]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/vrf.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
rtnetlink: enable IFLA_IF_NETNSID for RTM_DELLINK
[mirror_ubuntu-bionic-kernel.git] / drivers / net / vrf.c
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 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
136 /* set pkt_type to avoid skb hitting packet taps twice -
137 * once on Tx and again in Rx processing
138 */
139 skb->pkt_type = PACKET_LOOPBACK;
140
141 skb->protocol = eth_type_trans(skb, dev);
142
143 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
144 vrf_rx_stats(dev, len);
145 else
146 this_cpu_inc(dev->dstats->rx_drps);
147
148 return NETDEV_TX_OK;
149 }
150
151 #if IS_ENABLED(CONFIG_IPV6)
152 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
153 struct sk_buff *skb)
154 {
155 int err;
156
157 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
158 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
159
160 if (likely(err == 1))
161 err = dst_output(net, sk, skb);
162
163 return err;
164 }
165
166 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
167 struct net_device *dev)
168 {
169 const struct ipv6hdr *iph = ipv6_hdr(skb);
170 struct net *net = dev_net(skb->dev);
171 struct flowi6 fl6 = {
172 /* needed to match OIF rule */
173 .flowi6_oif = dev->ifindex,
174 .flowi6_iif = LOOPBACK_IFINDEX,
175 .daddr = iph->daddr,
176 .saddr = iph->saddr,
177 .flowlabel = ip6_flowinfo(iph),
178 .flowi6_mark = skb->mark,
179 .flowi6_proto = iph->nexthdr,
180 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
181 };
182 int ret = NET_XMIT_DROP;
183 struct dst_entry *dst;
184 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
185
186 dst = ip6_route_output(net, NULL, &fl6);
187 if (dst == dst_null)
188 goto err;
189
190 skb_dst_drop(skb);
191
192 /* if dst.dev is loopback or the VRF device again this is locally
193 * originated traffic destined to a local address. Short circuit
194 * to Rx path
195 */
196 if (dst->dev == dev)
197 return vrf_local_xmit(skb, dev, dst);
198
199 skb_dst_set(skb, dst);
200
201 /* strip the ethernet header added for pass through VRF device */
202 __skb_pull(skb, skb_network_offset(skb));
203
204 ret = vrf_ip6_local_out(net, skb->sk, skb);
205 if (unlikely(net_xmit_eval(ret)))
206 dev->stats.tx_errors++;
207 else
208 ret = NET_XMIT_SUCCESS;
209
210 return ret;
211 err:
212 vrf_tx_error(dev, skb);
213 return NET_XMIT_DROP;
214 }
215 #else
216 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
217 struct net_device *dev)
218 {
219 vrf_tx_error(dev, skb);
220 return NET_XMIT_DROP;
221 }
222 #endif
223
224 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
225 static int vrf_ip_local_out(struct net *net, struct sock *sk,
226 struct sk_buff *skb)
227 {
228 int err;
229
230 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
231 skb, NULL, skb_dst(skb)->dev, dst_output);
232 if (likely(err == 1))
233 err = dst_output(net, sk, skb);
234
235 return err;
236 }
237
238 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
239 struct net_device *vrf_dev)
240 {
241 struct iphdr *ip4h = ip_hdr(skb);
242 int ret = NET_XMIT_DROP;
243 struct flowi4 fl4 = {
244 /* needed to match OIF rule */
245 .flowi4_oif = vrf_dev->ifindex,
246 .flowi4_iif = LOOPBACK_IFINDEX,
247 .flowi4_tos = RT_TOS(ip4h->tos),
248 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
249 .flowi4_proto = ip4h->protocol,
250 .daddr = ip4h->daddr,
251 .saddr = ip4h->saddr,
252 };
253 struct net *net = dev_net(vrf_dev);
254 struct rtable *rt;
255
256 rt = ip_route_output_flow(net, &fl4, NULL);
257 if (IS_ERR(rt))
258 goto err;
259
260 skb_dst_drop(skb);
261
262 /* if dst.dev is loopback or the VRF device again this is locally
263 * originated traffic destined to a local address. Short circuit
264 * to Rx path
265 */
266 if (rt->dst.dev == vrf_dev)
267 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
268
269 skb_dst_set(skb, &rt->dst);
270
271 /* strip the ethernet header added for pass through VRF device */
272 __skb_pull(skb, skb_network_offset(skb));
273
274 if (!ip4h->saddr) {
275 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
