xref: /linux/drivers/net/vrf.c (revision 55223394d56bab42ebac71ba52e0fd8bfdc6fc07)
1 /*
2  * vrf.c: device driver to encapsulate a VRF space
3  *
4  * Copyright (c) 2015 Cumulus Networks. All rights reserved.
5  * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
6  * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
7  *
8  * Based on dummy, team and ipvlan drivers
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  */
15 
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/netdevice.h>
19 #include <linux/etherdevice.h>
20 #include <linux/ip.h>
21 #include <linux/init.h>
22 #include <linux/moduleparam.h>
23 #include <linux/netfilter.h>
24 #include <linux/rtnetlink.h>
25 #include <net/rtnetlink.h>
26 #include <linux/u64_stats_sync.h>
27 #include <linux/hashtable.h>
28 
29 #include <linux/inetdevice.h>
30 #include <net/arp.h>
31 #include <net/ip.h>
32 #include <net/ip_fib.h>
33 #include <net/ip6_fib.h>
34 #include <net/ip6_route.h>
35 #include <net/route.h>
36 #include <net/addrconf.h>
37 #include <net/l3mdev.h>
38 #include <net/fib_rules.h>
39 #include <net/netns/generic.h>
40 
41 #define DRV_NAME	"vrf"
42 #define DRV_VERSION	"1.0"
43 
44 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
45 
46 static unsigned int vrf_net_id;
47 
48 struct net_vrf {
49 	struct rtable __rcu	*rth;
50 	struct rt6_info	__rcu	*rt6;
51 #if IS_ENABLED(CONFIG_IPV6)
52 	struct fib6_table	*fib6_table;
53 #endif
54 	u32                     tb_id;
55 };
56 
57 struct pcpu_dstats {
58 	u64			tx_pkts;
59 	u64			tx_bytes;
60 	u64			tx_drps;
61 	u64			rx_pkts;
62 	u64			rx_bytes;
63 	u64			rx_drps;
64 	struct u64_stats_sync	syncp;
65 };
66 
67 static void vrf_rx_stats(struct net_device *dev, int len)
68 {
69 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
70 
71 	u64_stats_update_begin(&dstats->syncp);
72 	dstats->rx_pkts++;
73 	dstats->rx_bytes += len;
74 	u64_stats_update_end(&dstats->syncp);
75 }
76 
77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
78 {
79 	vrf_dev->stats.tx_errors++;
80 	kfree_skb(skb);
81 }
82 
83 static void vrf_get_stats64(struct net_device *dev,
84 			    struct rtnl_link_stats64 *stats)
85 {
86 	int i;
87 
88 	for_each_possible_cpu(i) {
89 		const struct pcpu_dstats *dstats;
90 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
91 		unsigned int start;
92 
93 		dstats = per_cpu_ptr(dev->dstats, i);
94 		do {
95 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
96 			tbytes = dstats->tx_bytes;
97 			tpkts = dstats->tx_pkts;
98 			tdrops = dstats->tx_drps;
99 			rbytes = dstats->rx_bytes;
100 			rpkts = dstats->rx_pkts;
101 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
102 		stats->tx_bytes += tbytes;
103 		stats->tx_packets += tpkts;
104 		stats->tx_dropped += tdrops;
105 		stats->rx_bytes += rbytes;
106 		stats->rx_packets += rpkts;
107 	}
108 }
109 
110 /* by default VRF devices do not have a qdisc and are expected
111  * to be created with only a single queue.
112  */
113 static bool qdisc_tx_is_default(const struct net_device *dev)
114 {
115 	struct netdev_queue *txq;
116 	struct Qdisc *qdisc;
117 
118 	if (dev->num_tx_queues > 1)
119 		return false;
120 
121 	txq = netdev_get_tx_queue(dev, 0);
122 	qdisc = rcu_access_pointer(txq->qdisc);
123 
124 	return !qdisc->enqueue;
125 }
126 
127 /* Local traffic destined to local address. Reinsert the packet to rx
128  * path, similar to loopback handling.
