xref: /openbmc/linux/drivers/net/vrf.c (revision fe0a5788)
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/module.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/etherdevice.h>
16 #include <linux/ip.h>
17 #include <linux/init.h>
18 #include <linux/moduleparam.h>
19 #include <linux/netfilter.h>
20 #include <linux/rtnetlink.h>
21 #include <net/rtnetlink.h>
22 #include <linux/u64_stats_sync.h>
23 #include <linux/hashtable.h>
24 #include <linux/spinlock_types.h>
25 
26 #include <linux/inetdevice.h>
27 #include <net/arp.h>
28 #include <net/ip.h>
29 #include <net/ip_fib.h>
30 #include <net/ip6_fib.h>
31 #include <net/ip6_route.h>
32 #include <net/route.h>
33 #include <net/addrconf.h>
34 #include <net/l3mdev.h>
35 #include <net/fib_rules.h>
36 #include <net/netns/generic.h>
37 
38 #define DRV_NAME	"vrf"
39 #define DRV_VERSION	"1.1"
40 
41 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
42 
43 #define HT_MAP_BITS	4
44 #define HASH_INITVAL	((u32)0xcafef00d)
45 
46 struct  vrf_map {
47 	DECLARE_HASHTABLE(ht, HT_MAP_BITS);
48 	spinlock_t vmap_lock;
49 
50 	/* shared_tables:
51 	 * count how many distinct tables do not comply with the strict mode
52 	 * requirement.
53 	 * shared_tables value must be 0 in order to enable the strict mode.
54 	 *
55 	 * example of the evolution of shared_tables:
56 	 *                                                        | time
57 	 * add  vrf0 --> table 100        shared_tables = 0       | t0
58 	 * add  vrf1 --> table 101        shared_tables = 0       | t1
59 	 * add  vrf2 --> table 100        shared_tables = 1       | t2
60 	 * add  vrf3 --> table 100        shared_tables = 1       | t3
61 	 * add  vrf4 --> table 101        shared_tables = 2       v t4
62 	 *
63 	 * shared_tables is a "step function" (or "staircase function")
64 	 * and it is increased by one when the second vrf is associated to a
65 	 * table.
66 	 *
67 	 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
68 	 *
69 	 * at t3, another dev (vrf3) is bound to the same table 100 but the
70 	 * value of shared_tables is still 1.
71 	 * This means that no matter how many new vrfs will register on the
72 	 * table 100, the shared_tables will not increase (considering only
73 	 * table 100).
74 	 *
75 	 * at t4, vrf4 is bound to table 101, and shared_tables = 2.
76 	 *
77 	 * Looking at the value of shared_tables we can immediately know if
78 	 * the strict_mode can or cannot be enforced. Indeed, strict_mode
79 	 * can be enforced iff shared_tables = 0.
80 	 *
81 	 * Conversely, shared_tables is decreased when a vrf is de-associated
82 	 * from a table with exactly two associated vrfs.
83 	 */
84 	u32 shared_tables;
85 
86 	bool strict_mode;
87 };
88 
89 struct vrf_map_elem {
90 	struct hlist_node hnode;
91 	struct list_head vrf_list;  /* VRFs registered to this table */
92 
93 	u32 table_id;
94 	int users;
95 	int ifindex;
96 };
97 
98 static unsigned int vrf_net_id;
99 
100 /* per netns vrf data */
101 struct netns_vrf {
102 	/* protected by rtnl lock */
103 	bool add_fib_rules;
104 
105 	struct vrf_map vmap;
106 	struct ctl_table_header	*ctl_hdr;
107 };
108 
109 struct net_vrf {
110 	struct rtable __rcu	*rth;
111 	struct rt6_info	__rcu	*rt6;
112 #if IS_ENABLED(CONFIG_IPV6)
113 	struct fib6_table	*fib6_table;
114 #endif
115 	u32                     tb_id;
116 
117 	struct list_head	me_list;   /* entry in vrf_map_elem */
118 	int			ifindex;
119 };
120 
121 struct pcpu_dstats {
122 	u64			tx_pkts;
123 	u64			tx_bytes;
124 	u64			tx_drps;
125 	u64			rx_pkts;
126 	u64			rx_bytes;
127 	u64			rx_drps;
128 	struct u64_stats_sync	syncp;
129 };
130 
131 static void vrf_rx_stats(struct net_device *dev, int len)
132 {
133 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
134 
135 	u64_stats_update_begin(&dstats->syncp);
136 	dstats->rx_pkts++;
137 	dstats->rx_bytes += len;
138 	u64_stats_update_end(&dstats->syncp);
139 }
140 
141 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
142 {
143 	vrf_dev->stats.tx_errors++;
144 	kfree_skb(skb);
145 }
146 
147 static void vrf_get_stats64(struct net_device *dev,
148 			    struct rtnl_link_stats64 *stats)
149 {
150 	int i;
151 
152 	for_each_possible_cpu(i) {
153 		const struct pcpu_dstats *dstats;
154 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
155 		unsigned int start;
156 
157 		dstats = per_cpu_ptr(dev->dstats, i);
158 		do {
159 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
160 			tbytes = dstats->tx_bytes;
161 			tpkts = dstats->tx_pkts;
162 			tdrops = dstats->tx_drps;
163 			rbytes = dstats->rx_bytes;
164 			rpkts = dstats->rx_pkts;
165 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
166 		stats->tx_bytes += tbytes;
167 		stats->tx_packets += tpkts;
168 		stats->tx_dropped += tdrops;
169 		stats->rx_bytes += rbytes;
170 		stats->rx_packets += rpkts;
171 	}
172 }
173 
174 static struct vrf_map *netns_vrf_map(struct net *net)
175 {
176 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
177 
178 	return &nn_vrf->vmap;
179 }
180 
181 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
182 {
183 	return netns_vrf_map(dev_net(dev));
184 }
185 
186 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
187 {
188 	struct list_head *me_head = &me->vrf_list;
189 	struct net_vrf *vrf;
190 
191 	if (list_empty(me_head))
192 		return -ENODEV;
193 
194 	vrf = list_first_entry(me_head, struct net_vrf, me_list);
195 
196 	return vrf->ifindex;
197 }
198 
199 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
200 {
201 	struct vrf_map_elem *me;
202 
203 	me = kmalloc(sizeof(*me), flags);
204 	if (!me)
205 		return NULL;
206 
207 	return me;
208 }
209 
210 static void vrf_map_elem_free(struct vrf_map_elem *me)
211 {
212 	kfree(me);
213 }
214 
215 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
216 			      int ifindex, int users)
217 {
218 	me->table_id = table_id;
219 	me->ifindex = ifindex;
220 	me->users = users;
221 	INIT_LIST_HEAD(&me->vrf_list);
222 }
223 
224 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
225 						u32 table_id)
226 {
227 	struct vrf_map_elem *me;
228 	u32 key;
229 
230 	key = jhash_1word(table_id, HASH_INITVAL);
231 	hash_for_each_possible(vmap->ht, me, hnode, key) {
232 		if (me->table_id == table_id)
233 			return me;
234 	}
235 
236 	return NULL;
237 }
238 
239 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
240 {
241 	u32 table_id = me->table_id;
242 	u32 key;
243 
244 	key = jhash_1word(table_id, HASH_INITVAL);
245 	hash_add(vmap->ht, &me->hnode, key);
246 }
247 
248 static void vrf_map_del_elem(struct vrf_map_elem *me)
249 {
250 	hash_del(&me->hnode);
251 }
252 
253 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
254 {
255 	spin_lock(&vmap->vmap_lock);
256 }
257 
258 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
259 {
260 	spin_unlock(&vmap->vmap_lock);
261 }
262 
263 /* called with rtnl lock held */
264 static int
265 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
266 {
267 	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
268 	struct net_vrf *vrf = netdev_priv(dev);
269 	struct vrf_map_elem *new_me, *me;
270 	u32 table_id = vrf->tb_id;
271 	bool free_new_me = false;
272 	int users;
273 	int res;
274 
275 	/* we pre-allocate elements used in the spin-locked section (so that we
276 	 * keep the spinlock as short as possibile).
