2 * vrf.c: device driver to encapsulate a VRF space
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>
8 * Based on dummy, team and ipvlan drivers
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.
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/netdevice.h>
19 #include <linux/etherdevice.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>
29 #include <linux/inetdevice.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>
40 #define DRV_NAME "vrf"
41 #define DRV_VERSION "1.0"
43 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
44 static bool add_fib_rules = true;
47 struct rtable __rcu *rth;
48 struct rtable __rcu *rth_local;
49 struct rt6_info __rcu *rt6;
50 struct rt6_info __rcu *rt6_local;
61 struct u64_stats_sync syncp;
64 static void vrf_rx_stats(struct net_device *dev, int len)
66 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
68 u64_stats_update_begin(&dstats->syncp);
70 dstats->rx_bytes += len;
71 u64_stats_update_end(&dstats->syncp);
74 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
76 vrf_dev->stats.tx_errors++;
80 static void vrf_get_stats64(struct net_device *dev,
81 struct rtnl_link_stats64 *stats)
85 for_each_possible_cpu(i) {
86 const struct pcpu_dstats *dstats;
87 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
90 dstats = per_cpu_ptr(dev->dstats, i);
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;
107 /* by default VRF devices do not have a qdisc and are expected
108 * to be created with only a single queue.
110 static bool qdisc_tx_is_default(const struct net_device *dev)
112 struct netdev_queue *txq;
115 if (dev->num_tx_queues > 1)
118 txq = netdev_get_tx_queue(dev, 0);
119 qdisc = rcu_access_pointer(txq->qdisc);
121 return !qdisc->enqueue;
124 /* Local traffic destined to local address. Reinsert the packet to rx
125 * path, similar to loopback handling.
127 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
128 struct dst_entry *dst)
134 skb_dst_set(skb, dst);
137 /* set pkt_type to avoid skb hitting packet taps twice -
138 * once on Tx and again in Rx processing
140 skb->pkt_type = PACKET_LOOPBACK;
142 skb->protocol = eth_type_trans(skb, dev);
144 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
145 vrf_rx_stats(dev, len);
147 this_cpu_inc(dev->dstats->rx_drps);
152 #if IS_ENABLED(CONFIG_IPV6)
153 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
158 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
159 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
161 if (likely(err == 1))
162 err = dst_output(net, sk, skb);
167 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
168 struct net_device *dev)
170 const struct ipv6hdr *iph = ipv6_hdr(skb);
171 struct net *net = dev_net(skb->dev);
172 struct flowi6 fl6 = {
173 /* needed to match OIF rule */
174 .flowi6_oif = dev->ifindex,
175 .flowi6_iif = LOOPBACK_IFINDEX,
178 .flowlabel = ip6_flowinfo(iph),
179 .flowi6_mark = skb->mark,
180 .flowi6_proto = iph->nexthdr,
181 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
183 int ret = NET_XMIT_DROP;
184 struct dst_entry *dst;
185 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
187 dst = ip6_route_output(net, NULL, &fl6);
193 /* if dst.dev is loopback or the VRF device again this is locally
194 * originated traffic destined to a local address. Short circuit
195 * to Rx path using our local dst
197 if (dst->dev == net->loopback_dev || dst->dev == dev) {
198 struct net_vrf *vrf = netdev_priv(dev);
199 struct rt6_info *rt6_local;
201 /* release looked up dst and use cached local dst */
206 rt6_local = rcu_dereference(vrf->rt6_local);
207 if (unlikely(!rt6_local)) {
212 /* Ordering issue: cached local dst is created on newlink
213 * before the IPv6 initialization. Using the local dst
214 * requires rt6i_idev to be set so make sure it is.
