xref: /linux/drivers/net/vrf.c (revision 8c749ce93ee69e789e46b3be98de9e0cbfcf8ed8)
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/rtnetlink.h>
36 #include <net/route.h>
37 #include <net/addrconf.h>
38 #include <net/l3mdev.h>
39 
40 #define RT_FL_TOS(oldflp4) \
41 	((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
42 
43 #define DRV_NAME	"vrf"
44 #define DRV_VERSION	"1.0"
45 
46 #define vrf_master_get_rcu(dev) \
47 	((struct net_device *)rcu_dereference(dev->rx_handler_data))
48 
49 struct net_vrf {
50 	struct rtable           *rth;
51 	struct rt6_info		*rt6;
52 	u32                     tb_id;
53 };
54 
55 struct pcpu_dstats {
56 	u64			tx_pkts;
57 	u64			tx_bytes;
58 	u64			tx_drps;
59 	u64			rx_pkts;
60 	u64			rx_bytes;
61 	struct u64_stats_sync	syncp;
62 };
63 
64 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
65 {
66 	return dst;
67 }
68 
69 static int vrf_ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
70 {
71 	return ip_local_out(net, sk, skb);
72 }
73 
74 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
75 {
76 	/* TO-DO: return max ethernet size? */
77 	return dst->dev->mtu;
78 }
79 
80 static void vrf_dst_destroy(struct dst_entry *dst)
81 {
82 	/* our dst lives forever - or until the device is closed */
83 }
84 
85 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
86 {
87 	return 65535 - 40;
88 }
89 
90 static struct dst_ops vrf_dst_ops = {
91 	.family		= AF_INET,
92 	.local_out	= vrf_ip_local_out,
93 	.check		= vrf_ip_check,
94 	.mtu		= vrf_v4_mtu,
95 	.destroy	= vrf_dst_destroy,
96 	.default_advmss	= vrf_default_advmss,
97 };
98 
99 /* neighbor handling is done with actual device; do not want
100  * to flip skb->dev for those ndisc packets. This really fails
101  * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
102  * a start.
103  */
104 #if IS_ENABLED(CONFIG_IPV6)
105 static bool check_ipv6_frame(const struct sk_buff *skb)
106 {
107 	const struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb->data;
108 	size_t hlen = sizeof(*ipv6h);
109 	bool rc = true;
110 
111 	if (skb->len < hlen)
112 		goto out;
113 
114 	if (ipv6h->nexthdr == NEXTHDR_ICMP) {
115 		const struct icmp6hdr *icmph;
116 
117 		if (skb->len < hlen + sizeof(*icmph))
118 			goto out;
119 
120 		icmph = (struct icmp6hdr *)(skb->data + sizeof(*ipv6h));
121 		switch (icmph->icmp6_type) {
122 		case NDISC_ROUTER_SOLICITATION:
123 		case NDISC_ROUTER_ADVERTISEMENT:
124 		case NDISC_NEIGHBOUR_SOLICITATION:
125 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
126 		case NDISC_REDIRECT:
127 			rc = false;
128 			break;
129 		}
130 	}
131 
132 out:
133 	return rc;
134 }
135 #else
136 static bool check_ipv6_frame(const struct sk_buff *skb)
137 {
138 	return false;
139 }
140 #endif
141 
142 static bool is_ip_rx_frame(struct sk_buff *skb)
143 {
144 	switch (skb->protocol) {
145 	case htons(ETH_P_IP):
146 		return true;
147 	case htons(ETH_P_IPV6):
148 		return check_ipv6_frame(skb);
149 	}
150 	return false;
151 }
152 
153 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
154 {
155 	vrf_dev->stats.tx_errors++;
156 	kfree_skb(skb);
157 }
158 
159 /* note: already called with rcu_read_lock */
160 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
161 {
162 	struct sk_buff *skb = *pskb;
163 
164 	if (is_ip_rx_frame(skb)) {
165 		struct net_device *dev = vrf_master_get_rcu(skb->dev);
166 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
167 
168 		u64_stats_update_begin(&dstats->syncp);
169 		dstats->rx_pkts++;
170 		dstats->rx_bytes += skb->len;
