2 * vrf.c: device driver to encapsulate a VRF space
4 * Copyright (c) 2015 Cumulus Networks. All rights reserved.
5 * Copyright (c) 2015 Shrijeet Mukherjee <shm@cumulusnetworks.com>
6 * Copyright (c) 2015 David Ahern <dsa@cumulusnetworks.com>
8 * Based on dummy, team and ipvlan drivers
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/netdevice.h>
19 #include <linux/etherdevice.h>
21 #include <linux/init.h>
22 #include <linux/moduleparam.h>
23 #include <linux/netfilter.h>
24 #include <linux/rtnetlink.h>
25 #include <net/rtnetlink.h>
26 #include <linux/u64_stats_sync.h>
27 #include <linux/hashtable.h>
29 #include <linux/inetdevice.h>
32 #include <net/ip_fib.h>
33 #include <net/ip6_route.h>
34 #include <net/rtnetlink.h>
35 #include <net/route.h>
36 #include <net/addrconf.h>
39 #define DRV_NAME "vrf"
40 #define DRV_VERSION "1.0"
42 #define vrf_is_slave(dev) ((dev)->flags & IFF_SLAVE)
44 #define vrf_master_get_rcu(dev) \
45 ((struct net_device *)rcu_dereference(dev->rx_handler_data))
53 struct u64_stats_sync syncp;
56 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
61 static int vrf_ip_local_out(struct sk_buff *skb)
63 return ip_local_out(skb);
66 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
68 /* TO-DO: return max ethernet size? */
72 static void vrf_dst_destroy(struct dst_entry *dst)
74 /* our dst lives forever - or until the device is closed */
77 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
82 static struct dst_ops vrf_dst_ops = {
84 .local_out = vrf_ip_local_out,
85 .check = vrf_ip_check,
87 .destroy = vrf_dst_destroy,
88 .default_advmss = vrf_default_advmss,
91 static bool is_ip_rx_frame(struct sk_buff *skb)
93 switch (skb->protocol) {
95 case htons(ETH_P_IPV6):
101 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
103 vrf_dev->stats.tx_errors++;
107 /* note: already called with rcu_read_lock */
108 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
110 struct sk_buff *skb = *pskb;
112 if (is_ip_rx_frame(skb)) {
113 struct net_device *dev = vrf_master_get_rcu(skb->dev);
114 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
116 u64_stats_update_begin(&dstats->syncp);
118 dstats->rx_bytes += skb->len;
119 u64_stats_update_end(&dstats->syncp);
123 return RX_HANDLER_ANOTHER;
125 return RX_HANDLER_PASS;
128 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
129 struct rtnl_link_stats64 *stats)
133 for_each_possible_cpu(i) {
134 const struct pcpu_dstats *dstats;
135 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
138 dstats = per_cpu_ptr(dev->dstats, i);
140 start = u64_stats_fetch_begin_irq(&dstats->syncp);
141 tbytes = dstats->tx_bytes;
142 tpkts = dstats->tx_pkts;
143 tdrops = dstats->tx_drps;
144 rbytes = dstats->rx_bytes;
145 rpkts = dstats->rx_pkts;
146 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
147 stats->tx_bytes += tbytes;
148 stats->tx_packets += tpkts;
149 stats->tx_dropped += tdrops;
150 stats->rx_bytes += rbytes;
151 stats->rx_packets += rpkts;
156 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
157 struct net_device *dev)
159 vrf_tx_error(dev, skb);
160 return NET_XMIT_DROP;
163 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
164 struct net_device *vrf_dev)
169 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
173 /* TO-DO: what about broadcast ? */
174 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
180 skb_dst_set(skb, &rt->dst);
186 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
187 struct net_device *vrf_dev)
189 struct iphdr *ip4h = ip_hdr(skb);
190 int ret = NET_XMIT_DROP;
191 struct flowi4 fl4 = {
192 /* needed to match OIF rule */
193 .flowi4_oif = vrf_dev->ifindex,
194 .flowi4_iif = LOOPBACK_IFINDEX,
195 .flowi4_tos = RT_TOS(ip4h->tos),
196 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_VRFSRC |
197 FLOWI_FLAG_SKIP_NH_OIF,
198 .daddr = ip4h->daddr,
201 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
205 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
209 ret = ip_local_out(skb);
210 if (unlikely(net_xmit_eval(ret)))
211 vrf_dev->stats.tx_errors++;
213 ret = NET_XMIT_SUCCESS;
218 vrf_tx_error(vrf_dev, skb);
222 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
224 /* strip the ethernet header added for pass through VRF device */
225 __skb_pull(skb, skb_network_offset(skb));
227 switch (skb->protocol) {
228 case htons(ETH_P_IP):
229 return vrf_process_v4_outbound(skb, dev);
230 case htons(ETH_P_IPV6):
231 return vrf_process_v6_outbound(skb, dev);
233 vrf_tx_error(dev, skb);
234 return NET_XMIT_DROP;
238 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
240 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
242 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
243 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
