xref: /openbmc/linux/drivers/net/vrf.c (revision 19e528dc)
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 static bool vrf_strict_mode(struct vrf_map *vmap)
264 {
265 	bool strict_mode;
266 
267 	vrf_map_lock(vmap);
268 	strict_mode = vmap->strict_mode;
269 	vrf_map_unlock(vmap);
270 
271 	return strict_mode;
272 }
273 
274 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
275 {
276 	bool *cur_mode;
277 	int res = 0;
278 
279 	vrf_map_lock(vmap);
280 
281 	cur_mode = &vmap->strict_mode;
282 	if (*cur_mode == new_mode)
283 		goto unlock;
284 
285 	if (*cur_mode) {
286 		/* disable strict mode */
287 		*cur_mode = false;
288 	} else {
289 		if (vmap->shared_tables) {
290 			/* we cannot allow strict_mode because there are some
291 			 * vrfs that share one or more tables.
292 			 */
293 			res = -EBUSY;
294 			goto unlock;
295 		}
296 
297 		/* no tables are shared among vrfs, so we can go back
298 		 * to 1:1 association between a vrf with its table.
299 		 */
300 		*cur_mode = true;
301 	}
302 
303 unlock:
304 	vrf_map_unlock(vmap);
305 
306 	return res;
307 }
308 
309 /* called with rtnl lock held */
310 static int
311 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
312 {
313 	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
314 	struct net_vrf *vrf = netdev_priv(dev);
315 	struct vrf_map_elem *new_me, *me;
316 	u32 table_id = vrf->tb_id;
317 	bool free_new_me = false;
318 	int users;
319 	int res;
320 
321 	/* we pre-allocate elements used in the spin-locked section (so that we
322 	 * keep the spinlock as short as possibile).
323 	 */
324 	new_me = vrf_map_elem_alloc(GFP_KERNEL);
325 	if (!new_me)
326 		return -ENOMEM;
327 
328 	vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
329 
330 	vrf_map_lock(vmap);
331 
332 	me = vrf_map_lookup_elem(vmap, table_id);
333 	if (!me) {
334 		me = new_me;
335 		vrf_map_add_elem(vmap, me);
336 		goto link_vrf;
337 	}
338 
339 	/* we already have an entry in the vrf_map, so it means there is (at
340 	 * least) a vrf registered on the specific table.
341 	 */
342 	free_new_me = true;
343 	if (vmap->strict_mode) {
344 		/* vrfs cannot share the same table */
345 		NL_SET_ERR_MSG(extack, "Table is used by another VRF");
346 		res = -EBUSY;
347 		goto unlock;
348 	}
349 
350 link_vrf:
351 	users = ++me->users;
352 	if (users == 2)
353 		++vmap->shared_tables;
354 
355 	list_add(&vrf->me_list, &me->vrf_list);
356 
357 	res = 0;
358 
359 unlock:
360 	vrf_map_unlock(vmap);
361 
362 	/* clean-up, if needed */
363 	if (free_new_me)
364 		vrf_map_elem_free(new_me);
365 
366 	return res;
367 }
368 
369 /* called with rtnl lock held */
370 static void vrf_map_unregister_dev(struct net_device *dev)
371 {
372 	struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
373 	struct net_vrf *vrf = netdev_priv(dev);
374 	u32 table_id = vrf->tb_id;
375 	struct vrf_map_elem *me;
376 	int users;
377 
378 	vrf_map_lock(vmap);
379 
380 	me = vrf_map_lookup_elem(vmap, table_id);
381 	if (!me)
382 		goto unlock;
383 
384 	list_del(&vrf->me_list);
385 
386 	users = --me->users;
387 	if (users == 1) {
388 		--vmap->shared_tables;
389 	} else if (users == 0) {
390 		vrf_map_del_elem(me);
391 
392 		/* no one will refer to this element anymore */
393 		vrf_map_elem_free(me);
394 	}
395 
396 unlock:
397 	vrf_map_unlock(vmap);
398 }
399 
400 /* return the vrf device index associated with the table_id */
401 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
402 {
403 	struct vrf_map *vmap = netns_vrf_map(net);
404 	struct vrf_map_elem *me;
405 	int ifindex;
406 
407 	vrf_map_lock(vmap);
408 
409 	if (!vmap->strict_mode) {
410 		ifindex = -EPERM;
411 		goto unlock;
412 	}
413 
414 	me = vrf_map_lookup_elem(vmap, table_id);
415 	if (!me) {
416 		ifindex = -ENODEV;
417 		goto unlock;
418 	}
419 
420 	ifindex = vrf_map_elem_get_vrf_ifindex(me);
421 
422 unlock:
423 	vrf_map_unlock(vmap);
424 
425 	return ifindex;
426 }
427 
428 /* by default VRF devices do not have a qdisc and are expected
429  * to be created with only a single queue.
430  */
431 static bool qdisc_tx_is_default(const struct net_device *dev)
432 {
433 	struct netdev_queue *txq;
434 	struct Qdisc *qdisc;
435 
436 	if (dev->num_tx_queues > 1)
437 		return false;
438 
439 	txq = netdev_get_tx_queue(dev, 0);
440 	qdisc = rcu_access_pointer(txq->qdisc);
441 
442 	return !qdisc->enqueue;
443 }
444 
445 /* Local traffic destined to local address. Reinsert the packet to rx
446  * path, similar to loopback handling.
