?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / net / vrf.c
blobb55eeb8f8fa3a56fb945acad1b7eac7b03649191
1 /*
2 * vrf.c: device driver to encapsulate a VRF space
3 *
4 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
5 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
6 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
7 *
8 * Based on dummy, team and ipvlan drivers
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
14 */
15
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/netdevice.h>
19 #include <linux/etherdevice.h>
20 #include <linux/ip.h>
21 #include <linux/init.h>
22 #include <linux/moduleparam.h>
23 #include <linux/netfilter.h>
24 #include <linux/rtnetlink.h>
25 #include <net/rtnetlink.h>
26 #include <linux/u64_stats_sync.h>
27 #include <linux/hashtable.h>
28
29 #include <linux/inetdevice.h>
30 #include <net/arp.h>
31 #include <net/ip.h>
32 #include <net/ip_fib.h>
33 #include <net/ip6_fib.h>
34 #include <net/ip6_route.h>
35 #include <net/route.h>
36 #include <net/addrconf.h>
37 #include <net/l3mdev.h>
38 #include <net/fib_rules.h>
39 #include <net/netns/generic.h>
40
41 #define DRV_NAME "vrf"
42 #define DRV_VERSION "1.0"
43
44 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
45
46 static unsigned int vrf_net_id;
47
48 struct net_vrf {
49 struct rtable __rcu *rth;
50 struct rt6_info __rcu *rt6;
51 #if IS_ENABLED(CONFIG_IPV6)
52 struct fib6_table *fib6_table;
53 #endif
54 u32 tb_id;
55 };
56
57 struct pcpu_dstats {
58 u64 tx_pkts;
59 u64 tx_bytes;
60 u64 tx_drps;
61 u64 rx_pkts;
62 u64 rx_bytes;
63 u64 rx_drps;
64 struct u64_stats_sync syncp;
65 };
66
67 static void vrf_rx_stats(struct net_device *dev, int len)
68 {
69 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
70
71 u64_stats_update_begin(&dstats->syncp);
72 dstats->rx_pkts++;
73 dstats->rx_bytes += len;
74 u64_stats_update_end(&dstats->syncp);
75 }
76
77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
78 {
79 vrf_dev->stats.tx_errors++;
80 kfree_skb(skb);
81 }
82
83 static void vrf_get_stats64(struct net_device *dev,
84 struct rtnl_link_stats64 *stats)
85 {
86 int i;
87
88 for_each_possible_cpu(i) {
89 const struct pcpu_dstats *dstats;
90 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
91 unsigned int start;
92
93 dstats = per_cpu_ptr(dev->dstats, i);
94 do {
95 start = u64_stats_fetch_begin_irq(&dstats->syncp);
96 tbytes = dstats->tx_bytes;
97 tpkts = dstats->tx_pkts;
98 tdrops = dstats->tx_drps;
99 rbytes = dstats->rx_bytes;
100 rpkts = dstats->rx_pkts;
101 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
102 stats->tx_bytes += tbytes;
103 stats->tx_packets += tpkts;
104 stats->tx_dropped += tdrops;
105 stats->rx_bytes += rbytes;
106 stats->rx_packets += rpkts;
107 }
108 }
109
110 /* by default VRF devices do not have a qdisc and are expected
111 * to be created with only a single queue.
112 */
113 static bool qdisc_tx_is_default(const struct net_device *dev)
114 {
115 struct netdev_queue *txq;
116 struct Qdisc *qdisc;
117
118 if (dev->num_tx_queues > 1)
119 return false;
120
121 txq = netdev_get_tx_queue(dev, 0);
122 qdisc = rcu_access_pointer(txq->qdisc);
123
124 return !qdisc->enqueue;
125 }
126
127 /* Local traffic destined to local address. Reinsert the packet to rx
128 * path, similar to loopback handling.
