1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * vrf.c: device driver to encapsulate a VRF space
5 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
6 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
7 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
9 * Based on dummy, team and ipvlan drivers
12 #include <linux/ethtool.h>
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/netdevice.h>
16 #include <linux/etherdevice.h>
18 #include <linux/init.h>
19 #include <linux/moduleparam.h>
20 #include <linux/netfilter.h>
21 #include <linux/rtnetlink.h>
22 #include <net/rtnetlink.h>
23 #include <linux/u64_stats_sync.h>
24 #include <linux/hashtable.h>
25 #include <linux/spinlock_types.h>
27 #include <linux/inetdevice.h>
30 #include <net/ip_fib.h>
31 #include <net/ip6_fib.h>
32 #include <net/ip6_route.h>
33 #include <net/route.h>
34 #include <net/addrconf.h>
35 #include <net/l3mdev.h>
36 #include <net/fib_rules.h>
37 #include <net/netns/generic.h>
39 #define DRV_NAME "vrf"
40 #define DRV_VERSION "1.1"
42 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
45 #define HASH_INITVAL ((u32)0xcafef00d)
48 DECLARE_HASHTABLE(ht, HT_MAP_BITS);
52 * count how many distinct tables do not comply with the strict mode
54 * shared_tables value must be 0 in order to enable the strict mode.
56 * example of the evolution of shared_tables:
58 * add vrf0 --> table 100 shared_tables = 0 | t0
59 * add vrf1 --> table 101 shared_tables = 0 | t1
60 * add vrf2 --> table 100 shared_tables = 1 | t2
61 * add vrf3 --> table 100 shared_tables = 1 | t3
62 * add vrf4 --> table 101 shared_tables = 2 v t4
64 * shared_tables is a "step function" (or "staircase function")
65 * and it is increased by one when the second vrf is associated to a
68 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
70 * at t3, another dev (vrf3) is bound to the same table 100 but the
71 * value of shared_tables is still 1.
72 * This means that no matter how many new vrfs will register on the
73 * table 100, the shared_tables will not increase (considering only
76 * at t4, vrf4 is bound to table 101, and shared_tables = 2.
78 * Looking at the value of shared_tables we can immediately know if
79 * the strict_mode can or cannot be enforced. Indeed, strict_mode
80 * can be enforced iff shared_tables = 0.
82 * Conversely, shared_tables is decreased when a vrf is de-associated
83 * from a table with exactly two associated vrfs.
91 struct hlist_node hnode;
92 struct list_head vrf_list; /* VRFs registered to this table */
99 static unsigned int vrf_net_id;
101 /* per netns vrf data */
103 /* protected by rtnl lock */
107 struct ctl_table_header *ctl_hdr;
111 struct rtable __rcu *rth;
112 struct rt6_info __rcu *rt6;
113 #if IS_ENABLED(CONFIG_IPV6)
114 struct fib6_table *fib6_table;
118 struct list_head me_list; /* entry in vrf_map_elem */
129 struct u64_stats_sync syncp;
132 static void vrf_rx_stats(struct net_device *dev, int len)
134 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
136 u64_stats_update_begin(&dstats->syncp);
138 dstats->rx_bytes += len;
139 u64_stats_update_end(&dstats->syncp);
142 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
144 vrf_dev->stats.tx_errors++;
148 static void vrf_get_stats64(struct net_device *dev,
149 struct rtnl_link_stats64 *stats)
153 for_each_possible_cpu(i) {
154 const struct pcpu_dstats *dstats;
155 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
158 dstats = per_cpu_ptr(dev->dstats, i);
160 start = u64_stats_fetch_begin_irq(&dstats->syncp);
161 tbytes = dstats->tx_bytes;
162 tpkts = dstats->tx_pkts;
163 tdrops = dstats->tx_drps;
164 rbytes = dstats->rx_bytes;
165 rpkts = dstats->rx_pkts;
166 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
167 stats->tx_bytes += tbytes;
168 stats->tx_packets += tpkts;
169 stats->tx_dropped += tdrops;
170 stats->rx_bytes += rbytes;
171 stats->rx_packets += rpkts;
175 static struct vrf_map *netns_vrf_map(struct net *net)
177 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
179 return &nn_vrf->vmap;
182 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
184 return netns_vrf_map(dev_net(dev));
187 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
189 struct list_head *me_head = &me->vrf_list;
192 if (list_empty(me_head))
195 vrf = list_first_entry(me_head, struct net_vrf, me_list);
200 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
202 struct vrf_map_elem *me;
204 me = kmalloc(sizeof(*me), flags);
211 static void vrf_map_elem_free(struct vrf_map_elem *me)
216 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
217 int ifindex, int users)
219 me->table_id = table_id;
220 me->ifindex = ifindex;
222 INIT_LIST_HEAD(&me->vrf_list);
225 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
228 struct vrf_map_elem *me;
231 key = jhash_1word(table_id, HASH_INITVAL);
232 hash_for_each_possible(vmap->ht, me, hnode, key) {
233 if (me->table_id == table_id)
240 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
242 u32 table_id = me->table_id;
245 key = jhash_1word(table_id, HASH_INITVAL);
246 hash_add(vmap->ht, &me->hnode, key);
249 static void vrf_map_del_elem(struct vrf_map_elem *me)
251 hash_del(&me->hnode);
254 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
256 spin_lock(&vmap->vmap_lock);
259 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
261 spin_unlock(&vmap->vmap_lock);
264 /* called with rtnl lock held */
266 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
268 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
269 struct net_vrf *vrf = netdev_priv(dev);
270 struct vrf_map_elem *new_me, *me;
271 u32 table_id = vrf->tb_id;
272 bool free_new_me = false;
276 /* we pre-allocate elements used in the spin-locked section (so that we
277 * keep the spinlock as short as possibile).