216 if (unlikely(!rt6_local->rt6i_idev)) {
217 rt6_local->rt6i_idev = in6_dev_get(dev);
218 if (!rt6_local->rt6i_idev) {
224 dst = &rt6_local->dst;
229 return vrf_local_xmit(skb, dev, &rt6_local->dst);
232 skb_dst_set(skb, dst);
234 /* strip the ethernet header added for pass through VRF device */
235 __skb_pull(skb, skb_network_offset(skb));
237 ret = vrf_ip6_local_out(net, skb->sk, skb);
238 if (unlikely(net_xmit_eval(ret)))
239 dev->stats.tx_errors++;
241 ret = NET_XMIT_SUCCESS;
245 vrf_tx_error(dev, skb);
246 return NET_XMIT_DROP;
249 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
250 struct net_device *dev)
252 vrf_tx_error(dev, skb);
253 return NET_XMIT_DROP;
257 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
258 static int vrf_ip_local_out(struct net *net, struct sock *sk,
263 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
264 skb, NULL, skb_dst(skb)->dev, dst_output);
265 if (likely(err == 1))
266 err = dst_output(net, sk, skb);
271 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
272 struct net_device *vrf_dev)
274 struct iphdr *ip4h = ip_hdr(skb);
275 int ret = NET_XMIT_DROP;
276 struct flowi4 fl4 = {
277 /* needed to match OIF rule */
278 .flowi4_oif = vrf_dev->ifindex,
279 .flowi4_iif = LOOPBACK_IFINDEX,
280 .flowi4_tos = RT_TOS(ip4h->tos),
281 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
282 .flowi4_proto = ip4h->protocol,
283 .daddr = ip4h->daddr,
284 .saddr = ip4h->saddr,
286 struct net *net = dev_net(vrf_dev);
289 rt = ip_route_output_flow(net, &fl4, NULL);
295 /* if dst.dev is loopback or the VRF device again this is locally
296 * originated traffic destined to a local address. Short circuit
297 * to Rx path using our local dst
299 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
300 struct net_vrf *vrf = netdev_priv(vrf_dev);
301 struct rtable *rth_local;
302 struct dst_entry *dst = NULL;
308 rth_local = rcu_dereference(vrf->rth_local);
309 if (likely(rth_local)) {
310 dst = &rth_local->dst;
319 return vrf_local_xmit(skb, vrf_dev, dst);
322 skb_dst_set(skb, &rt->dst);
324 /* strip the ethernet header added for pass through VRF device */
325 __skb_pull(skb, skb_network_offset(skb));
328 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
332 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
333 if (unlikely(net_xmit_eval(ret)))
334 vrf_dev->stats.tx_errors++;
336 ret = NET_XMIT_SUCCESS;
341 vrf_tx_error(vrf_dev, skb);
345 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
347 switch (skb->protocol) {
348 case htons(ETH_P_IP):
349 return vrf_process_v4_outbound(skb, dev);
350 case htons(ETH_P_IPV6):
351 return vrf_process_v6_outbound(skb, dev);
353 vrf_tx_error(dev, skb);
354 return NET_XMIT_DROP;
358 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
361 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
363 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
364 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
366 u64_stats_update_begin(&dstats->syncp);
368 dstats->tx_bytes += len;
369 u64_stats_update_end(&dstats->syncp);
371 this_cpu_inc(dev->dstats->tx_drps);
377 static int vrf_finish_direct(struct net *net, struct sock *sk,
380 struct net_device *vrf_dev = skb->dev;
382 if (!list_empty(&vrf_dev->ptype_all) &&
383 likely(skb_headroom(skb) >= ETH_HLEN)) {
384 struct ethhdr *eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
386 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
387 eth_zero_addr(eth->h_dest);
388 eth->h_proto = skb->protocol;
391 dev_queue_xmit_nit(skb, vrf_dev);
392 rcu_read_unlock_bh();
394 skb_pull(skb, ETH_HLEN);
400 #if IS_ENABLED(CONFIG_IPV6)
401 /* modelled after ip6_finish_output2 */
402 static int vrf_finish_output6(struct net *net, struct sock *sk,
405 struct dst_entry *dst = skb_dst(skb);
406 struct net_device *dev = dst->dev;
407 struct neighbour *neigh;
408 struct in6_addr *nexthop;
413 skb->protocol = htons(ETH_P_IPV6);
417 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
418 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
419 if (unlikely(!neigh))
420 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
421 if (!IS_ERR(neigh)) {
422 sock_confirm_neigh(skb, neigh);
423 ret = neigh_output(neigh, skb);
424 rcu_read_unlock_bh();
427 rcu_read_unlock_bh();
429 IP6_INC_STATS(dev_net(dst->dev),
430 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
435 /* modelled after ip6_output */
436 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
438 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
439 net, sk, skb, NULL, skb_dst(skb)->dev,
441 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
444 /* set dst on skb to send packet to us via dev_xmit path. Allows
445 * packet to go through device based features such as qdisc, netfilter
446 * hooks and packet sockets with skb->dev set to vrf device.
