xref: /openbmc/linux/drivers/net/vrf.c (revision 82df5b73)
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_dst_lookup_flow(net, NULL, &fl6, NULL);
192 	if (IS_ERR(dst) || 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 	    IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
479 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
480 
481 	return vrf_ip6_out_redirect(vrf_dev, skb);
482 }
483 
484 /* holding rtnl */
485 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
486 {
487 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
488 	struct net *net = dev_net(dev);
489 	struct dst_entry *dst;
490 
491 	RCU_INIT_POINTER(vrf->rt6, NULL);
492 	synchronize_rcu();
493 
494 	/* move dev in dst's to loopback so this VRF device can be deleted
495 	 * - based on dst_ifdown
496 	 */
497 	if (rt6) {
498 		dst = &rt6->dst;
499 		dev_put(dst->dev);
500 		dst->dev = net->loopback_dev;
501 		dev_hold(dst->dev);
502 		dst_release(dst);
503 	}
504 }
505 
506 static int vrf_rt6_create(struct net_device *dev)
507 {
508 	int flags = DST_NOPOLICY | DST_NOXFRM;
509 	struct net_vrf *vrf = netdev_priv(dev);
510 	struct net *net = dev_net(dev);
511 	struct rt6_info *rt6;
512 	int rc = -ENOMEM;
513 
514 	/* IPv6 can be CONFIG enabled and then disabled runtime */
515 	if (!ipv6_mod_enabled())
516 		return 0;
517 
518 	vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
519 	if (!vrf->fib6_table)
520 		goto out;
521 
522 	/* create a dst for routing packets out a VRF device */
523 	rt6 = ip6_dst_alloc(net, dev, flags);
524 	if (!rt6)
525 		goto out;
526 
527 	rt6->dst.output	= vrf_output6;
528 
529 	rcu_assign_pointer(vrf->rt6, rt6);
530 
531 	rc = 0;
532 out:
533 	return rc;
534 }
535 #else
536 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
537 				   struct sock *sk,
538 				   struct sk_buff *skb)
539 {
540 	return skb;
541 }
542 
543 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
544 {
545 }
546 
547 static int vrf_rt6_create(struct net_device *dev)
548 {
549 	return 0;
550 }
551 #endif
552 
553 /* modelled after ip_finish_output2 */
554 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
555 {
556 	struct dst_entry *dst = skb_dst(skb);
557 	struct rtable *rt = (struct rtable *)dst;
558 	struct net_device *dev = dst->dev;
559 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
560 	struct neighbour *neigh;
561 	bool is_v6gw = false;
562 	int ret = -EINVAL;
563 
564 	nf_reset_ct(skb);
565 
566 	/* Be paranoid, rather than too clever. */
567 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
568 		struct sk_buff *skb2;
569 
570 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
571 		if (!skb2) {
572 			ret = -ENOMEM;
573 			goto err;
574 		}
575 		if (skb->sk)
576 			skb_set_owner_w(skb2, skb->sk);
577 
578 		consume_skb(skb);
579 		skb = skb2;
580 	}
581 
582 	rcu_read_lock_bh();
583 
584 	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
585 	if (!IS_ERR(neigh)) {
586 		sock_confirm_neigh(skb, neigh);
587 		/* if crossing protocols, can not use the cached header */
588 		ret = neigh_output(neigh, skb, is_v6gw);
589 		rcu_read_unlock_bh();
590 		return ret;
591 	}
592 
593 	rcu_read_unlock_bh();
594 err:
595 	vrf_tx_error(skb->dev, skb);
596 	return ret;
597 }
598 
599 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
600 {
601 	struct net_device *dev = skb_dst(skb)->dev;
602 
603 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
604 
605 	skb->dev = dev;
606 	skb->protocol = htons(ETH_P_IP);
607 
608 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
609 			    net, sk, skb, NULL, dev,
610 			    vrf_finish_output,
611 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
612 }
613 
614 /* set dst on skb to send packet to us via dev_xmit path. Allows
615  * packet to go through device based features such as qdisc, netfilter
616  * hooks and packet sockets with skb->dev set to vrf device.
