Loading...
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
46static unsigned int vrf_net_id;
47
48struct net_vrf {
49 struct rtable __rcu *rth;
50 struct rt6_info __rcu *rt6;
51 u32 tb_id;
52};
53
54struct 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
64static 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
74static 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
80static 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 */
110static 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 */
127static 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)
152static 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
166static 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;
211err:
212 vrf_tx_error(dev, skb);
213 return NET_XMIT_DROP;
214}
215#else
216static 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 */
225static 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
238static 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
285out:
286 return ret;
287err:
288 vrf_tx_error(vrf_dev, skb);
289 goto out;
290}
291
292static 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
305static 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
324static 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 */
349static 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 */
383static 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 */
395static 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
423static 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
434static 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
458static 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 */
473static 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
494static 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;
522out:
523 return rc;
524}
525#else
526static 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
533static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
534{
535}
536
537static int vrf_rt6_create(struct net_device *dev)
538{
539 return 0;
540}
541#endif
542
543/* modelled after ip_finish_output2 */
544static 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 rcu_read_unlock_bh();
582 return ret;
583 }
584
585 rcu_read_unlock_bh();
586err:
587 vrf_tx_error(skb->dev, skb);
588 return ret;
589}
590
591static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
592{
593 struct net_device *dev = skb_dst(skb)->dev;
594
595 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
596
597 skb->dev = dev;
598 skb->protocol = htons(ETH_P_IP);
599
600 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
601 net, sk, skb, NULL, dev,
602 vrf_finish_output,
603 !(IPCB(skb)->flags & IPSKB_REROUTED));
604}
605
606/* set dst on skb to send packet to us via dev_xmit path. Allows
607 * packet to go through device based features such as qdisc, netfilter
608 * hooks and packet sockets with skb->dev set to vrf device.
609 */
610static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
611 struct sk_buff *skb)
612{
613 struct net_vrf *vrf = netdev_priv(vrf_dev);
614 struct dst_entry *dst = NULL;
615 struct rtable *rth;
616
617 rcu_read_lock();
618
619 rth = rcu_dereference(vrf->rth);
620 if (likely(rth)) {
621 dst = &rth->dst;
622 dst_hold(dst);
623 }
624
625 rcu_read_unlock();
626
627 if (unlikely(!dst)) {
628 vrf_tx_error(vrf_dev, skb);
629 return NULL;
630 }
631
632 skb_dst_drop(skb);
633 skb_dst_set(skb, dst);
634
635 return skb;
636}
637
638static int vrf_output_direct(struct net *net, struct sock *sk,
639 struct sk_buff *skb)
640{
641 skb->protocol = htons(ETH_P_IP);
642
643 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
644 net, sk, skb, NULL, skb->dev,
645 vrf_finish_direct,
646 !(IPCB(skb)->flags & IPSKB_REROUTED));
647}
648
649static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
650 struct sock *sk,
651 struct sk_buff *skb)
652{
653 struct net *net = dev_net(vrf_dev);
654 int err;
655
656 skb->dev = vrf_dev;
657
658 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
659 skb, NULL, vrf_dev, vrf_output_direct);
660
661 if (likely(err == 1))
662 err = vrf_output_direct(net, sk, skb);
663
664 /* reset skb device */
665 if (likely(err == 1))
666 nf_reset(skb);
667 else
668 skb = NULL;
669
670 return skb;
671}
672
673static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
674 struct sock *sk,
675 struct sk_buff *skb)
676{
677 /* don't divert multicast or local broadcast */
678 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
679 ipv4_is_lbcast(ip_hdr(skb)->daddr))
680 return skb;
681
682 if (qdisc_tx_is_default(vrf_dev))
683 return vrf_ip_out_direct(vrf_dev, sk, skb);
684
685 return vrf_ip_out_redirect(vrf_dev, skb);
686}
687
688/* called with rcu lock held */
689static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
690 struct sock *sk,
691 struct sk_buff *skb,
692 u16 proto)
693{
694 switch (proto) {
695 case AF_INET:
696 return vrf_ip_out(vrf_dev, sk, skb);
697 case AF_INET6:
698 return vrf_ip6_out(vrf_dev, sk, skb);
699 }
700
701 return skb;
702}
703
704/* holding rtnl */
705static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
706{
707 struct rtable *rth = rtnl_dereference(vrf->rth);
708 struct net *net = dev_net(dev);
709 struct dst_entry *dst;
710
711 RCU_INIT_POINTER(vrf->rth, NULL);
712 synchronize_rcu();
713
714 /* move dev in dst's to loopback so this VRF device can be deleted
715 * - based on dst_ifdown
716 */
717 if (rth) {
718 dst = &rth->dst;
719 dev_put(dst->dev);
720 dst->dev = net->loopback_dev;
721 dev_hold(dst->dev);
722 dst_release(dst);
723 }
724}
725
726static int vrf_rtable_create(struct net_device *dev)
727{
728 struct net_vrf *vrf = netdev_priv(dev);
729 struct rtable *rth;
730
731 if (!fib_new_table(dev_net(dev), vrf->tb_id))
732 return -ENOMEM;
733
734 /* create a dst for routing packets out through a VRF device */
735 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
736 if (!rth)
737 return -ENOMEM;
738
739 rth->dst.output = vrf_output;
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 */
749static 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
768static 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
791err:
792 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
793 return ret;
794}
795
796static 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 */
812static 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
822static 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
827static 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
838static 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
863out_rth:
864 vrf_rtable_release(dev, vrf);
865out_stats:
866 free_percpu(dev->dstats);
867 dev->dstats = NULL;
868out_nomem:
869 return -ENOMEM;
870}
871
872static 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
881static 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
888static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
889{
890 kfree_skb(skb);
891 return 0;
892}
893
894static 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 */
912static 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
937out:
938 return rc;
939}
940
941static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
942 const struct net_device *dev,
943 struct flowi6 *fl6,
944 int ifindex,
945 const struct sk_buff *skb,
946 int flags)
947{
948 struct net_vrf *vrf = netdev_priv(dev);
949 struct fib6_table *table = NULL;
950 struct rt6_info *rt6;
951
952 rcu_read_lock();
953
954 /* fib6_table does not have a refcnt and can not be freed */
955 rt6 = rcu_dereference(vrf->rt6);
956 if (likely(rt6))
957 table = rt6->rt6i_table;
958
959 rcu_read_unlock();
960
961 if (!table)
962 return NULL;
963
964 return ip6_pol_route(net, table, ifindex, fl6, skb, flags);
965}
966
967static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
968 int ifindex)
969{
970 const struct ipv6hdr *iph = ipv6_hdr(skb);
971 struct flowi6 fl6 = {
972 .flowi6_iif = ifindex,
973 .flowi6_mark = skb->mark,
974 .flowi6_proto = iph->nexthdr,
975 .daddr = iph->daddr,
976 .saddr = iph->saddr,
977 .flowlabel = ip6_flowinfo(iph),
978 };
979 struct net *net = dev_net(vrf_dev);
980 struct rt6_info *rt6;
981
982 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
983 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
984 if (unlikely(!rt6))
985 return;
986
987 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
988 return;
989
990 skb_dst_set(skb, &rt6->dst);
991}
992
993static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
994 struct sk_buff *skb)
995{
996 int orig_iif = skb->skb_iif;
997 bool need_strict;
998
999 /* loopback traffic; do not push through packet taps again.
1000 * Reset pkt_type for upper layers to process skb
1001 */
1002 if (skb->pkt_type == PACKET_LOOPBACK) {
1003 skb->dev = vrf_dev;
1004 skb->skb_iif = vrf_dev->ifindex;
1005 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1006 skb->pkt_type = PACKET_HOST;
1007 goto out;
1008 }
1009
1010 /* if packet is NDISC or addressed to multicast or link-local
1011 * then keep the ingress interface
1012 */
1013 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1014 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1015 vrf_rx_stats(vrf_dev, skb->len);
1016 skb->dev = vrf_dev;
1017 skb->skb_iif = vrf_dev->ifindex;
1018
1019 if (!list_empty(&vrf_dev->ptype_all)) {
1020 skb_push(skb, skb->mac_len);
1021 dev_queue_xmit_nit(skb, vrf_dev);
1022 skb_pull(skb, skb->mac_len);
1023 }
1024
1025 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1026 }
1027
1028 if (need_strict)
1029 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1030
1031 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1032out:
1033 return skb;
1034}
1035
1036#else
1037static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1038 struct sk_buff *skb)
1039{
1040 return skb;
1041}
1042#endif
1043
1044static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1045 struct sk_buff *skb)
1046{
1047 skb->dev = vrf_dev;
1048 skb->skb_iif = vrf_dev->ifindex;
1049 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1050
1051 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1052 goto out;
1053
1054 /* loopback traffic; do not push through packet taps again.