129  */
130 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
131 			  struct dst_entry *dst)
132 {
133 	int len = skb->len;
134 
135 	skb_orphan(skb);
136 
137 	skb_dst_set(skb, dst);
138 
139 	/* set pkt_type to avoid skb hitting packet taps twice -
140 	 * once on Tx and again in Rx processing
141 	 */
142 	skb->pkt_type = PACKET_LOOPBACK;
143 
144 	skb->protocol = eth_type_trans(skb, dev);
145 
146 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
147 		vrf_rx_stats(dev, len);
148 	else
149 		this_cpu_inc(dev->dstats->rx_drps);
150 
151 	return NETDEV_TX_OK;
152 }
153 
154 #if IS_ENABLED(CONFIG_IPV6)
155 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
156 			     struct sk_buff *skb)
157 {
158 	int err;
159 
160 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
161 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
162 
163 	if (likely(err == 1))
164 		err = dst_output(net, sk, skb);
165 
166 	return err;
167 }
168 
169 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
170 					   struct net_device *dev)
171 {
172 	const struct ipv6hdr *iph = ipv6_hdr(skb);
173 	struct net *net = dev_net(skb->dev);
174 	struct flowi6 fl6 = {
175 		/* needed to match OIF rule */
176 		.flowi6_oif = dev->ifindex,
177 		.flowi6_iif = LOOPBACK_IFINDEX,
178 		.daddr = iph->daddr,
179 		.saddr = iph->saddr,
180 		.flowlabel = ip6_flowinfo(iph),
181 		.flowi6_mark = skb->mark,
182 		.flowi6_proto = iph->nexthdr,
183 		.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
184 	};
185 	int ret = NET_XMIT_DROP;
186 	struct dst_entry *dst;
187 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
188 
189 	dst = ip6_route_output(net, NULL, &fl6);
190 	if (dst == dst_null)
191 		goto err;
192 
193 	skb_dst_drop(skb);
194 
195 	/* if dst.dev is loopback or the VRF device again this is locally
196 	 * originated traffic destined to a local address. Short circuit
197 	 * to Rx path
198 	 */
199 	if (dst->dev == dev)
200 		return vrf_local_xmit(skb, dev, dst);
201 
202 	skb_dst_set(skb, dst);
203 
204 	/* strip the ethernet header added for pass through VRF device */
205 	__skb_pull(skb, skb_network_offset(skb));
206 
207 	ret = vrf_ip6_local_out(net, skb->sk, skb);
208 	if (unlikely(net_xmit_eval(ret)))
209 		dev->stats.tx_errors++;
210 	else
211 		ret = NET_XMIT_SUCCESS;
212 
213 	return ret;
214 err:
215 	vrf_tx_error(dev, skb);
216 	return NET_XMIT_DROP;
217 }
218 #else
219 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
220 					   struct net_device *dev)
221 {
222 	vrf_tx_error(dev, skb);
223 	return NET_XMIT_DROP;
224 }
225 #endif
226 
227 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
228 static int vrf_ip_local_out(struct net *net, struct sock *sk,
229 			    struct sk_buff *skb)
230 {
231 	int err;
232 
233 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
234 		      skb, NULL, skb_dst(skb)->dev, dst_output);
235 	if (likely(err == 1))
236 		err = dst_output(net, sk, skb);
237 
238 	return err;
239 }
240 
241 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
242 					   struct net_device *vrf_dev)
243 {
244 	struct iphdr *ip4h = ip_hdr(skb);
245 	int ret = NET_XMIT_DROP;
246 	struct flowi4 fl4 = {
247 		/* needed to match OIF rule */
248 		.flowi4_oif = vrf_dev->ifindex,
249 		.flowi4_iif = LOOPBACK_IFINDEX,
250 		.flowi4_tos = RT_TOS(ip4h->tos),
251 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
252 		.flowi4_proto = ip4h->protocol,
253 		.daddr = ip4h->daddr,
254 		.saddr = ip4h->saddr,
255 	};
256 	struct net *net = dev_net(vrf_dev);
257 	struct rtable *rt;
258 
259 	rt = ip_route_output_flow(net, &fl4, NULL);
260 	if (IS_ERR(rt))
261 		goto err;
262 
263 	skb_dst_drop(skb);
264 
265 	/* if dst.dev is loopback or the VRF device again this is locally
266 	 * originated traffic destined to a local address. Short circuit
267 	 * to Rx path
268 	 */
269 	if (rt->dst.dev == vrf_dev)
270 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
271 
272 	skb_dst_set(skb, &rt->dst);
273 
274 	/* strip the ethernet header added for pass through VRF device */
275 	__skb_pull(skb, skb_network_offset(skb));
276 
277 	if (!ip4h->saddr) {
278 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
279 					       RT_SCOPE_LINK);
280 	}
281 
282 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
283 	if (unlikely(net_xmit_eval(ret)))
284 		vrf_dev->stats.tx_errors++;
285 	else
286 		ret = NET_XMIT_SUCCESS;
287 
288 out:
289 	return ret;
290 err:
291 	vrf_tx_error(vrf_dev, skb);
292 	goto out;
293 }
294 
295 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
296 {
297 	switch (skb->protocol) {
298 	case htons(ETH_P_IP):
299 		return vrf_process_v4_outbound(skb, dev);
300 	case htons(ETH_P_IPV6):
301 		return vrf_process_v6_outbound(skb, dev);
302 	default:
303 		vrf_tx_error(dev, skb);
304 		return NET_XMIT_DROP;
305 	}
306 }
307 
308 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
309 {
310 	int len = skb->len;
311 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
312 
313 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
314 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
315 
316 		u64_stats_update_begin(&dstats->syncp);
317 		dstats->tx_pkts++;
318 		dstats->tx_bytes += len;
319 		u64_stats_update_end(&dstats->syncp);
320 	} else {
321 		this_cpu_inc(dev->dstats->tx_drps);
322 	}
323 
324 	return ret;
325 }
326 
327 static int vrf_finish_direct(struct net *net, struct sock *sk,
328 			     struct sk_buff *skb)
329 {
330 	struct net_device *vrf_dev = skb->dev;
331 
332 	if (!list_empty(&vrf_dev->ptype_all) &&
333 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
334 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
335 
336 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
337 		eth_zero_addr(eth->h_dest);
338 		eth->h_proto = skb->protocol;
339 
340 		rcu_read_lock_bh();
341 		dev_queue_xmit_nit(skb, vrf_dev);
342 		rcu_read_unlock_bh();
343 
344 		skb_pull(skb, ETH_HLEN);
345 	}
346 
347 	return 1;
348 }
349 
350 #if IS_ENABLED(CONFIG_IPV6)
351 /* modelled after ip6_finish_output2 */
352 static int vrf_finish_output6(struct net *net, struct sock *sk,
353 			      struct sk_buff *skb)
354 {
355 	struct dst_entry *dst = skb_dst(skb);
356 	struct net_device *dev = dst->dev;
357 	struct neighbour *neigh;
358 	struct in6_addr *nexthop;
359 	int ret;
360 
361 	nf_reset(skb);
362 
363 	skb->protocol = htons(ETH_P_IPV6);
364 	skb->dev = dev;
365 
366 	rcu_read_lock_bh();
367 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
368 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
369 	if (unlikely(!neigh))
370 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
371 	if (!IS_ERR(neigh)) {
372 		sock_confirm_neigh(skb, neigh);
373 		ret = neigh_output(neigh, skb);
374 		rcu_read_unlock_bh();
375 		return ret;
376 	}
377 	rcu_read_unlock_bh();
378 
379 	IP6_INC_STATS(dev_net(dst->dev),
380 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
381 	kfree_skb(skb);
382 	return -EINVAL;
383 }
384 
385 /* modelled after ip6_output */
386 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
387 {
388 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
389 			    net, sk, skb, NULL, skb_dst(skb)->dev,
390 			    vrf_finish_output6,
391 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
392 }
393 
394 /* set dst on skb to send packet to us via dev_xmit path. Allows
395  * packet to go through device based features such as qdisc, netfilter
396  * hooks and packet sockets with skb->dev set to vrf device.
