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.
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/sch_generic.h>
38 #include <net/netns/generic.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/inet_dscp.h>
42 #define DRV_NAME "vrf"
43 #define DRV_VERSION "1.1"
45 #define FIB_RULE_PREF 1000 /* default preference for FIB rules */
48 #define HASH_INITVAL ((u32)0xcafef00d)
51 DECLARE_HASHTABLE(ht, HT_MAP_BITS);
55 * count how many distinct tables do not comply with the strict mode
57 * shared_tables value must be 0 in order to enable the strict mode.
59 * example of the evolution of shared_tables:
61 * add vrf0 --> table 100 shared_tables = 0 | t0
62 * add vrf1 --> table 101 shared_tables = 0 | t1
63 * add vrf2 --> table 100 shared_tables = 1 | t2
64 * add vrf3 --> table 100 shared_tables = 1 | t3
65 * add vrf4 --> table 101 shared_tables = 2 v t4
67 * shared_tables is a "step function" (or "staircase function")
68 * and it is increased by one when the second vrf is associated to a
71 * at t2, vrf0 and vrf2 are bound to table 100: shared_tables = 1.
73 * at t3, another dev (vrf3) is bound to the same table 100 but the
74 * value of shared_tables is still 1.
75 * This means that no matter how many new vrfs will register on the
76 * table 100, the shared_tables will not increase (considering only
79 * at t4, vrf4 is bound to table 101, and shared_tables = 2.
81 * Looking at the value of shared_tables we can immediately know if
82 * the strict_mode can or cannot be enforced. Indeed, strict_mode
83 * can be enforced iff shared_tables = 0.
85 * Conversely, shared_tables is decreased when a vrf is de-associated
86 * from a table with exactly two associated vrfs.
94 struct hlist_node hnode;
95 struct list_head vrf_list; /* VRFs registered to this table */
102 static unsigned int vrf_net_id;
104 /* per netns vrf data */
106 /* protected by rtnl lock */
110 struct ctl_table_header *ctl_hdr;
114 struct rtable __rcu *rth;
115 struct rt6_info __rcu *rt6;
116 #if IS_ENABLED(CONFIG_IPV6)
117 struct fib6_table *fib6_table;
121 struct list_head me_list; /* entry in vrf_map_elem */
125 static void vrf_rx_stats(struct net_device *dev, int len)
127 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
129 u64_stats_update_begin(&dstats->syncp);
130 u64_stats_inc(&dstats->rx_packets);
131 u64_stats_add(&dstats->rx_bytes, len);
132 u64_stats_update_end(&dstats->syncp);
135 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
137 vrf_dev->stats.tx_errors++;
141 static struct vrf_map *netns_vrf_map(struct net *net)
143 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
145 return &nn_vrf->vmap;
148 static struct vrf_map *netns_vrf_map_by_dev(struct net_device *dev)
150 return netns_vrf_map(dev_net(dev));
153 static int vrf_map_elem_get_vrf_ifindex(struct vrf_map_elem *me)
155 struct list_head *me_head = &me->vrf_list;
158 if (list_empty(me_head))
161 vrf = list_first_entry(me_head, struct net_vrf, me_list);
166 static struct vrf_map_elem *vrf_map_elem_alloc(gfp_t flags)
168 struct vrf_map_elem *me;
170 me = kmalloc(sizeof(*me), flags);
177 static void vrf_map_elem_free(struct vrf_map_elem *me)
182 static void vrf_map_elem_init(struct vrf_map_elem *me, int table_id,
183 int ifindex, int users)
185 me->table_id = table_id;
186 me->ifindex = ifindex;
188 INIT_LIST_HEAD(&me->vrf_list);
191 static struct vrf_map_elem *vrf_map_lookup_elem(struct vrf_map *vmap,
194 struct vrf_map_elem *me;
197 key = jhash_1word(table_id, HASH_INITVAL);
198 hash_for_each_possible(vmap->ht, me, hnode, key) {
199 if (me->table_id == table_id)
206 static void vrf_map_add_elem(struct vrf_map *vmap, struct vrf_map_elem *me)
208 u32 table_id = me->table_id;
211 key = jhash_1word(table_id, HASH_INITVAL);
212 hash_add(vmap->ht, &me->hnode, key);
215 static void vrf_map_del_elem(struct vrf_map_elem *me)
217 hash_del(&me->hnode);
220 static void vrf_map_lock(struct vrf_map *vmap) __acquires(&vmap->vmap_lock)
222 spin_lock(&vmap->vmap_lock);
225 static void vrf_map_unlock(struct vrf_map *vmap) __releases(&vmap->vmap_lock)
227 spin_unlock(&vmap->vmap_lock);
230 /* called with rtnl lock held */
232 vrf_map_register_dev(struct net_device *dev, struct netlink_ext_ack *extack)
234 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
235 struct net_vrf *vrf = netdev_priv(dev);
236 struct vrf_map_elem *new_me, *me;
237 u32 table_id = vrf->tb_id;
238 bool free_new_me = false;
242 /* we pre-allocate elements used in the spin-locked section (so that we
243 * keep the spinlock as short as possible).
245 new_me = vrf_map_elem_alloc(GFP_KERNEL);
249 vrf_map_elem_init(new_me, table_id, dev->ifindex, 0);
253 me = vrf_map_lookup_elem(vmap, table_id);
256 vrf_map_add_elem(vmap, me);
260 /* we already have an entry in the vrf_map, so it means there is (at
261 * least) a vrf registered on the specific table.
