xref: /openbmc/linux/drivers/net/vrf.c (revision bb0eb050)
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 
40 #define DRV_NAME	"vrf"
41 #define DRV_VERSION	"1.0"
42 
43 #define FIB_RULE_PREF  1000       /* default preference for FIB rules */
44 static bool add_fib_rules = true;
45 
46 struct net_vrf {
47 	struct rtable __rcu	*rth;
48 	struct rtable __rcu	*rth_local;
49 	struct rt6_info	__rcu	*rt6;
50 	struct rt6_info	__rcu	*rt6_local;
51 	u32                     tb_id;
52 };
53 
54 struct pcpu_dstats {
55 	u64			tx_pkts;
56 	u64			tx_bytes;
57 	u64			tx_drps;
58 	u64			rx_pkts;
59 	u64			rx_bytes;
60 	u64			rx_drps;
61 	struct u64_stats_sync	syncp;
62 };
63 
64 static void vrf_rx_stats(struct net_device *dev, int len)
65 {
66 	struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
67 
68 	u64_stats_update_begin(&dstats->syncp);
69 	dstats->rx_pkts++;
70 	dstats->rx_bytes += len;
71 	u64_stats_update_end(&dstats->syncp);
72 }
73 
74 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
75 {
76 	vrf_dev->stats.tx_errors++;
77 	kfree_skb(skb);
78 }
79 
80 static void vrf_get_stats64(struct net_device *dev,
81 			    struct rtnl_link_stats64 *stats)
82 {
83 	int i;
84 
85 	for_each_possible_cpu(i) {
86 		const struct pcpu_dstats *dstats;
87 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
88 		unsigned int start;
89 
90 		dstats = per_cpu_ptr(dev->dstats, i);
91 		do {
92 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
93 			tbytes = dstats->tx_bytes;
94 			tpkts = dstats->tx_pkts;
95 			tdrops = dstats->tx_drps;
96 			rbytes = dstats->rx_bytes;
97 			rpkts = dstats->rx_pkts;
98 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
99 		stats->tx_bytes += tbytes;
100 		stats->tx_packets += tpkts;
101 		stats->tx_dropped += tdrops;
102 		stats->rx_bytes += rbytes;
103 		stats->rx_packets += rpkts;
104 	}
105 }
106 
107 /* by default VRF devices do not have a qdisc and are expected
108  * to be created with only a single queue.
109  */
110 static bool qdisc_tx_is_default(const struct net_device *dev)
111 {
112 	struct netdev_queue *txq;
113 	struct Qdisc *qdisc;
114 
115 	if (dev->num_tx_queues > 1)
116 		return false;
117 
118 	txq = netdev_get_tx_queue(dev, 0);
119 	qdisc = rcu_access_pointer(txq->qdisc);
120 
121 	return !qdisc->enqueue;
122 }
123 
124 /* Local traffic destined to local address. Reinsert the packet to rx
125  * path, similar to loopback handling.
126  */
127 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
128 			  struct dst_entry *dst)
129 {
130 	int len = skb->len;
131 
132 	skb_orphan(skb);
133 
134 	skb_dst_set(skb, dst);
135 	skb_dst_force(skb);
136 
137 	/* set pkt_type to avoid skb hitting packet taps twice -
138 	 * once on Tx and again in Rx processing
139 	 */
140 	skb->pkt_type = PACKET_LOOPBACK;
141 
142 	skb->protocol = eth_type_trans(skb, dev);
143 
144 	if (likely(netif_rx(skb) == NET_RX_SUCCESS))
145 		vrf_rx_stats(dev, len);
146 	else
147 		this_cpu_inc(dev->dstats->rx_drps);
148 
149 	return NETDEV_TX_OK;
150 }
151 
152 #if IS_ENABLED(CONFIG_IPV6)
153 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
154 			     struct sk_buff *skb)
155 {
156 	int err;
157 
158 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
159 		      sk, skb, NULL, skb_dst(skb)->dev, dst_output);
160 
161 	if (likely(err == 1))
162 		err = dst_output(net, sk, skb);
163 
164 	return err;
165 }
166 
167 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
168 					   struct net_device *dev)
169 {
170 	const struct ipv6hdr *iph = ipv6_hdr(skb);
171 	struct net *net = dev_net(skb->dev);
172 	struct flowi6 fl6 = {
173 		/* needed to match OIF rule */
174 		.flowi6_oif = dev->ifindex,
175 		.flowi6_iif = LOOPBACK_IFINDEX,
176 		.daddr = iph->daddr,
177 		.saddr = iph->saddr,
178 		.flowlabel = ip6_flowinfo(iph),
179 		.flowi6_mark = skb->mark,
180 		.flowi6_proto = iph->nexthdr,
181 		.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
182 	};
183 	int ret = NET_XMIT_DROP;
184 	struct dst_entry *dst;
185 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
186 
187 	dst = ip6_route_output(net, NULL, &fl6);
188 	if (dst == dst_null)
189 		goto err;
190 
191 	skb_dst_drop(skb);
192 
193 	/* if dst.dev is loopback or the VRF device again this is locally
194 	 * originated traffic destined to a local address. Short circuit
195 	 * to Rx path using our local dst
196 	 */
197 	if (dst->dev == net->loopback_dev || dst->dev == dev) {
198 		struct net_vrf *vrf = netdev_priv(dev);
199 		struct rt6_info *rt6_local;
200 
201 		/* release looked up dst and use cached local dst */
202 		dst_release(dst);
203 
204 		rcu_read_lock();
205 
206 		rt6_local = rcu_dereference(vrf->rt6_local);
207 		if (unlikely(!rt6_local)) {
208 			rcu_read_unlock();
209 			goto err;
210 		}
211 
212 		/* Ordering issue: cached local dst is created on newlink
213 		 * before the IPv6 initialization. Using the local dst
214 		 * requires rt6i_idev to be set so make sure it is.
