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 RT_FL_TOS(oldflp4) \
41 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
43 #define DRV_NAME "vrf"
44 #define DRV_VERSION "1.0"
46 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
47 static bool add_fib_rules = true;
50 struct rtable __rcu *rth;
51 struct rtable __rcu *rth_local;
52 struct rt6_info __rcu *rt6;
53 struct rt6_info __rcu *rt6_local;
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 struct rtnl_link_stats64 *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;
111 /* Local traffic destined to local address. Reinsert the packet to rx
112 * path, similar to loopback handling.
114 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
115 struct dst_entry *dst)
121 skb_dst_set(skb, dst);
124 /* set pkt_type to avoid skb hitting packet taps twice -
125 * once on Tx and again in Rx processing
127 skb->pkt_type = PACKET_LOOPBACK;
129 skb->protocol = eth_type_trans(skb, dev);
131 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
132 vrf_rx_stats(dev, len);
134 this_cpu_inc(dev->dstats->rx_drps);
139 #if IS_ENABLED(CONFIG_IPV6)
140 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
141 struct net_device *dev)
143 const struct ipv6hdr *iph = ipv6_hdr(skb);
144 struct net *net = dev_net(skb->dev);
145 struct flowi6 fl6 = {
146 /* needed to match OIF rule */
147 .flowi6_oif = dev->ifindex,
148 .flowi6_iif = LOOPBACK_IFINDEX,
151 .flowlabel = ip6_flowinfo(iph),
152 .flowi6_mark = skb->mark,
153 .flowi6_proto = iph->nexthdr,
154 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
156 int ret = NET_XMIT_DROP;
157 struct dst_entry *dst;
158 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
160 dst = ip6_route_output(net, NULL, &fl6);
166 /* if dst.dev is loopback or the VRF device again this is locally
167 * originated traffic destined to a local address. Short circuit
168 * to Rx path using our local dst
170 if (dst->dev == net->loopback_dev || dst->dev == dev) {
171 struct net_vrf *vrf = netdev_priv(dev);
172 struct rt6_info *rt6_local;
174 /* release looked up dst and use cached local dst */
179 rt6_local = rcu_dereference(vrf->rt6_local);
180 if (unlikely(!rt6_local)) {
185 /* Ordering issue: cached local dst is created on newlink
186 * before the IPv6 initialization. Using the local dst
187 * requires rt6i_idev to be set so make sure it is.
189 if (unlikely(!rt6_local->rt6i_idev)) {
190 rt6_local->rt6i_idev = in6_dev_get(dev);
191 if (!rt6_local->rt6i_idev) {
197 dst = &rt6_local->dst;
202 return vrf_local_xmit(skb, dev, &rt6_local->dst);
205 skb_dst_set(skb, dst);
207 /* strip the ethernet header added for pass through VRF device */
208 __skb_pull(skb, skb_network_offset(skb));
210 ret = ip6_local_out(net, skb->sk, skb);
211 if (unlikely(net_xmit_eval(ret)))
212 dev->stats.tx_errors++;
214 ret = NET_XMIT_SUCCESS;
218 vrf_tx_error(dev, skb);
219 return NET_XMIT_DROP;
222 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
223 struct net_device *dev)
225 vrf_tx_error(dev, skb);
226 return NET_XMIT_DROP;
230 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
231 struct net_device *vrf_dev)
233 struct iphdr *ip4h = ip_hdr(skb);
234 int ret = NET_XMIT_DROP;
235 struct flowi4 fl4 = {
236 /* needed to match OIF rule */
237 .flowi4_oif = vrf_dev->ifindex,
238 .flowi4_iif = LOOPBACK_IFINDEX,
239 .flowi4_tos = RT_TOS(ip4h->tos),
240 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
241 FLOWI_FLAG_SKIP_NH_OIF,
242 .daddr = ip4h->daddr,
244 struct net *net = dev_net(vrf_dev);
247 rt = ip_route_output_flow(net, &fl4, NULL);
251 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
258 /* if dst.dev is loopback or the VRF device again this is locally
259 * originated traffic destined to a local address. Short circuit
260 * to Rx path using our local dst
262 if (rt->dst.dev == net->loopback_dev || rt->dst.dev == vrf_dev) {
263 struct net_vrf *vrf = netdev_priv(vrf_dev);
264 struct rtable *rth_local;
265 struct dst_entry *dst = NULL;
271 rth_local = rcu_dereference(vrf->rth_local);
272 if (likely(rth_local)) {
273 dst = &rth_local->dst;
282 return vrf_local_xmit(skb, vrf_dev, dst);
285 skb_dst_set(skb, &rt->dst);
287 /* strip the ethernet header added for pass through VRF device */