171 		u64_stats_update_end(&dstats->syncp);
172 
173 		skb->dev = dev;
174 
175 		return RX_HANDLER_ANOTHER;
176 	}
177 	return RX_HANDLER_PASS;
178 }
179 
180 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
181 						 struct rtnl_link_stats64 *stats)
182 {
183 	int i;
184 
185 	for_each_possible_cpu(i) {
186 		const struct pcpu_dstats *dstats;
187 		u64 tbytes, tpkts, tdrops, rbytes, rpkts;
188 		unsigned int start;
189 
190 		dstats = per_cpu_ptr(dev->dstats, i);
191 		do {
192 			start = u64_stats_fetch_begin_irq(&dstats->syncp);
193 			tbytes = dstats->tx_bytes;
194 			tpkts = dstats->tx_pkts;
195 			tdrops = dstats->tx_drps;
196 			rbytes = dstats->rx_bytes;
197 			rpkts = dstats->rx_pkts;
198 		} while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
199 		stats->tx_bytes += tbytes;
200 		stats->tx_packets += tpkts;
201 		stats->tx_dropped += tdrops;
202 		stats->rx_bytes += rbytes;
203 		stats->rx_packets += rpkts;
204 	}
205 	return stats;
206 }
207 
208 #if IS_ENABLED(CONFIG_IPV6)
209 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
210 					   struct net_device *dev)
211 {
212 	const struct ipv6hdr *iph = ipv6_hdr(skb);
213 	struct net *net = dev_net(skb->dev);
214 	struct flowi6 fl6 = {
215 		/* needed to match OIF rule */
216 		.flowi6_oif = dev->ifindex,
217 		.flowi6_iif = LOOPBACK_IFINDEX,
218 		.daddr = iph->daddr,
219 		.saddr = iph->saddr,
220 		.flowlabel = ip6_flowinfo(iph),
221 		.flowi6_mark = skb->mark,
222 		.flowi6_proto = iph->nexthdr,
223 		.flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
224 	};
225 	int ret = NET_XMIT_DROP;
226 	struct dst_entry *dst;
227 	struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
228 
229 	dst = ip6_route_output(net, NULL, &fl6);
230 	if (dst == dst_null)
231 		goto err;
232 
233 	skb_dst_drop(skb);
234 	skb_dst_set(skb, dst);
235 
236 	ret = ip6_local_out(net, skb->sk, skb);
237 	if (unlikely(net_xmit_eval(ret)))
238 		dev->stats.tx_errors++;
239 	else
240 		ret = NET_XMIT_SUCCESS;
241 
242 	return ret;
243 err:
244 	vrf_tx_error(dev, skb);
245 	return NET_XMIT_DROP;
246 }
247 #else
248 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
249 					   struct net_device *dev)
250 {
251 	vrf_tx_error(dev, skb);
252 	return NET_XMIT_DROP;
253 }
254 #endif
255 
256 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
257 			    struct net_device *vrf_dev)
258 {
259 	struct rtable *rt;
260 	int err = 1;
261 
262 	rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
263 	if (IS_ERR(rt))
264 		goto out;
265 
266 	/* TO-DO: what about broadcast ? */
267 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
268 		ip_rt_put(rt);
269 		goto out;
270 	}
271 
272 	skb_dst_drop(skb);
273 	skb_dst_set(skb, &rt->dst);
274 	err = 0;
275 out:
276 	return err;
277 }
278 
279 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
280 					   struct net_device *vrf_dev)
281 {
282 	struct iphdr *ip4h = ip_hdr(skb);
283 	int ret = NET_XMIT_DROP;
284 	struct flowi4 fl4 = {
285 		/* needed to match OIF rule */
286 		.flowi4_oif = vrf_dev->ifindex,
287 		.flowi4_iif = LOOPBACK_IFINDEX,
288 		.flowi4_tos = RT_TOS(ip4h->tos),
289 		.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
290 				FLOWI_FLAG_SKIP_NH_OIF,
291 		.daddr = ip4h->daddr,
292 	};
293 
294 	if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
295 		goto err;
296 
297 	if (!ip4h->saddr) {
298 		ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
299 					       RT_SCOPE_LINK);
300 	}
301 