447  */
448 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
449 			  struct dst_entry *dst)
450 {
451 	int len = skb->len;
452 
453 	skb_orphan(skb);
454 
455 	skb_dst_set(skb, dst);
456 
457 	/* set pkt_type to avoid skb hitting packet taps twice -
458 	 * once on Tx and again in Rx processing
459 	 */
460 	skb->pkt_type = PACKET_LOOPBACK;
461 
462 	skb->protocol = eth_type_trans(skb, dev);
463 
464 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
465 		vrf_rx_stats(dev, len);
466 	else
467 		this_cpu_inc(dev->dstats->rx_drps);
468 
469 	return NETDEV_TX_OK;
470 }
471 
472 #if IS_ENABLED(CONFIG_IPV6)
473 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
474 			     struct sk_buff *skb)
475 {
476 	int err;
477 
478 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
479 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
480 
481 	if (likely(err == 1))
482 		err = dst_output(net, sk, skb);
483 
484 	return err;
485 }
486 
487 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
488 					   struct net_device *dev)
489 {
490 	const struct ipv6hdr *iph;
491 	struct net *net = dev_net(skb->dev);
492 	struct flowi6 fl6;
493 	int ret = NET_XMIT_DROP;
494 	struct dst_entry *dst;
495 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
496 
497 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
498 		goto err;
499 
500 	iph = ipv6_hdr(skb);
501 
502 	memset(&fl6, 0, sizeof(fl6));
503 	/* needed to match OIF rule */
504 	fl6.flowi6_oif = dev->ifindex;
505 	fl6.flowi6_iif = LOOPBACK_IFINDEX;
506 	fl6.daddr = iph->daddr;
507 	fl6.saddr = iph->saddr;
508 	fl6.flowlabel = ip6_flowinfo(iph);
509 	fl6.flowi6_mark = skb->mark;
510 	fl6.flowi6_proto = iph->nexthdr;
511 	fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
512 
513 	dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
514 	if (IS_ERR(dst) || dst == dst_null)
515 		goto err;
516 
517 	skb_dst_drop(skb);
518 
519 	/* if dst.dev is loopback or the VRF device again this is locally
520 	 * originated traffic destined to a local address. Short circuit
521 	 * to Rx path
522 	 */
523 	if (dst->dev == dev)
524 		return vrf_local_xmit(skb, dev, dst);
525 
526 	skb_dst_set(skb, dst);
527 
528 	/* strip the ethernet header added for pass through VRF device */
529 	__skb_pull(skb, skb_network_offset(skb));
530 
531 	ret = vrf_ip6_local_out(net, skb->sk, skb);
532 	if (unlikely(net_xmit_eval(ret)))
533 		dev->stats.tx_errors++;
534 	else
535 		ret = NET_XMIT_SUCCESS;
536 
537 	return ret;
538 err:
539 	vrf_tx_error(dev, skb);
540 	return NET_XMIT_DROP;
541 }
542 #else
543 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
544 					   struct net_device *dev)
545 {
546 	vrf_tx_error(dev, skb);
547 	return NET_XMIT_DROP;
548 }
549 #endif
550 
551 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
552 static int vrf_ip_local_out(struct net *net, struct sock *sk,
553 			    struct sk_buff *skb)
554 {
555 	int err;
556 
557 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
558 		      skb, NULL, skb_dst(skb)->dev, dst_output);
559 	if (likely(err == 1))
560 		err = dst_output(net, sk, skb);
561 
562 	return err;
563 }
564 
565 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
566 					   struct net_device *vrf_dev)
567 {
568 	struct iphdr *ip4h;
569 	int ret = NET_XMIT_DROP;
570 	struct flowi4 fl4;
571 	struct net *net = dev_net(vrf_dev);
572 	struct rtable *rt;
573 
574 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
575 		goto err;
576 
577 	ip4h = ip_hdr(skb);
578 
579 	memset(&fl4, 0, sizeof(fl4));
580 	/* needed to match OIF rule */
581 	fl4.flowi4_oif = vrf_dev->ifindex;
582 	fl4.flowi4_iif = LOOPBACK_IFINDEX;
583 	fl4.flowi4_tos = RT_TOS(ip4h->tos);
584 	fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
585 	fl4.flowi4_proto = ip4h->protocol;
586 	fl4.daddr = ip4h->daddr;
587 	fl4.saddr = ip4h->saddr;
588 
589 	rt = ip_route_output_flow(net, &fl4, NULL);
590 	if (IS_ERR(rt))
591 		goto err;
592 
593 	skb_dst_drop(skb);
594 
595 	/* if dst.dev is loopback or the VRF device again this is locally
596 	 * originated traffic destined to a local address. Short circuit
597 	 * to Rx path
598 	 */
599 	if (rt->dst.dev == vrf_dev)
600 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
601 
602 	skb_dst_set(skb, &rt->dst);
603 
604 	/* strip the ethernet header added for pass through VRF device */
605 	__skb_pull(skb, skb_network_offset(skb));
606 
607 	if (!ip4h->saddr) {
608 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
609 					       RT_SCOPE_LINK);
610 	}
611 
612 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
613 	if (unlikely(net_xmit_eval(ret)))
614 		vrf_dev->stats.tx_errors++;
615 	else
616 		ret = NET_XMIT_SUCCESS;
617 
618 out:
619 	return ret;
620 err:
621 	vrf_tx_error(vrf_dev, skb);
622 	goto out;
623 }
624 
625 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
626 {
627 	switch (skb->protocol) {
628 	case htons(ETH_P_IP):
629 		return vrf_process_v4_outbound(skb, dev);
630 	case htons(ETH_P_IPV6):
631 		return vrf_process_v6_outbound(skb, dev);
632 	default:
633 		vrf_tx_error(dev, skb);
634 		return NET_XMIT_DROP;
635 	}
636 }
637 
638 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
639 {
640 	int len = skb->len;
641 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
642 
643 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
644 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
645 
646 		u64_stats_update_begin(&dstats->syncp);
647 		dstats->tx_pkts++;
648 		dstats->tx_bytes += len;
649 		u64_stats_update_end(&dstats->syncp);
650 	} else {
651 		this_cpu_inc(dev->dstats->tx_drps);
652 	}
653 
654 	return ret;
655 }
656 
657 static int vrf_finish_direct(struct net *net, struct sock *sk,
658 			     struct sk_buff *skb)
659 {
660 	struct net_device *vrf_dev = skb->dev;
661 
662 	if (!list_empty(&vrf_dev->ptype_all) &&
663 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
664 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
665 
666 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
667 		eth_zero_addr(eth->h_dest);
668 		eth->h_proto = skb->protocol;
669 
670 		rcu_read_lock_bh();
671 		dev_queue_xmit_nit(skb, vrf_dev);
672 		rcu_read_unlock_bh();
673 
674 		skb_pull(skb, ETH_HLEN);
675 	}
676 
677 	return 1;
678 }
679 
680 #if IS_ENABLED(CONFIG_IPV6)
681 /* modelled after ip6_finish_output2 */
682 static int vrf_finish_output6(struct net *net, struct sock *sk,
683 			      struct sk_buff *skb)
684 {
685 	struct dst_entry *dst = skb_dst(skb);
686 	struct net_device *dev = dst->dev;
687 	const struct in6_addr *nexthop;
688 	struct neighbour *neigh;
689 	int ret;
690 
691 	nf_reset_ct(skb);
692 
693 	skb->protocol = htons(ETH_P_IPV6);
694 	skb->dev = dev;
695 
696 	rcu_read_lock_bh();
697 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
698 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
699 	if (unlikely(!neigh))
700 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
701 	if (!IS_ERR(neigh)) {
702 		sock_confirm_neigh(skb, neigh);
703 		ret = neigh_output(neigh, skb, false);
704 		rcu_read_unlock_bh();
705 		return ret;
706 	}
707 	rcu_read_unlock_bh();
708 
709 	IP6_INC_STATS(dev_net(dst->dev),
710 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
711 	kfree_skb(skb);
712 	return -EINVAL;
713 }
714 
715 /* modelled after ip6_output */
716 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
717 {
718 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
719 			    net, sk, skb, NULL, skb_dst(skb)->dev,
720 			    vrf_finish_output6,
721 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
722 }
723 
724 /* set dst on skb to send packet to us via dev_xmit path. Allows
725  * packet to go through device based features such as qdisc, netfilter
726  * hooks and packet sockets with skb->dev set to vrf device.