245 u64_stats_update_begin(&dstats->syncp);
247 dstats->tx_bytes += skb->len;
248 u64_stats_update_end(&dstats->syncp);
250 this_cpu_inc(dev->dstats->tx_drps);
256 /* modelled after ip_finish_output2 */
257 static int vrf_finish_output(struct sock *sk, struct sk_buff *skb)
259 struct dst_entry *dst = skb_dst(skb);
260 struct rtable *rt = (struct rtable *)dst;
261 struct net_device *dev = dst->dev;
262 unsigned int hh_len = LL_RESERVED_SPACE(dev);
263 struct neighbour *neigh;
267 /* Be paranoid, rather than too clever. */
268 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
269 struct sk_buff *skb2;
271 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
277 skb_set_owner_w(skb2, skb->sk);
285 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
286 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
287 if (unlikely(!neigh))
288 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
290 ret = dst_neigh_output(dst, neigh, skb);
292 rcu_read_unlock_bh();
294 if (unlikely(ret < 0))
295 vrf_tx_error(skb->dev, skb);
299 static int vrf_output(struct sock *sk, struct sk_buff *skb)
301 struct net_device *dev = skb_dst(skb)->dev;
303 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
306 skb->protocol = htons(ETH_P_IP);
308 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb,
311 !(IPCB(skb)->flags & IPSKB_REROUTED));
314 static void vrf_rtable_destroy(struct net_vrf *vrf)
316 struct dst_entry *dst = (struct dst_entry *)vrf->rth;
322 static struct rtable *vrf_rtable_create(struct net_device *dev)
326 rth = dst_alloc(&vrf_dst_ops, dev, 2,
328 (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
330 rth->dst.output = vrf_output;
331 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
333 rth->rt_type = RTN_UNICAST;
334 rth->rt_is_input = 0;
338 rth->rt_uses_gateway = 0;
339 INIT_LIST_HEAD(&rth->rt_uncached);
340 rth->rt_uncached_list = NULL;
346 /**************************** device handling ********************/
348 /* cycle interface to flush neighbor cache and move routes across tables */
349 static void cycle_netdev(struct net_device *dev)
351 unsigned int flags = dev->flags;
354 if (!netif_running(dev))
357 ret = dev_change_flags(dev, flags & ~IFF_UP);
359 ret = dev_change_flags(dev, flags);
363 "Failed to cycle device %s; route tables might be wrong!\n",
368 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
369 struct net_device *dev)
371 struct list_head *head = &queue->all_slaves;
374 list_for_each_entry(slave, head, list) {
375 if (slave->dev == dev)
382 /* inverse of __vrf_insert_slave */
383 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
385 list_del(&slave->list);
388 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
390 list_add(&slave->list, &queue->all_slaves);
393 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
395 struct net_vrf_dev *vrf_ptr = kmalloc(sizeof(*vrf_ptr), GFP_KERNEL);
396 struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
397 struct net_vrf *vrf = netdev_priv(dev);
398 struct slave_queue *queue = &vrf->queue;
401 if (!slave || !vrf_ptr)
404 slave->dev = port_dev;
405 vrf_ptr->ifindex = dev->ifindex;
406 vrf_ptr->tb_id = vrf->tb_id;
408 /* register the packet handler for slave ports */
409 ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
412 "Device %s failed to register rx_handler\n",
417 ret = netdev_master_upper_dev_link(port_dev, dev);
421 port_dev->flags |= IFF_SLAVE;
422 __vrf_insert_slave(queue, slave);
423 rcu_assign_pointer(port_dev->vrf_ptr, vrf_ptr);
424 cycle_netdev(port_dev);
429 netdev_rx_handler_unregister(port_dev);
436 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
438 if (netif_is_vrf(port_dev) || vrf_is_slave(port_dev))
441 return do_vrf_add_slave(dev, port_dev);
444 /* inverse of do_vrf_add_slave */
445 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
447 struct net_vrf_dev *vrf_ptr = rtnl_dereference(port_dev->vrf_ptr);
448 struct net_vrf *vrf = netdev_priv(dev);
449 struct slave_queue *queue = &vrf->queue;
452 RCU_INIT_POINTER(port_dev->vrf_ptr, NULL);
454 netdev_upper_dev_unlink(port_dev, dev);
455 port_dev->flags &= ~IFF_SLAVE;
457 netdev_rx_handler_unregister(port_dev);
459 /* after netdev_rx_handler_unregister for synchronize_rcu */
462 cycle_netdev(port_dev);
464 slave = __vrf_find_slave_dev(queue, port_dev);
466 __vrf_remove_slave(queue, slave);
473 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
475 return do_vrf_del_slave(dev, port_dev);
478 static void vrf_dev_uninit(struct net_device *dev)
480 struct net_vrf *vrf = netdev_priv(dev);
481 struct slave_queue *queue = &vrf->queue;
482 struct list_head *head = &queue->all_slaves;
483 struct slave *slave, *next;
485 vrf_rtable_destroy(vrf);
487 list_for_each_entry_safe(slave, next, head, list)