302 	ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
303 	if (unlikely(net_xmit_eval(ret)))
304 		vrf_dev->stats.tx_errors++;
305 	else
306 		ret = NET_XMIT_SUCCESS;
307 
308 out:
309 	return ret;
310 err:
311 	vrf_tx_error(vrf_dev, skb);
312 	goto out;
313 }
314 
315 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
316 {
317 	/* strip the ethernet header added for pass through VRF device */
318 	__skb_pull(skb, skb_network_offset(skb));
319 
320 	switch (skb->protocol) {
321 	case htons(ETH_P_IP):
322 		return vrf_process_v4_outbound(skb, dev);
323 	case htons(ETH_P_IPV6):
324 		return vrf_process_v6_outbound(skb, dev);
325 	default:
326 		vrf_tx_error(dev, skb);
327 		return NET_XMIT_DROP;
328 	}
329 }
330 
331 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
332 {
333 	netdev_tx_t ret = is_ip_tx_frame(skb, dev);
334 
335 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
336 		struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
337 
338 		u64_stats_update_begin(&dstats->syncp);
339 		dstats->tx_pkts++;
340 		dstats->tx_bytes += skb->len;
341 		u64_stats_update_end(&dstats->syncp);
342 	} else {
343 		this_cpu_inc(dev->dstats->tx_drps);
344 	}
345 
346 	return ret;
347 }
348 
349 #if IS_ENABLED(CONFIG_IPV6)
350 static struct dst_entry *vrf_ip6_check(struct dst_entry *dst, u32 cookie)
351 {
352 	return dst;
353 }
354 
355 static struct dst_ops vrf_dst_ops6 = {
356 	.family		= AF_INET6,
357 	.local_out	= ip6_local_out,
358 	.check		= vrf_ip6_check,
359 	.mtu		= vrf_v4_mtu,
360 	.destroy	= vrf_dst_destroy,
361 	.default_advmss	= vrf_default_advmss,
362 };
363 
364 static int init_dst_ops6_kmem_cachep(void)
365 {
366 	vrf_dst_ops6.kmem_cachep = kmem_cache_create("vrf_ip6_dst_cache",
367 						     sizeof(struct rt6_info),
368 						     0,
369 						     SLAB_HWCACHE_ALIGN,
370 						     NULL);
371 
372 	if (!vrf_dst_ops6.kmem_cachep)
373 		return -ENOMEM;
374 
375 	return 0;
376 }
377 
378 static void free_dst_ops6_kmem_cachep(void)
379 {
380 	kmem_cache_destroy(vrf_dst_ops6.kmem_cachep);
381 }
382 
383 static int vrf_input6(struct sk_buff *skb)
384 {
385 	skb->dev->stats.rx_errors++;
386 	kfree_skb(skb);
387 	return 0;
388 }
389 
390 /* modelled after ip6_finish_output2 */
391 static int vrf_finish_output6(struct net *net, struct sock *sk,
392 			      struct sk_buff *skb)
393 {
394 	struct dst_entry *dst = skb_dst(skb);
395 	struct net_device *dev = dst->dev;
396 	struct neighbour *neigh;
397 	struct in6_addr *nexthop;
398 	int ret;
399 
400 	skb->protocol = htons(ETH_P_IPV6);
401 	skb->dev = dev;
402 
403 	rcu_read_lock_bh();
404 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
405 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
406 	if (unlikely(!neigh))
407 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
408 	if (!IS_ERR(neigh)) {
409 		ret = dst_neigh_output(dst, neigh, skb);
410 		rcu_read_unlock_bh();
411 		return ret;
412 	}
413 	rcu_read_unlock_bh();
414 
415 	IP6_INC_STATS(dev_net(dst->dev),
416 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
417 	kfree_skb(skb);
418 	return -EINVAL;
419 }
420 
421 /* modelled after ip6_output */
422 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
423 {
424 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
425 			    net, sk, skb, NULL, skb_dst(skb)->dev,
426 			    vrf_finish_output6,
427 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
428 }
429 
430 static void vrf_rt6_destroy(struct net_vrf *vrf)
431 {
432 	dst_destroy(&vrf->rt6->dst);
433 	free_percpu(vrf->rt6->rt6i_pcpu);