727  */
728 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
729 					    struct sk_buff *skb)
730 {
731 	struct net_vrf *vrf = netdev_priv(vrf_dev);
732 	struct dst_entry *dst = NULL;
733 	struct rt6_info *rt6;
734 
735 	rcu_read_lock();
736 
737 	rt6 = rcu_dereference(vrf->rt6);
738 	if (likely(rt6)) {
739 		dst = &rt6->dst;
740 		dst_hold(dst);
741 	}
742 
743 	rcu_read_unlock();
744 
745 	if (unlikely(!dst)) {
746 		vrf_tx_error(vrf_dev, skb);
747 		return NULL;
748 	}
749 
750 	skb_dst_drop(skb);
751 	skb_dst_set(skb, dst);
752 
753 	return skb;
754 }
755 
756 static int vrf_output6_direct(struct net *net, struct sock *sk,
757 			      struct sk_buff *skb)
758 {
759 	skb->protocol = htons(ETH_P_IPV6);
760 
761 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
762 			    net, sk, skb, NULL, skb->dev,
763 			    vrf_finish_direct,
764 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
765 }
766 
767 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
768 					  struct sock *sk,
769 					  struct sk_buff *skb)
770 {
771 	struct net *net = dev_net(vrf_dev);
772 	int err;
773 
774 	skb->dev = vrf_dev;
775 
776 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
777 		      skb, NULL, vrf_dev, vrf_output6_direct);
778 
779 	if (likely(err == 1))
780 		err = vrf_output6_direct(net, sk, skb);
781 
782 	/* reset skb device */
783 	if (likely(err == 1))
784 		nf_reset_ct(skb);
785 	else
786 		skb = NULL;
787 
788 	return skb;
789 }
790 
791 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
792 				   struct sock *sk,
793 				   struct sk_buff *skb)
794 {
795 	/* don't divert link scope packets */
796 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
797 		return skb;
798 
799 	if (qdisc_tx_is_default(vrf_dev) ||
800 	    IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
801 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
802 
803 	return vrf_ip6_out_redirect(vrf_dev, skb);
804 }
805 
806 /* holding rtnl */
807 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
808 {
809 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
810 	struct net *net = dev_net(dev);
811 	struct dst_entry *dst;
812 
813 	RCU_INIT_POINTER(vrf->rt6, NULL);
814 	synchronize_rcu();
815 
816 	/* move dev in dst's to loopback so this VRF device can be deleted
817 	 * - based on dst_ifdown
818 	 */
819 	if (rt6) {
820 		dst = &rt6->dst;
821 		dev_put(dst->dev);
822 		dst->dev = net->loopback_dev;
823 		dev_hold(dst->dev);
824 		dst_release(dst);
825 	}
826 }
827 
828 static int vrf_rt6_create(struct net_device *dev)
829 {
830 	int flags = DST_NOPOLICY | DST_NOXFRM;
831 	struct net_vrf *vrf = netdev_priv(dev);
832 	struct net *net = dev_net(dev);
833 	struct rt6_info *rt6;
834 	int rc = -ENOMEM;
835 
836 	/* IPv6 can be CONFIG enabled and then disabled runtime */
837 	if (!ipv6_mod_enabled())
838 		return 0;
839 
840 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
841 	if (!vrf->fib6_table)
842 		goto out;
843 
844 	/* create a dst for routing packets out a VRF device */
845 	rt6 = ip6_dst_alloc(net, dev, flags);
846 	if (!rt6)
847 		goto out;
848 
849 	rt6->dst.output	= vrf_output6;
850 
851 	rcu_assign_pointer(vrf->rt6, rt6);
852 
853 	rc = 0;
854 out:
855 	return rc;
856 }
857 #else
858 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
859 				   struct sock *sk,
860 				   struct sk_buff *skb)
861 {
862 	return skb;
863 }
864 
865 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
866 {
867 }
868 
869 static int vrf_rt6_create(struct net_device *dev)
870 {
871 	return 0;
872 }
873 #endif
874 
875 /* modelled after ip_finish_output2 */
876 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
877 {
878 	struct dst_entry *dst = skb_dst(skb);
879 	struct rtable *rt = (struct rtable *)dst;
880 	struct net_device *dev = dst->dev;
881 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
882 	struct neighbour *neigh;
883 	bool is_v6gw = false;
884 	int ret = -EINVAL;
885 
886 	nf_reset_ct(skb);
887 
888 	/* Be paranoid, rather than too clever. */
889 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
890 		struct sk_buff *skb2;
891 
892 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
893 		if (!skb2) {
894 			ret = -ENOMEM;
895 			goto err;
896 		}
897 		if (skb->sk)
898 			skb_set_owner_w(skb2, skb->sk);
899 
900 		consume_skb(skb);
901 		skb = skb2;
902 	}
903 
904 	rcu_read_lock_bh();
905 
906 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
907 	if (!IS_ERR(neigh)) {
908 		sock_confirm_neigh(skb, neigh);
909 		/* if crossing protocols, can not use the cached header */
910 		ret = neigh_output(neigh, skb, is_v6gw);
911 		rcu_read_unlock_bh();
912 		return ret;
913 	}
914 
915 	rcu_read_unlock_bh();
916 err:
917 	vrf_tx_error(skb->dev, skb);
918 	return ret;
919 }
920 
921 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
922 {
923 	struct net_device *dev = skb_dst(skb)->dev;
924 
925 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
926 
927 	skb->dev = dev;
928 	skb->protocol = htons(ETH_P_IP);
929 
930 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
931 			    net, sk, skb, NULL, dev,
932 			    vrf_finish_output,
933 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
934 }
935 
936 /* set dst on skb to send packet to us via dev_xmit path. Allows
937  * packet to go through device based features such as qdisc, netfilter
938  * hooks and packet sockets with skb->dev set to vrf device.
