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>
39 #define RT_FL_TOS(oldflp4) \
40 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
42 #define DRV_NAME "vrf"
43 #define DRV_VERSION "1.0"
57 struct u64_stats_sync syncp;
60 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
62 vrf_dev->stats.tx_errors++;
66 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
67 struct rtnl_link_stats64 *stats)
71 for_each_possible_cpu(i) {
72 const struct pcpu_dstats *dstats;
73 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
76 dstats = per_cpu_ptr(dev->dstats, i);
78 start = u64_stats_fetch_begin_irq(&dstats->syncp);
79 tbytes = dstats->tx_bytes;
80 tpkts = dstats->tx_pkts;
81 tdrops = dstats->tx_drps;
82 rbytes = dstats->rx_bytes;
83 rpkts = dstats->rx_pkts;
84 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
85 stats->tx_bytes += tbytes;
86 stats->tx_packets += tpkts;
87 stats->tx_dropped += tdrops;
88 stats->rx_bytes += rbytes;
89 stats->rx_packets += rpkts;
94 #if IS_ENABLED(CONFIG_IPV6)
95 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
96 struct net_device *dev)
98 const struct ipv6hdr *iph = ipv6_hdr(skb);
99 struct net *net = dev_net(skb->dev);
100 struct flowi6 fl6 = {
101 /* needed to match OIF rule */
102 .flowi6_oif = dev->ifindex,
103 .flowi6_iif = LOOPBACK_IFINDEX,
106 .flowlabel = ip6_flowinfo(iph),
107 .flowi6_mark = skb->mark,
108 .flowi6_proto = iph->nexthdr,
109 .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF,
111 int ret = NET_XMIT_DROP;
112 struct dst_entry *dst;
113 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
115 dst = ip6_route_output(net, NULL, &fl6);
120 skb_dst_set(skb, dst);
122 ret = ip6_local_out(net, skb->sk, skb);
123 if (unlikely(net_xmit_eval(ret)))
124 dev->stats.tx_errors++;
126 ret = NET_XMIT_SUCCESS;
130 vrf_tx_error(dev, skb);
131 return NET_XMIT_DROP;
134 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
135 struct net_device *dev)
137 vrf_tx_error(dev, skb);
138 return NET_XMIT_DROP;
142 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
143 struct net_device *vrf_dev)
148 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
152 /* TO-DO: what about broadcast ? */
153 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
159 skb_dst_set(skb, &rt->dst);
165 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
166 struct net_device *vrf_dev)
168 struct iphdr *ip4h = ip_hdr(skb);
169 int ret = NET_XMIT_DROP;
170 struct flowi4 fl4 = {
171 /* needed to match OIF rule */
172 .flowi4_oif = vrf_dev->ifindex,
173 .flowi4_iif = LOOPBACK_IFINDEX,
174 .flowi4_tos = RT_TOS(ip4h->tos),
175 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
176 FLOWI_FLAG_SKIP_NH_OIF,
177 .daddr = ip4h->daddr,
180 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
184 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
188 ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
189 if (unlikely(net_xmit_eval(ret)))
190 vrf_dev->stats.tx_errors++;
192 ret = NET_XMIT_SUCCESS;
197 vrf_tx_error(vrf_dev, skb);
201 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
203 /* strip the ethernet header added for pass through VRF device */
204 __skb_pull(skb, skb_network_offset(skb));
206 switch (skb->protocol) {
207 case htons(ETH_P_IP):
208 return vrf_process_v4_outbound(skb, dev);
209 case htons(ETH_P_IPV6):
210 return vrf_process_v6_outbound(skb, dev);
212 vrf_tx_error(dev, skb);
213 return NET_XMIT_DROP;
217 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
219 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
221 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
222 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
224 u64_stats_update_begin(&dstats->syncp);
226 dstats->tx_bytes += skb->len;
227 u64_stats_update_end(&dstats->syncp);
229 this_cpu_inc(dev->dstats->tx_drps);
235 #if IS_ENABLED(CONFIG_IPV6)
236 /* modelled after ip6_finish_output2 */
237 static int vrf_finish_output6(struct net *net, struct sock *sk,
240 struct dst_entry *dst = skb_dst(skb);
241 struct net_device *dev = dst->dev;
242 struct neighbour *neigh;