277 	 */
278 	new_me = vrf_map_elem_alloc(GFP_KERNEL);
279 	if (!new_me)
280 		return -ENOMEM;
281 
282 	vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
283 
284 	vrf_map_lock(vmap);
285 
286 	me = vrf_map_lookup_elem(vmap, table_id);
287 	if (!me) {
288 		me = new_me;
289 		vrf_map_add_elem(vmap, me);
290 		goto link_vrf;
291 	}
292 
293 	/* we already have an entry in the vrf_map, so it means there is (at
294 	 * least) a vrf registered on the specific table.
295 	 */
296 	free_new_me = true;
297 	if (vmap->strict_mode) {
298 		/* vrfs cannot share the same table */
299 		NL_SET_ERR_MSG(extack, "Table is used by another VRF");
300 		res = -EBUSY;
301 		goto unlock;
302 	}
303 
304 link_vrf:
305 	users = ++me->users;
306 	if (users == 2)
307 		++vmap->shared_tables;
308 
309 	list_add(&vrf->me_list, &me->vrf_list);
310 
311 	res = 0;
312 
313 unlock:
314 	vrf_map_unlock(vmap);
315 
316 	/* clean-up, if needed */
317 	if (free_new_me)
318 		vrf_map_elem_free(new_me);
319 
320 	return res;
321 }
322 
323 /* called with rtnl lock held */
324 static void vrf_map_unregister_dev(struct net_device *dev)
325 {
326 	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
327 	struct net_vrf *vrf = netdev_priv(dev);
328 	u32 table_id = vrf->tb_id;
329 	struct vrf_map_elem *me;
330 	int users;
331 
332 	vrf_map_lock(vmap);
333 
334 	me = vrf_map_lookup_elem(vmap, table_id);
335 	if (!me)
336 		goto unlock;
337 
338 	list_del(&vrf->me_list);
339 
340 	users = --me->users;
341 	if (users == 1) {
342 		--vmap->shared_tables;
343 	} else if (users == 0) {
344 		vrf_map_del_elem(me);
345 
346 		/* no one will refer to this element anymore */
347 		vrf_map_elem_free(me);
348 	}
349 
350 unlock:
351 	vrf_map_unlock(vmap);
352 }
353 
354 /* return the vrf device index associated with the table_id */
355 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
356 {
357 	struct vrf_map *vmap = netns_vrf_map(net);
358 	struct vrf_map_elem *me;
359 	int ifindex;
360 
361 	vrf_map_lock(vmap);
362 
363 	if (!vmap->strict_mode) {
364 		ifindex = -EPERM;
365 		goto unlock;
366 	}
367 
368 	me = vrf_map_lookup_elem(vmap, table_id);
369 	if (!me) {
370 		ifindex = -ENODEV;
371 		goto unlock;
372 	}
373 
374 	ifindex = vrf_map_elem_get_vrf_ifindex(me);
375 
376 unlock:
377 	vrf_map_unlock(vmap);
378 
379 	return ifindex;
380 }
381 
382 /* by default VRF devices do not have a qdisc and are expected
383  * to be created with only a single queue.
384  */
385 static bool qdisc_tx_is_default(const struct net_device *dev)
386 {
387 	struct netdev_queue *txq;
388 	struct Qdisc *qdisc;
389 
390 	if (dev->num_tx_queues > 1)
391 		return false;
392 
393 	txq = netdev_get_tx_queue(dev, 0);
394 	qdisc = rcu_access_pointer(txq->qdisc);
395 
396 	return !qdisc->enqueue;
397 }
398 
399 /* Local traffic destined to local address. Reinsert the packet to rx
400  * path, similar to loopback handling.
401  */
402 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
403 			  struct dst_entry *dst)
404 {
405 	int len = skb->len;
406 
407 	skb_orphan(skb);
408 
409 	skb_dst_set(skb, dst);
410 
411 	/* set pkt_type to avoid skb hitting packet taps twice -
412 	 * once on Tx and again in Rx processing
413 	 */
414 	skb->pkt_type = PACKET_LOOPBACK;
415 
416 	skb->protocol = eth_type_trans(skb, dev);
417 
418 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
419 		vrf_rx_stats(dev, len);
420 	else
421 		this_cpu_inc(dev->dstats->rx_drps);
422 
423 	return NETDEV_TX_OK;
424 }
425 
426 #if IS_ENABLED(CONFIG_IPV6)
427 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
428 			     struct sk_buff *skb)
429 {
430 	int err;
431 
432 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
433 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
434 
435 	if (likely(err == 1))
436 		err = dst_output(net, sk, skb);
437 
438 	return err;
439 }
440 
441 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
442 					   struct net_device *dev)
443 {
444 	const struct ipv6hdr *iph;
445 	struct net *net = dev_net(skb->dev);
446 	struct flowi6 fl6;
447 	int ret = NET_XMIT_DROP;
448 	struct dst_entry *dst;
449 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
450 
451 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
452 		goto err;
453 
454 	iph = ipv6_hdr(skb);
455 
456 	memset(&fl6, 0, sizeof(fl6));
457 	/* needed to match OIF rule */
458 	fl6.flowi6_oif = dev->ifindex;
459 	fl6.flowi6_iif = LOOPBACK_IFINDEX;
460 	fl6.daddr = iph->daddr;
461 	fl6.saddr = iph->saddr;
462 	fl6.flowlabel = ip6_flowinfo(iph);
463 	fl6.flowi6_mark = skb->mark;
464 	fl6.flowi6_proto = iph->nexthdr;
465 	fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
466 
467 	dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
468 	if (IS_ERR(dst) || dst == dst_null)
469 		goto err;
470 
471 	skb_dst_drop(skb);
472 
473 	/* if dst.dev is loopback or the VRF device again this is locally
474 	 * originated traffic destined to a local address. Short circuit
475 	 * to Rx path
476 	 */
477 	if (dst->dev == dev)
478 		return vrf_local_xmit(skb, dev, dst);
479 
480 	skb_dst_set(skb, dst);
481 
482 	/* strip the ethernet header added for pass through VRF device */
483 	__skb_pull(skb, skb_network_offset(skb));
484 
485 	ret = vrf_ip6_local_out(net, skb->sk, skb);
486 	if (unlikely(net_xmit_eval(ret)))
487 		dev->stats.tx_errors++;
488 	else