129 */
130 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
131 struct dst_entry *dst)
132 {
133 int len = skb->len;
134
135 skb_orphan(skb);
136
137 skb_dst_set(skb, dst);
138
139 /* set pkt_type to avoid skb hitting packet taps twice -
140 * once on Tx and again in Rx processing
141 */
142 skb->pkt_type = PACKET_LOOPBACK;
143
144 skb->protocol = eth_type_trans(skb, dev);
145
146 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
147 vrf_rx_stats(dev, len);
148 else
149 this_cpu_inc(dev->dstats->rx_drps);
150
151 return NETDEV_TX_OK;
152 }
153
154 #if IS_ENABLED(CONFIG_IPV6)
155 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
156 struct sk_buff *skb)
157 {
158 int err;
159
160 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
161 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
162
163 if (likely(err == 1))
164 err = dst_output(net, sk, skb);
165
166 return err;
167 }
168
169 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
170 struct net_device *dev)
171 {
172 const struct ipv6hdr *iph;
173 struct net *net = dev_net(skb->dev);
174 struct flowi6 fl6;
175 int ret = NET_XMIT_DROP;
176 struct dst_entry *dst;
177 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
178
179 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
180 goto err;
181
182 iph = ipv6_hdr(skb);
183
184 memset(&fl6, 0, sizeof(fl6));
185 /* needed to match OIF rule */
186 fl6.flowi6_oif = dev->ifindex;
187 fl6.flowi6_iif = LOOPBACK_IFINDEX;
188 fl6.daddr = iph->daddr;
189 fl6.saddr = iph->saddr;
190 fl6.flowlabel = ip6_flowinfo(iph);
191 fl6.flowi6_mark = skb->mark;
192 fl6.flowi6_proto = iph->nexthdr;
193 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
194
195 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
196 if (IS_ERR(dst) || dst == dst_null)
197 goto err;
198
199 skb_dst_drop(skb);
200
201 /* if dst.dev is loopback or the VRF device again this is locally
202 * originated traffic destined to a local address. Short circuit
203 * to Rx path
204 */
205 if (dst->dev == dev)
206 return vrf_local_xmit(skb, dev, dst);
207
208 skb_dst_set(skb, dst);
209
210 /* strip the ethernet header added for pass through VRF device */
211 __skb_pull(skb, skb_network_offset(skb));
212
213 ret = vrf_ip6_local_out(net, skb->sk, skb);
214 if (unlikely(net_xmit_eval(ret)))
215 dev->stats.tx_errors++;
216 else
217 ret = NET_XMIT_SUCCESS;
218
219 return ret;
220 err:
221 vrf_tx_error(dev, skb);
222 return NET_XMIT_DROP;
223 }
224 #else
225 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
226 struct net_device *dev)
227 {
228 vrf_tx_error(dev, skb);
229 return NET_XMIT_DROP;
230 }
231 #endif
232
233 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
234 static int vrf_ip_local_out(struct net *net, struct sock *sk,
235 struct sk_buff *skb)
236 {
237 int err;
238
239 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
240 skb, NULL, skb_dst(skb)->dev, dst_output);
241 if (likely(err == 1))
242 err = dst_output(net, sk, skb);
243
244 return err;
245 }
246
247 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
248 struct net_device *vrf_dev)
249 {
250 struct iphdr *ip4h;
251 int ret = NET_XMIT_DROP;
252 struct flowi4 fl4;
253 struct net *net = dev_net(vrf_dev);
254 struct rtable *rt;
255
256 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
257 goto err;
258
259 ip4h = ip_hdr(skb);
260
261 memset(&fl4, 0, sizeof(fl4));
262 /* needed to match OIF rule */
263 fl4.flowi4_oif = vrf_dev->ifindex;
264 fl4.flowi4_iif = LOOPBACK_IFINDEX;
265 fl4.flowi4_tos = RT_TOS(ip4h->tos);
266 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
267 fl4.flowi4_proto = ip4h->protocol;
268 fl4.daddr = ip4h->daddr;
269 fl4.saddr = ip4h->saddr;
270
271 rt = ip_route_output_flow(net, &fl4, NULL);
272 if (IS_ERR(rt))
273 goto err;
274
275 skb_dst_drop(skb);
276
277 /* if dst.dev is loopback or the VRF device again this is locally
278 * originated traffic destined to a local address. Short circuit
279 * to Rx path
280 */
281 if (rt->dst.dev == vrf_dev)
282 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
283
284 skb_dst_set(skb, &rt->dst);
285
286 /* strip the ethernet header added for pass through VRF device */
287 __skb_pull(skb, skb_network_offset(skb));
288
289 if (!ip4h->saddr) {
290 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
291 RT_SCOPE_LINK);
292 }
293
294 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
295 if (unlikely(net_xmit_eval(ret)))
296 vrf_dev->stats.tx_errors++;
297 else
298 ret = NET_XMIT_SUCCESS;
299
300 out:
301 return ret;
302 err:
303 vrf_tx_error(vrf_dev, skb);
304 goto out;
305 }
306
307 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
308 {
309 switch (skb->protocol) {
310 case htons(ETH_P_IP):
311 return vrf_process_v4_outbound(skb, dev);
312 case htons(ETH_P_IPV6):
313 return vrf_process_v6_outbound(skb, dev);
314 default:
315 vrf_tx_error(dev, skb);
316 return NET_XMIT_DROP;
317 }
318 }
319
320 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
321 {
322 int len = skb->len;
323 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
324
325 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
326 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
327
328 u64_stats_update_begin(&dstats->syncp);
329 dstats->tx_pkts++;
330 dstats->tx_bytes += len;
331 u64_stats_update_end(&dstats->syncp);
332 } else {
333 this_cpu_inc(dev->dstats->tx_drps);
334 }
335
336 return ret;
337 }
338
339 static int vrf_finish_direct(struct net *net, struct sock *sk,
340 struct sk_buff *skb)
341 {
342 struct net_device *vrf_dev = skb->dev;
343
344 if (!list_empty(&vrf_dev->ptype_all) &&
345 likely(skb_headroom(skb) >= ETH_HLEN)) {
346 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
347
348 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
349 eth_zero_addr(eth->h_dest);
350 eth->h_proto = skb->protocol;
351
352 rcu_read_lock_bh();
353 dev_queue_xmit_nit(skb, vrf_dev);
354 rcu_read_unlock_bh();
355
356 skb_pull(skb, ETH_HLEN);
357 }
358
359 return 1;
360 }
361
362 #if IS_ENABLED(CONFIG_IPV6)
363 /* modelled after ip6_finish_output2 */
364 static int vrf_finish_output6(struct net *net, struct sock *sk,
365 struct sk_buff *skb)
366 {
367 struct dst_entry *dst = skb_dst(skb);
368 struct net_device *dev = dst->dev;
369 struct neighbour *neigh;
370 struct in6_addr *nexthop;
371 int ret;
372
373 nf_reset(skb);
374
375 skb->protocol = htons(ETH_P_IPV6);
376 skb->dev = dev;
377
378 rcu_read_lock_bh();
379 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
380 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
381 if (unlikely(!neigh))
382 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
383 if (!IS_ERR(neigh)) {
384 sock_confirm_neigh(skb, neigh);
385 ret = neigh_output(neigh, skb);
386 rcu_read_unlock_bh();
387 return ret;
388 }
389 rcu_read_unlock_bh();
390
391 IP6_INC_STATS(dev_net(dst->dev),
392 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
393 kfree_skb(skb);
394 return -EINVAL;
395 }
396
397 /* modelled after ip6_output */
398 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
399 {
400 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
401 net, sk, skb, NULL, skb_dst(skb)->dev,
402 vrf_finish_output6,
403 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
404 }
405
406 /* set dst on skb to send packet to us via dev_xmit path. Allows
407 * packet to go through device based features such as qdisc, netfilter
408 * hooks and packet sockets with skb->dev set to vrf device.
