2 * vrf.c: device driver to encapsulate a VRF space
4 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
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>
39 #include <net/netns/generic.h>
41 #define DRV_NAME "vrf"
42 #define DRV_VERSION "1.0"
44 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
46 static unsigned int vrf_net_id;
49 struct rtable __rcu *rth;
50 struct rt6_info __rcu *rt6;
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);
136 /* set pkt_type to avoid skb hitting packet taps twice -
137 * once on Tx and again in Rx processing
139 skb->pkt_type = PACKET_LOOPBACK;
141 skb->protocol = eth_type_trans(skb, dev);
143 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
144 vrf_rx_stats(dev, len);
146 this_cpu_inc(dev->dstats->rx_drps);
151 #if IS_ENABLED(CONFIG_IPV6)
152 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
157 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
158 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
160 if (likely(err == 1))
161 err = dst_output(net, sk, skb);
166 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
167 struct net_device *dev)
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,
177 .flowlabel = ip6_flowinfo(iph),
178 .flowi6_mark = skb->mark,
179 .flowi6_proto = iph->nexthdr,
180 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
182 int ret = NET_XMIT_DROP;
183 struct dst_entry *dst;
184 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
186 dst = ip6_route_output(net, NULL, &fl6);
192 /* if dst.dev is loopback or the VRF device again this is locally
193 * originated traffic destined to a local address. Short circuit
197 return vrf_local_xmit(skb, dev, dst);
199 skb_dst_set(skb, dst);
201 /* strip the ethernet header added for pass through VRF device */
202 __skb_pull(skb, skb_network_offset(skb));
204 ret = vrf_ip6_local_out(net, skb->sk, skb);
205 if (unlikely(net_xmit_eval(ret)))
206 dev->stats.tx_errors++;
208 ret = NET_XMIT_SUCCESS;
212 vrf_tx_error(dev, skb);
213 return NET_XMIT_DROP;
216 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
217 struct net_device *dev)
219 vrf_tx_error(dev, skb);
220 return NET_XMIT_DROP;
224 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
225 static int vrf_ip_local_out(struct net *net, struct sock *sk,
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);
238 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
239 struct net_device *vrf_dev)
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,
253 struct net *net = dev_net(vrf_dev);
256 rt = ip_route_output_flow(net, &fl4, NULL);
262 /* if dst.dev is loopback or the VRF device again this is locally
263 * originated traffic destined to a local address. Short circuit
266 if (rt->dst.dev == vrf_dev)
267 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
269 skb_dst_set(skb, &rt->dst);
271 /* strip the ethernet header added for pass through VRF device */
272 __skb_pull(skb, skb_network_offset(skb));
275 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
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++;
283 ret = NET_XMIT_SUCCESS;
288 vrf_tx_error(vrf_dev, skb);
292 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
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);
300 vrf_tx_error(dev, skb);
301 return NET_XMIT_DROP;
305 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
308 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
310 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
311 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
313 u64_stats_update_begin(&dstats->syncp);
315 dstats->tx_bytes += len;
316 u64_stats_update_end(&dstats->syncp);
318 this_cpu_inc(dev->dstats->tx_drps);
324 static int vrf_finish_direct(struct net *net, struct sock *sk,
327 struct net_device *vrf_dev = skb->dev;
329 if (!list_empty(&vrf_dev->ptype_all) &&
330 likely(skb_headroom(skb) >= ETH_HLEN)) {
331 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
333 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
334 eth_zero_addr(eth->h_dest);
335 eth->h_proto = skb->protocol;
338 dev_queue_xmit_nit(skb, vrf_dev);
339 rcu_read_unlock_bh();
341 skb_pull(skb, ETH_HLEN);
347 #if IS_ENABLED(CONFIG_IPV6)
348 /* modelled after ip6_finish_output2 */
349 static int vrf_finish_output6(struct net *net, struct sock *sk,
352 struct dst_entry *dst = skb_dst(skb);
353 struct net_device *dev = dst->dev;
354 struct neighbour *neigh;
355 struct in6_addr *nexthop;
360 skb->protocol = htons(ETH_P_IPV6);
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();
374 rcu_read_unlock_bh();
376 IP6_INC_STATS(dev_net(dst->dev),
377 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
382 /* modelled after ip6_output */
383 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
385 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
386 net, sk, skb, NULL, skb_dst(skb)->dev,
388 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
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.
