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/module.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/etherdevice.h>
17 #include <linux/init.h>
18 #include <linux/moduleparam.h>
19 #include <linux/netfilter.h>
20 #include <linux/rtnetlink.h>
21 #include <net/rtnetlink.h>
22 #include <linux/u64_stats_sync.h>
23 #include <linux/hashtable.h>
24 #include <linux/spinlock_types.h>
26 #include <linux/inetdevice.h>
29 #include <net/ip_fib.h>
30 #include <net/ip6_fib.h>
31 #include <net/ip6_route.h>
32 #include <net/route.h>
33 #include <net/addrconf.h>
34 #include <net/l3mdev.h>
35 #include <net/fib_rules.h>
36 #include <net/netns/generic.h>
37 #include <net/netfilter/nf_conntrack.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 static void vrf_nf_set_untracked(struct sk_buff *skb)
429 if (skb_get_nfct(skb) == 0)
430 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
433 static void vrf_nf_reset_ct(struct sk_buff *skb)
435 if (skb_get_nfct(skb) == IP_CT_UNTRACKED)
439 #if IS_ENABLED(CONFIG_IPV6)
440 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
445 vrf_nf_reset_ct(skb);
447 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
448 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
450 if (likely(err == 1))
451 err = dst_output(net, sk, skb);
456 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
457 struct net_device *dev)
459 const struct ipv6hdr *iph;
460 struct net *net = dev_net(skb->dev);
462 int ret = NET_XMIT_DROP;
463 struct dst_entry *dst;
464 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
466 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
471 memset(&fl6, 0, sizeof(fl6));
472 /* needed to match OIF rule */
473 fl6.flowi6_oif = dev->ifindex;
474 fl6.flowi6_iif = LOOPBACK_IFINDEX;
475 fl6.daddr = iph->daddr;
476 fl6.saddr = iph->saddr;
477 fl6.flowlabel = ip6_flowinfo(iph);
478 fl6.flowi6_mark = skb->mark;
479 fl6.flowi6_proto = iph->nexthdr;
480 fl6.flowi6_flags = FLOWI_FLAG_SKIP_NH_OIF;
482 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
483 if (IS_ERR(dst) || dst == dst_null)
488 /* if dst.dev is loopback or the VRF device again this is locally
489 * originated traffic destined to a local address. Short circuit
493 return vrf_local_xmit(skb, dev, dst);
495 skb_dst_set(skb, dst);
497 /* strip the ethernet header added for pass through VRF device */
498 __skb_pull(skb, skb_network_offset(skb));
500 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
501 ret = vrf_ip6_local_out(net, skb->sk, skb);
502 if (unlikely(net_xmit_eval(ret)))
503 dev->stats.tx_errors++;
505 ret = NET_XMIT_SUCCESS;
509 vrf_tx_error(dev, skb);
510 return NET_XMIT_DROP;
513 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
514 struct net_device *dev)
516 vrf_tx_error(dev, skb);
517 return NET_XMIT_DROP;
521 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
522 static int vrf_ip_local_out(struct net *net, struct sock *sk,
527 vrf_nf_reset_ct(skb);
529 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
530 skb, NULL, skb_dst(skb)->dev, dst_output);
531 if (likely(err == 1))
532 err = dst_output(net, sk, skb);
537 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
538 struct net_device *vrf_dev)
541 int ret = NET_XMIT_DROP;
543 struct net *net = dev_net(vrf_dev);
546 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
551 memset(&fl4, 0, sizeof(fl4));
552 /* needed to match OIF rule */
553 fl4.flowi4_oif = vrf_dev->ifindex;