489 		ret = NET_XMIT_SUCCESS;
490 
491 	return ret;
492 err:
493 	vrf_tx_error(dev, skb);
494 	return NET_XMIT_DROP;
495 }
496 #else
497 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
498 					   struct net_device *dev)
499 {
500 	vrf_tx_error(dev, skb);
501 	return NET_XMIT_DROP;
502 }
503 #endif
504 
505 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
506 static int vrf_ip_local_out(struct net *net, struct sock *sk,
507 			    struct sk_buff *skb)
508 {
509 	int err;
510 
511 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
512 		      skb, NULL, skb_dst(skb)->dev, dst_output);
513 	if (likely(err == 1))
514 		err = dst_output(net, sk, skb);
515 
516 	return err;
517 }
518 
519 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
520 					   struct net_device *vrf_dev)
521 {
522 	struct iphdr *ip4h;
523 	int ret = NET_XMIT_DROP;
524 	struct flowi4 fl4;
525 	struct net *net = dev_net(vrf_dev);
526 	struct rtable *rt;
527 
528 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
529 		goto err;
530 
531 	ip4h = ip_hdr(skb);
532 
533 	memset(&fl4, 0, sizeof(fl4));
534 	/* needed to match OIF rule */
535 	fl4.flowi4_oif = vrf_dev->ifindex;
536 	fl4.flowi4_iif = LOOPBACK_IFINDEX;
537 	fl4.flowi4_tos = RT_TOS(ip4h->tos);
538 	fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
539 	fl4.flowi4_proto = ip4h->protocol;
540 	fl4.daddr = ip4h->daddr;
541 	fl4.saddr = ip4h->saddr;
542 
543 	rt = ip_route_output_flow(net, &fl4, NULL);
544 	if (IS_ERR(rt))
545 		goto err;
546 
547 	skb_dst_drop(skb);
548 
549 	/* if dst.dev is loopback or the VRF device again this is locally
550 	 * originated traffic destined to a local address. Short circuit
551 	 * to Rx path
552 	 */
553 	if (rt->dst.dev == vrf_dev)
554 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
555 
556 	skb_dst_set(skb, &rt->dst);
557 
558 	/* strip the ethernet header added for pass through VRF device */
559 	__skb_pull(skb, skb_network_offset(skb));
560 
561 	if (!ip4h->saddr) {
562 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
563 					       RT_SCOPE_LINK);
564 	}
565 
566 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
567 	if (unlikely(net_xmit_eval(ret)))
568 		vrf_dev->stats.tx_errors++;
569 	else
570 		ret = NET_XMIT_SUCCESS;
571 
572 out:
573 	return ret;
574 err:
575 	vrf_tx_error(vrf_dev, skb);
576 	goto out;
577 }
578 
579 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
580 {
581 	switch (skb->protocol) {
582 	case htons(ETH_P_IP):
583 		return vrf_process_v4_outbound(skb, dev);
584 	case htons(ETH_P_IPV6):
585 		return vrf_process_v6_outbound(skb, dev);
586 	default:
587 		vrf_tx_error(dev, skb);
588 		return NET_XMIT_DROP;
589 	}
590 }
591 
592 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
593 {
594 	int len = skb->len;
595 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
596 
597 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
598 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
599 
600 		u64_stats_update_begin(&dstats->syncp);
601 		dstats->tx_pkts++;
602 		dstats->tx_bytes += len;
603 		u64_stats_update_end(&dstats->syncp);
604 	} else {
605 		this_cpu_inc(dev->dstats->tx_drps);
606 	}
607 
608 	return ret;
609 }
610 
611 static int vrf_finish_direct(struct net *net, struct sock *sk,
612 			     struct sk_buff *skb)
613 {
614 	struct net_device *vrf_dev = skb->dev;
615 
616 	if (!list_empty(&vrf_dev->ptype_all) &&
617 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
618 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
619 
620 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
621 		eth_zero_addr(eth->h_dest);
622 		eth->h_proto = skb->protocol;
623 
624 		rcu_read_lock_bh();
625 		dev_queue_xmit_nit(skb, vrf_dev);
626 		rcu_read_unlock_bh();
627 
628 		skb_pull(skb, ETH_HLEN);
629 	}
630 
631 	return 1;
632 }
633 
634 #if IS_ENABLED(CONFIG_IPV6)
635 /* modelled after ip6_finish_output2 */
636 static int vrf_finish_output6(struct net *net, struct sock *sk,
637 			      struct sk_buff *skb)
638 {
639 	struct dst_entry *dst = skb_dst(skb);
640 	struct net_device *dev = dst->dev;
641 	const struct in6_addr *nexthop;
642 	struct neighbour *neigh;
643 	int ret;
644 
645 	nf_reset_ct(skb);
646 
647 	skb->protocol = htons(ETH_P_IPV6);
648 	skb->dev = dev;
649 
650 	rcu_read_lock_bh();
651 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
652 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
653 	if (unlikely(!neigh))
654 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
655 	if (!IS_ERR(neigh)) {
656 		sock_confirm_neigh(skb, neigh);
657 		ret = neigh_output(neigh, skb, false);
658 		rcu_read_unlock_bh();
659 		return ret;
660 	}
661 	rcu_read_unlock_bh();
662 
663 	IP6_INC_STATS(dev_net(dst->dev),
664 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
665 	kfree_skb(skb);
666 	return -EINVAL;
667 }
668 
669 /* modelled after ip6_output */
670 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
671 {
672 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
673 			    net, sk, skb, NULL, skb_dst(skb)->dev,
674 			    vrf_finish_output6,
675 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
676 }
677 
678 /* set dst on skb to send packet to us via dev_xmit path. Allows
679  * packet to go through device based features such as qdisc, netfilter
680  * hooks and packet sockets with skb->dev set to vrf device.