288 __skb_pull(skb, skb_network_offset(skb));
291 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
295 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
296 if (unlikely(net_xmit_eval(ret)))
297 vrf_dev->stats.tx_errors++;
299 ret = NET_XMIT_SUCCESS;
304 vrf_tx_error(vrf_dev, skb);
308 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
310 switch (skb->protocol) {
311 case htons(ETH_P_IP):
312 return vrf_process_v4_outbound(skb, dev);
313 case htons(ETH_P_IPV6):
314 return vrf_process_v6_outbound(skb, dev);
316 vrf_tx_error(dev, skb);
317 return NET_XMIT_DROP;
321 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
323 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
325 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
326 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
328 u64_stats_update_begin(&dstats->syncp);
330 dstats->tx_bytes += skb->len;
331 u64_stats_update_end(&dstats->syncp);
333 this_cpu_inc(dev->dstats->tx_drps);
339 #if IS_ENABLED(CONFIG_IPV6)
340 /* modelled after ip6_finish_output2 */
341 static int vrf_finish_output6(struct net *net, struct sock *sk,
344 struct dst_entry *dst = skb_dst(skb);
345 struct net_device *dev = dst->dev;
346 struct neighbour *neigh;
347 struct in6_addr *nexthop;
350 skb->protocol = htons(ETH_P_IPV6);
354 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
355 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
356 if (unlikely(!neigh))
357 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
358 if (!IS_ERR(neigh)) {
359 ret = dst_neigh_output(dst, neigh, skb);
360 rcu_read_unlock_bh();
363 rcu_read_unlock_bh();
365 IP6_INC_STATS(dev_net(dst->dev),
366 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
371 /* modelled after ip6_output */
372 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
374 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
375 net, sk, skb, NULL, skb_dst(skb)->dev,
377 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
381 static void vrf_rt6_release(struct net_vrf *vrf)
383 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
384 struct rt6_info *rt6_local = rtnl_dereference(vrf->rt6_local);
386 RCU_INIT_POINTER(vrf->rt6, NULL);
387 RCU_INIT_POINTER(vrf->rt6_local, NULL);
391 dst_release(&rt6->dst);
394 if (rt6_local->rt6i_idev)
395 in6_dev_put(rt6_local->rt6i_idev);
397 dst_release(&rt6_local->dst);
401 static int vrf_rt6_create(struct net_device *dev)
403 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE;
404 struct net_vrf *vrf = netdev_priv(dev);
405 struct net *net = dev_net(dev);
406 struct fib6_table *rt6i_table;
407 struct rt6_info *rt6, *rt6_local;
410 /* IPv6 can be CONFIG enabled and then disabled runtime */
411 if (!ipv6_mod_enabled())
414 rt6i_table = fib6_new_table(net, vrf->tb_id);
418 /* create a dst for routing packets out a VRF device */
419 rt6 = ip6_dst_alloc(net, dev, flags);
425 rt6->rt6i_table = rt6i_table;
426 rt6->dst.output = vrf_output6;
428 /* create a dst for local routing - packets sent locally
429 * to local address via the VRF device as a loopback
431 rt6_local = ip6_dst_alloc(net, dev, flags);
433 dst_release(&rt6->dst);
437 dst_hold(&rt6_local->dst);
439 rt6_local->rt6i_idev = in6_dev_get(dev);
440 rt6_local->rt6i_flags = RTF_UP | RTF_NONEXTHOP | RTF_LOCAL;
441 rt6_local->rt6i_table = rt6i_table;
442 rt6_local->dst.input = ip6_input;
444 rcu_assign_pointer(vrf->rt6, rt6);
445 rcu_assign_pointer(vrf->rt6_local, rt6_local);
452 static void vrf_rt6_release(struct net_vrf *vrf)
456 static int vrf_rt6_create(struct net_device *dev)
462 /* modelled after ip_finish_output2 */
463 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
465 struct dst_entry *dst = skb_dst(skb);
466 struct rtable *rt = (struct rtable *)dst;
467 struct net_device *dev = dst->dev;
468 unsigned int hh_len = LL_RESERVED_SPACE(dev);
469 struct neighbour *neigh;
473 /* Be paranoid, rather than too clever. */
474 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
475 struct sk_buff *skb2;
477 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
483 skb_set_owner_w(skb2, skb->sk);
491 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
492 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
493 if (unlikely(!neigh))
494 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