279 new_me = vrf_map_elem_alloc(GFP_KERNEL);
283 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
287 me = vrf_map_lookup_elem(vmap, table_id);
290 vrf_map_add_elem(vmap, me);
294 /* we already have an entry in the vrf_map, so it means there is (at
295 * least) a vrf registered on the specific table.
298 if (vmap->strict_mode) {
299 /* vrfs cannot share the same table */
300 NL_SET_ERR_MSG(extack, "Table is used by another VRF");
308 ++vmap->shared_tables;
310 list_add(&vrf->me_list, &me->vrf_list);
315 vrf_map_unlock(vmap);
317 /* clean-up, if needed */
319 vrf_map_elem_free(new_me);
324 /* called with rtnl lock held */
325 static void vrf_map_unregister_dev(struct net_device *dev)
327 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
328 struct net_vrf *vrf = netdev_priv(dev);
329 u32 table_id = vrf->tb_id;
330 struct vrf_map_elem *me;
335 me = vrf_map_lookup_elem(vmap, table_id);
339 list_del(&vrf->me_list);
343 --vmap->shared_tables;
344 } else if (users == 0) {
345 vrf_map_del_elem(me);
347 /* no one will refer to this element anymore */
348 vrf_map_elem_free(me);
352 vrf_map_unlock(vmap);
355 /* return the vrf device index associated with the table_id */
356 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
358 struct vrf_map *vmap = netns_vrf_map(net);
359 struct vrf_map_elem *me;
364 if (!vmap->strict_mode) {
369 me = vrf_map_lookup_elem(vmap, table_id);
375 ifindex = vrf_map_elem_get_vrf_ifindex(me);
378 vrf_map_unlock(vmap);
383 /* by default VRF devices do not have a qdisc and are expected
384 * to be created with only a single queue.
386 static bool qdisc_tx_is_default(const struct net_device *dev)
388 struct netdev_queue *txq;
391 if (dev->num_tx_queues > 1)
394 txq = netdev_get_tx_queue(dev, 0);
395 qdisc = rcu_access_pointer(txq->qdisc);
397 return !qdisc->enqueue;
400 /* Local traffic destined to local address. Reinsert the packet to rx
401 * path, similar to loopback handling.
403 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
404 struct dst_entry *dst)
410 skb_dst_set(skb, dst);
412 /* set pkt_type to avoid skb hitting packet taps twice -
413 * once on Tx and again in Rx processing
415 skb->pkt_type = PACKET_LOOPBACK;
417 skb->protocol = eth_type_trans(skb, dev);
419 if (likely(netif_rx(skb) == NET_RX_SUCCESS))
420 vrf_rx_stats(dev, len);
422 this_cpu_inc(dev->dstats->rx_drps);
427 #if IS_ENABLED(CONFIG_IPV6)
428 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
433 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
434 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
436 if (likely(err == 1))
437 err = dst_output(net, sk, skb);
442 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
443 struct net_device *dev)
445 const struct ipv6hdr *iph;
446 struct net *net = dev_net(skb->dev);
448 int ret = NET_XMIT_DROP;
449 struct dst_entry *dst;
450 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
452 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
457 memset(&fl6, 0, sizeof(fl6));
458 /* needed to match OIF rule */
459 fl6.flowi6_oif = dev->ifindex;
460 fl6.flowi6_iif = LOOPBACK_IFINDEX;
461 fl6.daddr = iph->daddr;
462 fl6.saddr = iph->saddr;
463 fl6.flowlabel = ip6_flowinfo(iph);
464 fl6.flowi6_mark = skb->mark;
465 fl6.flowi6_proto = iph->nexthdr;
466 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
468 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
469 if (IS_ERR(dst) || dst == dst_null)
474 /* if dst.dev is loopback or the VRF device again this is locally
475 * originated traffic destined to a local address. Short circuit
479 return vrf_local_xmit(skb, dev, dst);
481 skb_dst_set(skb, dst);
483 /* strip the ethernet header added for pass through VRF device */
484 __skb_pull(skb, skb_network_offset(skb));
486 ret = vrf_ip6_local_out(net, skb->sk, skb);
487 if (unlikely(net_xmit_eval(ret)))
488 dev->stats.tx_errors++;
490 ret = NET_XMIT_SUCCESS;
494 vrf_tx_error(dev, skb);
495 return NET_XMIT_DROP;
498 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