276 RT_SCOPE_LINK);
277 }
278
279 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
280 if (unlikely(net_xmit_eval(ret)))
281 vrf_dev->stats.tx_errors++;
282 else
283 ret = NET_XMIT_SUCCESS;
284
285 out:
286 return ret;
287 err:
288 vrf_tx_error(vrf_dev, skb);
289 goto out;
290 }
291
292 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
293 {
294 switch (skb->protocol) {
295 case htons(ETH_P_IP):
296 return vrf_process_v4_outbound(skb, dev);
297 case htons(ETH_P_IPV6):
298 return vrf_process_v6_outbound(skb, dev);
299 default:
300 vrf_tx_error(dev, skb);
301 return NET_XMIT_DROP;
302 }
303 }
304
305 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
306 {
307 int len = skb->len;
308 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
309
310 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
311 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
312
313 u64_stats_update_begin(&dstats->syncp);
314 dstats->tx_pkts++;
315 dstats->tx_bytes += len;
316 u64_stats_update_end(&dstats->syncp);
317 } else {
318 this_cpu_inc(dev->dstats->tx_drps);
319 }
320
321 return ret;
322 }
323
324 static int vrf_finish_direct(struct net *net, struct sock *sk,
325 struct sk_buff *skb)
326 {
327 struct net_device *vrf_dev = skb->dev;
328
329 if (!list_empty(&vrf_dev->ptype_all) &&
330 likely(skb_headroom(skb) >= ETH_HLEN)) {
331 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
332
333 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
334 eth_zero_addr(eth->h_dest);
335 eth->h_proto = skb->protocol;
336
337 rcu_read_lock_bh();
338 dev_queue_xmit_nit(skb, vrf_dev);
339 rcu_read_unlock_bh();
340
341 skb_pull(skb, ETH_HLEN);
342 }
343
344 return 1;
345 }
346
347 #if IS_ENABLED(CONFIG_IPV6)
348 /* modelled after ip6_finish_output2 */
349 static int vrf_finish_output6(struct net *net, struct sock *sk,
350 struct sk_buff *skb)
351 {
352 struct dst_entry *dst = skb_dst(skb);
353 struct net_device *dev = dst->dev;
354 struct neighbour *neigh;
355 struct in6_addr *nexthop;
356 int ret;
357
358 nf_reset(skb);
359
360 skb->protocol = htons(ETH_P_IPV6);
361 skb->dev = dev;
362
363 rcu_read_lock_bh();
364 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
365 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
366 if (unlikely(!neigh))
367 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
368 if (!IS_ERR(neigh)) {
369 sock_confirm_neigh(skb, neigh);
370 ret = neigh_output(neigh, skb);
371 rcu_read_unlock_bh();
372 return ret;
373 }
374 rcu_read_unlock_bh();
375
376 IP6_INC_STATS(dev_net(dst->dev),
377 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
378 kfree_skb(skb);
379 return -EINVAL;
380 }
381
382 /* modelled after ip6_output */
383 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
384 {
385 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
386 net, sk, skb, NULL, skb_dst(skb)->dev,
387 vrf_finish_output6,
388 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
389 }
390
391 /* set dst on skb to send packet to us via dev_xmit path. Allows
392 * packet to go through device based features such as qdisc, netfilter
393 * hooks and packet sockets with skb->dev set to vrf device.
394 */
395 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
396 struct sk_buff *skb)
397 {
398 struct net_vrf *vrf = netdev_priv(vrf_dev);
399 struct dst_entry *dst = NULL;
400 struct rt6_info *rt6;
401
402 rcu_read_lock();
403
404 rt6 = rcu_dereference(vrf->rt6);
405 if (likely(rt6)) {
406 dst = &rt6->dst;
407 dst_hold(dst);
408 }
409
410 rcu_read_unlock();
411
412 if (unlikely(!dst)) {
413 vrf_tx_error(vrf_dev, skb);
414 return NULL;
415 }
416
417 skb_dst_drop(skb);
418 skb_dst_set(skb, dst);
419
420 return skb;
421 }
422
423 static int vrf_output6_direct(struct net *net, struct sock *sk,
424 struct sk_buff *skb)
425 {
426 skb->protocol = htons(ETH_P_IPV6);
427
428 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
429 net, sk, skb, NULL, skb->dev,
430 vrf_finish_direct,
431 !(IPCB(skb)->flags & IPSKB_REROUTED));
432 }
433
434 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
435 struct sock *sk,
436 struct sk_buff *skb)
437 {
438 struct net *net = dev_net(vrf_dev);
439 int err;
