xref: /openbmc/linux/drivers/net/vrf.c (revision 6c33a6f4)
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 
25 #include <linux/inetdevice.h>
26 #include <net/arp.h>
27 #include <net/ip.h>
28 #include <net/ip_fib.h>
29 #include <net/ip6_fib.h>
30 #include <net/ip6_route.h>
31 #include <net/route.h>
32 #include <net/addrconf.h>
33 #include <net/l3mdev.h>
34 #include <net/fib_rules.h>
35 #include <net/netns/generic.h>
36 
37 #define DRV_NAME	"vrf"
38 #define DRV_VERSION	"1.0"
39 
40 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
41 
42 static unsigned int vrf_net_id;
43 
44 struct net_vrf {
45 	struct rtable __rcu	*rth;
46 	struct rt6_info	__rcu	*rt6;
47 #if IS_ENABLED(CONFIG_IPV6)
48 	struct fib6_table	*fib6_table;
49 #endif
50 	u32                     tb_id;
51 };
52 
53 struct pcpu_dstats {
54 	u64			tx_pkts;
55 	u64			tx_bytes;
56 	u64			tx_drps;
57 	u64			rx_pkts;
58 	u64			rx_bytes;
59 	u64			rx_drps;
60 	struct u64_stats_sync	syncp;
61 };
62 
63 static void vrf_rx_stats(struct net_device *dev, int len)
64 {
65 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
66 
67 	u64_stats_update_begin(&dstats->syncp);
68 	dstats->rx_pkts++;
69 	dstats->rx_bytes += len;
70 	u64_stats_update_end(&dstats->syncp);
71 }
72 
73 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
74 {
75 	vrf_dev->stats.tx_errors++;
76 	kfree_skb(skb);
77 }
78 
79 static void vrf_get_stats64(struct net_device *dev,
80 			    struct rtnl_link_stats64 *stats)
81 {
82 	int i;
83 
84 	for_each_possible_cpu(i) {
85 		const struct pcpu_dstats *dstats;
86 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
87 		unsigned int start;
88 
89 		dstats = per_cpu_ptr(dev->dstats, i);
90 		do {
91 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
92 			tbytes = dstats->tx_bytes;
93 			tpkts = dstats->tx_pkts;
94 			tdrops = dstats->tx_drps;
95 			rbytes = dstats->rx_bytes;
96 			rpkts = dstats->rx_pkts;
97 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
98 		stats->tx_bytes += tbytes;
99 		stats->tx_packets += tpkts;
100 		stats->tx_dropped += tdrops;
101 		stats->rx_bytes += rbytes;
102 		stats->rx_packets += rpkts;
103 	}
104 }
105 
106 /* by default VRF devices do not have a qdisc and are expected
107  * to be created with only a single queue.
108  */
109 static bool qdisc_tx_is_default(const struct net_device *dev)
110 {
111 	struct netdev_queue *txq;
112 	struct Qdisc *qdisc;
113 
114 	if (dev->num_tx_queues > 1)
115 		return false;
116 
117 	txq = netdev_get_tx_queue(dev, 0);
118 	qdisc = rcu_access_pointer(txq->qdisc);
119 
120 	return !qdisc->enqueue;
121 }
122 
123 /* Local traffic destined to local address. Reinsert the packet to rx
124  * path, similar to loopback handling.
125  */
126 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
127 			  struct dst_entry *dst)
128 {
129 	int len = skb->len;
130 
131 	skb_orphan(skb);
132 
133 	skb_dst_set(skb, dst);
134 
135 	/* set pkt_type to avoid skb hitting packet taps twice -
136 	 * once on Tx and again in Rx processing
137 	 */
138 	skb->pkt_type = PACKET_LOOPBACK;
139 
140 	skb->protocol = eth_type_trans(skb, dev);
141 
142 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
143 		vrf_rx_stats(dev, len);
144 	else
145 		this_cpu_inc(dev->dstats->rx_drps);
146 
147 	return NETDEV_TX_OK;
148 }
149 
150 #if IS_ENABLED(CONFIG_IPV6)
151 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
152 			     struct sk_buff *skb)
153 {
154 	int err;
155 
156 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
157 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
158 
159 	if (likely(err == 1))
160 		err = dst_output(net, sk, skb);
161 
162 	return err;
163 }
164 
165 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
166 					   struct net_device *dev)
167 {
168 	const struct ipv6hdr *iph;
169 	struct net *net = dev_net(skb->dev);
170 	struct flowi6 fl6;
171 	int ret = NET_XMIT_DROP;
172 	struct dst_entry *dst;
173 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
174 
175 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
176 		goto err;
177 
178 	iph = ipv6_hdr(skb);
179 
180 	memset(&fl6, 0, sizeof(fl6));
181 	/* needed to match OIF rule */
182 	fl6.flowi6_oif = dev->ifindex;
183 	fl6.flowi6_iif = LOOPBACK_IFINDEX;
184 	fl6.daddr = iph->daddr;
185 	fl6.saddr = iph->saddr;
186 	fl6.flowlabel = ip6_flowinfo(iph);
187 	fl6.flowi6_mark = skb->mark;
188 	fl6.flowi6_proto = iph->nexthdr;
189 	fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
190 
191 	dst = ip6_route_output(net, NULL, &fl6);
192 	if (dst == dst_null)
193 		goto err;
194 
195 	skb_dst_drop(skb);
196 
197 	/* if dst.dev is loopback or the VRF device again this is locally
198 	 * originated traffic destined to a local address. Short circuit
199 	 * to Rx path
200 	 */
201 	if (dst->dev == dev)
202 		return vrf_local_xmit(skb, dev, dst);
203 
204 	skb_dst_set(skb, dst);
205 
206 	/* strip the ethernet header added for pass through VRF device */
207 	__skb_pull(skb, skb_network_offset(skb));
208 
209 	ret = vrf_ip6_local_out(net, skb->sk, skb);
210 	if (unlikely(net_xmit_eval(ret)))
211 		dev->stats.tx_errors++;
212 	else
213 		ret = NET_XMIT_SUCCESS;
214 
215 	return ret;
216 err:
217 	vrf_tx_error(dev, skb);
218 	return NET_XMIT_DROP;