1055 * Reset pkt_type for upper layers to process skb
1056 */
1057 if (skb->pkt_type == PACKET_LOOPBACK) {
1058 skb->pkt_type = PACKET_HOST;
1059 goto out;
1060 }
1061
1062 vrf_rx_stats(vrf_dev, skb->len);
1063
1064 if (!list_empty(&vrf_dev->ptype_all)) {
1065 skb_push(skb, skb->mac_len);
1066 dev_queue_xmit_nit(skb, vrf_dev);
1067 skb_pull(skb, skb->mac_len);
1068 }
1069
1070 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1071out:
1072 return skb;
1073}
1074
1075/* called with rcu lock held */
1076static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1077 struct sk_buff *skb,
1078 u16 proto)
1079{
1080 switch (proto) {
1081 case AF_INET:
1082 return vrf_ip_rcv(vrf_dev, skb);
1083 case AF_INET6:
1084 return vrf_ip6_rcv(vrf_dev, skb);
1085 }
1086
1087 return skb;
1088}
1089
1090#if IS_ENABLED(CONFIG_IPV6)
1091/* send to link-local or multicast address via interface enslaved to
1092 * VRF device. Force lookup to VRF table without changing flow struct
1093 */
1094static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1095 struct flowi6 *fl6)
1096{
1097 struct net *net = dev_net(dev);
1098 int flags = RT6_LOOKUP_F_IFACE;
1099 struct dst_entry *dst = NULL;
1100 struct rt6_info *rt;
1101
1102 /* VRF device does not have a link-local address and
1103 * sending packets to link-local or mcast addresses over
1104 * a VRF device does not make sense
1105 */
1106 if (fl6->flowi6_oif == dev->ifindex) {
1107 dst = &net->ipv6.ip6_null_entry->dst;
1108 dst_hold(dst);
1109 return dst;
1110 }
1111
1112 if (!ipv6_addr_any(&fl6->saddr))
1113 flags |= RT6_LOOKUP_F_HAS_SADDR;
1114
1115 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1116 if (rt)
1117 dst = &rt->dst;
1118
1119 return dst;
1120}
1121#endif
1122
1123static const struct l3mdev_ops vrf_l3mdev_ops = {
1124 .l3mdev_fib_table = vrf_fib_table,
1125 .l3mdev_l3_rcv = vrf_l3_rcv,
1126 .l3mdev_l3_out = vrf_l3_out,
1127#if IS_ENABLED(CONFIG_IPV6)
1128 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1129#endif
1130};
1131
1132static void vrf_get_drvinfo(struct net_device *dev,
1133 struct ethtool_drvinfo *info)
1134{
1135 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1136 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1137}
1138
1139static const struct ethtool_ops vrf_ethtool_ops = {
1140 .get_drvinfo = vrf_get_drvinfo,
1141};
1142
1143static inline size_t vrf_fib_rule_nl_size(void)
1144{
1145 size_t sz;
1146
1147 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1148 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1149 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1150 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1151
1152 return sz;
1153}
1154
1155static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1156{
1157 struct fib_rule_hdr *frh;
1158 struct nlmsghdr *nlh;
1159 struct sk_buff *skb;
1160 int err;
1161
1162 if (family == AF_INET6 && !ipv6_mod_enabled())
1163 return 0;
1164
1165 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1166 if (!skb)
1167 return -ENOMEM;
1168
1169 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1170 if (!nlh)
1171 goto nla_put_failure;
1172
1173 /* rule only needs to appear once */
1174 nlh->nlmsg_flags |= NLM_F_EXCL;
1175
1176 frh = nlmsg_data(nlh);
1177 memset(frh, 0, sizeof(*frh));
1178 frh->family = family;
1179 frh->action = FR_ACT_TO_TBL;
1180
1181 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1182 goto nla_put_failure;
1183
1184 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1185 goto nla_put_failure;
1186
1187 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1188 goto nla_put_failure;
1189
1190 nlmsg_end(skb, nlh);
1191
1192 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1193 skb->sk = dev_net(dev)->rtnl;
1194 if (add_it) {
1195 err = fib_nl_newrule(skb, nlh, NULL);
1196 if (err == -EEXIST)
1197 err = 0;
1198 } else {
1199 err = fib_nl_delrule(skb, nlh, NULL);
1200 if (err == -ENOENT)
1201 err = 0;
1202 }
1203 nlmsg_free(skb);
1204
1205 return err;
1206
1207nla_put_failure:
1208 nlmsg_free(skb);
1209
1210 return -EMSGSIZE;
1211}
1212
1213static int vrf_add_fib_rules(const struct net_device *dev)
1214{
1215 int err;
1216
1217 err = vrf_fib_rule(dev, AF_INET, true);
1218 if (err < 0)
1219 goto out_err;
1220
1221 err = vrf_fib_rule(dev, AF_INET6, true);
1222 if (err < 0)
1223 goto ipv6_err;
1224
1225#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1226 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1227 if (err < 0)
1228 goto ipmr_err;
1229#endif
1230
1231 return 0;
1232
1233#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1234ipmr_err:
1235 vrf_fib_rule(dev, AF_INET6, false);
1236#endif
1237
1238ipv6_err:
1239 vrf_fib_rule(dev, AF_INET, false);
1240
1241out_err:
1242 netdev_err(dev, "Failed to add FIB rules.\n");
1243 return err;
1244}
1245
1246static void vrf_setup(struct net_device *dev)
1247{
1248 ether_setup(dev);
1249
1250 /* Initialize the device structure. */
1251 dev->netdev_ops = &vrf_netdev_ops;
1252 dev->l3mdev_ops = &vrf_l3mdev_ops;
1253 dev->ethtool_ops = &vrf_ethtool_ops;
1254 dev->needs_free_netdev = true;
1255
1256 /* Fill in device structure with ethernet-generic values. */
1257 eth_hw_addr_random(dev);
1258
1259 /* don't acquire vrf device's netif_tx_lock when transmitting */
1260 dev->features |= NETIF_F_LLTX;
1261
1262 /* don't allow vrf devices to change network namespaces. */
1263 dev->features |= NETIF_F_NETNS_LOCAL;
1264
1265 /* does not make sense for a VLAN to be added to a vrf device */
1266 dev->features |= NETIF_F_VLAN_CHALLENGED;
1267
1268 /* enable offload features */
1269 dev->features |= NETIF_F_GSO_SOFTWARE;
1270 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1271 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1272
1273 dev->hw_features = dev->features;
1274 dev->hw_enc_features = dev->features;
1275
1276 /* default to no qdisc; user can add if desired */
1277 dev->priv_flags |= IFF_NO_QUEUE;
1278}
1279
1280static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1281 struct netlink_ext_ack *extack)
1282{
1283 if (tb[IFLA_ADDRESS]) {
1284 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1285 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1286 return -EINVAL;
1287 }
1288 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1289 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1290 return -EADDRNOTAVAIL;
1291 }
1292 }
1293 return 0;
1294}
1295
1296static void vrf_dellink(struct net_device *dev, struct list_head *head)
1297{
1298 struct net_device *port_dev;
1299 struct list_head *iter;
1300
1301 netdev_for_each_lower_dev(dev, port_dev, iter)
1302 vrf_del_slave(dev, port_dev);
1303
1304 unregister_netdevice_queue(dev, head);
1305}
1306
1307static int vrf_newlink(struct net *src_net, struct net_device *dev,
1308 struct nlattr *tb[], struct nlattr *data[],
1309 struct netlink_ext_ack *extack)
1310{
1311 struct net_vrf *vrf = netdev_priv(dev);
1312 bool *add_fib_rules;
1313 struct net *net;
1314 int err;
1315
1316 if (!data || !data[IFLA_VRF_TABLE]) {
1317 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1318 return -EINVAL;
1319 }
1320
1321 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1322 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1323 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1324 "Invalid VRF table id");
1325 return -EINVAL;
1326 }
1327
1328 dev->priv_flags |= IFF_L3MDEV_MASTER;
1329
1330 err = register_netdevice(dev);
1331 if (err)
1332 goto out;
1333
1334 net = dev_net(dev);
1335 add_fib_rules = net_generic(net, vrf_net_id);
1336 if (*add_fib_rules) {
1337 err = vrf_add_fib_rules(dev);
1338 if (err) {
1339 unregister_netdevice(dev);
1340 goto out;
1341 }
1342 *add_fib_rules = false;
1343 }
1344
1345out:
1346 return err;
1347}
1348
1349static size_t vrf_nl_getsize(const struct net_device *dev)
1350{
1351 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1352}
1353
1354static int vrf_fillinfo(struct sk_buff *skb,
1355 const struct net_device *dev)
1356{
1357 struct net_vrf *vrf = netdev_priv(dev);
1358
1359 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1360}
1361
1362static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1363 const struct net_device *slave_dev)
1364{
1365 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1366}
1367
1368static int vrf_fill_slave_info(struct sk_buff *skb,
1369 const struct net_device *vrf_dev,
1370 const struct net_device *slave_dev)
1371{
1372 struct net_vrf *vrf = netdev_priv(vrf_dev);
1373
1374 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1375 return -EMSGSIZE;
1376
1377 return 0;
1378}
1379
1380static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1381 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1382};
1383
1384static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1385 .kind = DRV_NAME,
1386 .priv_size = sizeof(struct net_vrf),
1387
1388 .get_size = vrf_nl_getsize,
1389 .policy = vrf_nl_policy,
1390 .validate = vrf_validate,
1391 .fill_info = vrf_fillinfo,
1392
1393 .get_slave_size = vrf_get_slave_size,
1394 .fill_slave_info = vrf_fill_slave_info,
1395
1396 .newlink = vrf_newlink,
1397 .dellink = vrf_dellink,
1398 .setup = vrf_setup,
1399 .maxtype = IFLA_VRF_MAX,
1400};
1401