264 if (vmap->strict_mode) {
265 /* vrfs cannot share the same table */
266 NL_SET_ERR_MSG(extack, "Table is used by another VRF");
274 ++vmap->shared_tables;
276 list_add(&vrf->me_list, &me->vrf_list);
281 vrf_map_unlock(vmap);
283 /* clean-up, if needed */
285 vrf_map_elem_free(new_me);
290 /* called with rtnl lock held */
291 static void vrf_map_unregister_dev(struct net_device *dev)
293 struct vrf_map *vmap = netns_vrf_map_by_dev(dev);
294 struct net_vrf *vrf = netdev_priv(dev);
295 u32 table_id = vrf->tb_id;
296 struct vrf_map_elem *me;
301 me = vrf_map_lookup_elem(vmap, table_id);
305 list_del(&vrf->me_list);
309 --vmap->shared_tables;
310 } else if (users == 0) {
311 vrf_map_del_elem(me);
313 /* no one will refer to this element anymore */
314 vrf_map_elem_free(me);
318 vrf_map_unlock(vmap);
321 /* return the vrf device index associated with the table_id */
322 static int vrf_ifindex_lookup_by_table_id(struct net *net, u32 table_id)
324 struct vrf_map *vmap = netns_vrf_map(net);
325 struct vrf_map_elem *me;
330 if (!vmap->strict_mode) {
335 me = vrf_map_lookup_elem(vmap, table_id);
341 ifindex = vrf_map_elem_get_vrf_ifindex(me);
344 vrf_map_unlock(vmap);
349 /* by default VRF devices do not have a qdisc and are expected
350 * to be created with only a single queue.
352 static bool qdisc_tx_is_default(const struct net_device *dev)
354 struct netdev_queue *txq;
357 if (dev->num_tx_queues > 1)
360 txq = netdev_get_tx_queue(dev, 0);
361 qdisc = rcu_access_pointer(txq->qdisc);
363 return !qdisc->enqueue;
366 /* Local traffic destined to local address. Reinsert the packet to rx
367 * path, similar to loopback handling.
369 static int vrf_local_xmit(struct sk_buff *skb, struct net_device *dev,
370 struct dst_entry *dst)
376 skb_dst_set(skb, dst);
378 /* set pkt_type to avoid skb hitting packet taps twice -
379 * once on Tx and again in Rx processing
381 skb->pkt_type = PACKET_LOOPBACK;
383 skb->protocol = eth_type_trans(skb, dev);
385 if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) {
386 vrf_rx_stats(dev, len);
388 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
390 u64_stats_update_begin(&dstats->syncp);
391 u64_stats_inc(&dstats->rx_drops);
392 u64_stats_update_end(&dstats->syncp);
398 static void vrf_nf_set_untracked(struct sk_buff *skb)
400 if (skb_get_nfct(skb) == 0)
401 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
404 static void vrf_nf_reset_ct(struct sk_buff *skb)
406 if (skb_get_nfct(skb) == IP_CT_UNTRACKED)
410 #if IS_ENABLED(CONFIG_IPV6)
411 static int vrf_ip6_local_out(struct net *net, struct sock *sk,
416 vrf_nf_reset_ct(skb);
418 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net,
419 sk, skb, NULL, skb_dst(skb)->dev, dst_output);
421 if (likely(err == 1))
422 err = dst_output(net, sk, skb);
427 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
428 struct net_device *dev)
430 const struct ipv6hdr *iph;
431 struct net *net = dev_net(skb->dev);
433 int ret = NET_XMIT_DROP;
434 struct dst_entry *dst;
435 struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst;
437 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
442 memset(&fl6, 0, sizeof(fl6));
443 /* needed to match OIF rule */
444 fl6.flowi6_l3mdev = dev->ifindex;
445 fl6.flowi6_iif = LOOPBACK_IFINDEX;
446 fl6.daddr = iph->daddr;
447 fl6.saddr = iph->saddr;
448 fl6.flowlabel = ip6_flowinfo(iph);
449 fl6.flowi6_mark = skb->mark;
450 fl6.flowi6_proto = iph->nexthdr;
452 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
453 if (IS_ERR(dst) || dst == dst_null)
458 /* if dst.dev is the VRF device again this is locally originated traffic
459 * destined to a local address. Short circuit to Rx path.
462 return vrf_local_xmit(skb, dev, dst);
464 skb_dst_set(skb, dst);
466 /* strip the ethernet header added for pass through VRF device */
467 __skb_pull(skb, skb_network_offset(skb));
469 memset(IP6CB(skb), 0, sizeof(*IP6CB(skb)));
470 ret = vrf_ip6_local_out(net, skb->sk, skb);
471 if (unlikely(net_xmit_eval(ret)))
472 dev->stats.tx_errors++;
474 ret = NET_XMIT_SUCCESS;
478 vrf_tx_error(dev, skb);
479 return NET_XMIT_DROP;
482 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
483 struct net_device *dev)
485 vrf_tx_error(dev, skb);
486 return NET_XMIT_DROP;
490 /* based on ip_local_out; can't use it b/c the dst is switched pointing to us */
491 static int vrf_ip_local_out(struct net *net, struct sock *sk,
496 vrf_nf_reset_ct(skb);
498 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
499 skb, NULL, skb_dst(skb)->dev, dst_output);
500 if (likely(err == 1))
501 err = dst_output(net, sk, skb);
506 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
507 struct net_device *vrf_dev)
510 int ret = NET_XMIT_DROP;
512 struct net *net = dev_net(vrf_dev);
515 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
520 memset(&fl4, 0, sizeof(fl4));
521 /* needed to match OIF rule */
522 fl4.flowi4_l3mdev = vrf_dev->ifindex;
523 fl4.flowi4_iif = LOOPBACK_IFINDEX;
524 fl4.flowi4_tos = ip4h->tos & INET_DSCP_MASK;
525 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
526 fl4.flowi4_proto = ip4h->protocol;
527 fl4.daddr = ip4h->daddr;
528 fl4.saddr = ip4h->saddr;
530 rt = ip_route_output_flow(net, &fl4, NULL);
536 /* if dst.dev is the VRF device again this is locally originated traffic
537 * destined to a local address. Short circuit to Rx path.
