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>
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);
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;
171 struct net *net = dev_net(skb->dev);
173 int ret = NET_XMIT_DROP;
174 struct dst_entry *dst;
175 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
177 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
182 memset(&fl6, 0, sizeof(fl6));
183 /* needed to match OIF rule */
184 fl6.flowi6_oif = dev->ifindex;
185 fl6.flowi6_iif = LOOPBACK_IFINDEX;
186 fl6.daddr = iph->daddr;
187 fl6.saddr = iph->saddr;
188 fl6.flowlabel = ip6_flowinfo(iph);
189 fl6.flowi6_mark = skb->mark;
190 fl6.flowi6_proto = iph->nexthdr;
191 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
193 dst = ip6_route_output(net, NULL, &fl6);
199 /* if dst.dev is loopback or the VRF device again this is locally
200 * originated traffic destined to a local address. Short circuit
204 return vrf_local_xmit(skb, dev, dst);
206 skb_dst_set(skb, dst);
208 /* strip the ethernet header added for pass through VRF device */
209 __skb_pull(skb, skb_network_offset(skb));
211 ret = vrf_ip6_local_out(net, skb->sk, skb);
212 if (unlikely(net_xmit_eval(ret)))
213 dev->stats.tx_errors++;
215 ret = NET_XMIT_SUCCESS;
219 vrf_tx_error(dev, skb);
220 return NET_XMIT_DROP;
223 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
224 struct net_device *dev)
226 vrf_tx_error(dev, skb);
227 return NET_XMIT_DROP;
231 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
232 static int vrf_ip_local_out(struct net *net, struct sock *sk,
237 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
238 skb, NULL, skb_dst(skb)->dev, dst_output);
239 if (likely(err == 1))
240 err = dst_output(net, sk, skb);
245 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
246 struct net_device *vrf_dev)
249 int ret = NET_XMIT_DROP;
251 struct net *net = dev_net(vrf_dev);
254 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
259 memset(&fl4, 0, sizeof(fl4));
260 /* needed to match OIF rule */
261 fl4.flowi4_oif = vrf_dev->ifindex;
262 fl4.flowi4_iif = LOOPBACK_IFINDEX;
263 fl4.flowi4_tos = RT_TOS(ip4h->tos);
264 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
265 fl4.flowi4_proto = ip4h->protocol;
266 fl4.daddr = ip4h->daddr;
267 fl4.saddr = ip4h->saddr;
269 rt = ip_route_output_flow(net, &fl4, NULL);
275 /* if dst.dev is loopback or the VRF device again this is locally
276 * originated traffic destined to a local address. Short circuit
279 if (rt->dst.dev == vrf_dev)
280 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
282 skb_dst_set(skb, &rt->dst);
284 /* strip the ethernet header added for pass through VRF device */
285 __skb_pull(skb, skb_network_offset(skb));
288 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
292 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
293 if (unlikely(net_xmit_eval(ret)))
294 vrf_dev->stats.tx_errors++;
296 ret = NET_XMIT_SUCCESS;
301 vrf_tx_error(vrf_dev, skb);
305 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
307 switch (skb->protocol) {
308 case htons(ETH_P_IP):
309 return vrf_process_v4_outbound(skb, dev);
310 case htons(ETH_P_IPV6):
311 return vrf_process_v6_outbound(skb, dev);
313 vrf_tx_error(dev, skb);
314 return NET_XMIT_DROP;
318 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
321 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
323 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
324 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
326 u64_stats_update_begin(&dstats->syncp);
328 dstats->tx_bytes += len;
329 u64_stats_update_end(&dstats->syncp);
331 this_cpu_inc(dev->dstats->tx_drps);
337 static void vrf_finish_direct(struct sk_buff *skb)
339 struct net_device *vrf_dev = skb->dev;
341 if (!list_empty(&vrf_dev->ptype_all) &&
342 likely(skb_headroom(skb) >= ETH_HLEN)) {
343 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
345 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
346 eth_zero_addr(eth->h_dest);
347 eth->h_proto = skb->protocol;
350 dev_queue_xmit_nit(skb, vrf_dev);
351 rcu_read_unlock_bh();
353 skb_pull(skb, ETH_HLEN);
356 /* reset skb device */
360 #if IS_ENABLED(CONFIG_IPV6)
361 /* modelled after ip6_finish_output2 */
362 static int vrf_finish_output6(struct net *net, struct sock *sk,
365 struct dst_entry *dst = skb_dst(skb);
366 struct net_device *dev = dst->dev;
367 struct neighbour *neigh;
368 struct in6_addr *nexthop;
373 skb->protocol = htons(ETH_P_IPV6);
377 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
378 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
379 if (unlikely(!neigh))
380 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
381 if (!IS_ERR(neigh)) {
382 sock_confirm_neigh(skb, neigh);
383 ret = neigh_output(neigh, skb);
384 rcu_read_unlock_bh();
387 rcu_read_unlock_bh();
389 IP6_INC_STATS(dev_net(dst->dev),
390 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
395 /* modelled after ip6_output */
396 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
398 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
399 net, sk, skb, NULL, skb_dst(skb)->dev,
401 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
404 /* set dst on skb to send packet to us via dev_xmit path. Allows
405 * packet to go through device based features such as qdisc, netfilter
406 * hooks and packet sockets with skb->dev set to vrf device.