215 		 */
216 		if (unlikely(!rt6_local->rt6i_idev)) {
217 			rt6_local->rt6i_idev = in6_dev_get(dev);
218 			if (!rt6_local->rt6i_idev) {
219 				rcu_read_unlock();
220 				goto err;
221 			}
222 		}
223 
224 		dst = &rt6_local->dst;
225 		dst_hold(dst);
226 
227 		rcu_read_unlock();
228 
229 		return vrf_local_xmit(skb, dev, &rt6_local->dst);
230 	}
231 
232 	skb_dst_set(skb, dst);
233 
234 	/* strip the ethernet header added for pass through VRF device */
235 	__skb_pull(skb, skb_network_offset(skb));
236 
237 	ret = vrf_ip6_local_out(net, skb->sk, skb);
238 	if (unlikely(net_xmit_eval(ret)))
239 		dev->stats.tx_errors++;
240 	else
241 		ret = NET_XMIT_SUCCESS;
242 
243 	return ret;
244 err:
245 	vrf_tx_error(dev, skb);
246 	return NET_XMIT_DROP;
247 }
248 #else
249 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
250 					   struct net_device *dev)
251 {
252 	vrf_tx_error(dev, skb);
253 	return NET_XMIT_DROP;
254 }
255 #endif
256 
257 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
258 static int vrf_ip_local_out(struct net *net, struct sock *sk,
259 			    struct sk_buff *skb)
260 {
261 	int err;
262 
263 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
264 		      skb, NULL, skb_dst(skb)->dev, dst_output);
265 	if (likely(err == 1))
266 		err = dst_output(net, sk, skb);
267 
268 	return err;
269 }
270 
271 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
272 					   struct net_device *vrf_dev)
273 {
274 	struct iphdr *ip4h = ip_hdr(skb);
275 	int ret = NET_XMIT_DROP;
276 	struct flowi4 fl4 = {
277 		/* needed to match OIF rule */
278 		.flowi4_oif = vrf_dev->ifindex,
279 		.flowi4_iif = LOOPBACK_IFINDEX,
280 		.flowi4_tos = RT_TOS(ip4h->tos),
281 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
282 		.flowi4_proto = ip4h->protocol,
283 		.daddr = ip4h->daddr,
284 		.saddr = ip4h->saddr,
285 	};
286 	struct net *net = dev_net(vrf_dev);
287 	struct rtable *rt;
288 
289 	rt = ip_route_output_flow(net, &fl4, NULL);
290 	if (IS_ERR(rt))
291 		goto err;
292 
293 	skb_dst_drop(skb);
294 
295 	/* if dst.dev is loopback or the VRF device again this is locally
296 	 * originated traffic destined to a local address. Short circuit
297 	 * to Rx path using our local dst
298 	 */
299 	if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
300 		struct net_vrf *vrf = netdev_priv(vrf_dev);
301 		struct rtable *rth_local;
302 		struct dst_entry *dst = NULL;
303 
304 		ip_rt_put(rt);
305 
306 		rcu_read_lock();
307 
308 		rth_local = rcu_dereference(vrf->rth_local);
309 		if (likely(rth_local)) {
310 			dst = &rth_local->dst;
311 			dst_hold(dst);
312 		}
313 
314 		rcu_read_unlock();
315 
316 		if (unlikely(!dst))
317 			goto err;
318 
319 		return vrf_local_xmit(skb, vrf_dev, dst);
320 	}
321 
322 	skb_dst_set(skb, &rt->dst);
323 
324 	/* strip the ethernet header added for pass through VRF device */
325 	__skb_pull(skb, skb_network_offset(skb));
326 
327 	if (!ip4h->saddr) {
328 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
329 					       RT_SCOPE_LINK);
330 	}
331 
332 	ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
333 	if (unlikely(net_xmit_eval(ret)))
334 		vrf_dev->stats.tx_errors++;
335 	else
336 		ret = NET_XMIT_SUCCESS;
337 
338 out:
339 	return ret;
340 err:
341 	vrf_tx_error(vrf_dev, skb);
342 	goto out;
343 }
344 
345 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
346 {
347 	switch (skb->protocol) {
348 	case htons(ETH_P_IP):
349 		return vrf_process_v4_outbound(skb, dev);
350 	case htons(ETH_P_IPV6):
351 		return vrf_process_v6_outbound(skb, dev);
352 	default:
353 		vrf_tx_error(dev, skb);
354 		return NET_XMIT_DROP;
355 	}
356 }
357 
358 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
359 {
360 	int len = skb->len;
361 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
362 
363 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
364 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
365 
366 		u64_stats_update_begin(&dstats->syncp);
367 		dstats->tx_pkts++;
368 		dstats->tx_bytes += len;
369 		u64_stats_update_end(&dstats->syncp);
370 	} else {
371 		this_cpu_inc(dev->dstats->tx_drps);
372 	}
373 
374 	return ret;
375 }
376 
377 static int vrf_finish_direct(struct net *net, struct sock *sk,
378 			     struct sk_buff *skb)
379 {
380 	struct net_device *vrf_dev = skb->dev;
381 
382 	if (!list_empty(&vrf_dev->ptype_all) &&
383 	    likely(skb_headroom(skb) >= ETH_HLEN)) {
384 		struct ethhdr *eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
385 
386 		ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
387 		eth_zero_addr(eth->h_dest);
388 		eth->h_proto = skb->protocol;
389 
390 		rcu_read_lock_bh();
391 		dev_queue_xmit_nit(skb, vrf_dev);
392 		rcu_read_unlock_bh();
393 
394 		skb_pull(skb, ETH_HLEN);
395 	}
396 
397 	return 1;
398 }
399 
400 #if IS_ENABLED(CONFIG_IPV6)
401 /* modelled after ip6_finish_output2 */
402 static int vrf_finish_output6(struct net *net, struct sock *sk,
403 			      struct sk_buff *skb)
404 {
405 	struct dst_entry *dst = skb_dst(skb);
406 	struct net_device *dev = dst->dev;
407 	struct neighbour *neigh;
408 	struct in6_addr *nexthop;
409 	int ret;
410 
411 	nf_reset(skb);
412 
413 	skb->protocol = htons(ETH_P_IPV6);
414 	skb->dev = dev;
415 
416 	rcu_read_lock_bh();
417 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
418 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
419 	if (unlikely(!neigh))
420 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
421 	if (!IS_ERR(neigh)) {
422 		sock_confirm_neigh(skb, neigh);
423 		ret = neigh_output(neigh, skb);
424 		rcu_read_unlock_bh();
425 		return ret;
426 	}
427 	rcu_read_unlock_bh();
428 
429 	IP6_INC_STATS(dev_net(dst->dev),
430 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
431 	kfree_skb(skb);
432 	return -EINVAL;
433 }
434 
435 /* modelled after ip6_output */
436 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
437 {
438 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
439 			    net, sk, skb, NULL, skb_dst(skb)->dev,
440 			    vrf_finish_output6,
441 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
442 }
443 
444 /* set dst on skb to send packet to us via dev_xmit path. Allows
445  * packet to go through device based features such as qdisc, netfilter
446  * hooks and packet sockets with skb->dev set to vrf device.