440
441 skb->dev = vrf_dev;
442
443 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
444 skb, NULL, vrf_dev, vrf_output6_direct);
445
446 if (likely(err == 1))
447 err = vrf_output6_direct(net, sk, skb);
448
449 /* reset skb device */
450 if (likely(err == 1))
451 nf_reset(skb);
452 else
453 skb = NULL;
454
455 return skb;
456 }
457
458 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
459 struct sock *sk,
460 struct sk_buff *skb)
461 {
462 /* don't divert link scope packets */
463 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
464 return skb;
465
466 if (qdisc_tx_is_default(vrf_dev))
467 return vrf_ip6_out_direct(vrf_dev, sk, skb);
468
469 return vrf_ip6_out_redirect(vrf_dev, skb);
470 }
471
472 /* holding rtnl */
473 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
474 {
475 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
476 struct net *net = dev_net(dev);
477 struct dst_entry *dst;
478
479 RCU_INIT_POINTER(vrf->rt6, NULL);
480 synchronize_rcu();
481
482 /* move dev in dst's to loopback so this VRF device can be deleted
483 * - based on dst_ifdown
484 */
485 if (rt6) {
486 dst = &rt6->dst;
487 dev_put(dst->dev);
488 dst->dev = net->loopback_dev;
489 dev_hold(dst->dev);
490 dst_release(dst);
491 }
492 }
493
494 static int vrf_rt6_create(struct net_device *dev)
495 {
496 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
497 struct net_vrf *vrf = netdev_priv(dev);
498 struct net *net = dev_net(dev);
499 struct fib6_table *rt6i_table;
500 struct rt6_info *rt6;
501 int rc = -ENOMEM;
502
503 /* IPv6 can be CONFIG enabled and then disabled runtime */
504 if (!ipv6_mod_enabled())
505 return 0;
506
507 rt6i_table = fib6_new_table(net, vrf->tb_id);
508 if (!rt6i_table)
509 goto out;
510
511 /* create a dst for routing packets out a VRF device */
512 rt6 = ip6_dst_alloc(net, dev, flags);
513 if (!rt6)
514 goto out;
515
516 rt6->rt6i_table = rt6i_table;
517 rt6->dst.output = vrf_output6;
518
519 rcu_assign_pointer(vrf->rt6, rt6);
520
521 rc = 0;
522 out:
523 return rc;
524 }
525 #else
526 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
527 struct sock *sk,
528 struct sk_buff *skb)
529 {
530 return skb;
531 }
532
533 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
534 {
535 }
536
537 static int vrf_rt6_create(struct net_device *dev)
538 {
539 return 0;
540 }
541 #endif
542
543 /* modelled after ip_finish_output2 */
544 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
545 {
546 struct dst_entry *dst = skb_dst(skb);
547 struct rtable *rt = (struct rtable *)dst;
548 struct net_device *dev = dst->dev;
549 unsigned int hh_len = LL_RESERVED_SPACE(dev);
550 struct neighbour *neigh;
551 u32 nexthop;
552 int ret = -EINVAL;
553
554 nf_reset(skb);
555
556 /* Be paranoid, rather than too clever. */
557 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
558 struct sk_buff *skb2;
559
560 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
561 if (!skb2) {
562 ret = -ENOMEM;
563 goto err;
564 }
565 if (skb->sk)
566 skb_set_owner_w(skb2, skb->sk);
567
568 consume_skb(skb);
569 skb = skb2;
570 }
571
572 rcu_read_lock_bh();
573
574 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
575 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
576 if (unlikely(!neigh))
577 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
578 if (!IS_ERR(neigh)) {
579 sock_confirm_neigh(skb, neigh);
580 ret = neigh_output(neigh, skb);
581 }
582
583 rcu_read_unlock_bh();
584 err:
585 if (unlikely(ret < 0))
586 vrf_tx_error(skb->dev, skb);
587 return ret;
588 }
589
590 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
591 {
592 struct net_device *dev = skb_dst(skb)->dev;
593
594 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
595
596 skb->dev = dev;
597 skb->protocol = htons(ETH_P_IP);
598
599 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
600 net, sk, skb, NULL, dev,
601 vrf_finish_output,
602 !(IPCB(skb)->flags & IPSKB_REROUTED));
603 }
604
605 /* set dst on skb to send packet to us via dev_xmit path. Allows
606 * packet to go through device based features such as qdisc, netfilter
607 * hooks and packet sockets with skb->dev set to vrf device.
