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>
37 #include <net/l3mdev.h>
39 #define DRV_NAME "vrf"
40 #define DRV_VERSION "1.0"
42 #define vrf_master_get_rcu(dev) \
43 ((struct net_device *)rcu_dereference(dev->rx_handler_data))
46 struct list_head list;
47 struct net_device *dev;
51 struct list_head all_slaves;
55 struct slave_queue queue;
66 struct u64_stats_sync syncp;
69 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
74 static int vrf_ip_local_out(struct sk_buff *skb)
76 return ip_local_out(skb);
79 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
81 /* TO-DO: return max ethernet size? */
85 static void vrf_dst_destroy(struct dst_entry *dst)
87 /* our dst lives forever - or until the device is closed */
90 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
95 static struct dst_ops vrf_dst_ops = {
97 .local_out = vrf_ip_local_out,
98 .check = vrf_ip_check,
100 .destroy = vrf_dst_destroy,
101 .default_advmss = vrf_default_advmss,
104 static bool is_ip_rx_frame(struct sk_buff *skb)
106 switch (skb->protocol) {
107 case htons(ETH_P_IP):
108 case htons(ETH_P_IPV6):
114 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
116 vrf_dev->stats.tx_errors++;
120 /* note: already called with rcu_read_lock */
121 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
123 struct sk_buff *skb = *pskb;
125 if (is_ip_rx_frame(skb)) {
126 struct net_device *dev = vrf_master_get_rcu(skb->dev);
127 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
129 u64_stats_update_begin(&dstats->syncp);
131 dstats->rx_bytes += skb->len;
132 u64_stats_update_end(&dstats->syncp);
136 return RX_HANDLER_ANOTHER;
138 return RX_HANDLER_PASS;
141 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
142 struct rtnl_link_stats64 *stats)
146 for_each_possible_cpu(i) {
147 const struct pcpu_dstats *dstats;
148 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
151 dstats = per_cpu_ptr(dev->dstats, i);
153 start = u64_stats_fetch_begin_irq(&dstats->syncp);
154 tbytes = dstats->tx_bytes;
155 tpkts = dstats->tx_pkts;
156 tdrops = dstats->tx_drps;
157 rbytes = dstats->rx_bytes;
158 rpkts = dstats->rx_pkts;
159 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
160 stats->tx_bytes += tbytes;
161 stats->tx_packets += tpkts;
162 stats->tx_dropped += tdrops;
163 stats->rx_bytes += rbytes;
164 stats->rx_packets += rpkts;
169 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
170 struct net_device *dev)
172 vrf_tx_error(dev, skb);
173 return NET_XMIT_DROP;
176 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
177 struct net_device *vrf_dev)
182 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
186 /* TO-DO: what about broadcast ? */
187 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
193 skb_dst_set(skb, &rt->dst);
199 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
200 struct net_device *vrf_dev)
202 struct iphdr *ip4h = ip_hdr(skb);
203 int ret = NET_XMIT_DROP;
204 struct flowi4 fl4 = {
205 /* needed to match OIF rule */
206 .flowi4_oif = vrf_dev->ifindex,
207 .flowi4_iif = LOOPBACK_IFINDEX,
208 .flowi4_tos = RT_TOS(ip4h->tos),
209 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC |
210 FLOWI_FLAG_SKIP_NH_OIF,
211 .daddr = ip4h->daddr,
214 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
218 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
222 ret = ip_local_out(skb);
223 if (unlikely(net_xmit_eval(ret)))
224 vrf_dev->stats.tx_errors++;
226 ret = NET_XMIT_SUCCESS;
231 vrf_tx_error(vrf_dev, skb);
235 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
237 /* strip the ethernet header added for pass through VRF device */
238 __skb_pull(skb, skb_network_offset(skb));
240 switch (skb->protocol) {
241 case htons(ETH_P_IP):
242 return vrf_process_v4_outbound(skb, dev);
243 case htons(ETH_P_IPV6):
244 return vrf_process_v6_outbound(skb, dev);
246 vrf_tx_error(dev, skb);
247 return NET_XMIT_DROP;
251 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
253 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
255 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
256 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
258 u64_stats_update_begin(&dstats->syncp);
260 dstats->tx_bytes += skb->len;
261 u64_stats_update_end(&dstats->syncp);
263 this_cpu_inc(dev->dstats->tx_drps);
269 /* modelled after ip_finish_output2 */
270 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
272 struct dst_entry *dst = skb_dst(skb);
273 struct rtable *rt = (struct rtable *)dst;
274 struct net_device *dev = dst->dev;