219 }
220 #else
221 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
222 					   struct net_device *dev)
223 {
224 	vrf_tx_error(dev, skb);
225 	return NET_XMIT_DROP;
226 }
227 #endif
228 
229 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
230 static int vrf_ip_local_out(struct net *net, struct sock *sk,
231 			    struct sk_buff *skb)
232 {
233 	int err;
234 
235 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
236 		      skb, NULL, skb_dst(skb)->dev, dst_output);
237 	if (likely(err == 1))
238 		err = dst_output(net, sk, skb);
239 
240 	return err;
241 }
242 
243 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
244 					   struct net_device *vrf_dev)
245 {
246 	struct iphdr *ip4h;
247 	int ret = NET_XMIT_DROP;
248 	struct flowi4 fl4;
249 	struct net *net = dev_net(vrf_dev);
250 	struct rtable *rt;
251 
252 	if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
253 		goto err;
254 
255 	ip4h = ip_hdr(skb);
256 
257 	memset(&fl4, 0, sizeof(fl4));
258 	/* needed to match OIF rule */
259 	fl4.flowi4_oif = vrf_dev->ifindex;
260 	fl4.flowi4_iif = LOOPBACK_IFINDEX;
261 	fl4.flowi4_tos = RT_TOS(ip4h->tos);
262 	fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
263 	fl4.flowi4_proto = ip4h->protocol;
264 	fl4.daddr = ip4h->daddr;
265 	fl4.saddr = ip4h->saddr;
266 
267 	rt = ip_route_output_flow(net, &fl4, NULL);
268 	if (IS_ERR(rt))
269 		goto err;
270 
271 	skb_dst_drop(skb);
272 
273 	/* if dst.dev is loopback or the VRF device again this is locally
274 	 * originated traffic destined to a local address. Short circuit
275 	 * to Rx path
276 	 */
277 	if (rt->dst.dev == vrf_dev)
278 		return vrf_local_xmit(skb, vrf_dev, &rt->dst);
279 
280 	skb_dst_set(skb, &rt->dst);
281 
282 	/* strip the ethernet header added for pass through VRF device */
283 	__skb_pull(skb, skb_network_offset(skb));
284 
285 	if (!ip4h->saddr) {
286 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
287 					       RT_SCOPE_LINK);
288 	}
289 
290 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
291 	if (unlikely(net_xmit_eval(ret)))
292 		vrf_dev->stats.tx_errors++;
293 	else
294 		ret = NET_XMIT_SUCCESS;
295 
296 out:
297 	return ret;
298 err:
299 	vrf_tx_error(vrf_dev, skb);
300 	goto out;
301 }
302 
303 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
304 {
305 	switch (skb->protocol) {
306 	case htons(ETH_P_IP):
307 		return vrf_process_v4_outbound(skb, dev);
308 	case htons(ETH_P_IPV6):
309 		return vrf_process_v6_outbound(skb, dev);
310 	default:
311 		vrf_tx_error(dev, skb);
312 		return NET_XMIT_DROP;
313 	}
314 }
315 
316 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
317 {
318 	int len = skb->len;
319 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
320 
321 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
322 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
323 
324 		u64_stats_update_begin(&dstats->syncp);
325 		dstats->tx_pkts++;
326 		dstats->tx_bytes += len;
327 		u64_stats_update_end(&dstats->syncp);
328 	} else {
329 		this_cpu_inc(dev->dstats->tx_drps);
330 	}
331 
332 	return ret;
333 }
334 
335 static int vrf_finish_direct(struct net *net, struct sock *sk,
336 			     struct sk_buff *skb)
337 {
338 	struct net_device *vrf_dev = skb->dev;
339 
340 	if (!list_empty(&vrf_dev->ptype_all) &&
341 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
342 		struct ethhdr *eth = skb_push(skb, ETH_HLEN);
343 
344 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
345 		eth_zero_addr(eth->h_dest);
346 		eth->h_proto = skb->protocol;
347 
348 		rcu_read_lock_bh();
349 		dev_queue_xmit_nit(skb, vrf_dev);
350 		rcu_read_unlock_bh();
351 
352 		skb_pull(skb, ETH_HLEN);
353 	}
354 
355 	return 1;
356 }
357 
358 #if IS_ENABLED(CONFIG_IPV6)
359 /* modelled after ip6_finish_output2 */
360 static int vrf_finish_output6(struct net *net, struct sock *sk,
361 			      struct sk_buff *skb)
362 {
363 	struct dst_entry *dst = skb_dst(skb);
364 	struct net_device *dev = dst->dev;
365 	const struct in6_addr *nexthop;
366 	struct neighbour *neigh;
367 	int ret;
368 
369 	nf_reset_ct(skb);
370 
371 	skb->protocol = htons(ETH_P_IPV6);
372 	skb->dev = dev;
373 
374 	rcu_read_lock_bh();
375 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
376 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
377 	if (unlikely(!neigh))
378 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
379 	if (!IS_ERR(neigh)) {
380 		sock_confirm_neigh(skb, neigh);
381 		ret = neigh_output(neigh, skb, false);
382 		rcu_read_unlock_bh();
383 		return ret;
384 	}
385 	rcu_read_unlock_bh();
386 
387 	IP6_INC_STATS(dev_net(dst->dev),
388 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
389 	kfree_skb(skb);
390 	return -EINVAL;
391 }
392 
393 /* modelled after ip6_output */
394 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
395 {
396 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
397 			    net, sk, skb, NULL, skb_dst(skb)->dev,
398 			    vrf_finish_output6,
399 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
400 }
401 
402 /* set dst on skb to send packet to us via dev_xmit path. Allows
403  * packet to go through device based features such as qdisc, netfilter
404  * hooks and packet sockets with skb->dev set to vrf device.