681  */
682 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
683 					    struct sk_buff *skb)
684 {
685 	struct net_vrf *vrf = netdev_priv(vrf_dev);
686 	struct dst_entry *dst = NULL;
687 	struct rt6_info *rt6;
688 
689 	rcu_read_lock();
690 
691 	rt6 = rcu_dereference(vrf->rt6);
692 	if (likely(rt6)) {
693 		dst = &rt6->dst;
694 		dst_hold(dst);
695 	}
696 
697 	rcu_read_unlock();
698 
699 	if (unlikely(!dst)) {
700 		vrf_tx_error(vrf_dev, skb);
701 		return NULL;
702 	}
703 
704 	skb_dst_drop(skb);
705 	skb_dst_set(skb, dst);
706 
707 	return skb;
708 }
709 
710 static int vrf_output6_direct(struct net *net, struct sock *sk,
711 			      struct sk_buff *skb)
712 {
713 	skb->protocol = htons(ETH_P_IPV6);
714 
715 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
716 			    net, sk, skb, NULL, skb->dev,
717 			    vrf_finish_direct,
718 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
719 }
720 
721 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
722 					  struct sock *sk,
723 					  struct sk_buff *skb)
724 {
725 	struct net *net = dev_net(vrf_dev);
726 	int err;
727 
728 	skb->dev = vrf_dev;
729 
730 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
731 		      skb, NULL, vrf_dev, vrf_output6_direct);
732 
733 	if (likely(err == 1))
734 		err = vrf_output6_direct(net, sk, skb);
735 
736 	/* reset skb device */
737 	if (likely(err == 1))
738 		nf_reset_ct(skb);
739 	else
740 		skb = NULL;
741 
742 	return skb;
743 }
744 
745 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
746 				   struct sock *sk,
747 				   struct sk_buff *skb)
748 {
749 	/* don't divert link scope packets */
750 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
751 		return skb;
752 
753 	if (qdisc_tx_is_default(vrf_dev) ||
754 	    IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
755 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
756 
757 	return vrf_ip6_out_redirect(vrf_dev, skb);
758 }
759 
760 /* holding rtnl */
761 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
762 {
763 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
764 	struct net *net = dev_net(dev);
765 	struct dst_entry *dst;
766 
767 	RCU_INIT_POINTER(vrf->rt6, NULL);
768 	synchronize_rcu();
769 
770 	/* move dev in dst's to loopback so this VRF device can be deleted
771 	 * - based on dst_ifdown
772 	 */
773 	if (rt6) {
774 		dst = &rt6->dst;
775 		dev_put(dst->dev);
776 		dst->dev = net->loopback_dev;
777 		dev_hold(dst->dev);
778 		dst_release(dst);
779 	}
780 }
781 
782 static int vrf_rt6_create(struct net_device *dev)
783 {
784 	int flags = DST_NOPOLICY | DST_NOXFRM;
785 	struct net_vrf *vrf = netdev_priv(dev);
786 	struct net *net = dev_net(dev);
787 	struct rt6_info *rt6;
788 	int rc = -ENOMEM;
789 
790 	/* IPv6 can be CONFIG enabled and then disabled runtime */
791 	if (!ipv6_mod_enabled())
792 		return 0;
793 
794 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
795 	if (!vrf->fib6_table)
796 		goto out;
797 
798 	/* create a dst for routing packets out a VRF device */
799 	rt6 = ip6_dst_alloc(net, dev, flags);
800 	if (!rt6)
801 		goto out;
802 
803 	rt6->dst.output	= vrf_output6;
804 
805 	rcu_assign_pointer(vrf->rt6, rt6);
806 
807 	rc = 0;
808 out:
809 	return rc;
810 }
811 #else
812 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
813 				   struct sock *sk,
814 				   struct sk_buff *skb)
815 {
816 	return skb;
817 }
818 
819 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
820 {
821 }
822 
823 static int vrf_rt6_create(struct net_device *dev)
824 {
825 	return 0;
826 }
827 #endif
828 
829 /* modelled after ip_finish_output2 */
830 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
831 {
832 	struct dst_entry *dst = skb_dst(skb);
833 	struct rtable *rt = (struct rtable *)dst;
834 	struct net_device *dev = dst->dev;
835 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
836 	struct neighbour *neigh;
837 	bool is_v6gw = false;
838 	int ret = -EINVAL;
839 
840 	nf_reset_ct(skb);
841 
842 	/* Be paranoid, rather than too clever. */
843 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
844 		struct sk_buff *skb2;
845 
846 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
847 		if (!skb2) {
848 			ret = -ENOMEM;
849 			goto err;
850 		}
851 		if (skb->sk)
852 			skb_set_owner_w(skb2, skb->sk);
853 
854 		consume_skb(skb);
855 		skb = skb2;
856 	}
857 
858 	rcu_read_lock_bh();
859 
860 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
861 	if (!IS_ERR(neigh)) {
862 		sock_confirm_neigh(skb, neigh);
863 		/* if crossing protocols, can not use the cached header */
864 		ret = neigh_output(neigh, skb, is_v6gw);
865 		rcu_read_unlock_bh();
866 		return ret;
867 	}
868 
869 	rcu_read_unlock_bh();
870 err:
871 	vrf_tx_error(skb->dev, skb);
872 	return ret;
873 }
874 
875 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
876 {
877 	struct net_device *dev = skb_dst(skb)->dev;
878 
879 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
880 
881 	skb->dev = dev;
882 	skb->protocol = htons(ETH_P_IP);
883 
884 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
885 			    net, sk, skb, NULL, dev,
886 			    vrf_finish_output,
887 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
888 }
889 
890 /* set dst on skb to send packet to us via dev_xmit path. Allows
891  * packet to go through device based features such as qdisc, netfilter
892  * hooks and packet sockets with skb->dev set to vrf device.
893  */
894 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
895 					   struct sk_buff *skb)
896 {
897 	struct net_vrf *vrf = netdev_priv(vrf_dev);
898 	struct dst_entry *dst = NULL;
899 	struct rtable *rth;
900 
901 	rcu_read_lock();
902 
903 	rth = rcu_dereference(vrf->rth);
904 	if (likely(rth)) {
905 		dst = &rth->dst;
906 		dst_hold(dst);
907 	}
908 
909 	rcu_read_unlock();
910 
911 	if (unlikely(!dst)) {
912 		vrf_tx_error(vrf_dev, skb);
913 		return NULL;
914 	}
915 
916 	skb_dst_drop(skb);
917 	skb_dst_set(skb, dst);
918 
919 	return skb;
920 }
921 
922 static int vrf_output_direct(struct net *net, struct sock *sk,
923 			     struct sk_buff *skb)
924 {
925 	skb->protocol = htons(ETH_P_IP);
926 
927 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
928 			    net, sk, skb, NULL, skb->dev,
929 			    vrf_finish_direct,
930 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
931 }
932 
933 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
934 					 struct sock *sk,
935 					 struct sk_buff *skb)
936 {
937 	struct net *net = dev_net(vrf_dev);
938 	int err;
939 
940 	skb->dev = vrf_dev;
941 
942 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
943 		      skb, NULL, vrf_dev, vrf_output_direct);
944 
945 	if (likely(err == 1))
946 		err = vrf_output_direct(net, sk, skb);