554 fl4.flowi4_iif = LOOPBACK_IFINDEX;
555 fl4.flowi4_tos = RT_TOS(ip4h->tos);
556 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_SKIP_NH_OIF;
557 fl4.flowi4_proto = ip4h->protocol;
558 fl4.daddr = ip4h->daddr;
559 fl4.saddr = ip4h->saddr;
561 rt = ip_route_output_flow(net, &fl4, NULL);
567 /* if dst.dev is loopback or the VRF device again this is locally
568 * originated traffic destined to a local address. Short circuit
571 if (rt->dst.dev == vrf_dev)
572 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
574 skb_dst_set(skb, &rt->dst);
576 /* strip the ethernet header added for pass through VRF device */
577 __skb_pull(skb, skb_network_offset(skb));
580 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
584 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
585 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
586 if (unlikely(net_xmit_eval(ret)))
587 vrf_dev->stats.tx_errors++;
589 ret = NET_XMIT_SUCCESS;
594 vrf_tx_error(vrf_dev, skb);
598 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
600 switch (skb->protocol) {
601 case htons(ETH_P_IP):
602 return vrf_process_v4_outbound(skb, dev);
603 case htons(ETH_P_IPV6):
604 return vrf_process_v6_outbound(skb, dev);
606 vrf_tx_error(dev, skb);
607 return NET_XMIT_DROP;
611 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
614 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
616 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
617 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
619 u64_stats_update_begin(&dstats->syncp);
621 dstats->tx_bytes += len;
622 u64_stats_update_end(&dstats->syncp);
624 this_cpu_inc(dev->dstats->tx_drps);
630 static void vrf_finish_direct(struct sk_buff *skb)
632 struct net_device *vrf_dev = skb->dev;
634 if (!list_empty(&vrf_dev->ptype_all) &&
635 likely(skb_headroom(skb) >= ETH_HLEN)) {
636 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
638 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
639 eth_zero_addr(eth->h_dest);
640 eth->h_proto = skb->protocol;
643 dev_queue_xmit_nit(skb, vrf_dev);
644 rcu_read_unlock_bh();
646 skb_pull(skb, ETH_HLEN);
649 vrf_nf_reset_ct(skb);
652 #if IS_ENABLED(CONFIG_IPV6)
653 /* modelled after ip6_finish_output2 */
654 static int vrf_finish_output6(struct net *net, struct sock *sk,
657 struct dst_entry *dst = skb_dst(skb);
658 struct net_device *dev = dst->dev;
659 const struct in6_addr *nexthop;
660 struct neighbour *neigh;
663 vrf_nf_reset_ct(skb);
665 skb->protocol = htons(ETH_P_IPV6);
669 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
670 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
671 if (unlikely(!neigh))
672 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
673 if (!IS_ERR(neigh)) {
674 sock_confirm_neigh(skb, neigh);
675 ret = neigh_output(neigh, skb, false);
676 rcu_read_unlock_bh();
679 rcu_read_unlock_bh();
681 IP6_INC_STATS(dev_net(dst->dev),
682 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
687 /* modelled after ip6_output */
688 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
690 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
691 net, sk, skb, NULL, skb_dst(skb)->dev,
693 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
696 /* set dst on skb to send packet to us via dev_xmit path. Allows
697 * packet to go through device based features such as qdisc, netfilter
698 * hooks and packet sockets with skb->dev set to vrf device.