448 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
451 struct net_vrf *vrf = netdev_priv(vrf_dev);
452 struct dst_entry *dst = NULL;
453 struct rt6_info *rt6;
457 rt6 = rcu_dereference(vrf->rt6);
465 if (unlikely(!dst)) {
466 vrf_tx_error(vrf_dev, skb);
471 skb_dst_set(skb, dst);
476 static int vrf_output6_direct(struct net *net, struct sock *sk,
479 skb->protocol = htons(ETH_P_IPV6);
481 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
482 net, sk, skb, NULL, skb->dev,
484 !(IPCB(skb)->flags & IPSKB_REROUTED));
487 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
491 struct net *net = dev_net(vrf_dev);
496 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
497 skb, NULL, vrf_dev, vrf_output6_direct);
499 if (likely(err == 1))
500 err = vrf_output6_direct(net, sk, skb);
502 /* reset skb device */
503 if (likely(err == 1))
511 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
515 /* don't divert link scope packets */
516 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
519 if (qdisc_tx_is_default(vrf_dev))
520 return vrf_ip6_out_direct(vrf_dev, sk, skb);
522 return vrf_ip6_out_redirect(vrf_dev, skb);
526 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
528 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
529 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
530 struct net *net = dev_net(dev);
531 struct dst_entry *dst;
533 RCU_INIT_POINTER(vrf->rt6, NULL);
534 RCU_INIT_POINTER(vrf->rt6_local, NULL);
537 /* move dev in dst's to loopback so this VRF device can be deleted
538 * - based on dst_ifdown
543 dst->dev = net->loopback_dev;
549 if (rt6_local->rt6i_idev) {
550 in6_dev_put(rt6_local->rt6i_idev);
551 rt6_local->rt6i_idev = NULL;
554 dst = &rt6_local->dst;
556 dst->dev = net->loopback_dev;
562 static int vrf_rt6_create(struct net_device *dev)
564 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
565 struct net_vrf *vrf = netdev_priv(dev);
566 struct net *net = dev_net(dev);
567 struct fib6_table *rt6i_table;
568 struct rt6_info *rt6, *rt6_local;
571 /* IPv6 can be CONFIG enabled and then disabled runtime */
572 if (!ipv6_mod_enabled())
575 rt6i_table = fib6_new_table(net, vrf->tb_id);
579 /* create a dst for routing packets out a VRF device */
580 rt6 = ip6_dst_alloc(net, dev, flags);
586 rt6->rt6i_table = rt6i_table;
587 rt6->dst.output = vrf_output6;
589 /* create a dst for local routing - packets sent locally
590 * to local address via the VRF device as a loopback
592 rt6_local = ip6_dst_alloc(net, dev, flags);
594 dst_release(&rt6->dst);
598 dst_hold(&rt6_local->dst);
600 rt6_local->rt6i_idev = in6_dev_get(dev);
601 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
602 rt6_local->rt6i_table = rt6i_table;
603 rt6_local->dst.input = ip6_input;
605 rcu_assign_pointer(vrf->rt6, rt6);
606 rcu_assign_pointer(vrf->rt6_local, rt6_local);
613 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
620 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
624 static int vrf_rt6_create(struct net_device *dev)
630 /* modelled after ip_finish_output2 */
631 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
633 struct dst_entry *dst = skb_dst(skb);
634 struct rtable *rt = (struct rtable *)dst;
635 struct net_device *dev = dst->dev;
636 unsigned int hh_len = LL_RESERVED_SPACE(dev);
637 struct neighbour *neigh;
643 /* Be paranoid, rather than too clever. */
644 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
645 struct sk_buff *skb2;
647 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
653 skb_set_owner_w(skb2, skb->sk);
661 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
662 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
663 if (unlikely(!neigh))
664 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
665 if (!IS_ERR(neigh)) {
666 sock_confirm_neigh(skb, neigh);
667 ret = neigh_output(neigh, skb);
670 rcu_read_unlock_bh();
672 if (unlikely(ret < 0))
673 vrf_tx_error(skb->dev, skb);
677 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
679 struct net_device *dev = skb_dst(skb)->dev;
681 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
684 skb->protocol = htons(ETH_P_IP);
686 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
687 net, sk, skb, NULL, dev,
689 !(IPCB(skb)->flags & IPSKB_REROUTED));
692 /* set dst on skb to send packet to us via dev_xmit path. Allows
693 * packet to go through device based features such as qdisc, netfilter
694 * hooks and packet sockets with skb->dev set to vrf device.
