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/sch_generic.h> 38 #include <net/netns/generic.h> 39 #include <net/netfilter/nf_conntrack.h> 40 41 #define DRV_NAME "vrf" 42 #define DRV_VERSION "1.1" 43 44 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 45 46 #define HT_MAP_BITS 4 47 #define HASH_INITVAL ((u32)0xcafef00d) 48 49 struct vrf_map { 50 DECLARE_HASHTABLE(ht, HT_MAP_BITS); 51 spinlock_t vmap_lock; 52 53 /* shared_tables: 54 * count how many distinct tables do not comply with the strict mode 55 * requirement. 56 * shared_tables value must be 0 in order to enable the strict mode. 57 * 58 * example of the evolution of shared_tables: 59 * | time 60 * add vrf0 --> table 100 shared_tables = 0 | t0 61 * add vrf1 --> table 101 shared_tables = 0 | t1 62 * add vrf2 --> table 100 shared_tables = 1 | t2 63 * add vrf3 --> table 100 shared_tables = 1 | t3 64 * add vrf4 --> table 101 shared_tables = 2 v t4 65 * 66 * shared_tables is a "step function" (or "staircase function") 67 * and it is increased by one when the second vrf is associated to a 68 * table. 69 * 70 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1. 71 * 72 * at t3, another dev (vrf3) is bound to the same table 100 but the 73 * value of shared_tables is still 1. 74 * This means that no matter how many new vrfs will register on the 75 * table 100, the shared_tables will not increase (considering only 76 * table 100). 77 * 78 * at t4, vrf4 is bound to table 101, and shared_tables = 2. 79 * 80 * Looking at the value of shared_tables we can immediately know if 81 * the strict_mode can or cannot be enforced. Indeed, strict_mode 82 * can be enforced iff shared_tables = 0. 83 * 84 * Conversely, shared_tables is decreased when a vrf is de-associated 85 * from a table with exactly two associated vrfs. 86 */ 87 u32 shared_tables; 88 89 bool strict_mode; 90 }; 91 92 struct vrf_map_elem { 93 struct hlist_node hnode; 94 struct list_head vrf_list; /* VRFs registered to this table */ 95 96 u32 table_id; 97 int users; 98 int ifindex; 99 }; 100 101 static unsigned int vrf_net_id; 102 103 /* per netns vrf data */ 104 struct netns_vrf { 105 /* protected by rtnl lock */ 106 bool add_fib_rules; 107 108 struct vrf_map vmap; 109 struct ctl_table_header *ctl_hdr; 110 }; 111 112 struct net_vrf { 113 struct rtable __rcu *rth; 114 struct rt6_info __rcu *rt6; 115 #if IS_ENABLED(CONFIG_IPV6) 116 struct fib6_table *fib6_table; 117 #endif 118 u32 tb_id; 119 120 struct list_head me_list; /* entry in vrf_map_elem */ 121 int ifindex; 122 }; 123 124 struct pcpu_dstats { 125 u64 tx_pkts; 126 u64 tx_bytes; 127 u64 tx_drps; 128 u64 rx_pkts; 129 u64 rx_bytes; 130 u64 rx_drps; 131 struct u64_stats_sync syncp; 132 }; 133 134 static void vrf_rx_stats(struct net_device *dev, int len) 135 { 136 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 137 138 u64_stats_update_begin(&dstats->syncp); 139 dstats->rx_pkts++; 140 dstats->rx_bytes += len; 141 u64_stats_update_end(&dstats->syncp); 142 } 143 144 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 145 { 146 vrf_dev->stats.tx_errors++; 147 kfree_skb(skb); 148 } 149 150 static void vrf_get_stats64(struct net_device *dev, 151 struct rtnl_link_stats64 *stats) 152 { 153 int i; 154 155 for_each_possible_cpu(i) { 156 const struct pcpu_dstats *dstats; 157 u64 tbytes, tpkts, tdrops, rbytes, rpkts; 158 unsigned int start; 159 160 dstats = per_cpu_ptr(dev->dstats, i); 161 do { 162 start = u64_stats_fetch_begin(&dstats->syncp); 163 tbytes = dstats->tx_bytes; 164 tpkts = dstats->tx_pkts; 165 tdrops = dstats->tx_drps; 166 rbytes = dstats->rx_bytes; 167 rpkts = dstats->rx_pkts; 168 } while (u64_stats_fetch_retry(&dstats->syncp, start)); 169 stats->tx_bytes += tbytes; 170 stats->tx_packets += tpkts; 171 stats->tx_dropped += tdrops; 172 stats->rx_bytes += rbytes; 173 stats->rx_packets += rpkts; 174 } 175 } 176 177 static struct vrf_map *netns_vrf_map(struct net *net) 178 { 179 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 180 181 return &nn_vrf->vmap; 182 } 183 184 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 185 { 186 return netns_vrf_map(dev_net(dev)); 187 } 188 189 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 190 { 191 struct list_head *me_head = &me->vrf_list; 192 struct net_vrf *vrf; 193 194 if (list_empty(me_head)) 195 return -ENODEV; 196 197 vrf = list_first_entry(me_head, struct net_vrf, me_list); 198 199 return vrf->ifindex; 200 } 201 202 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 203 { 204 struct vrf_map_elem *me; 205 206 me = kmalloc(sizeof(*me), flags); 207 if (!me) 208 return NULL; 209 210 return me; 211 } 212 213 static void vrf_map_elem_free(struct vrf_map_elem *me) 214 { 215 kfree(me); 216 } 217 218 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 219 int ifindex, int users) 220 { 221 me->table_id = table_id; 222 me->ifindex = ifindex; 223 me->users = users; 224 INIT_LIST_HEAD(&me->vrf_list); 225 } 226 227 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 228 u32 table_id) 229 { 230 struct vrf_map_elem *me; 231 u32 key; 232 233 key = jhash_1word(table_id, HASH_INITVAL); 234 hash_for_each_possible(vmap->ht, me, hnode, key) { 235 if (me->table_id == table_id) 236 return me; 237 } 238 239 return NULL; 240 } 241 242 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 243 { 244 u32 table_id = me->table_id; 245 u32 key; 246 247 key = jhash_1word(table_id, HASH_INITVAL); 248 hash_add(vmap->ht, &me->hnode, key); 249 } 250 251 static void vrf_map_del_elem(struct vrf_map_elem *me) 252 { 253 hash_del(&me->hnode); 254 } 255 256 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 257 { 258 spin_lock(&vmap->vmap_lock); 259 } 260 261 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 262 { 263 spin_unlock(&vmap->vmap_lock); 264 } 265 266 /* called with rtnl lock held */ 267 static int 268 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 269 { 270 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 271 struct net_vrf *vrf = netdev_priv(dev); 272 struct vrf_map_elem *new_me, *me; 273 u32 table_id = vrf->tb_id; 274 bool free_new_me = false; 275 int users; 276 int res; 277 278 /* we pre-allocate elements used in the spin-locked section (so that we 279 * keep the spinlock as short as possible). 280 */ 281 new_me = vrf_map_elem_alloc(GFP_KERNEL); 282 if (!new_me) 283 return -ENOMEM; 284 285 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 286 287 vrf_map_lock(vmap); 288 289 me = vrf_map_lookup_elem(vmap, table_id); 290 if (!me) { 291 me = new_me; 292 vrf_map_add_elem(vmap, me); 293 goto link_vrf; 294 } 295 296 /* we already have an entry in the vrf_map, so it means there is (at 297 * least) a vrf registered on the specific table. 