700 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
703 struct net_vrf *vrf = netdev_priv(vrf_dev);
704 struct dst_entry *dst = NULL;
705 struct rt6_info *rt6;
709 rt6 = rcu_dereference(vrf->rt6);
717 if (unlikely(!dst)) {
718 vrf_tx_error(vrf_dev, skb);
723 skb_dst_set(skb, dst);
728 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
731 vrf_finish_direct(skb);
733 return vrf_ip6_local_out(net, sk, skb);
736 static int vrf_output6_direct(struct net *net, struct sock *sk,
741 skb->protocol = htons(ETH_P_IPV6);
743 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
744 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
745 NULL, skb->dev, vrf_output6_direct_finish);
747 if (likely(err == 1))
748 vrf_finish_direct(skb);
753 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
758 err = vrf_output6_direct(net, sk, skb);
759 if (likely(err == 1))
760 err = vrf_ip6_local_out(net, sk, skb);
765 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
769 struct net *net = dev_net(vrf_dev);
774 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
775 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
777 if (likely(err == 1))
778 err = vrf_output6_direct(net, sk, skb);
780 if (likely(err == 1))
786 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
790 /* don't divert link scope packets */
791 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
794 vrf_nf_set_untracked(skb);
796 if (qdisc_tx_is_default(vrf_dev) ||
797 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
798 return vrf_ip6_out_direct(vrf_dev, sk, skb);
800 return vrf_ip6_out_redirect(vrf_dev, skb);
804 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
806 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
807 struct net *net = dev_net(dev);
808 struct dst_entry *dst;
810 RCU_INIT_POINTER(vrf->rt6, NULL);
813 /* move dev in dst's to loopback so this VRF device can be deleted
814 * - based on dst_ifdown
819 dst->dev = net->loopback_dev;
825 static int vrf_rt6_create(struct net_device *dev)
827 int flags = DST_NOPOLICY | DST_NOXFRM;
828 struct net_vrf *vrf = netdev_priv(dev);
829 struct net *net = dev_net(dev);
830 struct rt6_info *rt6;
833 /* IPv6 can be CONFIG enabled and then disabled runtime */
834 if (!ipv6_mod_enabled())
837 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
838 if (!vrf->fib6_table)
841 /* create a dst for routing packets out a VRF device */
842 rt6 = ip6_dst_alloc(net, dev, flags);
846 rt6->dst.output = vrf_output6;
848 rcu_assign_pointer(vrf->rt6, rt6);
855 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
862 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
866 static int vrf_rt6_create(struct net_device *dev)
872 /* modelled after ip_finish_output2 */
873 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
875 struct dst_entry *dst = skb_dst(skb);
876 struct rtable *rt = (struct rtable *)dst;
877 struct net_device *dev = dst->dev;
878 unsigned int hh_len = LL_RESERVED_SPACE(dev);
879 struct neighbour *neigh;
880 bool is_v6gw = false;
883 vrf_nf_reset_ct(skb);
885 /* Be paranoid, rather than too clever. */
886 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
887 struct sk_buff *skb2;
889 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
895 skb_set_owner_w(skb2, skb->sk);
903 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
904 if (!IS_ERR(neigh)) {
905 sock_confirm_neigh(skb, neigh);
906 /* if crossing protocols, can not use the cached header */
907 ret = neigh_output(neigh, skb, is_v6gw);
908 rcu_read_unlock_bh();
912 rcu_read_unlock_bh();
914 vrf_tx_error(skb->dev, skb);
918 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
920 struct net_device *dev = skb_dst(skb)->dev;
922 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
925 skb->protocol = htons(ETH_P_IP);
927 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
928 net, sk, skb, NULL, dev,
930 !(IPCB(skb)->flags & IPSKB_REROUTED));
933 /* set dst on skb to send packet to us via dev_xmit path. Allows
934 * packet to go through device based features such as qdisc, netfilter
935 * hooks and packet sockets with skb->dev set to vrf device.
937 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
940 struct net_vrf *vrf = netdev_priv(vrf_dev);
941 struct dst_entry *dst = NULL;
946 rth = rcu_dereference(vrf->rth);
954 if (unlikely(!dst)) {
955 vrf_tx_error(vrf_dev, skb);
960 skb_dst_set(skb, dst);
