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/netdev_lock.h> 38 #include <net/sch_generic.h> 39 #include <net/netns/generic.h> 40 #include <net/netfilter/nf_conntrack.h> 41 #include <net/inet_dscp.h> 42 43 #define DRV_NAME "vrf" 44 #define DRV_VERSION "1.1" 45 46 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */ 47 48 #define HT_MAP_BITS 4 49 #define HASH_INITVAL ((u32)0xcafef00d) 50 51 struct vrf_map { 52 DECLARE_HASHTABLE(ht, HT_MAP_BITS); 53 spinlock_t vmap_lock; 54 55 /* shared_tables: 56 * count how many distinct tables do not comply with the strict mode 57 * requirement. 58 * shared_tables value must be 0 in order to enable the strict mode. 59 * 60 * example of the evolution of shared_tables: 61 * | time 62 * add vrf0 --> table 100 shared_tables = 0 | t0 63 * add vrf1 --> table 101 shared_tables = 0 | t1 64 * add vrf2 --> table 100 shared_tables = 1 | t2 65 * add vrf3 --> table 100 shared_tables = 1 | t3 66 * add vrf4 --> table 101 shared_tables = 2 v t4 67 * 68 * shared_tables is a "step function" (or "staircase function") 69 * and it is increased by one when the second vrf is associated to a 70 * table. 71 * 72 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1. 73 * 74 * at t3, another dev (vrf3) is bound to the same table 100 but the 75 * value of shared_tables is still 1. 76 * This means that no matter how many new vrfs will register on the 77 * table 100, the shared_tables will not increase (considering only 78 * table 100). 79 * 80 * at t4, vrf4 is bound to table 101, and shared_tables = 2. 81 * 82 * Looking at the value of shared_tables we can immediately know if 83 * the strict_mode can or cannot be enforced. Indeed, strict_mode 84 * can be enforced iff shared_tables = 0. 85 * 86 * Conversely, shared_tables is decreased when a vrf is de-associated 87 * from a table with exactly two associated vrfs. 88 */ 89 u32 shared_tables; 90 91 bool strict_mode; 92 }; 93 94 struct vrf_map_elem { 95 struct hlist_node hnode; 96 struct list_head vrf_list; /* VRFs registered to this table */ 97 98 u32 table_id; 99 int users; 100 int ifindex; 101 }; 102 103 static unsigned int vrf_net_id; 104 105 /* per netns vrf data */ 106 struct netns_vrf { 107 /* protected by rtnl lock */ 108 bool add_fib_rules; 109 110 struct vrf_map vmap; 111 struct ctl_table_header *ctl_hdr; 112 }; 113 114 struct net_vrf { 115 struct rtable __rcu *rth; 116 struct rt6_info __rcu *rt6; 117 #if IS_ENABLED(CONFIG_IPV6) 118 struct fib6_table *fib6_table; 119 #endif 120 u32 tb_id; 121 122 struct list_head me_list; /* entry in vrf_map_elem */ 123 int ifindex; 124 }; 125 126 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) 127 { 128 vrf_dev->stats.tx_errors++; 129 kfree_skb(skb); 130 } 131 132 static struct vrf_map *netns_vrf_map(struct net *net) 133 { 134 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 135 136 return &nn_vrf->vmap; 137 } 138 139 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev) 140 { 141 return netns_vrf_map(dev_net(dev)); 142 } 143 144 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me) 145 { 146 struct list_head *me_head = &me->vrf_list; 147 struct net_vrf *vrf; 148 149 if (list_empty(me_head)) 150 return -ENODEV; 151 152 vrf = list_first_entry(me_head, struct net_vrf, me_list); 153 154 return vrf->ifindex; 155 } 156 157 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags) 158 { 159 struct vrf_map_elem *me; 160 161 me = kmalloc(sizeof(*me), flags); 162 if (!me) 163 return NULL; 164 165 return me; 166 } 167 168 static void vrf_map_elem_free(struct vrf_map_elem *me) 169 { 170 kfree(me); 171 } 172 173 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id, 174 int ifindex, int users) 175 { 176 me->table_id = table_id; 177 me->ifindex = ifindex; 178 me->users = users; 179 INIT_LIST_HEAD(&me->vrf_list); 180 } 181 182 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap, 183 u32 table_id) 184 { 185 struct vrf_map_elem *me; 186 u32 key; 187 188 key = jhash_1word(table_id, HASH_INITVAL); 189 hash_for_each_possible(vmap->ht, me, hnode, key) { 190 if (me->table_id == table_id) 191 return me; 192 } 193 194 return NULL; 195 } 196 197 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me) 198 { 199 u32 table_id = me->table_id; 200 u32 key; 201 202 key = jhash_1word(table_id, HASH_INITVAL); 203 hash_add(vmap->ht, &me->hnode, key); 204 } 205 206 static void vrf_map_del_elem(struct vrf_map_elem *me) 207 { 208 hash_del(&me->hnode); 209 } 210 211 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock) 212 { 213 spin_lock(&vmap->vmap_lock); 214 } 215 216 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock) 217 { 218 spin_unlock(&vmap->vmap_lock); 219 } 220 221 /* called with rtnl lock held */ 222 static int 223 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack) 224 { 225 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 226 struct net_vrf *vrf = netdev_priv(dev); 227 struct vrf_map_elem *new_me, *me; 228 u32 table_id = vrf->tb_id; 229 bool free_new_me = false; 230 int users; 231 int res; 232 233 /* we pre-allocate elements used in the spin-locked section (so that we 234 * keep the spinlock as short as possible). 235 */ 236 new_me = vrf_map_elem_alloc(GFP_KERNEL); 237 if (!new_me) 238 return -ENOMEM; 239 240 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0); 241 242 vrf_map_lock(vmap); 243 244 me = vrf_map_lookup_elem(vmap, table_id); 245 if (!me) { 246 me = new_me; 247 vrf_map_add_elem(vmap, me); 248 goto link_vrf; 249 } 250 251 /* we already have an entry in the vrf_map, so it means there is (at 252 * least) a vrf registered on the specific table. 