298 */ 299 free_new_me = true; 300 if (vmap->strict_mode) { 301 /* vrfs cannot share the same table */ 302 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 303 res = -EBUSY; 304 goto unlock; 305 } 306 307 link_vrf: 308 users = ++me->users; 309 if (users == 2) 310 ++vmap->shared_tables; 311 312 list_add(&vrf->me_list, &me->vrf_list); 313 314 res = 0; 315 316 unlock: 317 vrf_map_unlock(vmap); 318 319 /* clean-up, if needed */ 320 if (free_new_me) 321 vrf_map_elem_free(new_me); 322 323 return res; 324 } 325 326 /* called with rtnl lock held */ 327 static void vrf_map_unregister_dev(struct net_device *dev) 328 { 329 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 330 struct net_vrf *vrf = netdev_priv(dev); 331 u32 table_id = vrf->tb_id; 332 struct vrf_map_elem *me; 333 int users; 334 335 vrf_map_lock(vmap); 336 337 me = vrf_map_lookup_elem(vmap, table_id); 338 if (!me) 339 goto unlock; 340 341 list_del(&vrf->me_list); 342 343 users = --me->users; 344 if (users == 1) { 345 --vmap->shared_tables; 346 } else if (users == 0) { 347 vrf_map_del_elem(me); 348 349 /* no one will refer to this element anymore */ 350 vrf_map_elem_free(me); 351 } 352 353 unlock: 354 vrf_map_unlock(vmap); 355 } 356 357 /* return the vrf device index associated with the table_id */ 358 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 359 { 360 struct vrf_map *vmap = netns_vrf_map(net); 361 struct vrf_map_elem *me; 362 int ifindex; 363 364 vrf_map_lock(vmap); 365 366 if (!vmap->strict_mode) { 367 ifindex = -EPERM; 368 goto unlock; 369 } 370 371 me = vrf_map_lookup_elem(vmap, table_id); 372 if (!me) { 373 ifindex = -ENODEV; 374 goto unlock; 375 } 376 377 ifindex = vrf_map_elem_get_vrf_ifindex(me); 378 379 unlock: 380 vrf_map_unlock(vmap); 381 382 return ifindex; 383 } 384 385 /* by default VRF devices do not have a qdisc and are expected 386 * to be created with only a single queue. 387 */ 388 static bool qdisc_tx_is_default(const struct net_device *dev) 389 { 390 struct netdev_queue *txq; 391 struct Qdisc *qdisc; 392 393 if (dev->num_tx_queues > 1) 394 return false; 395 396 txq = netdev_get_tx_queue(dev, 0); 397 qdisc = rcu_access_pointer(txq->qdisc); 398 399 return !qdisc->enqueue; 400 } 401 402 /* Local traffic destined to local address. Reinsert the packet to rx 403 * path, similar to loopback handling. 404 */ 405 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 406 struct dst_entry *dst) 407 { 408 int len = skb->len; 409 410 skb_orphan(skb); 411 412 skb_dst_set(skb, dst); 413 414 /* set pkt_type to avoid skb hitting packet taps twice - 415 * once on Tx and again in Rx processing 416 */ 417 skb->pkt_type = PACKET_LOOPBACK; 418 419 skb->protocol = eth_type_trans(skb, dev); 420 421 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) 422 vrf_rx_stats(dev, len); 423 else 424 this_cpu_inc(dev->dstats->rx_drps); 425 426 return NETDEV_TX_OK; 427 } 428 429 static void vrf_nf_set_untracked(struct sk_buff *skb) 430 { 431 if (skb_get_nfct(skb) == 0) 432 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 433 } 434 435 static void vrf_nf_reset_ct(struct sk_buff *skb) 436 { 437 if (skb_get_nfct(skb) == IP_CT_UNTRACKED) 438 nf_reset_ct(skb); 439 } 440 441 #if IS_ENABLED(CONFIG_IPV6) 442 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 443 struct sk_buff *skb) 444 { 445 int err; 446 447 vrf_nf_reset_ct(skb); 448 449 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 450 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 451 452 if (likely(err == 1)) 453 err = dst_output(net, sk, skb); 454 455 return err; 456 } 457 458 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 459 struct net_device *dev) 460 { 461 const struct ipv6hdr *iph; 462 struct net *net = dev_net(skb->dev); 463 struct flowi6 fl6; 464 int ret = NET_XMIT_DROP; 465 struct dst_entry *dst; 466 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 467 468 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 469 goto err; 470 471 iph = ipv6_hdr(skb); 472 473 memset(&fl6, 0, sizeof(fl6)); 474 /* needed to match OIF rule */ 475 fl6.flowi6_l3mdev = dev->ifindex; 476 fl6.flowi6_iif = LOOPBACK_IFINDEX; 477 fl6.daddr = iph->daddr; 478 fl6.saddr = iph->saddr; 479 fl6.flowlabel = ip6_flowinfo(iph); 480 fl6.flowi6_mark = skb->mark; 481 fl6.flowi6_proto = iph->nexthdr; 482 483 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 484 if (IS_ERR(dst) || dst == dst_null) 485 goto err; 486 487 skb_dst_drop(skb); 488 489 /* if dst.dev is the VRF device again this is locally originated traffic 490 * destined to a local address. Short circuit to Rx path. 491 */ 492 if (dst->dev == dev) 493 return vrf_local_xmit(skb, dev, dst); 494 495 skb_dst_set(skb, dst); 496 497 /* strip the ethernet header added for pass through VRF device */ 498 __skb_pull(skb, skb_network_offset(skb)); 499 500 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb))); 501 ret = vrf_ip6_local_out(net, skb->sk, skb); 502 if (unlikely(net_xmit_eval(ret))) 503 dev->stats.tx_errors++; 504 else 505 ret = NET_XMIT_SUCCESS; 506 507 return ret; 508 err: 509 vrf_tx_error(dev, skb); 510 return NET_XMIT_DROP; 511 } 512 #else 513 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 514 struct net_device *dev) 515 { 516 vrf_tx_error(dev, skb); 517 return NET_XMIT_DROP; 518 } 519 #endif 520 521 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 522 static int vrf_ip_local_out(struct net *net, struct sock *sk, 523 struct sk_buff *skb) 524 { 525 int err; 526 527 vrf_nf_reset_ct(skb); 528 529 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 530 skb, NULL, skb_dst(skb)->dev, dst_output); 531 if (likely(err == 1)) 532 err = dst_output(net, sk, skb); 533 534 return err; 535 } 536 537 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 538 struct net_device *vrf_dev) 539 { 540 struct iphdr *ip4h; 541 int ret = NET_XMIT_DROP; 542 struct flowi4 fl4; 543 struct net *net = dev_net(vrf_dev); 544 struct rtable *rt; 545 546 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 547 goto err; 548 549 ip4h = ip_hdr(skb); 550 551 memset(&fl4, 0, sizeof(fl4)); 552 /* needed to match OIF rule */ 553 fl4.flowi4_l3mdev = vrf_dev->ifindex; 554 fl4.flowi4_iif = LOOPBACK_IFINDEX; 555 fl4.flowi4_tos = RT_TOS(ip4h->tos); 556 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC; 557 fl4.flowi4_proto = ip4h->protocol; 558 fl4.daddr = ip4h->daddr; 559 fl4.saddr = ip4h->saddr; 560 561 rt = ip_route_output_flow(net, &fl4, NULL); 562 if (IS_ERR(rt)) 563 goto err; 564 565 skb_dst_drop(skb); 566 567 /* if dst.dev is the VRF device again this is locally originated traffic 568 * destined to a local address. Short circuit to Rx path. 569 */ 570 if (rt->dst.dev == vrf_dev) 571 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 572 573 skb_dst_set(skb, &rt->dst); 574 575 /* strip the ethernet header added for pass through VRF device */ 576 __skb_pull(skb, skb_network_offset(skb)); 577 578 if (!ip4h->saddr) { 579 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 580 RT_SCOPE_LINK); 581 } 582 583 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 584 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 585 if (unlikely(net_xmit_eval(ret))) 586 vrf_dev->stats.tx_errors++; 587 else 588 ret = NET_XMIT_SUCCESS; 589 590 out: 591 return ret; 592 err: 593 vrf_tx_error(vrf_dev, skb); 594 goto out; 595 } 596 597 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 598 { 599 switch (skb->protocol) { 600 case htons(ETH_P_IP): 601 return vrf_process_v4_outbound(skb, dev); 602 case htons(ETH_P_IPV6): 603 return vrf_process_v6_outbound(skb, dev); 604 default: 605 vrf_tx_error(dev, skb); 606 return NET_XMIT_DROP; 607 } 608 } 609 610 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 611 { 612 int len = skb->len; 613 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 614 615 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 616 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 617 618 u64_stats_update_begin(&dstats->syncp); 619 dstats->tx_pkts++; 620 dstats->tx_bytes += len; 621 u64_stats_update_end(&dstats->syncp); 622 } else { 623 this_cpu_inc(dev->dstats->tx_drps); 624 } 625 626 return ret; 627 } 628 629 static void vrf_finish_direct(struct sk_buff *skb) 630 { 631 struct net_device *vrf_dev = skb->dev; 632 633 if (!list_empty(&vrf_dev->ptype_all) && 634 likely(skb_headroom(skb) >= ETH_HLEN)) { 635 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 636 637 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 638 eth_zero_addr(eth->h_dest); 639 eth->h_proto = skb->protocol; 640 641 dev_queue_xmit_nit(skb, vrf_dev); 642 643 skb_pull(skb, ETH_HLEN); 644 } 645 646 vrf_nf_reset_ct(skb); 647 } 648 649 #if IS_ENABLED(CONFIG_IPV6) 650 /* modelled after ip6_finish_output2 */ 651 static int vrf_finish_output6(struct net *net, struct sock *sk, 652 struct sk_buff *skb) 653 { 654 struct dst_entry *dst = skb_dst(skb); 655 struct net_device *dev = dst->dev; 656 const struct in6_addr *nexthop; 657 struct neighbour *neigh; 658 int ret; 659 660 vrf_nf_reset_ct(skb); 661 662 skb->protocol = htons(ETH_P_IPV6); 663 skb->dev = dev; 664 665 rcu_read_lock(); 666 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 667 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 668 if (unlikely(!neigh)) 669 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 670 if (!IS_ERR(neigh)) { 671 sock_confirm_neigh(skb, neigh); 672 ret = neigh_output(neigh, skb, false); 673 rcu_read_unlock(); 674 return ret; 675 } 676 rcu_read_unlock(); 677 678 IP6_INC_STATS(dev_net(dst->dev), 679 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 680 kfree_skb(skb); 681 return -EINVAL; 682 } 683 684 /* modelled after ip6_output */ 685 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 686 { 687 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 688 net, sk, skb, NULL, skb_dst(skb)->dev, 689 vrf_finish_output6, 690 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 691 } 692 693 /* set dst on skb to send packet to us via dev_xmit path. Allows 694 * packet to go through device based features such as qdisc, netfilter 695 * hooks and packet sockets with skb->dev set to vrf device. 696 */ 697 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 698 struct sk_buff *skb) 699 { 700 struct net_vrf *vrf = netdev_priv(vrf_dev); 701 struct dst_entry *dst = NULL; 702 struct rt6_info *rt6; 703 704 rcu_read_lock(); 705 706 rt6 = rcu_dereference(vrf->rt6); 707 if (likely(rt6)) { 708 dst = &rt6->dst; 709 dst_hold(dst); 710 } 711 712 rcu_read_unlock(); 713 714 if (unlikely(!dst)) { 715 vrf_tx_error(vrf_dev, skb); 716 return NULL; 717 } 718 719 skb_dst_drop(skb); 720 skb_dst_set(skb, dst); 721 722 return skb; 723 } 724 725 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 726 struct sk_buff *skb) 727 { 728 vrf_finish_direct(skb); 729 730 return vrf_ip6_local_out(net, sk, skb); 731 } 732 733 static int vrf_output6_direct(struct net *net, struct sock *sk, 734 struct sk_buff *skb) 735 { 736 int err = 1; 737 738 skb->protocol = htons(ETH_P_IPV6); 739 740 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 741 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 742 NULL, skb->dev, vrf_output6_direct_finish); 743 744 if (likely(err == 1)) 745 vrf_finish_direct(skb); 746 747 return err; 748 } 749 750 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 751 struct sk_buff *skb) 752 { 753 int err; 754 755 err = vrf_output6_direct(net, sk, skb); 756 if (likely(err == 1)) 757 err = vrf_ip6_local_out(net, sk, skb); 758 759 return err; 760 } 761 762 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 763 struct sock *sk, 764 struct sk_buff *skb) 765 { 766 struct net *net = dev_net(vrf_dev); 767 int err; 768 769 skb->dev = vrf_dev; 770 771 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 772 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 773 774 if (likely(err == 1)) 775 err = vrf_output6_direct(net, sk, skb); 776 777 if (likely(err == 1)) 778 return skb; 779 780 return NULL; 781 } 782 783 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 784 struct sock *sk, 785 struct sk_buff *skb) 786 { 787 /* don't divert link scope packets */ 788 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 789 return skb; 790 791 vrf_nf_set_untracked(skb); 792 793 if (qdisc_tx_is_default(vrf_dev) || 794 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 795 return vrf_ip6_out_direct(vrf_dev, sk, skb); 796 797 return vrf_ip6_out_redirect(vrf_dev, skb); 798 } 799 800 /* holding rtnl */ 801 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 802 { 803 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 804 struct net *net = dev_net(dev); 805 struct dst_entry *dst; 806 807 RCU_INIT_POINTER(vrf->rt6, NULL); 808 synchronize_rcu(); 809 810 /* move dev in dst's to loopback so this VRF device can be deleted 811 * - based on dst_ifdown 812 */ 813 if (rt6) { 814 dst = &rt6->dst; 815 netdev_ref_replace(dst->dev, net->loopback_dev, 816 &dst->dev_tracker, GFP_KERNEL); 