409 */
410 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
411 struct sk_buff *skb)
412 {
413 struct net_vrf *vrf = netdev_priv(vrf_dev);
414 struct dst_entry *dst = NULL;
415 struct rt6_info *rt6;
416
417 rcu_read_lock();
418
419 rt6 = rcu_dereference(vrf->rt6);
420 if (likely(rt6)) {
421 dst = &rt6->dst;
422 dst_hold(dst);
423 }
424
425 rcu_read_unlock();
426
427 if (unlikely(!dst)) {
428 vrf_tx_error(vrf_dev, skb);
429 return NULL;
430 }
431
432 skb_dst_drop(skb);
433 skb_dst_set(skb, dst);
434
435 return skb;
436 }
437
438 static int vrf_output6_direct(struct net *net, struct sock *sk,
439 struct sk_buff *skb)
440 {
441 skb->protocol = htons(ETH_P_IPV6);
442
443 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
444 net, sk, skb, NULL, skb->dev,
445 vrf_finish_direct,
446 !(IPCB(skb)->flags & IPSKB_REROUTED));
447 }
448
449 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
450 struct sock *sk,
451 struct sk_buff *skb)
452 {
453 struct net *net = dev_net(vrf_dev);
454 int err;
455
456 skb->dev = vrf_dev;
457
458 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
459 skb, NULL, vrf_dev, vrf_output6_direct);
460
461 if (likely(err == 1))
462 err = vrf_output6_direct(net, sk, skb);
463
464 /* reset skb device */
465 if (likely(err == 1))
466 nf_reset(skb);
467 else
468 skb = NULL;
469
470 return skb;
471 }
472
473 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
474 struct sock *sk,
475 struct sk_buff *skb)
476 {
477 /* don't divert link scope packets */
478 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
479 return skb;
480
481 if (qdisc_tx_is_default(vrf_dev) ||
482 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
483 return vrf_ip6_out_direct(vrf_dev, sk, skb);
484
485 return vrf_ip6_out_redirect(vrf_dev, skb);
486 }
487
488 /* holding rtnl */
489 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
490 {
491 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
492 struct net *net = dev_net(dev);
493 struct dst_entry *dst;
494
495 RCU_INIT_POINTER(vrf->rt6, NULL);
496 synchronize_rcu();
497
498 /* move dev in dst's to loopback so this VRF device can be deleted
499 * - based on dst_ifdown
500 */
501 if (rt6) {
502 dst = &rt6->dst;
503 dev_put(dst->dev);
504 dst->dev = net->loopback_dev;
505 dev_hold(dst->dev);
506 dst_release(dst);
507 }
508 }
509
510 static int vrf_rt6_create(struct net_device *dev)
511 {
512 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
513 struct net_vrf *vrf = netdev_priv(dev);
514 struct net *net = dev_net(dev);
515 struct rt6_info *rt6;
516 int rc = -ENOMEM;
517
518 /* IPv6 can be CONFIG enabled and then disabled runtime */
519 if (!ipv6_mod_enabled())
520 return 0;
521
522 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
523 if (!vrf->fib6_table)
524 goto out;
525
526 /* create a dst for routing packets out a VRF device */
527 rt6 = ip6_dst_alloc(net, dev, flags);
528 if (!rt6)
529 goto out;
530
531 rt6->dst.output = vrf_output6;
532
533 rcu_assign_pointer(vrf->rt6, rt6);
534
535 rc = 0;
536 out:
537 return rc;
538 }
539 #else
540 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
541 struct sock *sk,
542 struct sk_buff *skb)
543 {
544 return skb;
545 }
546
547 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
548 {
549 }
550
551 static int vrf_rt6_create(struct net_device *dev)
552 {
553 return 0;
554 }
555 #endif
556
557 /* modelled after ip_finish_output2 */
558 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
559 {
560 struct dst_entry *dst = skb_dst(skb);
561 struct rtable *rt = (struct rtable *)dst;
562 struct net_device *dev = dst->dev;
563 unsigned int hh_len = LL_RESERVED_SPACE(dev);
564 struct neighbour *neigh;
565 u32 nexthop;
566 int ret = -EINVAL;
567
568 nf_reset(skb);
569
570 /* Be paranoid, rather than too clever. */
571 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
572 struct sk_buff *skb2;
573
574 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
575 if (!skb2) {
576 ret = -ENOMEM;
577 goto err;
578 }
579 if (skb->sk)
580 skb_set_owner_w(skb2, skb->sk);
581
582 consume_skb(skb);
583 skb = skb2;
584 }
585
586 rcu_read_lock_bh();
587
588 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
589 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
590 if (unlikely(!neigh))
591 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
592 if (!IS_ERR(neigh)) {
593 sock_confirm_neigh(skb, neigh);
594 ret = neigh_output(neigh, skb);
595 rcu_read_unlock_bh();
596 return ret;
597 }
598
599 rcu_read_unlock_bh();
600 err:
601 vrf_tx_error(skb->dev, skb);