939  */
940 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
941 					   struct sk_buff *skb)
942 {
943 	struct net_vrf *vrf = netdev_priv(vrf_dev);
944 	struct dst_entry *dst = NULL;
945 	struct rtable *rth;
946 
947 	rcu_read_lock();
948 
949 	rth = rcu_dereference(vrf->rth);
950 	if (likely(rth)) {
951 		dst = &rth->dst;
952 		dst_hold(dst);
953 	}
954 
955 	rcu_read_unlock();
956 
957 	if (unlikely(!dst)) {
958 		vrf_tx_error(vrf_dev, skb);
959 		return NULL;
960 	}
961 
962 	skb_dst_drop(skb);
963 	skb_dst_set(skb, dst);
964 
965 	return skb;
966 }
967 
968 static int vrf_output_direct(struct net *net, struct sock *sk,
969 			     struct sk_buff *skb)
970 {
971 	skb->protocol = htons(ETH_P_IP);
972 
973 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
974 			    net, sk, skb, NULL, skb->dev,
975 			    vrf_finish_direct,
976 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
977 }
978 
979 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
980 					 struct sock *sk,
981 					 struct sk_buff *skb)
982 {
983 	struct net *net = dev_net(vrf_dev);
984 	int err;
985 
986 	skb->dev = vrf_dev;
987 
988 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
989 		      skb, NULL, vrf_dev, vrf_output_direct);
990 
991 	if (likely(err == 1))
992 		err = vrf_output_direct(net, sk, skb);
993 
994 	/* reset skb device */
995 	if (likely(err == 1))
996 		nf_reset_ct(skb);
997 	else
998 		skb = NULL;
999 
1000 	return skb;
1001 }
1002 
1003 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
1004 				  struct sock *sk,
1005 				  struct sk_buff *skb)
1006 {
1007 	/* don't divert multicast or local broadcast */
1008 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
1009 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
1010 		return skb;
1011 
1012 	if (qdisc_tx_is_default(vrf_dev) ||
1013 	    IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
1014 		return vrf_ip_out_direct(vrf_dev, sk, skb);
1015 
1016 	return vrf_ip_out_redirect(vrf_dev, skb);
1017 }
1018 
1019 /* called with rcu lock held */
1020 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1021 				  struct sock *sk,
1022 				  struct sk_buff *skb,
1023 				  u16 proto)
1024 {
1025 	switch (proto) {
1026 	case AF_INET:
1027 		return vrf_ip_out(vrf_dev, sk, skb);
1028 	case AF_INET6:
1029 		return vrf_ip6_out(vrf_dev, sk, skb);
1030 	}
1031 
1032 	return skb;
1033 }
1034 
1035 /* holding rtnl */
1036 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
1037 {
1038 	struct rtable *rth = rtnl_dereference(vrf->rth);
1039 	struct net *net = dev_net(dev);
1040 	struct dst_entry *dst;
1041 
1042 	RCU_INIT_POINTER(vrf->rth, NULL);
1043 	synchronize_rcu();
1044 
1045 	/* move dev in dst's to loopback so this VRF device can be deleted
1046 	 * - based on dst_ifdown
1047 	 */
1048 	if (rth) {
1049 		dst = &rth->dst;
1050 		dev_put(dst->dev);
1051 		dst->dev = net->loopback_dev;
1052 		dev_hold(dst->dev);
1053 		dst_release(dst);
1054 	}
1055 }
1056 
1057 static int vrf_rtable_create(struct net_device *dev)
1058 {
1059 	struct net_vrf *vrf = netdev_priv(dev);
1060 	struct rtable *rth;
1061 
1062 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
1063 		return -ENOMEM;
1064 
1065 	/* create a dst for routing packets out through a VRF device */
1066 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
1067 	if (!rth)
1068 		return -ENOMEM;
1069 
1070 	rth->dst.output	= vrf_output;
1071 
1072 	rcu_assign_pointer(vrf->rth, rth);
1073 
1074 	return 0;
1075 }
1076 
1077 /**************************** device handling ********************/
1078 
1079 /* cycle interface to flush neighbor cache and move routes across tables */
1080 static void cycle_netdev(struct net_device *dev,
1081 			 struct netlink_ext_ack *extack)
1082 {
1083 	unsigned int flags = dev->flags;
1084 	int ret;
1085 
1086 	if (!netif_running(dev))
1087 		return;
1088 
1089 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1090 	if (ret >= 0)
1091 		ret = dev_change_flags(dev, flags, extack);
1092 
1093 	if (ret < 0) {
1094 		netdev_err(dev,
1095 			   "Failed to cycle device %s; route tables might be wrong!\n",
1096 			   dev->name);
1097 	}
1098 }
1099 
1100 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1101 			    struct netlink_ext_ack *extack)
1102 {
1103 	int ret;
1104 
1105 	/* do not allow loopback device to be enslaved to a VRF.