447  */
448 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
449 					    struct sk_buff *skb)
450 {
451 	struct net_vrf *vrf = netdev_priv(vrf_dev);
452 	struct dst_entry *dst = NULL;
453 	struct rt6_info *rt6;
454 
455 	rcu_read_lock();
456 
457 	rt6 = rcu_dereference(vrf->rt6);
458 	if (likely(rt6)) {
459 		dst = &rt6->dst;
460 		dst_hold(dst);
461 	}
462 
463 	rcu_read_unlock();
464 
465 	if (unlikely(!dst)) {
466 		vrf_tx_error(vrf_dev, skb);
467 		return NULL;
468 	}
469 
470 	skb_dst_drop(skb);
471 	skb_dst_set(skb, dst);
472 
473 	return skb;
474 }
475 
476 static int vrf_output6_direct(struct net *net, struct sock *sk,
477 			      struct sk_buff *skb)
478 {
479 	skb->protocol = htons(ETH_P_IPV6);
480 
481 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
482 			    net, sk, skb, NULL, skb->dev,
483 			    vrf_finish_direct,
484 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
485 }
486 
487 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
488 					  struct sock *sk,
489 					  struct sk_buff *skb)
490 {
491 	struct net *net = dev_net(vrf_dev);
492 	int err;
493 
494 	skb->dev = vrf_dev;
495 
496 	err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
497 		      skb, NULL, vrf_dev, vrf_output6_direct);
498 
499 	if (likely(err == 1))
500 		err = vrf_output6_direct(net, sk, skb);
501 
502 	/* reset skb device */
503 	if (likely(err == 1))
504 		nf_reset(skb);
505 	else
506 		skb = NULL;
507 
508 	return skb;
509 }
510 
511 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
512 				   struct sock *sk,
513 				   struct sk_buff *skb)
514 {
515 	/* don't divert link scope packets */
516 	if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
517 		return skb;
518 
519 	if (qdisc_tx_is_default(vrf_dev))
520 		return vrf_ip6_out_direct(vrf_dev, sk, skb);
521 
522 	return vrf_ip6_out_redirect(vrf_dev, skb);
523 }
524 
525 /* holding rtnl */
526 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
527 {
528 	struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
529 	struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
530 	struct net *net = dev_net(dev);
531 	struct dst_entry *dst;
532 
533 	RCU_INIT_POINTER(vrf->rt6, NULL);
534 	RCU_INIT_POINTER(vrf->rt6_local, NULL);
535 	synchronize_rcu();
536 
537 	/* move dev in dst's to loopback so this VRF device can be deleted
538 	 * - based on dst_ifdown
539 	 */
540 	if (rt6) {
541 		dst = &rt6->dst;
542 		dev_put(dst->dev);
543 		dst->dev = net->loopback_dev;
544 		dev_hold(dst->dev);
545 		dst_release(dst);
546 	}
547 
548 	if (rt6_local) {
549 		if (rt6_local->rt6i_idev) {
550 			in6_dev_put(rt6_local->rt6i_idev);
551 			rt6_local->rt6i_idev = NULL;
552 		}
553 
554 		dst = &rt6_local->dst;
555 		dev_put(dst->dev);
556 		dst->dev = net->loopback_dev;
557 		dev_hold(dst->dev);
558 		dst_release(dst);
559 	}
560 }
561 
562 static int vrf_rt6_create(struct net_device *dev)
563 {
564 	int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
565 	struct net_vrf *vrf = netdev_priv(dev);
566 	struct net *net = dev_net(dev);
567 	struct fib6_table *rt6i_table;
568 	struct rt6_info *rt6, *rt6_local;
569 	int rc = -ENOMEM;
570 
571 	/* IPv6 can be CONFIG enabled and then disabled runtime */
572 	if (!ipv6_mod_enabled())
573 		return 0;
574 
575 	rt6i_table = fib6_new_table(net, vrf->tb_id);
576 	if (!rt6i_table)
577 		goto out;
578 
579 	/* create a dst for routing packets out a VRF device */
580 	rt6 = ip6_dst_alloc(net, dev, flags);
581 	if (!rt6)
582 		goto out;
583 
584 	dst_hold(&rt6->dst);
585 
586 	rt6->rt6i_table = rt6i_table;
587 	rt6->dst.output	= vrf_output6;
588 
589 	/* create a dst for local routing - packets sent locally
590 	 * to local address via the VRF device as a loopback
591 	 */
592 	rt6_local = ip6_dst_alloc(net, dev, flags);
593 	if (!rt6_local) {
594 		dst_release(&rt6->dst);
595 		goto out;
596 	}
597 
598 	dst_hold(&rt6_local->dst);
599 
600 	rt6_local->rt6i_idev  = in6_dev_get(dev);
601 	rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
602 	rt6_local->rt6i_table = rt6i_table;
603 	rt6_local->dst.input  = ip6_input;
604 
605 	rcu_assign_pointer(vrf->rt6, rt6);
606 	rcu_assign_pointer(vrf->rt6_local, rt6_local);
607 
608 	rc = 0;
609 out:
610 	return rc;
611 }
612 #else
613 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
614 				   struct sock *sk,
615 				   struct sk_buff *skb)
616 {
617 	return skb;
618 }
619 
620 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
621 {
622 }
623 
624 static int vrf_rt6_create(struct net_device *dev)
625 {
626 	return 0;
627 }
628 #endif
629 
630 /* modelled after ip_finish_output2 */
631 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
632 {
633 	struct dst_entry *dst = skb_dst(skb);
634 	struct rtable *rt = (struct rtable *)dst;
635 	struct net_device *dev = dst->dev;
636 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
637 	struct neighbour *neigh;