408 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
411 struct net_vrf *vrf = netdev_priv(vrf_dev);
412 struct dst_entry *dst = NULL;
413 struct rt6_info *rt6;
417 rt6 = rcu_dereference(vrf->rt6);
425 if (unlikely(!dst)) {
426 vrf_tx_error(vrf_dev, skb);
431 skb_dst_set(skb, dst);
436 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
439 vrf_finish_direct(skb);
441 return vrf_ip6_local_out(net, sk, skb);
444 static int vrf_output6_direct(struct net *net, struct sock *sk,
449 skb->protocol = htons(ETH_P_IPV6);
451 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
452 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
453 NULL, skb->dev, vrf_output6_direct_finish);
455 if (likely(err == 1))
456 vrf_finish_direct(skb);
461 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
466 err = vrf_output6_direct(net, sk, skb);
467 if (likely(err == 1))
468 err = vrf_ip6_local_out(net, sk, skb);
473 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
477 struct net *net = dev_net(vrf_dev);
482 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
483 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
485 if (likely(err == 1))
486 err = vrf_output6_direct(net, sk, skb);
488 if (likely(err == 1))
494 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
498 /* don't divert link scope packets */
499 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
502 if (qdisc_tx_is_default(vrf_dev) ||
503 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
504 return vrf_ip6_out_direct(vrf_dev, sk, skb);
506 return vrf_ip6_out_redirect(vrf_dev, skb);
510 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
512 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
513 struct net *net = dev_net(dev);
514 struct dst_entry *dst;
516 RCU_INIT_POINTER(vrf->rt6, NULL);
519 /* move dev in dst's to loopback so this VRF device can be deleted
520 * - based on dst_ifdown
525 dst->dev = net->loopback_dev;
531 static int vrf_rt6_create(struct net_device *dev)
533 int flags = DST_HOST | DST_NOPOLICY | DST_NOXFRM;
534 struct net_vrf *vrf = netdev_priv(dev);
535 struct net *net = dev_net(dev);
536 struct fib6_table *rt6i_table;
537 struct rt6_info *rt6;
540 /* IPv6 can be CONFIG enabled and then disabled runtime */
541 if (!ipv6_mod_enabled())
544 rt6i_table = fib6_new_table(net, vrf->tb_id);
548 /* create a dst for routing packets out a VRF device */
549 rt6 = ip6_dst_alloc(net, dev, flags);
553 rt6->rt6i_table = rt6i_table;
554 rt6->dst.output = vrf_output6;
556 rcu_assign_pointer(vrf->rt6, rt6);
563 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
570 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
574 static int vrf_rt6_create(struct net_device *dev)
580 /* modelled after ip_finish_output2 */
581 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
583 struct dst_entry *dst = skb_dst(skb);
584 struct rtable *rt = (struct rtable *)dst;
585 struct net_device *dev = dst->dev;
586 unsigned int hh_len = LL_RESERVED_SPACE(dev);
587 struct neighbour *neigh;
593 /* Be paranoid, rather than too clever. */
594 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
595 struct sk_buff *skb2;
597 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
603 skb_set_owner_w(skb2, skb->sk);
611 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
612 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
613 if (unlikely(!neigh))
614 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
615 if (!IS_ERR(neigh)) {
616 sock_confirm_neigh(skb, neigh);
617 ret = neigh_output(neigh, skb);
618 rcu_read_unlock_bh();
622 rcu_read_unlock_bh();
624 vrf_tx_error(skb->dev, skb);
628 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
630 struct net_device *dev = skb_dst(skb)->dev;
632 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
635 skb->protocol = htons(ETH_P_IP);
637 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
638 net, sk, skb, NULL, dev,
640 !(IPCB(skb)->flags & IPSKB_REROUTED));
643 /* set dst on skb to send packet to us via dev_xmit path. Allows
644 * packet to go through device based features such as qdisc, netfilter
645 * hooks and packet sockets with skb->dev set to vrf device.