395 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
398 struct net_vrf *vrf = netdev_priv(vrf_dev);
399 struct dst_entry *dst = NULL;
400 struct rt6_info *rt6;
404 rt6 = rcu_dereference(vrf->rt6);
412 if (unlikely(!dst)) {
413 vrf_tx_error(vrf_dev, skb);
418 skb_dst_set(skb, dst);
423 static int vrf_output6_direct(struct net *net, struct sock *sk,
426 skb->protocol = htons(ETH_P_IPV6);
428 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
429 net, sk, skb, NULL, skb->dev,
431 !(IPCB(skb)->flags & IPSKB_REROUTED));
434 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
438 struct net *net = dev_net(vrf_dev);
443 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
444 skb, NULL, vrf_dev, vrf_output6_direct);
446 if (likely(err == 1))
447 err = vrf_output6_direct(net, sk, skb);
449 /* reset skb device */
450 if (likely(err == 1))
458 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
462 /* don't divert link scope packets */
463 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
466 if (qdisc_tx_is_default(vrf_dev))
467 return vrf_ip6_out_direct(vrf_dev, sk, skb);
469 return vrf_ip6_out_redirect(vrf_dev, skb);
473 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
475 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
476 struct net *net = dev_net(dev);
477 struct dst_entry *dst;
479 RCU_INIT_POINTER(vrf->rt6, NULL);
482 /* move dev in dst's to loopback so this VRF device can be deleted
483 * - based on dst_ifdown
488 dst->dev = net->loopback_dev;
494 static int vrf_rt6_create(struct net_device *dev)
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;
503 /* IPv6 can be CONFIG enabled and then disabled runtime */
504 if (!ipv6_mod_enabled())
507 rt6i_table = fib6_new_table(net, vrf->tb_id);
511 /* create a dst for routing packets out a VRF device */
512 rt6 = ip6_dst_alloc(net, dev, flags);
516 rt6->rt6i_table = rt6i_table;
517 rt6->dst.output = vrf_output6;
519 rcu_assign_pointer(vrf->rt6, rt6);
526 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
533 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
537 static int vrf_rt6_create(struct net_device *dev)
543 /* modelled after ip_finish_output2 */
544 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
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;
556 /* Be paranoid, rather than too clever. */
557 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
558 struct sk_buff *skb2;
560 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
566 skb_set_owner_w(skb2, skb->sk);
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);
583 rcu_read_unlock_bh();
585 if (unlikely(ret < 0))
586 vrf_tx_error(skb->dev, skb);
590 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
592 struct net_device *dev = skb_dst(skb)->dev;
594 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
597 skb->protocol = htons(ETH_P_IP);
599 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
600 net, sk, skb, NULL, dev,
602 !(IPCB(skb)->flags & IPSKB_REROUTED));
605 /* set dst on skb to send packet to us via dev_xmit path. Allows
606 * packet to go through device based features such as qdisc, netfilter
607 * hooks and packet sockets with skb->dev set to vrf device.