539 if (rt->dst.dev == vrf_dev)
540 return vrf_local_xmit(skb, vrf_dev, &rt->dst);
542 skb_dst_set(skb, &rt->dst);
544 /* strip the ethernet header added for pass through VRF device */
545 __skb_pull(skb, skb_network_offset(skb));
548 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
552 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
553 ret = vrf_ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
554 if (unlikely(net_xmit_eval(ret)))
555 vrf_dev->stats.tx_errors++;
557 ret = NET_XMIT_SUCCESS;
562 vrf_tx_error(vrf_dev, skb);
566 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
568 switch (skb->protocol) {
569 case htons(ETH_P_IP):
570 return vrf_process_v4_outbound(skb, dev);
571 case htons(ETH_P_IPV6):
572 return vrf_process_v6_outbound(skb, dev);
574 vrf_tx_error(dev, skb);
575 return NET_XMIT_DROP;
579 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
581 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
584 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
586 u64_stats_update_begin(&dstats->syncp);
587 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
589 u64_stats_inc(&dstats->tx_packets);
590 u64_stats_add(&dstats->tx_bytes, len);
592 u64_stats_inc(&dstats->tx_drops);
594 u64_stats_update_end(&dstats->syncp);
599 static void vrf_finish_direct(struct sk_buff *skb)
601 struct net_device *vrf_dev = skb->dev;
603 if (!list_empty(&vrf_dev->ptype_all) &&
604 likely(skb_headroom(skb) >= ETH_HLEN)) {
605 struct ethhdr *eth = skb_push(skb, ETH_HLEN);
607 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
608 eth_zero_addr(eth->h_dest);
609 eth->h_proto = skb->protocol;
611 dev_queue_xmit_nit(skb, vrf_dev);
613 skb_pull(skb, ETH_HLEN);
616 vrf_nf_reset_ct(skb);
619 #if IS_ENABLED(CONFIG_IPV6)
620 /* modelled after ip6_finish_output2 */
621 static int vrf_finish_output6(struct net *net, struct sock *sk,
624 struct dst_entry *dst = skb_dst(skb);
625 struct net_device *dev = dst->dev;
626 const struct in6_addr *nexthop;
627 struct neighbour *neigh;
630 vrf_nf_reset_ct(skb);
632 skb->protocol = htons(ETH_P_IPV6);
636 nexthop = rt6_nexthop(dst_rt6_info(dst), &ipv6_hdr(skb)->daddr);
637 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
638 if (unlikely(!neigh))
639 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
640 if (!IS_ERR(neigh)) {
641 sock_confirm_neigh(skb, neigh);
642 ret = neigh_output(neigh, skb, false);
648 IP6_INC_STATS(dev_net(dst->dev),
649 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
654 /* modelled after ip6_output */
655 static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb)
657 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
658 net, sk, skb, NULL, skb_dst(skb)->dev,
660 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
663 /* set dst on skb to send packet to us via dev_xmit path. Allows
664 * packet to go through device based features such as qdisc, netfilter
665 * hooks and packet sockets with skb->dev set to vrf device.
667 static struct sk_buff *vrf_ip6_out_redirect(struct net_device *vrf_dev,
670 struct net_vrf *vrf = netdev_priv(vrf_dev);
671 struct dst_entry *dst = NULL;
672 struct rt6_info *rt6;
676 rt6 = rcu_dereference(vrf->rt6);
684 if (unlikely(!dst)) {
685 vrf_tx_error(vrf_dev, skb);
690 skb_dst_set(skb, dst);
695 static int vrf_output6_direct_finish(struct net *net, struct sock *sk,
698 vrf_finish_direct(skb);
700 return vrf_ip6_local_out(net, sk, skb);
703 static int vrf_output6_direct(struct net *net, struct sock *sk,
708 skb->protocol = htons(ETH_P_IPV6);
710 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
711 err = nf_hook(NFPROTO_IPV6, NF_INET_POST_ROUTING, net, sk, skb,
712 NULL, skb->dev, vrf_output6_direct_finish);
714 if (likely(err == 1))
715 vrf_finish_direct(skb);
720 static int vrf_ip6_out_direct_finish(struct net *net, struct sock *sk,
725 err = vrf_output6_direct(net, sk, skb);
726 if (likely(err == 1))
727 err = vrf_ip6_local_out(net, sk, skb);
732 static struct sk_buff *vrf_ip6_out_direct(struct net_device *vrf_dev,
736 struct net *net = dev_net(vrf_dev);
741 err = nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk,
742 skb, NULL, vrf_dev, vrf_ip6_out_direct_finish);
744 if (likely(err == 1))
745 err = vrf_output6_direct(net, sk, skb);
747 if (likely(err == 1))
753 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
757 /* don't divert link scope packets */
758 if (rt6_need_strict(&ipv6_hdr(skb)->daddr))
761 vrf_nf_set_untracked(skb);
763 if (qdisc_tx_is_default(vrf_dev) ||
764 IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
765 return vrf_ip6_out_direct(vrf_dev, sk, skb);
767 return vrf_ip6_out_redirect(vrf_dev, skb);
771 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
773 struct rt6_info *rt6 = rtnl_dereference(vrf->rt6);
774 struct net *net = dev_net(dev);
775 struct dst_entry *dst;
777 RCU_INIT_POINTER(vrf->rt6, NULL);
780 /* move dev in dst's to loopback so this VRF device can be deleted
781 * - based on dst_ifdown
785 netdev_ref_replace(dst->dev, net->loopback_dev,
786 &dst->dev_tracker, GFP_KERNEL);
787 dst->dev = net->loopback_dev;
792 static int vrf_rt6_create(struct net_device *dev)
794 int flags = DST_NOPOLICY | DST_NOXFRM;
795 struct net_vrf *vrf = netdev_priv(dev);
796 struct net *net = dev_net(dev);
797 struct rt6_info *rt6;
800 /* IPv6 can be CONFIG enabled and then disabled runtime */
801 if (!ipv6_mod_enabled())
804 vrf->fib6_table = fib6_new_table(net, vrf->tb_id);
805 if (!vrf->fib6_table)
808 /* create a dst for routing packets out a VRF device */
809 rt6 = ip6_dst_alloc(net, dev, flags);
813 rt6->dst.output = vrf_output6;
815 rcu_assign_pointer(vrf->rt6, rt6);
822 static struct sk_buff *vrf_ip6_out(struct net_device *vrf_dev,
829 static void vrf_rt6_release(struct net_device *dev, struct net_vrf *vrf)
833 static int vrf_rt6_create(struct net_device *dev)
839 /* modelled after ip_finish_output2 */
840 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
842 struct dst_entry *dst = skb_dst(skb);
843 struct rtable *rt = dst_rtable(dst);
844 struct net_device *dev = dst->dev;
845 unsigned int hh_len = LL_RESERVED_SPACE(dev);
846 struct neighbour *neigh;
847 bool is_v6gw = false;
849 vrf_nf_reset_ct(skb);
851 /* Be paranoid, rather than too clever. */
852 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
853 skb = skb_expand_head(skb, hh_len);
855 dev->stats.tx_errors++;
862 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
863 if (!IS_ERR(neigh)) {
866 sock_confirm_neigh(skb, neigh);
867 /* if crossing protocols, can not use the cached header */
868 ret = neigh_output(neigh, skb, is_v6gw);
874 vrf_tx_error(skb->dev, skb);
878 static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
880 struct net_device *dev = skb_dst(skb)->dev;
882 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
885 skb->protocol = htons(ETH_P_IP);
887 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
888 net, sk, skb, NULL, dev,
890 !(IPCB(skb)->flags & IPSKB_REROUTED));
893 /* set dst on skb to send packet to us via dev_xmit path. Allows
894 * packet to go through device based features such as qdisc, netfilter
895 * hooks and packet sockets with skb->dev set to vrf device.