647 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
650 struct net_vrf *vrf = netdev_priv(vrf_dev);
651 struct dst_entry *dst = NULL;
656 rth = rcu_dereference(vrf->rth);
664 if (unlikely(!dst)) {
665 vrf_tx_error(vrf_dev, skb);
670 skb_dst_set(skb, dst);
675 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
678 vrf_finish_direct(skb);
680 return vrf_ip_local_out(net, sk, skb);
683 static int vrf_output_direct(struct net *net, struct sock *sk,
688 skb->protocol = htons(ETH_P_IP);
690 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
691 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
692 NULL, skb->dev, vrf_output_direct_finish);
694 if (likely(err == 1))
695 vrf_finish_direct(skb);
700 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
705 err = vrf_output_direct(net, sk, skb);
706 if (likely(err == 1))
707 err = vrf_ip_local_out(net, sk, skb);
712 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
716 struct net *net = dev_net(vrf_dev);
721 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
722 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
724 if (likely(err == 1))
725 err = vrf_output_direct(net, sk, skb);
727 if (likely(err == 1))
733 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
737 /* don't divert multicast or local broadcast */
738 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
739 ipv4_is_lbcast(ip_hdr(skb)->daddr))
742 if (qdisc_tx_is_default(vrf_dev) ||
743 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
744 return vrf_ip_out_direct(vrf_dev, sk, skb);
746 return vrf_ip_out_redirect(vrf_dev, skb);
749 /* called with rcu lock held */
750 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
757 return vrf_ip_out(vrf_dev, sk, skb);
759 return vrf_ip6_out(vrf_dev, sk, skb);
766 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
768 struct rtable *rth = rtnl_dereference(vrf->rth);
769 struct net *net = dev_net(dev);
770 struct dst_entry *dst;
772 RCU_INIT_POINTER(vrf->rth, NULL);
775 /* move dev in dst's to loopback so this VRF device can be deleted
776 * - based on dst_ifdown
781 dst->dev = net->loopback_dev;
787 static int vrf_rtable_create(struct net_device *dev)
789 struct net_vrf *vrf = netdev_priv(dev);
792 if (!fib_new_table(dev_net(dev), vrf->tb_id))
795 /* create a dst for routing packets out through a VRF device */
796 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
800 rth->dst.output = vrf_output;
801 rth->rt_table_id = vrf->tb_id;
803 rcu_assign_pointer(vrf->rth, rth);
808 /**************************** device handling ********************/
810 /* cycle interface to flush neighbor cache and move routes across tables */
811 static void cycle_netdev(struct net_device *dev)
813 unsigned int flags = dev->flags;
816 if (!netif_running(dev))
819 ret = dev_change_flags(dev, flags & ~IFF_UP);
821 ret = dev_change_flags(dev, flags);
825 "Failed to cycle device %s; route tables might be wrong!\n",
830 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
834 /* do not allow loopback device to be enslaved to a VRF.
835 * The vrf device acts as the loopback for the vrf.