609 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
612 struct net_vrf *vrf = netdev_priv(vrf_dev);
613 struct dst_entry *dst = NULL;
618 rth = rcu_dereference(vrf->rth);
626 if (unlikely(!dst)) {
627 vrf_tx_error(vrf_dev, skb);
632 skb_dst_set(skb, dst);
637 static int vrf_output_direct(struct net *net, struct sock *sk,
640 skb->protocol = htons(ETH_P_IP);
642 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
643 net, sk, skb, NULL, skb->dev,
645 !(IPCB(skb)->flags & IPSKB_REROUTED));
648 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
652 struct net *net = dev_net(vrf_dev);
657 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
658 skb, NULL, vrf_dev, vrf_output_direct);
660 if (likely(err == 1))
661 err = vrf_output_direct(net, sk, skb);
663 /* reset skb device */
664 if (likely(err == 1))
672 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
676 /* don't divert multicast */
677 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
680 if (qdisc_tx_is_default(vrf_dev))
681 return vrf_ip_out_direct(vrf_dev, sk, skb);
683 return vrf_ip_out_redirect(vrf_dev, skb);
686 /* called with rcu lock held */
687 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
694 return vrf_ip_out(vrf_dev, sk, skb);
696 return vrf_ip6_out(vrf_dev, sk, skb);
703 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
705 struct rtable *rth = rtnl_dereference(vrf->rth);
706 struct net *net = dev_net(dev);
707 struct dst_entry *dst;
709 RCU_INIT_POINTER(vrf->rth, NULL);
712 /* move dev in dst's to loopback so this VRF device can be deleted
713 * - based on dst_ifdown
718 dst->dev = net->loopback_dev;
724 static int vrf_rtable_create(struct net_device *dev)
726 struct net_vrf *vrf = netdev_priv(dev);
729 if (!fib_new_table(dev_net(dev), vrf->tb_id))
732 /* create a dst for routing packets out through a VRF device */
733 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
737 rth->dst.output = vrf_output;
738 rth->rt_table_id = vrf->tb_id;
740 rcu_assign_pointer(vrf->rth, rth);
745 /**************************** device handling ********************/
747 /* cycle interface to flush neighbor cache and move routes across tables */
748 static void cycle_netdev(struct net_device *dev)
750 unsigned int flags = dev->flags;
753 if (!netif_running(dev))
756 ret = dev_change_flags(dev, flags & ~IFF_UP);
758 ret = dev_change_flags(dev, flags);
762 "Failed to cycle device %s; route tables might be wrong!\n",
767 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
768 struct netlink_ext_ack *extack)
772 /* do not allow loopback device to be enslaved to a VRF.
773 * The vrf device acts as the loopback for the vrf.
775 if (port_dev == dev_net(dev)->loopback_dev) {
776 NL_SET_ERR_MSG(extack,
777 "Can not enslave loopback device to a VRF");
781 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
782 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
786 cycle_netdev(port_dev);
791 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
795 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
796 struct netlink_ext_ack *extack)
798 if (netif_is_l3_master(port_dev)) {
799 NL_SET_ERR_MSG(extack,
800 "Can not enslave an L3 master device to a VRF");
804 if (netif_is_l3_slave(port_dev))
807 return do_vrf_add_slave(dev, port_dev, extack);
810 /* inverse of do_vrf_add_slave */
811 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
813 netdev_upper_dev_unlink(port_dev, dev);
814 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
816 cycle_netdev(port_dev);
821 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
823 return do_vrf_del_slave(dev, port_dev);
826 static void vrf_dev_uninit(struct net_device *dev)
828 struct net_vrf *vrf = netdev_priv(dev);
830 vrf_rtable_release(dev, vrf);
831 vrf_rt6_release(dev, vrf);
833 free_percpu(dev->dstats);
837 static int vrf_dev_init(struct net_device *dev)
839 struct net_vrf *vrf = netdev_priv(dev);
841 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
845 /* create the default dst which points back to us */
846 if (vrf_rtable_create(dev) != 0)