397  */
398 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
399 					    struct sk_buff *skb)
400 {
401 	struct net_vrf *vrf = netdev_priv(vrf_dev);
402 	struct dst_entry *dst = NULL;
403 	struct rt6_info *rt6;
404 
405 	rcu_read_lock();
406 
407 	rt6 = rcu_dereference(vrf->rt6);
408 	if (likely(rt6)) {
409 		dst = &rt6->dst;
410 		dst_hold(dst);
411 	}
412 
413 	rcu_read_unlock();
414 
415 	if (unlikely(!dst)) {
416 		vrf_tx_error(vrf_dev, skb);
417 		return NULL;
418 	}
419 
420 	skb_dst_drop(skb);
421 	skb_dst_set(skb, dst);
422 
423 	return skb;
424 }
425 
426 static int vrf_output6_direct(struct net *net, struct sock *sk,
427 			      struct sk_buff *skb)
428 {
429 	skb->protocol = htons(ETH_P_IPV6);
430 
431 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
432 			    net, sk, skb, NULL, skb->dev,
433 			    vrf_finish_direct,
434 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
435 }
436 
437 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
438 					  struct sock *sk,
439 					  struct sk_buff *skb)
440 {
441 	struct net *net = dev_net(vrf_dev);
442 	int err;
443 
444 	skb->dev = vrf_dev;
445 
446 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
447 		      skb, NULL, vrf_dev, vrf_output6_direct);
448 
449 	if (likely(err == 1))
450 		err = vrf_output6_direct(net, sk, skb);
451 
452 	/* reset skb device */
453 	if (likely(err == 1))
454 		nf_reset(skb);
455 	else
456 		skb = NULL;
457 
458 	return skb;
459 }
460 
461 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
462 				   struct sock *sk,
463 				   struct sk_buff *skb)
464 {
465 	/* don't divert link scope packets */
466 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
467 		return skb;
468 
469 	if (qdisc_tx_is_default(vrf_dev))
470 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
471 
472 	return vrf_ip6_out_redirect(vrf_dev, skb);
473 }
474 
475 /* holding rtnl */
476 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
477 {
478 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
479 	struct net *net = dev_net(dev);
480 	struct dst_entry *dst;
481 
482 	RCU_INIT_POINTER(vrf->rt6, NULL);
483 	synchronize_rcu();
484 
485 	/* move dev in dst's to loopback so this VRF device can be deleted
486 	 * - based on dst_ifdown
487 	 */
488 	if (rt6) {
489 		dst = &rt6->dst;
490 		dev_put(dst->dev);
491 		dst->dev = net->loopback_dev;
492 		dev_hold(dst->dev);
493 		dst_release(dst);
494 	}
495 }
496 
497 static int vrf_rt6_create(struct net_device *dev)
498 {
499 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
500 	struct net_vrf *vrf = netdev_priv(dev);
501 	struct net *net = dev_net(dev);
502 	struct rt6_info *rt6;
503 	int rc = -ENOMEM;
504 
505 	/* IPv6 can be CONFIG enabled and then disabled runtime */
506 	if (!ipv6_mod_enabled())
507 		return 0;
508 
509 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
510 	if (!vrf->fib6_table)
511 		goto out;
512 
513 	/* create a dst for routing packets out a VRF device */
514 	rt6 = ip6_dst_alloc(net, dev, flags);
515 	if (!rt6)
516 		goto out;
517 
518 	rt6->dst.output	= vrf_output6;
519 
520 	rcu_assign_pointer(vrf->rt6, rt6);
521 
522 	rc = 0;
523 out:
524 	return rc;
525 }
526 #else
527 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
528 				   struct sock *sk,
529 				   struct sk_buff *skb)
530 {
531 	return skb;
532 }
533 
534 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
535 {
536 }
537 
538 static int vrf_rt6_create(struct net_device *dev)
539 {
540 	return 0;
541 }
542 #endif
543 
544 /* modelled after ip_finish_output2 */
545 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
546 {
547 	struct dst_entry *dst = skb_dst(skb);
548 	struct rtable *rt = (struct rtable *)dst;
549 	struct net_device *dev = dst->dev;
550 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
551 	struct neighbour *neigh;
552 	u32 nexthop;
553 	int ret = -EINVAL;
554 
555 	nf_reset(skb);
556 
557 	/* Be paranoid, rather than too clever. */
558 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
559 		struct sk_buff *skb2;
560 
561 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
562 		if (!skb2) {
563 			ret = -ENOMEM;
564 			goto err;
565 		}
566 		if (skb->sk)
567 			skb_set_owner_w(skb2, skb->sk);
568 
569 		consume_skb(skb);
570 		skb = skb2;
571 	}
572 
573 	rcu_read_lock_bh();
574 
575 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
576 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
577 	if (unlikely(!neigh))
578 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
579 	if (!IS_ERR(neigh)) {
580 		sock_confirm_neigh(skb, neigh);
581 		ret = neigh_output(neigh, skb);
582 		rcu_read_unlock_bh();
583 		return ret;
584 	}
585 
586 	rcu_read_unlock_bh();
587 err:
588 	vrf_tx_error(skb->dev, skb);
589 	return ret;
590 }
591 