696 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
699 struct net_vrf *vrf = netdev_priv(vrf_dev);
700 struct dst_entry *dst = NULL;
705 rth = rcu_dereference(vrf->rth);
713 if (unlikely(!dst)) {
714 vrf_tx_error(vrf_dev, skb);
719 skb_dst_set(skb, dst);
724 static int vrf_output_direct(struct net *net, struct sock *sk,
727 skb->protocol = htons(ETH_P_IP);
729 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
730 net, sk, skb, NULL, skb->dev,
732 !(IPCB(skb)->flags & IPSKB_REROUTED));
735 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
739 struct net *net = dev_net(vrf_dev);
744 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
745 skb, NULL, vrf_dev, vrf_output_direct);
747 if (likely(err == 1))
748 err = vrf_output_direct(net, sk, skb);
750 /* reset skb device */
751 if (likely(err == 1))
759 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
763 /* don't divert multicast */
764 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
767 if (qdisc_tx_is_default(vrf_dev))
768 return vrf_ip_out_direct(vrf_dev, sk, skb);
770 return vrf_ip_out_redirect(vrf_dev, skb);
773 /* called with rcu lock held */
774 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
781 return vrf_ip_out(vrf_dev, sk, skb);
783 return vrf_ip6_out(vrf_dev, sk, skb);
790 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
792 struct rtable *rth = rtnl_dereference(vrf->rth);
793 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
794 struct net *net = dev_net(dev);
795 struct dst_entry *dst;
797 RCU_INIT_POINTER(vrf->rth, NULL);
798 RCU_INIT_POINTER(vrf->rth_local, NULL);
801 /* move dev in dst's to loopback so this VRF device can be deleted
802 * - based on dst_ifdown
807 dst->dev = net->loopback_dev;
813 dst = &rth_local->dst;
815 dst->dev = net->loopback_dev;
821 static int vrf_rtable_create(struct net_device *dev)
823 struct net_vrf *vrf = netdev_priv(dev);
824 struct rtable *rth, *rth_local;
826 if (!fib_new_table(dev_net(dev), vrf->tb_id))
829 /* create a dst for routing packets out through a VRF device */
830 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
834 /* create a dst for local ingress routing - packets sent locally
835 * to local address via the VRF device as a loopback
837 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
839 dst_release(&rth->dst);
843 rth->dst.output = vrf_output;
844 rth->rt_table_id = vrf->tb_id;
846 rth_local->rt_table_id = vrf->tb_id;
848 rcu_assign_pointer(vrf->rth, rth);
849 rcu_assign_pointer(vrf->rth_local, rth_local);
854 /**************************** device handling ********************/
856 /* cycle interface to flush neighbor cache and move routes across tables */
857 static void cycle_netdev(struct net_device *dev)
859 unsigned int flags = dev->flags;
862 if (!netif_running(dev))
865 ret = dev_change_flags(dev, flags & ~IFF_UP);
867 ret = dev_change_flags(dev, flags);
871 "Failed to cycle device %s; route tables might be wrong!\n",
876 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
880 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
881 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
885 cycle_netdev(port_dev);
890 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
894 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
896 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
899 return do_vrf_add_slave(dev, port_dev);
902 /* inverse of do_vrf_add_slave */
903 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
905 netdev_upper_dev_unlink(port_dev, dev);
906 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
908 cycle_netdev(port_dev);
913 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
