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;
51 #if IS_ENABLED(CONFIG_IPV6)
52 struct fib6_table *fib6_table;
64 struct u64_stats_sync syncp;
67 static void vrf_rx_stats(struct net_device *dev, int len)
69 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
71 u64_stats_update_begin(&dstats->syncp);
73 dstats->rx_bytes += len;
74 u64_stats_update_end(&dstats->syncp);
77 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
79 vrf_dev->stats.tx_errors++;
83 static void vrf_get_stats64(struct net_device *dev,
84 struct rtnl_link_stats64 *stats)
88 for_each_possible_cpu(i) {
89 const struct pcpu_dstats *dstats;
90 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
93 dstats = per_cpu_ptr(dev->dstats, i);
95 start = u64_stats_fetch_begin_irq(&dstats->syncp);
96 tbytes = dstats->tx_bytes;
97 tpkts = dstats->tx_pkts;
98 tdrops = dstats->tx_drps;
99 rbytes = dstats->rx_bytes;
100 rpkts = dstats->rx_pkts;
101 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
102 stats->tx_bytes += tbytes;
103 stats->tx_packets += tpkts;
104 stats->tx_dropped += tdrops;
105 stats->rx_bytes += rbytes;
106 stats->rx_packets += rpkts;
110 /* by default VRF devices do not have a qdisc and are expected
111 * to be created with only a single queue.
113 static bool qdisc_tx_is_default(const struct net_device *dev)
115 struct netdev_queue *txq;
118 if (dev->num_tx_queues > 1)
121 txq = netdev_get_tx_queue(dev, 0);
122 qdisc = rcu_access_pointer(txq->qdisc);
124 return !qdisc->enqueue;
127 /* Local traffic destined to local address. Reinsert the packet to rx
128 * path, similar to loopback handling.
130 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
131 struct dst_entry *dst)
137 skb_dst_set(skb, dst);
139 /* set pkt_type to avoid skb hitting packet taps twice -
140 * once on Tx and again in Rx processing
142 skb->pkt_type = PACKET_LOOPBACK;
144 skb->protocol = eth_type_trans(skb, dev);
146 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
147 vrf_rx_stats(dev, len);
149 this_cpu_inc(dev->dstats->rx_drps);
154 #if IS_ENABLED(CONFIG_IPV6)
155 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
160 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
161 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
163 if (likely(err == 1))
164 err = dst_output(net, sk, skb);
169 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
170 struct net_device *dev)
172 const struct ipv6hdr *iph = ipv6_hdr(skb);
173 struct net *net = dev_net(skb->dev);
174 struct flowi6 fl6 = {
175 /* needed to match OIF rule */
176 .flowi6_oif = dev->ifindex,
177 .flowi6_iif = LOOPBACK_IFINDEX,
180 .flowlabel = ip6_flowinfo(iph),
181 .flowi6_mark = skb->mark,
182 .flowi6_proto = iph->nexthdr,
183 .flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF,
185 int ret = NET_XMIT_DROP;
186 struct dst_entry *dst;
187 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
189 dst = ip6_route_output(net, NULL, &fl6);
195 /* if dst.dev is loopback or the VRF device again this is locally
196 * originated traffic destined to a local address. Short circuit
200 return vrf_local_xmit(skb, dev, dst);
202 skb_dst_set(skb, dst);
204 /* strip the ethernet header added for pass through VRF device */
205 __skb_pull(skb, skb_network_offset(skb));
207 ret = vrf_ip6_local_out(net, skb->sk, skb);
208 if (unlikely(net_xmit_eval(ret)))
209 dev->stats.tx_errors++;
211 ret = NET_XMIT_SUCCESS;
215 vrf_tx_error(dev, skb);
216 return NET_XMIT_DROP;
219 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
220 struct net_device *dev)
222 vrf_tx_error(dev, skb);
223 return NET_XMIT_DROP;
227 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
228 static int vrf_ip_local_out(struct net *net, struct sock *sk,