592 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
593 {
594 	struct net_device *dev = skb_dst(skb)->dev;
595 
596 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
597 
598 	skb->dev = dev;
599 	skb->protocol = htons(ETH_P_IP);
600 
601 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
602 			    net, sk, skb, NULL, dev,
603 			    vrf_finish_output,
604 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
605 }
606 
607 /* set dst on skb to send packet to us via dev_xmit path. Allows
608  * packet to go through device based features such as qdisc, netfilter
609  * hooks and packet sockets with skb->dev set to vrf device.
610  */
611 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
612 					   struct sk_buff *skb)
613 {
614 	struct net_vrf *vrf = netdev_priv(vrf_dev);
615 	struct dst_entry *dst = NULL;
616 	struct rtable *rth;
617 
618 	rcu_read_lock();
619 
620 	rth = rcu_dereference(vrf->rth);
621 	if (likely(rth)) {
622 		dst = &rth->dst;
623 		dst_hold(dst);
624 	}
625 
626 	rcu_read_unlock();
627 
628 	if (unlikely(!dst)) {
629 		vrf_tx_error(vrf_dev, skb);
630 		return NULL;
631 	}
632 
633 	skb_dst_drop(skb);
634 	skb_dst_set(skb, dst);
635 
636 	return skb;
637 }
638 
639 static int vrf_output_direct(struct net *net, struct sock *sk,
640 			     struct sk_buff *skb)
641 {
642 	skb->protocol = htons(ETH_P_IP);
643 
644 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
645 			    net, sk, skb, NULL, skb->dev,
646 			    vrf_finish_direct,
647 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
648 }
649 
650 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
651 					 struct sock *sk,
652 					 struct sk_buff *skb)
653 {
654 	struct net *net = dev_net(vrf_dev);
655 	int err;
656 
657 	skb->dev = vrf_dev;
658 
659 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
660 		      skb, NULL, vrf_dev, vrf_output_direct);
661 
662 	if (likely(err == 1))
663 		err = vrf_output_direct(net, sk, skb);
664 
665 	/* reset skb device */
666 	if (likely(err == 1))
667 		nf_reset(skb);
668 	else
669 		skb = NULL;
670 
671 	return skb;
672 }
673 
674 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
675 				  struct sock *sk,
676 				  struct sk_buff *skb)
677 {
678 	/* don't divert multicast or local broadcast */
679 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
680 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
681 		return skb;
682 
683 	if (qdisc_tx_is_default(vrf_dev))
684 		return vrf_ip_out_direct(vrf_dev, sk, skb);
685 
686 	return vrf_ip_out_redirect(vrf_dev, skb);
687 }
688 
689 /* called with rcu lock held */
690 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
691 				  struct sock *sk,
692 				  struct sk_buff *skb,
693 				  u16 proto)
694 {
695 	switch (proto) {
696 	case AF_INET:
697 		return vrf_ip_out(vrf_dev, sk, skb);
698 	case AF_INET6:
699 		return vrf_ip6_out(vrf_dev, sk, skb);
700 	}
701 
702 	return skb;
703 }
704 
705 /* holding rtnl */
706 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
707 {
708 	struct rtable *rth = rtnl_dereference(vrf->rth);
709 	struct net *net = dev_net(dev);
710 	struct dst_entry *dst;
711 
712 	RCU_INIT_POINTER(vrf->rth, NULL);
713 	synchronize_rcu();
714 
715 	/* move dev in dst's to loopback so this VRF device can be deleted
716 	 * - based on dst_ifdown
717 	 */
718 	if (rth) {
719 		dst = &rth->dst;
720 		dev_put(dst->dev);
721 		dst->dev = net->loopback_dev;
722 		dev_hold(dst->dev);
723 		dst_release(dst);
724 	}
725 }
726 
727 static int vrf_rtable_create(struct net_device *dev)
728 {
729 	struct net_vrf *vrf = netdev_priv(dev);
730 	struct rtable *rth;
731 
732 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
733 		return -ENOMEM;
734 
735 	/* create a dst for routing packets out through a VRF device */
736 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
737 	if (!rth)
738 		return -ENOMEM;
739 
740 	rth->dst.output	= vrf_output;
741 
742 	rcu_assign_pointer(vrf->rth, rth);
743 
744 	return 0;
745 }
746 
747 /**************************** device handling ********************/
748 
749 /* cycle interface to flush neighbor cache and move routes across tables */
750 static void cycle_netdev(struct net_device *dev,
751 			 struct netlink_ext_ack *extack)
752 {
753 	unsigned int flags = dev->flags;
754 	int ret;
755 
756 	if (!netif_running(dev))
757 		return;
758 
759 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
760 	if (ret >= 0)
761 		ret = dev_change_flags(dev, flags, extack);
762 
763 	if (ret < 0) {
764 		netdev_err(dev,
765 			   "Failed to cycle device %s; route tables might be wrong!\n",
766 			   dev->name);
767 	}
768 }
769 
770 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
771 			    struct netlink_ext_ack *extack)
772 {
773 	int ret;
774 
775 	/* do not allow loopback device to be enslaved to a VRF.
