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 static void vrf_rx_stats(struct net_device *dev, int len) 125 { 126 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 127 128 u64_stats_update_begin(&dstats->syncp); 129 u64_stats_inc(&dstats->rx_packets); 130 u64_stats_add(&dstats->rx_bytes, len); 131 u64_stats_update_end(&dstats->syncp); 132 } 133 134 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 135 { 136 vrf_dev->stats.tx_errors++; 137 kfree_skb(skb); 138 } 139 140 static struct vrf_map *netns_vrf_map(struct net *net) 141 { 142 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 143 144 return &nn_vrf->vmap; 145 } 146 147 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 148 { 149 return netns_vrf_map(dev_net(dev)); 150 } 151 152 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 153 { 154 struct list_head *me_head = &me->vrf_list; 155 struct net_vrf *vrf; 156 157 if (list_empty(me_head)) 158 return -ENODEV; 159 160 vrf = list_first_entry(me_head, struct net_vrf, me_list); 161 162 return vrf->ifindex; 163 } 164 165 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 166 { 167 struct vrf_map_elem *me; 168 169 me = kmalloc(sizeof(*me), flags); 170 if (!me) 171 return NULL; 172 173 return me; 174 } 175 176 static void vrf_map_elem_free(struct vrf_map_elem *me) 177 { 178 kfree(me); 179 } 180 181 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 182 int ifindex, int users) 183 { 184 me->table_id = table_id; 185 me->ifindex = ifindex; 186 me->users = users; 187 INIT_LIST_HEAD(&me->vrf_list); 188 } 189 190 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 191 u32 table_id) 192 { 193 struct vrf_map_elem *me; 194 u32 key; 195 196 key = jhash_1word(table_id, HASH_INITVAL); 197 hash_for_each_possible(vmap->ht, me, hnode, key) { 198 if (me->table_id == table_id) 199 return me; 200 } 201 202 return NULL; 203 } 204 205 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 206 { 207 u32 table_id = me->table_id; 208 u32 key; 209 210 key = jhash_1word(table_id, HASH_INITVAL); 211 hash_add(vmap->ht, &me->hnode, key); 212 } 213 214 static void vrf_map_del_elem(struct vrf_map_elem *me) 215 { 216 hash_del(&me->hnode); 217 } 218 219 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 220 { 221 spin_lock(&vmap->vmap_lock); 222 } 223 224 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 225 { 226 spin_unlock(&vmap->vmap_lock); 227 } 228 229 /* called with rtnl lock held */ 230 static int 231 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 232 { 233 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 234 struct net_vrf *vrf = netdev_priv(dev); 235 struct vrf_map_elem *new_me, *me; 236 u32 table_id = vrf->tb_id; 237 bool free_new_me = false; 238 int users; 239 int res; 240 241 /* we pre-allocate elements used in the spin-locked section (so that we 242 * keep the spinlock as short as possible). 243 */ 244 new_me = vrf_map_elem_alloc(GFP_KERNEL); 245 if (!new_me) 246 return -ENOMEM; 247 248 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 249 250 vrf_map_lock(vmap); 251 252 me = vrf_map_lookup_elem(vmap, table_id); 253 if (!me) { 254 me = new_me; 255 vrf_map_add_elem(vmap, me); 256 goto link_vrf; 257 } 258 259 /* we already have an entry in the vrf_map, so it means there is (at 260 * least) a vrf registered on the specific table. 261 */ 262 free_new_me = true; 263 if (vmap->strict_mode) { 264 /* vrfs cannot share the same table */ 265 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 266 res = -EBUSY; 267 goto unlock; 268 } 269 270 link_vrf: 271 users = ++me->users; 272 if (users == 2) 273 ++vmap->shared_tables; 274 275 list_add(&vrf->me_list, &me->vrf_list); 276 277 res = 0; 278 279 unlock: 280 vrf_map_unlock(vmap); 281 282 /* clean-up, if needed */ 283 if (free_new_me) 284 vrf_map_elem_free(new_me); 285 286 return res; 287 } 288 289 /* called with rtnl lock held */ 290 static void vrf_map_unregister_dev(struct net_device *dev) 291 { 292 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 293 struct net_vrf *vrf = netdev_priv(dev); 294 u32 table_id = vrf->tb_id; 295 struct vrf_map_elem *me; 296 int users; 297 298 vrf_map_lock(vmap); 299 300 me = vrf_map_lookup_elem(vmap, table_id); 301 if (!me) 302 goto unlock; 303 304 list_del(&vrf->me_list); 305 306 users = --me->users; 307 if (users == 1) { 308 --vmap->shared_tables; 309 } else if (users == 0) { 310 vrf_map_del_elem(me); 311 312 /* no one will refer to this element anymore */ 313 vrf_map_elem_free(me); 314 } 315 316 unlock: 317 vrf_map_unlock(vmap); 318 } 319 320 /* return the vrf device index associated with the table_id */ 321 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 322 { 323 struct vrf_map *vmap = netns_vrf_map(net); 324 struct vrf_map_elem *me; 325 int ifindex; 326 327 vrf_map_lock(vmap); 328 329 if (!vmap->strict_mode) { 330 ifindex = -EPERM; 331 goto unlock; 332 } 333 334 me = vrf_map_lookup_elem(vmap, table_id); 335 if (!me) { 336 ifindex = -ENODEV; 337 goto unlock; 338 } 339 340 ifindex = vrf_map_elem_get_vrf_ifindex(me); 341 342 unlock: 343 vrf_map_unlock(vmap); 344 345 return ifindex; 346 } 347 348 /* by default VRF devices do not have a qdisc and are expected 349 * to be created with only a single queue. 350 */ 351 static bool qdisc_tx_is_default(const struct net_device *dev) 352 { 353 struct netdev_queue *txq; 354 struct Qdisc *qdisc; 355 356 if (dev->num_tx_queues > 1) 357 return false; 358 359 txq = netdev_get_tx_queue(dev, 0); 360 qdisc = rcu_access_pointer(txq->qdisc); 361 362 return !qdisc->enqueue; 363 } 364 365 /* Local traffic destined to local address. Reinsert the packet to rx 366 * path, similar to loopback handling. 