434 	vrf->rt6 = NULL;
435 }
436 
437 static int vrf_rt6_create(struct net_device *dev)
438 {
439 	struct net_vrf *vrf = netdev_priv(dev);
440 	struct dst_entry *dst;
441 	struct rt6_info *rt6;
442 	int cpu;
443 	int rc = -ENOMEM;
444 
445 	rt6 = dst_alloc(&vrf_dst_ops6, dev, 0,
446 			DST_OBSOLETE_NONE,
447 			(DST_HOST | DST_NOPOLICY | DST_NOXFRM));
448 	if (!rt6)
449 		goto out;
450 
451 	dst = &rt6->dst;
452 
453 	rt6->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, GFP_KERNEL);
454 	if (!rt6->rt6i_pcpu) {
455 		dst_destroy(dst);
456 		goto out;
457 	}
458 	for_each_possible_cpu(cpu) {
459 		struct rt6_info **p = per_cpu_ptr(rt6->rt6i_pcpu, cpu);
460 		*p =  NULL;
461 	}
462 
463 	memset(dst + 1, 0, sizeof(*rt6) - sizeof(*dst));
464 
465 	INIT_LIST_HEAD(&rt6->rt6i_siblings);
466 	INIT_LIST_HEAD(&rt6->rt6i_uncached);
467 
468 	rt6->dst.input	= vrf_input6;
469 	rt6->dst.output	= vrf_output6;
470 
471 	rt6->rt6i_table = fib6_get_table(dev_net(dev), vrf->tb_id);
472 
473 	atomic_set(&rt6->dst.__refcnt, 2);
474 
475 	vrf->rt6 = rt6;
476 	rc = 0;
477 out:
478 	return rc;
479 }
480 #else
481 static int init_dst_ops6_kmem_cachep(void)
482 {
483 	return 0;
484 }
485 
486 static void free_dst_ops6_kmem_cachep(void)
487 {
488 }
489 
490 static void vrf_rt6_destroy(struct net_vrf *vrf)
491 {
492 }
493 
494 static int vrf_rt6_create(struct net_device *dev)
495 {
496 	return 0;
497 }
498 #endif
499 
500 /* modelled after ip_finish_output2 */
501 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
502 {
503 	struct dst_entry *dst = skb_dst(skb);
504 	struct rtable *rt = (struct rtable *)dst;
505 	struct net_device *dev = dst->dev;
506 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
507 	struct neighbour *neigh;
508 	u32 nexthop;
509 	int ret = -EINVAL;
510 
511 	/* Be paranoid, rather than too clever. */
512 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
513 		struct sk_buff *skb2;
514 
515 		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
516 		if (!skb2) {
517 			ret = -ENOMEM;
518 			goto err;
519 		}
520 		if (skb->sk)
521 			skb_set_owner_w(skb2, skb->sk);
522 
523 		consume_skb(skb);
524 		skb = skb2;
525 	}
526 
527 	rcu_read_lock_bh();
528 
529 	nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
530 	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
531 	if (unlikely(!neigh))
532 		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
533 	if (!IS_ERR(neigh))
534 		ret = dst_neigh_output(dst, neigh, skb);
535 
536 	rcu_read_unlock_bh();
537 err:
538 	if (unlikely(ret < 0))
539 		vrf_tx_error(skb->dev, skb);
540 	return ret;
541 }
542 
543 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
544 {
545 	struct net_device *dev = skb_dst(skb)->dev;
546 
547 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
548 
549 	skb->dev = dev;
550 	skb->protocol = htons(ETH_P_IP);
551 
552 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
553 			    net, sk, skb, NULL, dev,
554 			    vrf_finish_output,
555 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
556 }
557 
558 static void vrf_rtable_destroy(struct net_vrf *vrf)
559 {
560 	struct dst_entry *dst = (struct dst_entry *)vrf->rth;
561 
562 	dst_destroy(dst);
563 	vrf->rth = NULL;
564 }
565 
566 static struct rtable *vrf_rtable_create(struct net_device *dev)
567 {
568 	struct net_vrf *vrf = netdev_priv(dev);
569 	struct rtable *rth;
570 
571 	rth = dst_alloc(&vrf_dst_ops, dev, 2,
572 			DST_OBSOLETE_NONE,
573 			(DST_HOST | DST_NOPOLICY | DST_NOXFRM));
