df872f4efb0d5d5453daddcaf8f6242bc6b8dabf
[deliverable/linux.git] / drivers / net / vrf.c
1 /*
2 * vrf.c: device driver to encapsulate a VRF space
3 *
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>
7 *
8 * Based on dummy, team and ipvlan drivers
9 *
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.
14 */
15
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/netdevice.h>
19 #include <linux/etherdevice.h>
20 #include <linux/ip.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>
28
29 #include <linux/inetdevice.h>
30 #include <net/arp.h>
31 #include <net/ip.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/vrf.h>
38 #include <net/l3mdev.h>
39
40 #define DRV_NAME "vrf"
41 #define DRV_VERSION "1.0"
42
43 #define vrf_is_slave(dev) ((dev)->flags & IFF_SLAVE)
44
45 #define vrf_master_get_rcu(dev) \
46 ((struct net_device *)rcu_dereference(dev->rx_handler_data))
47
48 struct pcpu_dstats {
49 u64 tx_pkts;
50 u64 tx_bytes;
51 u64 tx_drps;
52 u64 rx_pkts;
53 u64 rx_bytes;
54 struct u64_stats_sync syncp;
55 };
56
57 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
58 {
59 return dst;
60 }
61
62 static int vrf_ip_local_out(struct sk_buff *skb)
63 {
64 return ip_local_out(skb);
65 }
66
67 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
68 {
69 /* TO-DO: return max ethernet size? */
70 return dst->dev->mtu;
71 }
72
73 static void vrf_dst_destroy(struct dst_entry *dst)
74 {
75 /* our dst lives forever - or until the device is closed */
76 }
77
78 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
79 {
80 return 65535 - 40;
81 }
82
83 static struct dst_ops vrf_dst_ops = {
84 .family = AF_INET,
85 .local_out = vrf_ip_local_out,
86 .check = vrf_ip_check,
87 .mtu = vrf_v4_mtu,
88 .destroy = vrf_dst_destroy,
89 .default_advmss = vrf_default_advmss,
90 };
91
92 static bool is_ip_rx_frame(struct sk_buff *skb)
93 {
94 switch (skb->protocol) {
95 case htons(ETH_P_IP):
96 case htons(ETH_P_IPV6):
97 return true;
98 }
99 return false;
100 }
101
102 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
103 {
104 vrf_dev->stats.tx_errors++;
105 kfree_skb(skb);
106 }
107
108 /* note: already called with rcu_read_lock */
109 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
110 {
111 struct sk_buff *skb = *pskb;
112
113 if (is_ip_rx_frame(skb)) {
114 struct net_device *dev = vrf_master_get_rcu(skb->dev);
115 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
116
117 u64_stats_update_begin(&dstats->syncp);
118 dstats->rx_pkts++;
119 dstats->rx_bytes += skb->len;
120 u64_stats_update_end(&dstats->syncp);
121
122 skb->dev = dev;
123
124 return RX_HANDLER_ANOTHER;
125 }
126 return RX_HANDLER_PASS;
127 }
128
129 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
130 struct rtnl_link_stats64 *stats)
131 {
132 int i;
133
134 for_each_possible_cpu(i) {
135 const struct pcpu_dstats *dstats;
136 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
137 unsigned int start;
138
139 dstats = per_cpu_ptr(dev->dstats, i);
140 do {
141 start = u64_stats_fetch_begin_irq(&dstats->syncp);
142 tbytes = dstats->tx_bytes;
143 tpkts = dstats->tx_pkts;
144 tdrops = dstats->tx_drps;
145 rbytes = dstats->rx_bytes;
146 rpkts = dstats->rx_pkts;
147 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
148 stats->tx_bytes += tbytes;
149 stats->tx_packets += tpkts;
150 stats->tx_dropped += tdrops;
151 stats->rx_bytes += rbytes;
152 stats->rx_packets += rpkts;
153 }
154 return stats;
155 }
156
157 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
158 struct net_device *dev)