608 */
609 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
610 struct sk_buff *skb)
611 {
612 struct net_vrf *vrf = netdev_priv(vrf_dev);
613 struct dst_entry *dst = NULL;
614 struct rtable *rth;
615
616 rcu_read_lock();
617
618 rth = rcu_dereference(vrf->rth);
619 if (likely(rth)) {
620 dst = &rth->dst;
621 dst_hold(dst);
622 }
623
624 rcu_read_unlock();
625
626 if (unlikely(!dst)) {
627 vrf_tx_error(vrf_dev, skb);
628 return NULL;
629 }
630
631 skb_dst_drop(skb);
632 skb_dst_set(skb, dst);
633
634 return skb;
635 }
636
637 static int vrf_output_direct(struct net *net, struct sock *sk,
638 struct sk_buff *skb)
639 {
640 skb->protocol = htons(ETH_P_IP);
641
642 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
643 net, sk, skb, NULL, skb->dev,
644 vrf_finish_direct,
645 !(IPCB(skb)->flags & IPSKB_REROUTED));
646 }
647
648 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
649 struct sock *sk,
650 struct sk_buff *skb)
651 {
652 struct net *net = dev_net(vrf_dev);
653 int err;
654
655 skb->dev = vrf_dev;
656
657 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
658 skb, NULL, vrf_dev, vrf_output_direct);
659
660 if (likely(err == 1))
661 err = vrf_output_direct(net, sk, skb);
662
663 /* reset skb device */
664 if (likely(err == 1))
665 nf_reset(skb);
666 else
667 skb = NULL;
668
669 return skb;
670 }
671
672 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
673 struct sock *sk,
674 struct sk_buff *skb)
675 {
676 /* don't divert multicast or local broadcast */
677 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
678 ipv4_is_lbcast(ip_hdr(skb)->daddr))
679 return skb;
680
681 if (qdisc_tx_is_default(vrf_dev))
682 return vrf_ip_out_direct(vrf_dev, sk, skb);
683
684 return vrf_ip_out_redirect(vrf_dev, skb);
685 }
686
687 /* called with rcu lock held */
688 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
689 struct sock *sk,
690 struct sk_buff *skb,
691 u16 proto)
692 {
693 switch (proto) {
694 case AF_INET:
695 return vrf_ip_out(vrf_dev, sk, skb);
696 case AF_INET6:
697 return vrf_ip6_out(vrf_dev, sk, skb);
698 }
699
700 return skb;
701 }
702
703 /* holding rtnl */
704 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
705 {
706 struct rtable *rth = rtnl_dereference(vrf->rth);
707 struct net *net = dev_net(dev);
708 struct dst_entry *dst;
709
710 RCU_INIT_POINTER(vrf->rth, NULL);
711 synchronize_rcu();
712
713 /* move dev in dst's to loopback so this VRF device can be deleted
714 * - based on dst_ifdown
715 */
716 if (rth) {
717 dst = &rth->dst;
718 dev_put(dst->dev);
719 dst->dev = net->loopback_dev;
720 dev_hold(dst->dev);
721 dst_release(dst);
722 }
723 }
724
725 static int vrf_rtable_create(struct net_device *dev)
726 {
727 struct net_vrf *vrf = netdev_priv(dev);
728 struct rtable *rth;
729
730 if (!fib_new_table(dev_net(dev), vrf->tb_id))
731 return -ENOMEM;
732
733 /* create a dst for routing packets out through a VRF device */
734 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
735 if (!rth)
736 return -ENOMEM;
737
738 rth->dst.output = vrf_output;
739 rth->rt_table_id = vrf->tb_id;
740
741 rcu_assign_pointer(vrf->rth, rth);
742
743 return 0;
744 }
745
746 /**************************** device handling ********************/
747
748 /* cycle interface to flush neighbor cache and move routes across tables */
749 static void cycle_netdev(struct net_device *dev)
750 {
751 unsigned int flags = dev->flags;
752 int ret;
753
754 if (!netif_running(dev))
755 return;
756
757 ret = dev_change_flags(dev, flags & ~IFF_UP);
758 if (ret >= 0)
759 ret = dev_change_flags(dev, flags);
760
761 if (ret < 0) {
762 netdev_err(dev,
763 "Failed to cycle device %s; route tables might be wrong!\n",
764 dev->name);
765 }
766 }
767
768 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
769 struct netlink_ext_ack *extack)
770 {
771 int ret;
772
773 /* do not allow loopback device to be enslaved to a VRF.
774 * The vrf device acts as the loopback for the vrf.