253 */ 254 free_new_me = true; 255 if (vmap->strict_mode) { 256 /* vrfs cannot share the same table */ 257 NL_SET_ERR_MSG(extack, "Table is used by another VRF"); 258 res = -EBUSY; 259 goto unlock; 260 } 261 262 link_vrf: 263 users = ++me->users; 264 if (users == 2) 265 ++vmap->shared_tables; 266 267 list_add(&vrf->me_list, &me->vrf_list); 268 269 res = 0; 270 271 unlock: 272 vrf_map_unlock(vmap); 273 274 /* clean-up, if needed */ 275 if (free_new_me) 276 vrf_map_elem_free(new_me); 277 278 return res; 279 } 280 281 /* called with rtnl lock held */ 282 static void vrf_map_unregister_dev(struct net_device *dev) 283 { 284 struct vrf_map *vmap = netns_vrf_map_by_dev(dev); 285 struct net_vrf *vrf = netdev_priv(dev); 286 u32 table_id = vrf->tb_id; 287 struct vrf_map_elem *me; 288 int users; 289 290 vrf_map_lock(vmap); 291 292 me = vrf_map_lookup_elem(vmap, table_id); 293 if (!me) 294 goto unlock; 295 296 list_del(&vrf->me_list); 297 298 users = --me->users; 299 if (users == 1) { 300 --vmap->shared_tables; 301 } else if (users == 0) { 302 vrf_map_del_elem(me); 303 304 /* no one will refer to this element anymore */ 305 vrf_map_elem_free(me); 306 } 307 308 unlock: 309 vrf_map_unlock(vmap); 310 } 311 312 /* return the vrf device index associated with the table_id */ 313 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id) 314 { 315 struct vrf_map *vmap = netns_vrf_map(net); 316 struct vrf_map_elem *me; 317 int ifindex; 318 319 vrf_map_lock(vmap); 320 321 if (!vmap->strict_mode) { 322 ifindex = -EPERM; 323 goto unlock; 324 } 325 326 me = vrf_map_lookup_elem(vmap, table_id); 327 if (!me) { 328 ifindex = -ENODEV; 329 goto unlock; 330 } 331 332 ifindex = vrf_map_elem_get_vrf_ifindex(me); 333 334 unlock: 335 vrf_map_unlock(vmap); 336 337 return ifindex; 338 } 339 340 /* by default VRF devices do not have a qdisc and are expected 341 * to be created with only a single queue. 342 */ 343 static bool qdisc_tx_is_default(const struct net_device *dev) 344 { 345 struct netdev_queue *txq; 346 347 if (dev->num_tx_queues > 1) 348 return false; 349 350 txq = netdev_get_tx_queue(dev, 0); 351 352 return qdisc_txq_has_no_queue(txq); 353 } 354 355 /* Local traffic destined to local address. Reinsert the packet to rx 356 * path, similar to loopback handling. 357 */ 358 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev, 359 struct dst_entry *dst) 360 { 361 unsigned int len = skb->len; 362 363 skb_orphan(skb); 364 365 skb_dst_set(skb, dst); 366 367 /* set pkt_type to avoid skb hitting packet taps twice - 368 * once on Tx and again in Rx processing 369 */ 370 skb->pkt_type = PACKET_LOOPBACK; 371 372 skb->protocol = eth_type_trans(skb, dev); 373 374 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) 375 dev_dstats_rx_add(dev, len); 376 else 377 dev_dstats_rx_dropped(dev); 378 379 return NETDEV_TX_OK; 380 } 381 382 static void vrf_nf_set_untracked(struct sk_buff *skb) 383 { 384 if (skb_get_nfct(skb) == 0) 385 nf_ct_set(skb, NULL, IP_CT_UNTRACKED); 386 } 387 388 static void vrf_nf_reset_ct(struct sk_buff *skb) 389 { 390 if (skb_get_nfct(skb) == IP_CT_UNTRACKED) 391 nf_reset_ct(skb); 392 } 393 394 #if IS_ENABLED(CONFIG_IPV6) 395 static int vrf_ip6_local_out(struct net *net, struct sock *sk, 396 struct sk_buff *skb) 397 { 398 int err; 399 400 vrf_nf_reset_ct(skb); 401 402 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, 403 sk, skb, NULL, skb_dst(skb)->dev, dst_output); 404 405 if (likely(err == 1)) 406 err = dst_output(net, sk, skb); 407 408 return err; 409 } 410 411 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 412 struct net_device *dev) 413 { 414 const struct ipv6hdr *iph; 415 struct net *net = dev_net(skb->dev); 416 struct flowi6 fl6; 417 int ret = NET_XMIT_DROP; 418 struct dst_entry *dst; 419 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; 420 421 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 422 goto err; 423 424 iph = ipv6_hdr(skb); 425 426 memset(&fl6, 0, sizeof(fl6)); 427 /* needed to match OIF rule */ 428 fl6.flowi6_l3mdev = dev->ifindex; 429 fl6.flowi6_iif = LOOPBACK_IFINDEX; 430 fl6.daddr = iph->daddr; 431 fl6.saddr = iph->saddr; 432 fl6.flowlabel = ip6_flowinfo(iph); 433 fl6.flowi6_mark = skb->mark; 434 fl6.flowi6_proto = iph->nexthdr; 435 436 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 437 if (IS_ERR(dst) || dst == dst_null) 438 goto err; 439 440 skb_dst_drop(skb); 441 442 /* if dst.dev is the VRF device again this is locally originated traffic 443 * destined to a local address. Short circuit to Rx path. 444 */ 445 if (dst->dev == dev) 446 return vrf_local_xmit(skb, dev, dst); 447 448 skb_dst_set(skb, dst); 449 450 /* strip the ethernet header added for pass through VRF device */ 451 __skb_pull(skb, skb_network_offset(skb)); 452 453 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb))); 454 ret = vrf_ip6_local_out(net, skb->sk, skb); 455 if (unlikely(net_xmit_eval(ret))) 456 dev->stats.tx_errors++; 457 else 458 ret = NET_XMIT_SUCCESS; 459 460 return ret; 461 err: 462 vrf_tx_error(dev, skb); 463 return NET_XMIT_DROP; 464 } 465 #else 466 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, 467 struct net_device *dev) 468 { 469 vrf_tx_error(dev, skb); 470 return NET_XMIT_DROP; 471 } 472 #endif 473 474 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */ 475 static int vrf_ip_local_out(struct net *net, struct sock *sk, 476 struct sk_buff *skb) 477 { 478 int err; 479 480 vrf_nf_reset_ct(skb); 481 482 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 483 skb, NULL, skb_dst(skb)->dev, dst_output); 484 if (likely(err == 1)) 485 err = dst_output(net, sk, skb); 486 487 return err; 488 } 489 490 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, 491 struct net_device *vrf_dev) 492 { 493 struct iphdr *ip4h; 494 int ret = NET_XMIT_DROP; 495 struct flowi4 fl4; 496 struct net *net = dev_net(vrf_dev); 497 struct rtable *rt; 498 499 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 500 goto err; 501 502 ip4h = ip_hdr(skb); 503 504 memset(&fl4, 0, sizeof(fl4)); 505 /* needed to match OIF rule */ 506 fl4.flowi4_l3mdev = vrf_dev->ifindex; 507 fl4.flowi4_iif = LOOPBACK_IFINDEX; 508 fl4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip4h)); 509 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC; 510 fl4.flowi4_proto = ip4h->protocol; 511 fl4.daddr = ip4h->daddr; 512 fl4.saddr = ip4h->saddr; 513 514 rt = ip_route_output_flow(net, &fl4, NULL); 515 if (IS_ERR(rt)) 516 goto err; 517 518 skb_dst_drop(skb); 519 520 /* if dst.dev is the VRF device again this is locally originated traffic 521 * destined to a local address. Short circuit to Rx path. 522 */ 523 if (rt->dst.dev == vrf_dev) 524 return vrf_local_xmit(skb, vrf_dev, &rt->dst); 525 526 skb_dst_set(skb, &rt->dst); 527 528 /* strip the ethernet header added for pass through VRF device */ 529 __skb_pull(skb, skb_network_offset(skb)); 530 531 if (!ip4h->saddr) { 532 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, 533 RT_SCOPE_LINK); 534 } 535 536 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 537 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 538 if (unlikely(net_xmit_eval(ret))) 539 vrf_dev->stats.tx_errors++; 540 else 541 