496 ret = dst_neigh_output(dst, neigh, skb);
498 rcu_read_unlock_bh();
500 if (unlikely(ret < 0))
501 vrf_tx_error(skb->dev, skb);
505 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
507 struct net_device *dev = skb_dst(skb)->dev;
509 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
512 skb->protocol = htons(ETH_P_IP);
514 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
515 net, sk, skb, NULL, dev,
517 !(IPCB(skb)->flags & IPSKB_REROUTED));
521 static void vrf_rtable_release(struct net_vrf *vrf)
523 struct rtable *rth = rtnl_dereference(vrf->rth);
524 struct rtable *rth_local = rtnl_dereference(vrf->rth_local);
526 RCU_INIT_POINTER(vrf->rth, NULL);
527 RCU_INIT_POINTER(vrf->rth_local, NULL);
531 dst_release(&rth->dst);
534 dst_release(&rth_local->dst);
537 static int vrf_rtable_create(struct net_device *dev)
539 struct net_vrf *vrf = netdev_priv(dev);
540 struct rtable *rth, *rth_local;
542 if (!fib_new_table(dev_net(dev), vrf->tb_id))
545 /* create a dst for routing packets out through a VRF device */
546 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
550 /* create a dst for local ingress routing - packets sent locally
551 * to local address via the VRF device as a loopback
553 rth_local = rt_dst_alloc(dev, RTCF_LOCAL, RTN_LOCAL, 1, 1, 0);
555 dst_release(&rth->dst);
559 rth->dst.output = vrf_output;
560 rth->rt_table_id = vrf->tb_id;
562 rth_local->rt_table_id = vrf->tb_id;
564 rcu_assign_pointer(vrf->rth, rth);
565 rcu_assign_pointer(vrf->rth_local, rth_local);
570 /**************************** device handling ********************/
572 /* cycle interface to flush neighbor cache and move routes across tables */
573 static void cycle_netdev(struct net_device *dev)
575 unsigned int flags = dev->flags;
578 if (!netif_running(dev))
581 ret = dev_change_flags(dev, flags & ~IFF_UP);
583 ret = dev_change_flags(dev, flags);
587 "Failed to cycle device %s; route tables might be wrong!\n",
592 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
596 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
600 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
601 cycle_netdev(port_dev);
606 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
608 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
611 return do_vrf_add_slave(dev, port_dev);
614 /* inverse of do_vrf_add_slave */
615 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
617 netdev_upper_dev_unlink(port_dev, dev);
618 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
620 cycle_netdev(port_dev);
625 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
627 return do_vrf_del_slave(dev, port_dev);
630 static void vrf_dev_uninit(struct net_device *dev)
632 struct net_vrf *vrf = netdev_priv(dev);
633 struct net_device *port_dev;
634 struct list_head *iter;
636 vrf_rtable_release(vrf);
637 vrf_rt6_release(vrf);
639 netdev_for_each_lower_dev(dev, port_dev, iter)
640 vrf_del_slave(dev, port_dev);
642 free_percpu(dev->dstats);
646 static int vrf_dev_init(struct net_device *dev)
648 struct net_vrf *vrf = netdev_priv(dev);
650 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
654 /* create the default dst which points back to us */
655 if (vrf_rtable_create(dev) != 0)
658 if (vrf_rt6_create(dev) != 0)
661 dev->flags = IFF_MASTER | IFF_NOARP;
663 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
664 dev->mtu = 64 * 1024;
666 /* similarly, oper state is irrelevant; set to up to avoid confusion */
667 dev->operstate = IF_OPER_UP;
668 netdev_lockdep_set_classes(dev);
672 vrf_rtable_release(vrf);
674 free_percpu(dev->dstats);
680 static const struct net_device_ops vrf_netdev_ops = {
681 .ndo_init = vrf_dev_init,
682 .ndo_uninit = vrf_dev_uninit,
683 .ndo_start_xmit = vrf_xmit,
684 .ndo_get_stats64 = vrf_get_stats64,
685 .ndo_add_slave = vrf_add_slave,
686 .ndo_del_slave = vrf_del_slave,
689 static u32 vrf_fib_table(const struct net_device *dev)
691 struct net_vrf *vrf = netdev_priv(dev);
696 static struct rtable *vrf_get_rtable(const struct net_device *dev,
697 const struct flowi4 *fl4)
699 struct rtable *rth = NULL;
701 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
702 struct net_vrf *vrf = netdev_priv(dev);
706 rth = rcu_dereference(vrf->rth);
716 /* called under rcu_read_lock */