1402static int vrf_device_event(struct notifier_block *unused,
1403 unsigned long event, void *ptr)
1404{
1405 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1406
1407 /* only care about unregister events to drop slave references */
1408 if (event == NETDEV_UNREGISTER) {
1409 struct net_device *vrf_dev;
1410
1411 if (!netif_is_l3_slave(dev))
1412 goto out;
1413
1414 vrf_dev = netdev_master_upper_dev_get(dev);
1415 vrf_del_slave(vrf_dev, dev);
1416 }
1417out:
1418 return NOTIFY_DONE;
1419}
1420
1421static struct notifier_block vrf_notifier_block __read_mostly = {
1422 .notifier_call = vrf_device_event,
1423};
1424
1425/* Initialize per network namespace state */
1426static int __net_init vrf_netns_init(struct net *net)
1427{
1428 bool *add_fib_rules = net_generic(net, vrf_net_id);
1429
1430 *add_fib_rules = true;
1431
1432 return 0;
1433}
1434
1435static struct pernet_operations vrf_net_ops __net_initdata = {
1436 .init = vrf_netns_init,
1437 .id = &vrf_net_id,
1438 .size = sizeof(bool),
1439};
1440
1441static int __init vrf_init_module(void)
1442{
1443 int rc;
1444
1445 register_netdevice_notifier(&vrf_notifier_block);
1446
1447 rc = register_pernet_subsys(&vrf_net_ops);
1448 if (rc < 0)
1449 goto error;
1450
1451 rc = rtnl_link_register(&vrf_link_ops);
1452 if (rc < 0) {
1453 unregister_pernet_subsys(&vrf_net_ops);
1454 goto error;
1455 }
1456
1457 return 0;
1458
1459error:
1460 unregister_netdevice_notifier(&vrf_notifier_block);
1461 return rc;
1462}
1463
1464module_init(vrf_init_module);
1465MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1466MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1467MODULE_LICENSE("GPL");
1468MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1469MODULE_VERSION(DRV_VERSION);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * vrf.c: device driver to encapsulate a VRF space
4 *
5 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
8 *
9 * Based on dummy, team and ipvlan drivers
10 */
11
12#include <linux/ethtool.h>
13#include <linux/module.h>
14#include <linux/kernel.h>
15#include <linux/netdevice.h>
16#include <linux/etherdevice.h>
17#include <linux/ip.h>
18#include <linux/init.h>
19#include <linux/moduleparam.h>
20#include <linux/netfilter.h>
21#include <linux/rtnetlink.h>
22#include <net/rtnetlink.h>
23#include <linux/u64_stats_sync.h>
24#include <linux/hashtable.h>
25#include <linux/spinlock_types.h>
26
27#include <linux/inetdevice.h>
28#include <net/arp.h>
29#include <net/ip.h>
30#include <net/ip_fib.h>
31#include <net/ip6_fib.h>
32#include <net/ip6_route.h>
33#include <net/route.h>
34#include <net/addrconf.h>
35#include <net/l3mdev.h>
36#include <net/fib_rules.h>
37#include <net/netns/generic.h>
38
39#define DRV_NAME "vrf"
40#define DRV_VERSION "1.1"
41
42#define FIB_RULE_PREF 1000 /* default preference for FIB rules */
43
44#define HT_MAP_BITS 4
45#define HASH_INITVAL ((u32)0xcafef00d)
46
47struct vrf_map {
48 DECLARE_HASHTABLE(ht, HT_MAP_BITS);
49 spinlock_t vmap_lock;
50
51 /* shared_tables:
52 * count how many distinct tables do not comply with the strict mode
53 * requirement.
54 * shared_tables value must be 0 in order to enable the strict mode.
55 *
56 * example of the evolution of shared_tables:
57 * | time
58 * add vrf0 --> table 100 shared_tables = 0 | t0
59 * add vrf1 --> table 101 shared_tables = 0 | t1
60 * add vrf2 --> table 100 shared_tables = 1 | t2
61 * add vrf3 --> table 100 shared_tables = 1 | t3
62 * add vrf4 --> table 101 shared_tables = 2 v t4
63 *
64 * shared_tables is a "step function" (or "staircase function")
65 * and it is increased by one when the second vrf is associated to a
66 * table.
67 *
68 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
69 *
70 * at t3, another dev (vrf3) is bound to the same table 100 but the
71 * value of shared_tables is still 1.
72 * This means that no matter how many new vrfs will register on the
73 * table 100, the shared_tables will not increase (considering only
74 * table 100).
75 *
76 * at t4, vrf4 is bound to table 101, and shared_tables = 2.
77 *
78 * Looking at the value of shared_tables we can immediately know if
79 * the strict_mode can or cannot be enforced. Indeed, strict_mode
80 * can be enforced iff shared_tables = 0.
81 *
82 * Conversely, shared_tables is decreased when a vrf is de-associated
83 * from a table with exactly two associated vrfs.
84 */
85 u32 shared_tables;
86
87 bool strict_mode;
88};
89
90struct vrf_map_elem {
91 struct hlist_node hnode;
92 struct list_head vrf_list; /* VRFs registered to this table */
93
94 u32 table_id;
95 int users;
96 int ifindex;
97};
98
99static unsigned int vrf_net_id;
100
101/* per netns vrf data */
102struct netns_vrf {
103 /* protected by rtnl lock */
104 bool add_fib_rules;
105
106 struct vrf_map vmap;
107 struct ctl_table_header *ctl_hdr;
108};
109
110struct net_vrf {
111 struct rtable __rcu *rth;
112 struct rt6_info __rcu *rt6;
113#if IS_ENABLED(CONFIG_IPV6)
114 struct fib6_table *fib6_table;
115#endif
116 u32 tb_id;
117
118 struct list_head me_list; /* entry in vrf_map_elem */
119 int ifindex;
120};
121
122struct pcpu_dstats {
123 u64 tx_pkts;
124 u64 tx_bytes;
125 u64 tx_drps;
126 u64 rx_pkts;
127 u64 rx_bytes;
128 u64 rx_drps;
129 struct u64_stats_sync syncp;
130};
131
132static void vrf_rx_stats(struct net_device *dev, int len)
133{
134 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
135
136 u64_stats_update_begin(&dstats->syncp);
137 dstats->rx_pkts++;
138 dstats->rx_bytes += len;
139 u64_stats_update_end(&dstats->syncp);
140}
141
142static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
143{
144 vrf_dev->stats.tx_errors++;
145 kfree_skb(skb);
146}
147
148static void vrf_get_stats64(struct net_device *dev,
149 struct rtnl_link_stats64 *stats)
150{
151 int i;
152
153 for_each_possible_cpu(i) {
154 const struct pcpu_dstats *dstats;
155 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
156 unsigned int start;
157
158 dstats = per_cpu_ptr(dev->dstats, i);
159 do {
160 start = u64_stats_fetch_begin_irq(&dstats->syncp);
161 tbytes = dstats->tx_bytes;
162 tpkts = dstats->tx_pkts;
163 tdrops = dstats->tx_drps;
164 rbytes = dstats->rx_bytes;
165 rpkts = dstats->rx_pkts;
166 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
167 stats->tx_bytes += tbytes;
168 stats->tx_packets += tpkts;
169 stats->tx_dropped += tdrops;
170 stats->rx_bytes += rbytes;
171 stats->rx_packets += rpkts;
172 }
173}
174
175static struct vrf_map *netns_vrf_map(struct net *net)
176{
177 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
178
179 return &nn_vrf->vmap;
180}
181
182static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
183{
184 return netns_vrf_map(dev_net(dev));
185}
186
187static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
188{
189 struct list_head *me_head = &me->vrf_list;
190 struct net_vrf *vrf;
191
192 if (list_empty(me_head))
193 return -ENODEV;
194
195 vrf = list_first_entry(me_head, struct net_vrf, me_list);
196
197 return vrf->ifindex;
198}
199
200static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
201{
202 struct vrf_map_elem *me;
203
204 me = kmalloc(sizeof(*me), flags);
205 if (!me)
206 return NULL;
207
208 return me;
209}
210
211static void vrf_map_elem_free(struct vrf_map_elem *me)
212{
213 kfree(me);
214}
215
216static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
217 int ifindex, int users)
218{
219 me->table_id = table_id;
220 me->ifindex = ifindex;
221 me->users = users;
222 INIT_LIST_HEAD(&me->vrf_list);
223}
224
225static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
226 u32 table_id)
227{
228 struct vrf_map_elem *me;
229 u32 key;
230
231 key = jhash_1word(table_id, HASH_INITVAL);
232 hash_for_each_possible(vmap->ht, me, hnode, key) {
233 if (me->table_id == table_id)
234 return me;
235 }
236
237 return NULL;
238}
239
240static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
241{
242 u32 table_id = me->table_id;
243 u32 key;
244
245 key = jhash_1word(table_id, HASH_INITVAL);
246 hash_add(vmap->ht, &me->hnode, key);
247}
248
249static void vrf_map_del_elem(struct vrf_map_elem *me)
250{
251 hash_del(&me->hnode);
252}
253
254static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
255{
256 spin_lock(&vmap->vmap_lock);
257}
258
259static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
260{
261 spin_unlock(&vmap->vmap_lock);
262}
263
264/* called with rtnl lock held */
265static int
266vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
267{
268 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
269 struct net_vrf *vrf = netdev_priv(dev);
270 struct vrf_map_elem *new_me, *me;
271 u32 table_id = vrf->tb_id;
272 bool free_new_me = false;
273 int users;
274 int res;
275
276 /* we pre-allocate elements used in the spin-locked section (so that we
277 * keep the spinlock as short as possible).
278 */
279 new_me = vrf_map_elem_alloc(GFP_KERNEL);
280 if (!new_me)
281 return -ENOMEM;
282
283 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
284
285 vrf_map_lock(vmap);
286
287 me = vrf_map_lookup_elem(vmap, table_id);
288 if (!me) {
289 me = new_me;
290 vrf_map_add_elem(vmap, me);
291 goto link_vrf;
292 }
293
294 /* we already have an entry in the vrf_map, so it means there is (at
295 * least) a vrf registered on the specific table.
