skb_push(skb, ETH_HLEN); 1250 eth = (struct ethhdr *)skb->data; 1251 1252 skb_reset_mac_header(skb); 1253 1254 /* we set the ethernet destination and the source addresses to the 1255 * address of the VRF device. 1256 */ 1257 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1258 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1259 eth->h_proto = htons(proto); 1260 1261 /* the destination address of the Ethernet frame corresponds to the 1262 * address set on the VRF interface; therefore, the packet is intended 1263 * to be processed locally. 1264 */ 1265 skb->protocol = eth->h_proto; 1266 skb->pkt_type = PACKET_HOST; 1267 1268 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1269 1270 skb_pull_inline(skb, ETH_HLEN); 1271 1272 return 0; 1273 } 1274 1275 /* prepare and add the mac header to the packet if it was not set previously. 1276 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1277 * If the mac header was already set, the original mac header is left 1278 * untouched and the function returns immediately. 1279 */ 1280 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1281 struct net_device *vrf_dev, 1282 u16 proto) 1283 { 1284 if (skb_mac_header_was_set(skb)) 1285 return 0; 1286 1287 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1288 } 1289 1290 #if IS_ENABLED(CONFIG_IPV6) 1291 /* neighbor handling is done with actual device; do not want 1292 * to flip skb->dev for those ndisc packets. This really fails 1293 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1294 * a start. 1295 */ 1296 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1297 { 1298 const struct ipv6hdr *iph = ipv6_hdr(skb); 1299 bool rc = false; 1300 1301 if (iph->nexthdr == NEXTHDR_ICMP) { 1302 const struct icmp6hdr *icmph; 1303 struct icmp6hdr _icmph; 1304 1305 icmph = skb_header_pointer(skb, sizeof(*iph), 1306 sizeof(_icmph), &_icmph); 1307 if (!icmph) 1308 goto out; 1309 1310 switch (icmph->icmp6_type) { 1311 case NDISC_ROUTER_SOLICITATION: 1312 case NDISC_ROUTER_ADVERTISEMENT: 1313 case NDISC_NEIGHBOUR_SOLICITATION: 1314 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1315 case NDISC_REDIRECT: 1316 rc = true; 1317 break; 1318 } 1319 } 1320 1321 out: 1322 return rc; 1323 } 1324 1325 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1326 const struct net_device *dev, 1327 struct flowi6 *fl6, 1328 int ifindex, 1329 const struct sk_buff *skb, 1330 int flags) 1331 { 1332 struct net_vrf *vrf = netdev_priv(dev); 1333 1334 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1335 } 1336 1337 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1338 int ifindex) 1339 { 1340 const struct ipv6hdr *iph = ipv6_hdr(skb); 1341 struct flowi6 fl6 = { 1342 .flowi6_iif = ifindex, 1343 .flowi6_mark = skb->mark, 1344 .flowi6_proto = iph->nexthdr, 1345 .daddr = iph->daddr, 1346 .saddr = iph->saddr, 1347 .flowlabel = ip6_flowinfo(iph), 1348 }; 1349 struct net *net = dev_net(vrf_dev); 1350 struct rt6_info *rt6; 1351 1352 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1353 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1354 if (unlikely(!rt6)) 1355 return; 1356 1357 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1358 return; 1359 1360 skb_dst_set(skb, &rt6->dst); 1361 } 1362 1363 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1364 struct sk_buff *skb) 1365 { 1366 int orig_iif = skb->skb_iif; 1367 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1368 bool is_ndisc = ipv6_ndisc_frame(skb); 1369 1370 /* loopback, multicast & non-ND link-local traffic; do not push through 1371 * packet taps again. Reset pkt_type for upper layers to process skb. 1372 * For strict packets with a source LLA, determine the dst using the 1373 * original ifindex. 