ret = NET_XMIT_SUCCESS; 542 543 out: 544 return ret; 545 err: 546 vrf_tx_error(vrf_dev, skb); 547 goto out; 548 } 549 550 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) 551 { 552 switch (skb->protocol) { 553 case htons(ETH_P_IP): 554 return vrf_process_v4_outbound(skb, dev); 555 case htons(ETH_P_IPV6): 556 return vrf_process_v6_outbound(skb, dev); 557 default: 558 vrf_tx_error(dev, skb); 559 return NET_XMIT_DROP; 560 } 561 } 562 563 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) 564 { 565 unsigned int len = skb->len; 566 netdev_tx_t ret; 567 568 ret = is_ip_tx_frame(skb, dev); 569 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) 570 dev_dstats_tx_add(dev, len); 571 else 572 dev_dstats_tx_dropped(dev); 573 574 return ret; 575 } 576 577 static void vrf_finish_direct(struct sk_buff *skb) 578 { 579 struct net_device *vrf_dev = skb->dev; 580 581 if (!list_empty(&vrf_dev->ptype_all) && 582 likely(skb_headroom(skb) >= ETH_HLEN)) { 583 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 584 585 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 586 eth_zero_addr(eth->h_dest); 587 eth->h_proto = skb->protocol; 588 589 rcu_read_lock_bh(); 590 dev_queue_xmit_nit(skb, vrf_dev); 591 rcu_read_unlock_bh(); 592 593 skb_pull(skb, ETH_HLEN); 594 } 595 596 vrf_nf_reset_ct(skb); 597 } 598 599 #if IS_ENABLED(CONFIG_IPV6) 600 /* modelled after ip6_finish_output2 */ 601 static int vrf_finish_output6(struct net *net, struct sock *sk, 602 struct sk_buff *skb) 603 { 604 struct dst_entry *dst = skb_dst(skb); 605 struct net_device *dev = dst->dev; 606 const struct in6_addr *nexthop; 607 struct neighbour *neigh; 608 int ret; 609 610 vrf_nf_reset_ct(skb); 611 612 skb->protocol = htons(ETH_P_IPV6); 613 skb->dev = dev; 614 615 rcu_read_lock(); 616 nexthop = rt6_nexthop(dst_rt6_info(dst), &ipv6_hdr(skb)->daddr); 617 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 618 if (unlikely(!neigh)) 619 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 620 if (!IS_ERR(neigh)) { 621 sock_confirm_neigh(skb, neigh); 622 ret = neigh_output(neigh, skb, false); 623 rcu_read_unlock(); 624 return ret; 625 } 626 rcu_read_unlock(); 627 628 IP6_INC_STATS(dev_net(dst->dev), 629 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 630 kfree_skb(skb); 631 return -EINVAL; 632 } 633 634 /* modelled after ip6_output */ 635 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) 636 { 637 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 638 net, sk, skb, NULL, skb_dst(skb)->dev, 639 vrf_finish_output6, 640 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 641 } 642 643 /* set dst on skb to send packet to us via dev_xmit path. Allows 644 * packet to go through device based features such as qdisc, netfilter 645 * hooks and packet sockets with skb->dev set to vrf device. 646 */ 647 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev, 648 struct sk_buff *skb) 649 { 650 struct net_vrf *vrf = netdev_priv(vrf_dev); 651 struct dst_entry *dst = NULL; 652 struct rt6_info *rt6; 653 654 rcu_read_lock(); 655 656 rt6 = rcu_dereference(vrf->rt6); 657 if (likely(rt6)) { 658 dst = &rt6->dst; 659 dst_hold(dst); 660 } 661 662 rcu_read_unlock(); 663 664 if (unlikely(!dst)) { 665 vrf_tx_error(vrf_dev, skb); 666 return NULL; 667 } 668 669 skb_dst_drop(skb); 670 skb_dst_set(skb, dst); 671 672 return skb; 673 } 674 675 static int vrf_output6_direct_finish(struct net *net, struct sock *sk, 676 struct sk_buff *skb) 677 { 678 vrf_finish_direct(skb); 679 680 return vrf_ip6_local_out(net, sk, skb); 681 } 682 683 static int vrf_output6_direct(struct net *net, struct sock *sk, 684 struct sk_buff *skb) 685 { 686 int err = 1; 687 688 skb->protocol = htons(ETH_P_IPV6); 689 690 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 691 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb, 692 NULL, skb->dev, vrf_output6_direct_finish); 693 694 if (likely(err == 1)) 695 vrf_finish_direct(skb); 696 697 return err; 698 } 699 700 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk, 701 struct sk_buff *skb) 702 { 703 int err; 704 705 err = vrf_output6_direct(net, sk, skb); 706 if (likely(err == 1)) 707 err = vrf_ip6_local_out(net, sk, skb); 708 709 return err; 710 } 711 712 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev, 713 struct sock *sk, 714 struct sk_buff *skb) 715 { 716 struct net *net = dev_net(vrf_dev); 717 int err; 718 719 skb->dev = vrf_dev; 720 721 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, 722 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish); 723 724 if (likely(err == 1)) 725 err = vrf_output6_direct(net, sk, skb); 726 727 if (likely(err == 1)) 728 return skb; 729 730 return NULL; 731 } 732 733 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 734 struct sock *sk, 735 struct sk_buff *skb) 736 { 737 /* don't divert link scope packets */ 738 if (rt6_need_strict(&ipv6_hdr(skb)->daddr)) 739 return skb; 740 741 vrf_nf_set_untracked(skb); 742 743 if (qdisc_tx_is_default(vrf_dev) || 744 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) 745 return vrf_ip6_out_direct(vrf_dev, sk, skb); 746 747 return vrf_ip6_out_redirect(vrf_dev, skb); 748 } 749 750 /* holding rtnl */ 751 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 752 { 753 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); 754 struct net *net = dev_net(dev); 755 struct dst_entry *dst; 756 757 RCU_INIT_POINTER(vrf->rt6, NULL); 758 synchronize_rcu(); 759 760 /* move dev in dst's to loopback so this VRF device can be deleted 761 * - based on dst_ifdown 762 */ 763 if (rt6) { 764 dst = &rt6->dst; 765 netdev_ref_replace(dst->dev, net->loopback_dev, 766 &dst->dev_tracker, GFP_KERNEL); 767 dst->dev = net->loopback_dev; 768 dst_release(dst); 769 } 770 } 771 772 static int vrf_rt6_create(struct net_device *dev) 773 { 774 int flags = DST_NOPOLICY | DST_NOXFRM; 775 struct net_vrf *vrf = netdev_priv(dev); 776 struct net *net = dev_net(dev); 777 struct rt6_info *rt6; 778 int rc = -ENOMEM; 779 780 /* IPv6 can be CONFIG enabled and then disabled runtime */ 781 if (!ipv6_mod_enabled()) 782 return 0; 783 784 vrf->fib6_table = fib6_new_table(net, vrf->tb_id); 785 if (!vrf->fib6_table) 786 goto out; 787 788 /* create a dst for routing packets out a VRF device */ 789 rt6 = ip6_dst_alloc(net, dev, flags); 790 if (!rt6) 791 goto out; 792 793 rt6->dst.output = vrf_output6; 794 795 rcu_assign_pointer(vrf->rt6, rt6); 796 