488 vrf_del_slave(dev, slave->dev);
490 free_percpu(dev->dstats);
494 static int vrf_dev_init(struct net_device *dev)
496 struct net_vrf *vrf = netdev_priv(dev);
498 INIT_LIST_HEAD(&vrf->queue.all_slaves);
500 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
504 /* create the default dst which points back to us */
505 vrf->rth = vrf_rtable_create(dev);
509 dev->flags = IFF_MASTER | IFF_NOARP;
514 free_percpu(dev->dstats);
520 static const struct net_device_ops vrf_netdev_ops = {
521 .ndo_init = vrf_dev_init,
522 .ndo_uninit = vrf_dev_uninit,
523 .ndo_start_xmit = vrf_xmit,
524 .ndo_get_stats64 = vrf_get_stats64,
525 .ndo_add_slave = vrf_add_slave,
526 .ndo_del_slave = vrf_del_slave,
529 static void vrf_get_drvinfo(struct net_device *dev,
530 struct ethtool_drvinfo *info)
532 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
533 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
536 static const struct ethtool_ops vrf_ethtool_ops = {
537 .get_drvinfo = vrf_get_drvinfo,
540 static void vrf_setup(struct net_device *dev)
544 /* Initialize the device structure. */
545 dev->netdev_ops = &vrf_netdev_ops;
546 dev->ethtool_ops = &vrf_ethtool_ops;
547 dev->destructor = free_netdev;
549 /* Fill in device structure with ethernet-generic values. */
550 eth_hw_addr_random(dev);
552 /* don't acquire vrf device's netif_tx_lock when transmitting */
553 dev->features |= NETIF_F_LLTX;
555 /* don't allow vrf devices to change network namespaces. */
556 dev->features |= NETIF_F_NETNS_LOCAL;
559 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
561 if (tb[IFLA_ADDRESS]) {
562 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
564 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
565 return -EADDRNOTAVAIL;
570 static void vrf_dellink(struct net_device *dev, struct list_head *head)
572 struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr);
574 RCU_INIT_POINTER(dev->vrf_ptr, NULL);
575 kfree_rcu(vrf_ptr, rcu);
576 unregister_netdevice_queue(dev, head);
579 static int vrf_newlink(struct net *src_net, struct net_device *dev,
580 struct nlattr *tb[], struct nlattr *data[])
582 struct net_vrf *vrf = netdev_priv(dev);
583 struct net_vrf_dev *vrf_ptr;
586 if (!data || !data[IFLA_VRF_TABLE])
589 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
591 dev->priv_flags |= IFF_VRF_MASTER;
594 vrf_ptr = kmalloc(sizeof(*dev->vrf_ptr), GFP_KERNEL);
598 vrf_ptr->ifindex = dev->ifindex;
599 vrf_ptr->tb_id = vrf->tb_id;
601 err = register_netdevice(dev);
605 rcu_assign_pointer(dev->vrf_ptr, vrf_ptr);
615 static size_t vrf_nl_getsize(const struct net_device *dev)
617 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
620 static int vrf_fillinfo(struct sk_buff *skb,
621 const struct net_device *dev)
623 struct net_vrf *vrf = netdev_priv(dev);
625 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
628 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
629 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
632 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
634 .priv_size = sizeof(struct net_vrf),
636 .get_size = vrf_nl_getsize,
637 .policy = vrf_nl_policy,
638 .validate = vrf_validate,
639 .fill_info = vrf_fillinfo,
641 .newlink = vrf_newlink,
642 .dellink = vrf_dellink,
644 .maxtype = IFLA_VRF_MAX,
647 static int vrf_device_event(struct notifier_block *unused,
648 unsigned long event, void *ptr)
650 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
652 /* only care about unregister events to drop slave references */
653 if (event == NETDEV_UNREGISTER) {
654 struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr);
655 struct net_device *vrf_dev;
657 if (!vrf_ptr || netif_is_vrf(dev))
660 vrf_dev = netdev_master_upper_dev_get(dev);
661 vrf_del_slave(vrf_dev, dev);
667 static struct notifier_block vrf_notifier_block __read_mostly = {
668 .notifier_call = vrf_device_event,
671 static int __init vrf_init_module(void)
675 vrf_dst_ops.kmem_cachep =
676 kmem_cache_create("vrf_ip_dst_cache",
677 sizeof(struct rtable), 0,
681 if (!vrf_dst_ops.kmem_cachep)
684 register_netdevice_notifier(&vrf_notifier_block);
686 rc = rtnl_link_register(&vrf_link_ops);
693 unregister_netdevice_notifier(&vrf_notifier_block);
694 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
698 static void __exit vrf_cleanup_module(void)
700 rtnl_link_unregister(&vrf_link_ops);
701 unregister_netdevice_notifier(&vrf_notifier_block);
702 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
705 module_init(vrf_init_module);
706 module_exit(vrf_cleanup_module);
707 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
708 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
709 MODULE_LICENSE("GPL");
710 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
711 MODULE_VERSION(DRV_VERSION);