617  */
618 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
619 					   struct sk_buff *skb)
620 {
621 	struct net_vrf *vrf = netdev_priv(vrf_dev);
622 	struct dst_entry *dst = NULL;
623 	struct rtable *rth;
624 
625 	rcu_read_lock();
626 
627 	rth = rcu_dereference(vrf->rth);
628 	if (likely(rth)) {
629 		dst = &rth->dst;
630 		dst_hold(dst);
631 	}
632 
633 	rcu_read_unlock();
634 
635 	if (unlikely(!dst)) {
636 		vrf_tx_error(vrf_dev, skb);
637 		return NULL;
638 	}
639 
640 	skb_dst_drop(skb);
641 	skb_dst_set(skb, dst);
642 
643 	return skb;
644 }
645 
646 static int vrf_output_direct(struct net *net, struct sock *sk,
647 			     struct sk_buff *skb)
648 {
649 	skb->protocol = htons(ETH_P_IP);
650 
651 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
652 			    net, sk, skb, NULL, skb->dev,
653 			    vrf_finish_direct,
654 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
655 }
656 
657 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
658 					 struct sock *sk,
659 					 struct sk_buff *skb)
660 {
661 	struct net *net = dev_net(vrf_dev);
662 	int err;
663 
664 	skb->dev = vrf_dev;
665 
666 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
667 		      skb, NULL, vrf_dev, vrf_output_direct);
668 
669 	if (likely(err == 1))
670 		err = vrf_output_direct(net, sk, skb);
671 
672 	/* reset skb device */
673 	if (likely(err == 1))
674 		nf_reset_ct(skb);
675 	else
676 		skb = NULL;
677 
678 	return skb;
679 }
680 
681 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
682 				  struct sock *sk,
683 				  struct sk_buff *skb)
684 {
685 	/* don't divert multicast or local broadcast */
686 	if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
687 	    ipv4_is_lbcast(ip_hdr(skb)->daddr))
688 		return skb;
689 
690 	if (qdisc_tx_is_default(vrf_dev) ||
691 	    IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
692 		return vrf_ip_out_direct(vrf_dev, sk, skb);
693 
694 	return vrf_ip_out_redirect(vrf_dev, skb);
695 }
696 
697 /* called with rcu lock held */
698 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
699 				  struct sock *sk,
700 				  struct sk_buff *skb,
701 				  u16 proto)
702 {
703 	switch (proto) {
704 	case AF_INET:
705 		return vrf_ip_out(vrf_dev, sk, skb);
706 	case AF_INET6:
707 		return vrf_ip6_out(vrf_dev, sk, skb);
708 	}
709 
710 	return skb;
711 }
712 
713 /* holding rtnl */
714 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
715 {
716 	struct rtable *rth = rtnl_dereference(vrf->rth);
717 	struct net *net = dev_net(dev);
718 	struct dst_entry *dst;
719 
720 	RCU_INIT_POINTER(vrf->rth, NULL);
721 	synchronize_rcu();
722 
723 	/* move dev in dst's to loopback so this VRF device can be deleted
724 	 * - based on dst_ifdown
725 	 */
726 	if (rth) {
727 		dst = &rth->dst;
728 		dev_put(dst->dev);
729 		dst->dev = net->loopback_dev;
730 		dev_hold(dst->dev);
731 		dst_release(dst);
732 	}
733 }
734 
735 static int vrf_rtable_create(struct net_device *dev)
736 {
737 	struct net_vrf *vrf = netdev_priv(dev);
738 	struct rtable *rth;
739 
740 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
741 		return -ENOMEM;
742 
743 	/* create a dst for routing packets out through a VRF device */
744 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
745 	if (!rth)
746 		return -ENOMEM;
747 
748 	rth->dst.output	= vrf_output;
749 
750 	rcu_assign_pointer(vrf->rth, rth);
751 
752 	return 0;
753 }
754 
755 /**************************** device handling ********************/
756 
757 /* cycle interface to flush neighbor cache and move routes across tables */
758 static void cycle_netdev(struct net_device *dev,
759 			 struct netlink_ext_ack *extack)
760 {
761 	unsigned int flags = dev->flags;
762 	int ret;
763 
764 	if (!netif_running(dev))
765 		return;
766 
767 	ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
768 	if (ret >= 0)
769 		ret = dev_change_flags(dev, flags, extack);
770 
771 	if (ret < 0) {
772 		netdev_err(dev,
773 			   "Failed to cycle device %s; route tables might be wrong!\n",
774 			   dev->name);
775 	}
776 }
777 
778 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
779 			    struct netlink_ext_ack *extack)
780 {
781 	int ret;
782 
783 	/* do not allow loopback device to be enslaved to a VRF.