797 rc = 0; 798 out: 799 return rc; 800 } 801 #else 802 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev, 803 struct sock *sk, 804 struct sk_buff *skb) 805 { 806 return skb; 807 } 808 809 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf) 810 { 811 } 812 813 static int vrf_rt6_create(struct net_device *dev) 814 { 815 return 0; 816 } 817 #endif 818 819 /* modelled after ip_finish_output2 */ 820 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 821 { 822 struct dst_entry *dst = skb_dst(skb); 823 struct rtable *rt = dst_rtable(dst); 824 struct net_device *dev = dst->dev; 825 unsigned int hh_len = LL_RESERVED_SPACE(dev); 826 struct neighbour *neigh; 827 bool is_v6gw = false; 828 829 vrf_nf_reset_ct(skb); 830 831 /* Be paranoid, rather than too clever. */ 832 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 833 skb = skb_expand_head(skb, hh_len); 834 if (!skb) { 835 dev->stats.tx_errors++; 836 return -ENOMEM; 837 } 838 } 839 840 rcu_read_lock(); 841 842 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 843 if (!IS_ERR(neigh)) { 844 int ret; 845 846 sock_confirm_neigh(skb, neigh); 847 /* if crossing protocols, can not use the cached header */ 848 ret = neigh_output(neigh, skb, is_v6gw); 849 rcu_read_unlock(); 850 return ret; 851 } 852 853 rcu_read_unlock(); 854 vrf_tx_error(skb->dev, skb); 855 return -EINVAL; 856 } 857 858 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) 859 { 860 struct net_device *dev = skb_dst(skb)->dev; 861 862 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 863 864 skb->dev = dev; 865 skb->protocol = htons(ETH_P_IP); 866 867 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 868 net, sk, skb, NULL, dev, 869 vrf_finish_output, 870 !(IPCB(skb)->flags & IPSKB_REROUTED)); 871 } 872 873 /* set dst on skb to send packet to us via dev_xmit path. Allows 874 * packet to go through device based features such as qdisc, netfilter 875 * hooks and packet sockets with skb->dev set to vrf device. 876 */ 877 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev, 878 struct sk_buff *skb) 879 { 880 struct net_vrf *vrf = netdev_priv(vrf_dev); 881 struct dst_entry *dst = NULL; 882 struct rtable *rth; 883 884 rcu_read_lock(); 885 886 rth = rcu_dereference(vrf->rth); 887 if (likely(rth)) { 888 dst = &rth->dst; 889 dst_hold(dst); 890 } 891 892 rcu_read_unlock(); 893 894 if (unlikely(!dst)) { 895 vrf_tx_error(vrf_dev, skb); 896 return NULL; 897 } 898 899 skb_dst_drop(skb); 900 skb_dst_set(skb, dst); 901 902 return skb; 903 } 904 905 static int vrf_output_direct_finish(struct net *net, struct sock *sk, 906 struct sk_buff *skb) 907 { 908 vrf_finish_direct(skb); 909 910 return vrf_ip_local_out(net, sk, skb); 911 } 912 913 static int vrf_output_direct(struct net *net, struct sock *sk, 914 struct sk_buff *skb) 915 { 916 int err = 1; 917 918 skb->protocol = htons(ETH_P_IP); 919 920 if (!(IPCB(skb)->flags & IPSKB_REROUTED)) 921 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb, 922 NULL, skb->dev, vrf_output_direct_finish); 923 924 if (likely(err == 1)) 925 vrf_finish_direct(skb); 926 927 return err; 928 } 929 930 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk, 931 struct sk_buff *skb) 932 { 933 int err; 934 935 err = vrf_output_direct(net, sk, skb); 936 if (likely(err == 1)) 937 err = vrf_ip_local_out(net, sk, skb); 938 939 return err; 940 } 941 942 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev, 943 struct sock *sk, 944 struct sk_buff *skb) 945 { 946 struct net *net = dev_net(vrf_dev); 947 int err; 948 949 skb->dev = vrf_dev; 950 951 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk, 952 skb, NULL, vrf_dev, vrf_ip_out_direct_finish); 953 954 if (likely(err == 1)) 955 err = vrf_output_direct(net, sk, skb); 956 957 if (likely(err == 1)) 958 return skb; 959 960 return NULL; 961 } 962 963 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev, 964 struct sock *sk, 965 struct sk_buff *skb) 966 { 967 /* don't divert multicast or local broadcast */ 968 if (ipv4_is_multicast(ip_hdr(skb)->daddr) || 969 ipv4_is_lbcast(ip_hdr(skb)->daddr)) 970 return skb; 971 972 vrf_nf_set_untracked(skb); 973 974 if (qdisc_tx_is_default(vrf_dev) || 975 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) 976 return vrf_ip_out_direct(vrf_dev, sk, skb); 977 978 return vrf_ip_out_redirect(vrf_dev, skb); 979 } 980 981 /* called with rcu lock held */ 982 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev, 983 struct sock *sk, 984 struct sk_buff *skb, 985 u16 proto) 986 { 987 switch (proto) { 988 case AF_INET: 989 return vrf_ip_out(vrf_dev, sk, skb); 990 case AF_INET6: 991 return vrf_ip6_out(vrf_dev, sk, skb); 992 } 993 994 return skb; 995 } 996 997 /* holding rtnl */ 998 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf) 999 { 1000 struct rtable *rth = rtnl_dereference(vrf->rth); 1001 struct net *net = dev_net(dev); 1002 struct dst_entry *dst; 1003 1004 RCU_INIT_POINTER(vrf->rth, NULL); 1005 synchronize_rcu(); 1006 1007 /* move dev in dst's to loopback so this VRF device can be deleted 1008 * - based on dst_ifdown 1009 */ 1010 if (rth) { 1011 dst = &rth->dst; 1012 netdev_ref_replace(dst->dev, net->loopback_dev, 1013 &dst->dev_tracker, GFP_KERNEL); 1014 dst->dev = net->loopback_dev; 1015 dst_release(dst); 1016 } 1017 } 1018 1019 static int vrf_rtable_create(struct net_device *dev) 1020 { 1021 struct net_vrf *vrf = netdev_priv(dev); 1022 struct rtable *rth; 1023 1024 if (!fib_new_table(dev_net(dev), vrf->tb_id)) 1025 return -ENOMEM; 1026 1027 /* create a dst for routing packets out through a VRF device */ 1028 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1); 1029 if (!rth) 1030 return -ENOMEM; 1031 1032 rth->dst.output = vrf_output; 1033 1034 rcu_assign_pointer(vrf->rth, rth); 1035 1036 return 0; 1037 } 1038 1039 /**************************** device handling ********************/ 1040 1041 /* cycle interface to flush neighbor cache and move routes across tables */ 1042 static void cycle_netdev(struct net_device *dev, 1043 struct netlink_ext_ack *extack) 1044 { 1045 unsigned int flags = dev->flags; 1046 int ret; 1047 1048 if (!netif_running(dev)) 1049 return; 1050 1051 ret = dev_change_flags(dev, flags & ~IFF_UP, extack); 1052 if (ret >= 0) 1053 ret = dev_change_flags(dev, flags, extack); 1054 1055 if (ret < 0) { 1056 netdev_err(dev, 1057 "Failed to cycle device %s; route tables might be wrong!