243 struct in6_addr *nexthop;
246 skb->protocol = htons(ETH_P_IPV6);
250 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
251 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
252 if (unlikely(!neigh))
253 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
254 if (!IS_ERR(neigh)) {
255 ret = dst_neigh_output(dst, neigh, skb);
256 rcu_read_unlock_bh();
259 rcu_read_unlock_bh();
261 IP6_INC_STATS(dev_net(dst->dev),
262 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
267 /* modelled after ip6_output */
268 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
270 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
271 net, sk, skb, NULL, skb_dst(skb)->dev,
273 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
276 static void vrf_rt6_release(struct net_vrf *vrf)
278 dst_release(&vrf->rt6->dst);
282 static int vrf_rt6_create(struct net_device *dev)
284 struct net_vrf *vrf = netdev_priv(dev);
285 struct net *net = dev_net(dev);
286 struct fib6_table *rt6i_table;
287 struct rt6_info *rt6;
290 rt6i_table = fib6_new_table(net, vrf->tb_id);
294 rt6 = ip6_dst_alloc(net, dev,
295 DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE);
301 rt6->rt6i_table = rt6i_table;
302 rt6->dst.output = vrf_output6;
309 static void vrf_rt6_release(struct net_vrf *vrf)
313 static int vrf_rt6_create(struct net_device *dev)
319 /* modelled after ip_finish_output2 */
320 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
322 struct dst_entry *dst = skb_dst(skb);
323 struct rtable *rt = (struct rtable *)dst;
324 struct net_device *dev = dst->dev;
325 unsigned int hh_len = LL_RESERVED_SPACE(dev);
326 struct neighbour *neigh;
330 /* Be paranoid, rather than too clever. */
331 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
332 struct sk_buff *skb2;
334 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
340 skb_set_owner_w(skb2, skb->sk);
348 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
349 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
350 if (unlikely(!neigh))
351 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
353 ret = dst_neigh_output(dst, neigh, skb);
355 rcu_read_unlock_bh();
357 if (unlikely(ret < 0))
358 vrf_tx_error(skb->dev, skb);
362 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
364 struct net_device *dev = skb_dst(skb)->dev;
366 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
369 skb->protocol = htons(ETH_P_IP);
371 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
372 net, sk, skb, NULL, dev,
374 !(IPCB(skb)->flags & IPSKB_REROUTED));
377 static void vrf_rtable_release(struct net_vrf *vrf)
379 struct dst_entry *dst = (struct dst_entry *)vrf->rth;
385 static struct rtable *vrf_rtable_create(struct net_device *dev)
387 struct net_vrf *vrf = netdev_priv(dev);
390 if (!fib_new_table(dev_net(dev), vrf->tb_id))
393 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0);
395 rth->dst.output = vrf_output;
396 rth->rt_table_id = vrf->tb_id;
402 /**************************** device handling ********************/
404 /* cycle interface to flush neighbor cache and move routes across tables */
405 static void cycle_netdev(struct net_device *dev)
407 unsigned int flags = dev->flags;
410 if (!netif_running(dev))
413 ret = dev_change_flags(dev, flags & ~IFF_UP);
415 ret = dev_change_flags(dev, flags);
419 "Failed to cycle device %s; route tables might be wrong!\n",
424 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
428 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL);
432 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
433 cycle_netdev(port_dev);
438 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
440 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
443 return do_vrf_add_slave(dev, port_dev);
446 /* inverse of do_vrf_add_slave */
447 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
449 netdev_upper_dev_unlink(port_dev, dev);
450 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
452 cycle_netdev(port_dev);
457 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