775 */
776 if (port_dev == dev_net(dev)->loopback_dev) {
777 NL_SET_ERR_MSG(extack,
778 "Can not enslave loopback device to a VRF");
779 return -EOPNOTSUPP;
780 }
781
782 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
783 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
784 if (ret < 0)
785 goto err;
786
787 cycle_netdev(port_dev);
788
789 return 0;
790
791 err:
792 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
793 return ret;
794 }
795
796 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
797 struct netlink_ext_ack *extack)
798 {
799 if (netif_is_l3_master(port_dev)) {
800 NL_SET_ERR_MSG(extack,
801 "Can not enslave an L3 master device to a VRF");
802 return -EINVAL;
803 }
804
805 if (netif_is_l3_slave(port_dev))
806 return -EINVAL;
807
808 return do_vrf_add_slave(dev, port_dev, extack);
809 }
810
811 /* inverse of do_vrf_add_slave */
812 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
813 {
814 netdev_upper_dev_unlink(port_dev, dev);
815 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
816
817 cycle_netdev(port_dev);
818
819 return 0;
820 }
821
822 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
823 {
824 return do_vrf_del_slave(dev, port_dev);
825 }
826
827 static void vrf_dev_uninit(struct net_device *dev)
828 {
829 struct net_vrf *vrf = netdev_priv(dev);
830
831 vrf_rtable_release(dev, vrf);
832 vrf_rt6_release(dev, vrf);
833
834 free_percpu(dev->dstats);
835 dev->dstats = NULL;
836 }
837
838 static int vrf_dev_init(struct net_device *dev)
839 {
840 struct net_vrf *vrf = netdev_priv(dev);
841
842 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
843 if (!dev->dstats)
844 goto out_nomem;
845
846 /* create the default dst which points back to us */
847 if (vrf_rtable_create(dev) != 0)
848 goto out_stats;
849
850 if (vrf_rt6_create(dev) != 0)
851 goto out_rth;
852
853 dev->flags = IFF_MASTER | IFF_NOARP;
854
855 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
856 dev->mtu = 64 * 1024;
857
858 /* similarly, oper state is irrelevant; set to up to avoid confusion */
859 dev->operstate = IF_OPER_UP;
860 netdev_lockdep_set_classes(dev);
861 return 0;
862
863 out_rth:
864 vrf_rtable_release(dev, vrf);
865 out_stats:
866 free_percpu(dev->dstats);
867 dev->dstats = NULL;
868 out_nomem:
869 return -ENOMEM;
870 }
871
872 static const struct net_device_ops vrf_netdev_ops = {
873 .ndo_init = vrf_dev_init,
874 .ndo_uninit = vrf_dev_uninit,
875 .ndo_start_xmit = vrf_xmit,
876 .ndo_get_stats64 = vrf_get_stats64,
877 .ndo_add_slave = vrf_add_slave,
878 .ndo_del_slave = vrf_del_slave,
879 };
880
881 static u32 vrf_fib_table(const struct net_device *dev)
882 {
883 struct net_vrf *vrf = netdev_priv(dev);
884
885 return vrf->tb_id;
886 }
887
888 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
889 {
890 kfree_skb(skb);
891 return 0;
892 }
893
894 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
895 struct sk_buff *skb,
896 struct net_device *dev)
897 {
898 struct net *net = dev_net(dev);
899
900 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
901 skb = NULL; /* kfree_skb(skb) handled by nf code */
902
903 return skb;
904 }
905
906 #if IS_ENABLED(CONFIG_IPV6)
907 /* neighbor handling is done with actual device; do not want
908 * to flip skb->dev for those ndisc packets. This really fails
909 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
910 * a start.
911 */
912 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
913 {
914 const struct ipv6hdr *iph = ipv6_hdr(skb);
915 bool rc = false;
916
917 if (iph->nexthdr == NEXTHDR_ICMP) {
918 const struct icmp6hdr *icmph;
919 struct icmp6hdr _icmph;
920
921 icmph = skb_header_pointer(skb, sizeof(*iph),
922 sizeof(_icmph), &_icmph);
923 if (!icmph)
924 goto out;
925
926 switch (icmph->icmp6_type) {
927 case NDISC_ROUTER_SOLICITATION:
928 case NDISC_ROUTER_ADVERTISEMENT:
929 case NDISC_NEIGHBOUR_SOLICITATION:
930 case NDISC_NEIGHBOUR_ADVERTISEMENT:
931 case NDISC_REDIRECT:
932 rc = true;
933 break;
934 }
935 }
936
937 out:
938 return rc;
939 }
940
941 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
942 const struct net_device *dev,
943 struct flowi6 *fl6,
944 int ifindex,
945 int flags)
946 {
947 struct net_vrf *vrf = netdev_priv(dev);
948 struct fib6_table *table = NULL;
949 struct rt6_info *rt6;
950
951 rcu_read_lock();
952
953 /* fib6_table does not have a refcnt and can not be freed */
954 rt6 = rcu_dereference(vrf->rt6);
955 if (likely(rt6))
956 table = rt6->rt6i_table;
957
958 rcu_read_unlock();
959
960 if (!table)
961 return NULL;
962
963 return ip6_pol_route(net, table, ifindex, fl6, flags);
964 }
965
966 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
967 int ifindex)
968 {
969 const struct ipv6hdr *iph = ipv6_hdr(skb);
970 struct flowi6 fl6 = {
971 .flowi6_iif = ifindex,
972 .flowi6_mark = skb->mark,
973 .flowi6_proto = iph->nexthdr,
974 .daddr = iph->daddr,
975 .saddr = iph->saddr,
976 .flowlabel = ip6_flowinfo(iph),
977 };
978 struct net *net = dev_net(vrf_dev);
979 struct rt6_info *rt6;
980
981 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
982 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
983 if (unlikely(!rt6))
984 return;
985
986 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
987 return;
988
989 skb_dst_set(skb, &rt6->dst);
990 }
991
992 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
993 struct sk_buff *skb)
994 {
995 int orig_iif = skb->skb_iif;
996 bool need_strict;
997
998 /* loopback traffic; do not push through packet taps again.