638 	u32 nexthop;
639 	int ret = -EINVAL;
640 
641 	nf_reset(skb);
642 
643 	/* Be paranoid, rather than too clever. */
644 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
645 		struct sk_buff *skb2;
646 
647 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
648 		if (!skb2) {
649 			ret = -ENOMEM;
650 			goto err;
651 		}
652 		if (skb->sk)
653 			skb_set_owner_w(skb2, skb->sk);
654 
655 		consume_skb(skb);
656 		skb = skb2;
657 	}
658 
659 	rcu_read_lock_bh();
660 
661 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
662 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
663 	if (unlikely(!neigh))
664 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
665 	if (!IS_ERR(neigh)) {
666 		sock_confirm_neigh(skb, neigh);
667 		ret = neigh_output(neigh, skb);
668 	}
669 
670 	rcu_read_unlock_bh();
671 err:
672 	if (unlikely(ret < 0))
673 		vrf_tx_error(skb->dev, skb);
674 	return ret;
675 }
676 
677 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
678 {
679 	struct net_device *dev = skb_dst(skb)->dev;
680 
681 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
682 
683 	skb->dev = dev;
684 	skb->protocol = htons(ETH_P_IP);
685 
686 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
687 			    net, sk, skb, NULL, dev,
688 			    vrf_finish_output,
689 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
690 }
691 
692 /* set dst on skb to send packet to us via dev_xmit path. Allows
693  * packet to go through device based features such as qdisc, netfilter
694  * hooks and packet sockets with skb->dev set to vrf device.
695  */
696 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
697 					   struct sk_buff *skb)
698 {
699 	struct net_vrf *vrf = netdev_priv(vrf_dev);
700 	struct dst_entry *dst = NULL;
701 	struct rtable *rth;
702 
703 	rcu_read_lock();
704 
705 	rth = rcu_dereference(vrf->rth);
706 	if (likely(rth)) {
707 		dst = &rth->dst;
708 		dst_hold(dst);
709 	}
710 
711 	rcu_read_unlock();
712 
713 	if (unlikely(!dst)) {
714 		vrf_tx_error(vrf_dev, skb);
715 		return NULL;
716 	}
717 
718 	skb_dst_drop(skb);
719 	skb_dst_set(skb, dst);
720 
721 	return skb;
722 }
723 
724 static int vrf_output_direct(struct net *net, struct sock *sk,
725 			     struct sk_buff *skb)
726 {
727 	skb->protocol = htons(ETH_P_IP);
728 
729 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
730 			    net, sk, skb, NULL, skb->dev,
731 			    vrf_finish_direct,
732 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
733 }
734 
735 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
736 					 struct sock *sk,
737 					 struct sk_buff *skb)
738 {
739 	struct net *net = dev_net(vrf_dev);
740 	int err;
741 
742 	skb->dev = vrf_dev;
743 
744 	err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
745 		      skb, NULL, vrf_dev, vrf_output_direct);
746 
747 	if (likely(err == 1))
748 		err = vrf_output_direct(net, sk, skb);
749 
750 	/* reset skb device */
751 	if (likely(err == 1))
752 		nf_reset(skb);
753 	else
754 		skb = NULL;
755 
756 	return skb;
757 }
758 
759 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
760 				  struct sock *sk,
761 				  struct sk_buff *skb)
762 {
763 	/* don't divert multicast */
764 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
765 		return skb;
766 
767 	if (qdisc_tx_is_default(vrf_dev))
768 		return vrf_ip_out_direct(vrf_dev, sk, skb);
769 
770 	return vrf_ip_out_redirect(vrf_dev, skb);
771 }
772 
773 /* called with rcu lock held */
774 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
775 				  struct sock *sk,
776 				  struct sk_buff *skb,
777 				  u16 proto)
778 {
779 	switch (proto) {
780 	case AF_INET:
781 		return vrf_ip_out(vrf_dev, sk, skb);
782 	case AF_INET6:
783 		return vrf_ip6_out(vrf_dev, sk, skb);
784 	}
785 
786 	return skb;
787 }
788 
789 /* holding rtnl */
790 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
791 {
792 	struct rtable *rth = rtnl_dereference(vrf->rth);
793 	struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
794 	struct net *net = dev_net(dev);
795 	struct dst_entry *dst;
796 
797 	RCU_INIT_POINTER(vrf->rth, NULL);
798 	RCU_INIT_POINTER(vrf->rth_local, NULL);
799 	synchronize_rcu();
800 
801 	/* move dev in dst's to loopback so this VRF device can be deleted
802 	 * - based on dst_ifdown
803 	 */
804 	if (rth) {
805 		dst = &rth->dst;
806 		dev_put(dst->dev);
807 		dst->dev = net->loopback_dev;
808 		dev_hold(dst->dev);
809 		dst_release(dst);
810 	}
811 
812 	if (rth_local) {
813 		dst = &rth_local->dst;
814 		dev_put(dst->dev);
815 		dst->dev = net->loopback_dev;
816 		dev_hold(dst->dev);
817 		dst_release(dst);
818 	}
819 }
820 
821 static int vrf_rtable_create(struct net_device *dev)
822 {
823 	struct net_vrf *vrf = netdev_priv(dev);
824 	struct rtable *rth, *rth_local;
825 
826 	if (!fib_new_table(dev_net(dev), vrf->tb_id))
827 		return -ENOMEM;
828 