405  */
406 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
407 					    struct sk_buff *skb)
408 {
409 	struct net_vrf *vrf = netdev_priv(vrf_dev);
410 	struct dst_entry *dst = NULL;
411 	struct rt6_info *rt6;
412 
413 	rcu_read_lock();
414 
415 	rt6 = rcu_dereference(vrf->rt6);
416 	if (likely(rt6)) {
417 		dst = &rt6->dst;
418 		dst_hold(dst);
419 	}
420 
421 	rcu_read_unlock();
422 
423 	if (unlikely(!dst)) {
424 		vrf_tx_error(vrf_dev, skb);
425 		return NULL;
426 	}
427 
428 	skb_dst_drop(skb);
429 	skb_dst_set(skb, dst);
430 
431 	return skb;
432 }
433 
434 static int vrf_output6_direct(struct net *net, struct sock *sk,
435 			      struct sk_buff *skb)
436 {
437 	skb->protocol = htons(ETH_P_IPV6);
438 
439 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
440 			    net, sk, skb, NULL, skb->dev,
441 			    vrf_finish_direct,
442 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
443 }
444 
445 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
446 					  struct sock *sk,
447 					  struct sk_buff *skb)
448 {
449 	struct net *net = dev_net(vrf_dev);
450 	int err;
451 
452 	skb->dev = vrf_dev;
453 
454 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
455 		      skb, NULL, vrf_dev, vrf_output6_direct);
456 
457 	if (likely(err == 1))
458 		err = vrf_output6_direct(net, sk, skb);
459 
460 	/* reset skb device */
461 	if (likely(err == 1))
462 		nf_reset_ct(skb);
463 	else
464 		skb = NULL;
465 
466 	return skb;
467 }
468 
469 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
470 				   struct sock *sk,
471 				   struct sk_buff *skb)
472 {
473 	/* don't divert link scope packets */
474 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
475 		return skb;
476 
477 	if (qdisc_tx_is_default(vrf_dev))
478 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
479 
480 	return vrf_ip6_out_redirect(vrf_dev, skb);
481 }
482 
483 /* holding rtnl */
484 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
485 {
486 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
487 	struct net *net = dev_net(dev);
488 	struct dst_entry *dst;
489 
490 	RCU_INIT_POINTER(vrf->rt6, NULL);
491 	synchronize_rcu();
492 
493 	/* move dev in dst's to loopback so this VRF device can be deleted
494 	 * - based on dst_ifdown
495 	 */
496 	if (rt6) {
497 		dst = &rt6->dst;
498 		dev_put(dst->dev);
499 		dst->dev = net->loopback_dev;
500 		dev_hold(dst->dev);
501 		dst_release(dst);
502 	}
503 }
504 
505 static int vrf_rt6_create(struct net_device *dev)
506 {
507 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
508 	struct net_vrf *vrf = netdev_priv(dev);
509 	struct net *net = dev_net(dev);
510 	struct rt6_info *rt6;
511 	int rc = -ENOMEM;
512 
513 	/* IPv6 can be CONFIG enabled and then disabled runtime */
514 	if (!ipv6_mod_enabled())
515 		return 0;
516 
517 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
518 	if (!vrf->fib6_table)
519 		goto out;
520 
521 	/* create a dst for routing packets out a VRF device */
522 	rt6 = ip6_dst_alloc(net, dev, flags);
523 	if (!rt6)
524 		goto out;
525 
526 	rt6->dst.output	= vrf_output6;
527 
528 	rcu_assign_pointer(vrf->rt6, rt6);
529 
530 	rc = 0;
531 out:
532 	return rc;
533 }
534 #else
535 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
536 				   struct sock *sk,
537 				   struct sk_buff *skb)
538 {
539 	return skb;
540 }
541 
542 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
543 {
544 }
545 
546 static int vrf_rt6_create(struct net_device *dev)
547 {
548 	return 0;
549 }
550 #endif
551 
552 /* modelled after ip_finish_output2 */
553 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
554 {
555 	struct dst_entry *dst = skb_dst(skb);
556 	struct rtable *rt = (struct rtable *)dst;
557 	struct net_device *dev = dst->dev;
558 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
559 	struct neighbour *neigh;
560 	bool is_v6gw = false;
561 	int ret = -EINVAL;
562 
563 	nf_reset_ct(skb);
564 
565 	/* Be paranoid, rather than too clever. */
566 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
567 		struct sk_buff *skb2;
568 
569 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
570 		if (!skb2) {
571 			ret = -ENOMEM;
572 			goto err;
573 		}
574 		if (skb->sk)
575 			skb_set_owner_w(skb2, skb->sk);
576 
577 		consume_skb(skb);
578 		skb = skb2;
579 	}
580 
581 	rcu_read_lock_bh();
582 
583 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
584 	if (!IS_ERR(neigh)) {
585 		sock_confirm_neigh(skb, neigh);
586 		/* if crossing protocols, can not use the cached header */
587 		ret = neigh_output(neigh, skb, is_v6gw);
588 		rcu_read_unlock_bh();
589 		return ret;
590 	}
591 
592 	rcu_read_unlock_bh();
593 err:
594 	vrf_tx_error(skb->dev, skb);
595 	return ret;
596 }
597 
598 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