999 * Reset pkt_type for upper layers to process skb
1000 */
1001 if (skb->pkt_type == PACKET_LOOPBACK) {
1002 skb->dev = vrf_dev;
1003 skb->skb_iif = vrf_dev->ifindex;
1004 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1005 skb->pkt_type = PACKET_HOST;
1006 goto out;
1007 }
1008
1009 /* if packet is NDISC or addressed to multicast or link-local
1010 * then keep the ingress interface
1011 */
1012 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1013 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1014 vrf_rx_stats(vrf_dev, skb->len);
1015 skb->dev = vrf_dev;
1016 skb->skb_iif = vrf_dev->ifindex;
1017
1018 if (!list_empty(&vrf_dev->ptype_all)) {
1019 skb_push(skb, skb->mac_len);
1020 dev_queue_xmit_nit(skb, vrf_dev);
1021 skb_pull(skb, skb->mac_len);
1022 }
1023
1024 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1025 }
1026
1027 if (need_strict)
1028 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1029
1030 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1031 out:
1032 return skb;
1033 }
1034
1035 #else
1036 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1037 struct sk_buff *skb)
1038 {
1039 return skb;
1040 }
1041 #endif
1042
1043 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1044 struct sk_buff *skb)
1045 {
1046 skb->dev = vrf_dev;
1047 skb->skb_iif = vrf_dev->ifindex;
1048 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1049
1050 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1051 goto out;
1052
1053 /* loopback traffic; do not push through packet taps again.
1054 * Reset pkt_type for upper layers to process skb
1055 */
1056 if (skb->pkt_type == PACKET_LOOPBACK) {
1057 skb->pkt_type = PACKET_HOST;
1058 goto out;
1059 }
1060
1061 vrf_rx_stats(vrf_dev, skb->len);
1062
1063 if (!list_empty(&vrf_dev->ptype_all)) {
1064 skb_push(skb, skb->mac_len);
1065 dev_queue_xmit_nit(skb, vrf_dev);
1066 skb_pull(skb, skb->mac_len);
1067 }
1068
1069 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1070 out:
1071 return skb;
1072 }
1073
1074 /* called with rcu lock held */
1075 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1076 struct sk_buff *skb,
1077 u16 proto)
1078 {
1079 switch (proto) {
1080 case AF_INET:
1081 return vrf_ip_rcv(vrf_dev, skb);
1082 case AF_INET6:
1083 return vrf_ip6_rcv(vrf_dev, skb);
1084 }
1085
1086 return skb;
1087 }
1088
1089 #if IS_ENABLED(CONFIG_IPV6)
1090 /* send to link-local or multicast address via interface enslaved to
1091 * VRF device. Force lookup to VRF table without changing flow struct
1092 */
1093 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1094 struct flowi6 *fl6)
1095 {
1096 struct net *net = dev_net(dev);
1097 int flags = RT6_LOOKUP_F_IFACE;
1098 struct dst_entry *dst = NULL;
1099 struct rt6_info *rt;
1100
1101 /* VRF device does not have a link-local address and
1102 * sending packets to link-local or mcast addresses over
1103 * a VRF device does not make sense
1104 */
1105 if (fl6->flowi6_oif == dev->ifindex) {
1106 dst = &net->ipv6.ip6_null_entry->dst;
1107 dst_hold(dst);
1108 return dst;
1109 }
1110
1111 if (!ipv6_addr_any(&fl6->saddr))
1112 flags |= RT6_LOOKUP_F_HAS_SADDR;
1113
1114 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1115 if (rt)
1116 dst = &rt->dst;
1117
1118 return dst;
1119 }
1120 #endif
1121
1122 static const struct l3mdev_ops vrf_l3mdev_ops = {
1123 .l3mdev_fib_table = vrf_fib_table,
1124 .l3mdev_l3_rcv = vrf_l3_rcv,
1125 .l3mdev_l3_out = vrf_l3_out,
1126 #if IS_ENABLED(CONFIG_IPV6)
1127 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1128 #endif
1129 };
1130
1131 static void vrf_get_drvinfo(struct net_device *dev,
1132 struct ethtool_drvinfo *info)
1133 {
1134 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1135 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1136 }
1137
1138 static const struct ethtool_ops vrf_ethtool_ops = {
1139 .get_drvinfo = vrf_get_drvinfo,
1140 };
1141
1142 static inline size_t vrf_fib_rule_nl_size(void)
1143 {
1144 size_t sz;
1145
1146 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1147 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1148 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1149