837 if (port_dev == dev_net(dev)->loopback_dev)
840 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
841 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
845 cycle_netdev(port_dev);
850 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
854 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
856 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
859 return do_vrf_add_slave(dev, port_dev);
862 /* inverse of do_vrf_add_slave */
863 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
865 netdev_upper_dev_unlink(port_dev, dev);
866 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
868 cycle_netdev(port_dev);
873 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
875 return do_vrf_del_slave(dev, port_dev);
878 static void vrf_dev_uninit(struct net_device *dev)
880 struct net_vrf *vrf = netdev_priv(dev);
882 vrf_rtable_release(dev, vrf);
883 vrf_rt6_release(dev, vrf);
885 free_percpu(dev->dstats);
889 static int vrf_dev_init(struct net_device *dev)
891 struct net_vrf *vrf = netdev_priv(dev);
893 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
897 /* create the default dst which points back to us */
898 if (vrf_rtable_create(dev) != 0)
901 if (vrf_rt6_create(dev) != 0)
904 dev->flags = IFF_MASTER | IFF_NOARP;
906 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
907 dev->mtu = 64 * 1024;
909 /* similarly, oper state is irrelevant; set to up to avoid confusion */
910 dev->operstate = IF_OPER_UP;
911 netdev_lockdep_set_classes(dev);
915 vrf_rtable_release(dev, vrf);
917 free_percpu(dev->dstats);
923 static const struct net_device_ops vrf_netdev_ops = {
924 .ndo_init = vrf_dev_init,
925 .ndo_uninit = vrf_dev_uninit,
926 .ndo_start_xmit = vrf_xmit,
927 .ndo_get_stats64 = vrf_get_stats64,
928 .ndo_add_slave = vrf_add_slave,
929 .ndo_del_slave = vrf_del_slave,
932 static u32 vrf_fib_table(const struct net_device *dev)
934 struct net_vrf *vrf = netdev_priv(dev);
939 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
945 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
947 struct net_device *dev)
949 struct net *net = dev_net(dev);
951 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
952 skb = NULL; /* kfree_skb(skb) handled by nf code */
957 #if IS_ENABLED(CONFIG_IPV6)
958 /* neighbor handling is done with actual device; do not want
959 * to flip skb->dev for those ndisc packets. This really fails
960 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
963 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
965 const struct ipv6hdr *iph = ipv6_hdr(skb);
968 if (iph->nexthdr == NEXTHDR_ICMP) {
969 const struct icmp6hdr *icmph;
970 struct icmp6hdr _icmph;
972 icmph = skb_header_pointer(skb, sizeof(*iph),
973 sizeof(_icmph), &_icmph);
977 switch (icmph->icmp6_type) {
978 case NDISC_ROUTER_SOLICITATION:
979 case NDISC_ROUTER_ADVERTISEMENT:
980 case NDISC_NEIGHBOUR_SOLICITATION:
981 case NDISC_NEIGHBOUR_ADVERTISEMENT:
992 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
993 const struct net_device *dev,
998 struct net_vrf *vrf = netdev_priv(dev);
999 struct fib6_table *table = NULL;
1000 struct rt6_info *rt6;
1004 /* fib6_table does not have a refcnt and can not be freed */
1005 rt6 = rcu_dereference(vrf->rt6);
1007 table = rt6->rt6i_table;
1014 return ip6_pol_route(net, table, ifindex, fl6, flags);
1017 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1020 const struct ipv6hdr *iph = ipv6_hdr(skb);
1021 struct flowi6 fl6 = {
1022 .flowi6_iif = ifindex,
1023 .flowi6_mark = skb->mark,
1024 .flowi6_proto = iph->nexthdr,
1025 .daddr = iph->daddr,
1026 .saddr = iph->saddr,
1027 .flowlabel = ip6_flowinfo(iph),
1029 struct net *net = dev_net(vrf_dev);
1030 struct rt6_info *rt6;
1032 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex,
1033 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1037 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1040 skb_dst_set(skb, &rt6->dst);
1043 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1044 struct sk_buff *skb)
1046 int orig_iif = skb->skb_iif;
1049 /* loopback traffic; do not push through packet taps again.
1050 * Reset pkt_type for upper layers to process skb
1052 if (skb->pkt_type == PACKET_LOOPBACK) {
1054 skb->skb_iif = vrf_dev->ifindex;
1055 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1056 skb->pkt_type = PACKET_HOST;
1060 /* if packet is NDISC or addressed to multicast or link-local
1061 * then keep the ingress interface
1063 need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1064 if (!ipv6_ndisc_frame(skb) && !need_strict) {
1065 vrf_rx_stats(vrf_dev, skb->len);
1067 skb->skb_iif = vrf_dev->ifindex;
1069 if (!list_empty(&vrf_dev->ptype_all)) {
1070 skb_push(skb, skb->mac_len);
1071 dev_queue_xmit_nit(skb, vrf_dev);
1072 skb_pull(skb, skb->mac_len);
1075 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1079 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1081 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1087 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1088 struct sk_buff *skb)
1094 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1095 struct sk_buff *skb)
1098 skb->skb_iif = vrf_dev->ifindex;
1099 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1101 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1104 /* loopback traffic; do not push through packet taps again.