717 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
719 struct fib_result res = { .tclassid = 0 };
720 struct net *net = dev_net(dev);
721 u32 orig_tos = fl4->flowi4_tos;
722 u8 flags = fl4->flowi4_flags;
723 u8 scope = fl4->flowi4_scope;
724 u8 tos = RT_FL_TOS(fl4);
727 if (unlikely(!fl4->daddr))
730 fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
731 fl4->flowi4_iif = LOOPBACK_IFINDEX;
732 /* make sure oif is set to VRF device for lookup */
733 fl4->flowi4_oif = dev->ifindex;
734 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
735 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
736 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
738 rc = fib_lookup(net, fl4, &res, 0);
740 if (res.type == RTN_LOCAL)
741 fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
743 fib_select_path(net, &res, fl4, -1);
746 fl4->flowi4_flags = flags;
747 fl4->flowi4_tos = orig_tos;
748 fl4->flowi4_scope = scope;
753 #if IS_ENABLED(CONFIG_IPV6)
754 /* neighbor handling is done with actual device; do not want
755 * to flip skb->dev for those ndisc packets. This really fails
756 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
759 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
761 const struct ipv6hdr *iph = ipv6_hdr(skb);
764 if (iph->nexthdr == NEXTHDR_ICMP) {
765 const struct icmp6hdr *icmph;
766 struct icmp6hdr _icmph;
768 icmph = skb_header_pointer(skb, sizeof(*iph),
769 sizeof(_icmph), &_icmph);
773 switch (icmph->icmp6_type) {
774 case NDISC_ROUTER_SOLICITATION:
775 case NDISC_ROUTER_ADVERTISEMENT:
776 case NDISC_NEIGHBOUR_SOLICITATION:
777 case NDISC_NEIGHBOUR_ADVERTISEMENT:
788 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
791 /* loopback traffic; do not push through packet taps again.
792 * Reset pkt_type for upper layers to process skb
794 if (skb->pkt_type == PACKET_LOOPBACK) {
796 skb->skb_iif = vrf_dev->ifindex;
797 skb->pkt_type = PACKET_HOST;
801 /* if packet is NDISC keep the ingress interface */
802 if (!ipv6_ndisc_frame(skb)) {
804 skb->skb_iif = vrf_dev->ifindex;
806 skb_push(skb, skb->mac_len);
807 dev_queue_xmit_nit(skb, vrf_dev);
808 skb_pull(skb, skb->mac_len);
810 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
818 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
825 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
829 skb->skb_iif = vrf_dev->ifindex;
831 /* loopback traffic; do not push through packet taps again.
832 * Reset pkt_type for upper layers to process skb
834 if (skb->pkt_type == PACKET_LOOPBACK) {
835 skb->pkt_type = PACKET_HOST;
839 skb_push(skb, skb->mac_len);
840 dev_queue_xmit_nit(skb, vrf_dev);
841 skb_pull(skb, skb->mac_len);
847 /* called with rcu lock held */
848 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
854 return vrf_ip_rcv(vrf_dev, skb);
856 return vrf_ip6_rcv(vrf_dev, skb);
862 #if IS_ENABLED(CONFIG_IPV6)
863 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
866 struct dst_entry *dst = NULL;
868 if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
869 struct net_vrf *vrf = netdev_priv(dev);
874 rt = rcu_dereference(vrf->rt6);
887 static const struct l3mdev_ops vrf_l3mdev_ops = {
888 .l3mdev_fib_table = vrf_fib_table,
889 .l3mdev_get_rtable = vrf_get_rtable,
890 .l3mdev_get_saddr = vrf_get_saddr,
891 .l3mdev_l3_rcv = vrf_l3_rcv,
892 #if IS_ENABLED(CONFIG_IPV6)
893 .l3mdev_get_rt6_dst = vrf_get_rt6_dst,
897 static void vrf_get_drvinfo(struct net_device *dev,
898 struct ethtool_drvinfo *info)
900 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
901 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
904 static const struct ethtool_ops vrf_ethtool_ops = {
905 .get_drvinfo = vrf_get_drvinfo,
908 static inline size_t vrf_fib_rule_nl_size(void)
912 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
913 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
914 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
919 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
921 struct fib_rule_hdr *frh;
922 struct nlmsghdr *nlh;
926 if (family == AF_INET6 && !ipv6_mod_enabled())
929 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
933 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
935 goto nla_put_failure;
937 /* rule only needs to appear once */
938 nlh->nlmsg_flags &= NLM_F_EXCL;