574 	if (rth) {
575 		rth->dst.output	= vrf_output;
576 		rth->rt_genid	= rt_genid_ipv4(dev_net(dev));
577 		rth->rt_flags	= 0;
578 		rth->rt_type	= RTN_UNICAST;
579 		rth->rt_is_input = 0;
580 		rth->rt_iif	= 0;
581 		rth->rt_pmtu	= 0;
582 		rth->rt_gateway	= 0;
583 		rth->rt_uses_gateway = 0;
584 		rth->rt_table_id = vrf->tb_id;
585 		INIT_LIST_HEAD(&rth->rt_uncached);
586 		rth->rt_uncached_list = NULL;
587 	}
588 
589 	return rth;
590 }
591 
592 /**************************** device handling ********************/
593 
594 /* cycle interface to flush neighbor cache and move routes across tables */
595 static void cycle_netdev(struct net_device *dev)
596 {
597 	unsigned int flags = dev->flags;
598 	int ret;
599 
600 	if (!netif_running(dev))
601 		return;
602 
603 	ret = dev_change_flags(dev, flags & ~IFF_UP);
604 	if (ret >= 0)
605 		ret = dev_change_flags(dev, flags);
606 
607 	if (ret < 0) {
608 		netdev_err(dev,
609 			   "Failed to cycle device %s; route tables might be wrong!\n",
610 			   dev->name);
611 	}
612 }
613 
614 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
615 {
616 	int ret;
617 
618 	/* register the packet handler for slave ports */
619 	ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
620 	if (ret) {
621 		netdev_err(port_dev,
622 			   "Device %s failed to register rx_handler\n",
623 			   port_dev->name);
624 		goto out_fail;
625 	}
626 
627 	ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
628 	if (ret < 0)
629 		goto out_unregister;
630 
631 	port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
632 	cycle_netdev(port_dev);
633 
634 	return 0;
635 
636 out_unregister:
637 	netdev_rx_handler_unregister(port_dev);
638 out_fail:
639 	return ret;
640 }
641 
642 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
643 {
644 	if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
645 		return -EINVAL;
646 
647 	return do_vrf_add_slave(dev, port_dev);
648 }
649 
650 /* inverse of do_vrf_add_slave */
651 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
652 {
653 	netdev_upper_dev_unlink(port_dev, dev);
654 	port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
655 
656 	netdev_rx_handler_unregister(port_dev);
657 
658 	cycle_netdev(port_dev);
659 
660 	return 0;
661 }
662 
663 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
664 {
665 	return do_vrf_del_slave(dev, port_dev);
666 }
667 
668 static void vrf_dev_uninit(struct net_device *dev)
669 {
670 	struct net_vrf *vrf = netdev_priv(dev);
671 	struct net_device *port_dev;
672 	struct list_head *iter;
673 
674 	vrf_rtable_destroy(vrf);
675 	vrf_rt6_destroy(vrf);
676 
677 	netdev_for_each_lower_dev(dev, port_dev, iter)
678 		vrf_del_slave(dev, port_dev);
679 
680 	free_percpu(dev->dstats);
681 	dev->dstats = NULL;
682 }
683 
684 static int vrf_dev_init(struct net_device *dev)
685 {
686 	struct net_vrf *vrf = netdev_priv(dev);
687 
688 	dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
689 	if (!dev->dstats)
690 		goto out_nomem;
691 
692 	/* create the default dst which points back to us */
693 	vrf->rth = vrf_rtable_create(dev);
694 	if (!vrf->rth)
695 		goto out_stats;
696 
697 	if (vrf_rt6_create(dev) != 0)
698 		goto out_rth;
699 
700 	dev->flags = IFF_MASTER | IFF_NOARP;
701 
702 	return 0;
703 
704 out_rth:
705 	vrf_rtable_destroy(vrf);
706 out_stats:
707 	free_percpu(dev->dstats);
708 	dev->dstats = NULL;
709 out_nomem:
710 	return -ENOMEM;
711 }
712 
713 static const struct net_device_ops vrf_netdev_ops = {
714 	.ndo_init		= vrf_dev_init,