1105 * Reset pkt_type for upper layers to process skb
1107 if (skb->pkt_type == PACKET_LOOPBACK) {
1108 skb->pkt_type = PACKET_HOST;
1112 vrf_rx_stats(vrf_dev, skb->len);
1114 if (!list_empty(&vrf_dev->ptype_all)) {
1115 skb_push(skb, skb->mac_len);
1116 dev_queue_xmit_nit(skb, vrf_dev);
1117 skb_pull(skb, skb->mac_len);
1120 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1125 /* called with rcu lock held */
1126 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1127 struct sk_buff *skb,
1132 return vrf_ip_rcv(vrf_dev, skb);
1134 return vrf_ip6_rcv(vrf_dev, skb);
1140 #if IS_ENABLED(CONFIG_IPV6)
1141 /* send to link-local or multicast address via interface enslaved to
1142 * VRF device. Force lookup to VRF table without changing flow struct
1144 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1147 struct net *net = dev_net(dev);
1148 int flags = RT6_LOOKUP_F_IFACE;
1149 struct dst_entry *dst = NULL;
1150 struct rt6_info *rt;
1152 /* VRF device does not have a link-local address and
1153 * sending packets to link-local or mcast addresses over
1154 * a VRF device does not make sense
1156 if (fl6->flowi6_oif == dev->ifindex) {
1157 dst = &net->ipv6.ip6_null_entry->dst;
1162 if (!ipv6_addr_any(&fl6->saddr))
1163 flags |= RT6_LOOKUP_F_HAS_SADDR;
1165 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, flags);
1173 static const struct l3mdev_ops vrf_l3mdev_ops = {
1174 .l3mdev_fib_table = vrf_fib_table,
1175 .l3mdev_l3_rcv = vrf_l3_rcv,
1176 .l3mdev_l3_out = vrf_l3_out,
1177 #if IS_ENABLED(CONFIG_IPV6)
1178 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1182 static void vrf_get_drvinfo(struct net_device *dev,
1183 struct ethtool_drvinfo *info)
1185 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1186 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1189 static const struct ethtool_ops vrf_ethtool_ops = {
1190 .get_drvinfo = vrf_get_drvinfo,
1193 static inline size_t vrf_fib_rule_nl_size(void)
1197 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1198 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1199 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1204 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1206 struct fib_rule_hdr *frh;
1207 struct nlmsghdr *nlh;
1208 struct sk_buff *skb;
1211 if (family == AF_INET6 && !ipv6_mod_enabled())
1214 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1218 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1220 goto nla_put_failure;
1222 /* rule only needs to appear once */
1223 nlh->nlmsg_flags |= NLM_F_EXCL;
1225 frh = nlmsg_data(nlh);
1226 memset(frh, 0, sizeof(*frh));
1227 frh->family = family;
1228 frh->action = FR_ACT_TO_TBL;
1230 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1231 goto nla_put_failure;
1233 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1234 goto nla_put_failure;
1236 nlmsg_end(skb, nlh);
1238 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1239 skb->sk = dev_net(dev)->rtnl;
1241 err = fib_nl_newrule(skb, nlh, NULL);
1245 err = fib_nl_delrule(skb, nlh, NULL);
1259 static int vrf_add_fib_rules(const struct net_device *dev)
1263 err = vrf_fib_rule(dev, AF_INET, true);
1267 err = vrf_fib_rule(dev, AF_INET6, true);
1271 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1272 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1279 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1281 vrf_fib_rule(dev, AF_INET6, false);
1285 vrf_fib_rule(dev, AF_INET, false);
1288 netdev_err(dev, "Failed to add FIB rules.\n");
1292 static void vrf_setup(struct net_device *dev)
1296 /* Initialize the device structure. */
1297 dev->netdev_ops = &vrf_netdev_ops;
1298 dev->l3mdev_ops = &vrf_l3mdev_ops;
1299 dev->ethtool_ops = &vrf_ethtool_ops;
1300 dev->needs_free_netdev = true;
1302 /* Fill in device structure with ethernet-generic values. */
1303 eth_hw_addr_random(dev);
1305 /* don't acquire vrf device's netif_tx_lock when transmitting */
1306 dev->features |= NETIF_F_LLTX;
1308 /* don't allow vrf devices to change network namespaces. */
1309 dev->features |= NETIF_F_NETNS_LOCAL;
1311 /* does not make sense for a VLAN to be added to a vrf device */
1312 dev->features |= NETIF_F_VLAN_CHALLENGED;
1314 /* enable offload features */
1315 dev->features |= NETIF_F_GSO_SOFTWARE;
1316 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM;