499 struct net_device *dev)
501 vrf_tx_error(dev, skb);
502 return NET_XMIT_DROP;
506 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
507 static int vrf_ip_local_out(struct net *net, struct sock *sk,
512 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
513 skb, NULL, skb_dst(skb)->dev, dst_output);
514 if (likely(err == 1))
515 err = dst_output(net, sk, skb);
520 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
521 struct net_device *vrf_dev)
524 int ret = NET_XMIT_DROP;
526 struct net *net = dev_net(vrf_dev);
529 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
534 memset(&fl4, 0, sizeof(fl4));
535 /* needed to match OIF rule */
536 fl4.flowi4_oif = vrf_dev->ifindex;
537 fl4.flowi4_iif = LOOPBACK_IFINDEX;
538 fl4.flowi4_tos = RT_TOS(ip4h->tos);
539 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
540 fl4.flowi4_proto = ip4h->protocol;
541 fl4.daddr = ip4h->daddr;
542 fl4.saddr = ip4h->saddr;
544 rt = ip_route_output_flow(net, &fl4, NULL);
550 /* if dst.dev is loopback or the VRF device again this is locally
551 * originated traffic destined to a local address. Short circuit
554 if (rt->dst.dev == vrf_dev)
555 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
557 skb_dst_set(skb, &rt->dst);
559 /* strip the ethernet header added for pass through VRF device */
560 __skb_pull(skb, skb_network_offset(skb));
563 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
567 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
568 if (unlikely(net_xmit_eval(ret)))
569 vrf_dev->stats.tx_errors++;
571 ret = NET_XMIT_SUCCESS;
576 vrf_tx_error(vrf_dev, skb);
580 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
582 switch (skb->protocol) {
583 case htons(ETH_P_IP):
584 return vrf_process_v4_outbound(skb, dev);
585 case htons(ETH_P_IPV6):
586 return vrf_process_v6_outbound(skb, dev);
588 vrf_tx_error(dev, skb);
589 return NET_XMIT_DROP;
593 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
596 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
598 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
599 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
601 u64_stats_update_begin(&dstats->syncp);
603 dstats->tx_bytes += len;
604 u64_stats_update_end(&dstats->syncp);
606 this_cpu_inc(dev->dstats->tx_drps);
612 static void vrf_finish_direct(struct sk_buff *skb)
614 struct net_device *vrf_dev = skb->dev;
616 if (!list_empty(&vrf_dev->ptype_all) &&
617 likely(skb_headroom(skb) >= ETH_HLEN)) {
618 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
620 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
621 eth_zero_addr(eth->h_dest);
622 eth->h_proto = skb->protocol;
625 dev_queue_xmit_nit(skb, vrf_dev);
626 rcu_read_unlock_bh();
628 skb_pull(skb, ETH_HLEN);
631 /* reset skb device */
635 #if IS_ENABLED(CONFIG_IPV6)
636 /* modelled after ip6_finish_output2 */
637 static int vrf_finish_output6(struct net *net, struct sock *sk,
640 struct dst_entry *dst = skb_dst(skb);
641 struct net_device *dev = dst->dev;
642 const struct in6_addr *nexthop;
643 struct neighbour *neigh;
648 skb->protocol = htons(ETH_P_IPV6);
652 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
653 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
654 if (unlikely(!neigh))
655 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
656 if (!IS_ERR(neigh)) {
657 sock_confirm_neigh(skb, neigh);
658 ret = neigh_output(neigh, skb, false);
659 rcu_read_unlock_bh();
662 rcu_read_unlock_bh();
664 IP6_INC_STATS(dev_net(dst->dev),
665 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
670 /* modelled after ip6_output */
671 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
673 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
674 net, sk, skb, NULL, skb_dst(skb)->dev,
676 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
679 /* set dst on skb to send packet to us via dev_xmit path. Allows
680 * packet to go through device based features such as qdisc, netfilter
681 * hooks and packet sockets with skb->dev set to vrf device.