1374 */ 1375 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1376 skb->dev = vrf_dev; 1377 skb->skb_iif = vrf_dev->ifindex; 1378 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1379 1380 if (skb->pkt_type == PACKET_LOOPBACK) 1381 skb->pkt_type = PACKET_HOST; 1382 else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL) 1383 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1384 1385 goto out; 1386 } 1387 1388 /* if packet is NDISC then keep the ingress interface */ 1389 if (!is_ndisc) { 1390 vrf_rx_stats(vrf_dev, skb->len); 1391 skb->dev = vrf_dev; 1392 skb->skb_iif = vrf_dev->ifindex; 1393 1394 if (!list_empty(&vrf_dev->ptype_all)) { 1395 int err; 1396 1397 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1398 ETH_P_IPV6); 1399 if (likely(!err)) { 1400 skb_push(skb, skb->mac_len); 1401 dev_queue_xmit_nit(skb, vrf_dev); 1402 skb_pull(skb, skb->mac_len); 1403 } 1404 } 1405 1406 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1407 } 1408 1409 if (need_strict) 1410 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1411 1412 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1413 out: 1414 return skb; 1415 } 1416 1417 #else 1418 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1419 struct sk_buff *skb) 1420 { 1421 return skb; 1422 } 1423 #endif 1424 1425 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1426 struct sk_buff *skb) 1427 { 1428 skb->dev = vrf_dev; 1429 skb->skb_iif = vrf_dev->ifindex; 1430 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1431 1432 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1433 goto out; 1434 1435 /* loopback traffic; do not push through packet taps again. 1436 * Reset pkt_type for upper layers to process skb 1437 */ 1438 if (skb->pkt_type == PACKET_LOOPBACK) { 1439 skb->pkt_type = PACKET_HOST; 1440 goto out; 1441 } 1442 1443 vrf_rx_stats(vrf_dev, skb->len); 1444 1445 if (!list_empty(&vrf_dev->ptype_all)) { 1446 int err; 1447 1448 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP); 1449 if (likely(!err)) { 1450 skb_push(skb, skb->mac_len); 1451 dev_queue_xmit_nit(skb, vrf_dev); 1452 skb_pull(skb, skb->mac_len); 1453 } 1454 } 1455 1456 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1457 out: 1458 return skb; 1459 } 1460 1461 /* called with rcu lock held */ 1462 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1463 struct sk_buff *skb, 1464 u16 proto) 1465 { 1466 switch (proto) { 1467 case AF_INET: 1468 return vrf_ip_rcv(vrf_dev, skb); 1469 case AF_INET6: 1470 return vrf_ip6_rcv(vrf_dev, skb); 1471 } 1472 1473 return skb; 1474 } 1475 1476 #if IS_ENABLED(CONFIG_IPV6) 1477 /* send to link-local or multicast address via interface enslaved to 1478 * VRF device. Force lookup to VRF table without changing flow struct 1479 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1480 * is taken on the dst by this function. 1481 */ 1482 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1483 struct flowi6 *fl6) 1484 { 1485 struct net *net = dev_net(dev); 1486 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1487 struct dst_entry *dst = NULL; 1488 struct rt6_info *rt; 1489 1490 /* VRF device does not have a link-local address and 1491 * sending packets to link-local or mcast addresses over 1492 * a VRF device does not make sense 1493 */ 1494 if (fl6->flowi6_oif == dev->ifindex) { 1495 dst = &net->ipv6.ip6_null_entry->dst; 1496 return dst; 1497 } 1498 1499 if (!ipv6_addr_any(&fl6->saddr)) 1500 flags |= RT6_LOOKUP_F_HAS_SADDR; 1501 1502 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1503 if (rt) 1504 dst = &rt->dst; 1505 1506 return dst; 1507 } 1508 #endif 1509 1510 static const struct l3mdev_ops vrf_l3mdev_ops = { 1511 .l3mdev_fib_table = vrf_fib_table, 1512 .l3mdev_l3_rcv = vrf_l3_rcv, 1513 .l3mdev_l3_out = vrf_l3_out, 1514 #if IS_ENABLED(CONFIG_IPV6) 1515 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1516 #endif 1517 }; 1518 1519 static void vrf_get_drvinfo(struct net_device *dev, 1520 struct ethtool_drvinfo *info) 1521 { 1522 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 1523 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 1524 } 1525 1526 static const struct ethtool_ops vrf_ethtool_ops = { 1527 .get_drvinfo = vrf_get_drvinfo, 1528 }; 1529 1530 static inline size_t vrf_fib_rule_nl_size(void) 1531 { 1532 size_t sz; 1533 1534 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1535 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1536 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1537 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1538 1539 return sz; 1540 } 1541 1542 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1543 { 1544 struct fib_rule_hdr *frh; 1545 struct nlmsghdr *nlh; 1546 struct sk_buff *skb; 1547 int err; 1548 1549 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1550 !ipv6_mod_enabled()) 1551 return 0; 1552 1553 