965 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
968 vrf_finish_direct(skb);
970 return vrf_ip_local_out(net, sk, skb);
973 static int vrf_output_direct(struct net *net, struct sock *sk,
978 skb->protocol = htons(ETH_P_IP);
980 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
981 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
982 NULL, skb->dev, vrf_output_direct_finish);
984 if (likely(err == 1))
985 vrf_finish_direct(skb);
990 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
995 err = vrf_output_direct(net, sk, skb);
996 if (likely(err == 1))
997 err = vrf_ip_local_out(net, sk, skb);
1002 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
1004 struct sk_buff *skb)
1006 struct net *net = dev_net(vrf_dev);
1011 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
1012 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
1014 if (likely(err == 1))
1015 err = vrf_output_direct(net, sk, skb);
1017 if (likely(err == 1))
1023 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
1025 struct sk_buff *skb)
1027 /* don't divert multicast or local broadcast */
1028 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
1029 ipv4_is_lbcast(ip_hdr(skb)->daddr))
1032 vrf_nf_set_untracked(skb);
1034 if (qdisc_tx_is_default(vrf_dev) ||
1035 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
1036 return vrf_ip_out_direct(vrf_dev, sk, skb);
1038 return vrf_ip_out_redirect(vrf_dev, skb);
1041 /* called with rcu lock held */
1042 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1044 struct sk_buff *skb,
1049 return vrf_ip_out(vrf_dev, sk, skb);
1051 return vrf_ip6_out(vrf_dev, sk, skb);
1058 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
1060 struct rtable *rth = rtnl_dereference(vrf->rth);
1061 struct net *net = dev_net(dev);
1062 struct dst_entry *dst;
1064 RCU_INIT_POINTER(vrf->rth, NULL);
1067 /* move dev in dst's to loopback so this VRF device can be deleted
1068 * - based on dst_ifdown
1073 dst->dev = net->loopback_dev;
1079 static int vrf_rtable_create(struct net_device *dev)
1081 struct net_vrf *vrf = netdev_priv(dev);
1084 if (!fib_new_table(dev_net(dev), vrf->tb_id))
1087 /* create a dst for routing packets out through a VRF device */
1088 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1);
1092 rth->dst.output = vrf_output;
1094 rcu_assign_pointer(vrf->rth, rth);
1099 /**************************** device handling ********************/
1101 /* cycle interface to flush neighbor cache and move routes across tables */
1102 static void cycle_netdev(struct net_device *dev,
1103 struct netlink_ext_ack *extack)
1105 unsigned int flags = dev->flags;
1108 if (!netif_running(dev))
1111 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1113 ret = dev_change_flags(dev, flags, extack);
1117 "Failed to cycle device %s; route tables might be wrong!\n",
1122 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1123 struct netlink_ext_ack *extack)
1127 /* do not allow loopback device to be enslaved to a VRF.
1128 * The vrf device acts as the loopback for the vrf.
1130 if (port_dev == dev_net(dev)->loopback_dev) {
1131 NL_SET_ERR_MSG(extack,
1132 "Can not enslave loopback device to a VRF");
1136 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1137 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1141 cycle_netdev(port_dev, extack);
1146 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1150 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1151 struct netlink_ext_ack *extack)
1153 if (netif_is_l3_master(port_dev)) {
1154 NL_SET_ERR_MSG(extack,
1155 "Can not enslave an L3 master device to a VRF");
1159 if (netif_is_l3_slave(port_dev))
1162 return do_vrf_add_slave(dev, port_dev, extack);
1165 /* inverse of do_vrf_add_slave */
1166 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1168 netdev_upper_dev_unlink(port_dev, dev);
1169 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1171 cycle_netdev(port_dev, NULL);
1176 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1178 return do_vrf_del_slave(dev, port_dev);
1181 static void vrf_dev_uninit(struct net_device *dev)
1183 struct net_vrf *vrf = netdev_priv(dev);
1185 vrf_rtable_release(dev, vrf);
1186 vrf_rt6_release(dev, vrf);
1188 free_percpu(dev->dstats);
1192 static int vrf_dev_init(struct net_device *dev)
1194 struct net_vrf *vrf = netdev_priv(dev);
1196 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