602 return ret;
603 }
604
605 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
606 {
607 struct net_device *dev = skb_dst(skb)->dev;
608
609 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
610
611 skb->dev = dev;
612 skb->protocol = htons(ETH_P_IP);
613
614 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
615 net, sk, skb, NULL, dev,
616 vrf_finish_output,
617 !(IPCB(skb)->flags & IPSKB_REROUTED));
618 }
619
620 /* set dst on skb to send packet to us via dev_xmit path. Allows
621 * packet to go through device based features such as qdisc, netfilter
622 * hooks and packet sockets with skb->dev set to vrf device.
623 */
624 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
625 struct sk_buff *skb)
626 {
627 struct net_vrf *vrf = netdev_priv(vrf_dev);
628 struct dst_entry *dst = NULL;
629 struct rtable *rth;
630
631 rcu_read_lock();
632
633 rth = rcu_dereference(vrf->rth);
634 if (likely(rth)) {
635 dst = &rth->dst;
636 dst_hold(dst);
637 }
638
639 rcu_read_unlock();
640
641 if (unlikely(!dst)) {
642 vrf_tx_error(vrf_dev, skb);
643 return NULL;
644 }
645
646 skb_dst_drop(skb);
647 skb_dst_set(skb, dst);
648
649 return skb;
650 }
651
652 static int vrf_output_direct(struct net *net, struct sock *sk,
653 struct sk_buff *skb)
654 {
655 skb->protocol = htons(ETH_P_IP);
656
657 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
658 net, sk, skb, NULL, skb->dev,
659 vrf_finish_direct,
660 !(IPCB(skb)->flags & IPSKB_REROUTED));
661 }
662
663 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
664 struct sock *sk,
665 struct sk_buff *skb)
666 {
667 struct net *net = dev_net(vrf_dev);
668 int err;
669
670 skb->dev = vrf_dev;
671
672 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
673 skb, NULL, vrf_dev, vrf_output_direct);
674
675 if (likely(err == 1))
676 err = vrf_output_direct(net, sk, skb);
677
678 /* reset skb device */
679 if (likely(err == 1))
680 nf_reset(skb);
681 else
682 skb = NULL;
683
684 return skb;
685 }
686
687 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
688 struct sock *sk,
689 struct sk_buff *skb)
690 {
691 /* don't divert multicast or local broadcast */
692 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
693 ipv4_is_lbcast(ip_hdr(skb)->daddr))
694 return skb;
695
696 if (qdisc_tx_is_default(vrf_dev) ||
697 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
698 return vrf_ip_out_direct(vrf_dev, sk, skb);
699
700 return vrf_ip_out_redirect(vrf_dev, skb);
701 }
702
703 /* called with rcu lock held */
704 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
705 struct sock *sk,
706 struct sk_buff *skb,
707 u16 proto)
708 {
709 switch (proto) {
710 case AF_INET:
711 return vrf_ip_out(vrf_dev, sk, skb);
712 case AF_INET6:
713 return vrf_ip6_out(vrf_dev, sk, skb);
714 }
715
716 return skb;
717 }
718
719 /* holding rtnl */
720 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
721 {
722 struct rtable *rth = rtnl_dereference(vrf->rth);
723 struct net *net = dev_net(dev);
724 struct dst_entry *dst;
725
726 RCU_INIT_POINTER(vrf->rth, NULL);
727 synchronize_rcu();
728
729 /* move dev in dst's to loopback so this VRF device can be deleted
730 * - based on dst_ifdown
731 */
732 if (rth) {
733 dst = &rth->dst;
734 dev_put(dst->dev);
735 dst->dev = net->loopback_dev;
736 dev_hold(dst->dev);
737 dst_release(dst);
738 }
739 }
740
741 static int vrf_rtable_create(struct net_device *dev)
742 {
743 struct net_vrf *vrf = netdev_priv(dev);
744 struct rtable *rth;
745
746 if (!fib_new_table(dev_net(dev), vrf->tb_id))
747 return -ENOMEM;
748
749 /* create a dst for routing packets out through a VRF device */
750 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
751 if (!rth)
752 return -ENOMEM;
753
754 rth->dst.output = vrf_output;
755
756 rcu_assign_pointer(vrf->rth, rth);
757
758 return 0;
759 }
760
761 /**************************** device handling ********************/
762
763 /* cycle interface to flush neighbor cache and move routes across tables */
764 static void cycle_netdev(struct net_device *dev)
765 {
766 unsigned int flags = dev->flags;
767 int ret;
768
769 if (!netif_running(dev))
770 return;
771
772 ret = dev_change_flags(dev, flags & ~IFF_UP);
773 if (ret >= 0)
774 ret = dev_change_flags(dev, flags);
775
776 if (ret < 0) {
777 netdev_err(dev,
778 "Failed to cycle device %s; route tables might be wrong!\n",
779 dev->name);
780 }
781 }
782
783 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
784 struct netlink_ext_ack *extack)
785 {
786 int ret;
787
788 /* do not allow loopback device to be enslaved to a VRF.