1106 	 * The vrf device acts as the loopback for the vrf.
1107 	 */
1108 	if (port_dev == dev_net(dev)->loopback_dev) {
1109 		NL_SET_ERR_MSG(extack,
1110 			       "Can not enslave loopback device to a VRF");
1111 		return -EOPNOTSUPP;
1112 	}
1113 
1114 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1115 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1116 	if (ret < 0)
1117 		goto err;
1118 
1119 	cycle_netdev(port_dev, extack);
1120 
1121 	return 0;
1122 
1123 err:
1124 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1125 	return ret;
1126 }
1127 
1128 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1129 			 struct netlink_ext_ack *extack)
1130 {
1131 	if (netif_is_l3_master(port_dev)) {
1132 		NL_SET_ERR_MSG(extack,
1133 			       "Can not enslave an L3 master device to a VRF");
1134 		return -EINVAL;
1135 	}
1136 
1137 	if (netif_is_l3_slave(port_dev))
1138 		return -EINVAL;
1139 
1140 	return do_vrf_add_slave(dev, port_dev, extack);
1141 }
1142 
1143 /* inverse of do_vrf_add_slave */
1144 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1145 {
1146 	netdev_upper_dev_unlink(port_dev, dev);
1147 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1148 
1149 	cycle_netdev(port_dev, NULL);
1150 
1151 	return 0;
1152 }
1153 
1154 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1155 {
1156 	return do_vrf_del_slave(dev, port_dev);
1157 }
1158 
1159 static void vrf_dev_uninit(struct net_device *dev)
1160 {
1161 	struct net_vrf *vrf = netdev_priv(dev);
1162 
1163 	vrf_rtable_release(dev, vrf);
1164 	vrf_rt6_release(dev, vrf);
1165 
1166 	free_percpu(dev->dstats);
1167 	dev->dstats = NULL;
1168 }
1169 
1170 static int vrf_dev_init(struct net_device *dev)
1171 {
1172 	struct net_vrf *vrf = netdev_priv(dev);
1173 
1174 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1175 	if (!dev->dstats)
1176 		goto out_nomem;
1177 
1178 	/* create the default dst which points back to us */
1179 	if (vrf_rtable_create(dev) != 0)
1180 		goto out_stats;
1181 
1182 	if (vrf_rt6_create(dev) != 0)
1183 		goto out_rth;
1184 
1185 	dev->flags = IFF_MASTER | IFF_NOARP;
1186 
1187 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
1188 	dev->mtu = 64 * 1024;
1189 
1190 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
1191 	dev->operstate = IF_OPER_UP;
1192 	netdev_lockdep_set_classes(dev);
1193 	return 0;
1194 
1195 out_rth:
1196 	vrf_rtable_release(dev, vrf);
1197 out_stats:
1198 	free_percpu(dev->dstats);
1199 	dev->dstats = NULL;
1200 out_nomem:
1201 	return -ENOMEM;
1202 }
1203 
1204 static const struct net_device_ops vrf_netdev_ops = {
1205 	.ndo_init		= vrf_dev_init,
1206 	.ndo_uninit		= vrf_dev_uninit,
1207 	.ndo_start_xmit		= vrf_xmit,
1208 	.ndo_set_mac_address	= eth_mac_addr,
1209 	.ndo_get_stats64	= vrf_get_stats64,
1210 	.ndo_add_slave		= vrf_add_slave,
1211 	.ndo_del_slave		= vrf_del_slave,
1212 };
1213 
1214 static u32 vrf_fib_table(const struct net_device *dev)
1215 {
1216 	struct net_vrf *vrf = netdev_priv(dev);
1217 
1218 	return vrf->tb_id;
1219 }
1220 
1221 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1222 {
1223 	kfree_skb(skb);
1224 	return 0;
1225 }
1226 
1227 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1228 				      struct sk_buff *skb,
1229 				      struct net_device *dev)
1230 {
1231 	struct net *net = dev_net(dev);
1232 
1233 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1234 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
1235 
1236 	return skb;
1237 }
1238 
1239 #if IS_ENABLED(CONFIG_IPV6)
1240 /* neighbor handling is done with actual device; do not want
1241  * to flip skb->dev for those ndisc packets. This really fails
1242  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1243  * a start.