915 return do_vrf_del_slave(dev, port_dev);
918 static void vrf_dev_uninit(struct net_device *dev)
920 struct net_vrf *vrf = netdev_priv(dev);
921 struct net_device *port_dev;
922 struct list_head *iter;
924 vrf_rtable_release(dev, vrf);
925 vrf_rt6_release(dev, vrf);
927 netdev_for_each_lower_dev(dev, port_dev, iter)
928 vrf_del_slave(dev, port_dev);
930 free_percpu(dev->dstats);
934 static int vrf_dev_init(struct net_device *dev)
936 struct net_vrf *vrf = netdev_priv(dev);
938 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
942 /* create the default dst which points back to us */
943 if (vrf_rtable_create(dev) != 0)
946 if (vrf_rt6_create(dev) != 0)
949 dev->flags = IFF_MASTER | IFF_NOARP;
951 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
952 dev->mtu = 64 * 1024;
954 /* similarly, oper state is irrelevant; set to up to avoid confusion */
955 dev->operstate = IF_OPER_UP;
956 netdev_lockdep_set_classes(dev);
960 vrf_rtable_release(dev, vrf);
962 free_percpu(dev->dstats);
968 static const struct net_device_ops vrf_netdev_ops = {
969 .ndo_init = vrf_dev_init,
970 .ndo_uninit = vrf_dev_uninit,
971 .ndo_start_xmit = vrf_xmit,
972 .ndo_get_stats64 = vrf_get_stats64,
973 .ndo_add_slave = vrf_add_slave,
974 .ndo_del_slave = vrf_del_slave,
977 static u32 vrf_fib_table(const struct net_device *dev)
979 struct net_vrf *vrf = netdev_priv(dev);
984 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
989 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
991 struct net_device *dev)
993 struct net *net = dev_net(dev);
995 if (NF_HOOK(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) < 0)
996 skb = NULL; /* kfree_skb(skb) handled by nf code */
1001 #if IS_ENABLED(CONFIG_IPV6)
1002 /* neighbor handling is done with actual device; do not want
1003 * to flip skb->dev for those ndisc packets. This really fails
1004 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1007 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1009 const struct ipv6hdr *iph = ipv6_hdr(skb);
1012 if (iph->nexthdr == NEXTHDR_ICMP) {
1013 const struct icmp6hdr *icmph;
1014 struct icmp6hdr _icmph;
1016 icmph = skb_header_pointer(skb, sizeof(*iph),
1017 sizeof(_icmph), &_icmph);
1021 switch (icmph->icmp6_type) {
1022 case NDISC_ROUTER_SOLICITATION:
1023 case NDISC_ROUTER_ADVERTISEMENT:
1024 case NDISC_NEIGHBOUR_SOLICITATION:
1025 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1026 case NDISC_REDIRECT:
1036 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1037 const struct net_device *dev,
1042 struct net_vrf *vrf = netdev_priv(dev);
1043 struct fib6_table *table = NULL;
1044 struct rt6_info *rt6;
1048 /* fib6_table does not have a refcnt and can not be freed */
1049 rt6 = rcu_dereference(vrf->rt6);
1051 table = rt6->rt6i_table;
1058 return ip6_pol_route(net, table, ifindex, fl6, flags);
1061 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1064 const struct ipv6hdr *iph = ipv6_hdr(skb);
1065 struct flowi6 fl6 = {
1066 .daddr = iph->daddr,
1067 .saddr = iph->saddr,
1068 .flowlabel = ip6_flowinfo(iph),
1069 .flowi6_mark = skb->mark,
1070 .flowi6_proto = iph->nexthdr,
1071 .flowi6_iif = ifindex,
1073 struct net *net = dev_net(vrf_dev);
1074 struct rt6_info *rt6;
1076 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
1077 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1081 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1084 skb_dst_set(skb, &rt6->dst);
1087 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1088 struct sk_buff *skb)
1090 int orig_iif = skb->skb_iif;
1093 /* loopback traffic; do not push through packet taps again.