275 unsigned int hh_len = LL_RESERVED_SPACE(dev);
276 struct neighbour *neigh;
280 /* Be paranoid, rather than too clever. */
281 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
282 struct sk_buff *skb2;
284 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
290 skb_set_owner_w(skb2, skb->sk);
298 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
299 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
300 if (unlikely(!neigh))
301 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
303 ret = dst_neigh_output(dst, neigh, skb);
305 rcu_read_unlock_bh();
307 if (unlikely(ret < 0))
308 vrf_tx_error(skb->dev, skb);
312 static int vrf_output(struct sock *sk, struct sk_buff *skb)
314 struct net_device *dev = skb_dst(skb)->dev;
315 struct net *net = dev_net(dev);
317 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
320 skb->protocol = htons(ETH_P_IP);
322 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
323 net, sk, skb, NULL, dev,
325 !(IPCB(skb)->flags & IPSKB_REROUTED));
328 static void vrf_rtable_destroy(struct net_vrf *vrf)
330 struct dst_entry *dst = (struct dst_entry *)vrf->rth;
336 static struct rtable *vrf_rtable_create(struct net_device *dev)
338 struct net_vrf *vrf = netdev_priv(dev);
341 rth = dst_alloc(&vrf_dst_ops, dev, 2,
343 (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
345 rth->dst.output = vrf_output;
346 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
348 rth->rt_type = RTN_UNICAST;
349 rth->rt_is_input = 0;
353 rth->rt_uses_gateway = 0;
354 rth->rt_table_id = vrf->tb_id;
355 INIT_LIST_HEAD(&rth->rt_uncached);
356 rth->rt_uncached_list = NULL;
362 /**************************** device handling ********************/
364 /* cycle interface to flush neighbor cache and move routes across tables */
365 static void cycle_netdev(struct net_device *dev)
367 unsigned int flags = dev->flags;
370 if (!netif_running(dev))
373 ret = dev_change_flags(dev, flags & ~IFF_UP);
375 ret = dev_change_flags(dev, flags);
379 "Failed to cycle device %s; route tables might be wrong!\n",
384 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
385 struct net_device *dev)
387 struct list_head *head = &queue->all_slaves;
390 list_for_each_entry(slave, head, list) {
391 if (slave->dev == dev)
398 /* inverse of __vrf_insert_slave */
399 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
401 list_del(&slave->list);
404 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
406 list_add(&slave->list, &queue->all_slaves);
409 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
411 struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
412 struct net_vrf *vrf = netdev_priv(dev);
413 struct slave_queue *queue = &vrf->queue;
419 slave->dev = port_dev;
421 /* register the packet handler for slave ports */
422 ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
425 "Device %s failed to register rx_handler\n",
430 ret = netdev_master_upper_dev_link(port_dev, dev);
434 port_dev->priv_flags |= IFF_L3MDEV_SLAVE;
435 __vrf_insert_slave(queue, slave);
436 cycle_netdev(port_dev);
441 netdev_rx_handler_unregister(port_dev);
447 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
449 if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev))
452 return do_vrf_add_slave(dev, port_dev);
455 /* inverse of do_vrf_add_slave */
456 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
458 struct net_vrf *vrf = netdev_priv(dev);
459 struct slave_queue *queue = &vrf->queue;
462 netdev_upper_dev_unlink(port_dev, dev);
463 port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE;
465 netdev_rx_handler_unregister(port_dev);
467 cycle_netdev(port_dev);
469 slave = __vrf_find_slave_dev(queue, port_dev);
471 __vrf_remove_slave(queue, slave);
478 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
480 return do_vrf_del_slave(dev, port_dev);
483 static void vrf_dev_uninit(struct net_device *dev)
485 struct net_vrf *vrf = netdev_priv(dev);
486 struct slave_queue *queue = &vrf->queue;
487 struct list_head *head = &queue->all_slaves;
488 struct slave *slave, *next;
490 vrf_rtable_destroy(vrf);
492 list_for_each_entry_safe(slave, next, head, list)
493 vrf_del_slave(dev, slave->dev);
495 free_percpu(dev->dstats);
499 static int vrf_dev_init(struct net_device *dev)
501 struct net_vrf *vrf = netdev_priv(dev);
503 INIT_LIST_HEAD(&vrf->queue.all_slaves);
505 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
509 /* create the default dst which points back to us */
510 vrf->rth = vrf_rtable_create(dev);
514 dev->flags = IFF_MASTER | IFF_NOARP;