897 static struct sk_buff *vrf_ip_out_redirect(struct net_device *vrf_dev,
900 struct net_vrf *vrf = netdev_priv(vrf_dev);
901 struct dst_entry *dst = NULL;
906 rth = rcu_dereference(vrf->rth);
914 if (unlikely(!dst)) {
915 vrf_tx_error(vrf_dev, skb);
920 skb_dst_set(skb, dst);
925 static int vrf_output_direct_finish(struct net *net, struct sock *sk,
928 vrf_finish_direct(skb);
930 return vrf_ip_local_out(net, sk, skb);
933 static int vrf_output_direct(struct net *net, struct sock *sk,
938 skb->protocol = htons(ETH_P_IP);
940 if (!(IPCB(skb)->flags & IPSKB_REROUTED))
941 err = nf_hook(NFPROTO_IPV4, NF_INET_POST_ROUTING, net, sk, skb,
942 NULL, skb->dev, vrf_output_direct_finish);
944 if (likely(err == 1))
945 vrf_finish_direct(skb);
950 static int vrf_ip_out_direct_finish(struct net *net, struct sock *sk,
955 err = vrf_output_direct(net, sk, skb);
956 if (likely(err == 1))
957 err = vrf_ip_local_out(net, sk, skb);
962 static struct sk_buff *vrf_ip_out_direct(struct net_device *vrf_dev,
966 struct net *net = dev_net(vrf_dev);
971 err = nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, net, sk,
972 skb, NULL, vrf_dev, vrf_ip_out_direct_finish);
974 if (likely(err == 1))
975 err = vrf_output_direct(net, sk, skb);
977 if (likely(err == 1))
983 static struct sk_buff *vrf_ip_out(struct net_device *vrf_dev,
987 /* don't divert multicast or local broadcast */
988 if (ipv4_is_multicast(ip_hdr(skb)->daddr) ||
989 ipv4_is_lbcast(ip_hdr(skb)->daddr))
992 vrf_nf_set_untracked(skb);
994 if (qdisc_tx_is_default(vrf_dev) ||
995 IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
996 return vrf_ip_out_direct(vrf_dev, sk, skb);
998 return vrf_ip_out_redirect(vrf_dev, skb);
1001 /* called with rcu lock held */
1002 static struct sk_buff *vrf_l3_out(struct net_device *vrf_dev,
1004 struct sk_buff *skb,
1009 return vrf_ip_out(vrf_dev, sk, skb);
1011 return vrf_ip6_out(vrf_dev, sk, skb);
1018 static void vrf_rtable_release(struct net_device *dev, struct net_vrf *vrf)
1020 struct rtable *rth = rtnl_dereference(vrf->rth);
1021 struct net *net = dev_net(dev);
1022 struct dst_entry *dst;
1024 RCU_INIT_POINTER(vrf->rth, NULL);
1027 /* move dev in dst's to loopback so this VRF device can be deleted
1028 * - based on dst_ifdown
1032 netdev_ref_replace(dst->dev, net->loopback_dev,
1033 &dst->dev_tracker, GFP_KERNEL);
1034 dst->dev = net->loopback_dev;
1039 static int vrf_rtable_create(struct net_device *dev)
1041 struct net_vrf *vrf = netdev_priv(dev);
1044 if (!fib_new_table(dev_net(dev), vrf->tb_id))
1047 /* create a dst for routing packets out through a VRF device */
1048 rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1);
1052 rth->dst.output = vrf_output;
1054 rcu_assign_pointer(vrf->rth, rth);
1059 /**************************** device handling ********************/
1061 /* cycle interface to flush neighbor cache and move routes across tables */
1062 static void cycle_netdev(struct net_device *dev,
1063 struct netlink_ext_ack *extack)
1065 unsigned int flags = dev->flags;
1068 if (!netif_running(dev))
1071 ret = dev_change_flags(dev, flags & ~IFF_UP, extack);
1073 ret = dev_change_flags(dev, flags, extack);
1077 "Failed to cycle device %s; route tables might be wrong!\n",
1082 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1083 struct netlink_ext_ack *extack)
1087 /* do not allow loopback device to be enslaved to a VRF.
1088 * The vrf device acts as the loopback for the vrf.