1200 /* create the default dst which points back to us */
1201 if (vrf_rtable_create(dev) != 0)
1204 if (vrf_rt6_create(dev) != 0)
1207 dev->flags = IFF_MASTER | IFF_NOARP;
1209 /* similarly, oper state is irrelevant; set to up to avoid confusion */
1210 dev->operstate = IF_OPER_UP;
1211 netdev_lockdep_set_classes(dev);
1215 vrf_rtable_release(dev, vrf);
1217 free_percpu(dev->dstats);
1223 static const struct net_device_ops vrf_netdev_ops = {
1224 .ndo_init = vrf_dev_init,
1225 .ndo_uninit = vrf_dev_uninit,
1226 .ndo_start_xmit = vrf_xmit,
1227 .ndo_set_mac_address = eth_mac_addr,
1228 .ndo_get_stats64 = vrf_get_stats64,
1229 .ndo_add_slave = vrf_add_slave,
1230 .ndo_del_slave = vrf_del_slave,
1233 static u32 vrf_fib_table(const struct net_device *dev)
1235 struct net_vrf *vrf = netdev_priv(dev);
1240 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1246 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1247 struct sk_buff *skb,
1248 struct net_device *dev)
1250 struct net *net = dev_net(dev);
1252 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1253 skb = NULL; /* kfree_skb(skb) handled by nf code */
1258 #if IS_ENABLED(CONFIG_IPV6)
1259 /* neighbor handling is done with actual device; do not want
1260 * to flip skb->dev for those ndisc packets. This really fails
1261 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1264 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1266 const struct ipv6hdr *iph = ipv6_hdr(skb);
1269 if (iph->nexthdr == NEXTHDR_ICMP) {
1270 const struct icmp6hdr *icmph;
1271 struct icmp6hdr _icmph;
1273 icmph = skb_header_pointer(skb, sizeof(*iph),
1274 sizeof(_icmph), &_icmph);
1278 switch (icmph->icmp6_type) {
1279 case NDISC_ROUTER_SOLICITATION:
1280 case NDISC_ROUTER_ADVERTISEMENT:
1281 case NDISC_NEIGHBOUR_SOLICITATION:
1282 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1283 case NDISC_REDIRECT:
1293 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1294 const struct net_device *dev,
1297 const struct sk_buff *skb,
1300 struct net_vrf *vrf = netdev_priv(dev);
1302 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1305 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1308 const struct ipv6hdr *iph = ipv6_hdr(skb);
1309 struct flowi6 fl6 = {
1310 .flowi6_iif = ifindex,
1311 .flowi6_mark = skb->mark,
1312 .flowi6_proto = iph->nexthdr,
1313 .daddr = iph->daddr,
1314 .saddr = iph->saddr,
1315 .flowlabel = ip6_flowinfo(iph),
1317 struct net *net = dev_net(vrf_dev);
1318 struct rt6_info *rt6;
1320 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1321 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1325 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1328 skb_dst_set(skb, &rt6->dst);
1331 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1332 struct sk_buff *skb)
1334 int orig_iif = skb->skb_iif;
1335 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1336 bool is_ndisc = ipv6_ndisc_frame(skb);
1338 /* loopback, multicast & non-ND link-local traffic; do not push through
1339 * packet taps again. Reset pkt_type for upper layers to process skb.
1340 * For strict packets with a source LLA, determine the dst using the
1343 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1345 skb->skb_iif = vrf_dev->ifindex;
1346 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1348 if (skb->pkt_type == PACKET_LOOPBACK)
1349 skb->pkt_type = PACKET_HOST;
1350 else if (ipv6_addr_type(&ipv6_hdr(skb)->saddr) & IPV6_ADDR_LINKLOCAL)
1351 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1356 /* if packet is NDISC then keep the ingress interface */
1358 vrf_rx_stats(vrf_dev, skb->len);
1360 skb->skb_iif = vrf_dev->ifindex;
1362 if (!list_empty(&vrf_dev->ptype_all)) {
1363 skb_push(skb, skb->mac_len);
1364 dev_queue_xmit_nit(skb, vrf_dev);
1365 skb_pull(skb, skb->mac_len);
1368 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1372 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1374 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1380 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1381 struct sk_buff *skb)
1387 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1388 struct sk_buff *skb)
1391 skb->skb_iif = vrf_dev->ifindex;
1392 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1394 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1397 /* loopback traffic; do not push through packet taps again.