849 if (vrf_rt6_create(dev) != 0)
852 dev->flags = IFF_MASTER | IFF_NOARP;
854 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
855 dev->mtu = 64 * 1024;
857 /* similarly, oper state is irrelevant; set to up to avoid confusion */
858 dev->operstate = IF_OPER_UP;
859 netdev_lockdep_set_classes(dev);
863 vrf_rtable_release(dev, vrf);
865 free_percpu(dev->dstats);
871 static const struct net_device_ops vrf_netdev_ops = {
872 .ndo_init = vrf_dev_init,
873 .ndo_uninit = vrf_dev_uninit,
874 .ndo_start_xmit = vrf_xmit,
875 .ndo_get_stats64 = vrf_get_stats64,
876 .ndo_add_slave = vrf_add_slave,
877 .ndo_del_slave = vrf_del_slave,
880 static u32 vrf_fib_table(const struct net_device *dev)
882 struct net_vrf *vrf = netdev_priv(dev);
887 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
893 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
895 struct net_device *dev)
897 struct net *net = dev_net(dev);
899 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
900 skb = NULL; /* kfree_skb(skb) handled by nf code */
905 #if IS_ENABLED(CONFIG_IPV6)
906 /* neighbor handling is done with actual device; do not want
907 * to flip skb->dev for those ndisc packets. This really fails
908 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
911 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
913 const struct ipv6hdr *iph = ipv6_hdr(skb);
916 if (iph->nexthdr == NEXTHDR_ICMP) {
917 const struct icmp6hdr *icmph;
918 struct icmp6hdr _icmph;
920 icmph = skb_header_pointer(skb, sizeof(*iph),
921 sizeof(_icmph), &_icmph);
925 switch (icmph->icmp6_type) {
926 case NDISC_ROUTER_SOLICITATION:
927 case NDISC_ROUTER_ADVERTISEMENT:
928 case NDISC_NEIGHBOUR_SOLICITATION:
929 case NDISC_NEIGHBOUR_ADVERTISEMENT:
940 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
941 const struct net_device *dev,
946 struct net_vrf *vrf = netdev_priv(dev);
947 struct fib6_table *table = NULL;
948 struct rt6_info *rt6;
952 /* fib6_table does not have a refcnt and can not be freed */
953 rt6 = rcu_dereference(vrf->rt6);
955 table = rt6->rt6i_table;
962 return ip6_pol_route(net, table, ifindex, fl6, flags);
965 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
968 const struct ipv6hdr *iph = ipv6_hdr(skb);
969 struct flowi6 fl6 = {
970 .flowi6_iif = ifindex,
971 .flowi6_mark = skb->mark,
972 .flowi6_proto = iph->nexthdr,
975 .flowlabel = ip6_flowinfo(iph),
977 struct net *net = dev_net(vrf_dev);
978 struct rt6_info *rt6;
980 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
981 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
985 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
988 skb_dst_set(skb, &rt6->dst);
991 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
994 int orig_iif = skb->skb_iif;
997 /* loopback traffic; do not push through packet taps again.
998 * Reset pkt_type for upper layers to process skb
1000 if (skb->pkt_type == PACKET_LOOPBACK) {
1002 skb->skb_iif = vrf_dev->ifindex;
1003 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1004 skb->pkt_type = PACKET_HOST;
1008 /* if packet is NDISC or addressed to multicast or link-local
1009 * then keep the ingress interface
1011 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1012 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1013 vrf_rx_stats(vrf_dev, skb->len);
1015 skb->skb_iif = vrf_dev->ifindex;
1017 if (!list_empty(&vrf_dev->ptype_all)) {
1018 skb_push(skb, skb->mac_len);
1019 dev_queue_xmit_nit(skb, vrf_dev);
1020 skb_pull(skb, skb->mac_len);
1023 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1027 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1029 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1035 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1036 struct sk_buff *skb)
1042 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1043 struct sk_buff *skb)
1046 skb->skb_iif = vrf_dev->ifindex;
1047 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1049 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1052 /* loopback traffic; do not push through packet taps again.