459 return do_vrf_del_slave(dev, port_dev);
462 static void vrf_dev_uninit(struct net_device *dev)
464 struct net_vrf *vrf = netdev_priv(dev);
465 struct net_device *port_dev;
466 struct list_head *iter;
468 vrf_rtable_release(vrf);
469 vrf_rt6_release(vrf);
471 netdev_for_each_lower_dev(dev, port_dev, iter)
472 vrf_del_slave(dev, port_dev);
474 free_percpu(dev->dstats);
478 static int vrf_dev_init(struct net_device *dev)
480 struct net_vrf *vrf = netdev_priv(dev);
482 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
486 /* create the default dst which points back to us */
487 vrf->rth = vrf_rtable_create(dev);
491 if (vrf_rt6_create(dev) != 0)
494 dev->flags = IFF_MASTER | IFF_NOARP;
499 vrf_rtable_release(vrf);
501 free_percpu(dev->dstats);
507 static const struct net_device_ops vrf_netdev_ops = {
508 .ndo_init = vrf_dev_init,
509 .ndo_uninit = vrf_dev_uninit,
510 .ndo_start_xmit = vrf_xmit,
511 .ndo_get_stats64 = vrf_get_stats64,
512 .ndo_add_slave = vrf_add_slave,
513 .ndo_del_slave = vrf_del_slave,
516 static u32 vrf_fib_table(const struct net_device *dev)
518 struct net_vrf *vrf = netdev_priv(dev);
523 static struct rtable *vrf_get_rtable(const struct net_device *dev,
524 const struct flowi4 *fl4)
526 struct rtable *rth = NULL;
528 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
529 struct net_vrf *vrf = netdev_priv(dev);
538 /* called under rcu_read_lock */
539 static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4)
541 struct fib_result res = { .tclassid = 0 };
542 struct net *net = dev_net(dev);
543 u32 orig_tos = fl4->flowi4_tos;
544 u8 flags = fl4->flowi4_flags;
545 u8 scope = fl4->flowi4_scope;
546 u8 tos = RT_FL_TOS(fl4);
549 if (unlikely(!fl4->daddr))
552 fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF;
553 fl4->flowi4_iif = LOOPBACK_IFINDEX;
554 /* make sure oif is set to VRF device for lookup */
555 fl4->flowi4_oif = dev->ifindex;
556 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
557 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
558 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
560 rc = fib_lookup(net, fl4, &res, 0);
562 if (res.type == RTN_LOCAL)
563 fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr;
565 fib_select_path(net, &res, fl4, -1);
568 fl4->flowi4_flags = flags;
569 fl4->flowi4_tos = orig_tos;
570 fl4->flowi4_scope = scope;
575 #if IS_ENABLED(CONFIG_IPV6)
576 /* neighbor handling is done with actual device; do not want
577 * to flip skb->dev for those ndisc packets. This really fails
578 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
581 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
583 const struct ipv6hdr *iph = ipv6_hdr(skb);
586 if (iph->nexthdr == NEXTHDR_ICMP) {
587 const struct icmp6hdr *icmph;
588 struct icmp6hdr _icmph;
590 icmph = skb_header_pointer(skb, sizeof(*iph),
591 sizeof(_icmph), &_icmph);
595 switch (icmph->icmp6_type) {
596 case NDISC_ROUTER_SOLICITATION:
597 case NDISC_ROUTER_ADVERTISEMENT:
598 case NDISC_NEIGHBOUR_SOLICITATION:
599 case NDISC_NEIGHBOUR_ADVERTISEMENT:
610 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
613 /* if packet is NDISC keep the ingress interface */
614 if (!ipv6_ndisc_frame(skb)) {
616 skb->skb_iif = vrf_dev->ifindex;
618 skb_push(skb, skb->mac_len);
619 dev_queue_xmit_nit(skb, vrf_dev);
620 skb_pull(skb, skb->mac_len);
622 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
629 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
636 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
640 skb->skb_iif = vrf_dev->ifindex;
642 skb_push(skb, skb->mac_len);
643 dev_queue_xmit_nit(skb, vrf_dev);
644 skb_pull(skb, skb->mac_len);
649 /* called with rcu lock held */
650 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
656 return vrf_ip_rcv(vrf_dev, skb);
658 return vrf_ip6_rcv(vrf_dev, skb);
664 #if IS_ENABLED(CONFIG_IPV6)
665 static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev,
666 const struct flowi6 *fl6)
668 struct rt6_info *rt = NULL;
670 if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) {
671 struct net_vrf *vrf = netdev_priv(dev);