296 */
297 free_new_me = true;
298 if (vmap->strict_mode) {
299 /* vrfs cannot share the same table */
300 NL_SET_ERR_MSG(extack, "Table is used by another VRF");
301 res = -EBUSY;
302 goto unlock;
303 }
304
305link_vrf:
306 users = ++me->users;
307 if (users == 2)
308 ++vmap->shared_tables;
309
310 list_add(&vrf->me_list, &me->vrf_list);
311
312 res = 0;
313
314unlock:
315 vrf_map_unlock(vmap);
316
317 /* clean-up, if needed */
318 if (free_new_me)
319 vrf_map_elem_free(new_me);
320
321 return res;
322}
323
324/* called with rtnl lock held */
325static void vrf_map_unregister_dev(struct net_device *dev)
326{
327 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
328 struct net_vrf *vrf = netdev_priv(dev);
329 u32 table_id = vrf->tb_id;
330 struct vrf_map_elem *me;
331 int users;
332
333 vrf_map_lock(vmap);
334
335 me = vrf_map_lookup_elem(vmap, table_id);
336 if (!me)
337 goto unlock;
338
339 list_del(&vrf->me_list);
340
341 users = --me->users;
342 if (users == 1) {
343 --vmap->shared_tables;
344 } else if (users == 0) {
345 vrf_map_del_elem(me);
346
347 /* no one will refer to this element anymore */
348 vrf_map_elem_free(me);
349 }
350
351unlock:
352 vrf_map_unlock(vmap);
353}
354
355/* return the vrf device index associated with the table_id */
356static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
357{
358 struct vrf_map *vmap = netns_vrf_map(net);
359 struct vrf_map_elem *me;
360 int ifindex;
361
362 vrf_map_lock(vmap);
363
364 if (!vmap->strict_mode) {
365 ifindex = -EPERM;
366 goto unlock;
367 }
368
369 me = vrf_map_lookup_elem(vmap, table_id);
370 if (!me) {
371 ifindex = -ENODEV;
372 goto unlock;
373 }
374
375 ifindex = vrf_map_elem_get_vrf_ifindex(me);
376
377unlock:
378 vrf_map_unlock(vmap);
379
380 return ifindex;
381}
382
383/* by default VRF devices do not have a qdisc and are expected
384 * to be created with only a single queue.
385 */
386static bool qdisc_tx_is_default(const struct net_device *dev)
387{
388 struct netdev_queue *txq;
389 struct Qdisc *qdisc;
390
391 if (dev->num_tx_queues > 1)
392 return false;
393
394 txq = netdev_get_tx_queue(dev, 0);
395 qdisc = rcu_access_pointer(txq->qdisc);
396
397 return !qdisc->enqueue;
398}
399
400/* Local traffic destined to local address. Reinsert the packet to rx
401 * path, similar to loopback handling.
402 */
403static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
404 struct dst_entry *dst)
405{
406 int len = skb->len;
407
408 skb_orphan(skb);
409
410 skb_dst_set(skb, dst);
411
412 /* set pkt_type to avoid skb hitting packet taps twice -
413 * once on Tx and again in Rx processing
414 */
415 skb->pkt_type = PACKET_LOOPBACK;
416
417 skb->protocol = eth_type_trans(skb, dev);
418
419 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
420 vrf_rx_stats(dev, len);
421 else
422 this_cpu_inc(dev->dstats->rx_drps);
423
424 return NETDEV_TX_OK;
425}
426
427#if IS_ENABLED(CONFIG_IPV6)
428static int vrf_ip6_local_out(struct net *net, struct sock *sk,
429 struct sk_buff *skb)
430{
431 int err;
432
433 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
434 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
435
436 if (likely(err == 1))
437 err = dst_output(net, sk, skb);
438
439 return err;
440}
441
442static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
443 struct net_device *dev)
444{
445 const struct ipv6hdr *iph;
446 struct net *net = dev_net(skb->dev);
447 struct flowi6 fl6;
448 int ret = NET_XMIT_DROP;
449 struct dst_entry *dst;
450 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
451
452 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
453 goto err;
454
455 iph = ipv6_hdr(skb);
456
457 memset(&fl6, 0, sizeof(fl6));
458 /* needed to match OIF rule */
459 fl6.flowi6_oif = dev->ifindex;
460 fl6.flowi6_iif = LOOPBACK_IFINDEX;
461 fl6.daddr = iph->daddr;
462 fl6.saddr = iph->saddr;
463 fl6.flowlabel = ip6_flowinfo(iph);
464 fl6.flowi6_mark = skb->mark;
465 fl6.flowi6_proto = iph->nexthdr;
466 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
467
468 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
469 if (IS_ERR(dst) || dst == dst_null)
470 goto err;
471
472 skb_dst_drop(skb);
473
474 /* if dst.dev is the VRF device again this is locally originated traffic
475 * destined to a local address. Short circuit to Rx path.
476 */
477 if (dst->dev == dev)
478 return vrf_local_xmit(skb, dev, dst);
479
480 skb_dst_set(skb, dst);
481
482 /* strip the ethernet header added for pass through VRF device */
483 __skb_pull(skb, skb_network_offset(skb));
484
485 ret = vrf_ip6_local_out(net, skb->sk, skb);
486 if (unlikely(net_xmit_eval(ret)))
487 dev->stats.tx_errors++;
488 else
489 ret = NET_XMIT_SUCCESS;
490
491 return ret;
492err:
493 vrf_tx_error(dev, skb);
494 return NET_XMIT_DROP;
495}
496#else
497static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
498 struct net_device *dev)
499{
500 vrf_tx_error(dev, skb);
501 return NET_XMIT_DROP;
502}
503#endif
504
505/* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
506static int vrf_ip_local_out(struct net *net, struct sock *sk,
507 struct sk_buff *skb)
508{
509 int err;
510
511 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
512 skb, NULL, skb_dst(skb)->dev, dst_output);
513 if (likely(err == 1))
514 err = dst_output(net, sk, skb);
515
516 return err;
517}
518
519static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
520 struct net_device *vrf_dev)
521{
522 struct iphdr *ip4h;
523 int ret = NET_XMIT_DROP;
524 struct flowi4 fl4;
525 struct net *net = dev_net(vrf_dev);
526 struct rtable *rt;
527
528 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
529 goto err;
530
531 ip4h = ip_hdr(skb);
532
533 memset(&fl4, 0, sizeof(fl4));
534 /* needed to match OIF rule */
535 fl4.flowi4_oif = vrf_dev->ifindex;
536 fl4.flowi4_iif = LOOPBACK_IFINDEX;
537 fl4.flowi4_tos = RT_TOS(ip4h->tos);
538 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
539 fl4.flowi4_proto = ip4h->protocol;
540 fl4.daddr = ip4h->daddr;
541 fl4.saddr = ip4h->saddr;
542
543 rt = ip_route_output_flow(net, &fl4, NULL);
544 if (IS_ERR(rt))
545 goto err;
546
547 skb_dst_drop(skb);
548
549 /* if dst.dev is the VRF device again this is locally originated traffic
550 * destined to a local address. Short circuit to Rx path.
551 */
552 if (rt->dst.dev == vrf_dev)
553 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
554
555 skb_dst_set(skb, &rt->dst);
556
557 /* strip the ethernet header added for pass through VRF device */
558 __skb_pull(skb, skb_network_offset(skb));
559
560 if (!ip4h->saddr) {
561 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
562 RT_SCOPE_LINK);
563 }
564
565 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
566 if (unlikely(net_xmit_eval(ret)))
567 vrf_dev->stats.tx_errors++;
568 else
569 ret = NET_XMIT_SUCCESS;
570
571out:
572 return ret;
573err:
574 vrf_tx_error(vrf_dev, skb);
575 goto out;
576}
577
578static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
579{
580 switch (skb->protocol) {
581 case htons(ETH_P_IP):
582 return vrf_process_v4_outbound(skb, dev);
583 case htons(ETH_P_IPV6):
584 return vrf_process_v6_outbound(skb, dev);
585 default:
586 vrf_tx_error(dev, skb);
587 return NET_XMIT_DROP;
588 }
589}
590
591static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
592{
593 int len = skb->len;
594 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
595
596 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
597 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
598
599 u64_stats_update_begin(&dstats->syncp);
600 dstats->tx_pkts++;
601 dstats->tx_bytes += len;
602 u64_stats_update_end(&dstats->syncp);
603 } else {
604 this_cpu_inc(dev->dstats->tx_drps);
605 }
606
607 return ret;
608}
609
610static void vrf_finish_direct(struct sk_buff *skb)
611{
612 struct net_device *vrf_dev = skb->dev;
613
614 if (!list_empty(&vrf_dev->ptype_all) &&
615 likely(skb_headroom(skb) >= ETH_HLEN)) {
616 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
617
618 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
619 eth_zero_addr(eth->h_dest);
620 eth->h_proto = skb->protocol;
621
622 rcu_read_lock_bh();
623 dev_queue_xmit_nit(skb, vrf_dev);
624 rcu_read_unlock_bh();
625
626 skb_pull(skb, ETH_HLEN);
627 }
628
629 /* reset skb device */
630 nf_reset_ct(skb);
631}
632
633#if IS_ENABLED(CONFIG_IPV6)
634/* modelled after ip6_finish_output2 */
635static int vrf_finish_output6(struct net *net, struct sock *sk,
636 struct sk_buff *skb)
637{
638 struct dst_entry *dst = skb_dst(skb);
639 struct net_device *dev = dst->dev;
640 const struct in6_addr *nexthop;
641 struct neighbour *neigh;
642 int ret;
643
644 nf_reset_ct(skb);
645
646 skb->protocol = htons(ETH_P_IPV6);
647 skb->dev = dev;
648
649 rcu_read_lock_bh();
650 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
651 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
652 if (unlikely(!neigh))
653 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
654 if (!IS_ERR(neigh)) {
655 sock_confirm_neigh(skb, neigh);
656 ret = neigh_output(neigh, skb, false);
657 rcu_read_unlock_bh();
658 return ret;
659 }
660 rcu_read_unlock_bh();
661
662 IP6_INC_STATS(dev_net(dst->dev),
663 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
664 kfree_skb(skb);
665 return -EINVAL;
666}
667
668/* modelled after ip6_output */
669static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
670{
671 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
672 net, sk, skb, NULL, skb_dst(skb)->dev,
673 vrf_finish_output6,
674 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
675}
676
677/* set dst on skb to send packet to us via dev_xmit path. Allows
678 * packet to go through device based features such as qdisc, netfilter
679 * hooks and packet sockets with skb->dev set to vrf device.