817 dst->dev = net->loopback_dev; 818 dst_release(dst); 819 } 820 } 821 822 static int vrf_rt6_create(struct net_device *dev) 823 { 824 int flags = DST_NOPOLICY | DST_NOXFRM; 825 struct net_vrf *vrf = netdev_priv(dev); 826 struct net *net = dev_net(dev); 827 struct rt6_info *rt6; 828 int rc = -ENOMEM; 829 830 /* IPv6 can be CONFIG enabled and then disabled runtime */ 831 if (!ipv6_mod_enabled()) 832 return 0; 833 834 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 835 if (!vrf->fib6_table) 836 goto out; 837 838 /* create a dst for routing packets out a VRF device */ 839 rt6 = ip6_dst_alloc(net, dev, flags); 840 if (!rt6) 841 goto out; 842 843 rt6->dst.output = vrf_output6; 844 845 rcu_assign_pointer(vrf->rt6, rt6); 846 847 rc = 0; 848 out: 849 return rc; 850 } 851 #else 852 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 853 struct sock *sk, 854 struct sk_buff *skb) 855 { 856 return skb; 857 } 858 859 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 860 { 861 } 862 863 static int vrf_rt6_create(struct net_device *dev) 864 { 865 return 0; 866 } 867 #endif 868 869 /* modelled after ip_finish_output2 */ 870 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 871 { 872 struct dst_entry *dst = skb_dst(skb); 873 struct rtable *rt = (struct rtable *)dst; 874 struct net_device *dev = dst->dev; 875 unsigned int hh_len = LL_RESERVED_SPACE(dev); 876 struct neighbour *neigh; 877 bool is_v6gw = false; 878 879 vrf_nf_reset_ct(skb); 880 881 /* Be paranoid, rather than too clever. */ 882 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 883 skb = skb_expand_head(skb, hh_len); 884 if (!skb) { 885 dev->stats.tx_errors++; 886 return -ENOMEM; 887 } 888 } 889 890 rcu_read_lock(); 891 892 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 893 if (!IS_ERR(neigh)) { 894 int ret; 895 896 sock_confirm_neigh(skb, neigh); 897 /* if crossing protocols, can not use the cached header */ 898 ret = neigh_output(neigh, skb, is_v6gw); 899 rcu_read_unlock(); 900 return ret; 901 } 902 903 rcu_read_unlock(); 904 vrf_tx_error(skb->dev, skb); 905 return -EINVAL; 906 } 907 908 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 909 { 910 struct net_device *dev = skb_dst(skb)->dev; 911 912 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 913 914 skb->dev = dev; 915 skb->protocol = htons(ETH_P_IP); 916 917 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 918 net, sk, skb, NULL, dev, 919 vrf_finish_output, 920 !(IPCB(skb)->flags & IPSKB_REROUTED)); 921 } 922 923 /* set dst on skb to send packet to us via dev_xmit path. Allows 924 * packet to go through device based features such as qdisc, netfilter 925 * hooks and packet sockets with skb->dev set to vrf device. 926 */ 927 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 928 struct sk_buff *skb) 929 { 930 struct net_vrf *vrf = netdev_priv(vrf_dev); 931 struct dst_entry *dst = NULL; 932 struct rtable *rth; 933 934 rcu_read_lock(); 935 936 rth = rcu_dereference(vrf->rth); 937 if (likely(rth)) { 938 dst = &rth->dst; 939 dst_hold(dst); 940 } 941 942 rcu_read_unlock(); 943 944 if (unlikely(!dst)) { 945 vrf_tx_error(vrf_dev, skb); 946 return NULL; 947 } 948 949 skb_dst_drop(skb); 950 skb_dst_set(skb, dst); 951 952 return skb; 953 } 954 955 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 956 struct sk_buff *skb) 957 { 958 vrf_finish_direct(skb); 959 960 return vrf_ip_local_out(net, sk, skb); 961 } 962 963 static int vrf_output_direct(struct net *net, struct sock *sk, 964 struct sk_buff *skb) 965 { 966 int err = 1; 967 968 skb->protocol = htons(ETH_P_IP); 969 970 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 971 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 972 NULL, skb->dev, vrf_output_direct_finish); 973 974 if (likely(err == 1)) 975 vrf_finish_direct(skb); 976 977 return err; 978 } 979 980 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 981 struct sk_buff *skb) 982 { 983 int err; 984 985 err = vrf_output_direct(net, sk, skb); 986 if (likely(err == 1)) 987 err = vrf_ip_local_out(net, sk, skb); 988 989 return err; 990 } 991 992 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 993 struct sock *sk, 994 struct sk_buff *skb) 995 { 996 struct net *net = dev_net(vrf_dev); 997 int err; 998 999 skb->dev = vrf_dev; 1000 1001 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 1002 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 1003 1004 if (likely(err == 1)) 1005 err = vrf_output_direct(net, sk, skb); 1006 1007 if (likely(err == 1)) 1008 return skb; 1009 1010 return NULL; 1011 } 1012 1013 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 1014 struct sock *sk, 1015 struct sk_buff *skb) 1016 { 1017 /* don't divert multicast or local broadcast */ 1018 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 1019 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 1020 return skb; 1021 1022 vrf_nf_set_untracked(skb); 1023 1024 if (qdisc_tx_is_default(vrf_dev) || 1025 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 1026 return vrf_ip_out_direct(vrf_dev, sk, skb); 1027 1028 return vrf_ip_out_redirect(vrf_dev, skb); 1029 } 1030 1031 /* called with rcu lock held */ 1032 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1033 struct sock *sk, 1034 struct sk_buff *skb, 1035 u16 proto) 1036 { 1037 switch (proto) { 1038 case AF_INET: 1039 return vrf_ip_out(vrf_dev, sk, skb); 1040 case AF_INET6: 1041 return vrf_ip6_out(vrf_dev, sk, skb); 1042 } 1043 1044 return skb; 1045 } 1046 1047 /* holding rtnl */ 1048 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1049 { 1050 struct rtable *rth = rtnl_dereference(vrf->rth); 1051 struct net *net = dev_net(dev); 1052 struct dst_entry *dst; 1053 1054 RCU_INIT_POINTER(vrf->rth, NULL); 1055 synchronize_rcu(); 1056 1057 /* move dev in dst's to loopback so this VRF device can be deleted 1058 * - based on dst_ifdown 1059 */ 1060 if (rth) { 1061 dst = &rth->dst; 1062 netdev_ref_replace(dst->dev, net->loopback_dev, 1063 &dst->dev_tracker, GFP_KERNEL); 1064 dst->dev = net->loopback_dev; 1065 dst_release(dst); 1066 } 1067 } 1068 1069 static int vrf_rtable_create(struct net_device *dev) 1070 { 1071 struct net_vrf *vrf = netdev_priv(dev); 1072 struct rtable *rth; 1073 1074 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1075 return -ENOMEM; 1076 1077 /* create a dst for routing packets out through a VRF device */ 1078 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1); 1079 if (!rth) 1080 return -ENOMEM; 1081 1082 rth->dst.output = vrf_output; 1083 1084 rcu_assign_pointer(vrf->rth, rth); 1085 1086 return 0; 1087 } 1088 1089 /**************************** device handling ********************/ 1090 1091 /* cycle interface to flush neighbor cache and move routes across tables */ 1092 static void cycle_netdev(struct net_device *dev, 1093 struct netlink_ext_ack *extack) 1094 { 1095 unsigned int flags = dev->flags; 1096 int ret; 1097 1098 if (!netif_running(dev)) 1099 return; 1100 1101 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1102 if (ret >= 0) 1103 ret = dev_change_flags(dev, flags, extack); 1104 1105 if (ret < 0) { 1106 netdev_err(dev, 1107 "Failed to cycle device %s; route tables might be wrong!