599 {
600 	struct net_device *dev = skb_dst(skb)->dev;
601 
602 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
603 
604 	skb->dev = dev;
605 	skb->protocol = htons(ETH_P_IP);
606 
607 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
608 			    net, sk, skb, NULL, dev,
609 			    vrf_finish_output,
610 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
611 }
612 
613 /* set dst on skb to send packet to us via dev_xmit path. Allows
614  * packet to go through device based features such as qdisc, netfilter
615  * hooks and packet sockets with skb->dev set to vrf device.
616  */
617 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
618 					   struct sk_buff *skb)
619 {
620 	struct net_vrf *vrf = netdev_priv(vrf_dev);
621 	struct dst_entry *dst = NULL;
622 	struct rtable *rth;
623 
624 	rcu_read_lock();
625 
626 	rth = rcu_dereference(vrf->rth);
627 	if (likely(rth)) {
628 		dst = &rth->dst;
629 		dst_hold(dst);
630 	}
631 
632 	rcu_read_unlock();
633 
634 	if (unlikely(!dst)) {
635 		vrf_tx_error(vrf_dev, skb);
636 		return NULL;
637 	}
638 
639 	skb_dst_drop(skb);
640 	skb_dst_set(skb, dst);
641 
642 	return skb;
643 }
644 
645 static int vrf_output_direct(struct net *net, struct sock *sk,
646 			     struct sk_buff *skb)
647 {
648 	skb->protocol = htons(ETH_P_IP);
649 
650 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
651 			    net, sk, skb, NULL, skb->dev,
652 			    vrf_finish_direct,
653 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
654 }
655 
656 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
657 					 struct sock *sk,
658 					 struct sk_buff *skb)
659 {
660 	struct net *net = dev_net(vrf_dev);
661 	int err;
662 
663 	skb->dev = vrf_dev;
664 
665 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
666 		      skb, NULL, vrf_dev, vrf_output_direct);
667 
668 	if (likely(err == 1))
669 		err = vrf_output_direct(net, sk, skb);
670 
671 	/* reset skb device */
672 	if (likely(err == 1))
673 		nf_reset_ct(skb);
674 	else
675 		skb = NULL;
676 
677 	return skb;
678 }
679 
680 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
681 				  struct sock *sk,
682 				  struct sk_buff *skb)
683 {
684 	/* don't divert multicast or local broadcast */
685 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
686 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
687 		return skb;
688 
689 	if (qdisc_tx_is_default(vrf_dev))
690 		return vrf_ip_out_direct(vrf_dev, sk, skb);
691 
692 	return vrf_ip_out_redirect(vrf_dev, skb);
693 }
694 
695 /* called with rcu lock held */
696 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
697 				  struct sock *sk,
698 				  struct sk_buff *skb,
699 				  u16 proto)
700 {
701 	switch (proto) {
702 	case AF_INET:
703 		return vrf_ip_out(vrf_dev, sk, skb);
704 	case AF_INET6:
705 		return vrf_ip6_out(vrf_dev, sk, skb);
706 	}
707 
708 	return skb;
709 }
710 
711 /* holding rtnl */
712 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
713 {
714 	struct rtable *rth = rtnl_dereference(vrf->rth);
715 	struct net *net = dev_net(dev);
716 	struct dst_entry *dst;
717 
718 	RCU_INIT_POINTER(vrf->rth, NULL);
719 	synchronize_rcu();
720 
721 	/* move dev in dst's to loopback so this VRF device can be deleted
722 	 * - based on dst_ifdown
723 	 */
724 	if (rth) {
725 		dst = &rth->dst;
726 		dev_put(dst->dev);
727 		dst->dev = net->loopback_dev;
728 		dev_hold(dst->dev);
729 		dst_release(dst);
730 	}
731 }
732 
733 static int vrf_rtable_create(struct net_device *dev)
734 {
735 	struct net_vrf *vrf = netdev_priv(dev);
736 	struct rtable *rth;
737 
738 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
739 		return -ENOMEM;
740 
741 	/* create a dst for routing packets out through a VRF device */
742 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
743 	if (!rth)
744 		return -ENOMEM;
745 
746 	rth->dst.output	= vrf_output;
747 
748 	rcu_assign_pointer(vrf->rth, rth);
749 
750 	return 0;
751 }
752 
753 /**************************** device handling ********************/
754 
755 /* cycle interface to flush neighbor cache and move routes across tables */
756 static void cycle_netdev(struct net_device *dev,
757 			 struct netlink_ext_ack *extack)
758 {
759 	unsigned int flags = dev->flags;
760 	int ret;
761 
762 	if (!netif_running(dev))
763 		return;
764 
765 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
766 	if (ret >= 0)
767 		ret = dev_change_flags(dev, flags, extack);
768 
769 	if (ret < 0) {
770 		netdev_err(dev,
771 			   "Failed to cycle device %s; route tables might be wrong!\n",
772 			   dev->name);
773 	}
774 }
775 
776 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
777 			    struct netlink_ext_ack *extack)
778 {
779 	int ret;
780 
781 	/* do not allow loopback device to be enslaved to a VRF.