680 */
681static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
682 struct sk_buff *skb)
683{
684 struct net_vrf *vrf = netdev_priv(vrf_dev);
685 struct dst_entry *dst = NULL;
686 struct rt6_info *rt6;
687
688 rcu_read_lock();
689
690 rt6 = rcu_dereference(vrf->rt6);
691 if (likely(rt6)) {
692 dst = &rt6->dst;
693 dst_hold(dst);
694 }
695
696 rcu_read_unlock();
697
698 if (unlikely(!dst)) {
699 vrf_tx_error(vrf_dev, skb);
700 return NULL;
701 }
702
703 skb_dst_drop(skb);
704 skb_dst_set(skb, dst);
705
706 return skb;
707}
708
709static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
710 struct sk_buff *skb)
711{
712 vrf_finish_direct(skb);
713
714 return vrf_ip6_local_out(net, sk, skb);
715}
716
717static int vrf_output6_direct(struct net *net, struct sock *sk,
718 struct sk_buff *skb)
719{
720 int err = 1;
721
722 skb->protocol = htons(ETH_P_IPV6);
723
724 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
725 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
726 NULL, skb->dev, vrf_output6_direct_finish);
727
728 if (likely(err == 1))
729 vrf_finish_direct(skb);
730
731 return err;
732}
733
734static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
735 struct sk_buff *skb)
736{
737 int err;
738
739 err = vrf_output6_direct(net, sk, skb);
740 if (likely(err == 1))
741 err = vrf_ip6_local_out(net, sk, skb);
742
743 return err;
744}
745
746static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
747 struct sock *sk,
748 struct sk_buff *skb)
749{
750 struct net *net = dev_net(vrf_dev);
751 int err;
752
753 skb->dev = vrf_dev;
754
755 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
756 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
757
758 if (likely(err == 1))
759 err = vrf_output6_direct(net, sk, skb);
760
761 if (likely(err == 1))
762 return skb;
763
764 return NULL;
765}
766
767static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
768 struct sock *sk,
769 struct sk_buff *skb)
770{
771 /* don't divert link scope packets */
772 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
773 return skb;
774
775 if (qdisc_tx_is_default(vrf_dev) ||
776 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
777 return vrf_ip6_out_direct(vrf_dev, sk, skb);
778
779 return vrf_ip6_out_redirect(vrf_dev, skb);
780}
781
782/* holding rtnl */
783static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
784{
785 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
786 struct net *net = dev_net(dev);
787 struct dst_entry *dst;
788
789 RCU_INIT_POINTER(vrf->rt6, NULL);
790 synchronize_rcu();
791
792 /* move dev in dst's to loopback so this VRF device can be deleted
793 * - based on dst_ifdown
794 */
795 if (rt6) {
796 dst = &rt6->dst;
797 dev_put(dst->dev);
798 dst->dev = net->loopback_dev;
799 dev_hold(dst->dev);
800 dst_release(dst);
801 }
802}
803
804static int vrf_rt6_create(struct net_device *dev)
805{
806 int flags = DST_NOPOLICY | DST_NOXFRM;
807 struct net_vrf *vrf = netdev_priv(dev);
808 struct net *net = dev_net(dev);
809 struct rt6_info *rt6;
810 int rc = -ENOMEM;
811
812 /* IPv6 can be CONFIG enabled and then disabled runtime */
813 if (!ipv6_mod_enabled())
814 return 0;
815
816 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
817 if (!vrf->fib6_table)
818 goto out;
819
820 /* create a dst for routing packets out a VRF device */
821 rt6 = ip6_dst_alloc(net, dev, flags);
822 if (!rt6)
823 goto out;
824
825 rt6->dst.output = vrf_output6;
826
827 rcu_assign_pointer(vrf->rt6, rt6);
828
829 rc = 0;
830out:
831 return rc;
832}
833#else
834static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
835 struct sock *sk,
836 struct sk_buff *skb)
837{
838 return skb;
839}
840
841static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
842{
843}
844
845static int vrf_rt6_create(struct net_device *dev)
846{
847 return 0;
848}
849#endif
850
851/* modelled after ip_finish_output2 */
852static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
853{
854 struct dst_entry *dst = skb_dst(skb);
855 struct rtable *rt = (struct rtable *)dst;
856 struct net_device *dev = dst->dev;
857 unsigned int hh_len = LL_RESERVED_SPACE(dev);
858 struct neighbour *neigh;
859 bool is_v6gw = false;
860 int ret = -EINVAL;
861
862 nf_reset_ct(skb);
863
864 /* Be paranoid, rather than too clever. */
865 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
866 struct sk_buff *skb2;
867
868 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
869 if (!skb2) {
870 ret = -ENOMEM;
871 goto err;
872 }
873 if (skb->sk)
874 skb_set_owner_w(skb2, skb->sk);
875
876 consume_skb(skb);
877 skb = skb2;
878 }
879
880 rcu_read_lock_bh();
881
882 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
883 if (!IS_ERR(neigh)) {
884 sock_confirm_neigh(skb, neigh);
885 /* if crossing protocols, can not use the cached header */
886 ret = neigh_output(neigh, skb, is_v6gw);
887 rcu_read_unlock_bh();
888 return ret;
889 }
890
891 rcu_read_unlock_bh();
892err:
893 vrf_tx_error(skb->dev, skb);
894 return ret;
895}
896
897static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
898{
899 struct net_device *dev = skb_dst(skb)->dev;
900
901 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
902
903 skb->dev = dev;
904 skb->protocol = htons(ETH_P_IP);
905
906 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
907 net, sk, skb, NULL, dev,
908 vrf_finish_output,
909 !(IPCB(skb)->flags & IPSKB_REROUTED));
910}
911
912/* set dst on skb to send packet to us via dev_xmit path. Allows
913 * packet to go through device based features such as qdisc, netfilter
914 * hooks and packet sockets with skb->dev set to vrf device.
915 */
916static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
917 struct sk_buff *skb)
918{
919 struct net_vrf *vrf = netdev_priv(vrf_dev);
920 struct dst_entry *dst = NULL;
921 struct rtable *rth;
922
923 rcu_read_lock();
924
925 rth = rcu_dereference(vrf->rth);
926 if (likely(rth)) {
927 dst = &rth->dst;
928 dst_hold(dst);
929 }
930
931 rcu_read_unlock();
932
933 if (unlikely(!dst)) {
934 vrf_tx_error(vrf_dev, skb);
935 return NULL;
936 }
937
938 skb_dst_drop(skb);
939 skb_dst_set(skb, dst);
940
941 return skb;
942}
943
944static int vrf_output_direct_finish(struct net *net, struct sock *sk,
945 struct sk_buff *skb)
946{
947 vrf_finish_direct(skb);
948
949 return vrf_ip_local_out(net, sk, skb);
950}
951
952static int vrf_output_direct(struct net *net, struct sock *sk,
953 struct sk_buff *skb)
954{
955 int err = 1;
956
957 skb->protocol = htons(ETH_P_IP);
958
959 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
960 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
961 NULL, skb->dev, vrf_output_direct_finish);
962
963 if (likely(err == 1))
964 vrf_finish_direct(skb);
965
966 return err;
967}
968
969static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
970 struct sk_buff *skb)
971{
972 int err;
973
974 err = vrf_output_direct(net, sk, skb);
975 if (likely(err == 1))
976 err = vrf_ip_local_out(net, sk, skb);
977
978 return err;
979}
980
981static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
982 struct sock *sk,
983 struct sk_buff *skb)
984{
985 struct net *net = dev_net(vrf_dev);
986 int err;
987
988 skb->dev = vrf_dev;
989
990 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
991 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
992
993 if (likely(err == 1))
994 err = vrf_output_direct(net, sk, skb);
995
996 if (likely(err == 1))
997 return skb;
998
999 return NULL;
1000}
1001
1002static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
1003 struct sock *sk,
1004 struct sk_buff *skb)
1005{
1006 /* don't divert multicast or local broadcast */
1007 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
1008 ipv4_is_lbcast(ip_hdr(skb)->daddr))
1009 return skb;
1010
1011 if (qdisc_tx_is_default(vrf_dev) ||
1012 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
1013 return vrf_ip_out_direct(vrf_dev, sk, skb);
1014
1015 return vrf_ip_out_redirect(vrf_dev, skb);
1016}
1017
1018/* called with rcu lock held */
1019static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1020 struct sock *sk,
1021 struct sk_buff *skb,
1022 u16 proto)
1023{
1024 switch (proto) {
1025 case AF_INET:
1026 return vrf_ip_out(vrf_dev, sk, skb);
1027 case AF_INET6:
1028 return vrf_ip6_out(vrf_dev, sk, skb);
1029 }
1030
1031 return skb;
1032}
1033
1034/* holding rtnl */
1035static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
1036{
1037 struct rtable *rth = rtnl_dereference(vrf->rth);
1038 struct net *net = dev_net(dev);
1039 struct dst_entry *dst;
1040
1041 RCU_INIT_POINTER(vrf->rth, NULL);
1042 synchronize_rcu();
1043
1044 /* move dev in dst's to loopback so this VRF device can be deleted
1045 * - based on dst_ifdown
1046 */
1047 if (rth) {
1048 dst = &rth->dst;
1049 dev_put(dst->dev);
1050 dst->dev = net->loopback_dev;
1051 dev_hold(dst->dev);
1052 dst_release(dst);
1053 }
1054}
1055
1056static int vrf_rtable_create(struct net_device *dev)
1057{
1058 struct net_vrf *vrf = netdev_priv(dev);
1059 struct rtable *rth;
1060
1061 if (!fib_new_table(dev_net(dev), vrf->tb_id))
1062 return -ENOMEM;
1063
1064 /* create a dst for routing packets out through a VRF device */
1065 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
1066 if (!rth)
1067 return -ENOMEM;
1068
1069 rth->dst.output = vrf_output;
1070
1071 rcu_assign_pointer(vrf->rth, rth);
1072
1073 return 0;
1074}
1075
1076/**************************** device handling ********************/
1077
1078/* cycle interface to flush neighbor cache and move routes across tables */
1079static void cycle_netdev(struct net_device *dev,
1080 struct netlink_ext_ack *extack)
1081{
1082 unsigned int flags = dev->flags;
1083 int ret;
1084
1085 if (!netif_running(dev))
1086 return;
1087
1088 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1089 if (ret >= 0)
1090 ret = dev_change_flags(dev, flags, extack);
1091
1092 if (ret < 0) {
1093 netdev_err(dev,
1094 "Failed to cycle device %s; route tables might be wrong!\n",
1095 dev->name);
1096 }
1097}
1098
1099static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1100 struct netlink_ext_ack *extack)
1101{
1102 int ret;
1103
1104 /* do not allow loopback device to be enslaved to a VRF.