1244  */
1245 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1246 {
1247 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1248 	bool rc = false;
1249 
1250 	if (iph->nexthdr == NEXTHDR_ICMP) {
1251 		const struct icmp6hdr *icmph;
1252 		struct icmp6hdr _icmph;
1253 
1254 		icmph = skb_header_pointer(skb, sizeof(*iph),
1255 					   sizeof(_icmph), &_icmph);
1256 		if (!icmph)
1257 			goto out;
1258 
1259 		switch (icmph->icmp6_type) {
1260 		case NDISC_ROUTER_SOLICITATION:
1261 		case NDISC_ROUTER_ADVERTISEMENT:
1262 		case NDISC_NEIGHBOUR_SOLICITATION:
1263 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
1264 		case NDISC_REDIRECT:
1265 			rc = true;
1266 			break;
1267 		}
1268 	}
1269 
1270 out:
1271 	return rc;
1272 }
1273 
1274 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1275 					     const struct net_device *dev,
1276 					     struct flowi6 *fl6,
1277 					     int ifindex,
1278 					     const struct sk_buff *skb,
1279 					     int flags)
1280 {
1281 	struct net_vrf *vrf = netdev_priv(dev);
1282 
1283 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1284 }
1285 
1286 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1287 			      int ifindex)
1288 {
1289 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1290 	struct flowi6 fl6 = {
1291 		.flowi6_iif     = ifindex,
1292 		.flowi6_mark    = skb->mark,
1293 		.flowi6_proto   = iph->nexthdr,
1294 		.daddr          = iph->daddr,
1295 		.saddr          = iph->saddr,
1296 		.flowlabel      = ip6_flowinfo(iph),
1297 	};
1298 	struct net *net = dev_net(vrf_dev);
1299 	struct rt6_info *rt6;
1300 
1301 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1302 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1303 	if (unlikely(!rt6))
1304 		return;
1305 
1306 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1307 		return;
1308 
1309 	skb_dst_set(skb, &rt6->dst);
1310 }
1311 
1312 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1313 				   struct sk_buff *skb)
1314 {
1315 	int orig_iif = skb->skb_iif;
1316 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1317 	bool is_ndisc = ipv6_ndisc_frame(skb);
1318 
1319 	/* loopback, multicast & non-ND link-local traffic; do not push through
1320 	 * packet taps again. Reset pkt_type for upper layers to process skb
1321 	 */
1322 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1323 		skb->dev = vrf_dev;
1324 		skb->skb_iif = vrf_dev->ifindex;
1325 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1326 		if (skb->pkt_type == PACKET_LOOPBACK)
1327 			skb->pkt_type = PACKET_HOST;
1328 		goto out;
1329 	}
1330 
1331 	/* if packet is NDISC then keep the ingress interface */
1332 	if (!is_ndisc) {
1333 		vrf_rx_stats(vrf_dev, skb->len);
1334 		skb->dev = vrf_dev;
1335 		skb->skb_iif = vrf_dev->ifindex;
1336 
1337 		if (!list_empty(&vrf_dev->ptype_all)) {
1338 			skb_push(skb, skb->mac_len);
1339 			dev_queue_xmit_nit(skb, vrf_dev);
1340 			skb_pull(skb, skb->mac_len);
1341 		}
1342 
1343 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1344 	}
1345 
1346 	if (need_strict)
1347 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1348 
1349 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1350 out:
1351 	return skb;
1352 }
1353 
1354 #else
1355 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1356 				   struct sk_buff *skb)
1357 {
1358 	return skb;
1359 }
1360 #endif
1361 
1362 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1363 				  struct sk_buff *skb)
1364 {
1365 	skb->dev = vrf_dev;
1366 	skb->skb_iif = vrf_dev->ifindex;
1367 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1368 
1369 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1370 		goto out;
1371 
1372 	/* loopback traffic; do not push through packet taps again.
1373 	 * Reset pkt_type for upper layers to process skb
1374 	 */
1375 	if (skb->pkt_type == PACKET_LOOPBACK) {
1376 		skb->pkt_type = PACKET_HOST;
1377 		goto out;
1378 	}
1379 
1380 	vrf_rx_stats(vrf_dev, skb->len);
1381 
1382 	if (!list_empty(&vrf_dev->ptype_all)) {
1383 		skb_push(skb, skb->mac_len);
1384 		dev_queue_xmit_nit(skb, vrf_dev);
1385 		skb_pull(skb, skb->mac_len);
1386 	}
1387 
1388 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1389 out:
1390 	return skb;
1391 }
1392 
1393 /* called with rcu lock held */
1394 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1395 				  struct sk_buff *skb,
1396 				  u16 proto)
1397 {
1398 	switch (proto) {
1399 	case AF_INET:
1400 		return vrf_ip_rcv(vrf_dev, skb);
1401 	case AF_INET6:
1402 		return vrf_ip6_rcv(vrf_dev, skb);
1403 	}
1404 
1405 	return skb;
1406 }
1407 
1408 #if IS_ENABLED(CONFIG_IPV6)
1409 /* send to link-local or multicast address via interface enslaved to
1410  * VRF device. Force lookup to VRF table without changing flow struct
1411  * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1412  * is taken on the dst by this function.
1413  */
1414 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1415 					      struct flowi6 *fl6)
1416 {
1417 	struct net *net = dev_net(dev);
1418 	int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1419 	struct dst_entry *dst = NULL;
1420 	struct rt6_info *rt;
1421 
1422 	/* VRF device does not have a link-local address and
1423 	 * sending packets to link-local or mcast addresses over
1424 	 * a VRF device does not make sense
1425 	 */
1426 	if (fl6->flowi6_oif == dev->ifindex) {
1427 		dst = &net->ipv6.ip6_null_entry->dst;
1428 		return dst;
1429 	}
1430 
1431 	if (!ipv6_addr_any(&fl6->saddr))
1432 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1433 
1434 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1435 	if (rt)
1436 		dst = &rt->dst;
1437 
1438 	return dst;
1439 }
1440 #endif
1441 
1442 static const struct l3mdev_ops vrf_l3mdev_ops = {
1443 	.l3mdev_fib_table	= vrf_fib_table,
1444 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1445 	.l3mdev_l3_out		= vrf_l3_out,
1446 #if IS_ENABLED(CONFIG_IPV6)
1447 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1448 #endif
1449 };
1450 
1451 static void vrf_get_drvinfo(struct net_device *dev,
1452 			    struct ethtool_drvinfo *info)
1453 {
1454 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1455 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1456 }
1457 
1458 static const struct ethtool_ops vrf_ethtool_ops = {
1459 	.get_drvinfo	= vrf_get_drvinfo,