683 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
686 struct net_vrf *vrf = netdev_priv(vrf_dev);
687 struct dst_entry *dst = NULL;
688 struct rt6_info *rt6;
692 rt6 = rcu_dereference(vrf->rt6);
700 if (unlikely(!dst)) {
701 vrf_tx_error(vrf_dev, skb);
706 skb_dst_set(skb, dst);
711 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
714 vrf_finish_direct(skb);
716 return vrf_ip6_local_out(net, sk, skb);
719 static int vrf_output6_direct(struct net *net, struct sock *sk,
724 skb->protocol = htons(ETH_P_IPV6);
726 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
727 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
728 NULL, skb->dev, vrf_output6_direct_finish);
730 if (likely(err == 1))
731 vrf_finish_direct(skb);
736 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
741 err = vrf_output6_direct(net, sk, skb);
742 if (likely(err == 1))
743 err = vrf_ip6_local_out(net, sk, skb);
748 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
752 struct net *net = dev_net(vrf_dev);
757 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
758 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
760 if (likely(err == 1))
761 err = vrf_output6_direct(net, sk, skb);
763 if (likely(err == 1))
769 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
773 /* don't divert link scope packets */
774 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
777 if (qdisc_tx_is_default(vrf_dev) ||
778 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
779 return vrf_ip6_out_direct(vrf_dev, sk, skb);
781 return vrf_ip6_out_redirect(vrf_dev, skb);
785 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
787 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
788 struct net *net = dev_net(dev);
789 struct dst_entry *dst;
791 RCU_INIT_POINTER(vrf->rt6, NULL);
794 /* move dev in dst's to loopback so this VRF device can be deleted
795 * - based on dst_ifdown
800 dst->dev = net->loopback_dev;
806 static int vrf_rt6_create(struct net_device *dev)
808 int flags = DST_NOPOLICY | DST_NOXFRM;
809 struct net_vrf *vrf = netdev_priv(dev);
810 struct net *net = dev_net(dev);
811 struct rt6_info *rt6;
814 /* IPv6 can be CONFIG enabled and then disabled runtime */
815 if (!ipv6_mod_enabled())
818 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
819 if (!vrf->fib6_table)
822 /* create a dst for routing packets out a VRF device */
823 rt6 = ip6_dst_alloc(net, dev, flags);
827 rt6->dst.output = vrf_output6;
829 rcu_assign_pointer(vrf->rt6, rt6);
836 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
843 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
847 static int vrf_rt6_create(struct net_device *dev)
853 /* modelled after ip_finish_output2 */
854 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
856 struct dst_entry *dst = skb_dst(skb);
857 struct rtable *rt = (struct rtable *)dst;
858 struct net_device *dev = dst->dev;
859 unsigned int hh_len = LL_RESERVED_SPACE(dev);
860 struct neighbour *neigh;
861 bool is_v6gw = false;
866 /* Be paranoid, rather than too clever. */
867 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
868 struct sk_buff *skb2;
870 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
876 skb_set_owner_w(skb2, skb->sk);
884 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
885 if (!IS_ERR(neigh)) {
886 sock_confirm_neigh(skb, neigh);
887 /* if crossing protocols, can not use the cached header */
888 ret = neigh_output(neigh, skb, is_v6gw);
889 rcu_read_unlock_bh();
893 rcu_read_unlock_bh();
895 vrf_tx_error(skb->dev, skb);
899 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
901 struct net_device *dev = skb_dst(skb)->dev;
903 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
906 skb->protocol = htons(ETH_P_IP);
908 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
909 net, sk, skb, NULL, dev,
911 !(IPCB(skb)->flags & IPSKB_REROUTED));
914 /* set dst on skb to send packet to us via dev_xmit path. Allows
915 * packet to go through device based features such as qdisc, netfilter
916 * hooks and packet sockets with skb->dev set to vrf device.
918 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
921 struct net_vrf *vrf = netdev_priv(vrf_dev);
922 struct dst_entry *dst = NULL;
927 rth = rcu_dereference(vrf->rth);
935 if (unlikely(!dst)) {
936 vrf_tx_error(vrf_dev, skb);
941 skb_dst_set(skb, dst);
946 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
949 vrf_finish_direct(skb);
951 return vrf_ip_local_out(net, sk, skb);
954 static int vrf_output_direct(struct net *net, struct sock *sk,
959 skb->protocol = htons(ETH_P_IP);
961 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
962 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
963 NULL, skb->dev, vrf_output_direct_finish);
965 if (likely(err == 1))
966 vrf_finish_direct(skb);
971 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
976 err = vrf_output_direct(net, sk, skb);
977 if (likely(err == 1))
978 err = vrf_ip_local_out(net, sk, skb);
983 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
987 struct net *net = dev_net(vrf_dev);
992 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
993 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
995 if (likely(err == 1))
996 err = vrf_output_direct(net, sk, skb);
998 if (likely(err == 1))
1004 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
1006 struct sk_buff *skb)
1008 /* don't divert multicast or local broadcast */
1009 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
1010 ipv4_is_lbcast(ip_hdr(skb)->daddr))
1013 if (qdisc_tx_is_default(vrf_dev) ||
1014 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
1015 return vrf_ip_out_direct(vrf_dev, sk, skb);
1017 return vrf_ip_out_redirect(vrf_dev, skb);
1020 /* called with rcu lock held */
1021 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1023 struct sk_buff *skb,
1028 return vrf_ip_out(vrf_dev, sk, skb);
1030 return vrf_ip6_out(vrf_dev, sk, skb);
1037 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
1039 struct rtable *rth = rtnl_dereference(vrf->rth);
1040 struct net *net = dev_net(dev);
1041 struct dst_entry *dst;
1043 RCU_INIT_POINTER(vrf->rth, NULL);
1046 /* move dev in dst's to loopback so this VRF device can be deleted
1047 * - based on dst_ifdown
1052 dst->dev = net->loopback_dev;
1058 static int vrf_rtable_create(struct net_device *dev)
1060 struct net_vrf *vrf = netdev_priv(dev);
1063 if (!fib_new_table(dev_net(dev), vrf->tb_id))
1066 /* create a dst for routing packets out through a VRF device */
1067 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
1071 rth->dst.output = vrf_output;
1073 rcu_assign_pointer(vrf->rth, rth);
1078 /**************************** device handling ********************/
1080 /* cycle interface to flush neighbor cache and move routes across tables */
1081 static void cycle_netdev(struct net_device *dev,
1082 struct netlink_ext_ack *extack)
1084 unsigned int flags = dev->flags;
1087 if (!netif_running(dev))
1090 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1092 ret = dev_change_flags(dev, flags, extack);
1096 "Failed to cycle device %s; route tables might be wrong!\n",
1101 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1102 struct netlink_ext_ack *extack)
1106 /* do not allow loopback device to be enslaved to a VRF.