233 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
234 skb, NULL, skb_dst(skb)->dev, dst_output);
235 if (likely(err == 1))
236 err = dst_output(net, sk, skb);
241 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
242 struct net_device *vrf_dev)
244 struct iphdr *ip4h = ip_hdr(skb);
245 int ret = NET_XMIT_DROP;
246 struct flowi4 fl4 = {
247 /* needed to match OIF rule */
248 .flowi4_oif = vrf_dev->ifindex,
249 .flowi4_iif = LOOPBACK_IFINDEX,
250 .flowi4_tos = RT_TOS(ip4h->tos),
251 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF,
252 .flowi4_proto = ip4h->protocol,
253 .daddr = ip4h->daddr,
254 .saddr = ip4h->saddr,
256 struct net *net = dev_net(vrf_dev);
259 rt = ip_route_output_flow(net, &fl4, NULL);
265 /* if dst.dev is loopback or the VRF device again this is locally
266 * originated traffic destined to a local address. Short circuit
269 if (rt->dst.dev == vrf_dev)
270 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
272 skb_dst_set(skb, &rt->dst);
274 /* strip the ethernet header added for pass through VRF device */
275 __skb_pull(skb, skb_network_offset(skb));
278 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
282 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
283 if (unlikely(net_xmit_eval(ret)))
284 vrf_dev->stats.tx_errors++;
286 ret = NET_XMIT_SUCCESS;
291 vrf_tx_error(vrf_dev, skb);
295 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
297 switch (skb->protocol) {
298 case htons(ETH_P_IP):
299 return vrf_process_v4_outbound(skb, dev);
300 case htons(ETH_P_IPV6):
301 return vrf_process_v6_outbound(skb, dev);
303 vrf_tx_error(dev, skb);
304 return NET_XMIT_DROP;
308 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
311 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
313 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
314 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
316 u64_stats_update_begin(&dstats->syncp);
318 dstats->tx_bytes += len;
319 u64_stats_update_end(&dstats->syncp);
321 this_cpu_inc(dev->dstats->tx_drps);
327 static int vrf_finish_direct(struct net *net, struct sock *sk,
330 struct net_device *vrf_dev = skb->dev;
332 if (!list_empty(&vrf_dev->ptype_all) &&
333 likely(skb_headroom(skb) >= ETH_HLEN)) {
334 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
336 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
337 eth_zero_addr(eth->h_dest);
338 eth->h_proto = skb->protocol;
341 dev_queue_xmit_nit(skb, vrf_dev);
342 rcu_read_unlock_bh();
344 skb_pull(skb, ETH_HLEN);
350 #if IS_ENABLED(CONFIG_IPV6)
351 /* modelled after ip6_finish_output2 */
352 static int vrf_finish_output6(struct net *net, struct sock *sk,
355 struct dst_entry *dst = skb_dst(skb);
356 struct net_device *dev = dst->dev;
357 struct neighbour *neigh;
358 struct in6_addr *nexthop;
363 skb->protocol = htons(ETH_P_IPV6);
367 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
368 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
369 if (unlikely(!neigh))
370 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
371 if (!IS_ERR(neigh)) {
372 sock_confirm_neigh(skb, neigh);
373 ret = neigh_output(neigh, skb);
374 rcu_read_unlock_bh();
377 rcu_read_unlock_bh();
379 IP6_INC_STATS(dev_net(dst->dev),
380 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
385 /* modelled after ip6_output */
386 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
388 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
389 net, sk, skb, NULL, skb_dst(skb)->dev,
391 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
394 /* set dst on skb to send packet to us via dev_xmit path. Allows
395 * packet to go through device based features such as qdisc, netfilter
396 * hooks and packet sockets with skb->dev set to vrf device.