159 {
160 vrf_tx_error(dev, skb);
161 return NET_XMIT_DROP;
162 }
163
164 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
165 struct net_device *vrf_dev)
166 {
167 struct rtable *rt;
168 int err = 1;
169
170 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
171 if (IS_ERR(rt))
172 goto out;
173
174 /* TO-DO: what about broadcast ? */
175 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
176 ip_rt_put(rt);
177 goto out;
178 }
179
180 skb_dst_drop(skb);
181 skb_dst_set(skb, &rt->dst);
182 err = 0;
183 out:
184 return err;
185 }
186
187 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
188 struct net_device *vrf_dev)
189 {
190 struct iphdr *ip4h = ip_hdr(skb);
191 int ret = NET_XMIT_DROP;
192 struct flowi4 fl4 = {
193 /* needed to match OIF rule */
194 .flowi4_oif = vrf_dev->ifindex,
195 .flowi4_iif = LOOPBACK_IFINDEX,
196 .flowi4_tos = RT_TOS(ip4h->tos),
197 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_VRFSRC |
198 FLOWI_FLAG_SKIP_NH_OIF,
199 .daddr = ip4h->daddr,
200 };
201
202 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
203 goto err;
204
205 if (!ip4h->saddr) {
206 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
207 RT_SCOPE_LINK);
208 }
209
210 ret = ip_local_out(skb);
211 if (unlikely(net_xmit_eval(ret)))
212 vrf_dev->stats.tx_errors++;
213 else
214 ret = NET_XMIT_SUCCESS;
215
216 out:
217 return ret;
218 err:
219 vrf_tx_error(vrf_dev, skb);
220 goto out;
221 }
222
223 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
224 {
225 /* strip the ethernet header added for pass through VRF device */
226 __skb_pull(skb, skb_network_offset(skb));
227
228 switch (skb->protocol) {
229 case htons(ETH_P_IP):
230 return vrf_process_v4_outbound(skb, dev);
231 case htons(ETH_P_IPV6):
232 return vrf_process_v6_outbound(skb, dev);
233 default:
234 vrf_tx_error(dev, skb);
235 return NET_XMIT_DROP;
236 }
237 }
238
239 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
240 {
241 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
242
243 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
244 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
245
246 u64_stats_update_begin(&dstats->syncp);
247 dstats->tx_pkts++;
248 dstats->tx_bytes += skb->len;
249 u64_stats_update_end(&dstats->syncp);
250 } else {
251 this_cpu_inc(dev->dstats->tx_drps);
252 }
253
254 return ret;
255 }
256
257 /* modelled after ip_finish_output2 */
258 static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
259 {
260 struct dst_entry *dst = skb_dst(skb);
261 struct rtable *rt = (struct rtable *)dst;
262 struct net_device *dev = dst->dev;
263 unsigned int hh_len = LL_RESERVED_SPACE(dev);
264 struct neighbour *neigh;
265 u32 nexthop;
266 int ret = -EINVAL;
267
268 /* Be paranoid, rather than too clever. */
269 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
270 struct sk_buff *skb2;
271
272 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
273 if (!skb2) {
274 ret = -ENOMEM;
275 goto err;
276 }
277 if (skb->sk)
278 skb_set_owner_w(skb2, skb->sk);
279
280 consume_skb(skb);
281 skb = skb2;
282 }
283
284 rcu_read_lock_bh();
285
286 nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr);
287 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
288 if (unlikely(!neigh))
289 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
290 if (!IS_ERR(neigh))
291 ret = dst_neigh_output(dst, neigh, skb);
292
293 rcu_read_unlock_bh();
294 err:
295 if (unlikely(ret < 0))
296 vrf_tx_error(skb->dev, skb);
297 return ret;
298 }
299