784 	 * The vrf device acts as the loopback for the vrf.
785 	 */
786 	if (port_dev == dev_net(dev)->loopback_dev) {
787 		NL_SET_ERR_MSG(extack,
788 			       "Can not enslave loopback device to a VRF");
789 		return -EOPNOTSUPP;
790 	}
791 
792 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
793 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
794 	if (ret < 0)
795 		goto err;
796 
797 	cycle_netdev(port_dev, extack);
798 
799 	return 0;
800 
801 err:
802 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
803 	return ret;
804 }
805 
806 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
807 			 struct netlink_ext_ack *extack)
808 {
809 	if (netif_is_l3_master(port_dev)) {
810 		NL_SET_ERR_MSG(extack,
811 			       "Can not enslave an L3 master device to a VRF");
812 		return -EINVAL;
813 	}
814 
815 	if (netif_is_l3_slave(port_dev))
816 		return -EINVAL;
817 
818 	return do_vrf_add_slave(dev, port_dev, extack);
819 }
820 
821 /* inverse of do_vrf_add_slave */
822 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
823 {
824 	netdev_upper_dev_unlink(port_dev, dev);
825 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
826 
827 	cycle_netdev(port_dev, NULL);
828 
829 	return 0;
830 }
831 
832 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
833 {
834 	return do_vrf_del_slave(dev, port_dev);
835 }
836 
837 static void vrf_dev_uninit(struct net_device *dev)
838 {
839 	struct net_vrf *vrf = netdev_priv(dev);
840 
841 	vrf_rtable_release(dev, vrf);
842 	vrf_rt6_release(dev, vrf);
843 
844 	free_percpu(dev->dstats);
845 	dev->dstats = NULL;
846 }
847 
848 static int vrf_dev_init(struct net_device *dev)
849 {
850 	struct net_vrf *vrf = netdev_priv(dev);
851 
852 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
853 	if (!dev->dstats)
854 		goto out_nomem;
855 
856 	/* create the default dst which points back to us */
857 	if (vrf_rtable_create(dev) != 0)
858 		goto out_stats;
859 
860 	if (vrf_rt6_create(dev) != 0)
861 		goto out_rth;
862 
863 	dev->flags = IFF_MASTER | IFF_NOARP;
864 
865 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
866 	dev->mtu = 64 * 1024;
867 
868 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
869 	dev->operstate = IF_OPER_UP;
870 	netdev_lockdep_set_classes(dev);
871 	return 0;
872 
873 out_rth:
874 	vrf_rtable_release(dev, vrf);
875 out_stats:
876 	free_percpu(dev->dstats);
877 	dev->dstats = NULL;
878 out_nomem:
879 	return -ENOMEM;
880 }
881 
882 static const struct net_device_ops vrf_netdev_ops = {
883 	.ndo_init		= vrf_dev_init,
884 	.ndo_uninit		= vrf_dev_uninit,
885 	.ndo_start_xmit		= vrf_xmit,
886 	.ndo_set_mac_address	= eth_mac_addr,
887 	.ndo_get_stats64	= vrf_get_stats64,
888 	.ndo_add_slave		= vrf_add_slave,
889 	.ndo_del_slave		= vrf_del_slave,
890 };
891 
892 static u32 vrf_fib_table(const struct net_device *dev)
893 {
894 	struct net_vrf *vrf = netdev_priv(dev);
895 
896 	return vrf->tb_id;
897 }
898 
899 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
900 {
901 	kfree_skb(skb);
902 	return 0;
903 }
904 
905 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
906 				      struct sk_buff *skb,
907 				      struct net_device *dev)
908 {
909 	struct net *net = dev_net(dev);
910 
911 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
912 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
913 
914 	return skb;
915 }
916 
917 #if IS_ENABLED(CONFIG_IPV6)
918 /* neighbor handling is done with actual device; do not want
919  * to flip skb->dev for those ndisc packets. This really fails
920  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
921  * a start.