398 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
401 struct net_vrf *vrf = netdev_priv(vrf_dev);
402 struct dst_entry *dst = NULL;
403 struct rt6_info *rt6;
407 rt6 = rcu_dereference(vrf->rt6);
415 if (unlikely(!dst)) {
416 vrf_tx_error(vrf_dev, skb);
421 skb_dst_set(skb, dst);
426 static int vrf_output6_direct(struct net *net, struct sock *sk,
429 skb->protocol = htons(ETH_P_IPV6);
431 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
432 net, sk, skb, NULL, skb->dev,
434 !(IPCB(skb)->flags & IPSKB_REROUTED));
437 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
441 struct net *net = dev_net(vrf_dev);
446 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
447 skb, NULL, vrf_dev, vrf_output6_direct);
449 if (likely(err == 1))
450 err = vrf_output6_direct(net, sk, skb);
452 /* reset skb device */
453 if (likely(err == 1))
461 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
465 /* don't divert link scope packets */
466 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
469 if (qdisc_tx_is_default(vrf_dev))
470 return vrf_ip6_out_direct(vrf_dev, sk, skb);
472 return vrf_ip6_out_redirect(vrf_dev, skb);
476 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
478 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
479 struct net *net = dev_net(dev);
480 struct dst_entry *dst;
482 RCU_INIT_POINTER(vrf->rt6, NULL);
485 /* move dev in dst's to loopback so this VRF device can be deleted
486 * - based on dst_ifdown
491 dst->dev = net->loopback_dev;
497 static int vrf_rt6_create(struct net_device *dev)
499 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
500 struct net_vrf *vrf = netdev_priv(dev);
501 struct net *net = dev_net(dev);
502 struct rt6_info *rt6;
505 /* IPv6 can be CONFIG enabled and then disabled runtime */
506 if (!ipv6_mod_enabled())
509 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
510 if (!vrf->fib6_table)
513 /* create a dst for routing packets out a VRF device */
514 rt6 = ip6_dst_alloc(net, dev, flags);
518 rt6->dst.output = vrf_output6;
520 rcu_assign_pointer(vrf->rt6, rt6);
527 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
534 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
538 static int vrf_rt6_create(struct net_device *dev)
544 /* modelled after ip_finish_output2 */
545 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
547 struct dst_entry *dst = skb_dst(skb);
548 struct rtable *rt = (struct rtable *)dst;
549 struct net_device *dev = dst->dev;
550 unsigned int hh_len = LL_RESERVED_SPACE(dev);
551 struct neighbour *neigh;
557 /* Be paranoid, rather than too clever. */
558 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
559 struct sk_buff *skb2;
561 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
567 skb_set_owner_w(skb2, skb->sk);
575 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
576 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
577 if (unlikely(!neigh))
578 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
579 if (!IS_ERR(neigh)) {
580 sock_confirm_neigh(skb, neigh);
581 ret = neigh_output(neigh, skb);
582 rcu_read_unlock_bh();
586 rcu_read_unlock_bh();
588 vrf_tx_error(skb->dev, skb);
592 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
594 struct net_device *dev = skb_dst(skb)->dev;
596 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
599 skb->protocol = htons(ETH_P_IP);
601 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
602 net, sk, skb, NULL, dev,
604 !(IPCB(skb)->flags & IPSKB_REROUTED));
607 /* set dst on skb to send packet to us via dev_xmit path. Allows
608 * packet to go through device based features such as qdisc, netfilter
609 * hooks and packet sockets with skb->dev set to vrf device.