677 return (struct dst_entry *)rt;
681 static const struct l3mdev_ops vrf_l3mdev_ops = {
682 .l3mdev_fib_table = vrf_fib_table,
683 .l3mdev_get_rtable = vrf_get_rtable,
684 .l3mdev_get_saddr = vrf_get_saddr,
685 .l3mdev_l3_rcv = vrf_l3_rcv,
686 #if IS_ENABLED(CONFIG_IPV6)
687 .l3mdev_get_rt6_dst = vrf_get_rt6_dst,
691 static void vrf_get_drvinfo(struct net_device *dev,
692 struct ethtool_drvinfo *info)
694 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
695 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
698 static const struct ethtool_ops vrf_ethtool_ops = {
699 .get_drvinfo = vrf_get_drvinfo,
702 static void vrf_setup(struct net_device *dev)
706 /* Initialize the device structure. */
707 dev->netdev_ops = &vrf_netdev_ops;
708 dev->l3mdev_ops = &vrf_l3mdev_ops;
709 dev->ethtool_ops = &vrf_ethtool_ops;
710 dev->destructor = free_netdev;
712 /* Fill in device structure with ethernet-generic values. */
713 eth_hw_addr_random(dev);
715 /* don't acquire vrf device's netif_tx_lock when transmitting */
716 dev->features |= NETIF_F_LLTX;
718 /* don't allow vrf devices to change network namespaces. */
719 dev->features |= NETIF_F_NETNS_LOCAL;
722 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
724 if (tb[IFLA_ADDRESS]) {
725 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
727 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
728 return -EADDRNOTAVAIL;
733 static void vrf_dellink(struct net_device *dev, struct list_head *head)
735 unregister_netdevice_queue(dev, head);
738 static int vrf_newlink(struct net *src_net, struct net_device *dev,
739 struct nlattr *tb[], struct nlattr *data[])
741 struct net_vrf *vrf = netdev_priv(dev);
743 if (!data || !data[IFLA_VRF_TABLE])
746 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
748 dev->priv_flags |= IFF_L3MDEV_MASTER;
750 return register_netdevice(dev);
753 static size_t vrf_nl_getsize(const struct net_device *dev)
755 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
758 static int vrf_fillinfo(struct sk_buff *skb,
759 const struct net_device *dev)
761 struct net_vrf *vrf = netdev_priv(dev);
763 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
766 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
767 const struct net_device *slave_dev)
769 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
772 static int vrf_fill_slave_info(struct sk_buff *skb,
773 const struct net_device *vrf_dev,
774 const struct net_device *slave_dev)
776 struct net_vrf *vrf = netdev_priv(vrf_dev);
778 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
784 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
785 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
788 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
790 .priv_size = sizeof(struct net_vrf),
792 .get_size = vrf_nl_getsize,
793 .policy = vrf_nl_policy,
794 .validate = vrf_validate,
795 .fill_info = vrf_fillinfo,
797 .get_slave_size = vrf_get_slave_size,
798 .fill_slave_info = vrf_fill_slave_info,
800 .newlink = vrf_newlink,
801 .dellink = vrf_dellink,
803 .maxtype = IFLA_VRF_MAX,
806 static int vrf_device_event(struct notifier_block *unused,
807 unsigned long event, void *ptr)
809 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
811 /* only care about unregister events to drop slave references */
812 if (event == NETDEV_UNREGISTER) {
813 struct net_device *vrf_dev;
815 if (!netif_is_l3_slave(dev))
818 vrf_dev = netdev_master_upper_dev_get(dev);
819 vrf_del_slave(vrf_dev, dev);
825 static struct notifier_block vrf_notifier_block __read_mostly = {
826 .notifier_call = vrf_device_event,
829 static int __init vrf_init_module(void)
833 register_netdevice_notifier(&vrf_notifier_block);
835 rc = rtnl_link_register(&vrf_link_ops);
842 unregister_netdevice_notifier(&vrf_notifier_block);
846 module_init(vrf_init_module);
847 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
848 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
849 MODULE_LICENSE("GPL");
850 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
851 MODULE_VERSION(DRV_VERSION);