782 	 * The vrf device acts as the loopback for the vrf.
783 	 */
784 	if (port_dev == dev_net(dev)->loopback_dev) {
785 		NL_SET_ERR_MSG(extack,
786 			       "Can not enslave loopback device to a VRF");
787 		return -EOPNOTSUPP;
788 	}
789 
790 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
791 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
792 	if (ret < 0)
793 		goto err;
794 
795 	cycle_netdev(port_dev, extack);
796 
797 	return 0;
798 
799 err:
800 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
801 	return ret;
802 }
803 
804 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
805 			 struct netlink_ext_ack *extack)
806 {
807 	if (netif_is_l3_master(port_dev)) {
808 		NL_SET_ERR_MSG(extack,
809 			       "Can not enslave an L3 master device to a VRF");
810 		return -EINVAL;
811 	}
812 
813 	if (netif_is_l3_slave(port_dev))
814 		return -EINVAL;
815 
816 	return do_vrf_add_slave(dev, port_dev, extack);
817 }
818 
819 /* inverse of do_vrf_add_slave */
820 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
821 {
822 	netdev_upper_dev_unlink(port_dev, dev);
823 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
824 
825 	cycle_netdev(port_dev, NULL);
826 
827 	return 0;
828 }
829 
830 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
831 {
832 	return do_vrf_del_slave(dev, port_dev);
833 }
834 
835 static void vrf_dev_uninit(struct net_device *dev)
836 {
837 	struct net_vrf *vrf = netdev_priv(dev);
838 
839 	vrf_rtable_release(dev, vrf);
840 	vrf_rt6_release(dev, vrf);
841 
842 	free_percpu(dev->dstats);
843 	dev->dstats = NULL;
844 }
845 
846 static int vrf_dev_init(struct net_device *dev)
847 {
848 	struct net_vrf *vrf = netdev_priv(dev);
849 
850 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
851 	if (!dev->dstats)
852 		goto out_nomem;
853 
854 	/* create the default dst which points back to us */
855 	if (vrf_rtable_create(dev) != 0)
856 		goto out_stats;
857 
858 	if (vrf_rt6_create(dev) != 0)
859 		goto out_rth;
860 
861 	dev->flags = IFF_MASTER | IFF_NOARP;
862 
863 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
864 	dev->mtu = 64 * 1024;
865 
866 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
867 	dev->operstate = IF_OPER_UP;
868 	return 0;
869 
870 out_rth:
871 	vrf_rtable_release(dev, vrf);
872 out_stats:
873 	free_percpu(dev->dstats);
874 	dev->dstats = NULL;
875 out_nomem:
876 	return -ENOMEM;
877 }
878 
879 static const struct net_device_ops vrf_netdev_ops = {
880 	.ndo_init		= vrf_dev_init,
881 	.ndo_uninit		= vrf_dev_uninit,
882 	.ndo_start_xmit		= vrf_xmit,
883 	.ndo_set_mac_address	= eth_mac_addr,
884 	.ndo_get_stats64	= vrf_get_stats64,
885 	.ndo_add_slave		= vrf_add_slave,
886 	.ndo_del_slave		= vrf_del_slave,
887 };
888 
889 static u32 vrf_fib_table(const struct net_device *dev)
890 {
891 	struct net_vrf *vrf = netdev_priv(dev);
892 
893 	return vrf->tb_id;
894 }
895 
896 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
897 {
898 	kfree_skb(skb);
899 	return 0;
900 }
901 
902 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
903 				      struct sk_buff *skb,
904 				      struct net_device *dev)
905 {
906 	struct net *net = dev_net(dev);
907 
908 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
909 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
910 
911 	return skb;
912 }
913 
914 #if IS_ENABLED(CONFIG_IPV6)
915 /* neighbor handling is done with actual device; do not want
916  * to flip skb->dev for those ndisc packets. This really fails
917  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
918  * a start.