367 */ 368 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 369 struct dst_entry *dst) 370 { 371 int len = skb->len; 372 373 skb_orphan(skb); 374 375 skb_dst_set(skb, dst); 376 377 /* set pkt_type to avoid skb hitting packet taps twice - 378 * once on Tx and again in Rx processing 379 */ 380 skb->pkt_type = PACKET_LOOPBACK; 381 382 skb->protocol = eth_type_trans(skb, dev); 383 384 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) { 385 vrf_rx_stats(dev, len); 386 } else { 387 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 388 389 u64_stats_update_begin(&dstats->syncp); 390 u64_stats_inc(&dstats->rx_drops); 391 u64_stats_update_end(&dstats->syncp); 392 } 393 394 return NETDEV_TX_OK; 395 } 396 397 static void vrf_nf_set_untracked(struct sk_buff *skb) 398 { 399 if (skb_get_nfct(skb) == 0) 400 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 401 } 402 403 static void vrf_nf_reset_ct(struct sk_buff *skb) 404 { 405 if (skb_get_nfct(skb) == IP_CT_UNTRACKED) 406 nf_reset_ct(skb); 407 } 408 409 #if IS_ENABLED(CONFIG_IPV6) 410 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 411 struct sk_buff *skb) 412 { 413 int err; 414 415 vrf_nf_reset_ct(skb); 416 417 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 418 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 419 420 if (likely(err == 1)) 421 err = dst_output(net, sk, skb); 422 423 return err; 424 } 425 426 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 427 struct net_device *dev) 428 { 429 const struct ipv6hdr *iph; 430 struct net *net = dev_net(skb->dev); 431 struct flowi6 fl6; 432 int ret = NET_XMIT_DROP; 433 struct dst_entry *dst; 434 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 435 436 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 437 goto err; 438 439 iph = ipv6_hdr(skb); 440 441 memset(&fl6, 0, sizeof(fl6)); 442 /* needed to match OIF rule */ 443 fl6.flowi6_l3mdev = dev->ifindex; 444 fl6.flowi6_iif = LOOPBACK_IFINDEX; 445 fl6.daddr = iph->daddr; 446 fl6.saddr = iph->saddr; 447 fl6.flowlabel = ip6_flowinfo(iph); 448 fl6.flowi6_mark = skb->mark; 449 fl6.flowi6_proto = iph->nexthdr; 450 451 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 452 if (IS_ERR(dst) || dst == dst_null) 453 goto err; 454 455 skb_dst_drop(skb); 456 457 /* if dst.dev is the VRF device again this is locally originated traffic 458 * destined to a local address. Short circuit to Rx path. 459 */ 460 if (dst->dev == dev) 461 return vrf_local_xmit(skb, dev, dst); 462 463 skb_dst_set(skb, dst); 464 465 /* strip the ethernet header added for pass through VRF device */ 466 __skb_pull(skb, skb_network_offset(skb)); 467 468 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb))); 469 ret = vrf_ip6_local_out(net, skb->sk, skb); 470 if (unlikely(net_xmit_eval(ret))) 471 dev->stats.tx_errors++; 472 else 473 ret = NET_XMIT_SUCCESS; 474 475 return ret; 476 err: 477 vrf_tx_error(dev, skb); 478 return NET_XMIT_DROP; 479 } 480 #else 481 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 482 struct net_device *dev) 483 { 484 vrf_tx_error(dev, skb); 485 return NET_XMIT_DROP; 486 } 487 #endif 488 489 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 490 static int vrf_ip_local_out(struct net *net, struct sock *sk, 491 struct sk_buff *skb) 492 { 493 int err; 494 495 vrf_nf_reset_ct(skb); 496 497 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 498 skb, NULL, skb_dst(skb)->dev, dst_output); 499 if (likely(err == 1)) 500 err = dst_output(net, sk, skb); 501 502 return err; 503 } 504 505 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 506 struct net_device *vrf_dev) 507 { 508 struct iphdr *ip4h; 509 int ret = NET_XMIT_DROP; 510 struct flowi4 fl4; 511 struct net *net = dev_net(vrf_dev); 512 struct rtable *rt; 513 514 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 515 goto err; 516 517 ip4h = ip_hdr(skb); 518 519 memset(&fl4, 0, sizeof(fl4)); 520 /* needed to match OIF rule */ 521 fl4.flowi4_l3mdev = vrf_dev->ifindex; 522 fl4.flowi4_iif = LOOPBACK_IFINDEX; 523 fl4.flowi4_tos = RT_TOS(ip4h->tos); 524 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC; 525 fl4.flowi4_proto = ip4h->protocol; 526 fl4.daddr = ip4h->daddr; 527 fl4.saddr = ip4h->saddr; 528 529 rt = ip_route_output_flow(net, &fl4, NULL); 530 if (IS_ERR(rt)) 531 goto err; 532 533 skb_dst_drop(skb); 534 535 /* if dst.dev is the VRF device again this is locally originated traffic 536 * destined to a local address. Short circuit to Rx path. 537 */ 538 if (rt->dst.dev == vrf_dev) 539 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 540 541 skb_dst_set(skb, &rt->dst); 542 543 /* strip the ethernet header added for pass through VRF device */ 544 __skb_pull(skb, skb_network_offset(skb)); 545 546 if (!ip4h->saddr) { 547 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 548 RT_SCOPE_LINK); 549 } 550 551 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 552 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 553 if (unlikely(net_xmit_eval(ret))) 554 vrf_dev->stats.tx_errors++; 555 else 556 ret = NET_XMIT_SUCCESS; 557 558 out: 559 return ret; 560 err: 561 vrf_tx_error(vrf_dev, skb); 562 goto out; 563 } 564 565 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 566 { 567 switch (skb->protocol) { 568 case htons(ETH_P_IP): 569 return vrf_process_v4_outbound(skb, dev); 570 case htons(ETH_P_IPV6): 571 return vrf_process_v6_outbound(skb, dev); 572 default: 573 vrf_tx_error(dev, skb); 574 return NET_XMIT_DROP; 575 } 576 } 577 578 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 579 { 580 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); 581 582 int len = skb->len; 583 netdev_tx_t ret = is_ip_tx_frame(skb, dev); 584 585 u64_stats_update_begin(&dstats->syncp); 586 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 587 588 u64_stats_inc(&dstats->tx_packets); 589 u64_stats_add(&dstats->tx_bytes, len); 590 } else { 591 u64_stats_inc(&dstats->tx_drops); 592 } 593 u64_stats_update_end(&dstats->syncp); 594 595 return ret; 596 } 597 598 static void vrf_finish_direct(struct sk_buff *skb) 599 { 600 struct net_device *vrf_dev = skb->dev; 601 602 if (!list_empty(&vrf_dev->ptype_all) && 603 likely(skb_headroom(skb) >= ETH_HLEN)) { 604 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 605 606 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 607 eth_zero_addr(eth->h_dest); 608 eth->h_proto = skb->protocol; 609 610 dev_queue_xmit_nit(skb, vrf_dev); 611 612 skb_pull(skb, ETH_HLEN); 613 } 614 615 vrf_nf_reset_ct(skb); 616 } 617 618 #if IS_ENABLED(CONFIG_IPV6) 619 /* modelled after ip6_finish_output2 */ 620 static int