611 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
614 struct net_vrf *vrf = netdev_priv(vrf_dev);
615 struct dst_entry *dst = NULL;
620 rth = rcu_dereference(vrf->rth);
628 if (unlikely(!dst)) {
629 vrf_tx_error(vrf_dev, skb);
634 skb_dst_set(skb, dst);
639 static int vrf_output_direct(struct net *net, struct sock *sk,
642 skb->protocol = htons(ETH_P_IP);
644 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
645 net, sk, skb, NULL, skb->dev,
647 !(IPCB(skb)->flags & IPSKB_REROUTED));
650 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
654 struct net *net = dev_net(vrf_dev);
659 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
660 skb, NULL, vrf_dev, vrf_output_direct);
662 if (likely(err == 1))
663 err = vrf_output_direct(net, sk, skb);
665 /* reset skb device */
666 if (likely(err == 1))
674 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
678 /* don't divert multicast or local broadcast */
679 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
680 ipv4_is_lbcast(ip_hdr(skb)->daddr))
683 if (qdisc_tx_is_default(vrf_dev))
684 return vrf_ip_out_direct(vrf_dev, sk, skb);
686 return vrf_ip_out_redirect(vrf_dev, skb);
689 /* called with rcu lock held */
690 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
697 return vrf_ip_out(vrf_dev, sk, skb);
699 return vrf_ip6_out(vrf_dev, sk, skb);
706 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
708 struct rtable *rth = rtnl_dereference(vrf->rth);
709 struct net *net = dev_net(dev);
710 struct dst_entry *dst;
712 RCU_INIT_POINTER(vrf->rth, NULL);
715 /* move dev in dst's to loopback so this VRF device can be deleted
716 * - based on dst_ifdown
721 dst->dev = net->loopback_dev;
727 static int vrf_rtable_create(struct net_device *dev)
729 struct net_vrf *vrf = netdev_priv(dev);
732 if (!fib_new_table(dev_net(dev), vrf->tb_id))
735 /* create a dst for routing packets out through a VRF device */
736 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
740 rth->dst.output = vrf_output;
742 rcu_assign_pointer(vrf->rth, rth);
747 /**************************** device handling ********************/
749 /* cycle interface to flush neighbor cache and move routes across tables */
750 static void cycle_netdev(struct net_device *dev,
751 struct netlink_ext_ack *extack)
753 unsigned int flags = dev->flags;
756 if (!netif_running(dev))
759 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
761 ret = dev_change_flags(dev, flags, extack);
765 "Failed to cycle device %s; route tables might be wrong!\n",
770 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
771 struct netlink_ext_ack *extack)
775 /* do not allow loopback device to be enslaved to a VRF.
776 * The vrf device acts as the loopback for the vrf.
778 if (port_dev == dev_net(dev)->loopback_dev) {
779 NL_SET_ERR_MSG(extack,
780 "Can not enslave loopback device to a VRF");
784 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
785 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
789 cycle_netdev(port_dev, extack);
794 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
798 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
799 struct netlink_ext_ack *extack)
801 if (netif_is_l3_master(port_dev)) {
802 NL_SET_ERR_MSG(extack,
803 "Can not enslave an L3 master device to a VRF");
807 if (netif_is_l3_slave(port_dev))
810 return do_vrf_add_slave(dev, port_dev, extack);
813 /* inverse of do_vrf_add_slave */
814 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
816 netdev_upper_dev_unlink(port_dev, dev);
817 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
819 cycle_netdev(port_dev, NULL);
824 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
826 return do_vrf_del_slave(dev, port_dev);
829 static void vrf_dev_uninit(struct net_device *dev)
831 struct net_vrf *vrf = netdev_priv(dev);
833 vrf_rtable_release(dev, vrf);
834 vrf_rt6_release(dev, vrf);
836 free_percpu(dev->dstats);
840 static int vrf_dev_init(struct net_device *dev)
842 struct net_vrf *vrf = netdev_priv(dev);
844 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