1150 return sz;
1151 }
1152
1153 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1154 {
1155 struct fib_rule_hdr *frh;
1156 struct nlmsghdr *nlh;
1157 struct sk_buff *skb;
1158 int err;
1159
1160 if (family == AF_INET6 && !ipv6_mod_enabled())
1161 return 0;
1162
1163 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1164 if (!skb)
1165 return -ENOMEM;
1166
1167 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1168 if (!nlh)
1169 goto nla_put_failure;
1170
1171 /* rule only needs to appear once */
1172 nlh->nlmsg_flags |= NLM_F_EXCL;
1173
1174 frh = nlmsg_data(nlh);
1175 memset(frh, 0, sizeof(*frh));
1176 frh->family = family;
1177 frh->action = FR_ACT_TO_TBL;
1178
1179 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1180 goto nla_put_failure;
1181
1182 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1183 goto nla_put_failure;
1184
1185 nlmsg_end(skb, nlh);
1186
1187 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1188 skb->sk = dev_net(dev)->rtnl;
1189 if (add_it) {
1190 err = fib_nl_newrule(skb, nlh, NULL);
1191 if (err == -EEXIST)
1192 err = 0;
1193 } else {
1194 err = fib_nl_delrule(skb, nlh, NULL);
1195 if (err == -ENOENT)
1196 err = 0;
1197 }
1198 nlmsg_free(skb);
1199
1200 return err;
1201
1202 nla_put_failure:
1203 nlmsg_free(skb);
1204
1205 return -EMSGSIZE;
1206 }
1207
1208 static int vrf_add_fib_rules(const struct net_device *dev)
1209 {
1210 int err;
1211
1212 err = vrf_fib_rule(dev, AF_INET, true);
1213 if (err < 0)
1214 goto out_err;
1215
1216 err = vrf_fib_rule(dev, AF_INET6, true);
1217 if (err < 0)
1218 goto ipv6_err;
1219
1220 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1221 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1222 if (err < 0)
1223 goto ipmr_err;
1224 #endif
1225
1226 return 0;
1227
1228 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1229 ipmr_err:
1230 vrf_fib_rule(dev, AF_INET6, false);
1231 #endif
1232
1233 ipv6_err:
1234 vrf_fib_rule(dev, AF_INET, false);
1235
1236 out_err:
1237 netdev_err(dev, "Failed to add FIB rules.\n");
1238 return err;
1239 }
1240
1241 static void vrf_setup(struct net_device *dev)
1242 {
1243 ether_setup(dev);
1244
1245 /* Initialize the device structure. */
1246 dev->netdev_ops = &vrf_netdev_ops;
1247 dev->l3mdev_ops = &vrf_l3mdev_ops;
1248 dev->ethtool_ops = &vrf_ethtool_ops;
1249 dev->needs_free_netdev = true;
1250
1251 /* Fill in device structure with ethernet-generic values. */
1252 eth_hw_addr_random(dev);
1253
1254 /* don't acquire vrf device's netif_tx_lock when transmitting */
1255 dev->features |= NETIF_F_LLTX;
1256
1257 /* don't allow vrf devices to change network namespaces. */
1258 dev->features |= NETIF_F_NETNS_LOCAL;
1259
1260 /* does not make sense for a VLAN to be added to a vrf device */
1261 dev->features |= NETIF_F_VLAN_CHALLENGED;
1262
1263 /* enable offload features */
1264 dev->features |= NETIF_F_GSO_SOFTWARE;
1265 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1266 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1267
1268 dev->hw_features = dev->features;
1269 dev->hw_enc_features = dev->features;
1270
1271 /* default to no qdisc; user can add if desired */
1272 dev->priv_flags |= IFF_NO_QUEUE;
1273 }
1274
1275 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1276 struct netlink_ext_ack *extack)
1277 {
1278 if (tb[IFLA_ADDRESS]) {
1279 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1280 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1281 return -EINVAL;
1282 }
1283 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1284 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1285 return -EADDRNOTAVAIL;
1286 }
1287 }
1288 return 0;
1289 }
1290
1291 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1292 {
1293 struct net_device *port_dev;
1294 struct list_head *iter;
1295
1296 netdev_for_each_lower_dev(dev, port_dev, iter)
1297 vrf_del_slave(dev, port_dev);
1298
1299 unregister_netdevice_queue(dev, head);
1300 }
1301
1302 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1303 struct nlattr *tb[], struct nlattr *data[],
1304 struct netlink_ext_ack *extack)
1305 {