\n", 1058 dev->name); 1059 } 1060 } 1061 1062 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1063 struct netlink_ext_ack *extack) 1064 { 1065 int ret; 1066 1067 /* do not allow loopback device to be enslaved to a VRF. 1068 * The vrf device acts as the loopback for the vrf. 1069 */ 1070 if (port_dev == dev_net(dev)->loopback_dev) { 1071 NL_SET_ERR_MSG(extack, 1072 "Can not enslave loopback device to a VRF"); 1073 return -EOPNOTSUPP; 1074 } 1075 1076 port_dev->priv_flags |= IFF_L3MDEV_SLAVE; 1077 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack); 1078 if (ret < 0) 1079 goto err; 1080 1081 cycle_netdev(port_dev, extack); 1082 1083 return 0; 1084 1085 err: 1086 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1087 return ret; 1088 } 1089 1090 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev, 1091 struct netlink_ext_ack *extack) 1092 { 1093 if (netif_is_l3_master(port_dev)) { 1094 NL_SET_ERR_MSG(extack, 1095 "Can not enslave an L3 master device to a VRF"); 1096 return -EINVAL; 1097 } 1098 1099 if (netif_is_l3_slave(port_dev)) 1100 return -EINVAL; 1101 1102 return do_vrf_add_slave(dev, port_dev, extack); 1103 } 1104 1105 /* inverse of do_vrf_add_slave */ 1106 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1107 { 1108 netdev_upper_dev_unlink(port_dev, dev); 1109 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; 1110 1111 cycle_netdev(port_dev, NULL); 1112 1113 return 0; 1114 } 1115 1116 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) 1117 { 1118 return do_vrf_del_slave(dev, port_dev); 1119 } 1120 1121 static void vrf_dev_uninit(struct net_device *dev) 1122 { 1123 struct net_vrf *vrf = netdev_priv(dev); 1124 1125 vrf_rtable_release(dev, vrf); 1126 vrf_rt6_release(dev, vrf); 1127 } 1128 1129 static int vrf_dev_init(struct net_device *dev) 1130 { 1131 struct net_vrf *vrf = netdev_priv(dev); 1132 1133 /* create the default dst which points back to us */ 1134 if (vrf_rtable_create(dev) != 0) 1135 goto out_nomem; 1136 1137 if (vrf_rt6_create(dev) != 0) 1138 goto out_rth; 1139 1140 dev->flags = IFF_MASTER | IFF_NOARP; 1141 1142 /* similarly, oper state is irrelevant; set to up to avoid confusion */ 1143 dev->operstate = IF_OPER_UP; 1144 netdev_lockdep_set_classes(dev); 1145 return 0; 1146 1147 out_rth: 1148 vrf_rtable_release(dev, vrf); 1149 out_nomem: 1150 return -ENOMEM; 1151 } 1152 1153 static const struct net_device_ops vrf_netdev_ops = { 1154 .ndo_init = vrf_dev_init, 1155 .ndo_uninit = vrf_dev_uninit, 1156 .ndo_start_xmit = vrf_xmit, 1157 .ndo_set_mac_address = eth_mac_addr, 1158 .ndo_add_slave = vrf_add_slave, 1159 .ndo_del_slave = vrf_del_slave, 1160 }; 1161 1162 static u32 vrf_fib_table(const struct net_device *dev) 1163 { 1164 struct net_vrf *vrf = netdev_priv(dev); 1165 1166 return vrf->tb_id; 1167 } 1168 1169 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb) 1170 { 1171 kfree_skb(skb); 1172 return 0; 1173 } 1174 1175 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook, 1176 struct sk_buff *skb, 1177 struct net_device *dev) 1178 { 1179 struct net *net = dev_net(dev); 1180 1181 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1) 1182 skb = NULL; /* kfree_skb(skb) handled by nf code */ 1183 1184 return skb; 1185 } 1186 1187 static int vrf_prepare_mac_header(struct sk_buff *skb, 1188 struct net_device *vrf_dev, u16 proto) 1189 { 1190 struct ethhdr *eth; 1191 int err; 1192 1193 /* in general, we do not know if there is enough space in the head of 1194 * the packet for hosting the mac header. 1195 */ 1196 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev)); 1197 if (unlikely(err)) 1198 /* no space in the skb head */ 1199 return -ENOBUFS; 1200 1201 __skb_push(skb, ETH_HLEN); 1202 eth = (struct ethhdr *)skb->data; 1203 1204 skb_reset_mac_header(skb); 1205 skb_reset_mac_len(skb); 1206 1207 /* we set the ethernet destination and the source addresses to the 1208 * address of the VRF device. 1209 */ 1210 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr); 1211 ether_addr_copy(eth->h_source, vrf_dev->dev_addr); 1212 eth->h_proto = htons(proto); 1213 1214 /* the destination address of the Ethernet frame corresponds to the 1215 * address set on the VRF interface; therefore, the packet is intended 1216 * to be processed locally. 1217 */ 1218 skb->protocol = eth->h_proto; 1219 skb->pkt_type = PACKET_HOST; 1220 1221 skb_postpush_rcsum(skb, skb->data, ETH_HLEN); 1222 1223 skb_pull_inline(skb, ETH_HLEN); 1224 1225 return 0; 1226 } 1227 1228 /* prepare and add the mac header to the packet if it was not set previously. 1229 * In this way, packet sniffers such as tcpdump can parse the packet correctly. 1230 * If the mac header was already set, the original mac header is left 1231 * untouched and the function returns immediately. 1232 */ 1233 static int vrf_add_mac_header_if_unset(struct sk_buff *skb, 1234 struct net_device *vrf_dev, 1235 u16 proto, struct net_device *orig_dev) 1236 { 1237 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev)) 1238 return 0; 1239 1240 return vrf_prepare_mac_header(skb, vrf_dev, proto); 1241 } 1242 1243 #if IS_ENABLED(CONFIG_IPV6) 1244 /* neighbor handling is done with actual device; do not want 1245 * to flip skb->dev for those ndisc packets. This really fails 1246 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is 1247 * a start. 1248 */ 1249 static bool ipv6_ndisc_frame(const struct sk_buff *skb) 1250 { 1251 const struct ipv6hdr *iph = ipv6_hdr(skb); 1252 bool rc = false; 1253 1254 if (iph->nexthdr == NEXTHDR_ICMP) { 1255 const struct icmp6hdr *icmph; 1256 struct icmp6hdr _icmph; 1257 1258 icmph = skb_header_pointer(skb, sizeof(*iph), 1259 sizeof(_icmph), &_icmph); 1260 if (!icmph) 1261 goto out; 1262 1263 switch (icmph->icmp6_type) { 1264 case NDISC_ROUTER_SOLICITATION: 1265 case NDISC_ROUTER_ADVERTISEMENT: 1266 case NDISC_NEIGHBOUR_SOLICITATION: 1267 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1268 case NDISC_REDIRECT: 1269 rc = true; 1270 break; 1271 } 1272 } 1273 1274 out: 1275 return rc; 1276 } 1277 1278 static struct rt6_info *vrf_ip6_route_lookup(struct net *net, 1279 const struct net_device *dev, 1280 struct flowi6 *fl6, 1281 int ifindex, 1282 const struct sk_buff *skb, 1283 int flags) 1284 { 1285 struct net_vrf *vrf = netdev_priv(dev); 1286 1287 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags); 1288 } 1289 1290 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev, 1291 