1053 * Reset pkt_type for upper layers to process skb
1055 if (skb->pkt_type == PACKET_LOOPBACK) {
1056 skb->pkt_type = PACKET_HOST;
1060 vrf_rx_stats(vrf_dev, skb->len);
1062 if (!list_empty(&vrf_dev->ptype_all)) {
1063 skb_push(skb, skb->mac_len);
1064 dev_queue_xmit_nit(skb, vrf_dev);
1065 skb_pull(skb, skb->mac_len);
1068 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1073 /* called with rcu lock held */
1074 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1075 struct sk_buff *skb,
1080 return vrf_ip_rcv(vrf_dev, skb);
1082 return vrf_ip6_rcv(vrf_dev, skb);
1088 #if IS_ENABLED(CONFIG_IPV6)
1089 /* send to link-local or multicast address via interface enslaved to
1090 * VRF device. Force lookup to VRF table without changing flow struct
1092 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1095 struct net *net = dev_net(dev);
1096 int flags = RT6_LOOKUP_F_IFACE;
1097 struct dst_entry *dst = NULL;
1098 struct rt6_info *rt;
1100 /* VRF device does not have a link-local address and
1101 * sending packets to link-local or mcast addresses over
1102 * a VRF device does not make sense
1104 if (fl6->flowi6_oif == dev->ifindex) {
1105 dst = &net->ipv6.ip6_null_entry->dst;
1110 if (!ipv6_addr_any(&fl6->saddr))
1111 flags |= RT6_LOOKUP_F_HAS_SADDR;
1113 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1121 static const struct l3mdev_ops vrf_l3mdev_ops = {
1122 .l3mdev_fib_table = vrf_fib_table,
1123 .l3mdev_l3_rcv = vrf_l3_rcv,
1124 .l3mdev_l3_out = vrf_l3_out,
1125 #if IS_ENABLED(CONFIG_IPV6)
1126 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1130 static void vrf_get_drvinfo(struct net_device *dev,
1131 struct ethtool_drvinfo *info)
1133 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1134 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1137 static const struct ethtool_ops vrf_ethtool_ops = {
1138 .get_drvinfo = vrf_get_drvinfo,
1141 static inline size_t vrf_fib_rule_nl_size(void)
1145 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1146 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1147 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1152 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1154 struct fib_rule_hdr *frh;
1155 struct nlmsghdr *nlh;
1156 struct sk_buff *skb;
1159 if (family == AF_INET6 && !ipv6_mod_enabled())
1162 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1166 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1168 goto nla_put_failure;
1170 /* rule only needs to appear once */
1171 nlh->nlmsg_flags |= NLM_F_EXCL;
1173 frh = nlmsg_data(nlh);
1174 memset(frh, 0, sizeof(*frh));
1175 frh->family = family;
1176 frh->action = FR_ACT_TO_TBL;
1178 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1179 goto nla_put_failure;
1181 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1182 goto nla_put_failure;
1184 nlmsg_end(skb, nlh);
1186 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1187 skb->sk = dev_net(dev)->rtnl;
1189 err = fib_nl_newrule(skb, nlh, NULL);
1193 err = fib_nl_delrule(skb, nlh, NULL);
1207 static int vrf_add_fib_rules(const struct net_device *dev)
1211 err = vrf_fib_rule(dev, AF_INET, true);
1215 err = vrf_fib_rule(dev, AF_INET6, true);
1219 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1220 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1227 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1229 vrf_fib_rule(dev, AF_INET6, false);
1233 vrf_fib_rule(dev, AF_INET, false);
1236 netdev_err(dev, "Failed to add FIB rules.\n");
1240 static void vrf_setup(struct net_device *dev)
1244 /* Initialize the device structure. */
1245 dev->netdev_ops = &vrf_netdev_ops;
1246 dev->l3mdev_ops = &vrf_l3mdev_ops;
1247 dev->ethtool_ops = &vrf_ethtool_ops;
1248 dev->needs_free_netdev = true;
1250 /* Fill in device structure with ethernet-generic values. */
1251 eth_hw_addr_random(dev);
1253 /* don't acquire vrf device's netif_tx_lock when transmitting */
1254 dev->features |= NETIF_F_LLTX;
1256 /* don't allow vrf devices to change network namespaces. */
1257 dev->features |= NETIF_F_NETNS_LOCAL;
1259 /* does not make sense for a VLAN to be added to a vrf device */
1260 dev->features |= NETIF_F_VLAN_CHALLENGED;
1262 /* enable offload features */
1263 dev->features |= NETIF_F_GSO_SOFTWARE;