1094 * Reset pkt_type for upper layers to process skb
1096 if (skb->pkt_type == PACKET_LOOPBACK) {
1098 skb->skb_iif = vrf_dev->ifindex;
1099 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1100 skb->pkt_type = PACKET_HOST;
1104 /* if packet is NDISC or addressed to multicast or link-local
1105 * then keep the ingress interface
1107 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1108 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1109 vrf_rx_stats(vrf_dev, skb->len);
1111 skb->skb_iif = vrf_dev->ifindex;
1113 if (!list_empty(&vrf_dev->ptype_all)) {
1114 skb_push(skb, skb->mac_len);
1115 dev_queue_xmit_nit(skb, vrf_dev);
1116 skb_pull(skb, skb->mac_len);
1119 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1123 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1125 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1131 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1132 struct sk_buff *skb)
1138 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1139 struct sk_buff *skb)
1142 skb->skb_iif = vrf_dev->ifindex;
1143 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1145 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1148 /* loopback traffic; do not push through packet taps again.
1149 * Reset pkt_type for upper layers to process skb
1151 if (skb->pkt_type == PACKET_LOOPBACK) {
1152 skb->pkt_type = PACKET_HOST;
1156 vrf_rx_stats(vrf_dev, skb->len);
1158 if (!list_empty(&vrf_dev->ptype_all)) {
1159 skb_push(skb, skb->mac_len);
1160 dev_queue_xmit_nit(skb, vrf_dev);
1161 skb_pull(skb, skb->mac_len);
1164 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1169 /* called with rcu lock held */
1170 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1171 struct sk_buff *skb,
1176 return vrf_ip_rcv(vrf_dev, skb);
1178 return vrf_ip6_rcv(vrf_dev, skb);
1184 #if IS_ENABLED(CONFIG_IPV6)
1185 /* send to link-local or multicast address via interface enslaved to
1186 * VRF device. Force lookup to VRF table without changing flow struct
1188 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1191 struct net *net = dev_net(dev);
1192 int flags = RT6_LOOKUP_F_IFACE;
1193 struct dst_entry *dst = NULL;
1194 struct rt6_info *rt;
1196 /* VRF device does not have a link-local address and
1197 * sending packets to link-local or mcast addresses over
1198 * a VRF device does not make sense
1200 if (fl6->flowi6_oif == dev->ifindex) {
1201 dst = &net->ipv6.ip6_null_entry->dst;
1206 if (!ipv6_addr_any(&fl6->saddr))
1207 flags |= RT6_LOOKUP_F_HAS_SADDR;
1209 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1217 static const struct l3mdev_ops vrf_l3mdev_ops = {
1218 .l3mdev_fib_table = vrf_fib_table,
1219 .l3mdev_l3_rcv = vrf_l3_rcv,
1220 .l3mdev_l3_out = vrf_l3_out,
1221 #if IS_ENABLED(CONFIG_IPV6)
1222 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1226 static void vrf_get_drvinfo(struct net_device *dev,
1227 struct ethtool_drvinfo *info)
1229 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1230 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1233 static const struct ethtool_ops vrf_ethtool_ops = {
1234 .get_drvinfo = vrf_get_drvinfo,
1237 static inline size_t vrf_fib_rule_nl_size(void)
1241 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1242 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1243 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1248 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1250 struct fib_rule_hdr *frh;
1251 struct nlmsghdr *nlh;
1252 struct sk_buff *skb;
1255 if (family == AF_INET6 && !ipv6_mod_enabled())
1258 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1262 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1264 goto nla_put_failure;
1266 /* rule only needs to appear once */
1267 nlh->nlmsg_flags &= NLM_F_EXCL;
1269 frh = nlmsg_data(nlh);
1270 memset(frh, 0, sizeof(*frh));
1271 frh->family = family;
1272 frh->action = FR_ACT_TO_TBL;
1274 if (nla_put_u32(skb, FRA_L3MDEV, 1))
1275 goto nla_put_failure;
1277 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1278 goto nla_put_failure;
1280 nlmsg_end(skb, nlh);
1282 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1283 skb->sk = dev_net(dev)->rtnl;
1285 err = fib_nl_newrule(skb, nlh, NULL);
1289 err = fib_nl_delrule(skb, nlh, NULL);
1303 static int vrf_add_fib_rules(const struct net_device *dev)
1307 err = vrf_fib_rule(dev, AF_INET, true);
1311 err = vrf_fib_rule(dev, AF_INET6, true);