919  */
920 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
921 {
922 	const struct ipv6hdr *iph = ipv6_hdr(skb);
923 	bool rc = false;
924 
925 	if (iph->nexthdr == NEXTHDR_ICMP) {
926 		const struct icmp6hdr *icmph;
927 		struct icmp6hdr _icmph;
928 
929 		icmph = skb_header_pointer(skb, sizeof(*iph),
930 					   sizeof(_icmph), &_icmph);
931 		if (!icmph)
932 			goto out;
933 
934 		switch (icmph->icmp6_type) {
935 		case NDISC_ROUTER_SOLICITATION:
936 		case NDISC_ROUTER_ADVERTISEMENT:
937 		case NDISC_NEIGHBOUR_SOLICITATION:
938 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
939 		case NDISC_REDIRECT:
940 			rc = true;
941 			break;
942 		}
943 	}
944 
945 out:
946 	return rc;
947 }
948 
949 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
950 					     const struct net_device *dev,
951 					     struct flowi6 *fl6,
952 					     int ifindex,
953 					     const struct sk_buff *skb,
954 					     int flags)
955 {
956 	struct net_vrf *vrf = netdev_priv(dev);
957 
958 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
959 }
960 
961 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
962 			      int ifindex)
963 {
964 	const struct ipv6hdr *iph = ipv6_hdr(skb);
965 	struct flowi6 fl6 = {
966 		.flowi6_iif     = ifindex,
967 		.flowi6_mark    = skb->mark,
968 		.flowi6_proto   = iph->nexthdr,
969 		.daddr          = iph->daddr,
970 		.saddr          = iph->saddr,
971 		.flowlabel      = ip6_flowinfo(iph),
972 	};
973 	struct net *net = dev_net(vrf_dev);
974 	struct rt6_info *rt6;
975 
976 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
977 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
978 	if (unlikely(!rt6))
979 		return;
980 
981 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
982 		return;
983 
984 	skb_dst_set(skb, &rt6->dst);
985 }
986 
987 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
988 				   struct sk_buff *skb)
989 {
990 	int orig_iif = skb->skb_iif;
991 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
992 	bool is_ndisc = ipv6_ndisc_frame(skb);
993 
994 	/* loopback, multicast & non-ND link-local traffic; do not push through
995 	 * packet taps again. Reset pkt_type for upper layers to process skb
996 	 */
997 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
998 		skb->dev = vrf_dev;
999 		skb->skb_iif = vrf_dev->ifindex;
1000 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1001 		if (skb->pkt_type == PACKET_LOOPBACK)
1002 			skb->pkt_type = PACKET_HOST;
1003 		goto out;
1004 	}
1005 
1006 	/* if packet is NDISC then keep the ingress interface */
1007 	if (!is_ndisc) {
1008 		vrf_rx_stats(vrf_dev, skb->len);
1009 		skb->dev = vrf_dev;
1010 		skb->skb_iif = vrf_dev->ifindex;
1011 
1012 		if (!list_empty(&vrf_dev->ptype_all)) {
1013 			skb_push(skb, skb->mac_len);
1014 			dev_queue_xmit_nit(skb, vrf_dev);
1015 			skb_pull(skb, skb->mac_len);
1016 		}
1017 
1018 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1019 	}
1020 
1021 	if (need_strict)
1022 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1023 
1024 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1025 out:
1026 	return skb;
1027 }
1028 
1029 #else
1030 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1031 				   struct sk_buff *skb)
1032 {
1033 	return skb;
1034 }
1035 #endif
1036 
1037 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1038 				  struct sk_buff *skb)
1039 {
1040 	skb->dev = vrf_dev;
1041 	skb->skb_iif = vrf_dev->ifindex;
1042 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1043 
1044 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1045 		goto out;
1046 
1047 	/* loopback traffic; do not push through packet taps again.
1048 	 * Reset pkt_type for upper layers to process skb
1049 	 */
1050 	if (skb->pkt_type == PACKET_LOOPBACK) {
1051 		skb->pkt_type = PACKET_HOST;
1052 		goto out;
1053 	}
1054 
1055 	vrf_rx_stats(vrf_dev, skb->len);
1056 
1057 	if (!list_empty(&vrf_dev->ptype_all)) {
1058 		skb_push(skb, skb->mac_len);
1059 		dev_queue_xmit_nit(skb, vrf_dev);
1060 		skb_pull(skb, skb->mac_len);
1061 	}
1062 
1063 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1064 out:
1065 	return skb;
1066 }
1067 
1068 /* called with rcu lock held */
1069 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1070 				  struct sk_buff *skb,
1071 				  u16 proto)
1072 {
1073 	switch (proto) {
1074 	case AF_INET:
1075 		return vrf_ip_rcv(vrf_dev, skb);
1076 	case AF_INET6:
1077 		return vrf_ip6_rcv(vrf_dev, skb);
1078 	}
1079 
1080 	return skb;
1081 }
1082 
1083 #if IS_ENABLED(CONFIG_IPV6)
1084 /* send to link-local or multicast address via interface enslaved to
1085  * VRF device. Force lookup to VRF table without changing flow struct
1086  * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1087  * is taken on the dst by this function.