776 	 * The vrf device acts as the loopback for the vrf.
777 	 */
778 	if (port_dev == dev_net(dev)->loopback_dev) {
779 		NL_SET_ERR_MSG(extack,
780 			       "Can not enslave loopback device to a VRF");
781 		return -EOPNOTSUPP;
782 	}
783 
784 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
785 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
786 	if (ret < 0)
787 		goto err;
788 
789 	cycle_netdev(port_dev, extack);
790 
791 	return 0;
792 
793 err:
794 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
795 	return ret;
796 }
797 
798 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
799 			 struct netlink_ext_ack *extack)
800 {
801 	if (netif_is_l3_master(port_dev)) {
802 		NL_SET_ERR_MSG(extack,
803 			       "Can not enslave an L3 master device to a VRF");
804 		return -EINVAL;
805 	}
806 
807 	if (netif_is_l3_slave(port_dev))
808 		return -EINVAL;
809 
810 	return do_vrf_add_slave(dev, port_dev, extack);
811 }
812 
813 /* inverse of do_vrf_add_slave */
814 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
815 {
816 	netdev_upper_dev_unlink(port_dev, dev);
817 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
818 
819 	cycle_netdev(port_dev, NULL);
820 
821 	return 0;
822 }
823 
824 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
825 {
826 	return do_vrf_del_slave(dev, port_dev);
827 }
828 
829 static void vrf_dev_uninit(struct net_device *dev)
830 {
831 	struct net_vrf *vrf = netdev_priv(dev);
832 
833 	vrf_rtable_release(dev, vrf);
834 	vrf_rt6_release(dev, vrf);
835 
836 	free_percpu(dev->dstats);
837 	dev->dstats = NULL;
838 }
839 
840 static int vrf_dev_init(struct net_device *dev)
841 {
842 	struct net_vrf *vrf = netdev_priv(dev);
843 
844 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
845 	if (!dev->dstats)
846 		goto out_nomem;
847 
848 	/* create the default dst which points back to us */
849 	if (vrf_rtable_create(dev) != 0)
850 		goto out_stats;
851 
852 	if (vrf_rt6_create(dev) != 0)
853 		goto out_rth;
854 
855 	dev->flags = IFF_MASTER | IFF_NOARP;
856 
857 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
858 	dev->mtu = 64 * 1024;
859 
860 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
861 	dev->operstate = IF_OPER_UP;
862 	netdev_lockdep_set_classes(dev);
863 	return 0;
864 
865 out_rth:
866 	vrf_rtable_release(dev, vrf);
867 out_stats:
868 	free_percpu(dev->dstats);
869 	dev->dstats = NULL;
870 out_nomem:
871 	return -ENOMEM;
872 }
873 
874 static const struct net_device_ops vrf_netdev_ops = {
875 	.ndo_init		= vrf_dev_init,
876 	.ndo_uninit		= vrf_dev_uninit,
877 	.ndo_start_xmit		= vrf_xmit,
878 	.ndo_get_stats64	= vrf_get_stats64,
879 	.ndo_add_slave		= vrf_add_slave,
880 	.ndo_del_slave		= vrf_del_slave,
881 };
882 
883 static u32 vrf_fib_table(const struct net_device *dev)
884 {
885 	struct net_vrf *vrf = netdev_priv(dev);
886 
887 	return vrf->tb_id;
888 }
889 
890 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
891 {
892 	kfree_skb(skb);
893 	return 0;
894 }
895 
896 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
897 				      struct sk_buff *skb,
898 				      struct net_device *dev)
899 {
900 	struct net *net = dev_net(dev);
901 
902 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
903 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
904 
905 	return skb;
906 }
907 
908 #if IS_ENABLED(CONFIG_IPV6)
909 /* neighbor handling is done with actual device; do not want
910  * to flip skb->dev for those ndisc packets. This really fails
911  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
912  * a start.