922  */
923 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
924 {
925 	const struct ipv6hdr *iph = ipv6_hdr(skb);
926 	bool rc = false;
927 
928 	if (iph->nexthdr == NEXTHDR_ICMP) {
929 		const struct icmp6hdr *icmph;
930 		struct icmp6hdr _icmph;
931 
932 		icmph = skb_header_pointer(skb, sizeof(*iph),
933 					   sizeof(_icmph), &_icmph);
934 		if (!icmph)
935 			goto out;
936 
937 		switch (icmph->icmp6_type) {
938 		case NDISC_ROUTER_SOLICITATION:
939 		case NDISC_ROUTER_ADVERTISEMENT:
940 		case NDISC_NEIGHBOUR_SOLICITATION:
941 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
942 		case NDISC_REDIRECT:
943 			rc = true;
944 			break;
945 		}
946 	}
947 
948 out:
949 	return rc;
950 }
951 
952 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
953 					     const struct net_device *dev,
954 					     struct flowi6 *fl6,
955 					     int ifindex,
956 					     const struct sk_buff *skb,
957 					     int flags)
958 {
959 	struct net_vrf *vrf = netdev_priv(dev);
960 
961 	return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
962 }
963 
964 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
965 			      int ifindex)
966 {
967 	const struct ipv6hdr *iph = ipv6_hdr(skb);
968 	struct flowi6 fl6 = {
969 		.flowi6_iif     = ifindex,
970 		.flowi6_mark    = skb->mark,
971 		.flowi6_proto   = iph->nexthdr,
972 		.daddr          = iph->daddr,
973 		.saddr          = iph->saddr,
974 		.flowlabel      = ip6_flowinfo(iph),
975 	};
976 	struct net *net = dev_net(vrf_dev);
977 	struct rt6_info *rt6;
978 
979 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
980 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
981 	if (unlikely(!rt6))
982 		return;
983 
984 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
985 		return;
986 
987 	skb_dst_set(skb, &rt6->dst);
988 }
989 
990 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
991 				   struct sk_buff *skb)
992 {
993 	int orig_iif = skb->skb_iif;
994 	bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
995 	bool is_ndisc = ipv6_ndisc_frame(skb);
996 
997 	/* loopback, multicast & non-ND link-local traffic; do not push through
998 	 * packet taps again. Reset pkt_type for upper layers to process skb
999 	 */
1000 	if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1001 		skb->dev = vrf_dev;
1002 		skb->skb_iif = vrf_dev->ifindex;
1003 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1004 		if (skb->pkt_type == PACKET_LOOPBACK)
1005 			skb->pkt_type = PACKET_HOST;
1006 		goto out;
1007 	}
1008 
1009 	/* if packet is NDISC then keep the ingress interface */
1010 	if (!is_ndisc) {
1011 		vrf_rx_stats(vrf_dev, skb->len);
1012 		skb->dev = vrf_dev;
1013 		skb->skb_iif = vrf_dev->ifindex;
1014 
1015 		if (!list_empty(&vrf_dev->ptype_all)) {
1016 			skb_push(skb, skb->mac_len);
1017 			dev_queue_xmit_nit(skb, vrf_dev);
1018 			skb_pull(skb, skb->mac_len);
1019 		}
1020 
1021 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1022 	}
1023 
1024 	if (need_strict)
1025 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1026 
1027 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1028 out:
1029 	return skb;
1030 }
1031 
1032 #else
1033 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1034 				   struct sk_buff *skb)
1035 {
1036 	return skb;
1037 }
1038 #endif
1039 
1040 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1041 				  struct sk_buff *skb)
1042 {
1043 	skb->dev = vrf_dev;
1044 	skb->skb_iif = vrf_dev->ifindex;
1045 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1046 
1047 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1048 		goto out;
1049 
1050 	/* loopback traffic; do not push through packet taps again.