1090 if (port_dev == dev_net(dev)->loopback_dev) {
1091 NL_SET_ERR_MSG(extack,
1092 "Can not enslave loopback device to a VRF");
1096 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
1097 ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL, extack);
1101 cycle_netdev(port_dev, extack);
1106 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1110 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev,
1111 struct netlink_ext_ack *extack)
1113 if (netif_is_l3_master(port_dev)) {
1114 NL_SET_ERR_MSG(extack,
1115 "Can not enslave an L3 master device to a VRF");
1119 if (netif_is_l3_slave(port_dev))
1122 return do_vrf_add_slave(dev, port_dev, extack);
1125 /* inverse of do_vrf_add_slave */
1126 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1128 netdev_upper_dev_unlink(port_dev, dev);
1129 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
1131 cycle_netdev(port_dev, NULL);
1136 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
1138 return do_vrf_del_slave(dev, port_dev);
1141 static void vrf_dev_uninit(struct net_device *dev)
1143 struct net_vrf *vrf = netdev_priv(dev);
1145 vrf_rtable_release(dev, vrf);
1146 vrf_rt6_release(dev, vrf);
1149 static int vrf_dev_init(struct net_device *dev)
1151 struct net_vrf *vrf = netdev_priv(dev);
1153 /* create the default dst which points back to us */
1154 if (vrf_rtable_create(dev) != 0)
1157 if (vrf_rt6_create(dev) != 0)
1160 dev->flags = IFF_MASTER | IFF_NOARP;
1162 /* similarly, oper state is irrelevant; set to up to avoid confusion */
1163 dev->operstate = IF_OPER_UP;
1164 netdev_lockdep_set_classes(dev);
1168 vrf_rtable_release(dev, vrf);
1173 static const struct net_device_ops vrf_netdev_ops = {
1174 .ndo_init = vrf_dev_init,
1175 .ndo_uninit = vrf_dev_uninit,
1176 .ndo_start_xmit = vrf_xmit,
1177 .ndo_set_mac_address = eth_mac_addr,
1178 .ndo_add_slave = vrf_add_slave,
1179 .ndo_del_slave = vrf_del_slave,
1182 static u32 vrf_fib_table(const struct net_device *dev)
1184 struct net_vrf *vrf = netdev_priv(dev);
1189 static int vrf_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
1195 static struct sk_buff *vrf_rcv_nfhook(u8 pf, unsigned int hook,
1196 struct sk_buff *skb,
1197 struct net_device *dev)
1199 struct net *net = dev_net(dev);
1201 if (nf_hook(pf, hook, net, NULL, skb, dev, NULL, vrf_rcv_finish) != 1)
1202 skb = NULL; /* kfree_skb(skb) handled by nf code */
1207 static int vrf_prepare_mac_header(struct sk_buff *skb,
1208 struct net_device *vrf_dev, u16 proto)
1213 /* in general, we do not know if there is enough space in the head of
1214 * the packet for hosting the mac header.
1216 err = skb_cow_head(skb, LL_RESERVED_SPACE(vrf_dev));
1218 /* no space in the skb head */
1221 __skb_push(skb, ETH_HLEN);
1222 eth = (struct ethhdr *)skb->data;
1224 skb_reset_mac_header(skb);
1225 skb_reset_mac_len(skb);
1227 /* we set the ethernet destination and the source addresses to the
1228 * address of the VRF device.
1230 ether_addr_copy(eth->h_dest, vrf_dev->dev_addr);
1231 ether_addr_copy(eth->h_source, vrf_dev->dev_addr);
1232 eth->h_proto = htons(proto);
1234 /* the destination address of the Ethernet frame corresponds to the
1235 * address set on the VRF interface; therefore, the packet is intended
1236 * to be processed locally.
1238 skb->protocol = eth->h_proto;
1239 skb->pkt_type = PACKET_HOST;
1241 skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
1243 skb_pull_inline(skb, ETH_HLEN);
1248 /* prepare and add the mac header to the packet if it was not set previously.
1249 * In this way, packet sniffers such as tcpdump can parse the packet correctly.
1250 * If the mac header was already set, the original mac header is left
1251 * untouched and the function returns immediately.
1253 static int vrf_add_mac_header_if_unset(struct sk_buff *skb,
1254 struct net_device *vrf_dev,
1255 u16 proto, struct net_device *orig_dev)
1257 if (skb_mac_header_was_set(skb) && dev_has_header(orig_dev))
1260 return vrf_prepare_mac_header(skb, vrf_dev, proto);
1263 #if IS_ENABLED(CONFIG_IPV6)
1264 /* neighbor handling is done with actual device; do not want
1265 * to flip skb->dev for those ndisc packets. This really fails
1266 * for multiple next protocols (e.g., NEXTHDR_HOP). But it is
1269 static bool ipv6_ndisc_frame(const struct sk_buff *skb)
1271 const struct ipv6hdr *iph = ipv6_hdr(skb);
1274 if (iph->nexthdr == NEXTHDR_ICMP) {
1275 const struct icmp6hdr *icmph;
1276 struct icmp6hdr _icmph;
1278 icmph = skb_header_pointer(skb, sizeof(*iph),
1279 sizeof(_icmph), &_icmph);
1283 switch (icmph->icmp6_type) {
1284 case NDISC_ROUTER_SOLICITATION:
1285 case NDISC_ROUTER_ADVERTISEMENT:
1286 case NDISC_NEIGHBOUR_SOLICITATION:
1287 case NDISC_NEIGHBOUR_ADVERTISEMENT:
1288 case NDISC_REDIRECT:
1298 static struct rt6_info *vrf_ip6_route_lookup(struct net *net,
1299 const struct net_device *dev,
1302 const struct sk_buff *skb,
1305 struct net_vrf *vrf = netdev_priv(dev);
1307 return ip6_pol_route(net, vrf->fib6_table, ifindex, fl6, skb, flags);
1310 static void vrf_ip6_input_dst(struct sk_buff *skb, struct net_device *vrf_dev,
1313 const struct ipv6hdr *iph = ipv6_hdr(skb);
1314 struct flowi6 fl6 = {
1315 .flowi6_iif = ifindex,
1316 .flowi6_mark = skb->mark,
1317 .flowi6_proto = iph->nexthdr,
1318 .daddr = iph->daddr,
1319 .saddr = iph->saddr,
1320 .flowlabel = ip6_flowinfo(iph),
1322 struct net *net = dev_net(vrf_dev);
1323 struct rt6_info *rt6;
1325 rt6 = vrf_ip6_route_lookup(net, vrf_dev, &fl6, ifindex, skb,
1326 RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_IFACE);
1330 if (unlikely(&rt6->dst == &net->ipv6.ip6_null_entry->dst))
1333 skb_dst_set(skb, &rt6->dst);
1336 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1337 struct sk_buff *skb)
1339 int orig_iif = skb->skb_iif;
1340 bool need_strict = rt6_need_strict(&ipv6_hdr(skb)->daddr);
1341 bool is_ndisc = ipv6_ndisc_frame(skb);
1343 /* loopback, multicast & non-ND link-local traffic; do not push through
1344 * packet taps again. Reset pkt_type for upper layers to process skb.