789 * The vrf device acts as the loopback for the vrf.
790 */
791 if (port_dev == dev_net(dev)->loopback_dev) {
792 NL_SET_ERR_MSG(extack,
793 "Can not enslave loopback device to a VRF");
794 return -EOPNOTSUPP;
795 }
796
797 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
798 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
799 if (ret < 0)
800 goto err;
801
802 cycle_netdev(port_dev);
803
804 return 0;
805
806 err:
807 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
808 return ret;
809 }
810
811 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
812 struct netlink_ext_ack *extack)
813 {
814 if (netif_is_l3_master(port_dev)) {
815 NL_SET_ERR_MSG(extack,
816 "Can not enslave an L3 master device to a VRF");
817 return -EINVAL;
818 }
819
820 if (netif_is_l3_slave(port_dev))
821 return -EINVAL;
822
823 return do_vrf_add_slave(dev, port_dev, extack);
824 }
825
826 /* inverse of do_vrf_add_slave */
827 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
828 {
829 netdev_upper_dev_unlink(port_dev, dev);
830 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
831
832 cycle_netdev(port_dev);
833
834 return 0;
835 }
836
837 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
838 {
839 return do_vrf_del_slave(dev, port_dev);
840 }
841
842 static void vrf_dev_uninit(struct net_device *dev)
843 {
844 struct net_vrf *vrf = netdev_priv(dev);
845
846 vrf_rtable_release(dev, vrf);
847 vrf_rt6_release(dev, vrf);
848
849 free_percpu(dev->dstats);
850 dev->dstats = NULL;
851 }
852
853 static int vrf_dev_init(struct net_device *dev)
854 {
855 struct net_vrf *vrf = netdev_priv(dev);
856
857 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
858 if (!dev->dstats)
859 goto out_nomem;
860
861 /* create the default dst which points back to us */
862 if (vrf_rtable_create(dev) != 0)
863 goto out_stats;
864
865 if (vrf_rt6_create(dev) != 0)
866 goto out_rth;
867
868 dev->flags = IFF_MASTER | IFF_NOARP;
869
870 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
871 dev->mtu = 64 * 1024;
872
873 /* similarly, oper state is irrelevant; set to up to avoid confusion */
874 dev->operstate = IF_OPER_UP;
875 netdev_lockdep_set_classes(dev);
876 return 0;
877
878 out_rth:
879 vrf_rtable_release(dev, vrf);
880 out_stats:
881 free_percpu(dev->dstats);
882 dev->dstats = NULL;
883 out_nomem:
884 return -ENOMEM;
885 }
886
887 static const struct net_device_ops vrf_netdev_ops = {
888 .ndo_init = vrf_dev_init,
889 .ndo_uninit = vrf_dev_uninit,
890 .ndo_start_xmit = vrf_xmit,
891 .ndo_get_stats64 = vrf_get_stats64,
892 .ndo_add_slave = vrf_add_slave,
893 .ndo_del_slave = vrf_del_slave,
894 };
895
896 static u32 vrf_fib_table(const struct net_device *dev)
897 {
898 struct net_vrf *vrf = netdev_priv(dev);
899
900 return vrf->tb_id;
901 }
902
903 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
904 {
905 kfree_skb(skb);
906 return 0;
907 }
908
909 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
910 struct sk_buff *skb,
911 struct net_device *dev)
912 {
913 struct net *net = dev_net(dev);
914
915 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
916 skb = NULL; /* kfree_skb(skb) handled by nf code */
917
918 return skb;
919 }
920
921 #if IS_ENABLED(CONFIG_IPV6)
922 /* neighbor handling is done with actual device; do not want
923 * to flip skb->dev for those ndisc packets. This really fails
924 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
925 * a start.