1051 	 * Reset pkt_type for upper layers to process skb
1052 	 */
1053 	if (skb->pkt_type == PACKET_LOOPBACK) {
1054 		skb->pkt_type = PACKET_HOST;
1055 		goto out;
1056 	}
1057 
1058 	vrf_rx_stats(vrf_dev, skb->len);
1059 
1060 	if (!list_empty(&vrf_dev->ptype_all)) {
1061 		skb_push(skb, skb->mac_len);
1062 		dev_queue_xmit_nit(skb, vrf_dev);
1063 		skb_pull(skb, skb->mac_len);
1064 	}
1065 
1066 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1067 out:
1068 	return skb;
1069 }
1070 
1071 /* called with rcu lock held */
1072 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1073 				  struct sk_buff *skb,
1074 				  u16 proto)
1075 {
1076 	switch (proto) {
1077 	case AF_INET:
1078 		return vrf_ip_rcv(vrf_dev, skb);
1079 	case AF_INET6:
1080 		return vrf_ip6_rcv(vrf_dev, skb);
1081 	}
1082 
1083 	return skb;
1084 }
1085 
1086 #if IS_ENABLED(CONFIG_IPV6)
1087 /* send to link-local or multicast address via interface enslaved to
1088  * VRF device. Force lookup to VRF table without changing flow struct
1089  * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1090  * is taken on the dst by this function.
1091  */
1092 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1093 					      struct flowi6 *fl6)
1094 {
1095 	struct net *net = dev_net(dev);
1096 	int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1097 	struct dst_entry *dst = NULL;
1098 	struct rt6_info *rt;
1099 
1100 	/* VRF device does not have a link-local address and
1101 	 * sending packets to link-local or mcast addresses over
1102 	 * a VRF device does not make sense
1103 	 */
1104 	if (fl6->flowi6_oif == dev->ifindex) {
1105 		dst = &net->ipv6.ip6_null_entry->dst;
1106 		return dst;
1107 	}
1108 
1109 	if (!ipv6_addr_any(&fl6->saddr))
1110 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1111 
1112 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1113 	if (rt)
1114 		dst = &rt->dst;
1115 
1116 	return dst;
1117 }
1118 #endif
1119 
1120 static const struct l3mdev_ops vrf_l3mdev_ops = {
1121 	.l3mdev_fib_table	= vrf_fib_table,
1122 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1123 	.l3mdev_l3_out		= vrf_l3_out,
1124 #if IS_ENABLED(CONFIG_IPV6)
1125 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1126 #endif
1127 };
1128 
1129 static void vrf_get_drvinfo(struct net_device *dev,
1130 			    struct ethtool_drvinfo *info)
1131 {
1132 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1133 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1134 }
1135 
1136 static const struct ethtool_ops vrf_ethtool_ops = {
1137 	.get_drvinfo	= vrf_get_drvinfo,
1138 };
1139 
1140 static inline size_t vrf_fib_rule_nl_size(void)
1141 {
1142 	size_t sz;
1143 
1144 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1145 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1146 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1147 	sz += nla_total_size(sizeof(u8));       /* FRA_PROTOCOL */
1148 
1149 	return sz;
1150 }
1151 
1152 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1153 {
1154 	struct fib_rule_hdr *frh;
1155 	struct nlmsghdr *nlh;
1156 	struct sk_buff *skb;
1157 	int err;
1158 
1159 	if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1160 	    !ipv6_mod_enabled())
1161 		return 0;
1162 
1163 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1164 	if (!skb)
1165 		return -ENOMEM;
1166 
1167 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1168 	if (!nlh)
1169 		goto nla_put_failure;
1170 
1171 	/* rule only needs to appear once */
1172 	nlh->nlmsg_flags |= NLM_F_EXCL;
1173 
1174 	frh = nlmsg_data(nlh);
1175 	memset(frh, 0, sizeof(*frh));
1176 	frh->family = family;
1177 	frh->action = FR_ACT_TO_TBL;
1178 
1179 	if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1180 		goto nla_put_failure;
1181 
1182 	if (nla_put_u8(skb, FRA_L3MDEV, 1))
1183 		goto nla_put_failure;
1184 
1185 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1186 		goto nla_put_failure;
1187 
1188 	nlmsg_end(skb, nlh);
1189 
1190 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1191 	skb->sk = dev_net(dev)->rtnl;
1192 	if (add_it) {