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1554 if (!skb) 1555 return -ENOMEM; 1556 1557 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1558 if (!nlh) 1559 goto nla_put_failure; 1560 1561 /* rule only needs to appear once */ 1562 nlh->nlmsg_flags |= NLM_F_EXCL; 1563 1564 frh = nlmsg_data(nlh); 1565 memset(frh, 0, sizeof(*frh)); 1566 frh->family = family; 1567 frh->action = FR_ACT_TO_TBL; 1568 1569 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1570 goto nla_put_failure; 1571 1572 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1573 goto nla_put_failure; 1574 1575 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1576 goto nla_put_failure; 1577 1578 nlmsg_end(skb, nlh); 1579 1580 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1581 skb->sk = dev_net(dev)->rtnl; 1582 if (add_it) { 1583 err = fib_nl_newrule(skb, nlh, NULL); 1584 if (err == -EEXIST) 1585 err = 0; 1586 } else { 1587 err = fib_nl_delrule(skb, nlh, NULL); 1588 if (err == -ENOENT) 1589 err = 0; 1590 } 1591 nlmsg_free(skb); 1592 1593 return err; 1594 1595 nla_put_failure: 1596 nlmsg_free(skb); 1597 1598 return -EMSGSIZE; 1599 } 1600 1601 static int vrf_add_fib_rules(const struct net_device *dev) 1602 { 1603 int err; 1604 1605 err = vrf_fib_rule(dev, AF_INET, true); 1606 if (err < 0) 1607 goto out_err; 1608 1609 err = vrf_fib_rule(dev, AF_INET6, true); 1610 if (err < 0) 1611 goto ipv6_err; 1612 1613 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1614 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1615 if (err < 0) 1616 goto ipmr_err; 1617 #endif 1618 1619 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1620 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1621 if (err < 0) 1622 goto ip6mr_err; 1623 #endif 1624 1625 return 0; 1626 1627 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1628 ip6mr_err: 1629 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1630 #endif 1631 1632 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1633 ipmr_err: 1634 vrf_fib_rule(dev, AF_INET6, false); 1635 #endif 1636 1637 ipv6_err: 1638 vrf_fib_rule(dev, AF_INET, false); 1639 1640 out_err: 1641 netdev_err(dev, "Failed to add FIB rules.\n"); 1642 return err; 1643 } 1644 1645 static void vrf_setup(struct net_device *dev) 1646 { 1647 ether_setup(dev); 1648 1649 /* Initialize the device structure. */ 1650 dev->netdev_ops = &vrf_netdev_ops; 1651 dev->l3mdev_ops = &vrf_l3mdev_ops; 1652 dev->ethtool_ops = &vrf_ethtool_ops; 1653 dev->needs_free_netdev = true; 1654 1655 /* Fill in device structure with ethernet-generic values. */ 1656 eth_hw_addr_random(dev); 1657 1658 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1659 dev->features |= NETIF_F_LLTX; 1660 1661 /* don't allow vrf devices to change network namespaces. */ 1662 dev->features |= NETIF_F_NETNS_LOCAL; 1663 1664 /* does not make sense for a VLAN to be added to a vrf device */ 1665 dev->features |= NETIF_F_VLAN_CHALLENGED; 1666 1667 /* enable offload features */ 1668 dev->features |= NETIF_F_GSO_SOFTWARE; 1669 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1670 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1671 1672 dev->hw_features = dev->features; 1673 dev->hw_enc_features = dev->features; 1674 1675 /* default to no qdisc; user can add if desired */ 1676 dev->priv_flags |= IFF_NO_QUEUE; 1677 dev->priv_flags |= IFF_NO_RX_HANDLER; 1678 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1679 1680 /* VRF devices do not care about MTU, but if the MTU is set 1681 * too low then the ipv4 and ipv6 protocols are disabled 1682 * which breaks networking. 1683 */ 1684 dev->min_mtu = IPV6_MIN_MTU; 1685 dev->max_mtu = IP6_MAX_MTU; 1686 dev->mtu = dev->max_mtu; 1687 } 1688 1689 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1690 struct netlink_ext_ack *extack) 1691 { 1692 if (tb[IFLA_ADDRESS]) { 1693 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1694 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1695 return -EINVAL; 1696 } 1697 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1698 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1699 return -EADDRNOTAVAIL; 1700 } 1701 } 1702 return 0; 1703 } 1704 1705 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1706 { 1707 struct net_device *port_dev; 1708 struct list_head *iter; 1709 1710 