1264 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1265 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1267 dev->hw_features = dev->features;
1268 dev->hw_enc_features = dev->features;
1270 /* default to no qdisc; user can add if desired */
1271 dev->priv_flags |= IFF_NO_QUEUE;
1274 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1275 struct netlink_ext_ack *extack)
1277 if (tb[IFLA_ADDRESS]) {
1278 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1279 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1282 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1283 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1284 return -EADDRNOTAVAIL;
1290 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1292 struct net_device *port_dev;
1293 struct list_head *iter;
1295 netdev_for_each_lower_dev(dev, port_dev, iter)
1296 vrf_del_slave(dev, port_dev);
1298 unregister_netdevice_queue(dev, head);
1301 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1302 struct nlattr *tb[], struct nlattr *data[],
1303 struct netlink_ext_ack *extack)
1305 struct net_vrf *vrf = netdev_priv(dev);
1306 bool *add_fib_rules;
1310 if (!data || !data[IFLA_VRF_TABLE]) {
1311 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1315 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1316 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1317 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1318 "Invalid VRF table id");
1322 dev->priv_flags |= IFF_L3MDEV_MASTER;
1324 err = register_netdevice(dev);
1329 add_fib_rules = net_generic(net, vrf_net_id);
1330 if (*add_fib_rules) {
1331 err = vrf_add_fib_rules(dev);
1333 unregister_netdevice(dev);
1336 *add_fib_rules = false;
1343 static size_t vrf_nl_getsize(const struct net_device *dev)
1345 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1348 static int vrf_fillinfo(struct sk_buff *skb,
1349 const struct net_device *dev)
1351 struct net_vrf *vrf = netdev_priv(dev);
1353 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1356 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1357 const struct net_device *slave_dev)
1359 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1362 static int vrf_fill_slave_info(struct sk_buff *skb,
1363 const struct net_device *vrf_dev,
1364 const struct net_device *slave_dev)
1366 struct net_vrf *vrf = netdev_priv(vrf_dev);
1368 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1374 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1375 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1378 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1380 .priv_size = sizeof(struct net_vrf),
1382 .get_size = vrf_nl_getsize,
1383 .policy = vrf_nl_policy,
1384 .validate = vrf_validate,
1385 .fill_info = vrf_fillinfo,
1387 .get_slave_size = vrf_get_slave_size,
1388 .fill_slave_info = vrf_fill_slave_info,
1390 .newlink = vrf_newlink,
1391 .dellink = vrf_dellink,
1393 .maxtype = IFLA_VRF_MAX,
1396 static int vrf_device_event(struct notifier_block *unused,
1397 unsigned long event, void *ptr)
1399 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1401 /* only care about unregister events to drop slave references */
1402 if (event == NETDEV_UNREGISTER) {
1403 struct net_device *vrf_dev;
1405 if (!netif_is_l3_slave(dev))
1408 vrf_dev = netdev_master_upper_dev_get(dev);
1409 vrf_del_slave(vrf_dev, dev);
1415 static struct notifier_block vrf_notifier_block __read_mostly = {
1416 .notifier_call = vrf_device_event,
1419 /* Initialize per network namespace state */
1420 static int __net_init vrf_netns_init(struct net *net)
1422 bool *add_fib_rules = net_generic(net, vrf_net_id);
1424 *add_fib_rules = true;
1429 static struct pernet_operations vrf_net_ops __net_initdata = {
1430 .init = vrf_netns_init,
1432 .size = sizeof(bool),
1435 static int __init vrf_init_module(void)
1439 register_netdevice_notifier(&vrf_notifier_block);
1441 rc = register_pernet_subsys(&vrf_net_ops);
1445 rc = rtnl_link_register(&vrf_link_ops);
1447 unregister_pernet_subsys(&vrf_net_ops);
1454 unregister_netdevice_notifier(&vrf_notifier_block);
1458 module_init(vrf_init_module);
1459 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1460 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1461 MODULE_LICENSE("GPL");
1462 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1463 MODULE_VERSION(DRV_VERSION);