1105 * The vrf device acts as the loopback for the vrf.
1106 */
1107 if (port_dev == dev_net(dev)->loopback_dev) {
1108 NL_SET_ERR_MSG(extack,
1109 "Can not enslave loopback device to a VRF");
1110 return -EOPNOTSUPP;
1111 }
1112
1113 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1114 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1115 if (ret < 0)
1116 goto err;
1117
1118 cycle_netdev(port_dev, extack);
1119
1120 return 0;
1121
1122err:
1123 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1124 return ret;
1125}
1126
1127static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1128 struct netlink_ext_ack *extack)
1129{
1130 if (netif_is_l3_master(port_dev)) {
1131 NL_SET_ERR_MSG(extack,
1132 "Can not enslave an L3 master device to a VRF");
1133 return -EINVAL;
1134 }
1135
1136 if (netif_is_l3_slave(port_dev))
1137 return -EINVAL;
1138
1139 return do_vrf_add_slave(dev, port_dev, extack);
1140}
1141
1142/* inverse of do_vrf_add_slave */
1143static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1144{
1145 netdev_upper_dev_unlink(port_dev, dev);
1146 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1147
1148 cycle_netdev(port_dev, NULL);
1149
1150 return 0;
1151}
1152
1153static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1154{
1155 return do_vrf_del_slave(dev, port_dev);
1156}
1157
1158static void vrf_dev_uninit(struct net_device *dev)
1159{
1160 struct net_vrf *vrf = netdev_priv(dev);
1161
1162 vrf_rtable_release(dev, vrf);
1163 vrf_rt6_release(dev, vrf);
1164
1165 free_percpu(dev->dstats);
1166 dev->dstats = NULL;
1167}
1168
1169static int vrf_dev_init(struct net_device *dev)
1170{
1171 struct net_vrf *vrf = netdev_priv(dev);
1172
1173 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1174 if (!dev->dstats)
1175 goto out_nomem;
1176
1177 /* create the default dst which points back to us */
1178 if (vrf_rtable_create(dev) != 0)
1179 goto out_stats;
1180
1181 if (vrf_rt6_create(dev) != 0)
1182 goto out_rth;
1183
1184 dev->flags = IFF_MASTER | IFF_NOARP;
1185
1186 /* similarly, oper state is irrelevant; set to up to avoid confusion */
1187 dev->operstate = IF_OPER_UP;
1188 netdev_lockdep_set_classes(dev);
1189 return 0;
1190
1191out_rth:
1192 vrf_rtable_release(dev, vrf);
1193out_stats:
1194 free_percpu(dev->dstats);
1195 dev->dstats = NULL;
1196out_nomem:
1197 return -ENOMEM;
1198}
1199
1200static const struct net_device_ops vrf_netdev_ops = {
1201 .ndo_init = vrf_dev_init,
1202 .ndo_uninit = vrf_dev_uninit,
1203 .ndo_start_xmit = vrf_xmit,
1204 .ndo_set_mac_address = eth_mac_addr,
1205 .ndo_get_stats64 = vrf_get_stats64,
1206 .ndo_add_slave = vrf_add_slave,
1207 .ndo_del_slave = vrf_del_slave,
1208};
1209
1210static u32 vrf_fib_table(const struct net_device *dev)
1211{
1212 struct net_vrf *vrf = netdev_priv(dev);
1213
1214 return vrf->tb_id;
1215}
1216
1217static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1218{
1219 kfree_skb(skb);
1220 return 0;
1221}
1222
1223static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1224 struct sk_buff *skb,
1225 struct net_device *dev)
1226{
1227 struct net *net = dev_net(dev);
1228
1229 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1230 skb = NULL; /* kfree_skb(skb) handled by nf code */
1231
1232 return skb;
1233}
1234
1235static int vrf_prepare_mac_header(struct sk_buff *skb,
1236 struct net_device *vrf_dev, u16 proto)
1237{
1238 struct ethhdr *eth;
1239 int err;
1240
1241 /* in general, we do not know if there is enough space in the head of
1242 * the packet for hosting the mac header.
1243 */
1244 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
1245 if (unlikely(err))
1246 /* no space in the skb head */
1247 return -ENOBUFS;
1248
1249 __skb_push(skb, ETH_HLEN);
1250 eth = (struct ethhdr *)skb->data;
1251
1252 skb_reset_mac_header(skb);
1253
1254 /* we set the ethernet destination and the source addresses to the
1255 * address of the VRF device.
1256 */
1257 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
1258 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
1259 eth->h_proto = htons(proto);
1260
1261 /* the destination address of the Ethernet frame corresponds to the
1262 * address set on the VRF interface; therefore, the packet is intended
1263 * to be processed locally.
1264 */
1265 skb->protocol = eth->h_proto;
1266 skb->pkt_type = PACKET_HOST;
1267
1268 skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
1269
1270 skb_pull_inline(skb, ETH_HLEN);
1271
1272 return 0;
1273}
1274
1275/* prepare and add the mac header to the packet if it was not set previously.
1276 * In this way, packet sniffers such as tcpdump can parse the packet correctly.
1277 * If the mac header was already set, the original mac header is left
1278 * untouched and the function returns immediately.
1279 */
1280static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
1281 struct net_device *vrf_dev,
1282 u16 proto)
1283{
1284 if (skb_mac_header_was_set(skb))
1285 return 0;
1286
1287 return vrf_prepare_mac_header(skb, vrf_dev, proto);
1288}
1289
1290#if IS_ENABLED(CONFIG_IPV6)
1291/* neighbor handling is done with actual device; do not want
1292 * to flip skb->dev for those ndisc packets. This really fails
1293 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1294 * a start.
1295 */
1296static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1297{
1298 const struct ipv6hdr *iph = ipv6_hdr(skb);
1299 bool rc = false;
1300
1301 if (iph->nexthdr == NEXTHDR_ICMP) {
1302 const struct icmp6hdr *icmph;
1303 struct icmp6hdr _icmph;
1304
1305 icmph = skb_header_pointer(skb, sizeof(*iph),
1306 sizeof(_icmph), &_icmph);
1307 if (!icmph)
1308 goto out;
1309
1310 switch (icmph->icmp6_type) {
1311 case NDISC_ROUTER_SOLICITATION:
1312 case NDISC_ROUTER_ADVERTISEMENT:
1313 case NDISC_NEIGHBOUR_SOLICITATION:
1314 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1315 case NDISC_REDIRECT:
1316 rc = true;
1317 break;
1318 }
1319 }
1320
1321out:
1322 return rc;
1323}
1324
1325static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1326 const struct net_device *dev,
1327 struct flowi6 *fl6,
1328 int ifindex,
1329 const struct sk_buff *skb,
1330 int flags)
1331{
1332 struct net_vrf *vrf = netdev_priv(dev);
1333
1334 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1335}
1336
1337static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1338 int ifindex)
1339{
1340 const struct ipv6hdr *iph = ipv6_hdr(skb);
1341 struct flowi6 fl6 = {
1342 .flowi6_iif = ifindex,
1343 .flowi6_mark = skb->mark,
1344 .flowi6_proto = iph->nexthdr,
1345 .daddr = iph->daddr,
1346 .saddr = iph->saddr,
1347 .flowlabel = ip6_flowinfo(iph),
1348 };
1349 struct net *net = dev_net(vrf_dev);
1350 struct rt6_info *rt6;
1351
1352 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1353 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1354 if (unlikely(!rt6))
1355 return;
1356
1357 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1358 return;
1359
1360 skb_dst_set(skb, &rt6->dst);
1361}
1362
1363static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1364 struct sk_buff *skb)
1365{
1366 int orig_iif = skb->skb_iif;
1367 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1368 bool is_ndisc = ipv6_ndisc_frame(skb);
1369
1370 nf_reset_ct(skb);
1371
1372 /* loopback, multicast & non-ND link-local traffic; do not push through
1373 * packet taps again. Reset pkt_type for upper layers to process skb.