829 	/* create a dst for routing packets out through a VRF device */
830 	rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
831 	if (!rth)
832 		return -ENOMEM;
833 
834 	/* create a dst for local ingress routing - packets sent locally
835 	 * to local address via the VRF device as a loopback
836 	 */
837 	rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
838 	if (!rth_local) {
839 		dst_release(&rth->dst);
840 		return -ENOMEM;
841 	}
842 
843 	rth->dst.output	= vrf_output;
844 	rth->rt_table_id = vrf->tb_id;
845 
846 	rth_local->rt_table_id = vrf->tb_id;
847 
848 	rcu_assign_pointer(vrf->rth, rth);
849 	rcu_assign_pointer(vrf->rth_local, rth_local);
850 
851 	return 0;
852 }
853 
854 /**************************** device handling ********************/
855 
856 /* cycle interface to flush neighbor cache and move routes across tables */
857 static void cycle_netdev(struct net_device *dev)
858 {
859 	unsigned int flags = dev->flags;
860 	int ret;
861 
862 	if (!netif_running(dev))
863 		return;
864 
865 	ret = dev_change_flags(dev, flags & ~IFF_UP);
866 	if (ret >= 0)
867 		ret = dev_change_flags(dev, flags);
868 
869 	if (ret < 0) {
870 		netdev_err(dev,
871 			   "Failed to cycle device %s; route tables might be wrong!\n",
872 			   dev->name);
873 	}
874 }
875 
876 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
877 {
878 	int ret;
879 
880 	/* do not allow loopback device to be enslaved to a VRF.
881 	 * The vrf device acts as the loopback for the vrf.
882 	 */
883 	if (port_dev == dev_net(dev)->loopback_dev)
884 		return -EOPNOTSUPP;
885 
886 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
887 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
888 	if (ret < 0)
889 		goto err;
890 
891 	cycle_netdev(port_dev);
892 
893 	return 0;
894 
895 err:
896 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
897 	return ret;
898 }
899 
900 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
901 {
902 	if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
903 		return -EINVAL;
904 
905 	return do_vrf_add_slave(dev, port_dev);
906 }
907 
908 /* inverse of do_vrf_add_slave */
909 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
910 {
911 	netdev_upper_dev_unlink(port_dev, dev);
912 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
913 
914 	cycle_netdev(port_dev);
915 
916 	return 0;
917 }
918 
919 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
920 {
921 	return do_vrf_del_slave(dev, port_dev);
922 }
923 
924 static void vrf_dev_uninit(struct net_device *dev)
925 {
926 	struct net_vrf *vrf = netdev_priv(dev);
927 	struct net_device *port_dev;
928 	struct list_head *iter;
929 
930 	vrf_rtable_release(dev, vrf);
931 	vrf_rt6_release(dev, vrf);
932 
933 	netdev_for_each_lower_dev(dev, port_dev, iter)
934 		vrf_del_slave(dev, port_dev);
935 
936 	free_percpu(dev->dstats);
937 	dev->dstats = NULL;
938 }
939 
940 static int vrf_dev_init(struct net_device *dev)
941 {
942 	struct net_vrf *vrf = netdev_priv(dev);
943 
944 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
945 	if (!dev->dstats)
946 		goto out_nomem;
947 
948 	/* create the default dst which points back to us */
949 	if (vrf_rtable_create(dev) != 0)
950 		goto out_stats;
951 
952 	if (vrf_rt6_create(dev) != 0)
953 		goto out_rth;
954 
955 	dev->flags = IFF_MASTER | IFF_NOARP;
956 
957 	/* MTU is irrelevant for VRF device; set to 64k similar to lo */
958 	dev->mtu = 64 * 1024;
959 
960 	/* similarly, oper state is irrelevant; set to up to avoid confusion */
961 	dev->operstate = IF_OPER_UP;
962 	netdev_lockdep_set_classes(dev);
963 	return 0;
964 
965 out_rth:
966 	vrf_rtable_release(dev, vrf);
967 out_stats:
968 	free_percpu(dev->dstats);
969 	dev->dstats = NULL;
970 out_nomem:
971 	return -ENOMEM;
972 }
973 
974 static const struct net_device_ops vrf_netdev_ops = {
975 	.ndo_init		= vrf_dev_init,
976 	.ndo_uninit		= vrf_dev_uninit,
977 	.ndo_start_xmit		= vrf_xmit,
978 	.ndo_get_stats64	= vrf_get_stats64,
979 	.ndo_add_slave		= vrf_add_slave,
980 	.ndo_del_slave		= vrf_del_slave,
981 };
982 
983 static u32 vrf_fib_table(const struct net_device *dev)
984 {
985 	struct net_vrf *vrf = netdev_priv(dev);
986 
987 	return vrf->tb_id;
988 }
989 
990 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
991 {
992 	kfree_skb(skb);
993 	return 0;
994 }
995 
996 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
997 				      struct sk_buff *skb,
998 				      struct net_device *dev)
999 {
1000 	struct net *net = dev_net(dev);
1001 
1002 	if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1003 		skb = NULL;    /* kfree_skb(skb) handled by nf code */
1004 
1005 	return skb;
1006 }
1007 
1008 #if IS_ENABLED(CONFIG_IPV6)
1009 /* neighbor handling is done with actual device; do not want
1010  * to flip skb->dev for those ndisc packets. This really fails
1011  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1012  * a start.