300 static int vrf_output(struct sock *sk, struct sk_buff *skb)
301 {
302 struct net_device *dev = skb_dst(skb)->dev;
303 struct net *net = dev_net(dev);
304
305 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
306
307 skb->dev = dev;
308 skb->protocol = htons(ETH_P_IP);
309
310 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
311 net, sk, skb, NULL, dev,
312 vrf_finish_output,
313 !(IPCB(skb)->flags & IPSKB_REROUTED));
314 }
315
316 static void vrf_rtable_destroy(struct net_vrf *vrf)
317 {
318 struct dst_entry *dst = (struct dst_entry *)vrf->rth;
319
320 dst_destroy(dst);
321 vrf->rth = NULL;
322 }
323
324 static struct rtable *vrf_rtable_create(struct net_device *dev)
325 {
326 struct net_vrf *vrf = netdev_priv(dev);
327 struct rtable *rth;
328
329 rth = dst_alloc(&vrf_dst_ops, dev, 2,
330 DST_OBSOLETE_NONE,
331 (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
332 if (rth) {
333 rth->dst.output = vrf_output;
334 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
335 rth->rt_flags = 0;
336 rth->rt_type = RTN_UNICAST;
337 rth->rt_is_input = 0;
338 rth->rt_iif = 0;
339 rth->rt_pmtu = 0;
340 rth->rt_gateway = 0;
341 rth->rt_uses_gateway = 0;
342 rth->rt_table_id = vrf->tb_id;
343 INIT_LIST_HEAD(&rth->rt_uncached);
344 rth->rt_uncached_list = NULL;
345 }
346
347 return rth;
348 }
349
350 /**************************** device handling ********************/
351
352 /* cycle interface to flush neighbor cache and move routes across tables */
353 static void cycle_netdev(struct net_device *dev)
354 {
355 unsigned int flags = dev->flags;
356 int ret;
357
358 if (!netif_running(dev))
359 return;
360
361 ret = dev_change_flags(dev, flags & ~IFF_UP);
362 if (ret >= 0)
363 ret = dev_change_flags(dev, flags);
364
365 if (ret < 0) {
366 netdev_err(dev,
367 "Failed to cycle device %s; route tables might be wrong!\n",
368 dev->name);
369 }
370 }
371
372 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
373 struct net_device *dev)
374 {
375 struct list_head *head = &queue->all_slaves;
376 struct slave *slave;
377
378 list_for_each_entry(slave, head, list) {
379 if (slave->dev == dev)
380 return slave;
381 }
382
383 return NULL;
384 }
385
386 /* inverse of __vrf_insert_slave */
387 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
388 {
389 list_del(&slave->list);
390 }
391
392 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
393 {
394 list_add(&slave->list, &queue->all_slaves);
395 }
396
397 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
398 {
399 struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
400 struct net_vrf *vrf = netdev_priv(dev);
401 struct slave_queue *queue = &vrf->queue;
402 int ret = -ENOMEM;
403
404 if (!slave)
405 goto out_fail;
406
407 slave->dev = port_dev;
408
409 /* register the packet handler for slave ports */
410 ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
411 if (ret) {
412 netdev_err(port_dev,
413 "Device %s failed to register rx_handler\n",
414 port_dev->name);
415 goto out_fail;
416 }
417
418 ret = netdev_master_upper_dev_link(port_dev, dev);
419 if (ret < 0)
420 goto out_unregister;
421
422 port_dev->flags |= IFF_SLAVE;
423 __vrf_insert_slave(queue, slave);
424 cycle_netdev(port_dev);
425
426 return 0;
427
428 out_unregister:
429 netdev_rx_handler_unregister(port_dev);
430 out_fail:
431 kfree(slave);
432 return ret;
433 }
434
435 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
436 {
437 if (netif_is_l3_master(port_dev) || vrf_is_slave(port_dev))
438 return -EINVAL;
439
440 return do_vrf_add_slave(dev, port_dev);