1460 };
1461 
1462 static inline size_t vrf_fib_rule_nl_size(void)
1463 {
1464 	size_t sz;
1465 
1466 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1467 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1468 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1469 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1470 
1471 	return sz;
1472 }
1473 
1474 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1475 {
1476 	struct fib_rule_hdr *frh;
1477 	struct nlmsghdr *nlh;
1478 	struct sk_buff *skb;
1479 	int err;
1480 
1481 	if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1482 	    !ipv6_mod_enabled())
1483 		return 0;
1484 
1485 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1486 	if (!skb)
1487 		return -ENOMEM;
1488 
1489 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1490 	if (!nlh)
1491 		goto nla_put_failure;
1492 
1493 	/* rule only needs to appear once */
1494 	nlh->nlmsg_flags |= NLM_F_EXCL;
1495 
1496 	frh = nlmsg_data(nlh);
1497 	memset(frh, 0, sizeof(*frh));
1498 	frh->family = family;
1499 	frh->action = FR_ACT_TO_TBL;
1500 
1501 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1502 		goto nla_put_failure;
1503 
1504 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1505 		goto nla_put_failure;
1506 
1507 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1508 		goto nla_put_failure;
1509 
1510 	nlmsg_end(skb, nlh);
1511 
1512 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1513 	skb->sk = dev_net(dev)->rtnl;
1514 	if (add_it) {
1515 		err = fib_nl_newrule(skb, nlh, NULL);
1516 		if (err == -EEXIST)
1517 			err = 0;
1518 	} else {
1519 		err = fib_nl_delrule(skb, nlh, NULL);
1520 		if (err == -ENOENT)
1521 			err = 0;
1522 	}
1523 	nlmsg_free(skb);
1524 
1525 	return err;
1526 
1527 nla_put_failure:
1528 	nlmsg_free(skb);
1529 
1530 	return -EMSGSIZE;
1531 }
1532 
1533 static int vrf_add_fib_rules(const struct net_device *dev)
1534 {
1535 	int err;
1536 
1537 	err = vrf_fib_rule(dev, AF_INET,  true);
1538 	if (err < 0)
1539 		goto out_err;
1540 
1541 	err = vrf_fib_rule(dev, AF_INET6, true);
1542 	if (err < 0)
1543 		goto ipv6_err;
1544 
1545 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1546 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1547 	if (err < 0)
1548 		goto ipmr_err;
1549 #endif
1550 
1551 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1552 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1553 	if (err < 0)
1554 		goto ip6mr_err;
1555 #endif
1556 
1557 	return 0;
1558 
1559 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1560 ip6mr_err:
1561 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1562 #endif
1563 
1564 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1565 ipmr_err:
1566 	vrf_fib_rule(dev, AF_INET6,  false);
1567 #endif
1568 
1569 ipv6_err:
1570 	vrf_fib_rule(dev, AF_INET,  false);
1571 
1572 out_err:
1573 	netdev_err(dev, "Failed to add FIB rules.\n");
1574 	return err;
1575 }
1576 
1577 static void vrf_setup(struct net_device *dev)
1578 {
1579 	ether_setup(dev);
1580 
1581 	/* Initialize the device structure. */
1582 	dev->netdev_ops = &vrf_netdev_ops;
1583 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1584 	dev->ethtool_ops = &vrf_ethtool_ops;
1585 	dev->needs_free_netdev = true;
1586 
1587 	/* Fill in device structure with ethernet-generic values. */
1588 	eth_hw_addr_random(dev);
1589 
1590 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1591 	dev->features |= NETIF_F_LLTX;
1592 
1593 	/* don't allow vrf devices to change network namespaces. */
1594 	dev->features |= NETIF_F_NETNS_LOCAL;
1595 
1596 	/* does not make sense for a VLAN to be added to a vrf device */
1597 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1598 
1599 	/* enable offload features */
1600 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1601 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1602 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1603 
1604 	dev->hw_features = dev->features;
1605 	dev->hw_enc_features = dev->features;
1606 
1607 	/* default to no qdisc; user can add if desired */
1608 	dev->priv_flags |= IFF_NO_QUEUE;
1609 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1610 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1611 
1612 	/* VRF devices do not care about MTU, but if the MTU is set
1613 	 * too low then the ipv4 and ipv6 protocols are disabled
1614 	 * which breaks networking.
1615 	 */
1616 	dev->min_mtu = IPV6_MIN_MTU;
1617 	dev->max_mtu = ETH_MAX_MTU;
1618 }
1619 
1620 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1621 			struct netlink_ext_ack *extack)
1622 {
1623 	if (tb[IFLA_ADDRESS]) {
1624 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1625 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1626 			return -EINVAL;
1627 		}
1628 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1629 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1630 			return -EADDRNOTAVAIL;
1631 		}
1632 	}
1633 	return 0;
1634 }
1635 
1636 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1637 {
1638 	struct net_device *port_dev;
1639 	struct list_head *iter;
1640 
1641 	netdev_for_each_lower_dev(dev, port_dev, iter)
1642 		vrf_del_slave(dev, port_dev);
1643 
1644 	vrf_map_unregister_dev(dev);
1645 
1646 	unregister_netdevice_queue(dev, head);
1647 }
1648 
1649 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1650 		       struct nlattr *tb[], struct nlattr *data[],
1651 		       struct netlink_ext_ack *extack)
1652 {
1653 	struct net_vrf *vrf = netdev_priv(dev);
1654 	struct netns_vrf *nn_vrf;
1655 	bool *add_fib_rules;
1656 	struct net *net;
1657 	int err;
1658 
1659 	if (!data || !data[IFLA_VRF_TABLE]) {
1660 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1661 		return -EINVAL;
1662 	}
1663 
1664 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1665 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1666 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1667 				    "Invalid VRF table id");
1668 		return -EINVAL;
1669 	}
1670 
1671 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1672 
1673 	err = register_netdevice(dev);
1674 	if (err)
1675 		goto out;
1676 
1677 	/* mapping between table_id and vrf;
1678 	 * note: such binding could not be done in the dev init function
1679 	 * because dev->ifindex id is not available yet.