913  */
914 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
915 {
916 	const struct ipv6hdr *iph = ipv6_hdr(skb);
917 	bool rc = false;
918 
919 	if (iph->nexthdr == NEXTHDR_ICMP) {
920 		const struct icmp6hdr *icmph;
921 		struct icmp6hdr _icmph;
922 
923 		icmph = skb_header_pointer(skb, sizeof(*iph),
924 					   sizeof(_icmph), &_icmph);
925 		if (!icmph)
926 			goto out;
927 
928 		switch (icmph->icmp6_type) {
929 		case NDISC_ROUTER_SOLICITATION:
930 		case NDISC_ROUTER_ADVERTISEMENT:
931 		case NDISC_NEIGHBOUR_SOLICITATION:
932 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
933 		case NDISC_REDIRECT:
934 			rc = true;
935 			break;
936 		}
937 	}
938 
939 out:
940 	return rc;
941 }
942 
943 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
944 					     const struct net_device *dev,
945 					     struct flowi6 *fl6,
946 					     int ifindex,
947 					     const struct sk_buff *skb,
948 					     int flags)
949 {
950 	struct net_vrf *vrf = netdev_priv(dev);
951 
952 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
953 }
954 
955 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
956 			      int ifindex)
957 {
958 	const struct ipv6hdr *iph = ipv6_hdr(skb);
959 	struct flowi6 fl6 = {
960 		.flowi6_iif     = ifindex,
961 		.flowi6_mark    = skb->mark,
962 		.flowi6_proto   = iph->nexthdr,
963 		.daddr          = iph->daddr,
964 		.saddr          = iph->saddr,
965 		.flowlabel      = ip6_flowinfo(iph),
966 	};
967 	struct net *net = dev_net(vrf_dev);
968 	struct rt6_info *rt6;
969 
970 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
971 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
972 	if (unlikely(!rt6))
973 		return;
974 
975 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
976 		return;
977 
978 	skb_dst_set(skb, &rt6->dst);
979 }
980 
981 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
982 				   struct sk_buff *skb)
983 {
984 	int orig_iif = skb->skb_iif;
985 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
986 	bool is_ndisc = ipv6_ndisc_frame(skb);
987 
988 	/* loopback, multicast & non-ND link-local traffic; do not push through
989 	 * packet taps again. Reset pkt_type for upper layers to process skb
990 	 */
991 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
992 		skb->dev = vrf_dev;
993 		skb->skb_iif = vrf_dev->ifindex;
994 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
995 		if (skb->pkt_type == PACKET_LOOPBACK)
996 			skb->pkt_type = PACKET_HOST;
997 		goto out;
998 	}
999 
1000 	/* if packet is NDISC then keep the ingress interface */
1001 	if (!is_ndisc) {
1002 		vrf_rx_stats(vrf_dev, skb->len);
1003 		skb->dev = vrf_dev;
1004 		skb->skb_iif = vrf_dev->ifindex;
1005 
1006 		if (!list_empty(&vrf_dev->ptype_all)) {
1007 			skb_push(skb, skb->mac_len);
1008 			dev_queue_xmit_nit(skb, vrf_dev);
1009 			skb_pull(skb, skb->mac_len);
1010 		}
1011 
1012 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1013 	}
1014 
1015 	if (need_strict)
1016 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1017 
1018 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1019 out:
1020 	return skb;
1021 }
1022 
1023 #else
1024 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1025 				   struct sk_buff *skb)
1026 {
1027 	return skb;
1028 }
1029 #endif
1030 
1031 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1032 				  struct sk_buff *skb)
1033 {
1034 	skb->dev = vrf_dev;
1035 	skb->skb_iif = vrf_dev->ifindex;
1036 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1037 
1038 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1039 		goto out;
1040 
1041 	/* loopback traffic; do not push through packet taps again.
1042 	 * Reset pkt_type for upper layers to process skb
1043 	 */
1044 	if (skb->pkt_type == PACKET_LOOPBACK) {
1045 		skb->pkt_type = PACKET_HOST;
1046 		goto out;
1047 	}
1048 
1049 	vrf_rx_stats(vrf_dev, skb->len);
1050 
1051 	if (!list_empty(&vrf_dev->ptype_all)) {
1052 		skb_push(skb, skb->mac_len);
1053 		dev_queue_xmit_nit(skb, vrf_dev);
1054 		skb_pull(skb, skb->mac_len);
1055 	}
1056 
1057 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1058 out:
1059 	return skb;
1060 }
1061 
1062 /* called with rcu lock held */
1063 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1064 				  struct sk_buff *skb,
1065 				  u16 proto)
1066 {
1067 	switch (proto) {
1068 	case AF_INET:
1069 		return vrf_ip_rcv(vrf_dev, skb);
1070 	case AF_INET6:
1071 		return vrf_ip6_rcv(vrf_dev, skb);
1072 	}
1073 
1074 	return skb;
1075 }
1076 
1077 #if IS_ENABLED(CONFIG_IPV6)
1078 /* send to link-local or multicast address via interface enslaved to
1079  * VRF device. Force lookup to VRF table without changing flow struct
1080  */
1081 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1082 					      struct flowi6 *fl6)
1083 {
1084 	struct net *net = dev_net(dev);
1085 	int flags = RT6_LOOKUP_F_IFACE;
1086 	struct dst_entry *dst = NULL;
1087 	struct rt6_info *rt;
1088 
1089 	/* VRF device does not have a link-local address and
1090 	 * sending packets to link-local or mcast addresses over
1091 	 * a VRF device does not make sense
1092 	 */
1093 	if (fl6->flowi6_oif == dev->ifindex) {
1094 		dst = &net->ipv6.ip6_null_entry->dst;
1095 		dst_hold(dst);