1088  */
1089 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1090 					      struct flowi6 *fl6)
1091 {
1092 	struct net *net = dev_net(dev);
1093 	int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1094 	struct dst_entry *dst = NULL;
1095 	struct rt6_info *rt;
1096 
1097 	/* VRF device does not have a link-local address and
1098 	 * sending packets to link-local or mcast addresses over
1099 	 * a VRF device does not make sense
1100 	 */
1101 	if (fl6->flowi6_oif == dev->ifindex) {
1102 		dst = &net->ipv6.ip6_null_entry->dst;
1103 		return dst;
1104 	}
1105 
1106 	if (!ipv6_addr_any(&fl6->saddr))
1107 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1108 
1109 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1110 	if (rt)
1111 		dst = &rt->dst;
1112 
1113 	return dst;
1114 }
1115 #endif
1116 
1117 static const struct l3mdev_ops vrf_l3mdev_ops = {
1118 	.l3mdev_fib_table	= vrf_fib_table,
1119 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1120 	.l3mdev_l3_out		= vrf_l3_out,
1121 #if IS_ENABLED(CONFIG_IPV6)
1122 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1123 #endif
1124 };
1125 
1126 static void vrf_get_drvinfo(struct net_device *dev,
1127 			    struct ethtool_drvinfo *info)
1128 {
1129 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1130 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1131 }
1132 
1133 static const struct ethtool_ops vrf_ethtool_ops = {
1134 	.get_drvinfo	= vrf_get_drvinfo,
1135 };
1136 
1137 static inline size_t vrf_fib_rule_nl_size(void)
1138 {
1139 	size_t sz;
1140 
1141 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1142 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1143 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1144 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1145 
1146 	return sz;
1147 }
1148 
1149 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1150 {
1151 	struct fib_rule_hdr *frh;
1152 	struct nlmsghdr *nlh;
1153 	struct sk_buff *skb;
1154 	int err;
1155 
1156 	if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1157 	    !ipv6_mod_enabled())
1158 		return 0;
1159 
1160 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1161 	if (!skb)
1162 		return -ENOMEM;
1163 
1164 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1165 	if (!nlh)
1166 		goto nla_put_failure;
1167 
1168 	/* rule only needs to appear once */
1169 	nlh->nlmsg_flags |= NLM_F_EXCL;
1170 
1171 	frh = nlmsg_data(nlh);
1172 	memset(frh, 0, sizeof(*frh));
1173 	frh->family = family;
1174 	frh->action = FR_ACT_TO_TBL;
1175 
1176 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1177 		goto nla_put_failure;
1178 
1179 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1180 		goto nla_put_failure;
1181 
1182 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1183 		goto nla_put_failure;
1184 
1185 	nlmsg_end(skb, nlh);
1186 
1187 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1188 	skb->sk = dev_net(dev)->rtnl;
1189 	if (add_it) {
1190 		err = fib_nl_newrule(skb, nlh, NULL);
1191 		if (err == -EEXIST)
1192 			err = 0;
1193 	} else {
1194 		err = fib_nl_delrule(skb, nlh, NULL);
1195 		if (err == -ENOENT)
1196 			err = 0;
1197 	}
1198 	nlmsg_free(skb);
1199 
1200 	return err;
1201 
1202 nla_put_failure:
1203 	nlmsg_free(skb);
1204 
1205 	return -EMSGSIZE;
1206 }
1207 
1208 static int vrf_add_fib_rules(const struct net_device *dev)
1209 {
1210 	int err;
1211 
1212 	err = vrf_fib_rule(dev, AF_INET,  true);
1213 	if (err < 0)
1214 		goto out_err;
1215 
1216 	err = vrf_fib_rule(dev, AF_INET6, true);
1217 	if (err < 0)
1218 		goto ipv6_err;
1219 
1220 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1221 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1222 	if (err < 0)
1223 		goto ipmr_err;
1224 #endif
1225 
1226 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1227 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1228 	if (err < 0)
1229 		goto ip6mr_err;
1230 #endif
1231 
1232 	return 0;
1233 
1234 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1235 ip6mr_err:
1236 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1237 #endif
1238 
1239 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1240 ipmr_err:
1241 	vrf_fib_rule(dev, AF_INET6,  false);
1242 #endif
1243 
1244 ipv6_err:
1245 	vrf_fib_rule(dev, AF_INET,  false);
1246 
1247 out_err:
1248 	netdev_err(dev, "Failed to add FIB rules.\n");
1249 	return err;
1250 }
1251 
1252 static void vrf_setup(struct net_device *dev)
1253 {
1254 	ether_setup(dev);
1255 
1256 	/* Initialize the device structure. */
1257 	dev->netdev_ops = &vrf_netdev_ops;
1258 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1259 	dev->ethtool_ops = &vrf_ethtool_ops;
1260 	dev->needs_free_netdev = true;
1261 
1262 	/* Fill in device structure with ethernet-generic values. */
1263 	eth_hw_addr_random(dev);
1264 
1265 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1266 	dev->features |= NETIF_F_LLTX;
1267 
1268 	/* don't allow vrf devices to change network namespaces. */
1269 	dev->features |= NETIF_F_NETNS_LOCAL;
1270 
1271 	/* does not make sense for a VLAN to be added to a vrf device */
1272 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1273 
1274 	/* enable offload features */
1275 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1276 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1277 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1278 
1279 	dev->hw_features = dev->features;
1280 	dev->hw_enc_features = dev->features;
1281 
1282 	/* default to no qdisc; user can add if desired */
1283 	dev->priv_flags |= IFF_NO_QUEUE;
1284 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1285 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1286 
1287 	/* VRF devices do not care about MTU, but if the MTU is set
1288 	 * too low then the ipv4 and ipv6 protocols are disabled
1289 	 * which breaks networking.