1107 * The vrf device acts as the loopback for the vrf.
1109 if (port_dev == dev_net(dev)->loopback_dev) {
1110 NL_SET_ERR_MSG(extack,
1111 "Can not enslave loopback device to a VRF");
1115 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1116 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1120 cycle_netdev(port_dev, extack);
1125 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1129 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1130 struct netlink_ext_ack *extack)
1132 if (netif_is_l3_master(port_dev)) {
1133 NL_SET_ERR_MSG(extack,
1134 "Can not enslave an L3 master device to a VRF");
1138 if (netif_is_l3_slave(port_dev))
1141 return do_vrf_add_slave(dev, port_dev, extack);
1144 /* inverse of do_vrf_add_slave */
1145 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1147 netdev_upper_dev_unlink(port_dev, dev);
1148 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1150 cycle_netdev(port_dev, NULL);
1155 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1157 return do_vrf_del_slave(dev, port_dev);
1160 static void vrf_dev_uninit(struct net_device *dev)
1162 struct net_vrf *vrf = netdev_priv(dev);
1164 vrf_rtable_release(dev, vrf);
1165 vrf_rt6_release(dev, vrf);
1167 free_percpu(dev->dstats);
1171 static int vrf_dev_init(struct net_device *dev)
1173 struct net_vrf *vrf = netdev_priv(dev);
1175 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1179 /* create the default dst which points back to us */
1180 if (vrf_rtable_create(dev) != 0)
1183 if (vrf_rt6_create(dev) != 0)
1186 dev->flags = IFF_MASTER | IFF_NOARP;
1188 /* MTU is irrelevant for VRF device; set to 64k similar to lo */
1189 dev->mtu = 64 * 1024;
1191 /* similarly, oper state is irrelevant; set to up to avoid confusion */
1192 dev->operstate = IF_OPER_UP;
1193 netdev_lockdep_set_classes(dev);
1197 vrf_rtable_release(dev, vrf);
1199 free_percpu(dev->dstats);
1205 static const struct net_device_ops vrf_netdev_ops = {
1206 .ndo_init = vrf_dev_init,
1207 .ndo_uninit = vrf_dev_uninit,
1208 .ndo_start_xmit = vrf_xmit,
1209 .ndo_set_mac_address = eth_mac_addr,
1210 .ndo_get_stats64 = vrf_get_stats64,
1211 .ndo_add_slave = vrf_add_slave,
1212 .ndo_del_slave = vrf_del_slave,
1215 static u32 vrf_fib_table(const struct net_device *dev)
1217 struct net_vrf *vrf = netdev_priv(dev);
1222 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1228 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1229 struct sk_buff *skb,
1230 struct net_device *dev)
1232 struct net *net = dev_net(dev);
1234 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1235 skb = NULL; /* kfree_skb(skb) handled by nf code */
1240 #if IS_ENABLED(CONFIG_IPV6)
1241 /* neighbor handling is done with actual device; do not want
1242 * to flip skb->dev for those ndisc packets. This really fails
1243 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1246 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1248 const struct ipv6hdr *iph = ipv6_hdr(skb);
1251 if (iph->nexthdr == NEXTHDR_ICMP) {
1252 const struct icmp6hdr *icmph;
1253 struct icmp6hdr _icmph;
1255 icmph = skb_header_pointer(skb, sizeof(*iph),
1256 sizeof(_icmph), &_icmph);
1260 switch (icmph->icmp6_type) {
1261 case NDISC_ROUTER_SOLICITATION:
1262 case NDISC_ROUTER_ADVERTISEMENT:
1263 case NDISC_NEIGHBOUR_SOLICITATION:
1264 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1265 case NDISC_REDIRECT:
1275 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1276 const struct net_device *dev,
1279 const struct sk_buff *skb,
1282 struct net_vrf *vrf = netdev_priv(dev);
1284 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1287 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1290 const struct ipv6hdr *iph = ipv6_hdr(skb);
1291 struct flowi6 fl6 = {
1292 .flowi6_iif = ifindex,
1293 .flowi6_mark = skb->mark,
1294 .flowi6_proto = iph->nexthdr,
1295 .daddr = iph->daddr,
1296 .saddr = iph->saddr,
1297 .flowlabel = ip6_flowinfo(iph),
1299 struct net *net = dev_net(vrf_dev);
1300 struct rt6_info *rt6;
1302 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1303 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1307 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1310 skb_dst_set(skb, &rt6->dst);
1313 static int vrf_prepare_mac_header(struct sk_buff *skb,
1314 struct net_device *vrf_dev, u16 proto)
1319 /* in general, we do not know if there is enough space in the head of
1320 * the packet for hosting the mac header.