1013  */
1014 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1015 {
1016 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1017 	bool rc = false;
1018 
1019 	if (iph->nexthdr == NEXTHDR_ICMP) {
1020 		const struct icmp6hdr *icmph;
1021 		struct icmp6hdr _icmph;
1022 
1023 		icmph = skb_header_pointer(skb, sizeof(*iph),
1024 					   sizeof(_icmph), &_icmph);
1025 		if (!icmph)
1026 			goto out;
1027 
1028 		switch (icmph->icmp6_type) {
1029 		case NDISC_ROUTER_SOLICITATION:
1030 		case NDISC_ROUTER_ADVERTISEMENT:
1031 		case NDISC_NEIGHBOUR_SOLICITATION:
1032 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
1033 		case NDISC_REDIRECT:
1034 			rc = true;
1035 			break;
1036 		}
1037 	}
1038 
1039 out:
1040 	return rc;
1041 }
1042 
1043 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1044 					     const struct net_device *dev,
1045 					     struct flowi6 *fl6,
1046 					     int ifindex,
1047 					     int flags)
1048 {
1049 	struct net_vrf *vrf = netdev_priv(dev);
1050 	struct fib6_table *table = NULL;
1051 	struct rt6_info *rt6;
1052 
1053 	rcu_read_lock();
1054 
1055 	/* fib6_table does not have a refcnt and can not be freed */
1056 	rt6 = rcu_dereference(vrf->rt6);
1057 	if (likely(rt6))
1058 		table = rt6->rt6i_table;
1059 
1060 	rcu_read_unlock();
1061 
1062 	if (!table)
1063 		return NULL;
1064 
1065 	return ip6_pol_route(net, table, ifindex, fl6, flags);
1066 }
1067 
1068 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1069 			      int ifindex)
1070 {
1071 	const struct ipv6hdr *iph = ipv6_hdr(skb);
1072 	struct flowi6 fl6 = {
1073 		.daddr          = iph->daddr,
1074 		.saddr          = iph->saddr,
1075 		.flowlabel      = ip6_flowinfo(iph),
1076 		.flowi6_mark    = skb->mark,
1077 		.flowi6_proto   = iph->nexthdr,
1078 		.flowi6_iif     = ifindex,
1079 	};
1080 	struct net *net = dev_net(vrf_dev);
1081 	struct rt6_info *rt6;
1082 
1083 	rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
1084 				   RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1085 	if (unlikely(!rt6))
1086 		return;
1087 
1088 	if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1089 		return;
1090 
1091 	skb_dst_set(skb, &rt6->dst);
1092 }
1093 
1094 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1095 				   struct sk_buff *skb)
1096 {
1097 	int orig_iif = skb->skb_iif;
1098 	bool need_strict;
1099 
1100 	/* loopback traffic; do not push through packet taps again.
1101 	 * Reset pkt_type for upper layers to process skb
1102 	 */
1103 	if (skb->pkt_type == PACKET_LOOPBACK) {
1104 		skb->dev = vrf_dev;
1105 		skb->skb_iif = vrf_dev->ifindex;
1106 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1107 		skb->pkt_type = PACKET_HOST;
1108 		goto out;
1109 	}
1110 
1111 	/* if packet is NDISC or addressed to multicast or link-local
1112 	 * then keep the ingress interface
1113 	 */
1114 	need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1115 	if (!ipv6_ndisc_frame(skb) && !need_strict) {
1116 		vrf_rx_stats(vrf_dev, skb->len);
1117 		skb->dev = vrf_dev;
1118 		skb->skb_iif = vrf_dev->ifindex;
1119 
1120 		if (!list_empty(&vrf_dev->ptype_all)) {
1121 			skb_push(skb, skb->mac_len);
1122 			dev_queue_xmit_nit(skb, vrf_dev);
1123 			skb_pull(skb, skb->mac_len);
1124 		}
1125 
1126 		IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1127 	}
1128 
1129 	if (need_strict)
1130 		vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1131 
1132 	skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1133 out:
1134 	return skb;
1135 }
1136 
1137 #else
1138 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1139 				   struct sk_buff *skb)
1140 {
1141 	return skb;
1142 }
1143 #endif
1144 
1145 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1146 				  struct sk_buff *skb)
1147 {
1148 	skb->dev = vrf_dev;
1149 	skb->skb_iif = vrf_dev->ifindex;
1150 	IPCB(skb)->flags |= IPSKB_L3SLAVE;
1151 
1152 	if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1153 		goto out;
1154 
1155 	/* loopback traffic; do not push through packet taps again.