848 /* create the default dst which points back to us */
849 if (vrf_rtable_create(dev) != 0)
852 if (vrf_rt6_create(dev) != 0)
855 dev->flags = IFF_MASTER | IFF_NOARP;
857 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
858 dev->mtu = 64 * 1024;
860 /* similarly, oper state is irrelevant; set to up to avoid confusion */
861 dev->operstate = IF_OPER_UP;
862 netdev_lockdep_set_classes(dev);
866 vrf_rtable_release(dev, vrf);
868 free_percpu(dev->dstats);
874 static const struct net_device_ops vrf_netdev_ops = {
875 .ndo_init = vrf_dev_init,
876 .ndo_uninit = vrf_dev_uninit,
877 .ndo_start_xmit = vrf_xmit,
878 .ndo_get_stats64 = vrf_get_stats64,
879 .ndo_add_slave = vrf_add_slave,
880 .ndo_del_slave = vrf_del_slave,
883 static u32 vrf_fib_table(const struct net_device *dev)
885 struct net_vrf *vrf = netdev_priv(dev);
890 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
896 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
898 struct net_device *dev)
900 struct net *net = dev_net(dev);
902 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
903 skb = NULL; /* kfree_skb(skb) handled by nf code */
908 #if IS_ENABLED(CONFIG_IPV6)
909 /* neighbor handling is done with actual device; do not want
910 * to flip skb->dev for those ndisc packets. This really fails
911 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
914 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
916 const struct ipv6hdr *iph = ipv6_hdr(skb);
919 if (iph->nexthdr == NEXTHDR_ICMP) {
920 const struct icmp6hdr *icmph;
921 struct icmp6hdr _icmph;
923 icmph = skb_header_pointer(skb, sizeof(*iph),
924 sizeof(_icmph), &_icmph);
928 switch (icmph->icmp6_type) {
929 case NDISC_ROUTER_SOLICITATION:
930 case NDISC_ROUTER_ADVERTISEMENT:
931 case NDISC_NEIGHBOUR_SOLICITATION:
932 case NDISC_NEIGHBOUR_ADVERTISEMENT:
943 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
944 const struct net_device *dev,
947 const struct sk_buff *skb,
950 struct net_vrf *vrf = netdev_priv(dev);
952 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
955 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
958 const struct ipv6hdr *iph = ipv6_hdr(skb);
959 struct flowi6 fl6 = {
960 .flowi6_iif = ifindex,
961 .flowi6_mark = skb->mark,
962 .flowi6_proto = iph->nexthdr,
965 .flowlabel = ip6_flowinfo(iph),
967 struct net *net = dev_net(vrf_dev);
968 struct rt6_info *rt6;
970 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
971 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
975 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
978 skb_dst_set(skb, &rt6->dst);
981 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
984 int orig_iif = skb->skb_iif;
985 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
986 bool is_ndisc = ipv6_ndisc_frame(skb);
988 /* loopback, multicast & non-ND link-local traffic; do not push through
989 * packet taps again. Reset pkt_type for upper layers to process skb
991 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
993 skb->skb_iif = vrf_dev->ifindex;
994 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
995 if (skb->pkt_type == PACKET_LOOPBACK)
996 skb->pkt_type = PACKET_HOST;
1000 /* if packet is NDISC then keep the ingress interface */
1002 vrf_rx_stats(vrf_dev, skb->len);
1004 skb->skb_iif = vrf_dev->ifindex;
1006 if (!list_empty(&vrf_dev->ptype_all)) {
1007 skb_push(skb, skb->mac_len);
1008 dev_queue_xmit_nit(skb, vrf_dev);
1009 skb_pull(skb, skb->mac_len);
1012 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1016 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1018 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1024 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1025 struct sk_buff *skb)
1031 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1032 struct sk_buff *skb)
1035 skb->skb_iif = vrf_dev->ifindex;
1036 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1038 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1041 /* loopback traffic; do not push through packet taps again.