441 }
442
443 /* inverse of do_vrf_add_slave */
444 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
445 {
446 struct net_vrf *vrf = netdev_priv(dev);
447 struct slave_queue *queue = &vrf->queue;
448 struct slave *slave;
449
450 netdev_upper_dev_unlink(port_dev, dev);
451 port_dev->flags &= ~IFF_SLAVE;
452
453 netdev_rx_handler_unregister(port_dev);
454
455 cycle_netdev(port_dev);
456
457 slave = __vrf_find_slave_dev(queue, port_dev);
458 if (slave)
459 __vrf_remove_slave(queue, slave);
460
461 kfree(slave);
462
463 return 0;
464 }
465
466 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
467 {
468 return do_vrf_del_slave(dev, port_dev);
469 }
470
471 static void vrf_dev_uninit(struct net_device *dev)
472 {
473 struct net_vrf *vrf = netdev_priv(dev);
474 struct slave_queue *queue = &vrf->queue;
475 struct list_head *head = &queue->all_slaves;
476 struct slave *slave, *next;
477
478 vrf_rtable_destroy(vrf);
479
480 list_for_each_entry_safe(slave, next, head, list)
481 vrf_del_slave(dev, slave->dev);
482
483 free_percpu(dev->dstats);
484 dev->dstats = NULL;
485 }
486
487 static int vrf_dev_init(struct net_device *dev)
488 {
489 struct net_vrf *vrf = netdev_priv(dev);
490
491 INIT_LIST_HEAD(&vrf->queue.all_slaves);
492
493 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
494 if (!dev->dstats)
495 goto out_nomem;
496
497 /* create the default dst which points back to us */
498 vrf->rth = vrf_rtable_create(dev);
499 if (!vrf->rth)
500 goto out_stats;
501
502 dev->flags = IFF_MASTER | IFF_NOARP;
503
504 return 0;
505
506 out_stats:
507 free_percpu(dev->dstats);
508 dev->dstats = NULL;
509 out_nomem:
510 return -ENOMEM;
511 }
512
513 static const struct net_device_ops vrf_netdev_ops = {
514 .ndo_init = vrf_dev_init,
515 .ndo_uninit = vrf_dev_uninit,
516 .ndo_start_xmit = vrf_xmit,
517 .ndo_get_stats64 = vrf_get_stats64,
518 .ndo_add_slave = vrf_add_slave,
519 .ndo_del_slave = vrf_del_slave,
520 };
521
522 static u32 vrf_fib_table(const struct net_device *dev)
523 {
524 struct net_vrf *vrf = netdev_priv(dev);
525
526 return vrf->tb_id;
527 }
528
529 static struct rtable *vrf_get_rtable(const struct net_device *dev,
530 const struct flowi4 *fl4)
531 {
532 struct rtable *rth = NULL;
533
534 if (!(fl4->flowi4_flags & FLOWI_FLAG_VRFSRC)) {
535 struct net_vrf *vrf = netdev_priv(dev);
536
537 rth = vrf->rth;
538 atomic_inc(&rth->dst.__refcnt);
539 }
540
541 return rth;
542 }
543
544 static const struct l3mdev_ops vrf_l3mdev_ops = {
545 .l3mdev_fib_table = vrf_fib_table,
546 .l3mdev_get_rtable = vrf_get_rtable,
547 };
548
549 static void vrf_get_drvinfo(struct net_device *dev,
550 struct ethtool_drvinfo *info)
551 {
552 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
553 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
554 }
555
556 static const struct ethtool_ops vrf_ethtool_ops = {
557 .get_drvinfo = vrf_get_drvinfo,
558 };
559
560 static void vrf_setup(struct net_device *dev)
561 {
562 ether_setup(dev);
563
564 /* Initialize the device structure. */
565 dev->netdev_ops = &vrf_netdev_ops;
566 dev->l3mdev_ops = &vrf_l3mdev_ops;
567 dev->ethtool_ops = &vrf_ethtool_ops;
568 dev->destructor = free_netdev;
569
570 /* Fill in device structure with ethernet-generic values. */
571 eth_hw_addr_random(dev);
572
573 /* don't acquire vrf device's netif_tx_lock when transmitting */
574 dev->features |= NETIF_F_LLTX;
575
576 /* don't allow vrf devices to change network namespaces. */
577 dev->features |= NETIF_F_NETNS_LOCAL;
578 }
579