926 */
927 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
928 {
929 const struct ipv6hdr *iph = ipv6_hdr(skb);
930 bool rc = false;
931
932 if (iph->nexthdr == NEXTHDR_ICMP) {
933 const struct icmp6hdr *icmph;
934 struct icmp6hdr _icmph;
935
936 icmph = skb_header_pointer(skb, sizeof(*iph),
937 sizeof(_icmph), &_icmph);
938 if (!icmph)
939 goto out;
940
941 switch (icmph->icmp6_type) {
942 case NDISC_ROUTER_SOLICITATION:
943 case NDISC_ROUTER_ADVERTISEMENT:
944 case NDISC_NEIGHBOUR_SOLICITATION:
945 case NDISC_NEIGHBOUR_ADVERTISEMENT:
946 case NDISC_REDIRECT:
947 rc = true;
948 break;
949 }
950 }
951
952 out:
953 return rc;
954 }
955
956 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
957 const struct net_device *dev,
958 struct flowi6 *fl6,
959 int ifindex,
960 const struct sk_buff *skb,
961 int flags)
962 {
963 struct net_vrf *vrf = netdev_priv(dev);
964
965 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
966 }
967
968 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
969 int ifindex)
970 {
971 const struct ipv6hdr *iph = ipv6_hdr(skb);
972 struct flowi6 fl6 = {
973 .flowi6_iif = ifindex,
974 .flowi6_mark = skb->mark,
975 .flowi6_proto = iph->nexthdr,
976 .daddr = iph->daddr,
977 .saddr = iph->saddr,
978 .flowlabel = ip6_flowinfo(iph),
979 };
980 struct net *net = dev_net(vrf_dev);
981 struct rt6_info *rt6;
982
983 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
984 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
985 if (unlikely(!rt6))
986 return;
987
988 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
989 return;
990
991 skb_dst_set(skb, &rt6->dst);
992 }
993
994 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
995 struct sk_buff *skb)
996 {
997 int orig_iif = skb->skb_iif;
998 bool need_strict;
999
1000 /* loopback traffic; do not push through packet taps again.
1001 * Reset pkt_type for upper layers to process skb
1002 */
1003 if (skb->pkt_type == PACKET_LOOPBACK) {
1004 skb->dev = vrf_dev;
1005 skb->skb_iif = vrf_dev->ifindex;
1006 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1007 skb->pkt_type = PACKET_HOST;
1008 goto out;
1009 }
1010
1011 /* if packet is NDISC or addressed to multicast or link-local
1012 * then keep the ingress interface
1013 */
1014 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1015 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1016 vrf_rx_stats(vrf_dev, skb->len);
1017 skb->dev = vrf_dev;
1018 skb->skb_iif = vrf_dev->ifindex;
1019
1020 if (!list_empty(&vrf_dev->ptype_all)) {
1021 skb_push(skb, skb->mac_len);
1022 dev_queue_xmit_nit(skb, vrf_dev);
1023 skb_pull(skb, skb->mac_len);
1024 }
1025
1026 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1027 }
1028
1029 if (need_strict)
1030 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1031
1032 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1033 out:
1034 return skb;
1035 }
1036
1037 #else
1038 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1039 struct sk_buff *skb)
1040 {
1041 return skb;
1042 }
1043 #endif
1044
1045 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1046 struct sk_buff *skb)
1047 {
1048 skb->dev = vrf_dev;
1049 skb->skb_iif = vrf_dev->ifindex;
1050 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1051
1052 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1053 goto out;
1054
1055 /* loopback traffic; do not push through packet taps again.
1056 * Reset pkt_type for upper layers to process skb
1057 */
1058 if (skb->pkt_type == PACKET_LOOPBACK) {
1059 skb->pkt_type = PACKET_HOST;
1060 goto out;
1061 }
1062
1063 vrf_rx_stats(vrf_dev, skb->len);
1064
1065 if (!list_empty(&vrf_dev->ptype_all)) {
1066 skb_push(skb, skb->mac_len);
1067 dev_queue_xmit_nit(skb, vrf_dev);
1068 skb_pull(skb, skb->mac_len);
1069 }
1070
1071 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1072 out:
1073 return skb;
1074 }
1075
1076 /* called with rcu lock held */
1077 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1078 struct sk_buff *skb,
1079 u16 proto)
1080 {
1081 switch (proto) {
1082 case AF_INET:
1083 return vrf_ip_rcv(vrf_dev, skb);
1084 case AF_INET6:
1085 return vrf_ip6_rcv(vrf_dev, skb);
1086 }
1087
1088 return skb;
1089 }
1090
1091 #if IS_ENABLED(CONFIG_IPV6)
1092 /* send to link-local or multicast address via interface enslaved to
1093 * VRF device. Force lookup to VRF table without changing flow struct
1094 */
1095 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1096 struct flowi6 *fl6)
1097 {
1098 struct net *net = dev_net(dev);
1099 int flags = RT6_LOOKUP_F_IFACE;
1100 struct dst_entry *dst = NULL;
1101 struct rt6_info *rt;
1102
1103 /* VRF device does not have a link-local address and
1104 * sending packets to link-local or mcast addresses over
1105 * a VRF device does not make sense
1106 */
1107 if (fl6->flowi6_oif == dev->ifindex) {