netdev_for_each_lower_dev(dev, port_dev, iter) 1711 vrf_del_slave(dev, port_dev); 1712 1713 vrf_map_unregister_dev(dev); 1714 1715 unregister_netdevice_queue(dev, head); 1716 } 1717 1718 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1719 struct nlattr *tb[], struct nlattr *data[], 1720 struct netlink_ext_ack *extack) 1721 { 1722 struct net_vrf *vrf = netdev_priv(dev); 1723 struct netns_vrf *nn_vrf; 1724 bool *add_fib_rules; 1725 struct net *net; 1726 int err; 1727 1728 if (!data || !data[IFLA_VRF_TABLE]) { 1729 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1730 return -EINVAL; 1731 } 1732 1733 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1734 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1735 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1736 "Invalid VRF table id"); 1737 return -EINVAL; 1738 } 1739 1740 dev->priv_flags |= IFF_L3MDEV_MASTER; 1741 1742 err = register_netdevice(dev); 1743 if (err) 1744 goto out; 1745 1746 /* mapping between table_id and vrf; 1747 * note: such binding could not be done in the dev init function 1748 * because dev->ifindex id is not available yet. 1749 */ 1750 vrf->ifindex = dev->ifindex; 1751 1752 err = vrf_map_register_dev(dev, extack); 1753 if (err) { 1754 unregister_netdevice(dev); 1755 goto out; 1756 } 1757 1758 net = dev_net(dev); 1759 nn_vrf = net_generic(net, vrf_net_id); 1760 1761 add_fib_rules = &nn_vrf->add_fib_rules; 1762 if (*add_fib_rules) { 1763 err = vrf_add_fib_rules(dev); 1764 if (err) { 1765 vrf_map_unregister_dev(dev); 1766 unregister_netdevice(dev); 1767 goto out; 1768 } 1769 *add_fib_rules = false; 1770 } 1771 1772 out: 1773 return err; 1774 } 1775 1776 static size_t vrf_nl_getsize(const struct net_device *dev) 1777 { 1778 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1779 } 1780 1781 static int vrf_fillinfo(struct sk_buff *skb, 1782 const struct net_device *dev) 1783 { 1784 struct net_vrf *vrf = netdev_priv(dev); 1785 1786 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1787 } 1788 1789 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1790 const struct net_device *slave_dev) 1791 { 1792 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1793 } 1794 1795 static int vrf_fill_slave_info(struct sk_buff *skb, 1796 const struct net_device *vrf_dev, 1797 const struct net_device *slave_dev) 1798 { 1799 struct net_vrf *vrf = netdev_priv(vrf_dev); 1800 1801 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1802 return -EMSGSIZE; 1803 1804 return 0; 1805 } 1806 1807 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1808 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1809 }; 1810 1811 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1812 .kind = DRV_NAME, 1813 .priv_size = sizeof(struct net_vrf), 1814 1815 .get_size = vrf_nl_getsize, 1816 .policy = vrf_nl_policy, 1817 .validate = vrf_validate, 1818 .fill_info = vrf_fillinfo, 1819 1820 .get_slave_size = vrf_get_slave_size, 1821 .fill_slave_info = vrf_fill_slave_info, 1822 1823 .newlink = vrf_newlink, 1824 .dellink = vrf_dellink, 1825 .setup = vrf_setup, 1826 .maxtype = IFLA_VRF_MAX, 1827 }; 1828 1829 static int vrf_device_event(struct notifier_block *unused, 1830 unsigned long event, void *ptr) 1831 { 1832 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1833 1834 /* only care about unregister events to drop slave references */ 1835 if (event == NETDEV_UNREGISTER) { 1836 struct net_device *vrf_dev; 1837 1838 if (!netif_is_l3_slave(dev)) 1839 goto out; 1840 1841 vrf_dev = netdev_master_upper_dev_get(dev); 1842 vrf_del_slave(vrf_dev, dev); 1843 } 1844 out: 1845 return NOTIFY_DONE; 1846 } 1847 1848 static struct notifier_block vrf_notifier_block __read_mostly = { 1849 .notifier_call = vrf_device_event, 1850 }; 1851 1852 static int vrf_map_init(struct vrf_map *vmap) 1853 { 1854 spin_lock_init(&vmap->vmap_lock); 1855 hash_init(vmap->ht); 1856 1857 vmap->strict_mode = false; 1858 1859 return 0; 1860 } 1861 1862 #ifdef CONFIG_SYSCTL 1863 static bool vrf_strict_mode(struct vrf_map *vmap) 1864 { 1865 bool strict_mode; 1866 1867 vrf_map_lock(vmap); 1868 strict_mode = vmap->strict_mode; 1869 vrf_map_unlock(vmap); 1870 1871 return strict_mode; 1872 } 1873 1874 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1875 { 1876 bool *cur_mode; 1877 int res = 0; 1878 1879 vrf_map_lock(vmap); 1880 1881 cur_mode = &vmap->strict_mode; 1882 if (*cur_mode == new_mode) 1883 goto unlock; 1884 1885 if (*cur_mode) { 1886 /* disable strict mode */ 1887 *cur_mode = false; 1888 } else { 1889 if (vmap->shared_tables) { 1890 /* we cannot allow strict_mode because there are some 1891 * vrfs that share one or more tables. 