1374 * For strict packets with a source LLA, determine the dst using the
1375 * original ifindex.
1376 */
1377 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1378 skb->dev = vrf_dev;
1379 skb->skb_iif = vrf_dev->ifindex;
1380 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1381
1382 if (skb->pkt_type == PACKET_LOOPBACK)
1383 skb->pkt_type = PACKET_HOST;
1384 else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL)
1385 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1386
1387 goto out;
1388 }
1389
1390 /* if packet is NDISC then keep the ingress interface */
1391 if (!is_ndisc) {
1392 vrf_rx_stats(vrf_dev, skb->len);
1393 skb->dev = vrf_dev;
1394 skb->skb_iif = vrf_dev->ifindex;
1395
1396 if (!list_empty(&vrf_dev->ptype_all)) {
1397 int err;
1398
1399 err = vrf_add_mac_header_if_unset(skb, vrf_dev,
1400 ETH_P_IPV6);
1401 if (likely(!err)) {
1402 skb_push(skb, skb->mac_len);
1403 dev_queue_xmit_nit(skb, vrf_dev);
1404 skb_pull(skb, skb->mac_len);
1405 }
1406 }
1407
1408 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1409 }
1410
1411 if (need_strict)
1412 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1413
1414 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1415out:
1416 return skb;
1417}
1418
1419#else
1420static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1421 struct sk_buff *skb)
1422{
1423 return skb;
1424}
1425#endif
1426
1427static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1428 struct sk_buff *skb)
1429{
1430 skb->dev = vrf_dev;
1431 skb->skb_iif = vrf_dev->ifindex;
1432 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1433
1434 nf_reset_ct(skb);
1435
1436 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1437 goto out;
1438
1439 /* loopback traffic; do not push through packet taps again.
1440 * Reset pkt_type for upper layers to process skb
1441 */
1442 if (skb->pkt_type == PACKET_LOOPBACK) {
1443 skb->pkt_type = PACKET_HOST;
1444 goto out;
1445 }
1446
1447 vrf_rx_stats(vrf_dev, skb->len);
1448
1449 if (!list_empty(&vrf_dev->ptype_all)) {
1450 int err;
1451
1452 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP);
1453 if (likely(!err)) {
1454 skb_push(skb, skb->mac_len);
1455 dev_queue_xmit_nit(skb, vrf_dev);
1456 skb_pull(skb, skb->mac_len);
1457 }
1458 }
1459
1460 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1461out:
1462 return skb;
1463}
1464
1465/* called with rcu lock held */
1466static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1467 struct sk_buff *skb,
1468 u16 proto)
1469{
1470 switch (proto) {
1471 case AF_INET:
1472 return vrf_ip_rcv(vrf_dev, skb);
1473 case AF_INET6:
1474 return vrf_ip6_rcv(vrf_dev, skb);
1475 }
1476
1477 return skb;
1478}
1479
1480#if IS_ENABLED(CONFIG_IPV6)
1481/* send to link-local or multicast address via interface enslaved to
1482 * VRF device. Force lookup to VRF table without changing flow struct
1483 * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1484 * is taken on the dst by this function.
1485 */
1486static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1487 struct flowi6 *fl6)
1488{
1489 struct net *net = dev_net(dev);
1490 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1491 struct dst_entry *dst = NULL;
1492 struct rt6_info *rt;
1493
1494 /* VRF device does not have a link-local address and
1495 * sending packets to link-local or mcast addresses over
1496 * a VRF device does not make sense
1497 */
1498 if (fl6->flowi6_oif == dev->ifindex) {
1499 dst = &net->ipv6.ip6_null_entry->dst;
1500 return dst;
1501 }
1502
1503 if (!ipv6_addr_any(&fl6->saddr))
1504 flags |= RT6_LOOKUP_F_HAS_SADDR;
1505
1506 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1507 if (rt)
1508 dst = &rt->dst;
1509
1510 return dst;
1511}
1512#endif
1513
1514static const struct l3mdev_ops vrf_l3mdev_ops = {
1515 .l3mdev_fib_table = vrf_fib_table,
1516 .l3mdev_l3_rcv = vrf_l3_rcv,
1517 .l3mdev_l3_out = vrf_l3_out,
1518#if IS_ENABLED(CONFIG_IPV6)
1519 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1520#endif
1521};
1522
1523static void vrf_get_drvinfo(struct net_device *dev,
1524 struct ethtool_drvinfo *info)
1525{
1526 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1527 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1528}
1529
1530static const struct ethtool_ops vrf_ethtool_ops = {
1531 .get_drvinfo = vrf_get_drvinfo,
1532};
1533
1534static inline size_t vrf_fib_rule_nl_size(void)
1535{
1536 size_t sz;
1537
1538 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1539 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1540 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1541 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1542
1543 return sz;
1544}
1545
1546static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1547{
1548 struct fib_rule_hdr *frh;
1549 struct nlmsghdr *nlh;
1550 struct sk_buff *skb;
1551 int err;
1552
1553 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1554 !ipv6_mod_enabled())
1555 return 0;
1556
1557 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1558 if (!skb)
1559 return -ENOMEM;
1560
1561 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1562 if (!nlh)
1563 goto nla_put_failure;
1564
1565 /* rule only needs to appear once */
1566 nlh->nlmsg_flags |= NLM_F_EXCL;
1567
1568 frh = nlmsg_data(nlh);
1569 memset(frh, 0, sizeof(*frh));
1570 frh->family = family;
1571 frh->action = FR_ACT_TO_TBL;
1572
1573 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1574 goto nla_put_failure;
1575
1576 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1577 goto nla_put_failure;
1578
1579 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1580 goto nla_put_failure;
1581
1582 nlmsg_end(skb, nlh);
1583
1584 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1585 skb->sk = dev_net(dev)->rtnl;
1586 if (add_it) {
1587 err = fib_nl_newrule(skb, nlh, NULL);
1588 if (err == -EEXIST)
1589 err = 0;
1590 } else {
1591 err = fib_nl_delrule(skb, nlh, NULL);
1592 if (err == -ENOENT)
1593 err = 0;
1594 }
1595 nlmsg_free(skb);
1596
1597 return err;
1598
1599nla_put_failure:
1600 nlmsg_free(skb);
1601
1602 return -EMSGSIZE;
1603}
1604
1605static int vrf_add_fib_rules(const struct net_device *dev)
1606{
1607 int err;
1608
1609 err = vrf_fib_rule(dev, AF_INET, true);
1610 if (err < 0)
1611 goto out_err;
1612
1613 err = vrf_fib_rule(dev, AF_INET6, true);
1614 if (err < 0)
1615 goto ipv6_err;
1616
1617#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1618 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1619 if (err < 0)
1620 goto ipmr_err;
1621#endif
1622
1623#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1624 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1625 if (err < 0)
1626 goto ip6mr_err;
1627#endif
1628
1629 return 0;
1630
1631#if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1632ip6mr_err:
1633 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1634#endif
1635
1636#if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1637ipmr_err:
1638 vrf_fib_rule(dev, AF_INET6, false);
1639#endif
1640
1641ipv6_err:
1642 vrf_fib_rule(dev, AF_INET, false);
1643
1644out_err:
1645 netdev_err(dev, "Failed to add FIB rules.\n");
1646 return err;
1647}
1648
1649static void vrf_setup(struct net_device *dev)
1650{
1651 ether_setup(dev);
1652
1653 /* Initialize the device structure. */
1654 dev->netdev_ops = &vrf_netdev_ops;
1655 dev->l3mdev_ops = &vrf_l3mdev_ops;
1656 dev->ethtool_ops = &vrf_ethtool_ops;
1657 dev->needs_free_netdev = true;
1658
1659 /* Fill in device structure with ethernet-generic values. */
1660 eth_hw_addr_random(dev);
1661
1662 /* don't acquire vrf device's netif_tx_lock when transmitting */
1663 dev->features |= NETIF_F_LLTX;
1664
1665 /* don't allow vrf devices to change network namespaces. */
1666 dev->features |= NETIF_F_NETNS_LOCAL;
1667
1668 /* does not make sense for a VLAN to be added to a vrf device */
1669 dev->features |= NETIF_F_VLAN_CHALLENGED;
1670
1671 /* enable offload features */
1672 dev->features |= NETIF_F_GSO_SOFTWARE;
1673 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1674 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1675
1676 dev->hw_features = dev->features;
1677 dev->hw_enc_features = dev->features;
1678
1679 /* default to no qdisc; user can add if desired */
1680 dev->priv_flags |= IFF_NO_QUEUE;
1681 dev->priv_flags |= IFF_NO_RX_HANDLER;
1682 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1683
1684 /* VRF devices do not care about MTU, but if the MTU is set
1685 * too low then the ipv4 and ipv6 protocols are disabled
1686 * which breaks networking.