1156 	 * Reset pkt_type for upper layers to process skb
1157 	 */
1158 	if (skb->pkt_type == PACKET_LOOPBACK) {
1159 		skb->pkt_type = PACKET_HOST;
1160 		goto out;
1161 	}
1162 
1163 	vrf_rx_stats(vrf_dev, skb->len);
1164 
1165 	if (!list_empty(&vrf_dev->ptype_all)) {
1166 		skb_push(skb, skb->mac_len);
1167 		dev_queue_xmit_nit(skb, vrf_dev);
1168 		skb_pull(skb, skb->mac_len);
1169 	}
1170 
1171 	skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1172 out:
1173 	return skb;
1174 }
1175 
1176 /* called with rcu lock held */
1177 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1178 				  struct sk_buff *skb,
1179 				  u16 proto)
1180 {
1181 	switch (proto) {
1182 	case AF_INET:
1183 		return vrf_ip_rcv(vrf_dev, skb);
1184 	case AF_INET6:
1185 		return vrf_ip6_rcv(vrf_dev, skb);
1186 	}
1187 
1188 	return skb;
1189 }
1190 
1191 #if IS_ENABLED(CONFIG_IPV6)
1192 /* send to link-local or multicast address via interface enslaved to
1193  * VRF device. Force lookup to VRF table without changing flow struct
1194  */
1195 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1196 					      struct flowi6 *fl6)
1197 {
1198 	struct net *net = dev_net(dev);
1199 	int flags = RT6_LOOKUP_F_IFACE;
1200 	struct dst_entry *dst = NULL;
1201 	struct rt6_info *rt;
1202 
1203 	/* VRF device does not have a link-local address and
1204 	 * sending packets to link-local or mcast addresses over
1205 	 * a VRF device does not make sense
1206 	 */
1207 	if (fl6->flowi6_oif == dev->ifindex) {
1208 		dst = &net->ipv6.ip6_null_entry->dst;
1209 		dst_hold(dst);
1210 		return dst;
1211 	}
1212 
1213 	if (!ipv6_addr_any(&fl6->saddr))
1214 		flags |= RT6_LOOKUP_F_HAS_SADDR;
1215 
1216 	rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1217 	if (rt)
1218 		dst = &rt->dst;
1219 
1220 	return dst;
1221 }
1222 #endif
1223 
1224 static const struct l3mdev_ops vrf_l3mdev_ops = {
1225 	.l3mdev_fib_table	= vrf_fib_table,
1226 	.l3mdev_l3_rcv		= vrf_l3_rcv,
1227 	.l3mdev_l3_out		= vrf_l3_out,
1228 #if IS_ENABLED(CONFIG_IPV6)
1229 	.l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1230 #endif
1231 };
1232 
1233 static void vrf_get_drvinfo(struct net_device *dev,
1234 			    struct ethtool_drvinfo *info)
1235 {
1236 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1237 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1238 }
1239 
1240 static const struct ethtool_ops vrf_ethtool_ops = {
1241 	.get_drvinfo	= vrf_get_drvinfo,
1242 };
1243 
1244 static inline size_t vrf_fib_rule_nl_size(void)
1245 {
1246 	size_t sz;
1247 
1248 	sz  = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1249 	sz += nla_total_size(sizeof(u8));	/* FRA_L3MDEV */
1250 	sz += nla_total_size(sizeof(u32));	/* FRA_PRIORITY */
1251 
1252 	return sz;
1253 }
1254 
1255 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1256 {
1257 	struct fib_rule_hdr *frh;
1258 	struct nlmsghdr *nlh;
1259 	struct sk_buff *skb;
1260 	int err;
1261 
1262 	if (family == AF_INET6 && !ipv6_mod_enabled())
1263 		return 0;
1264 
1265 	skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1266 	if (!skb)
1267 		return -ENOMEM;
1268 
1269 	nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1270 	if (!nlh)
1271 		goto nla_put_failure;
1272 
1273 	/* rule only needs to appear once */
1274 	nlh->nlmsg_flags |= NLM_F_EXCL;
1275 
1276 	frh = nlmsg_data(nlh);
1277 	memset(frh, 0, sizeof(*frh));
1278 	frh->family = family;
1279 	frh->action = FR_ACT_TO_TBL;
1280 
1281 	if (nla_put_u32(skb, FRA_L3MDEV, 1))
1282 		goto nla_put_failure;
1283 
1284 	if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1285 		goto nla_put_failure;
1286 
1287 	nlmsg_end(skb, nlh);
1288 
1289 	/* fib_nl_{new,del}rule handling looks for net from skb->sk */
1290 	skb->sk = dev_net(dev)->rtnl;
1291 	if (add_it) {
1292 		err = fib_nl_newrule(skb, nlh, NULL);
1293 		if (err == -EEXIST)
1294 			err = 0;
1295 	} else {
1296 		err = fib_nl_delrule(skb, nlh, NULL);
1297 		if (err == -ENOENT)
1298 			err = 0;
1299 	}
1300 	nlmsg_free(skb);
1301 
1302 	return err;
1303 
1304 nla_put_failure:
1305 	nlmsg_free(skb);
1306 
1307 	return -EMSGSIZE;
1308 }
1309 
1310 static int vrf_add_fib_rules(const struct net_device *dev)
1311 {
1312 	int err;
1313 
1314 	err = vrf_fib_rule(dev, AF_INET,  true);
1315 	if (err < 0)
1316 		goto out_err;
1317 
1318 	err = vrf_fib_rule(dev, AF_INET6, true);
1319 	if (err < 0)
1320 		goto ipv6_err;
1321 
1322 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1323 	err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1324 	if (err < 0)
1325 		goto ipmr_err;
1326 #endif
1327 
1328 	return 0;
1329 
1330 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1331 ipmr_err:
1332 	vrf_fib_rule(dev, AF_INET6,  false);
1333 #endif
1334 
1335 ipv6_err:
1336 	vrf_fib_rule(dev, AF_INET,  false);
1337 
1338 out_err:
1339 	netdev_err(dev, "Failed to add FIB rules.\n");
1340 	return err;
1341 }
1342 
1343 static void vrf_setup(struct net_device *dev)
1344 {
1345 	ether_setup(dev);
1346 
1347 	/* Initialize the device structure. */