947 
948 	/* reset skb device */
949 	if (likely(err == 1))
950 		nf_reset_ct(skb);
951 	else
952 		skb = NULL;
953 
954 	return skb;
955 }
956 
957 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
958 				  struct sock *sk,
959 				  struct sk_buff *skb)
960 {
961 	/* don't divert multicast or local broadcast */
962 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
963 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
964 		return skb;
965 
966 	if (qdisc_tx_is_default(vrf_dev) ||
967 	    IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
968 		return vrf_ip_out_direct(vrf_dev, sk, skb);
969 
970 	return vrf_ip_out_redirect(vrf_dev, skb);
971 }
972 
973 /* called with rcu lock held */
974 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
975 				  struct sock *sk,
976 				  struct sk_buff *skb,
977 				  u16 proto)
978 {
979 	switch (proto) {
980 	case AF_INET:
981 		return vrf_ip_out(vrf_dev, sk, skb);
982 	case AF_INET6:
983 		return vrf_ip6_out(vrf_dev, sk, skb);
984 	}
985 
986 	return skb;
987 }
988 
989 /* holding rtnl */
990 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
991 {
992 	struct rtable *rth = rtnl_dereference(vrf->rth);
993 	struct net *net = dev_net(dev);
994 	struct dst_entry *dst;
995 
996 	RCU_INIT_POINTER(vrf->rth, NULL);
997 	synchronize_rcu();
998 
999 	/* move dev in dst's to loopback so this VRF device can be deleted
1000 	 * - based on dst_ifdown
1001 	 */
1002 	if (rth) {
1003 		dst = &rth->dst;
1004 		dev_put(dst->dev);
1005 		dst->dev = net->loopback_dev;
1006 		dev_hold(dst->dev);
1007 		dst_release(dst);
1008 	}
1009 }
1010 
1011 static int vrf_rtable_create(struct net_device *dev)
1012 {
1013 	struct net_vrf *vrf = netdev_priv(dev);
1014 	struct rtable *rth;
1015 
1016 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
1017 		return -ENOMEM;
1018 
1019 	/* create a dst for routing packets out through a VRF device */
1020 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
1021 	if (!rth)
1022 		return -ENOMEM;
1023 
1024 	rth->dst.output	= vrf_output;
1025 
1026 	rcu_assign_pointer(vrf->rth, rth);
1027 
1028 	return 0;
1029 }
1030 
1031 /**************************** device handling ********************/
1032 
1033 /* cycle interface to flush neighbor cache and move routes across tables */
1034 static void cycle_netdev(struct net_device *dev,
1035 			 struct netlink_ext_ack *extack)
1036 {
1037 	unsigned int flags = dev->flags;
1038 	int ret;
1039 
1040 	if (!netif_running(dev))
1041 		return;
1042 
1043 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1044 	if (ret >= 0)
1045 		ret = dev_change_flags(dev, flags, extack);
1046 
1047 	if (ret < 0) {
1048 		netdev_err(dev,
1049 			   "Failed to cycle device %s; route tables might be wrong!\n",
1050 			   dev->name);
1051 	}
1052 }
1053 
1054 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1055 			    struct netlink_ext_ack *extack)
1056 {
1057 	int ret;
1058 
1059 	/* do not allow loopback device to be enslaved to a VRF.
1060 	 * The vrf device acts as the loopback for the vrf.
1061 	 */
1062 	if (port_dev == dev_net(dev)->loopback_dev) {
1063 		NL_SET_ERR_MSG(extack,
1064 			       "Can not enslave loopback device to a VRF");
1065 		return -EOPNOTSUPP;
1066 	}
1067 
1068 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1069 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1070 	if (ret < 0)
1071 		goto err;
1072 
1073 	cycle_netdev(port_dev, extack);
1074 
1075 	return 0;
1076 
1077 err:
1078 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1079 	return ret;
1080 }
1081 
1082 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1083 			 struct netlink_ext_ack *extack)
1084 {
1085 	if (netif_is_l3_master(port_dev)) {
1086 		NL_SET_ERR_MSG(extack,
1087 			       "Can not enslave an L3 master device to a VRF");
1088 		return -EINVAL;
1089 	}
1090 
1091 	if (netif_is_l3_slave(port_dev))
1092 		return -EINVAL;
1093 
1094 	return do_vrf_add_slave(dev, port_dev, extack);
1095 }
1096 
1097 /* inverse of do_vrf_add_slave */
1098 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1099 {
1100 	netdev_upper_dev_unlink(port_dev, dev);
1101 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1102 
1103 	cycle_netdev(port_dev, NULL);
1104 
1105 	return 0;
1106 }
1107 
1108 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1109 {
1110 	return do_vrf_del_slave(dev, port_dev);
1111 }
1112 
1113 static void vrf_dev_uninit(struct net_device *dev)
1114 {
1115 	struct net_vrf *vrf = netdev_priv(dev);
1116 
1117 	vrf_rtable_release(dev, vrf);
1118 	vrf_rt6_release(dev, vrf);
1119 
1120 	free_percpu(dev->dstats);
1121 	dev->dstats = NULL;
1122 }
1123 
1124 static int vrf_dev_init(struct net_device *dev)
1125 {
1126 	struct net_vrf *vrf = netdev_priv(dev);
1127 
1128 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1129 	if (!dev->dstats)
1130 		goto out_nomem;
1131 
1132 	/* create the default dst which points back to us */
1133 	if (vrf_rtable_create(dev) != 0)
1134 		goto out_stats;
1135 
1136 	if (vrf_rt6_create(dev) != 0)
1137 		goto out_rth;
1138 
1139 	dev->flags = IFF_MASTER | IFF_NOARP;
1140 
1141 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
1142 	dev->mtu = 64 * 1024;
1143 
1144 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
1145 	dev->operstate = IF_OPER_UP;
1146 	netdev_lockdep_set_classes(dev);
1147 	return 0;
1148 
1149 out_rth:
1150 	vrf_rtable_release(dev, vrf);
1151 out_stats:
1152 	free_percpu(dev->dstats);
1153 	dev->dstats = NULL;
1154 out_nomem:
1155 	return -ENOMEM;
1156 }
1157 
1158 static const struct net_device_ops vrf_netdev_ops = {
1159 	.ndo_init		= vrf_dev_init,
1160 	.ndo_uninit		= vrf_dev_uninit,
1161 	.ndo_start_xmit		= vrf_xmit,
1162 	.ndo_set_mac_address	= eth_mac_addr,
1163 	.ndo_get_stats64	= vrf_get_stats64,
1164 	.ndo_add_slave		= vrf_add_slave,
1165 	.ndo_del_slave		= vrf_del_slave,
1166 };
1167 
1168 static u32 vrf_fib_table(const struct net_device *dev)
1169 {
1170 	struct net_vrf *vrf = netdev_priv(dev);
1171 
1172 	return vrf->tb_id;
1173 }
1174 
1175 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1176 {
1177 	kfree_skb(skb);
1178 	return 0;
1179 }
1180 
1181 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1182 				      struct sk_buff *skb,
1183 				      struct net_device *dev)
1184 {
1185 	struct net *net = dev_net(dev);
1186 
1187 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1188 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
1189 
1190 	return skb;
1191 }
1192 
1193 #if IS_ENABLED(CONFIG_IPV6)
1194 /* neighbor handling is done with actual device; do not want
1195  * to flip skb->dev for those ndisc packets. This really fails
1196  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1197  * a start.