int ifindex) 1292 { 1293 const struct ipv6hdr *iph = ipv6_hdr(skb); 1294 struct flowi6 fl6 = { 1295 .flowi6_iif = ifindex, 1296 .flowi6_mark = skb->mark, 1297 .flowi6_proto = iph->nexthdr, 1298 .daddr = iph->daddr, 1299 .saddr = iph->saddr, 1300 .flowlabel = ip6_flowinfo(iph), 1301 }; 1302 struct net *net = dev_net(vrf_dev); 1303 struct rt6_info *rt6; 1304 1305 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb, 1306 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE); 1307 if (unlikely(!rt6)) 1308 return; 1309 1310 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst)) 1311 return; 1312 1313 skb_dst_set(skb, &rt6->dst); 1314 } 1315 1316 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1317 struct sk_buff *skb) 1318 { 1319 int orig_iif = skb->skb_iif; 1320 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr); 1321 bool is_ndisc = ipv6_ndisc_frame(skb); 1322 1323 /* loopback, multicast & non-ND link-local traffic; do not push through 1324 * packet taps again. Reset pkt_type for upper layers to process skb. 1325 * For non-loopback strict packets, determine the dst using the original 1326 * ifindex. 1327 */ 1328 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) { 1329 skb->dev = vrf_dev; 1330 skb->skb_iif = vrf_dev->ifindex; 1331 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1332 1333 if (skb->pkt_type == PACKET_LOOPBACK) 1334 skb->pkt_type = PACKET_HOST; 1335 else 1336 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1337 1338 goto out; 1339 } 1340 1341 /* if packet is NDISC then keep the ingress interface */ 1342 if (!is_ndisc) { 1343 struct net_device *orig_dev = skb->dev; 1344 1345 dev_dstats_rx_add(vrf_dev, skb->len); 1346 skb->dev = vrf_dev; 1347 skb->skb_iif = vrf_dev->ifindex; 1348 1349 if (!list_empty(&vrf_dev->ptype_all)) { 1350 int err; 1351 1352 err = vrf_add_mac_header_if_unset(skb, vrf_dev, 1353 ETH_P_IPV6, 1354 orig_dev); 1355 if (likely(!err)) { 1356 skb_push(skb, skb->mac_len); 1357 dev_queue_xmit_nit(skb, vrf_dev); 1358 skb_pull(skb, skb->mac_len); 1359 } 1360 } 1361 1362 IP6CB(skb)->flags |= IP6SKB_L3SLAVE; 1363 } 1364 1365 if (need_strict) 1366 vrf_ip6_input_dst(skb, vrf_dev, orig_iif); 1367 1368 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev); 1369 out: 1370 return skb; 1371 } 1372 1373 #else 1374 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, 1375 struct sk_buff *skb) 1376 { 1377 return skb; 1378 } 1379 #endif 1380 1381 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, 1382 struct sk_buff *skb) 1383 { 1384 struct net_device *orig_dev = skb->dev; 1385 1386 skb->dev = vrf_dev; 1387 skb->skb_iif = vrf_dev->ifindex; 1388 IPCB(skb)->flags |= IPSKB_L3SLAVE; 1389 1390 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) 1391 goto out; 1392 1393 /* loopback traffic; do not push through packet taps again. 1394 * Reset pkt_type for upper layers to process skb 1395 */ 1396 if (skb->pkt_type == PACKET_LOOPBACK) { 1397 skb->pkt_type = PACKET_HOST; 1398 goto out; 1399 } 1400 1401 dev_dstats_rx_add(vrf_dev, skb->len); 1402 1403 if (!list_empty(&vrf_dev->ptype_all)) { 1404 int err; 1405 1406 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP, 1407 orig_dev); 1408 if (likely(!err)) { 1409 skb_push(skb, skb->mac_len); 1410 dev_queue_xmit_nit(skb, vrf_dev); 1411 skb_pull(skb, skb->mac_len); 1412 } 1413 } 1414 1415 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev); 1416 out: 1417 return skb; 1418 } 1419 1420 /* called with rcu lock held */ 1421 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, 1422 struct sk_buff *skb, 1423 u16 proto) 1424 { 1425 switch (proto) { 1426 case AF_INET: 1427 return vrf_ip_rcv(vrf_dev, skb); 1428 case AF_INET6: 1429 return vrf_ip6_rcv(vrf_dev, skb); 1430 } 1431 1432 return skb; 1433 } 1434 1435 #if IS_ENABLED(CONFIG_IPV6) 1436 /* send to link-local or multicast address via interface enslaved to 1437 * VRF device. Force lookup to VRF table without changing flow struct 1438 * Note: Caller to this function must hold rcu_read_lock() and no refcnt 1439 * is taken on the dst by this function. 1440 */ 1441 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev, 1442 struct flowi6 *fl6) 1443 { 1444 struct net *net = dev_net(dev); 1445 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF; 1446 struct dst_entry *dst = NULL; 1447 struct rt6_info *rt; 1448 1449 /* VRF device does not have a link-local address and 1450 * sending packets to link-local or mcast addresses over 1451 * a VRF device does not make sense 1452 */ 1453 if (fl6->flowi6_oif == dev->ifindex) { 1454 dst = &net->ipv6.ip6_null_entry->dst; 1455 return dst; 1456 } 1457 1458 if (!ipv6_addr_any(&fl6->saddr)) 1459 flags |= RT6_LOOKUP_F_HAS_SADDR; 1460 1461 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags); 1462 if (rt) 1463 dst = &rt->dst; 1464 1465 return dst; 1466 } 1467 #endif 1468 1469 static const struct l3mdev_ops vrf_l3mdev_ops = { 1470 .l3mdev_fib_table = vrf_fib_table, 1471 .l3mdev_l3_rcv = vrf_l3_rcv, 1472 .l3mdev_l3_out = vrf_l3_out, 1473 #if IS_ENABLED(CONFIG_IPV6) 1474 .l3mdev_link_scope_lookup = vrf_link_scope_lookup, 1475 #endif 1476 }; 1477 1478 static void vrf_get_drvinfo(struct net_device *dev, 1479 struct ethtool_drvinfo *info) 1480 { 1481 strscpy(info->driver, DRV_NAME, sizeof(info->driver)); 1482 strscpy(info->version, DRV_VERSION, sizeof(info->version)); 1483 } 1484 1485 static const struct ethtool_ops vrf_ethtool_ops = { 1486 .get_drvinfo = vrf_get_drvinfo, 1487 }; 1488 1489 static inline size_t vrf_fib_rule_nl_size(void) 1490 { 1491 size_t sz; 1492 1493 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr)); 1494 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */ 1495 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */ 1496 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */ 1497 1498 return sz; 1499 } 1500 1501 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it) 1502 { 1503 struct fib_rule_hdr *frh; 1504 struct nlmsghdr *nlh; 1505 struct sk_buff *skb; 1506 int err; 1507 1508 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) && 1509 !ipv6_mod_enabled()) 1510 return 0; 1511 1512 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL); 1513 if (!skb) 1514 return -ENOMEM; 1515 1516 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0); 1517 if (!nlh) 1518 goto nla_put_failure; 1519 1520 /* rule only needs to appear once */ 1521 nlh->nlmsg_flags |= NLM_F_EXCL; 1522 1523 frh = nlmsg_data(nlh); 1524 memset(frh, 0, sizeof(*frh)); 1525 frh->family = family; 1526 frh->action = FR_ACT_TO_TBL; 1527 1528 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL)) 1529 goto nla_put_failure; 1530 1531 if (nla_put_u8(skb, FRA_L3MDEV, 1)) 1532 goto nla_put_failure; 1533 1534 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF)) 1535 goto nla_put_failure; 1536 1537 nlmsg_end(skb, nlh); 1538 1539 if (add_it) { 1540 err = fib_newrule(dev_net(dev), skb, nlh, NULL, true); 1541 if (err == -EEXIST) 1542 err = 0; 1543 } else { 1544 err = fib_delrule(dev_net(dev), skb, nlh, NULL, true); 1545 if (err == -ENOENT) 1546 err = 0; 1547 } 1548 nlmsg_free(skb); 1549 1550 return err; 1551 1552 nla_put_failure: 1553 nlmsg_free(skb); 1554 1555 return -EMSGSIZE; 1556 } 1557 1558 static int vrf_add_fib_rules(const struct net_device *dev) 1559 { 1560 int err; 1561 1562 err = vrf_fib_rule(dev, AF_INET, true); 1563 if (err < 0) 1564 goto out_err; 1565 1566 err = vrf_fib_rule(dev, AF_INET6, true); 1567 if (err < 0) 1568 goto ipv6_err; 1569 1570 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1571 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true); 1572 if (err < 0) 1573 goto ipmr_err; 1574 #endif 1575 1576 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1577 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true); 1578 if (err < 0) 1579 goto ip6mr_err; 1580 #endif 1581 1582 return 0; 1583 1584 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES) 1585 ip6mr_err: 1586 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false); 1587 #endif 1588 1589 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES) 1590 ipmr_err: 1591 vrf_fib_rule(dev, AF_INET6, false); 1592 #endif 1593 1594 ipv6_err: 1595 vrf_fib_rule(dev, AF_INET, false); 1596 1597 out_err: 1598 netdev_err(dev, "Failed to add FIB rules.\n"); 1599 return err; 1600 } 1601 1602 static void vrf_setup(struct net_device *dev) 1603 { 1604 ether_setup(dev); 1605 1606 /* Initialize the device structure. */ 1607 dev->netdev_ops = &vrf_netdev_ops; 1608 dev->l3mdev_ops = &vrf_l3mdev_ops; 1609 dev->ethtool_ops = &vrf_ethtool_ops; 1610 dev->needs_free_netdev = true; 1611 1612 /* Fill in device structure with ethernet-generic values. */ 1613 eth_hw_addr_random(dev); 1614 1615 /* don't acquire vrf device's netif_tx_lock when transmitting */ 1616 dev->lltx = true; 1617 1618 /* don't allow vrf devices to change network namespaces. */ 1619 dev->netns_immutable = true; 1620 1621 /* does not make sense for a VLAN to be added to a vrf device */ 1622 dev->features |= NETIF_F_VLAN_CHALLENGED; 1623 1624 /* enable offload features */ 1625 dev->features |= NETIF_F_GSO_SOFTWARE; 1626 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC; 1627 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA; 1628 1629 dev->hw_features = dev->features; 1630 dev->hw_enc_features = dev->features; 1631 1632 /* default to no qdisc; user can add if desired */ 1633 dev->priv_flags |= IFF_NO_QUEUE; 1634 dev->priv_flags |= IFF_NO_RX_HANDLER; 1635 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1636 1637 /* VRF devices do not care about MTU, but if the MTU is set 1638 * too low then the ipv4 and ipv6 protocols are disabled 1639 * which breaks networking. 1640 */ 1641 dev->min_mtu = IPV6_MIN_MTU; 1642 dev->max_mtu = IP6_MAX_MTU; 1643 dev->mtu = dev->max_mtu; 1644 1645 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS; 1646 } 1647 1648 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[], 1649 struct netlink_ext_ack *extack) 1650 { 1651 if (tb[IFLA_ADDRESS]) { 1652 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1653 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1654 return -EINVAL; 1655 } 1656 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1657 NL_SET_ERR_MSG(extack, "Invalid hardware address"); 1658 return -EADDRNOTAVAIL; 1659 } 1660 } 1661 return 0; 1662 } 1663 1664 static void vrf_dellink(struct net_device *dev, struct list_head *head) 1665 { 1666 struct net_device *port_dev; 1667 struct list_head *iter; 1668 1669 netdev_for_each_lower_dev(dev, port_dev, iter) 1670 vrf_del_slave(dev, port_dev); 1671 1672 vrf_map_unregister_dev(dev); 1673 1674 unregister_netdevice_queue(dev, head); 1675 } 1676 1677 static int vrf_newlink(struct net_device *dev, 1678 struct rtnl_newlink_params *params, 1679 struct netlink_ext_ack *extack) 1680 { 1681 struct net_vrf *vrf = netdev_priv(dev); 1682 struct nlattr **data = params->data; 1683 struct netns_vrf *nn_vrf; 1684 bool *add_fib_rules; 1685 struct net *net; 1686 int err; 1687 1688 if (!data || !data[IFLA_VRF_TABLE]) { 1689 NL_SET_ERR_MSG(extack, "VRF table id is missing"); 1690 return -EINVAL; 1691 } 1692 1693 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); 1694 if (vrf->tb_id == RT_TABLE_UNSPEC) { 1695 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE], 1696 "Invalid VRF table id"); 1697 return -EINVAL; 1698 } 1699 1700 dev->priv_flags |= IFF_L3MDEV_MASTER; 1701 1702 err = register_netdevice(dev); 1703 if (err) 1704 goto out; 1705 1706 /* mapping between table_id and vrf; 1707 * note: such binding could not be done in the dev init function 1708 * because dev->ifindex id is not available yet. 1709 */ 1710 vrf->ifindex = dev->ifindex; 1711 1712 err = vrf_map_register_dev(dev, extack); 1713 if (err) { 1714 unregister_netdevice(dev); 1715 goto out; 1716 } 1717 1718 net = dev_net(dev); 1719 nn_vrf = net_generic(net, vrf_net_id); 1720 1721 add_fib_rules = &nn_vrf->add_fib_rules; 1722 if (*add_fib_rules) { 1723 err = vrf_add_fib_rules(dev); 1724 if (err) { 1725 vrf_map_unregister_dev(dev); 1726 unregister_netdevice(dev); 1727 goto out; 1728 } 1729 *add_fib_rules = false; 1730 } 1731 1732 out: 1733 return err; 1734 } 1735 1736 static size_t vrf_nl_getsize(const struct net_device *dev) 1737 { 1738 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ 1739 } 1740 1741 static int vrf_fillinfo(struct sk_buff *skb, 1742 const struct net_device *dev) 1743 { 1744 struct net_vrf *vrf = netdev_priv(dev); 1745 1746 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); 1747 } 1748 1749 static size_t vrf_get_slave_size(const struct net_device *bond_dev, 1750 const struct net_device *slave_dev) 1751 { 1752 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ 1753 } 1754 1755 static int vrf_fill_slave_info(struct sk_buff *skb, 1756 const struct