1322 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
1324 /* no space in the skb head */
1327 __skb_push(skb, ETH_HLEN);
1328 eth = (struct ethhdr *)skb->data;
1330 skb_reset_mac_header(skb);
1332 /* we set the ethernet destination and the source addresses to the
1333 * address of the VRF device.
1335 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
1336 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
1337 eth->h_proto = htons(proto);
1339 /* the destination address of the Ethernet frame corresponds to the
1340 * address set on the VRF interface; therefore, the packet is intended
1341 * to be processed locally.
1343 skb->protocol = eth->h_proto;
1344 skb->pkt_type = PACKET_HOST;
1346 skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
1348 skb_pull_inline(skb, ETH_HLEN);
1353 /* prepare and add the mac header to the packet if it was not set previously.
1354 * In this way, packet sniffers such as tcpdump can parse the packet correctly.
1355 * If the mac header was already set, the original mac header is left
1356 * untouched and the function returns immediately.
1358 static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
1359 struct net_device *vrf_dev,
1362 if (skb_mac_header_was_set(skb))
1365 return vrf_prepare_mac_header(skb, vrf_dev, proto);
1368 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1369 struct sk_buff *skb)
1371 int orig_iif = skb->skb_iif;
1372 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1373 bool is_ndisc = ipv6_ndisc_frame(skb);
1375 /* loopback, multicast & non-ND link-local traffic; do not push through
1376 * packet taps again. Reset pkt_type for upper layers to process skb
1378 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1380 skb->skb_iif = vrf_dev->ifindex;
1381 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1382 if (skb->pkt_type == PACKET_LOOPBACK)
1383 skb->pkt_type = PACKET_HOST;
1387 /* if packet is NDISC then keep the ingress interface */
1389 vrf_rx_stats(vrf_dev, skb->len);
1391 skb->skb_iif = vrf_dev->ifindex;
1393 if (!list_empty(&vrf_dev->ptype_all)) {
1396 err = vrf_add_mac_header_if_unset(skb, vrf_dev,
1399 skb_push(skb, skb->mac_len);
1400 dev_queue_xmit_nit(skb, vrf_dev);
1401 skb_pull(skb, skb->mac_len);
1405 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1409 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1411 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1417 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1418 struct sk_buff *skb)
1424 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1425 struct sk_buff *skb)
1428 skb->skb_iif = vrf_dev->ifindex;
1429 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1431 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1434 /* loopback traffic; do not push through packet taps again.
1435 * Reset pkt_type for upper layers to process skb
1437 if (skb->pkt_type == PACKET_LOOPBACK) {
1438 skb->pkt_type = PACKET_HOST;
1442 vrf_rx_stats(vrf_dev, skb->len);
1444 if (!list_empty(&vrf_dev->ptype_all)) {
1447 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP);
1449 skb_push(skb, skb->mac_len);
1450 dev_queue_xmit_nit(skb, vrf_dev);
1451 skb_pull(skb, skb->mac_len);
1455 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1460 /* called with rcu lock held */
1461 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1462 struct sk_buff *skb,
1467 return vrf_ip_rcv(vrf_dev, skb);
1469 return vrf_ip6_rcv(vrf_dev, skb);
1475 #if IS_ENABLED(CONFIG_IPV6)
1476 /* send to link-local or multicast address via interface enslaved to
1477 * VRF device. Force lookup to VRF table without changing flow struct
1478 * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1479 * is taken on the dst by this function.