1198  */
1199 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1200 {
1201 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1202 	bool rc = false;
1203 
1204 	if (iph->nexthdr == NEXTHDR_ICMP) {
1205 		const struct icmp6hdr *icmph;
1206 		struct icmp6hdr _icmph;
1207 
1208 		icmph = skb_header_pointer(skb, sizeof(*iph),
1209 					   sizeof(_icmph), &_icmph);
1210 		if (!icmph)
1211 			goto out;
1212 
1213 		switch (icmph->icmp6_type) {
1214 		case NDISC_ROUTER_SOLICITATION:
1215 		case NDISC_ROUTER_ADVERTISEMENT:
1216 		case NDISC_NEIGHBOUR_SOLICITATION:
1217 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
1218 		case NDISC_REDIRECT:
1219 			rc = true;
1220 			break;
1221 		}
1222 	}
1223 
1224 out:
1225 	return rc;
1226 }
1227 
1228 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1229 					     const struct net_device *dev,
1230 					     struct flowi6 *fl6,
1231 					     int ifindex,
1232 					     const struct sk_buff *skb,
1233 					     int flags)
1234 {
1235 	struct net_vrf *vrf = netdev_priv(dev);
1236 
1237 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1238 }
1239 
1240 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1241 			      int ifindex)
1242 {
1243 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1244 	struct flowi6 fl6 = {
1245 		.flowi6_iif     = ifindex,
1246 		.flowi6_mark    = skb->mark,
1247 		.flowi6_proto   = iph->nexthdr,
1248 		.daddr          = iph->daddr,
1249 		.saddr          = iph->saddr,
1250 		.flowlabel      = ip6_flowinfo(iph),
1251 	};
1252 	struct net *net = dev_net(vrf_dev);
1253 	struct rt6_info *rt6;
1254 
1255 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1256 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1257 	if (unlikely(!rt6))
1258 		return;
1259 
1260 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1261 		return;
1262 
1263 	skb_dst_set(skb, &rt6->dst);
1264 }
1265 
1266 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1267 				   struct sk_buff *skb)
1268 {
1269 	int orig_iif = skb->skb_iif;
1270 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1271 	bool is_ndisc = ipv6_ndisc_frame(skb);
1272 
1273 	/* loopback, multicast & non-ND link-local traffic; do not push through
1274 	 * packet taps again. Reset pkt_type for upper layers to process skb
1275 	 */
1276 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1277 		skb->dev = vrf_dev;
1278 		skb->skb_iif = vrf_dev->ifindex;
1279 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1280 		if (skb->pkt_type == PACKET_LOOPBACK)
1281 			skb->pkt_type = PACKET_HOST;
1282 		goto out;
1283 	}
1284 
1285 	/* if packet is NDISC then keep the ingress interface */
1286 	if (!is_ndisc) {
1287 		vrf_rx_stats(vrf_dev, skb->len);
1288 		skb->dev = vrf_dev;
1289 		skb->skb_iif = vrf_dev->ifindex;
1290 
1291 		if (!list_empty(&vrf_dev->ptype_all)) {
1292 			skb_push(skb, skb->mac_len);
1293 			dev_queue_xmit_nit(skb, vrf_dev);
1294 			skb_pull(skb, skb->mac_len);
1295 		}
1296 
1297 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1298 	}
1299 
1300 	if (need_strict)
1301 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1302 
1303 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1304 out:
1305 	return skb;
1306 }
1307 
1308 #else
1309 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1310 				   struct sk_buff *skb)
1311 {
1312 	return skb;
1313 }
1314 #endif
1315 
1316 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1317 				  struct sk_buff *skb)
1318 {
1319 	skb->dev = vrf_dev;
1320 	skb->skb_iif = vrf_dev->ifindex;
1321 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1322 
1323 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1324 		goto out;
1325 
1326 	/* loopback traffic; do not push through packet taps again.
1327 	 * Reset pkt_type for upper layers to process skb
1328 	 */
1329 	if (skb->pkt_type == PACKET_LOOPBACK) {
1330 		skb->pkt_type = PACKET_HOST;
1331 		goto out;
1332 	}
1333 
1334 	vrf_rx_stats(vrf_dev, skb->len);
1335 
1336 	if (!list_empty(&vrf_dev->ptype_all)) {
1337 		skb_push(skb, skb->mac_len);
1338 		dev_queue_xmit_nit(skb, vrf_dev);
1339 		skb_pull(skb, skb->mac_len);
1340 	}
1341 
1342 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1343 out:
1344 	return skb;
1345 }
1346 
1347 /* called with rcu lock held */
1348 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1349 				  struct sk_buff *skb,
1350 				  u16 proto)
1351 {
1352 	switch (proto) {
1353 	case AF_INET:
1354 		return vrf_ip_rcv(vrf_dev, skb);
1355 	case AF_INET6:
1356 		return vrf_ip6_rcv(vrf_dev, skb);
1357 	}
1358 
1359 	return skb;
1360 }
1361 
1362 #if IS_ENABLED(CONFIG_IPV6)
1363 /* send to link-local or multicast address via interface enslaved to
1364  * VRF device. Force lookup to VRF table without changing flow struct
1365  * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1366  * is taken on the dst by this function.
1367  */
1368 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1369 					      struct flowi6 *fl6)
1370 {
1371 	struct net *net = dev_net(dev);
1372 	int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1373 	struct dst_entry *dst = NULL;
1374 	struct rt6_info *rt;
1375 
1376 	/* VRF device does not have a link-local address and
1377 	 * sending packets to link-local or mcast addresses over
1378 	 * a VRF device does not make sense
1379 	 */
1380 	if (fl6->flowi6_oif == dev->ifindex) {
1381 		dst = &net->ipv6.ip6_null_entry->dst;
1382 		return dst;
1383 	}
1384 
1385 	if (!ipv6_addr_any(&fl6->saddr))
1386 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1387 
1388 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1389 	if (rt)
1390 		dst = &rt->dst;
1391 
1392 	return dst;
1393 }
1394 #endif
1395 
1396 static const struct l3mdev_ops vrf_l3mdev_ops = {
1397 	.l3mdev_fib_table	= vrf_fib_table,
1398 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1399 	.l3mdev_l3_out		= vrf_l3_out,
1400 #if IS_ENABLED(CONFIG_IPV6)
1401 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1402 #endif
1403 };
1404 
1405 static void vrf_get_drvinfo(struct net_device *dev,
1406 			    struct ethtool_drvinfo *info)
1407 {
1408 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1409 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1410 }
1411 
1412 static const struct ethtool_ops vrf_ethtool_ops = {
1413 	.get_drvinfo	= vrf_get_drvinfo,