715 	.ndo_uninit		= vrf_dev_uninit,
716 	.ndo_start_xmit		= vrf_xmit,
717 	.ndo_get_stats64	= vrf_get_stats64,
718 	.ndo_add_slave		= vrf_add_slave,
719 	.ndo_del_slave		= vrf_del_slave,
720 };
721 
722 static u32 vrf_fib_table(const struct net_device *dev)
723 {
724 	struct net_vrf *vrf = netdev_priv(dev);
725 
726 	return vrf->tb_id;
727 }
728 
729 static struct rtable *vrf_get_rtable(const struct net_device *dev,
730 				     const struct flowi4 *fl4)
731 {
732 	struct rtable *rth = NULL;
733 
734 	if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
735 		struct net_vrf *vrf = netdev_priv(dev);
736 
737 		rth = vrf->rth;
738 		atomic_inc(&rth->dst.__refcnt);
739 	}
740 
741 	return rth;
742 }
743 
744 /* called under rcu_read_lock */
745 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
746 {
747 	struct fib_result res = { .tclassid = 0 };
748 	struct net *net = dev_net(dev);
749 	u32 orig_tos = fl4->flowi4_tos;
750 	u8 flags = fl4->flowi4_flags;
751 	u8 scope = fl4->flowi4_scope;
752 	u8 tos = RT_FL_TOS(fl4);
753 	int rc;
754 
755 	if (unlikely(!fl4->daddr))
756 		return 0;
757 
758 	fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
759 	fl4->flowi4_iif = LOOPBACK_IFINDEX;
760 	fl4->flowi4_tos = tos & IPTOS_RT_MASK;
761 	fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
762 			     RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
763 
764 	rc = fib_lookup(net, fl4, &res, 0);
765 	if (!rc) {
766 		if (res.type == RTN_LOCAL)
767 			fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
768 		else
769 			fib_select_path(net, &res, fl4, -1);
770 	}
771 
772 	fl4->flowi4_flags = flags;
773 	fl4->flowi4_tos = orig_tos;
774 	fl4->flowi4_scope = scope;
775 
776 	return rc;
777 }
778 
779 #if IS_ENABLED(CONFIG_IPV6)
780 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
781 					 const struct flowi6 *fl6)
782 {
783 	struct rt6_info *rt = NULL;
784 
785 	if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
786 		struct net_vrf *vrf = netdev_priv(dev);
787 
788 		rt = vrf->rt6;
789 		atomic_inc(&rt->dst.__refcnt);
790 	}
791 
792 	return (struct dst_entry *)rt;
793 }
794 #endif
795 
796 static const struct l3mdev_ops vrf_l3mdev_ops = {
797 	.l3mdev_fib_table	= vrf_fib_table,
798 	.l3mdev_get_rtable	= vrf_get_rtable,
799 	.l3mdev_get_saddr	= vrf_get_saddr,
800 #if IS_ENABLED(CONFIG_IPV6)
801 	.l3mdev_get_rt6_dst	= vrf_get_rt6_dst,
802 #endif
803 };
804 
805 static void vrf_get_drvinfo(struct net_device *dev,
806 			    struct ethtool_drvinfo *info)
807 {
808 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
809 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
810 }
811 
812 static const struct ethtool_ops vrf_ethtool_ops = {
813 	.get_drvinfo	= vrf_get_drvinfo,
814 };
815 
816 static void vrf_setup(struct net_device *dev)
817 {
818 	ether_setup(dev);
819 
820 	/* Initialize the device structure. */
821 	dev->netdev_ops = &vrf_netdev_ops;
822 	dev->l3mdev_ops = &vrf_l3mdev_ops;
823 	dev->ethtool_ops = &vrf_ethtool_ops;
824 	dev->destructor = free_netdev;
825 
826 	/* Fill in device structure with ethernet-generic values. */
827 	eth_hw_addr_random(dev);
828 
829 	/* don't acquire vrf device's netif_tx_lock when transmitting */
830 	dev->features |= NETIF_F_LLTX;
831 
832 	/* don't allow vrf devices to change network namespaces. */
833 	dev->features |= NETIF_F_NETNS_LOCAL;
834 }
835 
836 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
837 {
838 	if (tb[IFLA_ADDRESS]) {
839 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