1481 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1484 struct net *net = dev_net(dev);
1485 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1486 struct dst_entry *dst = NULL;
1487 struct rt6_info *rt;
1489 /* VRF device does not have a link-local address and
1490 * sending packets to link-local or mcast addresses over
1491 * a VRF device does not make sense
1493 if (fl6->flowi6_oif == dev->ifindex) {
1494 dst = &net->ipv6.ip6_null_entry->dst;
1498 if (!ipv6_addr_any(&fl6->saddr))
1499 flags |= RT6_LOOKUP_F_HAS_SADDR;
1501 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1509 static const struct l3mdev_ops vrf_l3mdev_ops = {
1510 .l3mdev_fib_table = vrf_fib_table,
1511 .l3mdev_l3_rcv = vrf_l3_rcv,
1512 .l3mdev_l3_out = vrf_l3_out,
1513 #if IS_ENABLED(CONFIG_IPV6)
1514 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1518 static void vrf_get_drvinfo(struct net_device *dev,
1519 struct ethtool_drvinfo *info)
1521 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1522 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1525 static const struct ethtool_ops vrf_ethtool_ops = {
1526 .get_drvinfo = vrf_get_drvinfo,
1529 static inline size_t vrf_fib_rule_nl_size(void)
1533 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1534 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1535 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1536 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1541 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1543 struct fib_rule_hdr *frh;
1544 struct nlmsghdr *nlh;
1545 struct sk_buff *skb;
1548 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1549 !ipv6_mod_enabled())
1552 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1556 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1558 goto nla_put_failure;
1560 /* rule only needs to appear once */
1561 nlh->nlmsg_flags |= NLM_F_EXCL;
1563 frh = nlmsg_data(nlh);
1564 memset(frh, 0, sizeof(*frh));
1565 frh->family = family;
1566 frh->action = FR_ACT_TO_TBL;
1568 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1569 goto nla_put_failure;
1571 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1572 goto nla_put_failure;
1574 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1575 goto nla_put_failure;
1577 nlmsg_end(skb, nlh);
1579 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1580 skb->sk = dev_net(dev)->rtnl;
1582 err = fib_nl_newrule(skb, nlh, NULL);
1586 err = fib_nl_delrule(skb, nlh, NULL);
1600 static int vrf_add_fib_rules(const struct net_device *dev)
1604 err = vrf_fib_rule(dev, AF_INET, true);
1608 err = vrf_fib_rule(dev, AF_INET6, true);
1612 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1613 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1618 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1619 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1626 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1628 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1631 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1633 vrf_fib_rule(dev, AF_INET6, false);
1637 vrf_fib_rule(dev, AF_INET, false);
1640 netdev_err(dev, "Failed to add FIB rules.\n");
1644 static void vrf_setup(struct net_device *dev)
1648 /* Initialize the device structure. */
1649 dev->netdev_ops = &vrf_netdev_ops;
1650 dev->l3mdev_ops = &vrf_l3mdev_ops;
1651 dev->ethtool_ops = &vrf_ethtool_ops;
1652 dev->needs_free_netdev = true;
1654 /* Fill in device structure with ethernet-generic values. */
1655 eth_hw_addr_random(dev);
1657 /* don't acquire vrf device's netif_tx_lock when transmitting */
1658 dev->features |= NETIF_F_LLTX;
1660 /* don't allow vrf devices to change network namespaces. */
1661 dev->features |= NETIF_F_NETNS_LOCAL;
1663 /* does not make sense for a VLAN to be added to a vrf device */
1664 dev->features |= NETIF_F_VLAN_CHALLENGED;
1666 /* enable offload features */
1667 dev->features |= NETIF_F_GSO_SOFTWARE;
1668 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1669 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1671 dev->hw_features = dev->features;
1672 dev->hw_enc_features = dev->features;
1674 /* default to no qdisc; user can add if desired */
1675 dev->priv_flags |= IFF_NO_QUEUE;
1676 dev->priv_flags |= IFF_NO_RX_HANDLER;
1677 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1679 /* VRF devices do not care about MTU, but if the MTU is set
1680 * too low then the ipv4 and ipv6 protocols are disabled
1681 * which breaks networking.
1683 dev->min_mtu = IPV6_MIN_MTU;
1684 dev->max_mtu = ETH_MAX_MTU;
1687 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1688 struct netlink_ext_ack *extack)
1690 if (tb[IFLA_ADDRESS]) {
1691 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1692 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1695 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1696 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1697 return -EADDRNOTAVAIL;
1703 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1705 struct net_device *port_dev;
1706 struct list_head *iter;
1708 netdev_for_each_lower_dev(dev, port_dev, iter)
1709 vrf_del_slave(dev, port_dev);
1711 vrf_map_unregister_dev(dev);
1713 unregister_netdevice_queue(dev, head);
1716 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1717 struct nlattr *tb[], struct nlattr *data[],
1718 struct netlink_ext_ack *extack)
1720 struct net_vrf *vrf = netdev_priv(dev);
1721 struct netns_vrf *nn_vrf;
1722 bool *add_fib_rules;
1726 if (!data || !data[IFLA_VRF_TABLE]) {
1727 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1731 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1732 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1733 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1734 "Invalid VRF table id");
1738 dev->priv_flags |= IFF_L3MDEV_MASTER;
1740 err = register_netdevice(dev);
1744 /* mapping between table_id and vrf;
1745 * note: such binding could not be done in the dev init function
1746 * because dev->ifindex id is not available yet.