\n", 1108 dev->name); 1109 } 1110 } 1111 1112 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1113 struct netlink_ext_ack *extack) 1114 { 1115 int ret; 1116 1117 /* do not allow loopback device to be enslaved to a VRF. 1118 * The vrf device acts as the loopback for the vrf. 1119 */ 1120 if (port_dev == dev_net(dev)->loopback_dev) { 1121 NL_SET_ERR_MSG(extack, 1122 "Can not enslave loopback device to a VRF"); 1123 return -EOPNOTSUPP; 1124 } 1125 1126 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1127 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1128 if (ret < 0) 1129 goto err; 1130 1131 cycle_netdev(port_dev, extack); 1132 1133 return 0; 1134 1135 err: 1136 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1137 return ret; 1138 } 1139 1140 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1141 struct netlink_ext_ack *extack) 1142 { 1143 if (netif_is_l3_master(port_dev)) { 1144 NL_SET_ERR_MSG(extack, 1145 "Can not enslave an L3 master device to a VRF"); 1146 return -EINVAL; 1147 } 1148 1149 if (netif_is_l3_slave(port_dev)) 1150 return -EINVAL; 1151 1152 return do_vrf_add_slave(dev, port_dev, extack); 1153 } 1154 1155 /* inverse of do_vrf_add_slave */ 1156 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1157 { 1158 netdev_upper_dev_unlink(port_dev, dev); 1159 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1160 1161 cycle_netdev(port_dev, NULL); 1162 1163 return 0; 1164 } 1165 1166 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1167 { 1168 return do_vrf_del_slave(dev, port_dev); 1169 } 1170 1171 static void vrf_dev_uninit(struct net_device *dev) 1172 { 1173 struct net_vrf *vrf = netdev_priv(dev); 1174 1175 vrf_rtable_release(dev, vrf); 1176 vrf_rt6_release(dev, vrf); 1177 1178 free_percpu(dev->dstats); 1179 dev->dstats = NULL; 1180 } 1181 1182 static int vrf_dev_init(struct net_device *dev) 1183 { 1184 struct net_vrf *vrf = netdev_priv(dev); 1185 1186 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 1187 if (!dev->dstats) 1188 goto out_nomem; 1189 1190 /* create the default dst which points back to us */ 1191 if (vrf_rtable_create(dev) != 0) 1192 goto out_stats; 1193 1194 if (vrf_rt6_create(dev) != 0) 1195 goto out_rth; 1196 1197 dev->flags = IFF_MASTER | IFF_NOARP; 1198 1199 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1200 dev->operstate = IF_OPER_UP; 1201 netdev_lockdep_set_classes(dev); 1202 return 0; 1203 1204 out_rth: 1205 vrf_rtable_release(dev, vrf); 1206 out_stats: 1207 free_percpu(dev->dstats); 1208 dev->dstats = NULL; 1209 out_nomem: 1210 return -ENOMEM; 1211 } 1212 1213 static const struct net_device_ops vrf_netdev_ops = { 1214 .ndo_init = vrf_dev_init, 1215 .ndo_uninit = vrf_dev_uninit, 1216 .ndo_start_xmit = vrf_xmit, 1217 .ndo_set_mac_address = eth_mac_addr, 1218 .ndo_get_stats64 = vrf_get_stats64, 1219 .ndo_add_slave = vrf_add_slave, 1220 .ndo_del_slave = vrf_del_slave, 1221 }; 1222 1223 static u32 vrf_fib_table(const struct net_device *dev) 1224 { 1225 struct net_vrf *vrf = netdev_priv(dev); 1226 1227 return vrf->tb_id; 1228 } 1229 1230 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1231 { 1232 kfree_skb(skb); 1233 return 0; 1234 } 1235 1236 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1237 struct sk_buff *skb, 1238 struct net_device *dev) 1239 { 1240 struct net *net = dev_net(dev); 1241 1242 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1243 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1244 1245 return skb; 1246 } 1247 1248 static int vrf_prepare_mac_header(struct sk_buff *skb, 1249 struct net_device *vrf_dev, u16 proto) 1250 { 1251 struct ethhdr *eth; 1252 int err; 1253 1254 /* in general, we do not know if there is enough space in the head of 1255 * the packet for hosting the mac header. 1256 */ 1257 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1258 if (unlikely(err)) 1259 /* no space in the skb head */ 1260 return -ENOBUFS; 1261 1262 __skb_push(skb, ETH_HLEN); 1263 eth = (struct ethhdr *)skb->data; 1264 1265 skb_reset_mac_header(skb); 1266 skb_reset_mac_len(skb); 1267 1268 /* we set the ethernet destination and the source addresses to the 1269 * address of the VRF device. 1270 */ 1271 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1272 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1273 eth->h_proto = htons(proto); 1274 1275 /* the destination address of the Ethernet frame corresponds to the 1276 * address set on the VRF interface; therefore, the packet is intended 1277 * to be processed locally. 1278 */ 1279 skb->protocol = eth->h_proto; 1280 skb->pkt_type = PACKET_HOST; 1281 1282 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1283 1284 skb_pull_inline(skb, ETH_HLEN); 1285 1286 return 0; 1287 } 1288 1289 /* prepare and add the mac header to the packet if it was not set previously. 1290 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1291 * If the mac header was already set, the original mac header is left 1292 * untouched and the function returns immediately. 1293 */ 1294 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1295 struct net_device *vrf_dev, 1296 u16 proto, struct net_device *orig_dev) 1297 { 1298 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev)) 1299 return 0; 1300 1301 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1302 } 1303 1304 #if IS_ENABLED(CONFIG_IPV6) 1305 /* neighbor handling is done with actual device; do not want 1306 * to flip skb->dev for those ndisc packets. This really fails 1307 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1308 * a start. 