1096 		return dst;
1097 	}
1098 
1099 	if (!ipv6_addr_any(&fl6->saddr))
1100 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1101 
1102 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1103 	if (rt)
1104 		dst = &rt->dst;
1105 
1106 	return dst;
1107 }
1108 #endif
1109 
1110 static const struct l3mdev_ops vrf_l3mdev_ops = {
1111 	.l3mdev_fib_table	= vrf_fib_table,
1112 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1113 	.l3mdev_l3_out		= vrf_l3_out,
1114 #if IS_ENABLED(CONFIG_IPV6)
1115 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1116 #endif
1117 };
1118 
1119 static void vrf_get_drvinfo(struct net_device *dev,
1120 			    struct ethtool_drvinfo *info)
1121 {
1122 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1123 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1124 }
1125 
1126 static const struct ethtool_ops vrf_ethtool_ops = {
1127 	.get_drvinfo	= vrf_get_drvinfo,
1128 };
1129 
1130 static inline size_t vrf_fib_rule_nl_size(void)
1131 {
1132 	size_t sz;
1133 
1134 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1135 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1136 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1137 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1138 
1139 	return sz;
1140 }
1141 
1142 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1143 {
1144 	struct fib_rule_hdr *frh;
1145 	struct nlmsghdr *nlh;
1146 	struct sk_buff *skb;
1147 	int err;
1148 
1149 	if (family == AF_INET6 && !ipv6_mod_enabled())
1150 		return 0;
1151 
1152 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1153 	if (!skb)
1154 		return -ENOMEM;
1155 
1156 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1157 	if (!nlh)
1158 		goto nla_put_failure;
1159 
1160 	/* rule only needs to appear once */
1161 	nlh->nlmsg_flags |= NLM_F_EXCL;
1162 
1163 	frh = nlmsg_data(nlh);
1164 	memset(frh, 0, sizeof(*frh));
1165 	frh->family = family;
1166 	frh->action = FR_ACT_TO_TBL;
1167 
1168 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1169 		goto nla_put_failure;
1170 
1171 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1172 		goto nla_put_failure;
1173 
1174 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1175 		goto nla_put_failure;
1176 
1177 	nlmsg_end(skb, nlh);
1178 
1179 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1180 	skb->sk = dev_net(dev)->rtnl;
1181 	if (add_it) {
1182 		err = fib_nl_newrule(skb, nlh, NULL);
1183 		if (err == -EEXIST)
1184 			err = 0;
1185 	} else {
1186 		err = fib_nl_delrule(skb, nlh, NULL);
1187 		if (err == -ENOENT)
1188 			err = 0;
1189 	}
1190 	nlmsg_free(skb);
1191 
1192 	return err;
1193 
1194 nla_put_failure:
1195 	nlmsg_free(skb);
1196 
1197 	return -EMSGSIZE;
1198 }
1199 
1200 static int vrf_add_fib_rules(const struct net_device *dev)
1201 {
1202 	int err;
1203 
1204 	err = vrf_fib_rule(dev, AF_INET,  true);
1205 	if (err < 0)
1206 		goto out_err;
1207 
1208 	err = vrf_fib_rule(dev, AF_INET6, true);
1209 	if (err < 0)
1210 		goto ipv6_err;
1211 
1212 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1213 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1214 	if (err < 0)
1215 		goto ipmr_err;
1216 #endif
1217 
1218 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1219 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1220 	if (err < 0)
1221 		goto ip6mr_err;
1222 #endif
1223 
1224 	return 0;
1225 
1226 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1227 ip6mr_err:
1228 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1229 #endif
1230 
1231 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1232 ipmr_err:
1233 	vrf_fib_rule(dev, AF_INET6,  false);
1234 #endif
1235 
1236 ipv6_err:
1237 	vrf_fib_rule(dev, AF_INET,  false);
1238 
1239 out_err:
1240 	netdev_err(dev, "Failed to add FIB rules.\n");
1241 	return err;
1242 }
1243 
1244 static void vrf_setup(struct net_device *dev)
1245 {
1246 	ether_setup(dev);
1247 
1248 	/* Initialize the device structure. */
1249 	dev->netdev_ops = &vrf_netdev_ops;
1250 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1251 	dev->ethtool_ops = &vrf_ethtool_ops;
1252 	dev->needs_free_netdev = true;
1253 
1254 	/* Fill in device structure with ethernet-generic values. */
1255 	eth_hw_addr_random(dev);
1256 
1257 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1258 	dev->features |= NETIF_F_LLTX;
1259 
1260 	/* don't allow vrf devices to change network namespaces. */
1261 	dev->features |= NETIF_F_NETNS_LOCAL;
1262 
1263 	/* does not make sense for a VLAN to be added to a vrf device */
1264 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1265 
1266 	/* enable offload features */
1267 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1268 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1269 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1270 
1271 	dev->hw_features = dev->features;
1272 	dev->hw_enc_features = dev->features;
1273 
1274 	/* default to no qdisc; user can add if desired */
1275 	dev->priv_flags |= IFF_NO_QUEUE;
1276 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1277 
1278 	/* VRF devices do not care about MTU, but if the MTU is set
1279 	 * too low then the ipv4 and ipv6 protocols are disabled
1280 	 * which breaks networking.