1193 		err = fib_nl_newrule(skb, nlh, NULL);
1194 		if (err == -EEXIST)
1195 			err = 0;
1196 	} else {
1197 		err = fib_nl_delrule(skb, nlh, NULL);
1198 		if (err == -ENOENT)
1199 			err = 0;
1200 	}
1201 	nlmsg_free(skb);
1202 
1203 	return err;
1204 
1205 nla_put_failure:
1206 	nlmsg_free(skb);
1207 
1208 	return -EMSGSIZE;
1209 }
1210 
1211 static int vrf_add_fib_rules(const struct net_device *dev)
1212 {
1213 	int err;
1214 
1215 	err = vrf_fib_rule(dev, AF_INET,  true);
1216 	if (err < 0)
1217 		goto out_err;
1218 
1219 	err = vrf_fib_rule(dev, AF_INET6, true);
1220 	if (err < 0)
1221 		goto ipv6_err;
1222 
1223 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1224 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1225 	if (err < 0)
1226 		goto ipmr_err;
1227 #endif
1228 
1229 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1230 	err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1231 	if (err < 0)
1232 		goto ip6mr_err;
1233 #endif
1234 
1235 	return 0;
1236 
1237 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1238 ip6mr_err:
1239 	vrf_fib_rule(dev, RTNL_FAMILY_IPMR,  false);
1240 #endif
1241 
1242 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1243 ipmr_err:
1244 	vrf_fib_rule(dev, AF_INET6,  false);
1245 #endif
1246 
1247 ipv6_err:
1248 	vrf_fib_rule(dev, AF_INET,  false);
1249 
1250 out_err:
1251 	netdev_err(dev, "Failed to add FIB rules.\n");
1252 	return err;
1253 }
1254 
1255 static void vrf_setup(struct net_device *dev)
1256 {
1257 	ether_setup(dev);
1258 
1259 	/* Initialize the device structure. */
1260 	dev->netdev_ops = &vrf_netdev_ops;
1261 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1262 	dev->ethtool_ops = &vrf_ethtool_ops;
1263 	dev->needs_free_netdev = true;
1264 
1265 	/* Fill in device structure with ethernet-generic values. */
1266 	eth_hw_addr_random(dev);
1267 
1268 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1269 	dev->features |= NETIF_F_LLTX;
1270 
1271 	/* don't allow vrf devices to change network namespaces. */
1272 	dev->features |= NETIF_F_NETNS_LOCAL;
1273 
1274 	/* does not make sense for a VLAN to be added to a vrf device */
1275 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1276 
1277 	/* enable offload features */
1278 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1279 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1280 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1281 
1282 	dev->hw_features = dev->features;
1283 	dev->hw_enc_features = dev->features;
1284 
1285 	/* default to no qdisc; user can add if desired */
1286 	dev->priv_flags |= IFF_NO_QUEUE;
1287 	dev->priv_flags |= IFF_NO_RX_HANDLER;
1288 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1289 
1290 	/* VRF devices do not care about MTU, but if the MTU is set
1291 	 * too low then the ipv4 and ipv6 protocols are disabled
1292 	 * which breaks networking.
1293 	 */
1294 	dev->min_mtu = IPV6_MIN_MTU;
1295 	dev->max_mtu = ETH_MAX_MTU;
1296 }
1297 
1298 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1299 			struct netlink_ext_ack *extack)
1300 {
1301 	if (tb[IFLA_ADDRESS]) {
1302 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1303 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1304 			return -EINVAL;
1305 		}
1306 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1307 			NL_SET_ERR_MSG(extack, "Invalid hardware address");
1308 			return -EADDRNOTAVAIL;
1309 		}
1310 	}
1311 	return 0;
1312 }
1313 
1314 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1315 {
1316 	struct net_device *port_dev;
1317 	struct list_head *iter;
1318 
1319 	netdev_for_each_lower_dev(dev, port_dev, iter)
1320 		vrf_del_slave(dev, port_dev);
1321 
1322 	unregister_netdevice_queue(dev, head);
1323 }
1324 
1325 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1326 		       struct nlattr *tb[], struct nlattr *data[],
1327 		       struct netlink_ext_ack *extack)
1328 {
1329 	struct net_vrf *vrf = netdev_priv(dev);
1330 	bool *add_fib_rules;
1331 	struct net *net;
1332 	int err;
1333 
1334 	if (!data || !data[IFLA_VRF_TABLE]) {
1335 		NL_SET_ERR_MSG(extack, "VRF table id is missing");