1309 */ 1310 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1311 { 1312 const struct ipv6hdr *iph = ipv6_hdr(skb); 1313 bool rc = false; 1314 1315 if (iph->nexthdr == NEXTHDR_ICMP) { 1316 const struct icmp6hdr *icmph; 1317 struct icmp6hdr _icmph; 1318 1319 icmph = skb_header_pointer(skb, sizeof(*iph), 1320 sizeof(_icmph), &_icmph); 1321 if (!icmph) 1322 goto out; 1323 1324 switch (icmph->icmp6_type) { 1325 case NDISC_ROUTER_SOLICITATION: 1326 case NDISC_ROUTER_ADVERTISEMENT: 1327 case NDISC_NEIGHBOUR_SOLICITATION: 1328 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1329 case NDISC_REDIRECT: 1330 rc = true; 1331 break; 1332 } 1333 } 1334 1335 out: 1336 return rc; 1337 } 1338 1339 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1340 const struct net_device *dev, 1341 struct flowi6 *fl6, 1342 int ifindex, 1343 const struct sk_buff *skb, 1344 int flags) 1345 { 1346 struct net_vrf *vrf = netdev_priv(dev); 1347 1348 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1349 } 1350 1351 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1352 int ifindex) 1353 { 1354 const struct ipv6hdr *iph = ipv6_hdr(skb); 1355 struct flowi6 fl6 = { 1356 .flowi6_iif = ifindex, 1357 .flowi6_mark = skb->mark, 1358 .flowi6_proto = iph->nexthdr, 1359 .daddr = iph->daddr, 1360 .saddr = iph->saddr, 1361 .flowlabel = ip6_flowinfo(iph), 1362 }; 1363 struct net *net = dev_net(vrf_dev); 1364 struct rt6_info *rt6; 1365 1366 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1367 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1368 if (unlikely(!rt6)) 1369 return; 1370 1371 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1372 return; 1373 1374 skb_dst_set(skb, &rt6->dst); 1375 } 1376 1377 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1378 struct sk_buff *skb) 1379 { 1380 int orig_iif = skb->skb_iif; 1381 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1382 bool is_ndisc = ipv6_ndisc_frame(skb); 1383 1384 /* loopback, multicast & non-ND link-local traffic; do not push through 1385 * packet taps again. Reset pkt_type for upper layers to process skb. 1386 * For non-loopback strict packets, determine the dst using the original 1387 * ifindex. 1388 */ 1389 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1390 skb->dev = vrf_dev; 1391 skb->skb_iif = vrf_dev->ifindex; 1392 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1393 1394 if (skb->pkt_type == PACKET_LOOPBACK) 1395 skb->pkt_type = PACKET_HOST; 1396 else 1397 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1398 1399 goto out; 1400 } 1401 1402 /* if packet is NDISC then keep the ingress interface */ 1403 if (!is_ndisc) { 1404 struct net_device *orig_dev = skb->dev; 1405 1406 vrf_rx_stats(vrf_dev, skb->len); 1407 skb->dev = vrf_dev; 1408 skb->skb_iif = vrf_dev->ifindex; 1409 1410 if (!list_empty(&vrf_dev->ptype_all)) { 1411 int err; 1412 1413 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1414 ETH_P_IPV6, 1415 orig_dev); 1416 if (likely(!err)) { 1417 skb_push(skb, skb->mac_len); 1418 dev_queue_xmit_nit(skb, vrf_dev); 1419 skb_pull(skb, skb->mac_len); 1420 } 1421 } 1422 1423 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1424 } 1425 1426 if (need_strict) 1427 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1428 1429 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1430 out: 1431 return skb; 1432 } 1433 1434 #else 1435 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1436 struct sk_buff *skb) 1437 { 1438 return skb; 1439 } 1440 #endif 1441 1442 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1443 struct sk_buff *skb) 1444 { 1445 struct net_device *orig_dev = skb->dev; 1446 1447 skb->dev = vrf_dev; 1448 skb->skb_iif = vrf_dev->ifindex; 1449 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1450 1451 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1452 goto out; 1453 1454 /* loopback traffic; do not push through packet taps again. 1455 * Reset pkt_type for upper layers to process skb 1456 */ 1457 if (skb->pkt_type == PACKET_LOOPBACK) { 1458 skb->pkt_type = PACKET_HOST; 1459 goto out; 1460 } 1461 1462 vrf_rx_stats(vrf_dev, skb->len); 1463 1464 if (!list_empty(&vrf_dev->ptype_all)) { 1465 int err; 1466 1467 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP, 1468 orig_dev); 1469 if (likely(!err)) { 1470 skb_push(skb, skb->mac_len); 1471 dev_queue_xmit_nit(skb, vrf_dev); 1472 skb_pull(skb, skb->mac_len); 1473 } 1474 } 1475 1476 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1477 out: 1478 return skb; 1479 } 1480 1481 /* called with rcu lock held */ 1482 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1483 struct sk_buff *skb, 1484 u16 proto) 1485 { 1486 switch (proto) { 1487 case AF_INET: 1488 return vrf_ip_rcv(vrf_dev, skb); 1489 case AF_INET6: 1490 return vrf_ip6_rcv(vrf_dev, skb); 1491 } 1492 1493 return skb; 1494 } 1495 1496 #if IS_ENABLED(CONFIG_IPV6) 1497 /* send to link-local or multicast address via interface enslaved to 1498 * VRF device. Force lookup to VRF table without changing flow struct 1499 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1500 * is taken on the dst by this function. 1501 */ 1502 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1503 struct flowi6 *fl6) 1504 { 1505 struct net *net = dev_net(dev); 1506 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1507 struct dst_entry *dst = NULL; 1508 struct rt6_info *rt; 1509 1510 /* VRF device does not have a link-local address and 1511 * sending packets to link-local or mcast addresses over 1512 * a VRF device does not make sense 1513 */ 1514 if (fl6->flowi6_oif == dev->ifindex) { 1515 dst = &net->ipv6.ip6_null_entry->dst; 1516 return dst; 1517 } 1518 1519 if (!ipv6_addr_any(&fl6->saddr)) 1520 flags |= RT6_LOOKUP_F_HAS_SADDR; 1521 1522 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1523 if (rt) 1524 dst = &rt->dst; 1525 1526 return dst; 1527 } 1528 #endif 1529 1530 static const struct l3mdev_ops vrf_l3mdev_ops = { 1531 .l3mdev_fib_table = vrf_fib_table, 1532 .l3mdev_l3_rcv = vrf_l3_rcv, 1533 .l3mdev_l3_out = vrf_l3_out, 1534 #if IS_ENABLED(CONFIG_IPV6) 1535 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1536 #endif 1537 }; 1538 1539 static void vrf_get_drvinfo(struct net_device *dev, 1540 struct ethtool_drvinfo *info) 1541 { 1542 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 1543 strscpy(info->version, DRV_VERSION, sizeof(info->version)); 1544 } 1545 1546 static const struct ethtool_ops vrf_ethtool_ops = { 1547 .get_drvinfo = vrf_get_drvinfo, 1548 }; 1549 1550 static inline size_t vrf_fib_rule_nl_size(void) 1551 { 1552 size_t sz; 1553 1554 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1555 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1556 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1557 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1558 1559 return sz; 1560 } 1561 1562 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1563 { 1564 struct fib_rule_hdr *frh; 1565 struct nlmsghdr *nlh; 1566 struct sk_buff *skb; 1567 int err; 1568 1569 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1570 !ipv6_mod_enabled()) 1571 return 0; 1572 1573 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1574 if (!skb) 1575 return -ENOMEM; 1576 1577 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1578 if (!nlh) 1579 goto nla_put_failure; 1580 1581 /* rule only needs to appear once */ 1582 nlh->nlmsg_flags |= NLM_F_EXCL; 1583 1584 frh = nlmsg_data(nlh); 1585 memset(frh, 0, sizeof(*frh)); 1586 frh->family = family; 1587 frh->action = FR_ACT_TO_TBL; 1588 1589 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1590 goto nla_put_failure; 1591 1592 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1593 goto nla_put_failure; 1594 1595 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1596 goto nla_put_failure; 1597 1598 nlmsg_end(skb, nlh); 1599 1600 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1601 skb->sk = dev_net(dev)->rtnl; 1602 if (add_it) { 1603 err = fib_nl_newrule(skb, nlh, NULL); 1604 if (err == -EEXIST) 1605 err = 0; 1606 } else { 1607 err = fib_nl_delrule(skb, nlh, NULL); 1608 if (err == -ENOENT) 1609 err = 0; 1610 } 1611 nlmsg_free(skb); 1612 1613 return err; 1614 1615 nla_put_failure: 1616 nlmsg_free(skb); 1617 1618 return -EMSGSIZE; 1619 } 1620 1621 static int vrf_add_fib_rules(const struct net_device *dev) 1622 { 1623 int err; 1624 1625 err = vrf_fib_rule(dev, AF_INET, true); 1626 if (err < 0) 1627 goto out_err; 1628 1629 err = vrf_fib_rule(dev, AF_INET6, true); 1630 if (err < 0) 1631 goto ipv6_err; 1632 1633 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1634 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1635 if (err < 0) 1636 goto ipmr_err; 1637 #endif 1638 1639 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1640 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1641 if (err < 0) 1642 goto ip6mr_err; 1643 #endif 1644 1645 return 0; 1646 1647 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1648 ip6mr_err: 1649 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1650 #endif 1651 1652 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1653 ipmr_err: 1654 vrf_fib_rule(dev, AF_INET6, false); 1655 #endif 1656 1657 ipv6_err: 1658 vrf_fib_rule(dev, AF_INET, false); 1659 1660 out_err: 1661 netdev_err(dev, "Failed to add FIB rules.\n"); 1662 return err; 1663 } 1664 1665 static void vrf_setup(struct net_device *dev) 1666 { 1667 ether_setup(dev); 1668 1669 /* Initialize the device structure. */ 1670 dev->netdev_ops = &vrf_netdev_ops; 1671 dev->l3mdev_ops = &vrf_l3mdev_ops; 1672 dev->ethtool_ops = &vrf_ethtool_ops; 1673 dev->needs_free_netdev = true; 1674 1675 /* Fill in device structure with ethernet-generic values. */ 1676 eth_hw_addr_random(dev); 1677 1678 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1679 dev->features |= NETIF_F_LLTX; 1680 1681 /* don't allow vrf devices to change network namespaces. */ 1682 dev->features |= NETIF_F_NETNS_LOCAL; 1683 1684 /* does not make sense for a VLAN to be added to a vrf device */ 1685 dev->features |= NETIF_F_VLAN_CHALLENGED; 1686 1687 /* enable offload features */ 1688 dev->features |= NETIF_F_GSO_SOFTWARE; 1689 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1690 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1691 1692 dev->hw_features = dev->features; 1693 dev->hw_enc_features = dev->features; 1694 1695 /* default to no qdisc; user can add if desired */ 1696 dev->priv_flags |= IFF_NO_QUEUE; 1697 dev->priv_flags |= IFF_NO_RX_HANDLER; 1698 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1699 1700 /* VRF devices do not care about MTU, but if the MTU is set 1701 * too low then the ipv4 and ipv6 protocols are disabled 1702 * which breaks networking. 1703 */ 1704 dev->min_mtu = IPV6_MIN_MTU; 1705 dev->max_mtu = IP6_MAX_MTU; 1706 dev->mtu = dev->max_mtu; 1707 } 1708 1709 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1710 struct netlink_ext_ack *extack) 1711 { 1712 if (tb[IFLA_ADDRESS]) { 1713 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1714 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1715 return -EINVAL; 1716 } 1717 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1718 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1719 return -EADDRNOTAVAIL; 1720 } 1721 } 1722 return 0; 1723 } 1724 1725 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1726 { 1727 struct net_device *port_dev; 1728 struct list_head *iter; 1729 1730 netdev_for_each_lower_dev(dev, port_dev, iter) 1731 vrf_del_slave(dev, port_dev); 1732 1733 vrf_map_unregister_dev(dev); 1734 1735 unregister_netdevice_queue(dev, head); 1736 } 1737 1738 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1739 struct nlattr *tb[], struct nlattr *data[], 1740 struct netlink_ext_ack *extack) 1741 { 1742 struct net_vrf *vrf = netdev_priv(dev); 1743 struct netns_vrf *nn_vrf; 1744 bool *add_fib_rules; 1745 struct net *net; 1746 int err; 1747 1748 if (!data || !data[IFLA_VRF_TABLE]) { 1749 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1750 return -EINVAL; 1751 } 1752 1753 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1754 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1755 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1756 "Invalid VRF table id"); 1757 return -EINVAL; 1758 } 1759 1760 dev->priv_flags |= IFF_L3MDEV_MASTER; 1761 1762 err = register_netdevice(dev); 1763 if (err) 1764 goto out; 1765 1766 /* mapping between table_id and vrf; 1767 * note: such binding could not be done in the dev init function 1768 * because dev->ifindex id is not available yet. 1769 */ 1770 vrf->ifindex = dev->ifindex; 1771 1772 err = vrf_map_register_dev(dev, extack); 1773 if (err) { 1774 unregister_netdevice(dev); 1775 goto out; 1776 } 1777 1778 net = dev_net(dev); 1779 nn_vrf = net_generic(net, vrf_net_id); 1780 1781 add_fib_rules = &nn_vrf->add_fib_rules; 1782 if (*add_fib_rules) { 1783 err = vrf_add_fib_rules(dev); 1784 if (err) { 1785 vrf_map_unregister_dev(dev); 1786 unregister_netdevice(dev); 1787 goto out; 1788 } 1789 *add_fib_rules = false; 1790 } 1791 1792 out: 1793 return err; 1794 } 1795 1796 static size_t vrf_nl_getsize(const struct net_device *dev) 1797 { 1798 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1799 } 1800 1801 static int vrf_fillinfo(struct sk_buff *skb, 1802 const struct net_device *dev) 1803 { 1804 struct net_vrf *vrf = netdev_priv(dev); 1805 1806 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1807 } 1808 1809 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1810 