vrf_finish_output6(struct net *net, struct sock *sk, 621 struct sk_buff *skb) 622 { 623 struct dst_entry *dst = skb_dst(skb); 624 struct net_device *dev = dst->dev; 625 const struct in6_addr *nexthop; 626 struct neighbour *neigh; 627 int ret; 628 629 vrf_nf_reset_ct(skb); 630 631 skb->protocol = htons(ETH_P_IPV6); 632 skb->dev = dev; 633 634 rcu_read_lock(); 635 nexthop = rt6_nexthop(dst_rt6_info(dst), &ipv6_hdr(skb)->daddr); 636 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 637 if (unlikely(!neigh)) 638 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 639 if (!IS_ERR(neigh)) { 640 sock_confirm_neigh(skb, neigh); 641 ret = neigh_output(neigh, skb, false); 642 rcu_read_unlock(); 643 return ret; 644 } 645 rcu_read_unlock(); 646 647 IP6_INC_STATS(dev_net(dst->dev), 648 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 649 kfree_skb(skb); 650 return -EINVAL; 651 } 652 653 /* modelled after ip6_output */ 654 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 655 { 656 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 657 net, sk, skb, NULL, skb_dst(skb)->dev, 658 vrf_finish_output6, 659 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 660 } 661 662 /* set dst on skb to send packet to us via dev_xmit path. Allows 663 * packet to go through device based features such as qdisc, netfilter 664 * hooks and packet sockets with skb->dev set to vrf device. 665 */ 666 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 667 struct sk_buff *skb) 668 { 669 struct net_vrf *vrf = netdev_priv(vrf_dev); 670 struct dst_entry *dst = NULL; 671 struct rt6_info *rt6; 672 673 rcu_read_lock(); 674 675 rt6 = rcu_dereference(vrf->rt6); 676 if (likely(rt6)) { 677 dst = &rt6->dst; 678 dst_hold(dst); 679 } 680 681 rcu_read_unlock(); 682 683 if (unlikely(!dst)) { 684 vrf_tx_error(vrf_dev, skb); 685 return NULL; 686 } 687 688 skb_dst_drop(skb); 689 skb_dst_set(skb, dst); 690 691 return skb; 692 } 693 694 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 695 struct sk_buff *skb) 696 { 697 vrf_finish_direct(skb); 698 699 return vrf_ip6_local_out(net, sk, skb); 700 } 701 702 static int vrf_output6_direct(struct net *net, struct sock *sk, 703 struct sk_buff *skb) 704 { 705 int err = 1; 706 707 skb->protocol = htons(ETH_P_IPV6); 708 709 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 710 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 711 NULL, skb->dev, vrf_output6_direct_finish); 712 713 if (likely(err == 1)) 714 vrf_finish_direct(skb); 715 716 return err; 717 } 718 719 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 720 struct sk_buff *skb) 721 { 722 int err; 723 724 err = vrf_output6_direct(net, sk, skb); 725 if (likely(err == 1)) 726 err = vrf_ip6_local_out(net, sk, skb); 727 728 return err; 729 } 730 731 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 732 struct sock *sk, 733 struct sk_buff *skb) 734 { 735 struct net *net = dev_net(vrf_dev); 736 int err; 737 738 skb->dev = vrf_dev; 739 740 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 741 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 742 743 if (likely(err == 1)) 744 err = vrf_output6_direct(net, sk, skb); 745 746 if (likely(err == 1)) 747 return skb; 748 749 return NULL; 750 } 751 752 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 753 struct sock *sk, 754 struct sk_buff *skb) 755 { 756 /* don't divert link scope packets */ 757 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 758 return skb; 759 760 vrf_nf_set_untracked(skb); 761 762 if (qdisc_tx_is_default(vrf_dev) || 763 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 764 return vrf_ip6_out_direct(vrf_dev, sk, skb); 765 766 return vrf_ip6_out_redirect(vrf_dev, skb); 767 } 768 769 /* holding rtnl */ 770 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 771 { 772 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 773 struct net *net = dev_net(dev); 774 struct dst_entry *dst; 775 776 RCU_INIT_POINTER(vrf->rt6, NULL); 777 synchronize_rcu(); 778 779 /* move dev in dst's to loopback so this VRF device can be deleted 780 * - based on dst_ifdown 781 */ 782 if (rt6) { 783 dst = &rt6->dst; 784 netdev_ref_replace(dst->dev, net->loopback_dev, 785 &dst->dev_tracker, GFP_KERNEL); 786 dst->dev = net->loopback_dev; 787 dst_release(dst); 788 } 789 } 790 791 static int vrf_rt6_create(struct net_device *dev) 792 { 793 int flags = DST_NOPOLICY | DST_NOXFRM; 794 struct net_vrf *vrf = netdev_priv(dev); 795 struct net *net = dev_net(dev); 796 struct rt6_info *rt6; 797 int rc = -ENOMEM; 798 799 /* IPv6 can be CONFIG enabled and then disabled runtime */ 800 if (!ipv6_mod_enabled()) 801 return 0; 802 803 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 804 if (!vrf->fib6_table) 805 goto out; 806 807 /* create a dst for routing packets out a VRF device */ 808 rt6 = ip6_dst_alloc(net, dev, flags); 809 if (!rt6) 810 goto out; 811 812 rt6->dst.output = vrf_output6; 813 814 rcu_assign_pointer(vrf->rt6, rt6); 815 816 rc = 0; 817 out: 818 return rc; 819 } 820 #else 821 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 822 struct sock *sk, 823 struct sk_buff *skb) 824 { 825 return skb; 826 } 827 828 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 829 { 830 } 831 832 static int vrf_rt6_create(struct net_device *dev) 833 { 834 return 0; 835 } 836 #endif 837 838 /* modelled after ip_finish_output2 */ 839 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 840 { 841 struct dst_entry *dst = skb_dst(skb); 842 struct rtable *rt = dst_rtable(dst); 843 struct net_device *dev = dst->dev; 844 unsigned int hh_len = LL_RESERVED_SPACE(dev); 845 struct neighbour *neigh; 846 bool is_v6gw = false; 847 848 vrf_nf_reset_ct(skb); 849 850 /* Be paranoid, rather than too clever. */ 851 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 852 skb = skb_expand_head(skb, hh_len); 853 if (!skb) { 854 dev->stats.tx_errors++; 855 return -ENOMEM; 856 } 857 } 858 859 rcu_read_lock(); 860 861 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 862 if (!IS_ERR(neigh)) { 863 int ret; 864 865 sock_confirm_neigh(skb, neigh); 866 /* if crossing protocols, can not use the cached header */ 867 ret = neigh_output(neigh, skb, is_v6gw); 868 rcu_read_unlock(); 869 return ret; 870 } 871 872 rcu_read_unlock(); 873 vrf_tx_error(skb->dev, skb); 874 return -EINVAL; 875 } 876 877 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 878 { 879 struct net_device *dev = skb_dst(skb)->dev; 880 881 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 882 883 skb->dev = dev; 884 skb->protocol = htons(ETH_P_IP); 885 886 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 887 net, sk, skb, NULL, dev, 888 vrf_finish_output, 889 !