1281 	 */
1282 	dev->min_mtu = IPV6_MIN_MTU;
1283 	dev->max_mtu = ETH_MAX_MTU;
1284 }
1285 
1286 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1287 			struct netlink_ext_ack *extack)
1288 {
1289 	if (tb[IFLA_ADDRESS]) {
1290 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1291 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1292 			return -EINVAL;
1293 		}
1294 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1295 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1296 			return -EADDRNOTAVAIL;
1297 		}
1298 	}
1299 	return 0;
1300 }
1301 
1302 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1303 {
1304 	struct net_device *port_dev;
1305 	struct list_head *iter;
1306 
1307 	netdev_for_each_lower_dev(dev, port_dev, iter)
1308 		vrf_del_slave(dev, port_dev);
1309 
1310 	unregister_netdevice_queue(dev, head);
1311 }
1312 
1313 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1314 		       struct nlattr *tb[], struct nlattr *data[],
1315 		       struct netlink_ext_ack *extack)
1316 {
1317 	struct net_vrf *vrf = netdev_priv(dev);
1318 	bool *add_fib_rules;
1319 	struct net *net;
1320 	int err;
1321 
1322 	if (!data || !data[IFLA_VRF_TABLE]) {
1323 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1324 		return -EINVAL;
1325 	}
1326 
1327 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1328 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1329 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1330 				    "Invalid VRF table id");
1331 		return -EINVAL;
1332 	}
1333 
1334 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1335 
1336 	err = register_netdevice(dev);
1337 	if (err)
1338 		goto out;
1339 
1340 	net = dev_net(dev);
1341 	add_fib_rules = net_generic(net, vrf_net_id);
1342 	if (*add_fib_rules) {
1343 		err = vrf_add_fib_rules(dev);
1344 		if (err) {
1345 			unregister_netdevice(dev);
1346 			goto out;
1347 		}
1348 		*add_fib_rules = false;
1349 	}
1350 
1351 out:
1352 	return err;
1353 }
1354 
1355 static size_t vrf_nl_getsize(const struct net_device *dev)
1356 {
1357 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1358 }
1359 
1360 static int vrf_fillinfo(struct sk_buff *skb,
1361 			const struct net_device *dev)
1362 {
1363 	struct net_vrf *vrf = netdev_priv(dev);
1364 
1365 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1366 }
1367 
1368 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1369 				 const struct net_device *slave_dev)
1370 {
1371 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1372 }
1373 
1374 static int vrf_fill_slave_info(struct sk_buff *skb,
1375 			       const struct net_device *vrf_dev,
1376 			       const struct net_device *slave_dev)
1377 {
1378 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1379 
1380 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1381 		return -EMSGSIZE;
1382 
1383 	return 0;
1384 }
1385 
1386 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1387 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1388 };
1389 
1390 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1391 	.kind		= DRV_NAME,
1392 	.priv_size	= sizeof(struct net_vrf),
1393 
1394 	.get_size	= vrf_nl_getsize,
1395 	.policy		= vrf_nl_policy,
1396 	.validate	= vrf_validate,
1397 	.fill_info	= vrf_fillinfo,
1398 
1399 	.get_slave_size  = vrf_get_slave_size,
1400 	.fill_slave_info = vrf_fill_slave_info,
1401 
1402 	.newlink	= vrf_newlink,
1403 	.dellink	= vrf_dellink,
1404 	.setup		= vrf_setup,
1405 	.maxtype	= IFLA_VRF_MAX,
1406 };
1407 
1408 static int vrf_device_event(struct notifier_block *unused,
1409 			    unsigned long event, void *ptr)
1410 {
1411 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1412 
1413 	/* only care about unregister events to drop slave references */
1414 	if (event == NETDEV_UNREGISTER) {
1415 		struct net_device *vrf_dev;
1416 
1417 		if (!netif_is_l3_slave(dev))
1418 			goto out;
1419 
1420 		vrf_dev = netdev_master_upper_dev_get(dev);
1421 		vrf_del_slave(vrf_dev, dev);
1422 	}
1423 out:
1424 	return NOTIFY_DONE;
1425 }
1426 
1427 static struct notifier_block vrf_notifier_block __read_mostly = {
1428 	.notifier_call = vrf_device_event,
1429 };
1430 
1431 /* Initialize per network namespace state */
1432 static int __net_init vrf_netns_init(struct net *net)
1433 {
1434 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1435 
1436 	*add_fib_rules = true;
1437 
1438 	return 0;
1439 }
1440 
1441 static struct pernet_operations vrf_net_ops __net_initdata = {
1442 	.init = vrf_netns_init,
1443 	.id   = &vrf_net_id,
1444 	.size = sizeof(bool),
1445 };
1446 
1447 static int __init vrf_init_module(void)
1448 {
1449 	int rc;
1450 
1451 	register_netdevice_notifier(&vrf_notifier_block);
1452 
1453 	rc = register_pernet_subsys(&vrf_net_ops);
1454 	if (rc < 0)
1455 		goto error;
1456 
1457 	rc = rtnl_link_register(&vrf_link_ops);
1458 	if (rc < 0) {
1459 		unregister_pernet_subsys(&vrf_net_ops);
1460 		goto error;
1461 	}
1462 
1463 	return 0;
1464 
1465 error:
1466 	unregister_netdevice_notifier(&vrf_notifier_block);
1467 	return rc;
1468 }
1469 
1470 module_init(vrf_init_module);
1471 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1472 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1473 MODULE_LICENSE("GPL");
1474 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1475 MODULE_VERSION(DRV_VERSION);
1476