const struct net_device *slave_dev) 1811 { 1812 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1813 } 1814 1815 static int vrf_fill_slave_info(struct sk_buff *skb, 1816 const struct net_device *vrf_dev, 1817 const struct net_device *slave_dev) 1818 { 1819 struct net_vrf *vrf = netdev_priv(vrf_dev); 1820 1821 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1822 return -EMSGSIZE; 1823 1824 return 0; 1825 } 1826 1827 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1828 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1829 }; 1830 1831 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1832 .kind = DRV_NAME, 1833 .priv_size = sizeof(struct net_vrf), 1834 1835 .get_size = vrf_nl_getsize, 1836 .policy = vrf_nl_policy, 1837 .validate = vrf_validate, 1838 .fill_info = vrf_fillinfo, 1839 1840 .get_slave_size = vrf_get_slave_size, 1841 .fill_slave_info = vrf_fill_slave_info, 1842 1843 .newlink = vrf_newlink, 1844 .dellink = vrf_dellink, 1845 .setup = vrf_setup, 1846 .maxtype = IFLA_VRF_MAX, 1847 }; 1848 1849 static int vrf_device_event(struct notifier_block *unused, 1850 unsigned long event, void *ptr) 1851 { 1852 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1853 1854 /* only care about unregister events to drop slave references */ 1855 if (event == NETDEV_UNREGISTER) { 1856 struct net_device *vrf_dev; 1857 1858 if (!netif_is_l3_slave(dev)) 1859 goto out; 1860 1861 vrf_dev = netdev_master_upper_dev_get(dev); 1862 vrf_del_slave(vrf_dev, dev); 1863 } 1864 out: 1865 return NOTIFY_DONE; 1866 } 1867 1868 static struct notifier_block vrf_notifier_block __read_mostly = { 1869 .notifier_call = vrf_device_event, 1870 }; 1871 1872 static int vrf_map_init(struct vrf_map *vmap) 1873 { 1874 spin_lock_init(&vmap->vmap_lock); 1875 hash_init(vmap->ht); 1876 1877 vmap->strict_mode = false; 1878 1879 return 0; 1880 } 1881 1882 #ifdef CONFIG_SYSCTL 1883 static bool vrf_strict_mode(struct vrf_map *vmap) 1884 { 1885 bool strict_mode; 1886 1887 vrf_map_lock(vmap); 1888 strict_mode = vmap->strict_mode; 1889 vrf_map_unlock(vmap); 1890 1891 return strict_mode; 1892 } 1893 1894 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1895 { 1896 bool *cur_mode; 1897 int res = 0; 1898 1899 vrf_map_lock(vmap); 1900 1901 cur_mode = &vmap->strict_mode; 1902 if (*cur_mode == new_mode) 1903 goto unlock; 1904 1905 if (*cur_mode) { 1906 /* disable strict mode */ 1907 *cur_mode = false; 1908 } else { 1909 if (vmap->shared_tables) { 1910 /* we cannot allow strict_mode because there are some 1911 * vrfs that share one or more tables. 1912 */ 1913 res = -EBUSY; 1914 goto unlock; 1915 } 1916 1917 /* no tables are shared among vrfs, so we can go back 1918 * to 1:1 association between a vrf with its table. 1919 */ 1920 *cur_mode = true; 1921 } 1922 1923 unlock: 1924 vrf_map_unlock(vmap); 1925 1926 return res; 1927 } 1928 1929 static int vrf_shared_table_handler(struct ctl_table *table, int write, 1930 void *buffer, size_t *lenp, loff_t *ppos) 1931 { 1932 struct net *net = (struct net *)table->extra1; 1933 struct vrf_map *vmap = netns_vrf_map(net); 1934 int proc_strict_mode = 0; 1935 struct ctl_table tmp = { 1936 .procname = table->procname, 1937 .data = &proc_strict_mode, 1938 .maxlen = sizeof(int), 1939 .mode = table->mode, 1940 .extra1 = SYSCTL_ZERO, 1941 .extra2 = SYSCTL_ONE, 1942 }; 1943 int ret; 1944 1945 if (!write) 1946 proc_strict_mode = vrf_strict_mode(vmap); 1947 1948 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1949 1950 if (write && ret == 0) 1951 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1952 1953 return ret; 1954 } 1955 1956 static const struct ctl_table vrf_table[] = { 1957 { 1958 .procname = "strict_mode", 1959 .data = NULL, 1960 .maxlen = sizeof(int), 1961 .mode = 0644, 1962 .proc_handler = vrf_shared_table_handler, 1963 /* set by the vrf_netns_init */ 1964 .extra1 = NULL, 1965 }, 1966 { }, 1967 }; 1968 1969 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1970 { 1971 struct ctl_table *table; 1972 1973 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1974 if (!table) 1975 return -ENOMEM; 1976 1977 /* init the extra1 parameter with the reference to current netns */ 1978 table[0].extra1 = net; 1979 1980 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table); 1981 if (!nn_vrf->ctl_hdr) { 1982 kfree(table); 1983 return -ENOMEM; 1984 } 1985 1986 return 0; 1987 } 1988 1989 static void vrf_netns_exit_sysctl(struct net *net) 1990 { 1991 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1992 struct ctl_table *table; 1993 1994 table = nn_vrf->ctl_hdr->ctl_table_arg; 1995 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1996 kfree(table); 1997 } 1998 #else 1999 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 2000 { 2001 return 0; 2002 } 2003 2004 static void vrf_netns_exit_sysctl(struct net *net) 2005 { 2006 } 2007 #endif 2008 2009 /* Initialize per network namespace state */ 2010 static int __net_init vrf_netns_init(struct net *net) 2011 { 2012 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 2013 2014 nn_vrf->add_fib_rules = true; 2015 vrf_map_init(&nn_vrf->vmap); 2016 2017 return vrf_netns_init_sysctl(net, nn_vrf); 2018 } 2019 2020 static void __net_exit vrf_netns_exit(struct net *net) 2021 { 2022 vrf_netns_exit_sysctl(net); 2023 } 2024 2025 static struct pernet_operations vrf_net_ops __net_initdata = { 2026 .init = vrf_netns_init, 2027 .exit = vrf_netns_exit, 2028 .id = &vrf_net_id, 2029 .size = sizeof(struct netns_vrf), 2030 }; 2031 2032 static int __init vrf_init_module(void) 2033 { 2034 int rc; 2035 2036 register_netdevice_notifier(&vrf_notifier_block); 2037 2038 rc = register_pernet_subsys(&vrf_net_ops); 2039 if (rc < 0) 2040 goto error; 2041 2042 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2043 vrf_ifindex_lookup_by_table_id); 2044 if (rc < 0) 2045 goto unreg_pernet; 2046 2047 rc = rtnl_link_register(&vrf_link_ops); 2048 if (rc < 0) 2049 goto table_lookup_unreg; 2050 2051 return 0; 2052 2053 table_lookup_unreg: 2054 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2055 vrf_ifindex_lookup_by_table_id); 2056 2057 unreg_pernet: 2058 unregister_pernet_subsys(&vrf_net_ops); 2059 2060 error: 2061 unregister_netdevice_notifier(&vrf_notifier_block); 2062 return rc; 2063 } 2064 2065 module_init(vrf_init_module); 2066 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2067 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2068 MODULE_LICENSE("GPL"); 2069 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2070 MODULE_VERSION(DRV_VERSION); 2071