1345 * For non-loopback strict packets, determine the dst using the original
1348 if (skb->pkt_type == PACKET_LOOPBACK || (need_strict && !is_ndisc)) {
1350 skb->skb_iif = vrf_dev->ifindex;
1351 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1353 if (skb->pkt_type == PACKET_LOOPBACK)
1354 skb->pkt_type = PACKET_HOST;
1356 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1361 /* if packet is NDISC then keep the ingress interface */
1363 struct net_device *orig_dev = skb->dev;
1365 vrf_rx_stats(vrf_dev, skb->len);
1367 skb->skb_iif = vrf_dev->ifindex;
1369 if (!list_empty(&vrf_dev->ptype_all)) {
1372 err = vrf_add_mac_header_if_unset(skb, vrf_dev,
1376 skb_push(skb, skb->mac_len);
1377 dev_queue_xmit_nit(skb, vrf_dev);
1378 skb_pull(skb, skb->mac_len);
1382 IP6CB(skb)->flags |= IP6SKB_L3SLAVE;
1386 vrf_ip6_input_dst(skb, vrf_dev, orig_iif);
1388 skb = vrf_rcv_nfhook(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, vrf_dev);
1394 static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev,
1395 struct sk_buff *skb)
1401 static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev,
1402 struct sk_buff *skb)
1404 struct net_device *orig_dev = skb->dev;
1407 skb->skb_iif = vrf_dev->ifindex;
1408 IPCB(skb)->flags |= IPSKB_L3SLAVE;
1410 if (ipv4_is_multicast(ip_hdr(skb)->daddr))
1413 /* loopback traffic; do not push through packet taps again.
1414 * Reset pkt_type for upper layers to process skb
1416 if (skb->pkt_type == PACKET_LOOPBACK) {
1417 skb->pkt_type = PACKET_HOST;
1421 vrf_rx_stats(vrf_dev, skb->len);
1423 if (!list_empty(&vrf_dev->ptype_all)) {
1426 err = vrf_add_mac_header_if_unset(skb, vrf_dev, ETH_P_IP,
1429 skb_push(skb, skb->mac_len);
1430 dev_queue_xmit_nit(skb, vrf_dev);
1431 skb_pull(skb, skb->mac_len);
1435 skb = vrf_rcv_nfhook(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, vrf_dev);
1440 /* called with rcu lock held */
1441 static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev,
1442 struct sk_buff *skb,
1447 return vrf_ip_rcv(vrf_dev, skb);
1449 return vrf_ip6_rcv(vrf_dev, skb);
1455 #if IS_ENABLED(CONFIG_IPV6)
1456 /* send to link-local or multicast address via interface enslaved to
1457 * VRF device. Force lookup to VRF table without changing flow struct
1458 * Note: Caller to this function must hold rcu_read_lock() and no refcnt
1459 * is taken on the dst by this function.
1461 static struct dst_entry *vrf_link_scope_lookup(const struct net_device *dev,
1464 struct net *net = dev_net(dev);
1465 int flags = RT6_LOOKUP_F_IFACE | RT6_LOOKUP_F_DST_NOREF;
1466 struct dst_entry *dst = NULL;
1467 struct rt6_info *rt;
1469 /* VRF device does not have a link-local address and
1470 * sending packets to link-local or mcast addresses over
1471 * a VRF device does not make sense
1473 if (fl6->flowi6_oif == dev->ifindex) {
1474 dst = &net->ipv6.ip6_null_entry->dst;
1478 if (!ipv6_addr_any(&fl6->saddr))
1479 flags |= RT6_LOOKUP_F_HAS_SADDR;
1481 rt = vrf_ip6_route_lookup(net, dev, fl6, fl6->flowi6_oif, NULL, flags);
1489 static const struct l3mdev_ops vrf_l3mdev_ops = {
1490 .l3mdev_fib_table = vrf_fib_table,
1491 .l3mdev_l3_rcv = vrf_l3_rcv,
1492 .l3mdev_l3_out = vrf_l3_out,
1493 #if IS_ENABLED(CONFIG_IPV6)
1494 .l3mdev_link_scope_lookup = vrf_link_scope_lookup,
1498 static void vrf_get_drvinfo(struct net_device *dev,
1499 struct ethtool_drvinfo *info)
1501 strscpy(info->driver, DRV_NAME, sizeof(info->driver));
1502 strscpy(info->version, DRV_VERSION, sizeof(info->version));
1505 static const struct ethtool_ops vrf_ethtool_ops = {
1506 .get_drvinfo = vrf_get_drvinfo,
1509 static inline size_t vrf_fib_rule_nl_size(void)
1513 sz = NLMSG_ALIGN(sizeof(struct fib_rule_hdr));
1514 sz += nla_total_size(sizeof(u8)); /* FRA_L3MDEV */
1515 sz += nla_total_size(sizeof(u32)); /* FRA_PRIORITY */
1516 sz += nla_total_size(sizeof(u8)); /* FRA_PROTOCOL */
1521 static int vrf_fib_rule(const struct net_device *dev, __u8 family, bool add_it)
1523 struct fib_rule_hdr *frh;
1524 struct nlmsghdr *nlh;
1525 struct sk_buff *skb;
1528 if ((family == AF_INET6 || family == RTNL_FAMILY_IP6MR) &&
1529 !ipv6_mod_enabled())
1532 skb = nlmsg_new(vrf_fib_rule_nl_size(), GFP_KERNEL);
1536 nlh = nlmsg_put(skb, 0, 0, 0, sizeof(*frh), 0);
1538 goto nla_put_failure;