519 free_percpu(dev->dstats);
525 static const struct net_device_ops vrf_netdev_ops = {
526 .ndo_init = vrf_dev_init,
527 .ndo_uninit = vrf_dev_uninit,
528 .ndo_start_xmit = vrf_xmit,
529 .ndo_get_stats64 = vrf_get_stats64,
530 .ndo_add_slave = vrf_add_slave,
531 .ndo_del_slave = vrf_del_slave,
534 static u32 vrf_fib_table(const struct net_device *dev)
536 struct net_vrf *vrf = netdev_priv(dev);
541 static struct rtable *vrf_get_rtable(const struct net_device *dev,
542 const struct flowi4 *fl4)
544 struct rtable *rth = NULL;
546 if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) {
547 struct net_vrf *vrf = netdev_priv(dev);
550 atomic_inc(&rth->dst.__refcnt);
556 static const struct l3mdev_ops vrf_l3mdev_ops = {
557 .l3mdev_fib_table = vrf_fib_table,
558 .l3mdev_get_rtable = vrf_get_rtable,
561 static void vrf_get_drvinfo(struct net_device *dev,
562 struct ethtool_drvinfo *info)
564 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
565 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
568 static const struct ethtool_ops vrf_ethtool_ops = {
569 .get_drvinfo = vrf_get_drvinfo,
572 static void vrf_setup(struct net_device *dev)
576 /* Initialize the device structure. */
577 dev->netdev_ops = &vrf_netdev_ops;
578 dev->l3mdev_ops = &vrf_l3mdev_ops;
579 dev->ethtool_ops = &vrf_ethtool_ops;
580 dev->destructor = free_netdev;
582 /* Fill in device structure with ethernet-generic values. */
583 eth_hw_addr_random(dev);
585 /* don't acquire vrf device's netif_tx_lock when transmitting */
586 dev->features |= NETIF_F_LLTX;
588 /* don't allow vrf devices to change network namespaces. */
589 dev->features |= NETIF_F_NETNS_LOCAL;
592 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
594 if (tb[IFLA_ADDRESS]) {
595 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
597 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
598 return -EADDRNOTAVAIL;
603 static void vrf_dellink(struct net_device *dev, struct list_head *head)
605 unregister_netdevice_queue(dev, head);
608 static int vrf_newlink(struct net *src_net, struct net_device *dev,
609 struct nlattr *tb[], struct nlattr *data[])
611 struct net_vrf *vrf = netdev_priv(dev);
614 if (!data || !data[IFLA_VRF_TABLE])
617 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
619 dev->priv_flags |= IFF_L3MDEV_MASTER;
621 err = register_netdevice(dev);
632 static size_t vrf_nl_getsize(const struct net_device *dev)
634 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
637 static int vrf_fillinfo(struct sk_buff *skb,
638 const struct net_device *dev)
640 struct net_vrf *vrf = netdev_priv(dev);
642 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
645 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
646 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
649 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
651 .priv_size = sizeof(struct net_vrf),
653 .get_size = vrf_nl_getsize,
654 .policy = vrf_nl_policy,
655 .validate = vrf_validate,
656 .fill_info = vrf_fillinfo,
658 .newlink = vrf_newlink,
659 .dellink = vrf_dellink,
661 .maxtype = IFLA_VRF_MAX,
664 static int vrf_device_event(struct notifier_block *unused,
665 unsigned long event, void *ptr)
667 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
669 /* only care about unregister events to drop slave references */
670 if (event == NETDEV_UNREGISTER) {
671 struct net_device *vrf_dev;
673 if (!netif_is_l3_slave(dev))
676 vrf_dev = netdev_master_upper_dev_get(dev);
677 vrf_del_slave(vrf_dev, dev);
683 static struct notifier_block vrf_notifier_block __read_mostly = {
684 .notifier_call = vrf_device_event,
687 static int __init vrf_init_module(void)
691 vrf_dst_ops.kmem_cachep =
692 kmem_cache_create("vrf_ip_dst_cache",
693 sizeof(struct rtable), 0,
697 if (!vrf_dst_ops.kmem_cachep)
700 register_netdevice_notifier(&vrf_notifier_block);
702 rc = rtnl_link_register(&vrf_link_ops);
709 unregister_netdevice_notifier(&vrf_notifier_block);
710 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
714 static void __exit vrf_cleanup_module(void)
716 rtnl_link_unregister(&vrf_link_ops);
717 unregister_netdevice_notifier(&vrf_notifier_block);
718 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
721 module_init(vrf_init_module);
722 module_exit(vrf_cleanup_module);
723 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
724 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
725 MODULE_LICENSE("GPL");
726 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
727 MODULE_VERSION(DRV_VERSION);