net_device *vrf_dev, 1757 const struct net_device *slave_dev) 1758 { 1759 struct net_vrf *vrf = netdev_priv(vrf_dev); 1760 1761 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) 1762 return -EMSGSIZE; 1763 1764 return 0; 1765 } 1766 1767 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { 1768 [IFLA_VRF_TABLE] = { .type = NLA_U32 }, 1769 }; 1770 1771 static struct rtnl_link_ops vrf_link_ops __read_mostly = { 1772 .kind = DRV_NAME, 1773 .priv_size = sizeof(struct net_vrf), 1774 1775 .get_size = vrf_nl_getsize, 1776 .policy = vrf_nl_policy, 1777 .validate = vrf_validate, 1778 .fill_info = vrf_fillinfo, 1779 1780 .get_slave_size = vrf_get_slave_size, 1781 .fill_slave_info = vrf_fill_slave_info, 1782 1783 .newlink = vrf_newlink, 1784 .dellink = vrf_dellink, 1785 .setup = vrf_setup, 1786 .maxtype = IFLA_VRF_MAX, 1787 }; 1788 1789 static int vrf_device_event(struct notifier_block *unused, 1790 unsigned long event, void *ptr) 1791 { 1792 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1793 1794 /* only care about unregister events to drop slave references */ 1795 if (event == NETDEV_UNREGISTER) { 1796 struct net_device *vrf_dev; 1797 1798 if (!netif_is_l3_slave(dev)) 1799 goto out; 1800 1801 vrf_dev = netdev_master_upper_dev_get(dev); 1802 vrf_del_slave(vrf_dev, dev); 1803 } 1804 out: 1805 return NOTIFY_DONE; 1806 } 1807 1808 static struct notifier_block vrf_notifier_block __read_mostly = { 1809 .notifier_call = vrf_device_event, 1810 }; 1811 1812 static int vrf_map_init(struct vrf_map *vmap) 1813 { 1814 spin_lock_init(&vmap->vmap_lock); 1815 hash_init(vmap->ht); 1816 1817 vmap->strict_mode = false; 1818 1819 return 0; 1820 } 1821 1822 #ifdef CONFIG_SYSCTL 1823 static bool vrf_strict_mode(struct vrf_map *vmap) 1824 { 1825 bool strict_mode; 1826 1827 vrf_map_lock(vmap); 1828 strict_mode = vmap->strict_mode; 1829 vrf_map_unlock(vmap); 1830 1831 return strict_mode; 1832 } 1833 1834 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode) 1835 { 1836 bool *cur_mode; 1837 int res = 0; 1838 1839 vrf_map_lock(vmap); 1840 1841 cur_mode = &vmap->strict_mode; 1842 if (*cur_mode == new_mode) 1843 goto unlock; 1844 1845 if (*cur_mode) { 1846 /* disable strict mode */ 1847 *cur_mode = false; 1848 } else { 1849 if (vmap->shared_tables) { 1850 /* we cannot allow strict_mode because there are some 1851 * vrfs that share one or more tables. 1852 */ 1853 res = -EBUSY; 1854 goto unlock; 1855 } 1856 1857 /* no tables are shared among vrfs, so we can go back 1858 * to 1:1 association between a vrf with its table. 1859 */ 1860 *cur_mode = true; 1861 } 1862 1863 unlock: 1864 vrf_map_unlock(vmap); 1865 1866 return res; 1867 } 1868 1869 static int vrf_shared_table_handler(const struct ctl_table *table, int write, 1870 void *buffer, size_t *lenp, loff_t *ppos) 1871 { 1872 struct net *net = (struct net *)table->extra1; 1873 struct vrf_map *vmap = netns_vrf_map(net); 1874 int proc_strict_mode = 0; 1875 struct ctl_table tmp = { 1876 .procname = table->procname, 1877 .data = &proc_strict_mode, 1878 .maxlen = sizeof(int), 1879 .mode = table->mode, 1880 .extra1 = SYSCTL_ZERO, 1881 .extra2 = SYSCTL_ONE, 1882 }; 1883 int ret; 1884 1885 if (!write) 1886 proc_strict_mode = vrf_strict_mode(vmap); 1887 1888 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 1889 1890 if (write && ret == 0) 1891 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode); 1892 1893 return ret; 1894 } 1895 1896 static const struct ctl_table vrf_table[] = { 1897 { 1898 .procname = "strict_mode", 1899 .data = NULL, 1900 .maxlen = sizeof(int), 1901 .mode = 0644, 1902 .proc_handler = vrf_shared_table_handler, 1903 /* set by the vrf_netns_init */ 1904 .extra1 = NULL, 1905 }, 1906 }; 1907 1908 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1909 { 1910 struct ctl_table *table; 1911 1912 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL); 1913 if (!table) 1914 return -ENOMEM; 1915 1916 /* init the extra1 parameter with the reference to current netns */ 1917 table[0].extra1 = net; 1918 1919 nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table, 1920 ARRAY_SIZE(vrf_table)); 1921 if (!nn_vrf->ctl_hdr) { 1922 kfree(table); 1923 return -ENOMEM; 1924 } 1925 1926 return 0; 1927 } 1928 1929 static void vrf_netns_exit_sysctl(struct net *net) 1930 { 1931 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1932 const struct ctl_table *table; 1933 1934 table = nn_vrf->ctl_hdr->ctl_table_arg; 1935 unregister_net_sysctl_table(nn_vrf->ctl_hdr); 1936 kfree(table); 1937 } 1938 #else 1939 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf) 1940 { 1941 return 0; 1942 } 1943 1944 static void vrf_netns_exit_sysctl(struct net *net) 1945 { 1946 } 1947 #endif 1948 1949 /* Initialize per network namespace state */ 1950 static int __net_init vrf_netns_init(struct net *net) 1951 { 1952 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id); 1953 1954 nn_vrf->add_fib_rules = true; 1955 vrf_map_init(&nn_vrf->vmap); 1956 1957 return vrf_netns_init_sysctl(net, nn_vrf); 1958 } 1959 1960 static void __net_exit vrf_netns_exit(struct net *net) 1961 { 1962 vrf_netns_exit_sysctl(net); 1963 } 1964 1965 static struct pernet_operations vrf_net_ops __net_initdata = { 1966 .init = vrf_netns_init, 1967 .exit = vrf_netns_exit, 1968 .id = &vrf_net_id, 1969 .size = sizeof(struct netns_vrf), 1970 }; 1971 1972 static int __init vrf_init_module(void) 1973 { 1974 int rc; 1975 1976 register_netdevice_notifier(&vrf_notifier_block); 1977 1978 rc = register_pernet_subsys(&vrf_net_ops); 1979 if (rc < 0) 1980 goto error; 1981 1982 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF, 1983 vrf_ifindex_lookup_by_table_id); 1984 if (rc < 0) 1985 goto unreg_pernet; 1986 1987 rc = rtnl_link_register(&vrf_link_ops); 1988 if (rc < 0) 1989 goto table_lookup_unreg; 1990 1991 return 0; 1992 1993 table_lookup_unreg: 1994 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF, 1995 vrf_ifindex_lookup_by_table_id); 1996 1997 unreg_pernet: 1998 unregister_pernet_subsys(&vrf_net_ops); 1999 2000 error: 2001 unregister_netdevice_notifier(&vrf_notifier_block); 2002 return rc; 2003 } 2004 2005 module_init(vrf_init_module); 2006 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); 2007 MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); 2008 MODULE_LICENSE("GPL"); 2009 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 2010 MODULE_VERSION(DRV_VERSION); 2011