1748 vrf->ifindex = dev->ifindex;
1750 err = vrf_map_register_dev(dev, extack);
1752 unregister_netdevice(dev);
1757 nn_vrf = net_generic(net, vrf_net_id);
1759 add_fib_rules = &nn_vrf->add_fib_rules;
1760 if (*add_fib_rules) {
1761 err = vrf_add_fib_rules(dev);
1763 vrf_map_unregister_dev(dev);
1764 unregister_netdevice(dev);
1767 *add_fib_rules = false;
1774 static size_t vrf_nl_getsize(const struct net_device *dev)
1776 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1779 static int vrf_fillinfo(struct sk_buff *skb,
1780 const struct net_device *dev)
1782 struct net_vrf *vrf = netdev_priv(dev);
1784 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1787 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1788 const struct net_device *slave_dev)
1790 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1793 static int vrf_fill_slave_info(struct sk_buff *skb,
1794 const struct net_device *vrf_dev,
1795 const struct net_device *slave_dev)
1797 struct net_vrf *vrf = netdev_priv(vrf_dev);
1799 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1805 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1806 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1809 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1811 .priv_size = sizeof(struct net_vrf),
1813 .get_size = vrf_nl_getsize,
1814 .policy = vrf_nl_policy,
1815 .validate = vrf_validate,
1816 .fill_info = vrf_fillinfo,
1818 .get_slave_size = vrf_get_slave_size,
1819 .fill_slave_info = vrf_fill_slave_info,
1821 .newlink = vrf_newlink,
1822 .dellink = vrf_dellink,
1824 .maxtype = IFLA_VRF_MAX,
1827 static int vrf_device_event(struct notifier_block *unused,
1828 unsigned long event, void *ptr)
1830 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1832 /* only care about unregister events to drop slave references */
1833 if (event == NETDEV_UNREGISTER) {
1834 struct net_device *vrf_dev;
1836 if (!netif_is_l3_slave(dev))
1839 vrf_dev = netdev_master_upper_dev_get(dev);
1840 vrf_del_slave(vrf_dev, dev);
1846 static struct notifier_block vrf_notifier_block __read_mostly = {
1847 .notifier_call = vrf_device_event,
1850 static int vrf_map_init(struct vrf_map *vmap)
1852 spin_lock_init(&vmap->vmap_lock);
1853 hash_init(vmap->ht);
1855 vmap->strict_mode = false;
1860 #ifdef CONFIG_SYSCTL
1861 static bool vrf_strict_mode(struct vrf_map *vmap)
1866 strict_mode = vmap->strict_mode;
1867 vrf_map_unlock(vmap);
1872 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1879 cur_mode = &vmap->strict_mode;
1880 if (*cur_mode == new_mode)
1884 /* disable strict mode */
1887 if (vmap->shared_tables) {
1888 /* we cannot allow strict_mode because there are some
1889 * vrfs that share one or more tables.
1895 /* no tables are shared among vrfs, so we can go back
1896 * to 1:1 association between a vrf with its table.
1902 vrf_map_unlock(vmap);
1907 static int vrf_shared_table_handler(struct ctl_table *table, int write,
1908 void *buffer, size_t *lenp, loff_t *ppos)
1910 struct net *net = (struct net *)table->extra1;
1911 struct vrf_map *vmap = netns_vrf_map(net);
1912 int proc_strict_mode = 0;
1913 struct ctl_table tmp = {
1914 .procname = table->procname,
1915 .data = &proc_strict_mode,
1916 .maxlen = sizeof(int),
1917 .mode = table->mode,
1918 .extra1 = SYSCTL_ZERO,
1919 .extra2 = SYSCTL_ONE,
1924 proc_strict_mode = vrf_strict_mode(vmap);
1926 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1928 if (write && ret == 0)
1929 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1934 static const struct ctl_table vrf_table[] = {
1936 .procname = "strict_mode",
1938 .maxlen = sizeof(int),
1940 .proc_handler = vrf_shared_table_handler,
1941 /* set by the vrf_netns_init */
1947 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1949 struct ctl_table *table;
1951 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1955 /* init the extra1 parameter with the reference to current netns */
1956 table[0].extra1 = net;
1958 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1959 if (!nn_vrf->ctl_hdr) {
1967 static void vrf_netns_exit_sysctl(struct net *net)
1969 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1970 struct ctl_table *table;
1972 table = nn_vrf->ctl_hdr->ctl_table_arg;
1973 unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1977 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1982 static void vrf_netns_exit_sysctl(struct net *net)
1987 /* Initialize per network namespace state */
1988 static int __net_init vrf_netns_init(struct net *net)
1990 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1992 nn_vrf->add_fib_rules = true;
1993 vrf_map_init(&nn_vrf->vmap);
1995 return vrf_netns_init_sysctl(net, nn_vrf);
1998 static void __net_exit vrf_netns_exit(struct net *net)
2000 vrf_netns_exit_sysctl(net);
2003 static struct pernet_operations vrf_net_ops __net_initdata = {
2004 .init = vrf_netns_init,
2005 .exit = vrf_netns_exit,
2007 .size = sizeof(struct netns_vrf),
2010 static int __init vrf_init_module(void)
2014 register_netdevice_notifier(&vrf_notifier_block);
2016 rc = register_pernet_subsys(&vrf_net_ops);
2020 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
2021 vrf_ifindex_lookup_by_table_id);
2025 rc = rtnl_link_register(&vrf_link_ops);
2027 goto table_lookup_unreg;
2032 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
2033 vrf_ifindex_lookup_by_table_id);
2036 unregister_pernet_subsys(&vrf_net_ops);
2039 unregister_netdevice_notifier(&vrf_notifier_block);
2043 module_init(vrf_init_module);
2044 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
2045 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
2046 MODULE_LICENSE("GPL");
2047 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2048 MODULE_VERSION(DRV_VERSION);