1540 /* rule only needs to appear once */
1541 nlh->nlmsg_flags |= NLM_F_EXCL;
1543 frh = nlmsg_data(nlh);
1544 memset(frh, 0, sizeof(*frh));
1545 frh->family = family;
1546 frh->action = FR_ACT_TO_TBL;
1548 if (nla_put_u8(skb, FRA_PROTOCOL, RTPROT_KERNEL))
1549 goto nla_put_failure;
1551 if (nla_put_u8(skb, FRA_L3MDEV, 1))
1552 goto nla_put_failure;
1554 if (nla_put_u32(skb, FRA_PRIORITY, FIB_RULE_PREF))
1555 goto nla_put_failure;
1557 nlmsg_end(skb, nlh);
1559 /* fib_nl_{new,del}rule handling looks for net from skb->sk */
1560 skb->sk = dev_net(dev)->rtnl;
1562 err = fib_nl_newrule(skb, nlh, NULL);
1566 err = fib_nl_delrule(skb, nlh, NULL);
1580 static int vrf_add_fib_rules(const struct net_device *dev)
1584 err = vrf_fib_rule(dev, AF_INET, true);
1588 err = vrf_fib_rule(dev, AF_INET6, true);
1592 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1593 err = vrf_fib_rule(dev, RTNL_FAMILY_IPMR, true);
1598 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1599 err = vrf_fib_rule(dev, RTNL_FAMILY_IP6MR, true);
1606 #if IS_ENABLED(CONFIG_IPV6_MROUTE_MULTIPLE_TABLES)
1608 vrf_fib_rule(dev, RTNL_FAMILY_IPMR, false);
1611 #if IS_ENABLED(CONFIG_IP_MROUTE_MULTIPLE_TABLES)
1613 vrf_fib_rule(dev, AF_INET6, false);
1617 vrf_fib_rule(dev, AF_INET, false);
1620 netdev_err(dev, "Failed to add FIB rules.\n");
1624 static void vrf_setup(struct net_device *dev)
1628 /* Initialize the device structure. */
1629 dev->netdev_ops = &vrf_netdev_ops;
1630 dev->l3mdev_ops = &vrf_l3mdev_ops;
1631 dev->ethtool_ops = &vrf_ethtool_ops;
1632 dev->needs_free_netdev = true;
1634 /* Fill in device structure with ethernet-generic values. */
1635 eth_hw_addr_random(dev);
1637 /* don't acquire vrf device's netif_tx_lock when transmitting */
1640 /* don't allow vrf devices to change network namespaces. */
1641 dev->netns_local = true;
1643 /* does not make sense for a VLAN to be added to a vrf device */
1644 dev->features |= NETIF_F_VLAN_CHALLENGED;
1646 /* enable offload features */
1647 dev->features |= NETIF_F_GSO_SOFTWARE;
1648 dev->features |= NETIF_F_RXCSUM | NETIF_F_HW_CSUM | NETIF_F_SCTP_CRC;
1649 dev->features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA;
1651 dev->hw_features = dev->features;
1652 dev->hw_enc_features = dev->features;
1654 /* default to no qdisc; user can add if desired */
1655 dev->priv_flags |= IFF_NO_QUEUE;
1656 dev->priv_flags |= IFF_NO_RX_HANDLER;
1657 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1659 /* VRF devices do not care about MTU, but if the MTU is set
1660 * too low then the ipv4 and ipv6 protocols are disabled
1661 * which breaks networking.
1663 dev->min_mtu = IPV6_MIN_MTU;
1664 dev->max_mtu = IP6_MAX_MTU;
1665 dev->mtu = dev->max_mtu;
1667 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS;
1670 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[],
1671 struct netlink_ext_ack *extack)
1673 if (tb[IFLA_ADDRESS]) {
1674 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1675 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1678 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1679 NL_SET_ERR_MSG(extack, "Invalid hardware address");
1680 return -EADDRNOTAVAIL;
1686 static void vrf_dellink(struct net_device *dev, struct list_head *head)
1688 struct net_device *port_dev;
1689 struct list_head *iter;
1691 netdev_for_each_lower_dev(dev, port_dev, iter)
1692 vrf_del_slave(dev, port_dev);
1694 vrf_map_unregister_dev(dev);
1696 unregister_netdevice_queue(dev, head);
1699 static int vrf_newlink(struct net *src_net, struct net_device *dev,
1700 struct nlattr *tb[], struct nlattr *data[],
1701 struct netlink_ext_ack *extack)
1703 struct net_vrf *vrf = netdev_priv(dev);
1704 struct netns_vrf *nn_vrf;
1705 bool *add_fib_rules;
1709 if (!data || !data[IFLA_VRF_TABLE]) {
1710 NL_SET_ERR_MSG(extack, "VRF table id is missing");
1714 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
1715 if (vrf->tb_id == RT_TABLE_UNSPEC) {
1716 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VRF_TABLE],
1717 "Invalid VRF table id");
1721 dev->priv_flags |= IFF_L3MDEV_MASTER;
1723 err = register_netdevice(dev);
1727 /* mapping between table_id and vrf;
1728 * note: such binding could not be done in the dev init function
1729 * because dev->ifindex id is not available yet.