940 frh = nlmsg_data(nlh);
941 memset(frh, 0, sizeof(*frh));
942 frh->family = family;
943 frh->action = FR_ACT_TO_TBL;
945 if (nla_put_u32(skb, FRA_L3MDEV, 1))
946 goto nla_put_failure;
948 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
949 goto nla_put_failure;
953 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
954 skb->sk = dev_net(dev)->rtnl;
956 err = fib_nl_newrule(skb, nlh);
960 err = fib_nl_delrule(skb, nlh);
974 static int vrf_add_fib_rules(const struct net_device *dev)
978 err = vrf_fib_rule(dev, AF_INET, true);
982 err = vrf_fib_rule(dev, AF_INET6, true);
989 vrf_fib_rule(dev, AF_INET, false);
992 netdev_err(dev, "Failed to add FIB rules.\n");
996 static void vrf_setup(struct net_device *dev)
1000 /* Initialize the device structure. */
1001 dev->netdev_ops = &vrf_netdev_ops;
1002 dev->l3mdev_ops = &vrf_l3mdev_ops;
1003 dev->ethtool_ops = &vrf_ethtool_ops;
1004 dev->destructor = free_netdev;
1006 /* Fill in device structure with ethernet-generic values. */
1007 eth_hw_addr_random(dev);
1009 /* don't acquire vrf device's netif_tx_lock when transmitting */
1010 dev->features |= NETIF_F_LLTX;
1012 /* don't allow vrf devices to change network namespaces. */
1013 dev->features |= NETIF_F_NETNS_LOCAL;
1016 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
1018 if (tb[IFLA_ADDRESS]) {
1019 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1021 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1022 return -EADDRNOTAVAIL;
1027 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1029 unregister_netdevice_queue(dev, head);
1032 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1033 struct nlattr *tb[], struct nlattr *data[])
1035 struct net_vrf *vrf = netdev_priv(dev);
1038 if (!data || !data[IFLA_VRF_TABLE])
1041 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1043 dev->priv_flags |= IFF_L3MDEV_MASTER;
1045 err = register_netdevice(dev);
1049 if (add_fib_rules) {
1050 err = vrf_add_fib_rules(dev);
1052 unregister_netdevice(dev);
1055 add_fib_rules = false;
1062 static size_t vrf_nl_getsize(const struct net_device *dev)
1064 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1067 static int vrf_fillinfo(struct sk_buff *skb,
1068 const struct net_device *dev)
1070 struct net_vrf *vrf = netdev_priv(dev);
1072 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1075 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1076 const struct net_device *slave_dev)
1078 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1081 static int vrf_fill_slave_info(struct sk_buff *skb,
1082 const struct net_device *vrf_dev,
1083 const struct net_device *slave_dev)
1085 struct net_vrf *vrf = netdev_priv(vrf_dev);
1087 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1093 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1094 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1097 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1099 .priv_size = sizeof(struct net_vrf),
1101 .get_size = vrf_nl_getsize,
1102 .policy = vrf_nl_policy,
1103 .validate = vrf_validate,
1104 .fill_info = vrf_fillinfo,
1106 .get_slave_size = vrf_get_slave_size,
1107 .fill_slave_info = vrf_fill_slave_info,
1109 .newlink = vrf_newlink,
1110 .dellink = vrf_dellink,
1112 .maxtype = IFLA_VRF_MAX,
1115 static int vrf_device_event(struct notifier_block *unused,
1116 unsigned long event, void *ptr)
1118 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1120 /* only care about unregister events to drop slave references */
1121 if (event == NETDEV_UNREGISTER) {
1122 struct net_device *vrf_dev;
1124 if (!netif_is_l3_slave(dev))
1127 vrf_dev = netdev_master_upper_dev_get(dev);
1128 vrf_del_slave(vrf_dev, dev);
1134 static struct notifier_block vrf_notifier_block __read_mostly = {
1135 .notifier_call = vrf_device_event,
1138 static int __init vrf_init_module(void)
1142 register_netdevice_notifier(&vrf_notifier_block);
1144 rc = rtnl_link_register(&vrf_link_ops);
1151 unregister_netdevice_notifier(&vrf_notifier_block);
1155 module_init(vrf_init_module);
1156 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1157 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1158 MODULE_LICENSE("GPL");
1159 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1160 MODULE_VERSION(DRV_VERSION);