1687 */
1688 dev->min_mtu = IPV6_MIN_MTU;
1689 dev->max_mtu = IP6_MAX_MTU;
1690 dev->mtu = dev->max_mtu;
1691}
1692
1693static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1694 struct netlink_ext_ack *extack)
1695{
1696 if (tb[IFLA_ADDRESS]) {
1697 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1698 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1699 return -EINVAL;
1700 }
1701 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1702 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1703 return -EADDRNOTAVAIL;
1704 }
1705 }
1706 return 0;
1707}
1708
1709static void vrf_dellink(struct net_device *dev, struct list_head *head)
1710{
1711 struct net_device *port_dev;
1712 struct list_head *iter;
1713
1714 netdev_for_each_lower_dev(dev, port_dev, iter)
1715 vrf_del_slave(dev, port_dev);
1716
1717 vrf_map_unregister_dev(dev);
1718
1719 unregister_netdevice_queue(dev, head);
1720}
1721
1722static int vrf_newlink(struct net *src_net, struct net_device *dev,
1723 struct nlattr *tb[], struct nlattr *data[],
1724 struct netlink_ext_ack *extack)
1725{
1726 struct net_vrf *vrf = netdev_priv(dev);
1727 struct netns_vrf *nn_vrf;
1728 bool *add_fib_rules;
1729 struct net *net;
1730 int err;
1731
1732 if (!data || !data[IFLA_VRF_TABLE]) {
1733 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1734 return -EINVAL;
1735 }
1736
1737 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1738 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1739 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1740 "Invalid VRF table id");
1741 return -EINVAL;
1742 }
1743
1744 dev->priv_flags |= IFF_L3MDEV_MASTER;
1745
1746 err = register_netdevice(dev);
1747 if (err)
1748 goto out;
1749
1750 /* mapping between table_id and vrf;
1751 * note: such binding could not be done in the dev init function
1752 * because dev->ifindex id is not available yet.
1753 */
1754 vrf->ifindex = dev->ifindex;
1755
1756 err = vrf_map_register_dev(dev, extack);
1757 if (err) {
1758 unregister_netdevice(dev);
1759 goto out;
1760 }
1761
1762 net = dev_net(dev);
1763 nn_vrf = net_generic(net, vrf_net_id);
1764
1765 add_fib_rules = &nn_vrf->add_fib_rules;
1766 if (*add_fib_rules) {
1767 err = vrf_add_fib_rules(dev);
1768 if (err) {
1769 vrf_map_unregister_dev(dev);
1770 unregister_netdevice(dev);
1771 goto out;
1772 }
1773 *add_fib_rules = false;
1774 }
1775
1776out:
1777 return err;
1778}
1779
1780static size_t vrf_nl_getsize(const struct net_device *dev)
1781{
1782 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1783}
1784
1785static int vrf_fillinfo(struct sk_buff *skb,
1786 const struct net_device *dev)
1787{
1788 struct net_vrf *vrf = netdev_priv(dev);
1789
1790 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1791}
1792
1793static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1794 const struct net_device *slave_dev)
1795{
1796 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1797}
1798
1799static int vrf_fill_slave_info(struct sk_buff *skb,
1800 const struct net_device *vrf_dev,
1801 const struct net_device *slave_dev)
1802{
1803 struct net_vrf *vrf = netdev_priv(vrf_dev);
1804
1805 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1806 return -EMSGSIZE;
1807
1808 return 0;
1809}
1810
1811static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1812 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1813};
1814
1815static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1816 .kind = DRV_NAME,
1817 .priv_size = sizeof(struct net_vrf),
1818
1819 .get_size = vrf_nl_getsize,
1820 .policy = vrf_nl_policy,
1821 .validate = vrf_validate,
1822 .fill_info = vrf_fillinfo,
1823
1824 .get_slave_size = vrf_get_slave_size,
1825 .fill_slave_info = vrf_fill_slave_info,
1826
1827 .newlink = vrf_newlink,
1828 .dellink = vrf_dellink,
1829 .setup = vrf_setup,
1830 .maxtype = IFLA_VRF_MAX,
1831};
1832
1833static int vrf_device_event(struct notifier_block *unused,
1834 unsigned long event, void *ptr)
1835{
1836 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1837
1838 /* only care about unregister events to drop slave references */
1839 if (event == NETDEV_UNREGISTER) {
1840 struct net_device *vrf_dev;
1841
1842 if (!netif_is_l3_slave(dev))
1843 goto out;
1844
1845 vrf_dev = netdev_master_upper_dev_get(dev);
1846 vrf_del_slave(vrf_dev, dev);
1847 }
1848out:
1849 return NOTIFY_DONE;
1850}
1851
1852static struct notifier_block vrf_notifier_block __read_mostly = {
1853 .notifier_call = vrf_device_event,
1854};
1855
1856static int vrf_map_init(struct vrf_map *vmap)
1857{
1858 spin_lock_init(&vmap->vmap_lock);
1859 hash_init(vmap->ht);
1860
1861 vmap->strict_mode = false;
1862
1863 return 0;
1864}
1865
1866#ifdef CONFIG_SYSCTL
1867static bool vrf_strict_mode(struct vrf_map *vmap)
1868{
1869 bool strict_mode;
1870
1871 vrf_map_lock(vmap);
1872 strict_mode = vmap->strict_mode;
1873 vrf_map_unlock(vmap);
1874
1875 return strict_mode;
1876}
1877
1878static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1879{
1880 bool *cur_mode;
1881 int res = 0;
1882
1883 vrf_map_lock(vmap);
1884
1885 cur_mode = &vmap->strict_mode;
1886 if (*cur_mode == new_mode)
1887 goto unlock;
1888
1889 if (*cur_mode) {
1890 /* disable strict mode */
1891 *cur_mode = false;
1892 } else {
1893 if (vmap->shared_tables) {
1894 /* we cannot allow strict_mode because there are some
1895 * vrfs that share one or more tables.
1896 */
1897 res = -EBUSY;
1898 goto unlock;
1899 }
1900
1901 /* no tables are shared among vrfs, so we can go back
1902 * to 1:1 association between a vrf with its table.
1903 */
1904 *cur_mode = true;
1905 }
1906
1907unlock:
1908 vrf_map_unlock(vmap);
1909
1910 return res;
1911}
1912
1913static int vrf_shared_table_handler(struct ctl_table *table, int write,
1914 void *buffer, size_t *lenp, loff_t *ppos)
1915{
1916 struct net *net = (struct net *)table->extra1;
1917 struct vrf_map *vmap = netns_vrf_map(net);
1918 int proc_strict_mode = 0;
1919 struct ctl_table tmp = {
1920 .procname = table->procname,
1921 .data = &proc_strict_mode,
1922 .maxlen = sizeof(int),
1923 .mode = table->mode,
1924 .extra1 = SYSCTL_ZERO,
1925 .extra2 = SYSCTL_ONE,
1926 };
1927 int ret;
1928
1929 if (!write)
1930 proc_strict_mode = vrf_strict_mode(vmap);
1931
1932 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1933
1934 if (write && ret == 0)
1935 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1936
1937 return ret;
1938}
1939
1940static const struct ctl_table vrf_table[] = {
1941 {
1942 .procname = "strict_mode",
1943 .data = NULL,
1944 .maxlen = sizeof(int),
1945 .mode = 0644,
1946 .proc_handler = vrf_shared_table_handler,
1947 /* set by the vrf_netns_init */
1948 .extra1 = NULL,
1949 },
1950 { },
1951};
1952
1953static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1954{
1955 struct ctl_table *table;
1956
1957 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1958 if (!table)
1959 return -ENOMEM;
1960
1961 /* init the extra1 parameter with the reference to current netns */
1962 table[0].extra1 = net;
1963
1964 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1965 if (!nn_vrf->ctl_hdr) {
1966 kfree(table);
1967 return -ENOMEM;
1968 }
1969
1970 return 0;
1971}
1972
1973static void vrf_netns_exit_sysctl(struct net *net)
1974{
1975 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1976 struct ctl_table *table;
1977
1978 table = nn_vrf->ctl_hdr->ctl_table_arg;
1979 unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1980 kfree(table);
1981}
1982#else
1983static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1984{
1985 return 0;
1986}
1987
1988static void vrf_netns_exit_sysctl(struct net *net)
1989{
1990}
1991#endif
1992
1993/* Initialize per network namespace state */
1994static int __net_init vrf_netns_init(struct net *net)
1995{
1996 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1997
1998 nn_vrf->add_fib_rules = true;
1999 vrf_map_init(&nn_vrf->vmap);
2000
2001 return vrf_netns_init_sysctl(net, nn_vrf);
2002}
2003
2004static void __net_exit vrf_netns_exit(struct net *net)
2005{
2006 vrf_netns_exit_sysctl(net);
2007}
2008
2009static struct pernet_operations vrf_net_ops __net_initdata = {
2010 .init = vrf_netns_init,
2011 .exit = vrf_netns_exit,
2012 .id = &vrf_net_id,
2013 .size = sizeof(struct netns_vrf),
2014};
2015
2016static int __init vrf_init_module(void)
2017{
2018 int rc;
2019
2020 register_netdevice_notifier(&vrf_notifier_block);
2021
2022 rc = register_pernet_subsys(&vrf_net_ops);
2023 if (rc < 0)
2024 goto error;
2025
2026 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
2027 vrf_ifindex_lookup_by_table_id);
2028 if (rc < 0)
2029 goto unreg_pernet;
2030
2031 rc = rtnl_link_register(&vrf_link_ops);
2032 if (rc < 0)
2033 goto table_lookup_unreg;
2034
2035 return 0;
2036
2037table_lookup_unreg:
2038 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
2039 vrf_ifindex_lookup_by_table_id);
2040
2041unreg_pernet:
2042 unregister_pernet_subsys(&vrf_net_ops);
2043
2044error:
2045 unregister_netdevice_notifier(&vrf_notifier_block);
2046 return rc;
2047}
2048
2049module_init(vrf_init_module);
2050MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
2051MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
2052MODULE_LICENSE("GPL");
2053MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2054MODULE_VERSION(DRV_VERSION);