(IPCB(skb)->flags & IPSKB_REROUTED)); 890 } 891 892 /* set dst on skb to send packet to us via dev_xmit path. Allows 893 * packet to go through device based features such as qdisc, netfilter 894 * hooks and packet sockets with skb->dev set to vrf device. 895 */ 896 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 897 struct sk_buff *skb) 898 { 899 struct net_vrf *vrf = netdev_priv(vrf_dev); 900 struct dst_entry *dst = NULL; 901 struct rtable *rth; 902 903 rcu_read_lock(); 904 905 rth = rcu_dereference(vrf->rth); 906 if (likely(rth)) { 907 dst = &rth->dst; 908 dst_hold(dst); 909 } 910 911 rcu_read_unlock(); 912 913 if (unlikely(!dst)) { 914 vrf_tx_error(vrf_dev, skb); 915 return NULL; 916 } 917 918 skb_dst_drop(skb); 919 skb_dst_set(skb, dst); 920 921 return skb; 922 } 923 924 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 925 struct sk_buff *skb) 926 { 927 vrf_finish_direct(skb); 928 929 return vrf_ip_local_out(net, sk, skb); 930 } 931 932 static int vrf_output_direct(struct net *net, struct sock *sk, 933 struct sk_buff *skb) 934 { 935 int err = 1; 936 937 skb->protocol = htons(ETH_P_IP); 938 939 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 940 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 941 NULL, skb->dev, vrf_output_direct_finish); 942 943 if (likely(err == 1)) 944 vrf_finish_direct(skb); 945 946 return err; 947 } 948 949 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 950 struct sk_buff *skb) 951 { 952 int err; 953 954 err = vrf_output_direct(net, sk, skb); 955 if (likely(err == 1)) 956 err = vrf_ip_local_out(net, sk, skb); 957 958 return err; 959 } 960 961 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 962 struct sock *sk, 963 struct sk_buff *skb) 964 { 965 struct net *net = dev_net(vrf_dev); 966 int err; 967 968 skb->dev = vrf_dev; 969 970 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 971 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 972 973 if (likely(err == 1)) 974 err = vrf_output_direct(net, sk, skb); 975 976 if (likely(err == 1)) 977 return skb; 978 979 return NULL; 980 } 981 982 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 983 struct sock *sk, 984 struct sk_buff *skb) 985 { 986 /* don't divert multicast or local broadcast */ 987 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 988 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 989 return skb; 990 991 vrf_nf_set_untracked(skb); 992 993 if (qdisc_tx_is_default(vrf_dev) || 994 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 995 return vrf_ip_out_direct(vrf_dev, sk, skb); 996 997 return vrf_ip_out_redirect(vrf_dev, skb); 998 } 999 1000 /* called with rcu lock held */ 1001 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 1002 struct sock *sk, 1003 struct sk_buff *skb, 1004 u16 proto) 1005 { 1006 switch (proto) { 1007 case AF_INET: 1008 return vrf_ip_out(vrf_dev, sk, skb); 1009 case AF_INET6: 1010 return vrf_ip6_out(vrf_dev, sk, skb); 1011 } 1012 1013 return skb; 1014 } 1015 1016 /* holding rtnl */ 1017 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 1018 { 1019 struct rtable *rth = rtnl_dereference(vrf->rth); 1020 struct net *net = dev_net(dev); 1021 struct dst_entry *dst; 1022 1023 RCU_INIT_POINTER(vrf->rth, NULL); 1024 synchronize_rcu(); 1025 1026 /* move dev in dst's to loopback so this VRF device can be deleted 1027 * - based on dst_ifdown 1028 */ 1029 if (rth) { 1030 dst = &rth->dst; 1031 netdev_ref_replace(dst->dev, net->loopback_dev, 1032 &dst->dev_tracker, GFP_KERNEL); 1033 dst->dev = net->loopback_dev; 1034 dst_release(dst); 1035 } 1036 } 1037 1038 static int vrf_rtable_create(struct net_device *dev) 1039 { 1040 struct net_vrf *vrf = netdev_priv(dev); 1041 struct rtable *rth; 1042 1043 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1044 return -ENOMEM; 1045 1046 /* create a dst for routing packets out through a VRF device */ 1047 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1); 1048 if (!rth) 1049 return -ENOMEM; 1050 1051 rth->dst.output = vrf_output; 1052 1053 rcu_assign_pointer(vrf->rth, rth); 1054 1055 return 0; 1056 } 1057 1058 /**************************** device handling ********************/ 1059 1060 /* cycle interface to flush neighbor cache and move routes across tables */ 1061 static void cycle_netdev(struct net_device *dev, 1062 struct netlink_ext_ack *extack) 1063 { 1064 unsigned int flags = dev->flags; 1065 int ret; 1066 1067 if (!netif_running(dev)) 1068 return; 1069 1070 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1071 if (ret >= 0) 1072 ret = dev_change_flags(dev, flags, extack); 1073 1074 if (ret < 0) { 1075 netdev_err(dev, 1076 "Failed to cycle device %s; route tables might be wrong!