1398 * Reset pkt_type for upper layers to process skb
1400 if (skb->pkt_type == PACKET_LOOPBACK) {
1401 skb->pkt_type = PACKET_HOST;
1405 vrf_rx_stats(vrf_dev, skb->len);
1407 if (!list_empty(&vrf_dev->ptype_all)) {
1408 skb_push(skb, skb->mac_len);
1409 dev_queue_xmit_nit(skb, vrf_dev);
1410 skb_pull(skb, skb->mac_len);
1413 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1418 /* called with rcu lock held */
1419 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1420 struct sk_buff *skb,
1425 return vrf_ip_rcv(vrf_dev, skb);
1427 return vrf_ip6_rcv(vrf_dev, skb);
1433 #if IS_ENABLED(CONFIG_IPV6)
1434 /* send to link-local or multicast address via interface enslaved to
1435 * VRF device. Force lookup to VRF table without changing flow struct
1436 * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1437 * is taken on the dst by this function.
1439 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1442 struct net *net = dev_net(dev);
1443 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1444 struct dst_entry *dst = NULL;
1445 struct rt6_info *rt;
1447 /* VRF device does not have a link-local address and
1448 * sending packets to link-local or mcast addresses over
1449 * a VRF device does not make sense
1451 if (fl6->flowi6_oif == dev->ifindex) {
1452 dst = &net->ipv6.ip6_null_entry->dst;
1456 if (!ipv6_addr_any(&fl6->saddr))
1457 flags |= RT6_LOOKUP_F_HAS_SADDR;
1459 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1467 static const struct l3mdev_ops vrf_l3mdev_ops = {
1468 .l3mdev_fib_table = vrf_fib_table,
1469 .l3mdev_l3_rcv = vrf_l3_rcv,
1470 .l3mdev_l3_out = vrf_l3_out,
1471 #if IS_ENABLED(CONFIG_IPV6)
1472 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1476 static void vrf_get_drvinfo(struct net_device *dev,
1477 struct ethtool_drvinfo *info)
1479 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1480 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1483 static const struct ethtool_ops vrf_ethtool_ops = {
1484 .get_drvinfo = vrf_get_drvinfo,
1487 static inline size_t vrf_fib_rule_nl_size(void)
1491 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1492 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1493 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1494 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1499 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1501 struct fib_rule_hdr *frh;
1502 struct nlmsghdr *nlh;
1503 struct sk_buff *skb;
1506 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1507 !ipv6_mod_enabled())
1510 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1514 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1516 goto nla_put_failure;
1518 /* rule only needs to appear once */
1519 nlh->nlmsg_flags |= NLM_F_EXCL;
1521 frh = nlmsg_data(nlh);
1522 memset(frh, 0, sizeof(*frh));
1523 frh->family = family;
1524 frh->action = FR_ACT_TO_TBL;
1526 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1527 goto nla_put_failure;
1529 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1530 goto nla_put_failure;
1532 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1533 goto nla_put_failure;
1535 nlmsg_end(skb, nlh);
1537 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1538 skb->sk = dev_net(dev)->rtnl;
1540 err = fib_nl_newrule(skb, nlh, NULL);
1544 err = fib_nl_delrule(skb, nlh, NULL);
1558 static int vrf_add_fib_rules(const struct net_device *dev)
1562 err = vrf_fib_rule(dev, AF_INET, true);
1566 err = vrf_fib_rule(dev, AF_INET6, true);
1570 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1571 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1576 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1577 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1584 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1586 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1589 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1591 vrf_fib_rule(dev, AF_INET6, false);
1595 vrf_fib_rule(dev, AF_INET, false);
1598 netdev_err(dev, "Failed to add FIB rules.\n");
1602 static void vrf_setup(struct net_device *dev)
1606 /* Initialize the device structure. */
1607 dev->netdev_ops = &vrf_netdev_ops;
1608 dev->l3mdev_ops = &vrf_l3mdev_ops;
1609 dev->ethtool_ops = &vrf_ethtool_ops;
1610 dev->needs_free_netdev = true;
1612 /* Fill in device structure with ethernet-generic values. */
1613 eth_hw_addr_random(dev);
1615 /* don't acquire vrf device's netif_tx_lock when transmitting */
1616 dev->features |= NETIF_F_LLTX;
1618 /* don't allow vrf devices to change network namespaces. */
1619 dev->features |= NETIF_F_NETNS_LOCAL;
1621 /* does not make sense for a VLAN to be added to a vrf device */
1622 dev->features |= NETIF_F_VLAN_CHALLENGED;
1624 /* enable offload features */
1625 dev->features |= NETIF_F_GSO_SOFTWARE;
1626 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1627 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1629 dev->hw_features = dev->features;
1630 dev->hw_enc_features = dev->features;
1632 /* default to no qdisc; user can add if desired */
1633 dev->priv_flags |= IFF_NO_QUEUE;
1634 dev->priv_flags |= IFF_NO_RX_HANDLER;
1635 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1637 /* VRF devices do not care about MTU, but if the MTU is set
1638 * too low then the ipv4 and ipv6 protocols are disabled
1639 * which breaks networking.