1042 * Reset pkt_type for upper layers to process skb
1044 if (skb->pkt_type == PACKET_LOOPBACK) {
1045 skb->pkt_type = PACKET_HOST;
1049 vrf_rx_stats(vrf_dev, skb->len);
1051 if (!list_empty(&vrf_dev->ptype_all)) {
1052 skb_push(skb, skb->mac_len);
1053 dev_queue_xmit_nit(skb, vrf_dev);
1054 skb_pull(skb, skb->mac_len);
1057 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1062 /* called with rcu lock held */
1063 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1064 struct sk_buff *skb,
1069 return vrf_ip_rcv(vrf_dev, skb);
1071 return vrf_ip6_rcv(vrf_dev, skb);
1077 #if IS_ENABLED(CONFIG_IPV6)
1078 /* send to link-local or multicast address via interface enslaved to
1079 * VRF device. Force lookup to VRF table without changing flow struct
1081 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1084 struct net *net = dev_net(dev);
1085 int flags = RT6_LOOKUP_F_IFACE;
1086 struct dst_entry *dst = NULL;
1087 struct rt6_info *rt;
1089 /* VRF device does not have a link-local address and
1090 * sending packets to link-local or mcast addresses over
1091 * a VRF device does not make sense
1093 if (fl6->flowi6_oif == dev->ifindex) {
1094 dst = &net->ipv6.ip6_null_entry->dst;
1099 if (!ipv6_addr_any(&fl6->saddr))
1100 flags |= RT6_LOOKUP_F_HAS_SADDR;
1102 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1110 static const struct l3mdev_ops vrf_l3mdev_ops = {
1111 .l3mdev_fib_table = vrf_fib_table,
1112 .l3mdev_l3_rcv = vrf_l3_rcv,
1113 .l3mdev_l3_out = vrf_l3_out,
1114 #if IS_ENABLED(CONFIG_IPV6)
1115 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1119 static void vrf_get_drvinfo(struct net_device *dev,
1120 struct ethtool_drvinfo *info)
1122 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1123 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1126 static const struct ethtool_ops vrf_ethtool_ops = {
1127 .get_drvinfo = vrf_get_drvinfo,
1130 static inline size_t vrf_fib_rule_nl_size(void)
1134 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1135 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1136 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1137 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1142 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1144 struct fib_rule_hdr *frh;
1145 struct nlmsghdr *nlh;
1146 struct sk_buff *skb;
1149 if (family == AF_INET6 && !ipv6_mod_enabled())
1152 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1156 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1158 goto nla_put_failure;
1160 /* rule only needs to appear once */
1161 nlh->nlmsg_flags |= NLM_F_EXCL;
1163 frh = nlmsg_data(nlh);
1164 memset(frh, 0, sizeof(*frh));
1165 frh->family = family;
1166 frh->action = FR_ACT_TO_TBL;
1168 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1169 goto nla_put_failure;
1171 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1172 goto nla_put_failure;
1174 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1175 goto nla_put_failure;
1177 nlmsg_end(skb, nlh);
1179 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1180 skb->sk = dev_net(dev)->rtnl;
1182 err = fib_nl_newrule(skb, nlh, NULL);
1186 err = fib_nl_delrule(skb, nlh, NULL);
1200 static int vrf_add_fib_rules(const struct net_device *dev)
1204 err = vrf_fib_rule(dev, AF_INET, true);
1208 err = vrf_fib_rule(dev, AF_INET6, true);
1212 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1213 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1218 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1219 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1226 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1228 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1231 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1233 vrf_fib_rule(dev, AF_INET6, false);
1237 vrf_fib_rule(dev, AF_INET, false);
1240 netdev_err(dev, "Failed to add FIB rules.\n");
1244 static void vrf_setup(struct net_device *dev)
1248 /* Initialize the device structure. */
1249 dev->netdev_ops = &vrf_netdev_ops;
1250 dev->l3mdev_ops = &vrf_l3mdev_ops;
1251 dev->ethtool_ops = &vrf_ethtool_ops;
1252 dev->needs_free_netdev = true;
1254 /* Fill in device structure with ethernet-generic values. */
1255 eth_hw_addr_random(dev);
1257 /* don't acquire vrf device's netif_tx_lock when transmitting */
1258 dev->features |= NETIF_F_LLTX;
1260 /* don't allow vrf devices to change network namespaces. */
1261 dev->features |= NETIF_F_NETNS_LOCAL;
1263 /* does not make sense for a VLAN to be added to a vrf device */