1731 vrf->ifindex = dev->ifindex;
1733 err = vrf_map_register_dev(dev, extack);
1735 unregister_netdevice(dev);
1740 nn_vrf = net_generic(net, vrf_net_id);
1742 add_fib_rules = &nn_vrf->add_fib_rules;
1743 if (*add_fib_rules) {
1744 err = vrf_add_fib_rules(dev);
1746 vrf_map_unregister_dev(dev);
1747 unregister_netdevice(dev);
1750 *add_fib_rules = false;
1757 static size_t vrf_nl_getsize(const struct net_device *dev)
1759 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
1762 static int vrf_fillinfo(struct sk_buff *skb,
1763 const struct net_device *dev)
1765 struct net_vrf *vrf = netdev_priv(dev);
1767 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
1770 static size_t vrf_get_slave_size(const struct net_device *bond_dev,
1771 const struct net_device *slave_dev)
1773 return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */
1776 static int vrf_fill_slave_info(struct sk_buff *skb,
1777 const struct net_device *vrf_dev,
1778 const struct net_device *slave_dev)
1780 struct net_vrf *vrf = netdev_priv(vrf_dev);
1782 if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id))
1788 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
1789 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
1792 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
1794 .priv_size = sizeof(struct net_vrf),
1796 .get_size = vrf_nl_getsize,
1797 .policy = vrf_nl_policy,
1798 .validate = vrf_validate,
1799 .fill_info = vrf_fillinfo,
1801 .get_slave_size = vrf_get_slave_size,
1802 .fill_slave_info = vrf_fill_slave_info,
1804 .newlink = vrf_newlink,
1805 .dellink = vrf_dellink,
1807 .maxtype = IFLA_VRF_MAX,
1810 static int vrf_device_event(struct notifier_block *unused,
1811 unsigned long event, void *ptr)
1813 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1815 /* only care about unregister events to drop slave references */
1816 if (event == NETDEV_UNREGISTER) {
1817 struct net_device *vrf_dev;
1819 if (!netif_is_l3_slave(dev))
1822 vrf_dev = netdev_master_upper_dev_get(dev);
1823 vrf_del_slave(vrf_dev, dev);
1829 static struct notifier_block vrf_notifier_block __read_mostly = {
1830 .notifier_call = vrf_device_event,
1833 static int vrf_map_init(struct vrf_map *vmap)
1835 spin_lock_init(&vmap->vmap_lock);
1836 hash_init(vmap->ht);
1838 vmap->strict_mode = false;
1843 #ifdef CONFIG_SYSCTL
1844 static bool vrf_strict_mode(struct vrf_map *vmap)
1849 strict_mode = vmap->strict_mode;
1850 vrf_map_unlock(vmap);
1855 static int vrf_strict_mode_change(struct vrf_map *vmap, bool new_mode)
1862 cur_mode = &vmap->strict_mode;
1863 if (*cur_mode == new_mode)
1867 /* disable strict mode */
1870 if (vmap->shared_tables) {
1871 /* we cannot allow strict_mode because there are some
1872 * vrfs that share one or more tables.
1878 /* no tables are shared among vrfs, so we can go back
1879 * to 1:1 association between a vrf with its table.
1885 vrf_map_unlock(vmap);
1890 static int vrf_shared_table_handler(const struct ctl_table *table, int write,
1891 void *buffer, size_t *lenp, loff_t *ppos)
1893 struct net *net = (struct net *)table->extra1;
1894 struct vrf_map *vmap = netns_vrf_map(net);
1895 int proc_strict_mode = 0;
1896 struct ctl_table tmp = {
1897 .procname = table->procname,
1898 .data = &proc_strict_mode,
1899 .maxlen = sizeof(int),
1900 .mode = table->mode,
1901 .extra1 = SYSCTL_ZERO,
1902 .extra2 = SYSCTL_ONE,
1907 proc_strict_mode = vrf_strict_mode(vmap);
1909 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1911 if (write && ret == 0)
1912 ret = vrf_strict_mode_change(vmap, (bool)proc_strict_mode);
1917 static const struct ctl_table vrf_table[] = {
1919 .procname = "strict_mode",
1921 .maxlen = sizeof(int),
1923 .proc_handler = vrf_shared_table_handler,
1924 /* set by the vrf_netns_init */
1929 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1931 struct ctl_table *table;
1933 table = kmemdup(vrf_table, sizeof(vrf_table), GFP_KERNEL);
1937 /* init the extra1 parameter with the reference to current netns */
1938 table[0].extra1 = net;
1940 nn_vrf->ctl_hdr = register_net_sysctl_sz(net, "net/vrf", table,
1941 ARRAY_SIZE(vrf_table));
1942 if (!nn_vrf->ctl_hdr) {
1950 static void vrf_netns_exit_sysctl(struct net *net)
1952 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1953 const struct ctl_table *table;
1955 table = nn_vrf->ctl_hdr->ctl_table_arg;
1956 unregister_net_sysctl_table(nn_vrf->ctl_hdr);
1960 static int vrf_netns_init_sysctl(struct net *net, struct netns_vrf *nn_vrf)
1965 static void vrf_netns_exit_sysctl(struct net *net)
1970 /* Initialize per network namespace state */
1971 static int __net_init vrf_netns_init(struct net *net)
1973 struct netns_vrf *nn_vrf = net_generic(net, vrf_net_id);
1975 nn_vrf->add_fib_rules = true;
1976 vrf_map_init(&nn_vrf->vmap);
1978 return vrf_netns_init_sysctl(net, nn_vrf);
1981 static void __net_exit vrf_netns_exit(struct net *net)
1983 vrf_netns_exit_sysctl(net);
1986 static struct pernet_operations vrf_net_ops __net_initdata = {
1987 .init = vrf_netns_init,
1988 .exit = vrf_netns_exit,
1990 .size = sizeof(struct netns_vrf),
1993 static int __init vrf_init_module(void)
1997 register_netdevice_notifier(&vrf_notifier_block);
1999 rc = register_pernet_subsys(&vrf_net_ops);
2003 rc = l3mdev_table_lookup_register(L3MDEV_TYPE_VRF,
2004 vrf_ifindex_lookup_by_table_id);
2008 rc = rtnl_link_register(&vrf_link_ops);
2010 goto table_lookup_unreg;
2015 l3mdev_table_lookup_unregister(L3MDEV_TYPE_VRF,
2016 vrf_ifindex_lookup_by_table_id);
2019 unregister_pernet_subsys(&vrf_net_ops);
2022 unregister_netdevice_notifier(&vrf_notifier_block);
2026 module_init(vrf_init_module);
2027 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
2028 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
2029 MODULE_LICENSE("GPL");
2030 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2031 MODULE_VERSION(DRV_VERSION);