1414 };
1415 
1416 static inline size_t vrf_fib_rule_nl_size(void)
1417 {
1418 	size_t sz;
1419 
1420 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1421 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1422 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1423 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1424 
1425 	return sz;
1426 }
1427 
1428 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1429 {
1430 	struct fib_rule_hdr *frh;
1431 	struct nlmsghdr *nlh;
1432 	struct sk_buff *skb;
1433 	int err;
1434 
1435 	if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1436 	    !ipv6_mod_enabled())
1437 		return 0;
1438 
1439 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1440 	if (!skb)
1441 		return -ENOMEM;
1442 
1443 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1444 	if (!nlh)
1445 		goto nla_put_failure;
1446 
1447 	/* rule only needs to appear once */
1448 	nlh->nlmsg_flags |= NLM_F_EXCL;
1449 
1450 	frh = nlmsg_data(nlh);
1451 	memset(frh, 0, sizeof(*frh));
1452 	frh->family = family;
1453 	frh->action = FR_ACT_TO_TBL;
1454 
1455 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1456 		goto nla_put_failure;
1457 
1458 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1459 		goto nla_put_failure;
1460 
1461 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1462 		goto nla_put_failure;
1463 
1464 	nlmsg_end(skb, nlh);
1465 
1466 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1467 	skb->sk = dev_net(dev)->rtnl;
1468 	if (add_it) {
1469 		err = fib_nl_newrule(skb, nlh, NULL);
1470 		if (err == -EEXIST)
1471 			err = 0;
1472 	} else {
1473 		err = fib_nl_delrule(skb, nlh, NULL);
1474 		if (err == -ENOENT)
1475 			err = 0;
1476 	}
1477 	nlmsg_free(skb);
1478 
1479 	return err;
1480 
1481 nla_put_failure:
1482 	nlmsg_free(skb);
1483 
1484 	return -EMSGSIZE;
1485 }
1486 
1487 static int vrf_add_fib_rules(const struct net_device *dev)
1488 {
1489 	int err;
1490 
1491 	err = vrf_fib_rule(dev, AF_INET,  true);
1492 	if (err < 0)
1493 		goto out_err;
1494 
1495 	err = vrf_fib_rule(dev, AF_INET6, true);
1496 	if (err < 0)
1497 		goto ipv6_err;
1498 
1499 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1500 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1501 	if (err < 0)
1502 		goto ipmr_err;
1503 #endif
1504 
1505 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1506 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1507 	if (err < 0)
1508 		goto ip6mr_err;
1509 #endif
1510 
1511 	return 0;
1512 
1513 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1514 ip6mr_err:
1515 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1516 #endif
1517 
1518 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1519 ipmr_err:
1520 	vrf_fib_rule(dev, AF_INET6,  false);
1521 #endif
1522 
1523 ipv6_err:
1524 	vrf_fib_rule(dev, AF_INET,  false);
1525 
1526 out_err:
1527 	netdev_err(dev, "Failed to add FIB rules.\n");
1528 	return err;
1529 }
1530 
1531 static void vrf_setup(struct net_device *dev)
1532 {
1533 	ether_setup(dev);
1534 
1535 	/* Initialize the device structure. */
1536 	dev->netdev_ops = &vrf_netdev_ops;
1537 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1538 	dev->ethtool_ops = &vrf_ethtool_ops;
1539 	dev->needs_free_netdev = true;
1540 
1541 	/* Fill in device structure with ethernet-generic values. */
1542 	eth_hw_addr_random(dev);
1543 
1544 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1545 	dev->features |= NETIF_F_LLTX;
1546 
1547 	/* don't allow vrf devices to change network namespaces. */
1548 	dev->features |= NETIF_F_NETNS_LOCAL;
1549 
1550 	/* does not make sense for a VLAN to be added to a vrf device */
1551 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1552 
1553 	/* enable offload features */
1554 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1555 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1556 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1557 
1558 	dev->hw_features = dev->features;
1559 	dev->hw_enc_features = dev->features;
1560 
1561 	/* default to no qdisc; user can add if desired */
1562 	dev->priv_flags |= IFF_NO_QUEUE;
1563 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1564 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1565 
1566 	/* VRF devices do not care about MTU, but if the MTU is set
1567 	 * too low then the ipv4 and ipv6 protocols are disabled
1568 	 * which breaks networking.
1569 	 */
1570 	dev->min_mtu = IPV6_MIN_MTU;
1571 	dev->max_mtu = ETH_MAX_MTU;
1572 }
1573 
1574 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1575 			struct netlink_ext_ack *extack)
1576 {
1577 	if (tb[IFLA_ADDRESS]) {
1578 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1579 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1580 			return -EINVAL;
1581 		}
1582 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1583 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1584 			return -EADDRNOTAVAIL;
1585 		}
1586 	}
1587 	return 0;
1588 }
1589 
1590 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1591 {
1592 	struct net_device *port_dev;
1593 	struct list_head *iter;
1594 
1595 	netdev_for_each_lower_dev(dev, port_dev, iter)
1596 		vrf_del_slave(dev, port_dev);
1597 
1598 	vrf_map_unregister_dev(dev);
1599 
1600 	unregister_netdevice_queue(dev, head);
1601 }
1602 
1603 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1604 		       struct nlattr *tb[], struct nlattr *data[],
1605 		       struct netlink_ext_ack *extack)
1606 {
1607 	struct net_vrf *vrf = netdev_priv(dev);
1608 	struct netns_vrf *nn_vrf;
1609 	bool *add_fib_rules;
1610 	struct net *net;
1611 	int err;
1612 
1613 	if (!data || !data[IFLA_VRF_TABLE]) {
1614 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1615 		return -EINVAL;
1616 	}
1617 
1618 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1619 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1620 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1621 				    "Invalid VRF table id");
1622 		return -EINVAL;
1623 	}
1624 
1625 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1626 
1627 	err = register_netdevice(dev);
1628 	if (err)
1629 		goto out;
1630 
1631 	/* mapping between table_id and vrf;
1632 	 * note: such binding could not be done in the dev init function
1633 	 * because dev->ifindex id is not available yet.