840 			return -EINVAL;
841 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
842 			return -EADDRNOTAVAIL;
843 	}
844 	return 0;
845 }
846 
847 static void vrf_dellink(struct net_device *dev, struct list_head *head)
848 {
849 	unregister_netdevice_queue(dev, head);
850 }
851 
852 static int vrf_newlink(struct net *src_net, struct net_device *dev,
853 		       struct nlattr *tb[], struct nlattr *data[])
854 {
855 	struct net_vrf *vrf = netdev_priv(dev);
856 
857 	if (!data || !data[IFLA_VRF_TABLE])
858 		return -EINVAL;
859 
860 	vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
861 
862 	dev->priv_flags |= IFF_L3MDEV_MASTER;
863 
864 	return register_netdevice(dev);
865 }
866 
867 static size_t vrf_nl_getsize(const struct net_device *dev)
868 {
869 	return nla_total_size(sizeof(u32));  /* IFLA_VRF_TABLE */
870 }
871 
872 static int vrf_fillinfo(struct sk_buff *skb,
873 			const struct net_device *dev)
874 {
875 	struct net_vrf *vrf = netdev_priv(dev);
876 
877 	return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
878 }
879 
880 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
881 	[IFLA_VRF_TABLE] = { .type = NLA_U32 },
882 };
883 
884 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
885 	.kind		= DRV_NAME,
886 	.priv_size	= sizeof(struct net_vrf),
887 
888 	.get_size	= vrf_nl_getsize,
889 	.policy		= vrf_nl_policy,
890 	.validate	= vrf_validate,
891 	.fill_info	= vrf_fillinfo,
892 
893 	.newlink	= vrf_newlink,
894 	.dellink	= vrf_dellink,
895 	.setup		= vrf_setup,
896 	.maxtype	= IFLA_VRF_MAX,
897 };
898 
899 static int vrf_device_event(struct notifier_block *unused,
900 			    unsigned long event, void *ptr)
901 {
902 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
903 
904 	/* only care about unregister events to drop slave references */
905 	if (event == NETDEV_UNREGISTER) {
906 		struct net_device *vrf_dev;
907 
908 		if (!netif_is_l3_slave(dev))
909 			goto out;
910 
911 		vrf_dev = netdev_master_upper_dev_get(dev);
912 		vrf_del_slave(vrf_dev, dev);
913 	}
914 out:
915 	return NOTIFY_DONE;
916 }
917 
918 static struct notifier_block vrf_notifier_block __read_mostly = {
919 	.notifier_call = vrf_device_event,
920 };
921 
922 static int __init vrf_init_module(void)
923 {
924 	int rc;
925 
926 	vrf_dst_ops.kmem_cachep =
927 		kmem_cache_create("vrf_ip_dst_cache",
928 				  sizeof(struct rtable), 0,
929 				  SLAB_HWCACHE_ALIGN,
930 				  NULL);
931 
932 	if (!vrf_dst_ops.kmem_cachep)
933 		return -ENOMEM;
934 
935 	rc = init_dst_ops6_kmem_cachep();
936 	if (rc != 0)
937 		goto error2;
938 
939 	register_netdevice_notifier(&vrf_notifier_block);
940 
941 	rc = rtnl_link_register(&vrf_link_ops);
942 	if (rc < 0)
943 		goto error;
944 
945 	return 0;
946 
947 error:
948 	unregister_netdevice_notifier(&vrf_notifier_block);
949 	free_dst_ops6_kmem_cachep();
950 error2:
951 	kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
952 	return rc;
953 }
954 
955 static void __exit vrf_cleanup_module(void)
956 {
957 	rtnl_link_unregister(&vrf_link_ops);
958 	unregister_netdevice_notifier(&vrf_notifier_block);
959 	kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
960 	free_dst_ops6_kmem_cachep();
961 }
962 
963 module_init(vrf_init_module);
964 module_exit(vrf_cleanup_module);
965 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
966 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
967 MODULE_LICENSE("GPL");
968 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
969 MODULE_VERSION(DRV_VERSION);
970