1315 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1316 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1323 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1325 vrf_fib_rule(dev, AF_INET6, false);
1329 vrf_fib_rule(dev, AF_INET, false);
1332 netdev_err(dev, "Failed to add FIB rules.\n");
1336 static void vrf_setup(struct net_device *dev)
1340 /* Initialize the device structure. */
1341 dev->netdev_ops = &vrf_netdev_ops;
1342 dev->l3mdev_ops = &vrf_l3mdev_ops;
1343 dev->ethtool_ops = &vrf_ethtool_ops;
1344 dev->destructor = free_netdev;
1346 /* Fill in device structure with ethernet-generic values. */
1347 eth_hw_addr_random(dev);
1349 /* don't acquire vrf device's netif_tx_lock when transmitting */
1350 dev->features |= NETIF_F_LLTX;
1352 /* don't allow vrf devices to change network namespaces. */
1353 dev->features |= NETIF_F_NETNS_LOCAL;
1355 /* does not make sense for a VLAN to be added to a vrf device */
1356 dev->features |= NETIF_F_VLAN_CHALLENGED;
1358 /* enable offload features */
1359 dev->features |= NETIF_F_GSO_SOFTWARE;
1360 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1361 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1363 dev->hw_features = dev->features;
1364 dev->hw_enc_features = dev->features;
1366 /* default to no qdisc; user can add if desired */
1367 dev->priv_flags |= IFF_NO_QUEUE;
1370 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1372 if (tb[IFLA_ADDRESS]) {
1373 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1375 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1376 return -EADDRNOTAVAIL;
1381 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1383 unregister_netdevice_queue(dev, head);
1386 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1387 struct nlattr *tb[], struct nlattr *data[])
1389 struct net_vrf *vrf = netdev_priv(dev);
1392 if (!data || !data[IFLA_VRF_TABLE])
1395 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1396 if (vrf->tb_id == RT_TABLE_UNSPEC)
1399 dev->priv_flags |= IFF_L3MDEV_MASTER;
1401 err = register_netdevice(dev);
1405 if (add_fib_rules) {
1406 err = vrf_add_fib_rules(dev);
1408 unregister_netdevice(dev);
1411 add_fib_rules = false;
1418 static size_t vrf_nl_getsize(const struct net_device *dev)
1420 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1423 static int vrf_fillinfo(struct sk_buff *skb,
1424 const struct net_device *dev)
1426 struct net_vrf *vrf = netdev_priv(dev);
1428 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1431 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1432 const struct net_device *slave_dev)
1434 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1437 static int vrf_fill_slave_info(struct sk_buff *skb,
1438 const struct net_device *vrf_dev,
1439 const struct net_device *slave_dev)
1441 struct net_vrf *vrf = netdev_priv(vrf_dev);
1443 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1449 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1450 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1453 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1455 .priv_size = sizeof(struct net_vrf),
1457 .get_size = vrf_nl_getsize,
1458 .policy = vrf_nl_policy,
1459 .validate = vrf_validate,
1460 .fill_info = vrf_fillinfo,
1462 .get_slave_size = vrf_get_slave_size,
1463 .fill_slave_info = vrf_fill_slave_info,
1465 .newlink = vrf_newlink,
1466 .dellink = vrf_dellink,
1468 .maxtype = IFLA_VRF_MAX,
1471 static int vrf_device_event(struct notifier_block *unused,
1472 unsigned long event, void *ptr)
1474 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1476 /* only care about unregister events to drop slave references */
1477 if (event == NETDEV_UNREGISTER) {
1478 struct net_device *vrf_dev;
1480 if (!netif_is_l3_slave(dev))
1483 vrf_dev = netdev_master_upper_dev_get(dev);
1484 vrf_del_slave(vrf_dev, dev);
1490 static struct notifier_block vrf_notifier_block __read_mostly = {
1491 .notifier_call = vrf_device_event,
1494 static int __init vrf_init_module(void)
1498 register_netdevice_notifier(&vrf_notifier_block);
1500 rc = rtnl_link_register(&vrf_link_ops);
1507 unregister_netdevice_notifier(&vrf_notifier_block);
1511 module_init(vrf_init_module);
1512 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1513 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1514 MODULE_LICENSE("GPL");
1515 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1516 MODULE_VERSION(DRV_VERSION);