1108 dst = &net->ipv6.ip6_null_entry->dst;
1109 dst_hold(dst);
1110 return dst;
1111 }
1112
1113 if (!ipv6_addr_any(&fl6->saddr))
1114 flags |= RT6_LOOKUP_F_HAS_SADDR;
1115
1116 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1117 if (rt)
1118 dst = &rt->dst;
1119
1120 return dst;
1121 }
1122 #endif
1123
1124 static const struct l3mdev_ops vrf_l3mdev_ops = {
1125 .l3mdev_fib_table = vrf_fib_table,
1126 .l3mdev_l3_rcv = vrf_l3_rcv,
1127 .l3mdev_l3_out = vrf_l3_out,
1128 #if IS_ENABLED(CONFIG_IPV6)
1129 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1130 #endif
1131 };
1132
1133 static void vrf_get_drvinfo(struct net_device *dev,
1134 struct ethtool_drvinfo *info)
1135 {
1136 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1137 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1138 }
1139
1140 static const struct ethtool_ops vrf_ethtool_ops = {
1141 .get_drvinfo = vrf_get_drvinfo,
1142 };
1143
1144 static inline size_t vrf_fib_rule_nl_size(void)
1145 {
1146 size_t sz;
1147
1148 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1149 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1150 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1151 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1152
1153 return sz;
1154 }
1155
1156 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1157 {
1158 struct fib_rule_hdr *frh;
1159 struct nlmsghdr *nlh;
1160 struct sk_buff *skb;
1161 int err;
1162
1163 if (family == AF_INET6 && !ipv6_mod_enabled())
1164 return 0;
1165
1166 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1167 if (!skb)
1168 return -ENOMEM;
1169
1170 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1171 if (!nlh)
1172 goto nla_put_failure;
1173
1174 /* rule only needs to appear once */
1175 nlh->nlmsg_flags |= NLM_F_EXCL;
1176
1177 frh = nlmsg_data(nlh);
1178 memset(frh, 0, sizeof(*frh));
1179 frh->family = family;
1180 frh->action = FR_ACT_TO_TBL;
1181
1182 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1183 goto nla_put_failure;
1184
1185 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1186 goto nla_put_failure;
1187
1188 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1189 goto nla_put_failure;
1190
1191 nlmsg_end(skb, nlh);
1192
1193 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1194 skb->sk = dev_net(dev)->rtnl;
1195 if (add_it) {
1196 err = fib_nl_newrule(skb, nlh, NULL);
1197 if (err == -EEXIST)
1198 err = 0;
1199 } else {
1200 err = fib_nl_delrule(skb, nlh, NULL);
1201 if (err == -ENOENT)
1202 err = 0;
1203 }
1204 nlmsg_free(skb);
1205
1206 return err;
1207
1208 nla_put_failure:
1209 nlmsg_free(skb);
1210
1211 return -EMSGSIZE;
1212 }
1213
1214 static int vrf_add_fib_rules(const struct net_device *dev)
1215 {
1216 int err;
1217
1218 err = vrf_fib_rule(dev, AF_INET, true);
1219 if (err < 0)
1220 goto out_err;
1221
1222 err = vrf_fib_rule(dev, AF_INET6, true);
1223 if (err < 0)
1224 goto ipv6_err;
1225
1226 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1227 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1228 if (err < 0)
1229 goto ipmr_err;
1230 #endif
1231
1232 return 0;
1233
1234 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1235 ipmr_err:
1236 vrf_fib_rule(dev, AF_INET6, false);
1237 #endif
1238
1239 ipv6_err:
1240 vrf_fib_rule(dev, AF_INET, false);
1241
1242 out_err:
1243 netdev_err(dev, "Failed to add FIB rules.\n");
1244 return err;
1245 }
1246
1247 static void vrf_setup(struct net_device *dev)
1248 {
1249 ether_setup(dev);
1250
1251 /* Initialize the device structure. */
1252 dev->netdev_ops = &vrf_netdev_ops;
1253 dev->l3mdev_ops = &vrf_l3mdev_ops;
1254 dev->ethtool_ops = &vrf_ethtool_ops;
1255 dev->needs_free_netdev = true;
1256
1257 /* Fill in device structure with ethernet-generic values. */
1258 eth_hw_addr_random(dev);
1259
1260 /* don't acquire vrf device's netif_tx_lock when transmitting */
1261 dev->features |= NETIF_F_LLTX;
1262
1263 /* don't allow vrf devices to change network namespaces. */
1264 dev->features |= NETIF_F_NETNS_LOCAL;
1265
1266 /* does not make sense for a VLAN to be added to a vrf device */
1267 dev->features |= NETIF_F_VLAN_CHALLENGED;
1268
1269 /* enable offload features */
1270 dev->features |= NETIF_F_GSO_SOFTWARE;
1271 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1272 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1273
1274 dev->hw_features = dev->features;
1275 dev->hw_enc_features = dev->features;
1276
1277 /* default to no qdisc; user can add if desired */
1278 dev->priv_flags |= IFF_NO_QUEUE;
1279 dev->priv_flags |= IFF_NO_RX_HANDLER;
1280 }
1281
1282 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1283 struct netlink_ext_ack *extack)
1284 {
1285 if (tb[IFLA_ADDRESS]) {
1286 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1287 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1288 return -EINVAL;