1264 dev->features |= NETIF_F_VLAN_CHALLENGED;
1266 /* enable offload features */
1267 dev->features |= NETIF_F_GSO_SOFTWARE;
1268 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1269 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1271 dev->hw_features = dev->features;
1272 dev->hw_enc_features = dev->features;
1274 /* default to no qdisc; user can add if desired */
1275 dev->priv_flags |= IFF_NO_QUEUE;
1276 dev->priv_flags |= IFF_NO_RX_HANDLER;
1282 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1283 struct netlink_ext_ack *extack)
1285 if (tb[IFLA_ADDRESS]) {
1286 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1287 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1290 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1291 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1292 return -EADDRNOTAVAIL;
1298 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1300 struct net_device *port_dev;
1301 struct list_head *iter;
1303 netdev_for_each_lower_dev(dev, port_dev, iter)
1304 vrf_del_slave(dev, port_dev);
1306 unregister_netdevice_queue(dev, head);
1309 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1310 struct nlattr *tb[], struct nlattr *data[],
1311 struct netlink_ext_ack *extack)
1313 struct net_vrf *vrf = netdev_priv(dev);
1314 bool *add_fib_rules;
1318 if (!data || !data[IFLA_VRF_TABLE]) {
1319 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1323 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1324 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1325 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1326 "Invalid VRF table id");
1330 dev->priv_flags |= IFF_L3MDEV_MASTER;
1332 err = register_netdevice(dev);
1337 add_fib_rules = net_generic(net, vrf_net_id);
1338 if (*add_fib_rules) {
1339 err = vrf_add_fib_rules(dev);
1341 unregister_netdevice(dev);
1344 *add_fib_rules = false;
1351 static size_t vrf_nl_getsize(const struct net_device *dev)
1353 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1356 static int vrf_fillinfo(struct sk_buff *skb,
1357 const struct net_device *dev)
1359 struct net_vrf *vrf = netdev_priv(dev);
1361 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1364 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1365 const struct net_device *slave_dev)
1367 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1370 static int vrf_fill_slave_info(struct sk_buff *skb,
1371 const struct net_device *vrf_dev,
1372 const struct net_device *slave_dev)
1374 struct net_vrf *vrf = netdev_priv(vrf_dev);
1376 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1382 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1383 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1386 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1388 .priv_size = sizeof(struct net_vrf),
1390 .get_size = vrf_nl_getsize,
1391 .policy = vrf_nl_policy,
1392 .validate = vrf_validate,
1393 .fill_info = vrf_fillinfo,
1395 .get_slave_size = vrf_get_slave_size,
1396 .fill_slave_info = vrf_fill_slave_info,
1398 .newlink = vrf_newlink,
1399 .dellink = vrf_dellink,
1401 .maxtype = IFLA_VRF_MAX,
1404 static int vrf_device_event(struct notifier_block *unused,
1405 unsigned long event, void *ptr)
1407 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1409 /* only care about unregister events to drop slave references */
1410 if (event == NETDEV_UNREGISTER) {
1411 struct net_device *vrf_dev;
1413 if (!netif_is_l3_slave(dev))
1416 vrf_dev = netdev_master_upper_dev_get(dev);
1417 vrf_del_slave(vrf_dev, dev);
1423 static struct notifier_block vrf_notifier_block __read_mostly = {
1424 .notifier_call = vrf_device_event,
1427 /* Initialize per network namespace state */
1428 static int __net_init vrf_netns_init(struct net *net)
1430 bool *add_fib_rules = net_generic(net, vrf_net_id);
1432 *add_fib_rules = true;
1437 static struct pernet_operations vrf_net_ops __net_initdata = {
1438 .init = vrf_netns_init,
1440 .size = sizeof(bool),
1443 static int __init vrf_init_module(void)
1447 register_netdevice_notifier(&vrf_notifier_block);
1449 rc = register_pernet_subsys(&vrf_net_ops);
1453 rc = rtnl_link_register(&vrf_link_ops);
1455 unregister_pernet_subsys(&vrf_net_ops);
1462 unregister_netdevice_notifier(&vrf_notifier_block);
1466 module_init(vrf_init_module);
1467 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1468 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1469 MODULE_LICENSE("GPL");
1470 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1471 MODULE_VERSION(DRV_VERSION);