\n", 1077 dev->name); 1078 } 1079 } 1080 1081 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1082 struct netlink_ext_ack *extack) 1083 { 1084 int ret; 1085 1086 /* do not allow loopback device to be enslaved to a VRF. 1087 * The vrf device acts as the loopback for the vrf. 1088 */ 1089 if (port_dev == dev_net(dev)->loopback_dev) { 1090 NL_SET_ERR_MSG(extack, 1091 "Can not enslave loopback device to a VRF"); 1092 return -EOPNOTSUPP; 1093 } 1094 1095 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1096 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1097 if (ret < 0) 1098 goto err; 1099 1100 cycle_netdev(port_dev, extack); 1101 1102 return 0; 1103 1104 err: 1105 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1106 return ret; 1107 } 1108 1109 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1110 struct netlink_ext_ack *extack) 1111 { 1112 if (netif_is_l3_master(port_dev)) { 1113 NL_SET_ERR_MSG(extack, 1114 "Can not enslave an L3 master device to a VRF"); 1115 return -EINVAL; 1116 } 1117 1118 if (netif_is_l3_slave(port_dev)) 1119 return -EINVAL; 1120 1121 return do_vrf_add_slave(dev, port_dev, extack); 1122 } 1123 1124 /* inverse of do_vrf_add_slave */ 1125 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1126 { 1127 netdev_upper_dev_unlink(port_dev, dev); 1128 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1129 1130 cycle_netdev(port_dev, NULL); 1131 1132 return 0; 1133 } 1134 1135 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1136 { 1137 return do_vrf_del_slave(dev, port_dev); 1138 } 1139 1140 static void vrf_dev_uninit(struct net_device *dev) 1141 { 1142 struct net_vrf *vrf = netdev_priv(dev); 1143 1144 vrf_rtable_release(dev, vrf); 1145 vrf_rt6_release(dev, vrf); 1146 } 1147 1148 static int vrf_dev_init(struct net_device *dev) 1149 { 1150 struct net_vrf *vrf = netdev_priv(dev); 1151 1152 /* create the default dst which points back to us */ 1153 if (vrf_rtable_create(dev) != 0) 1154 goto out_nomem; 1155 1156 if (vrf_rt6_create(dev) != 0) 1157 goto out_rth; 1158 1159 dev->flags = IFF_MASTER | IFF_NOARP; 1160 1161 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1162 dev->operstate = IF_OPER_UP; 1163 netdev_lockdep_set_classes(dev); 1164 return 0; 1165 1166 out_rth: 1167 vrf_rtable_release(dev, vrf); 1168 out_nomem: 1169 return -ENOMEM; 1170 } 1171 1172 static const struct net_device_ops vrf_netdev_ops = { 1173 .ndo_init = vrf_dev_init, 1174 .ndo_uninit = vrf_dev_uninit, 1175 .ndo_start_xmit = vrf_xmit, 1176 .ndo_set_mac_address = eth_mac_addr, 1177 .ndo_add_slave = vrf_add_slave, 1178 .ndo_del_slave = vrf_del_slave, 1179 }; 1180 1181 static u32 vrf_fib_table(const struct net_device *dev) 1182 { 1183 struct net_vrf *vrf = netdev_priv(dev); 1184 1185 return vrf->tb_id; 1186 } 1187 1188 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1189 { 1190 kfree_skb(skb); 1191 return 0; 1192 } 1193 1194 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1195 struct sk_buff *skb, 1196 struct net_device *dev) 1197 { 1198 struct net *net = dev_net(dev); 1199 1200 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1201 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1202 1203 return skb; 1204 } 1205 1206 static int vrf_prepare_mac_header(struct sk_buff *skb, 1207 struct net_device *vrf_dev, u16 proto) 1208 { 1209 struct ethhdr *eth; 1210 int err; 1211 1212 /* in general, we do not know if there is enough space in the head of 1213 * the packet for hosting the mac header. 1214 */ 1215 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1216 if (unlikely(err)) 1217 /* no space in the skb head */ 1218 return -ENOBUFS; 1219 1220 __skb_push(skb, ETH_HLEN); 1221 eth = (struct ethhdr *)skb->data; 1222 1223 skb_reset_mac_header(skb); 1224 skb_reset_mac_len(skb); 1225 1226 /* we set the ethernet destination and the source addresses to the 1227 * address of the VRF device. 1228 */ 1229 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1230 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1231 eth->h_proto = htons(proto); 1232 1233 /* the destination address of the Ethernet frame corresponds to the 1234 * address set on the VRF interface; therefore, the packet is intended 1235 * to be processed locally. 1236 */ 1237 skb->protocol = eth->h_proto; 1238 skb->pkt_type = PACKET_HOST; 1239 1240 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1241 1242 skb_pull_inline(skb, ETH_HLEN); 1243 1244 return 0; 1245 } 1246 1247 /* prepare and add the mac header to the packet if it was not set previously. 1248 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1249 * If the mac header was already set, the original mac header is left 1250 * untouched and the function returns immediately. 1251 */ 1252 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1253 struct net_device *vrf_dev, 1254 u16 proto, struct net_device *orig_dev) 1255 { 1256 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev)) 1257 return 0; 1258 1259 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1260 } 1261 1262 #if IS_ENABLED(CONFIG_IPV6) 1263 /* neighbor handling is done with actual device; do not want 1264 * to flip skb->dev for those ndisc packets. This really fails 1265 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1266 * a start. 1267 */ 1268 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1269 { 1270 const struct ipv6hdr *iph = ipv6_hdr(skb); 1271 bool rc = false; 1272 1273 if (iph->nexthdr == NEXTHDR_ICMP) { 1274 const struct icmp6hdr *icmph; 1275 struct icmp6hdr _icmph; 1276 1277 icmph = skb_header_pointer(skb, sizeof(*iph), 1278 sizeof(_icmph), &_icmph); 1279 if (!icmph) 1280 goto out; 1281 1282 switch (icmph->icmp6_type) { 1283 case NDISC_ROUTER_SOLICITATION: 1284 case NDISC_ROUTER_ADVERTISEMENT: 1285 case NDISC_NEIGHBOUR_SOLICITATION: 1286 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1287 case NDISC_REDIRECT: 1288 rc = true; 1289 break; 1290 } 1291 } 1292 1293 out: 1294 return rc; 1295 } 1296 1297 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1298 const struct net_device *dev, 1299 struct flowi6 *fl6, 1300 int ifindex, 1301 const struct sk_buff *skb, 1302 int flags) 1303 { 1304 struct net_vrf *vrf = netdev_priv(dev); 1305 1306 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1307 } 1308 1309 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1310 int ifindex) 1311 { 1312 const struct ipv6hdr *iph = ipv6_hdr(skb); 1313 struct flowi6 fl6 = { 1314 .flowi6_iif = ifindex, 1315 .flowi6_mark = skb->mark, 1316 .flowi6_proto = iph->nexthdr, 1317 .daddr = iph->daddr, 1318 .saddr = iph->saddr, 1319 .flowlabel = ip6_flowinfo(iph), 1320 }; 1321 struct net *net = dev_net(vrf_dev); 1322 struct rt6_info *rt6; 1323 1324 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1325 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1326 if (unlikely(!rt6)) 1327 return; 1328 1329 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1330 return; 1331 1332 skb_dst_set(skb, &rt6->dst); 1333 } 1334 1335 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1336 struct sk_buff *skb) 1337 { 1338 int orig_iif = skb->skb_iif; 1339 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1340 bool is_ndisc = ipv6_ndisc_frame(skb); 1341 1342 /* loopback, multicast & non-ND link-local traffic; do not push through 1343 * packet taps again. Reset pkt_type for upper layers to process skb. 1344 * For non-loopback strict packets, determine the dst using the original 1345 * ifindex. 