1641 dev->min_mtu = IPV6_MIN_MTU;
1642 dev->max_mtu = IP6_MAX_MTU;
1643 dev->mtu = dev->max_mtu;
1646 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1647 struct netlink_ext_ack *extack)
1649 if (tb[IFLA_ADDRESS]) {
1650 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1651 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1654 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1655 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1656 return -EADDRNOTAVAIL;
1662 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1664 struct net_device *port_dev;
1665 struct list_head *iter;
1667 netdev_for_each_lower_dev(dev, port_dev, iter)
1668 vrf_del_slave(dev, port_dev);
1670 vrf_map_unregister_dev(dev);
1672 unregister_netdevice_queue(dev, head);
1675 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1676 struct nlattr *tb[], struct nlattr *data[],
1677 struct netlink_ext_ack *extack)
1679 struct net_vrf *vrf = netdev_priv(dev);
1680 struct netns_vrf *nn_vrf;
1681 bool *add_fib_rules;
1685 if (!data || !data[IFLA_VRF_TABLE]) {
1686 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1690 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1691 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1692 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1693 "Invalid VRF table id");
1697 dev->priv_flags |= IFF_L3MDEV_MASTER;
1699 err = register_netdevice(dev);
1703 /* mapping between table_id and vrf;
1704 * note: such binding could not be done in the dev init function
1705 * because dev->ifindex id is not available yet.
1707 vrf->ifindex = dev->ifindex;
1709 err = vrf_map_register_dev(dev, extack);
1711 unregister_netdevice(dev);
1716 nn_vrf = net_generic(net, vrf_net_id);
1718 add_fib_rules = &nn_vrf->add_fib_rules;
1719 if (*add_fib_rules) {
1720 err = vrf_add_fib_rules(dev);
1722 vrf_map_unregister_dev(dev);
1723 unregister_netdevice(dev);
1726 *add_fib_rules = false;
1733 static size_t vrf_nl_getsize(const struct net_device *dev)
1735 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1738 static int vrf_fillinfo(struct sk_buff *skb,
1739 const struct net_device *dev)
1741 struct net_vrf *vrf = netdev_priv(dev);
1743 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1746 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1747 const struct net_device *slave_dev)
1749 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1752 static int vrf_fill_slave_info(struct sk_buff *skb,
1753 const struct net_device *vrf_dev,
1754 const struct net_device *slave_dev)
1756 struct net_vrf *vrf = netdev_priv(vrf_dev);
1758 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1764 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1765 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1768 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1770 .priv_size = sizeof(struct net_vrf),
1772 .get_size = vrf_nl_getsize,
1773 .policy = vrf_nl_policy,
1774 .validate = vrf_validate,
1775 .fill_info = vrf_fillinfo,
1777 .get_slave_size = vrf_get_slave_size,
1778 .fill_slave_info = vrf_fill_slave_info,
1780 .newlink = vrf_newlink,
1781 .dellink = vrf_dellink,
1783 .maxtype = IFLA_VRF_MAX,
1786 static int vrf_device_event(struct notifier_block *unused,
1787 unsigned long event, void *ptr)
1789 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1791 /* only care about unregister events to drop slave references */
1792 if (event == NETDEV_UNREGISTER) {
1793 struct net_device *vrf_dev;
1795 if (!netif_is_l3_slave(dev))
1798 vrf_dev = netdev_master_upper_dev_get(dev);
1799 vrf_del_slave(vrf_dev, dev);
1805 static struct notifier_block vrf_notifier_block __read_mostly = {
1806 .notifier_call = vrf_device_event,
1809 static int vrf_map_init(struct vrf_map *vmap)
1811 spin_lock_init(&vmap->vmap_lock);
1812 hash_init(vmap->ht);
1814 vmap->strict_mode = false;
1819 #ifdef CONFIG_SYSCTL
1820 static bool vrf_strict_mode(struct vrf_map *vmap)
1825 strict_mode = vmap->strict_mode;
1826 vrf_map_unlock(vmap);
1831 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1838 cur_mode = &vmap->strict_mode;
1839 if (*cur_mode == new_mode)
1843 /* disable strict mode */
1846 if (vmap->shared_tables) {
1847 /* we cannot allow strict_mode because there are some
1848 * vrfs that share one or more tables.