1336 		return -EINVAL;
1337 	}
1338 
1339 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1340 	if (vrf->tb_id == RT_TABLE_UNSPEC) {
1341 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1342 				    "Invalid VRF table id");
1343 		return -EINVAL;
1344 	}
1345 
1346 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1347 
1348 	err = register_netdevice(dev);
1349 	if (err)
1350 		goto out;
1351 
1352 	net = dev_net(dev);
1353 	add_fib_rules = net_generic(net, vrf_net_id);
1354 	if (*add_fib_rules) {
1355 		err = vrf_add_fib_rules(dev);
1356 		if (err) {
1357 			unregister_netdevice(dev);
1358 			goto out;
1359 		}
1360 		*add_fib_rules = false;
1361 	}
1362 
1363 out:
1364 	return err;
1365 }
1366 
1367 static size_t vrf_nl_getsize(const struct net_device *dev)
1368 {
1369 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1370 }
1371 
1372 static int vrf_fillinfo(struct sk_buff *skb,
1373 			const struct net_device *dev)
1374 {
1375 	struct net_vrf *vrf = netdev_priv(dev);
1376 
1377 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1378 }
1379 
1380 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1381 				 const struct net_device *slave_dev)
1382 {
1383 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1384 }
1385 
1386 static int vrf_fill_slave_info(struct sk_buff *skb,
1387 			       const struct net_device *vrf_dev,
1388 			       const struct net_device *slave_dev)
1389 {
1390 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1391 
1392 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1393 		return -EMSGSIZE;
1394 
1395 	return 0;
1396 }
1397 
1398 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1399 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1400 };
1401 
1402 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1403 	.kind		= DRV_NAME,
1404 	.priv_size	= sizeof(struct net_vrf),
1405 
1406 	.get_size	= vrf_nl_getsize,
1407 	.policy		= vrf_nl_policy,
1408 	.validate	= vrf_validate,
1409 	.fill_info	= vrf_fillinfo,
1410 
1411 	.get_slave_size  = vrf_get_slave_size,
1412 	.fill_slave_info = vrf_fill_slave_info,
1413 
1414 	.newlink	= vrf_newlink,
1415 	.dellink	= vrf_dellink,
1416 	.setup		= vrf_setup,
1417 	.maxtype	= IFLA_VRF_MAX,
1418 };
1419 
1420 static int vrf_device_event(struct notifier_block *unused,
1421 			    unsigned long event, void *ptr)
1422 {
1423 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1424 
1425 	/* only care about unregister events to drop slave references */
1426 	if (event == NETDEV_UNREGISTER) {
1427 		struct net_device *vrf_dev;
1428 
1429 		if (!netif_is_l3_slave(dev))
1430 			goto out;
1431 
1432 		vrf_dev = netdev_master_upper_dev_get(dev);
1433 		vrf_del_slave(vrf_dev, dev);
1434 	}
1435 out:
1436 	return NOTIFY_DONE;
1437 }
1438 
1439 static struct notifier_block vrf_notifier_block __read_mostly = {
1440 	.notifier_call = vrf_device_event,
1441 };
1442 
1443 /* Initialize per network namespace state */
1444 static int __net_init vrf_netns_init(struct net *net)
1445 {
1446 	bool *add_fib_rules = net_generic(net, vrf_net_id);
1447 
1448 	*add_fib_rules = true;
1449 
1450 	return 0;
1451 }
1452 
1453 static struct pernet_operations vrf_net_ops __net_initdata = {
1454 	.init = vrf_netns_init,
1455 	.id   = &vrf_net_id,
1456 	.size = sizeof(bool),
1457 };
1458 
1459 static int __init vrf_init_module(void)
1460 {
1461 	int rc;
1462 
1463 	register_netdevice_notifier(&vrf_notifier_block);
1464 
1465 	rc = register_pernet_subsys(&vrf_net_ops);
1466 	if (rc < 0)
1467 		goto error;
1468 
1469 	rc = rtnl_link_register(&vrf_link_ops);
1470 	if (rc < 0) {
1471 		unregister_pernet_subsys(&vrf_net_ops);
1472 		goto error;
1473 	}
1474 
1475 	return 0;
1476 
1477 error:
1478 	unregister_netdevice_notifier(&vrf_notifier_block);
1479 	return rc;
1480 }
1481 
1482 module_init(vrf_init_module);
1483 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1484 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1485 MODULE_LICENSE("GPL");
1486 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1487 MODULE_VERSION(DRV_VERSION);
1488