1346 */ 1347 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1348 skb->dev = vrf_dev; 1349 skb->skb_iif = vrf_dev->ifindex; 1350 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1351 1352 if (skb->pkt_type == PACKET_LOOPBACK) 1353 skb->pkt_type = PACKET_HOST; 1354 else 1355 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1356 1357 goto out; 1358 } 1359 1360 /* if packet is NDISC then keep the ingress interface */ 1361 if (!is_ndisc) { 1362 struct net_device *orig_dev = skb->dev; 1363 1364 vrf_rx_stats(vrf_dev, skb->len); 1365 skb->dev = vrf_dev; 1366 skb->skb_iif = vrf_dev->ifindex; 1367 1368 if (!list_empty(&vrf_dev->ptype_all)) { 1369 int err; 1370 1371 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1372 ETH_P_IPV6, 1373 orig_dev); 1374 if (likely(!err)) { 1375 skb_push(skb, skb->mac_len); 1376 dev_queue_xmit_nit(skb, vrf_dev); 1377 skb_pull(skb, skb->mac_len); 1378 } 1379 } 1380 1381 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1382 } 1383 1384 if (need_strict) 1385 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1386 1387 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1388 out: 1389 return skb; 1390 } 1391 1392 #else 1393 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1394 struct sk_buff *skb) 1395 { 1396 return skb; 1397 } 1398 #endif 1399 1400 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1401 struct sk_buff *skb) 1402 { 1403 struct net_device *orig_dev = skb->dev; 1404 1405 skb->dev = vrf_dev; 1406 skb->skb_iif = vrf_dev->ifindex; 1407 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1408 1409 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1410 goto out; 1411 1412 /* loopback traffic; do not push through packet taps again. 1413 * Reset pkt_type for upper layers to process skb 1414 */ 1415 if (skb->pkt_type == PACKET_LOOPBACK) { 1416 skb->pkt_type = PACKET_HOST; 1417 goto out; 1418 } 1419 1420 vrf_rx_stats(vrf_dev, skb->len); 1421 1422 if (!list_empty(&vrf_dev->ptype_all)) { 1423 int err; 1424 1425 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP, 1426 orig_dev); 1427 if (likely(!err)) { 1428 skb_push(skb, skb->mac_len); 1429 dev_queue_xmit_nit(skb, vrf_dev); 1430 skb_pull(skb, skb->mac_len); 1431 } 1432 } 1433 1434 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1435 out: 1436 return skb; 1437 } 1438 1439 /* called with rcu lock held */ 1440 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1441 struct sk_buff *skb, 1442 u16 proto) 1443 { 1444 switch (proto) { 1445 case AF_INET: 1446 return vrf_ip_rcv(vrf_dev, skb); 1447 case AF_INET6: 1448 return vrf_ip6_rcv(vrf_dev, skb); 1449 } 1450 1451 return skb; 1452 } 1453 1454 #if IS_ENABLED(CONFIG_IPV6) 1455 /* send to link-local or multicast address via interface enslaved to 1456 * VRF device. Force lookup to VRF table without changing flow struct 1457 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1458 * is taken on the dst by this function. 1459 */ 1460 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1461 struct flowi6 *fl6) 1462 { 1463 struct net *net = dev_net(dev); 1464 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1465 struct dst_entry *dst = NULL; 1466 struct rt6_info *rt; 1467 1468 /* VRF device does not have a link-local address and 1469 * sending packets to link-local or mcast addresses over 1470 * a VRF device does not make sense 1471 */ 1472 if (fl6->flowi6_oif == dev->ifindex) { 1473 dst = &net->ipv6.ip6_null_entry->dst; 1474 return dst; 1475 } 1476 1477 if (!ipv6_addr_any(&fl6->saddr)) 1478 flags |= RT6_LOOKUP_F_HAS_SADDR; 1479 1480 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1481 if (rt) 1482 dst = &rt->dst; 1483 1484 return dst; 1485 } 1486 #endif 1487 1488 static const struct l3mdev_ops vrf_l3mdev_ops = { 1489 .l3mdev_fib_table = vrf_fib_table, 1490 .l3mdev_l3_rcv = vrf_l3_rcv, 1491 .l3mdev_l3_out = vrf_l3_out, 1492 #if IS_ENABLED(CONFIG_IPV6) 1493 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1494 #endif 1495 }; 1496 1497 static void vrf_get_drvinfo(struct net_device *dev, 1498 struct ethtool_drvinfo *info) 1499 { 1500 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 1501 strscpy(info->version, DRV_VERSION, sizeof(info->version)); 1502 } 1503 1504 static const struct ethtool_ops vrf_ethtool_ops = { 1505 .get_drvinfo = vrf_get_drvinfo, 1506 }; 1507 1508 static inline size_t vrf_fib_rule_nl_size(void) 1509 { 1510 size_t sz; 1511 1512 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1513 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1514 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1515 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1516 1517 return sz; 1518 } 1519 1520 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1521 { 1522 struct fib_rule_hdr *frh; 1523 struct nlmsghdr *nlh; 1524 struct sk_buff *skb; 1525 int err; 1526 1527 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1528 !ipv6_mod_enabled()) 1529 return 0; 1530 1531 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1532 if (!skb) 1533 return -ENOMEM; 1534 1535 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1536 if (!nlh) 1537 goto nla_put_failure; 1538 1539 /* rule only needs to appear once */ 1540 nlh->nlmsg_flags |= NLM_F_EXCL; 1541 1542 frh = nlmsg_data(nlh); 1543 memset(frh, 0, sizeof(*frh)); 1544 frh->family = family; 1545 frh->action = FR_ACT_TO_TBL; 1546 1547 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1548 goto nla_put_failure; 1549 1550 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1551 goto nla_put_failure; 1552 1553 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1554 goto nla_put_failure; 1555 1556 nlmsg_end(skb, nlh); 1557 1558 /* fib_nl_{new,del}rule handling looks for net from skb->sk */ 1559 skb->sk = dev_net(dev)->rtnl; 1560 if (add_it) { 1561 err = fib_nl_newrule(skb, nlh, NULL); 1562 if (err == -EEXIST) 1563 err = 0; 1564 } else { 1565 err = fib_nl_delrule(skb, nlh, NULL); 1566 if (err == -ENOENT) 1567 err = 0; 1568 } 1569 nlmsg_free(skb); 1570 1571 return err; 1572 1573 nla_put_failure: 1574 nlmsg_free(skb); 1575 1576 return -EMSGSIZE; 1577 } 1578 1579 static int vrf_add_fib_rules(const struct net_device *dev) 1580 { 1581 int err; 1582 1583 err = vrf_fib_rule(dev, AF_INET, true); 1584 if (err < 0) 1585 goto out_err; 1586 1587 err = vrf_fib_rule(dev, AF_INET6, true); 1588 if (err < 0) 1589 goto ipv6_err; 1590 1591 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1592 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1593 if (err < 0) 1594 goto ipmr_err; 1595 #endif 1596 1597 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1598 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1599 if (err < 0) 1600 goto ip6mr_err; 1601 #endif 1602 1603 return 0; 1604 1605 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1606 ip6mr_err: 1607 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1608 #endif 1609 1610 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1611 ipmr_err: 1612 vrf_fib_rule(dev, AF_INET6, false); 1613 #endif 1614 1615 ipv6_err: 1616 vrf_fib_rule(dev, AF_INET, false); 1617 1618 out_err: 1619 netdev_err(dev, "Failed to add FIB rules.\n"); 1620 return err; 1621 } 1622 1623 static void vrf_setup(struct net_device *dev) 1624 { 1625 ether_setup(dev); 1626 1627 /* Initialize the device structure. */ 1628 dev->netdev_ops = &vrf_netdev_ops; 1629 dev->l3mdev_ops = &vrf_l3mdev_ops; 1630 dev->ethtool_ops = &vrf_ethtool_ops; 1631 dev->needs_free_netdev = true; 1632 1633 /* Fill in device structure with ethernet-generic values. */ 1634 eth_hw_addr_random(dev); 1635 1636 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1637 dev->features |= NETIF_F_LLTX; 1638 1639 /* don't allow vrf devices to change network namespaces. */ 1640 dev->features |= NETIF_F_NETNS_LOCAL; 1641 1642 /* does not make sense for a VLAN to be added to a vrf device */ 1643 dev->features |= NETIF_F_VLAN_CHALLENGED; 1644 1645 /* enable offload features */ 1646 dev->features |= NETIF_F_GSO_SOFTWARE; 1647 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1648 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1649 1650 dev->hw_features = dev->features; 1651 dev->hw_enc_features = dev->features; 1652 1653 /* default to no qdisc; user can add if desired */ 1654 dev->priv_flags |= IFF_NO_QUEUE; 1655 dev->priv_flags |= IFF_NO_RX_HANDLER; 1656 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1657 1658 /* VRF devices do not care about MTU, but if the MTU is set 1659 * too low then the ipv4 and ipv6 protocols are disabled 1660 * which breaks networking. 1661 */ 1662 dev->min_mtu = IPV6_MIN_MTU; 1663 dev->max_mtu = IP6_MAX_MTU; 1664 dev->mtu = dev->max_mtu; 1665 1666 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS; 1667 } 1668 1669 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1670 struct netlink_ext_ack *extack) 1671 { 1672 if (tb[IFLA_ADDRESS]) { 1673 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1674 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1675 return -EINVAL; 1676 } 1677 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1678 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1679 return -EADDRNOTAVAIL; 1680 } 1681 } 1682 return 0; 1683 } 1684 1685 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1686 { 1687 struct net_device *port_dev; 1688 struct list_head *iter; 1689 1690 netdev_for_each_lower_dev(dev, port_dev, iter) 1691 vrf_del_slave(dev, port_dev); 1692 1693 vrf_map_unregister_dev(dev); 1694 1695 unregister_netdevice_queue(dev, head); 1696 } 1697 1698 static int vrf_newlink(struct net *src_net, struct net_device *dev, 1699 struct nlattr *tb[], struct nlattr *data[], 1700 struct netlink_ext_ack *extack) 1701 { 1702 struct net_vrf *vrf = netdev_priv(dev); 1703 struct netns_vrf *nn_vrf; 1704 bool *add_fib_rules; 1705 struct net *net; 1706 int err; 1707 1708 if (!data || !data[IFLA_VRF_TABLE]) { 1709 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1710 return -EINVAL; 1711 } 1712 1713 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1714 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1715 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1716 "Invalid VRF table id"); 1717 return -EINVAL; 1718 } 1719 1720 dev->priv_flags |= IFF_L3MDEV_MASTER; 1721 1722 err = register_netdevice(dev); 1723 if (err) 1724 goto out; 1725 1726 /* mapping between table_id and vrf; 1727 * note: such binding could not be done in the dev init function 1728 * because dev->ifindex id is not available yet. 1729 */ 1730 vrf->ifindex = dev->ifindex; 1731 1732 err = vrf_map_register_dev(dev, extack); 1733 if (err) { 1734 unregister_netdevice(dev); 1735 goto out; 1736 } 1737 1738 net = dev_net(dev); 1739 nn_vrf = net_generic(net, vrf_net_id); 1740 1741 add_fib_rules = &nn_vrf->add_fib_rules; 1742 if (*add_fib_rules) { 1743 err = vrf_add_fib_rules(dev); 1744 if (err) { 1745 vrf_map_unregister_dev(dev); 1746 unregister_netdevice(dev); 1747 goto out; 1748 } 1749 *add_fib_rules = false; 1750 } 1751 1752 out: 1753 return err; 1754 } 1755 1756 static size_t vrf_nl_getsize(const struct net_device *dev) 1757 { 1758 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1759 } 1760 1761 static int vrf_fillinfo(struct sk_buff *skb, 1762 const struct net_device *dev) 1763 { 1764 struct net_vrf *vrf = netdev_priv(dev); 1765 1766 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1767 } 1768 1769 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1770 const struct net_device *slave_dev) 1771 { 1772 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1773 } 1774 1775 static int vrf_fill_slave_info(struct sk_buff *skb, 1776 const struct net_device *vrf_dev, 1777 const struct net_device *slave_dev) 1778 { 1779 struct net_vrf *vrf = netdev_priv(vrf_dev); 1780 1781 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1782 return -EMSGSIZE; 1783 1784 return 0; 1785 } 1786 1787 