1854 /* no tables are shared among vrfs, so we can go back
1855 * to 1:1 association between a vrf with its table.
1861 vrf_map_unlock(vmap);
1866 static int vrf_shared_table_handler(struct ctl_table *table, int write,
1867 void *buffer, size_t *lenp, loff_t *ppos)
1869 struct net *net = (struct net *)table->extra1;
1870 struct vrf_map *vmap = netns_vrf_map(net);
1871 int proc_strict_mode = 0;
1872 struct ctl_table tmp = {
1873 .procname = table->procname,
1874 .data = &proc_strict_mode,
1875 .maxlen = sizeof(int),
1876 .mode = table->mode,
1877 .extra1 = SYSCTL_ZERO,
1878 .extra2 = SYSCTL_ONE,
1883 proc_strict_mode = vrf_strict_mode(vmap);
1885 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1887 if (write && ret == 0)
1888 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1893 static const struct ctl_table vrf_table[] = {
1895 .procname = "strict_mode",
1897 .maxlen = sizeof(int),
1899 .proc_handler = vrf_shared_table_handler,
1900 /* set by the vrf_netns_init */
1906 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1908 struct ctl_table *table;
1910 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1914 /* init the extra1 parameter with the reference to current netns */
1915 table[0].extra1 = net;
1917 nn_vrf->ctl_hdr = register_net_sysctl(net, "net/vrf", table);
1918 if (!nn_vrf->ctl_hdr) {
1926 static void vrf_netns_exit_sysctl(struct net *net)
1928 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1929 struct ctl_table *table;
1931 table = nn_vrf->ctl_hdr->ctl_table_arg;
1932 unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1936 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1941 static void vrf_netns_exit_sysctl(struct net *net)
1946 /* Initialize per network namespace state */
1947 static int __net_init vrf_netns_init(struct net *net)
1949 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1951 nn_vrf->add_fib_rules = true;
1952 vrf_map_init(&nn_vrf->vmap);
1954 return vrf_netns_init_sysctl(net, nn_vrf);
1957 static void __net_exit vrf_netns_exit(struct net *net)
1959 vrf_netns_exit_sysctl(net);
1962 static struct pernet_operations vrf_net_ops __net_initdata = {
1963 .init = vrf_netns_init,
1964 .exit = vrf_netns_exit,
1966 .size = sizeof(struct netns_vrf),
1969 static int __init vrf_init_module(void)
1973 register_netdevice_notifier(&vrf_notifier_block);
1975 rc = register_pernet_subsys(&vrf_net_ops);
1979 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
1980 vrf_ifindex_lookup_by_table_id);
1984 rc = rtnl_link_register(&vrf_link_ops);
1986 goto table_lookup_unreg;
1991 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
1992 vrf_ifindex_lookup_by_table_id);
1995 unregister_pernet_subsys(&vrf_net_ops);
1998 unregister_netdevice_notifier(&vrf_notifier_block);
2002 module_init(vrf_init_module);
2003 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
2004 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
2005 MODULE_LICENSE("GPL");
2006 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2007 MODULE_VERSION(DRV_VERSION);