1317 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1319 dev->hw_features = dev->features;
1320 dev->hw_enc_features = dev->features;
1322 /* default to no qdisc; user can add if desired */
1323 dev->priv_flags |= IFF_NO_QUEUE;
1326 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1327 struct netlink_ext_ack *extack)
1329 if (tb[IFLA_ADDRESS]) {
1330 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1331 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1334 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1335 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1336 return -EADDRNOTAVAIL;
1342 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1344 struct net_device *port_dev;
1345 struct list_head *iter;
1347 netdev_for_each_lower_dev(dev, port_dev, iter)
1348 vrf_del_slave(dev, port_dev);
1350 unregister_netdevice_queue(dev, head);
1353 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1354 struct nlattr *tb[], struct nlattr *data[],
1355 struct netlink_ext_ack *extack)
1357 struct net_vrf *vrf = netdev_priv(dev);
1358 bool *add_fib_rules;
1362 if (!data || !data[IFLA_VRF_TABLE]) {
1363 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1367 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1368 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1369 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1370 "Invalid VRF table id");
1374 dev->priv_flags |= IFF_L3MDEV_MASTER;
1376 err = register_netdevice(dev);
1381 add_fib_rules = net_generic(net, vrf_net_id);
1382 if (*add_fib_rules) {
1383 err = vrf_add_fib_rules(dev);
1385 unregister_netdevice(dev);
1388 *add_fib_rules = false;
1395 static size_t vrf_nl_getsize(const struct net_device *dev)
1397 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1400 static int vrf_fillinfo(struct sk_buff *skb,
1401 const struct net_device *dev)
1403 struct net_vrf *vrf = netdev_priv(dev);
1405 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1408 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1409 const struct net_device *slave_dev)
1411 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1414 static int vrf_fill_slave_info(struct sk_buff *skb,
1415 const struct net_device *vrf_dev,
1416 const struct net_device *slave_dev)
1418 struct net_vrf *vrf = netdev_priv(vrf_dev);
1420 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1426 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1427 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1430 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1432 .priv_size = sizeof(struct net_vrf),
1434 .get_size = vrf_nl_getsize,
1435 .policy = vrf_nl_policy,
1436 .validate = vrf_validate,
1437 .fill_info = vrf_fillinfo,
1439 .get_slave_size = vrf_get_slave_size,
1440 .fill_slave_info = vrf_fill_slave_info,
1442 .newlink = vrf_newlink,
1443 .dellink = vrf_dellink,
1445 .maxtype = IFLA_VRF_MAX,
1448 static int vrf_device_event(struct notifier_block *unused,
1449 unsigned long event, void *ptr)
1451 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1453 /* only care about unregister events to drop slave references */
1454 if (event == NETDEV_UNREGISTER) {
1455 struct net_device *vrf_dev;
1457 if (!netif_is_l3_slave(dev))
1460 vrf_dev = netdev_master_upper_dev_get(dev);
1461 vrf_del_slave(vrf_dev, dev);
1467 static struct notifier_block vrf_notifier_block __read_mostly = {
1468 .notifier_call = vrf_device_event,
1471 /* Initialize per network namespace state */
1472 static int __net_init vrf_netns_init(struct net *net)
1474 bool *add_fib_rules = net_generic(net, vrf_net_id);
1476 *add_fib_rules = true;
1481 static struct pernet_operations vrf_net_ops __net_initdata = {
1482 .init = vrf_netns_init,
1484 .size = sizeof(bool),
1487 static int __init vrf_init_module(void)
1491 register_netdevice_notifier(&vrf_notifier_block);
1493 rc = register_pernet_subsys(&vrf_net_ops);
1497 rc = rtnl_link_register(&vrf_link_ops);
1499 unregister_pernet_subsys(&vrf_net_ops);
1506 unregister_netdevice_notifier(&vrf_notifier_block);
1510 module_init(vrf_init_module);
1511 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
1512 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
1513 MODULE_LICENSE("GPL");
1514 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
1515 MODULE_VERSION(DRV_VERSION);