1634 	 */
1635 	vrf->ifindex = dev->ifindex;
1636 
1637 	err = vrf_map_register_dev(dev, extack);
1638 	if (err) {
1639 		unregister_netdevice(dev);
1640 		goto out;
1641 	}
1642 
1643 	net = dev_net(dev);
1644 	nn_vrf = net_generic(net, vrf_net_id);
1645 
1646 	add_fib_rules = &nn_vrf->add_fib_rules;
1647 	if (*add_fib_rules) {
1648 		err = vrf_add_fib_rules(dev);
1649 		if (err) {
1650 			vrf_map_unregister_dev(dev);
1651 			unregister_netdevice(dev);
1652 			goto out;
1653 		}
1654 		*add_fib_rules = false;
1655 	}
1656 
1657 out:
1658 	return err;
1659 }
1660 
1661 static size_t vrf_nl_getsize(const struct net_device *dev)
1662 {
1663 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1664 }
1665 
1666 static int vrf_fillinfo(struct sk_buff *skb,
1667 			const struct net_device *dev)
1668 {
1669 	struct net_vrf *vrf = netdev_priv(dev);
1670 
1671 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1672 }
1673 
1674 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1675 				 const struct net_device *slave_dev)
1676 {
1677 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1678 }
1679 
1680 static int vrf_fill_slave_info(struct sk_buff *skb,
1681 			       const struct net_device *vrf_dev,
1682 			       const struct net_device *slave_dev)
1683 {
1684 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1685 
1686 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1687 		return -EMSGSIZE;
1688 
1689 	return 0;
1690 }
1691 
1692 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1693 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1694 };
1695 
1696 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1697 	.kind		= DRV_NAME,
1698 	.priv_size	= sizeof(struct net_vrf),
1699 
1700 	.get_size	= vrf_nl_getsize,
1701 	.policy		= vrf_nl_policy,
1702 	.validate	= vrf_validate,
1703 	.fill_info	= vrf_fillinfo,
1704 
1705 	.get_slave_size  = vrf_get_slave_size,
1706 	.fill_slave_info = vrf_fill_slave_info,
1707 
1708 	.newlink	= vrf_newlink,
1709 	.dellink	= vrf_dellink,
1710 	.setup		= vrf_setup,
1711 	.maxtype	= IFLA_VRF_MAX,
1712 };
1713 
1714 static int vrf_device_event(struct notifier_block *unused,
1715 			    unsigned long event, void *ptr)
1716 {
1717 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1718 
1719 	/* only care about unregister events to drop slave references */
1720 	if (event == NETDEV_UNREGISTER) {
1721 		struct net_device *vrf_dev;
1722 
1723 		if (!netif_is_l3_slave(dev))
1724 			goto out;
1725 
1726 		vrf_dev = netdev_master_upper_dev_get(dev);
1727 		vrf_del_slave(vrf_dev, dev);
1728 	}
1729 out:
1730 	return NOTIFY_DONE;
1731 }
1732 
1733 static struct notifier_block vrf_notifier_block __read_mostly = {
1734 	.notifier_call = vrf_device_event,
1735 };
1736 
1737 static int vrf_map_init(struct vrf_map *vmap)
1738 {
1739 	spin_lock_init(&vmap->vmap_lock);
1740 	hash_init(vmap->ht);
1741 
1742 	vmap->strict_mode = false;
1743 
1744 	return 0;
1745 }
1746 
1747 #ifdef CONFIG_SYSCTL
1748 static bool vrf_strict_mode(struct vrf_map *vmap)
1749 {
1750 	bool strict_mode;
1751 
1752 	vrf_map_lock(vmap);
1753 	strict_mode = vmap->strict_mode;
1754 	vrf_map_unlock(vmap);
1755 
1756 	return strict_mode;
1757 }
1758 
1759 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1760 {
1761 	bool *cur_mode;
1762 	int res = 0;
1763 
1764 	vrf_map_lock(vmap);
1765 
1766 	cur_mode = &vmap->strict_mode;
1767 	if (*cur_mode == new_mode)
1768 		goto unlock;
1769 
1770 	if (*cur_mode) {
1771 		/* disable strict mode */
1772 		*cur_mode = false;
1773 	} else {
1774 		if (vmap->shared_tables) {
1775 			/* we cannot allow strict_mode because there are some
1776 			 * vrfs that share one or more tables.
1777 			 */
1778 			res = -EBUSY;
1779 			goto unlock;
1780 		}
1781 
1782 		/* no tables are shared among vrfs, so we can go back
1783 		 * to 1:1 association between a vrf with its table.
1784 		 */
1785 		*cur_mode = true;
1786 	}
1787 
1788 unlock:
1789 	vrf_map_unlock(vmap);
1790 
1791 	return res;
1792 }
1793 
1794 static int vrf_shared_table_handler(struct ctl_table *table, int write,
1795 				    void *buffer, size_t *lenp, loff_t *ppos)
1796 {
1797 	struct net *net = (struct net *)table->extra1;
1798 	struct vrf_map *vmap = netns_vrf_map(net);
1799 	int proc_strict_mode = 0;
1800 	struct ctl_table tmp = {
1801 		.procname	= table->procname,
1802 		.data		= &proc_strict_mode,
1803 		.maxlen		= sizeof(int),
1804 		.mode		= table->mode,
1805 		.extra1		= SYSCTL_ZERO,
1806 		.extra2		= SYSCTL_ONE,
1807 	};
1808 	int ret;
1809 
1810 	if (!write)
1811 		proc_strict_mode = vrf_strict_mode(vmap);
1812 
1813 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1814 
1815 	if (write && ret == 0)
1816 		ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1817 
1818 	return ret;
1819 }
1820 
1821 static const struct ctl_table vrf_table[] = {
1822 	{
1823 		.procname	= "strict_mode",
1824 		.data		= NULL,
1825 		.maxlen		= sizeof(int),
1826 		.mode		= 0644,
1827 		.proc_handler	= vrf_shared_table_handler,
1828 		/* set by the vrf_netns_init */
1829 		.extra1		= NULL,
1830 	},
1831 	{ },
1832 };
1833 
1834 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1835 {
1836 	struct ctl_table *table;
1837 
1838 	table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1839 	if (!table)
1840 		return -ENOMEM;
1841 
1842 	/* init the extra1 parameter with the reference to current netns */
1843 	table[0].extra1 = net;
1844 
1845 	nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1846 	if (!nn_vrf->ctl_hdr) {
1847 		kfree(table);
1848 		return -ENOMEM;
1849 	}
1850 
1851 	return 0;
1852 }
1853 
1854 static void vrf_netns_exit_sysctl(struct net *net)
1855 {
1856 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1857 	struct ctl_table *table;
1858 
1859 	table = nn_vrf->ctl_hdr->ctl_table_arg;
1860 	unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1861 	kfree(table);
1862 }
1863 #else
1864 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1865 {
1866 	return 0;
1867 }
1868 
1869 static void vrf_netns_exit_sysctl(struct net *net)
1870 {
1871 }
1872 #endif
1873 
1874 /* Initialize per network namespace state */
1875 static int __net_init vrf_netns_init(struct net *net)
1876 {
1877 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1878 
1879 	nn_vrf->add_fib_rules = true;
1880 	vrf_map_init(&nn_vrf->vmap);
1881 
1882 	return vrf_netns_init_sysctl(net, nn_vrf);
1883 }
1884 
1885 static void __net_exit vrf_netns_exit(struct net *net)
1886 {
1887 	vrf_netns_exit_sysctl(net);
1888 }
1889 
1890 static struct pernet_operations vrf_net_ops __net_initdata = {
1891 	.init = vrf_netns_init,
1892 	.exit = vrf_netns_exit,
1893 	.id   = &vrf_net_id,
1894 	.size = sizeof(struct netns_vrf),
1895 };
1896 
1897 static int __init vrf_init_module(void)
1898 {
1899 	int rc;
1900 
1901 	register_netdevice_notifier(&vrf_notifier_block);
1902 
1903 	rc = register_pernet_subsys(&vrf_net_ops);
1904 	if (rc < 0)
1905 		goto error;
1906 
1907 	rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
1908 					  vrf_ifindex_lookup_by_table_id);
1909 	if (rc < 0)
1910 		goto unreg_pernet;
1911 
1912 	rc = rtnl_link_register(&vrf_link_ops);
1913 	if (rc < 0)
1914 		goto table_lookup_unreg;
1915 
1916 	return 0;
1917 
1918 table_lookup_unreg:
1919 	l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
1920 				       vrf_ifindex_lookup_by_table_id);
1921 
1922 unreg_pernet:
1923 	unregister_pernet_subsys(&vrf_net_ops);
1924 
1925 error:
1926 	unregister_netdevice_notifier(&vrf_notifier_block);
1927 	return rc;
1928 }
1929 
1930 module_init(vrf_init_module);
1931 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1932 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1933 MODULE_LICENSE("GPL");
1934 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1935 MODULE_VERSION(DRV_VERSION);
1936