1306 struct net_vrf *vrf = netdev_priv(dev);
1307 bool *add_fib_rules;
1308 struct net *net;
1309 int err;
1310
1311 if (!data || !data[IFLA_VRF_TABLE]) {
1312 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1313 return -EINVAL;
1314 }
1315
1316 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1317 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1318 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1319 "Invalid VRF table id");
1320 return -EINVAL;
1321 }
1322
1323 dev->priv_flags |= IFF_L3MDEV_MASTER;
1324
1325 err = register_netdevice(dev);
1326 if (err)
1327 goto out;
1328
1329 net = dev_net(dev);
1330 add_fib_rules = net_generic(net, vrf_net_id);
1331 if (*add_fib_rules) {
1332 err = vrf_add_fib_rules(dev);
1333 if (err) {
1334 unregister_netdevice(dev);
1335 goto out;
1336 }
1337 *add_fib_rules = false;
1338 }
1339
1340 out:
1341 return err;
1342 }
1343
1344 static size_t vrf_nl_getsize(const struct net_device *dev)
1345 {
1346 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1347 }
1348
1349 static int vrf_fillinfo(struct sk_buff *skb,
1350 const struct net_device *dev)
1351 {
1352 struct net_vrf *vrf = netdev_priv(dev);
1353
1354 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1355 }
1356
1357 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1358 const struct net_device *slave_dev)
1359 {
1360 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1361 }
1362
1363 static int vrf_fill_slave_info(struct sk_buff *skb,
1364 const struct net_device *vrf_dev,
1365 const struct net_device *slave_dev)
1366 {
1367 struct net_vrf *vrf = netdev_priv(vrf_dev);
1368
1369 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1370 return -EMSGSIZE;
1371
1372 return 0;
1373 }
1374
1375 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1376 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1377 };
1378
1379 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1380 .kind = DRV_NAME,
1381 .priv_size = sizeof(struct net_vrf),
1382
1383 .get_size = vrf_nl_getsize,
1384 .policy = vrf_nl_policy,
1385 .validate = vrf_validate,
1386 .fill_info = vrf_fillinfo,
1387
1388 .get_slave_size = vrf_get_slave_size,
1389 .fill_slave_info = vrf_fill_slave_info,
1390
1391 .newlink = vrf_newlink,
1392 .dellink = vrf_dellink,
1393 .setup = vrf_setup,
1394 .maxtype = IFLA_VRF_MAX,
1395 };
1396
1397 static int vrf_device_event(struct notifier_block *unused,
1398 unsigned long event, void *ptr)
1399 {
1400 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1401
1402 /* only care about unregister events to drop slave references */
1403 if (event == NETDEV_UNREGISTER) {
1404 struct net_device *vrf_dev;
1405
1406 if (!netif_is_l3_slave(dev))
1407 goto out;
1408
1409 vrf_dev = netdev_master_upper_dev_get(dev);
1410 vrf_del_slave(vrf_dev, dev);
1411 }
1412 out:
1413 return NOTIFY_DONE;
1414 }
1415
1416 static struct notifier_block vrf_notifier_block __read_mostly = {
1417 .notifier_call = vrf_device_event,
1418 };
1419
1420 /* Initialize per network namespace state */
1421 static int __net_init vrf_netns_init(struct net *net)
1422 {
1423 bool *add_fib_rules = net_generic(net, vrf_net_id);
1424
1425 *add_fib_rules = true;
1426
1427 return 0;
1428 }
1429
1430 static struct pernet_operations vrf_net_ops __net_initdata = {
1431 .init = vrf_netns_init,
1432 .id = &vrf_net_id,
1433 .size = sizeof(bool),
1434 };
1435
1436 static int __init vrf_init_module(void)
1437 {
1438 int rc;
1439
1440 register_netdevice_notifier(&vrf_notifier_block);
1441
1442 rc = register_pernet_subsys(&vrf_net_ops);
1443 if (rc < 0)
1444 goto error;
1445
1446 rc = rtnl_link_register(&vrf_link_ops);
1447 if (rc < 0) {
1448 unregister_pernet_subsys(&vrf_net_ops);
1449 goto error;
1450 }
1451
1452 return 0;
1453
1454 error:
1455 unregister_netdevice_notifier(&vrf_notifier_block);
1456 return rc;
1457 }
1458
1459 module_init(vrf_init_module);
1460 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1461 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1462 MODULE_LICENSE("GPL");
1463 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1464 MODULE_VERSION(DRV_VERSION);
ubuntu-bionic-kernel mirror