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1788 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1789 }; 1790 1791 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1792 .kind = DRV_NAME, 1793 .priv_size = sizeof(struct net_vrf), 1794 1795 .get_size = vrf_nl_getsize, 1796 .policy = vrf_nl_policy, 1797 .validate = vrf_validate, 1798 .fill_info = vrf_fillinfo, 1799 1800 .get_slave_size = vrf_get_slave_size, 1801 .fill_slave_info = vrf_fill_slave_info, 1802 1803 .newlink = vrf_newlink, 1804 .dellink = vrf_dellink, 1805 .setup = vrf_setup, 1806 .maxtype = IFLA_VRF_MAX, 1807 }; 1808 1809 static int vrf_device_event(struct notifier_block *unused, 1810 unsigned long event, void *ptr) 1811 { 1812 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1813 1814 /* only care about unregister events to drop slave references */ 1815 if (event == NETDEV_UNREGISTER) { 1816 struct net_device *vrf_dev; 1817 1818 if (!netif_is_l3_slave(dev)) 1819 goto out; 1820 1821 vrf_dev = netdev_master_upper_dev_get(dev); 1822 vrf_del_slave(vrf_dev, dev); 1823 } 1824 out: 1825 return NOTIFY_DONE; 1826 } 1827 1828 static struct notifier_block vrf_notifier_block __read_mostly = { 1829 .notifier_call = vrf_device_event, 1830 }; 1831 1832 static int vrf_map_init(struct vrf_map *vmap) 1833 { 1834 spin_lock_init(&vmap->vmap_lock); 1835 hash_init(vmap->ht); 1836 1837 vmap->strict_mode = false; 1838 1839 return 0; 1840 } 1841 1842 #ifdef CONFIG_SYSCTL 1843 static bool vrf_strict_mode(struct vrf_map *vmap) 1844 { 1845 bool strict_mode; 1846 1847 vrf_map_lock(vmap); 1848 strict_mode = vmap->strict_mode; 1849 vrf_map_unlock(vmap); 1850 1851 return strict_mode; 1852 } 1853 1854 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1855 { 1856 bool *cur_mode; 1857 int res = 0; 1858 1859 vrf_map_lock(vmap); 1860 1861 cur_mode = &vmap->strict_mode; 1862 if (*cur_mode == new_mode) 1863 goto unlock; 1864 1865 if (*cur_mode) { 1866 /* disable strict mode */ 1867 *cur_mode = false; 1868 } else { 1869 if (vmap->shared_tables) { 1870 /* we cannot allow strict_mode because there are some 1871 * vrfs that share one or more tables. 1872 */ 1873 res = -EBUSY; 1874 goto unlock; 1875 } 1876 1877 /* no tables are shared among vrfs, so we can go back 1878 * to 1:1 association between a vrf with its table. 1879 */ 1880 *cur_mode = true; 1881 } 1882 1883 unlock: 1884 vrf_map_unlock(vmap); 1885 1886 return res; 1887 } 1888 1889 static int vrf_shared_table_handler(const struct ctl_table *table, int write, 1890 void *buffer, size_t *lenp, loff_t *ppos) 1891 { 1892 struct net *net = (struct net *)table->extra1; 1893 struct vrf_map *vmap = netns_vrf_map(net); 1894 int proc_strict_mode = 0; 1895 struct ctl_table tmp = { 1896 .procname = table->procname, 1897 .data = &proc_strict_mode, 1898 .maxlen = sizeof(int), 1899 .mode = table->mode, 1900 .extra1 = SYSCTL_ZERO, 1901 .extra2 = SYSCTL_ONE, 1902 }; 1903 int ret; 1904 1905 if (!write) 1906 proc_strict_mode = vrf_strict_mode(vmap); 1907 1908 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1909 1910 if (write && ret == 0) 1911 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1912 1913 return ret; 1914 } 1915 1916 static const struct ctl_table vrf_table[] = { 1917 { 1918 .procname = "strict_mode", 1919 .data = NULL, 1920 .maxlen = sizeof(int), 1921 .mode = 0644, 1922 .proc_handler = vrf_shared_table_handler, 1923 /* set by the vrf_netns_init */ 1924 .extra1 = NULL, 1925 }, 1926 }; 1927 1928 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1929 { 1930 struct ctl_table *table; 1931 1932 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1933 if (!table) 1934 return -ENOMEM; 1935 1936 /* init the extra1 parameter with the reference to current netns */ 1937 table[0].extra1 = net; 1938 1939 nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table, 1940 ARRAY_SIZE(vrf_table)); 1941 if (!nn_vrf->ctl_hdr) { 1942 kfree(table); 1943 return -ENOMEM; 1944 } 1945 1946 return 0; 1947 } 1948 1949 static void vrf_netns_exit_sysctl(struct net *net) 1950 { 1951 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1952 const struct ctl_table *table; 1953 1954 table = nn_vrf->ctl_hdr->ctl_table_arg; 1955 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1956 kfree(table); 1957 } 1958 #else 1959 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1960 { 1961 return 0; 1962 } 1963 1964 static void vrf_netns_exit_sysctl(struct net *net) 1965 { 1966 } 1967 #endif 1968 1969 /* Initialize per network namespace state */ 1970 static int __net_init vrf_netns_init(struct net *net) 1971 { 1972 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1973 1974 nn_vrf->add_fib_rules = true; 1975 vrf_map_init(&nn_vrf->vmap); 1976 1977 return vrf_netns_init_sysctl(net, nn_vrf); 1978 } 1979 1980 static void __net_exit vrf_netns_exit(struct net *net) 1981 { 1982 vrf_netns_exit_sysctl(net); 1983 } 1984 1985 static struct pernet_operations vrf_net_ops __net_initdata = { 1986 .init = vrf_netns_init, 1987 .exit = vrf_netns_exit, 1988 .id = &vrf_net_id, 1989 .size = sizeof(struct netns_vrf), 1990 }; 1991 1992 static int __init vrf_init_module(void) 1993 { 1994 int rc; 1995 1996 register_netdevice_notifier(&vrf_notifier_block); 1997 1998 rc = register_pernet_subsys(&vrf_net_ops); 1999 if (rc < 0) 2000 goto error; 2001 2002 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 2003 vrf_ifindex_lookup_by_table_id); 2004 if (rc < 0) 2005 goto unreg_pernet; 2006 2007 rc = rtnl_link_register(&vrf_link_ops); 2008 if (rc < 0) 2009 goto table_lookup_unreg; 2010 2011 return 0; 2012 2013 table_lookup_unreg: 2014 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 2015 vrf_ifindex_lookup_by_table_id); 2016 2017 unreg_pernet: 2018 unregister_pernet_subsys(&vrf_net_ops); 2019 2020 error: 2021 unregister_netdevice_notifier(&vrf_notifier_block); 2022 return rc; 2023 } 2024 2025 module_init(vrf_init_module); 2026 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2027 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2028 MODULE_LICENSE("GPL"); 2029 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2030 MODULE_VERSION(DRV_VERSION); 2031