1892 */ 1893 res = -EBUSY; 1894 goto unlock; 1895 } 1896 1897 /* no tables are shared among vrfs, so we can go back 1898 * to 1:1 association between a vrf with its table. 1899 */ 1900 *cur_mode = true; 1901 } 1902 1903 unlock: 1904 vrf_map_unlock(vmap); 1905 1906 return res; 1907 } 1908 1909 static int vrf_shared_table_handler(struct ctl_table *table, int write, 1910 void *buffer, size_t *lenp, loff_t *ppos) 1911 { 1912 struct net *net = (struct net *)table->extra1; 1913 struct vrf_map *vmap = netns_vrf_map(net); 1914 int proc_strict_mode = 0; 1915 struct ctl_table tmp = { 1916 .procname = table->procname, 1917 .data = &proc_strict_mode, 1918 .maxlen = sizeof(int), 1919 .mode = table->mode, 1920 .extra1 = SYSCTL_ZERO, 1921 .extra2 = SYSCTL_ONE, 1922 }; 1923 int ret; 1924 1925 if (!write) 1926 proc_strict_mode = vrf_strict_mode(vmap); 1927 1928 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1929 1930 if (write && ret == 0) 1931 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1932 1933 return ret; 1934 } 1935 1936 static const struct ctl_table vrf_table[] = { 1937 { 1938 .procname = "strict_mode", 1939 .data = NULL, 1940 .maxlen = sizeof(int), 1941 .mode = 0644, 1942 .proc_handler = vrf_shared_table_handler, 1943 /* set by the vrf_netns_init */ 1944 .extra1 = NULL, 1945 }, 1946 { }, 1947 }; 1948 1949 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1950 { 1951 struct ctl_table *table; 1952 1953 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1954 if (!table) 1955 return -ENOMEM; 1956 1957 /* init the extra1 parameter with the reference to current netns */ 1958 table[0].extra1 = net; 1959 1960 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table); 1961 if (!nn_vrf->ctl_hdr) { 1962 kfree(table); 1963 return -ENOMEM; 1964 } 1965 1966 return 0; 1967 } 1968 1969 static void vrf_netns_exit_sysctl(struct net *net) 1970 { 1971 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1972 struct ctl_table *table; 1973 1974 table = nn_vrf->ctl_hdr->ctl_table_arg; 1975 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1976 kfree(table); 1977 } 1978 #else 1979 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1980 { 1981 return 0; 1982 } 1983 1984 static void vrf_netns_exit_sysctl(struct net *net) 1985 { 1986 } 1987 #endif 1988 1989 /* Initialize per network namespace state */ 1990 static int __net_init vrf_netns_init(struct net *net) 1991 { 1992 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1993 1994 nn_vrf->add_fib_rules = true; 1995 vrf_map_init(&nn_vrf->vmap); 1996 1997 return vrf_netns_init_sysctl(net, nn_vrf); 1998 } 1999 2000 static void __net_exit vrf_netns_exit(struct net *net) 2001 { 2002 vrf_netns_exit_sysctl(net); 2003 } 2004 2005 static struct pernet_operations vrf_net_ops __net_initdata = { 2006 .init = vrf_netns_init, 2007 .exit = vrf_netns_exit, 2008 .id = &vrf_net_id, 2009 .size = sizeof(struct netns_vrf), 2010 }; 2011 2012 static int __init vrf_init_module(void) 2013 { 2014 int rc; 2015 2016 register_netdevice_notifier(&vrf_notifier_block); 2017 2018 rc = register_pernet_subsys(&vrf_net_ops); 2019 if (rc < 0) 2020 goto error; 2021 2022 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2023 vrf_ifindex_lookup_by_table_id); 2024 if (rc < 0) 2025 goto unreg_pernet; 2026 2027 rc = rtnl_link_register(&vrf_link_ops); 2028 if (rc < 0) 2029 goto table_lookup_unreg; 2030 2031 return 0; 2032 2033 table_lookup_unreg: 2034 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2035 vrf_ifindex_lookup_by_table_id); 2036 2037 unreg_pernet: 2038 unregister_pernet_subsys(&vrf_net_ops); 2039 2040 error: 2041 unregister_netdevice_notifier(&vrf_notifier_block); 2042 return rc; 2043 } 2044 2045 module_init(vrf_init_module); 2046 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2047 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2048 MODULE_LICENSE("GPL"); 2049 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2050 MODULE_VERSION(DRV_VERSION); 2051