1289 }
1290 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1291 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1292 return -EADDRNOTAVAIL;
1293 }
1294 }
1295 return 0;
1296 }
1297
1298 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1299 {
1300 struct net_device *port_dev;
1301 struct list_head *iter;
1302
1303 netdev_for_each_lower_dev(dev, port_dev, iter)
1304 vrf_del_slave(dev, port_dev);
1305
1306 unregister_netdevice_queue(dev, head);
1307 }
1308
1309 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1310 struct nlattr *tb[], struct nlattr *data[],
1311 struct netlink_ext_ack *extack)
1312 {
1313 struct net_vrf *vrf = netdev_priv(dev);
1314 bool *add_fib_rules;
1315 struct net *net;
1316 int err;
1317
1318 if (!data || !data[IFLA_VRF_TABLE]) {
1319 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1320 return -EINVAL;
1321 }
1322
1323 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1324 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1325 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1326 "Invalid VRF table id");
1327 return -EINVAL;
1328 }
1329
1330 dev->priv_flags |= IFF_L3MDEV_MASTER;
1331
1332 err = register_netdevice(dev);
1333 if (err)
1334 goto out;
1335
1336 net = dev_net(dev);
1337 add_fib_rules = net_generic(net, vrf_net_id);
1338 if (*add_fib_rules) {
1339 err = vrf_add_fib_rules(dev);
1340 if (err) {
1341 unregister_netdevice(dev);
1342 goto out;
1343 }
1344 *add_fib_rules = false;
1345 }
1346
1347 out:
1348 return err;
1349 }
1350
1351 static size_t vrf_nl_getsize(const struct net_device *dev)
1352 {
1353 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1354 }
1355
1356 static int vrf_fillinfo(struct sk_buff *skb,
1357 const struct net_device *dev)
1358 {
1359 struct net_vrf *vrf = netdev_priv(dev);
1360
1361 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1362 }
1363
1364 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1365 const struct net_device *slave_dev)
1366 {
1367 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1368 }
1369
1370 static int vrf_fill_slave_info(struct sk_buff *skb,
1371 const struct net_device *vrf_dev,
1372 const struct net_device *slave_dev)
1373 {
1374 struct net_vrf *vrf = netdev_priv(vrf_dev);
1375
1376 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1377 return -EMSGSIZE;
1378
1379 return 0;
1380 }
1381
1382 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1383 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1384 };
1385
1386 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1387 .kind = DRV_NAME,
1388 .priv_size = sizeof(struct net_vrf),
1389
1390 .get_size = vrf_nl_getsize,
1391 .policy = vrf_nl_policy,
1392 .validate = vrf_validate,
1393 .fill_info = vrf_fillinfo,
1394
1395 .get_slave_size = vrf_get_slave_size,
1396 .fill_slave_info = vrf_fill_slave_info,
1397
1398 .newlink = vrf_newlink,
1399 .dellink = vrf_dellink,
1400 .setup = vrf_setup,
1401 .maxtype = IFLA_VRF_MAX,
1402 };
1403
1404 static int vrf_device_event(struct notifier_block *unused,
1405 unsigned long event, void *ptr)
1406 {
1407 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1408
1409 /* only care about unregister events to drop slave references */
1410 if (event == NETDEV_UNREGISTER) {
1411 struct net_device *vrf_dev;
1412
1413 if (!netif_is_l3_slave(dev))
1414 goto out;
1415
1416 vrf_dev = netdev_master_upper_dev_get(dev);
1417 vrf_del_slave(vrf_dev, dev);
1418 }
1419 out:
1420 return NOTIFY_DONE;
1421 }
1422
1423 static struct notifier_block vrf_notifier_block __read_mostly = {
1424 .notifier_call = vrf_device_event,
1425 };
1426
1427 /* Initialize per network namespace state */
1428 static int __net_init vrf_netns_init(struct net *net)
1429 {
1430 bool *add_fib_rules = net_generic(net, vrf_net_id);
1431
1432 *add_fib_rules = true;
1433
1434 return 0;
1435 }
1436
1437 static struct pernet_operations vrf_net_ops __net_initdata = {
1438 .init = vrf_netns_init,
1439 .id = &vrf_net_id,
1440 .size = sizeof(bool),
1441 };
1442
1443 static int __init vrf_init_module(void)
1444 {
1445 int rc;
1446
1447 register_netdevice_notifier(&vrf_notifier_block);
1448
1449 rc = register_pernet_subsys(&vrf_net_ops);
1450 if (rc < 0)
1451 goto error;
1452
1453 rc = rtnl_link_register(&vrf_link_ops);
1454 if (rc < 0) {
1455 unregister_pernet_subsys(&vrf_net_ops);
1456 goto error;
1457 }
1458
1459 return 0;
1460
1461 error:
1462 unregister_netdevice_notifier(&vrf_notifier_block);
1463 return rc;
1464 }
1465
1466 module_init(vrf_init_module);
1467 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1468 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1469 MODULE_LICENSE("GPL");
1470 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1471 MODULE_VERSION(DRV_VERSION);
Linux with added FPC-III support