580 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
581 {
582 if (tb[IFLA_ADDRESS]) {
583 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
584 return -EINVAL;
585 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
586 return -EADDRNOTAVAIL;
587 }
588 return 0;
589 }
590
591 static void vrf_dellink(struct net_device *dev, struct list_head *head)
592 {
593 unregister_netdevice_queue(dev, head);
594 }
595
596 static int vrf_newlink(struct net *src_net, struct net_device *dev,
597 struct nlattr *tb[], struct nlattr *data[])
598 {
599 struct net_vrf *vrf = netdev_priv(dev);
600 int err;
601
602 if (!data || !data[IFLA_VRF_TABLE])
603 return -EINVAL;
604
605 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
606
607 dev->priv_flags |= IFF_L3MDEV_MASTER;
608
609 err = register_netdevice(dev);
610 if (err < 0)
611 goto out_fail;
612
613 return 0;
614
615 out_fail:
616 free_netdev(dev);
617 return err;
618 }
619
620 static size_t vrf_nl_getsize(const struct net_device *dev)
621 {
622 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
623 }
624
625 static int vrf_fillinfo(struct sk_buff *skb,
626 const struct net_device *dev)
627 {
628 struct net_vrf *vrf = netdev_priv(dev);
629
630 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
631 }
632
633 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
634 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
635 };
636
637 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
638 .kind = DRV_NAME,
639 .priv_size = sizeof(struct net_vrf),
640
641 .get_size = vrf_nl_getsize,
642 .policy = vrf_nl_policy,
643 .validate = vrf_validate,
644 .fill_info = vrf_fillinfo,
645
646 .newlink = vrf_newlink,
647 .dellink = vrf_dellink,
648 .setup = vrf_setup,
649 .maxtype = IFLA_VRF_MAX,
650 };
651
652 static int vrf_device_event(struct notifier_block *unused,
653 unsigned long event, void *ptr)
654 {
655 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
656
657 /* only care about unregister events to drop slave references */
658 if (event == NETDEV_UNREGISTER) {
659 struct net_device *vrf_dev;
660
661 if (netif_is_l3_master(dev))
662 goto out;
663
664 vrf_dev = netdev_master_upper_dev_get(dev);
665 vrf_del_slave(vrf_dev, dev);
666 }
667 out:
668 return NOTIFY_DONE;
669 }
670
671 static struct notifier_block vrf_notifier_block __read_mostly = {
672 .notifier_call = vrf_device_event,
673 };
674
675 static int __init vrf_init_module(void)
676 {
677 int rc;
678
679 vrf_dst_ops.kmem_cachep =
680 kmem_cache_create("vrf_ip_dst_cache",
681 sizeof(struct rtable), 0,
682 SLAB_HWCACHE_ALIGN,
683 NULL);
684
685 if (!vrf_dst_ops.kmem_cachep)
686 return -ENOMEM;
687
688 register_netdevice_notifier(&vrf_notifier_block);
689
690 rc = rtnl_link_register(&vrf_link_ops);
691 if (rc < 0)
692 goto error;
693
694 return 0;
695
696 error:
697 unregister_netdevice_notifier(&vrf_notifier_block);
698 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
699 return rc;
700 }
701
702 static void __exit vrf_cleanup_module(void)
703 {
704 rtnl_link_unregister(&vrf_link_ops);
705 unregister_netdevice_notifier(&vrf_notifier_block);
706 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
707 }
708
709 module_init(vrf_init_module);
710 module_exit(vrf_cleanup_module);
711 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
712 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
713 MODULE_LICENSE("GPL");
714 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
715 MODULE_VERSION(DRV_VERSION);
This page took 0.043321 seconds and 5 git commands to generate.