1680 	 */
1681 	vrf->ifindex = dev->ifindex;
1682 
1683 	err = vrf_map_register_dev(dev, extack);
1684 	if (err) {
1685 		unregister_netdevice(dev);
1686 		goto out;
1687 	}
1688 
1689 	net = dev_net(dev);
1690 	nn_vrf = net_generic(net, vrf_net_id);
1691 
1692 	add_fib_rules = &nn_vrf->add_fib_rules;
1693 	if (*add_fib_rules) {
1694 		err = vrf_add_fib_rules(dev);
1695 		if (err) {
1696 			vrf_map_unregister_dev(dev);
1697 			unregister_netdevice(dev);
1698 			goto out;
1699 		}
1700 		*add_fib_rules = false;
1701 	}
1702 
1703 out:
1704 	return err;
1705 }
1706 
1707 static size_t vrf_nl_getsize(const struct net_device *dev)
1708 {
1709 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1710 }
1711 
1712 static int vrf_fillinfo(struct sk_buff *skb,
1713 			const struct net_device *dev)
1714 {
1715 	struct net_vrf *vrf = netdev_priv(dev);
1716 
1717 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1718 }
1719 
1720 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1721 				 const struct net_device *slave_dev)
1722 {
1723 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1724 }
1725 
1726 static int vrf_fill_slave_info(struct sk_buff *skb,
1727 			       const struct net_device *vrf_dev,
1728 			       const struct net_device *slave_dev)
1729 {
1730 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1731 
1732 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1733 		return -EMSGSIZE;
1734 
1735 	return 0;
1736 }
1737 
1738 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1739 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1740 };
1741 
1742 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1743 	.kind		= DRV_NAME,
1744 	.priv_size	= sizeof(struct net_vrf),
1745 
1746 	.get_size	= vrf_nl_getsize,
1747 	.policy		= vrf_nl_policy,
1748 	.validate	= vrf_validate,
1749 	.fill_info	= vrf_fillinfo,
1750 
1751 	.get_slave_size  = vrf_get_slave_size,
1752 	.fill_slave_info = vrf_fill_slave_info,
1753 
1754 	.newlink	= vrf_newlink,
1755 	.dellink	= vrf_dellink,
1756 	.setup		= vrf_setup,
1757 	.maxtype	= IFLA_VRF_MAX,
1758 };
1759 
1760 static int vrf_device_event(struct notifier_block *unused,
1761 			    unsigned long event, void *ptr)
1762 {
1763 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1764 
1765 	/* only care about unregister events to drop slave references */
1766 	if (event == NETDEV_UNREGISTER) {
1767 		struct net_device *vrf_dev;
1768 
1769 		if (!netif_is_l3_slave(dev))
1770 			goto out;
1771 
1772 		vrf_dev = netdev_master_upper_dev_get(dev);
1773 		vrf_del_slave(vrf_dev, dev);
1774 	}
1775 out:
1776 	return NOTIFY_DONE;
1777 }
1778 
1779 static struct notifier_block vrf_notifier_block __read_mostly = {
1780 	.notifier_call = vrf_device_event,
1781 };
1782 
1783 static int vrf_map_init(struct vrf_map *vmap)
1784 {
1785 	spin_lock_init(&vmap->vmap_lock);
1786 	hash_init(vmap->ht);
1787 
1788 	vmap->strict_mode = false;
1789 
1790 	return 0;
1791 }
1792 
1793 static int vrf_shared_table_handler(struct ctl_table *table, int write,
1794 				    void *buffer, size_t *lenp, loff_t *ppos)
1795 {
1796 	struct net *net = (struct net *)table->extra1;
1797 	struct vrf_map *vmap = netns_vrf_map(net);
1798 	int proc_strict_mode = 0;
1799 	struct ctl_table tmp = {
1800 		.procname	= table->procname,
1801 		.data		= &proc_strict_mode,
1802 		.maxlen		= sizeof(int),
1803 		.mode		= table->mode,
1804 		.extra1		= SYSCTL_ZERO,
1805 		.extra2		= SYSCTL_ONE,
1806 	};
1807 	int ret;
1808 
1809 	if (!write)
1810 		proc_strict_mode = vrf_strict_mode(vmap);
1811 
1812 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1813 
1814 	if (write && ret == 0)
1815 		ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1816 
1817 	return ret;
1818 }
1819 
1820 static const struct ctl_table vrf_table[] = {
1821 	{
1822 		.procname	= "strict_mode",
1823 		.data		= NULL,
1824 		.maxlen		= sizeof(int),
1825 		.mode		= 0644,
1826 		.proc_handler	= vrf_shared_table_handler,
1827 		/* set by the vrf_netns_init */
1828 		.extra1		= NULL,
1829 	},
1830 	{ },
1831 };
1832 
1833 /* Initialize per network namespace state */
1834 static int __net_init vrf_netns_init(struct net *net)
1835 {
1836 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1837 	struct ctl_table *table;
1838 	int res;
1839 
1840 	nn_vrf->add_fib_rules = true;
1841 	vrf_map_init(&nn_vrf->vmap);
1842 
1843 	table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1844 	if (!table)
1845 		return -ENOMEM;
1846 
1847 	/* init the extra1 parameter with the reference to current netns */
1848 	table[0].extra1 = net;
1849 
1850 	nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1851 	if (!nn_vrf->ctl_hdr) {
1852 		res = -ENOMEM;
1853 		goto free_table;
1854 	}
1855 
1856 	return 0;
1857 
1858 free_table:
1859 	kfree(table);
1860 
1861 	return res;
1862 }
1863 
1864 static void __net_exit vrf_netns_exit(struct net *net)
1865 {
1866 	struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1867 	struct ctl_table *table;
1868 
1869 	table = nn_vrf->ctl_hdr->ctl_table_arg;
1870 	unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1871 	kfree(table);
1872 }
1873 
1874 static struct pernet_operations vrf_net_ops __net_initdata = {
1875 	.init = vrf_netns_init,
1876 	.exit = vrf_netns_exit,
1877 	.id   = &vrf_net_id,
1878 	.size = sizeof(struct netns_vrf),
1879 };
1880 
1881 static int __init vrf_init_module(void)
1882 {
1883 	int rc;
1884 
1885 	register_netdevice_notifier(&vrf_notifier_block);
1886 
1887 	rc = register_pernet_subsys(&vrf_net_ops);
1888 	if (rc < 0)
1889 		goto error;
1890 
1891 	rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
1892 					  vrf_ifindex_lookup_by_table_id);
1893 	if (rc < 0)
1894 		goto unreg_pernet;
1895 
1896 	rc = rtnl_link_register(&vrf_link_ops);
1897 	if (rc < 0)
1898 		goto table_lookup_unreg;
1899 
1900 	return 0;
1901 
1902 table_lookup_unreg:
1903 	l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
1904 				       vrf_ifindex_lookup_by_table_id);
1905 
1906 unreg_pernet:
1907 	unregister_pernet_subsys(&vrf_net_ops);
1908 
1909 error:
1910 	unregister_netdevice_notifier(&vrf_notifier_block);
1911 	return rc;
1912 }
1913 
1914 module_init(vrf_init_module);
1915 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1916 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1917 MODULE_LICENSE("GPL");
1918 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1919 MODULE_VERSION(DRV_VERSION);
1920