1290 	 */
1291 	dev->min_mtu = IPV6_MIN_MTU;
1292 	dev->max_mtu = ETH_MAX_MTU;
1293 }
1294 
1295 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1296 			struct netlink_ext_ack *extack)
1297 {
1298 	if (tb[IFLA_ADDRESS]) {
1299 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1300 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1301 			return -EINVAL;
1302 		}
1303 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1304 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1305 			return -EADDRNOTAVAIL;
1306 		}
1307 	}
1308 	return 0;
1309 }
1310 
1311 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1312 {
1313 	struct net_device *port_dev;
1314 	struct list_head *iter;
1315 
1316 	netdev_for_each_lower_dev(dev, port_dev, iter)
1317 		vrf_del_slave(dev, port_dev);
1318 
1319 	unregister_netdevice_queue(dev, head);
1320 }
1321 
1322 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1323 		       struct nlattr *tb[], struct nlattr *data[],
1324 		       struct netlink_ext_ack *extack)
1325 {
1326 	struct net_vrf *vrf = netdev_priv(dev);
1327 	bool *add_fib_rules;
1328 	struct net *net;
1329 	int err;
1330 
1331 	if (!data || !data[IFLA_VRF_TABLE]) {
1332 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1333 		return -EINVAL;
1334 	}
1335 
1336 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1337 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1338 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1339 				    "Invalid VRF table id");
1340 		return -EINVAL;
1341 	}
1342 
1343 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1344 
1345 	err = register_netdevice(dev);
1346 	if (err)
1347 		goto out;
1348 
1349 	net = dev_net(dev);
1350 	add_fib_rules = net_generic(net, vrf_net_id);
1351 	if (*add_fib_rules) {
1352 		err = vrf_add_fib_rules(dev);
1353 		if (err) {
1354 			unregister_netdevice(dev);
1355 			goto out;
1356 		}
1357 		*add_fib_rules = false;
1358 	}
1359 
1360 out:
1361 	return err;
1362 }
1363 
1364 static size_t vrf_nl_getsize(const struct net_device *dev)
1365 {
1366 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1367 }
1368 
1369 static int vrf_fillinfo(struct sk_buff *skb,
1370 			const struct net_device *dev)
1371 {
1372 	struct net_vrf *vrf = netdev_priv(dev);
1373 
1374 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1375 }
1376 
1377 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1378 				 const struct net_device *slave_dev)
1379 {
1380 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1381 }
1382 
1383 static int vrf_fill_slave_info(struct sk_buff *skb,
1384 			       const struct net_device *vrf_dev,
1385 			       const struct net_device *slave_dev)
1386 {
1387 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1388 
1389 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1390 		return -EMSGSIZE;
1391 
1392 	return 0;
1393 }
1394 
1395 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1396 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1397 };
1398 
1399 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1400 	.kind		= DRV_NAME,
1401 	.priv_size	= sizeof(struct net_vrf),
1402 
1403 	.get_size	= vrf_nl_getsize,
1404 	.policy		= vrf_nl_policy,
1405 	.validate	= vrf_validate,
1406 	.fill_info	= vrf_fillinfo,
1407 
1408 	.get_slave_size  = vrf_get_slave_size,
1409 	.fill_slave_info = vrf_fill_slave_info,
1410 
1411 	.newlink	= vrf_newlink,
1412 	.dellink	= vrf_dellink,
1413 	.setup		= vrf_setup,
1414 	.maxtype	= IFLA_VRF_MAX,
1415 };
1416 
1417 static int vrf_device_event(struct notifier_block *unused,
1418 			    unsigned long event, void *ptr)
1419 {
1420 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1421 
1422 	/* only care about unregister events to drop slave references */
1423 	if (event == NETDEV_UNREGISTER) {
1424 		struct net_device *vrf_dev;
1425 
1426 		if (!netif_is_l3_slave(dev))
1427 			goto out;
1428 
1429 		vrf_dev = netdev_master_upper_dev_get(dev);
1430 		vrf_del_slave(vrf_dev, dev);
1431 	}
1432 out:
1433 	return NOTIFY_DONE;
1434 }
1435 
1436 static struct notifier_block vrf_notifier_block __read_mostly = {
1437 	.notifier_call = vrf_device_event,
1438 };
1439 
1440 /* Initialize per network namespace state */
1441 static int __net_init vrf_netns_init(struct net *net)
1442 {
1443 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1444 
1445 	*add_fib_rules = true;
1446 
1447 	return 0;
1448 }
1449 
1450 static struct pernet_operations vrf_net_ops __net_initdata = {
1451 	.init = vrf_netns_init,
1452 	.id   = &vrf_net_id,
1453 	.size = sizeof(bool),
1454 };
1455 
1456 static int __init vrf_init_module(void)
1457 {
1458 	int rc;
1459 
1460 	register_netdevice_notifier(&vrf_notifier_block);
1461 
1462 	rc = register_pernet_subsys(&vrf_net_ops);
1463 	if (rc < 0)
1464 		goto error;
1465 
1466 	rc = rtnl_link_register(&vrf_link_ops);
1467 	if (rc < 0) {
1468 		unregister_pernet_subsys(&vrf_net_ops);
1469 		goto error;
1470 	}
1471 
1472 	return 0;
1473 
1474 error:
1475 	unregister_netdevice_notifier(&vrf_notifier_block);
1476 	return rc;
1477 }
1478 
1479 module_init(vrf_init_module);
1480 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1481 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1482 MODULE_LICENSE("GPL");
1483 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1484 MODULE_VERSION(DRV_VERSION);
1485