1348 	dev->netdev_ops = &vrf_netdev_ops;
1349 	dev->l3mdev_ops = &vrf_l3mdev_ops;
1350 	dev->ethtool_ops = &vrf_ethtool_ops;
1351 	dev->destructor = free_netdev;
1352 
1353 	/* Fill in device structure with ethernet-generic values. */
1354 	eth_hw_addr_random(dev);
1355 
1356 	/* don't acquire vrf device's netif_tx_lock when transmitting */
1357 	dev->features |= NETIF_F_LLTX;
1358 
1359 	/* don't allow vrf devices to change network namespaces. */
1360 	dev->features |= NETIF_F_NETNS_LOCAL;
1361 
1362 	/* does not make sense for a VLAN to be added to a vrf device */
1363 	dev->features   |= NETIF_F_VLAN_CHALLENGED;
1364 
1365 	/* enable offload features */
1366 	dev->features   |= NETIF_F_GSO_SOFTWARE;
1367 	dev->features   |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1368 	dev->features   |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1369 
1370 	dev->hw_features = dev->features;
1371 	dev->hw_enc_features = dev->features;
1372 
1373 	/* default to no qdisc; user can add if desired */
1374 	dev->priv_flags |= IFF_NO_QUEUE;
1375 }
1376 
1377 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1378 {
1379 	if (tb[IFLA_ADDRESS]) {
1380 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1381 			return -EINVAL;
1382 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1383 			return -EADDRNOTAVAIL;
1384 	}
1385 	return 0;
1386 }
1387 
1388 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1389 {
1390 	unregister_netdevice_queue(dev, head);
1391 }
1392 
1393 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1394 		       struct nlattr *tb[], struct nlattr *data[])
1395 {
1396 	struct net_vrf *vrf = netdev_priv(dev);
1397 	int err;
1398 
1399 	if (!data || !data[IFLA_VRF_TABLE])
1400 		return -EINVAL;
1401 
1402 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1403 	if (vrf->tb_id == RT_TABLE_UNSPEC)
1404 		return -EINVAL;
1405 
1406 	dev->priv_flags |= IFF_L3MDEV_MASTER;
1407 
1408 	err = register_netdevice(dev);
1409 	if (err)
1410 		goto out;
1411 
1412 	if (add_fib_rules) {
1413 		err = vrf_add_fib_rules(dev);
1414 		if (err) {
1415 			unregister_netdevice(dev);
1416 			goto out;
1417 		}
1418 		add_fib_rules = false;
1419 	}
1420 
1421 out:
1422 	return err;
1423 }
1424 
1425 static size_t vrf_nl_getsize(const struct net_device *dev)
1426 {
1427 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
1428 }
1429 
1430 static int vrf_fillinfo(struct sk_buff *skb,
1431 			const struct net_device *dev)
1432 {
1433 	struct net_vrf *vrf = netdev_priv(dev);
1434 
1435 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1436 }
1437 
1438 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1439 				 const struct net_device *slave_dev)
1440 {
1441 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_PORT_TABLE */
1442 }
1443 
1444 static int vrf_fill_slave_info(struct sk_buff *skb,
1445 			       const struct net_device *vrf_dev,
1446 			       const struct net_device *slave_dev)
1447 {
1448 	struct net_vrf *vrf = netdev_priv(vrf_dev);
1449 
1450 	if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1451 		return -EMSGSIZE;
1452 
1453 	return 0;
1454 }
1455 
1456 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1457 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
1458 };
1459 
1460 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1461 	.kind		= DRV_NAME,
1462 	.priv_size	= sizeof(struct net_vrf),
1463 
1464 	.get_size	= vrf_nl_getsize,
1465 	.policy		= vrf_nl_policy,
1466 	.validate	= vrf_validate,
1467 	.fill_info	= vrf_fillinfo,
1468 
1469 	.get_slave_size  = vrf_get_slave_size,
1470 	.fill_slave_info = vrf_fill_slave_info,
1471 
1472 	.newlink	= vrf_newlink,
1473 	.dellink	= vrf_dellink,
1474 	.setup		= vrf_setup,
1475 	.maxtype	= IFLA_VRF_MAX,
1476 };
1477 
1478 static int vrf_device_event(struct notifier_block *unused,
1479 			    unsigned long event, void *ptr)
1480 {
1481 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1482 
1483 	/* only care about unregister events to drop slave references */
1484 	if (event == NETDEV_UNREGISTER) {
1485 		struct net_device *vrf_dev;
1486 
1487 		if (!netif_is_l3_slave(dev))
1488 			goto out;
1489 
1490 		vrf_dev = netdev_master_upper_dev_get(dev);
1491 		vrf_del_slave(vrf_dev, dev);
1492 	}
1493 out:
1494 	return NOTIFY_DONE;
1495 }
1496 
1497 static struct notifier_block vrf_notifier_block __read_mostly = {
1498 	.notifier_call = vrf_device_event,
1499 };
1500 
1501 static int __init vrf_init_module(void)
1502 {
1503 	int rc;
1504 
1505 	register_netdevice_notifier(&vrf_notifier_block);
1506 
1507 	rc = rtnl_link_register(&vrf_link_ops);
1508 	if (rc < 0)
1509 		goto error;
1510 
1511 	return 0;
1512 
1513 error:
1514 	unregister_netdevice_notifier(&vrf_notifier_block);
1515 	return rc;
1516 }
1517 
1518 module_init(vrf_init_module);
1519 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1520 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1521 MODULE_LICENSE("GPL");
1522 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1523 MODULE_VERSION(DRV_VERSION);
1524