projects / deliverable / linux.git / blob
? search:


97605eab14ae3f4abbc454c42fc0222a36592d41
[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/ip.h>
31 #include <net/ip_fib.h>
32 #include <net/ip6_route.h>
33 #include <net/rtnetlink.h>
34 #include <net/route.h>
35 #include <net/addrconf.h>
36 #include <net/vrf.h>
37
38 #define DRV_NAME "vrf"
39 #define DRV_VERSION "1.0"
40
41 #define vrf_is_slave(dev) ((dev)->flags & IFF_SLAVE)
42
43 #define vrf_master_get_rcu(dev) \
44 ((struct net_device *)rcu_dereference(dev->rx_handler_data))
45
46 struct pcpu_dstats {
47 u64 tx_pkts;
48 u64 tx_bytes;
49 u64 tx_drps;
50 u64 rx_pkts;
51 u64 rx_bytes;
52 struct u64_stats_sync syncp;
53 };
54
55 static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie)
56 {
57 return dst;
58 }
59
60 static int vrf_ip_local_out(struct sk_buff *skb)
61 {
62 return ip_local_out(skb);
63 }
64
65 static unsigned int vrf_v4_mtu(const struct dst_entry *dst)
66 {
67 /* TO-DO: return max ethernet size? */
68 return dst->dev->mtu;
69 }
70
71 static void vrf_dst_destroy(struct dst_entry *dst)
72 {
73 /* our dst lives forever - or until the device is closed */
74 }
75
76 static unsigned int vrf_default_advmss(const struct dst_entry *dst)
77 {
78 return 65535 - 40;
79 }
80
81 static struct dst_ops vrf_dst_ops = {
82 .family = AF_INET,
83 .local_out = vrf_ip_local_out,
84 .check = vrf_ip_check,
85 .mtu = vrf_v4_mtu,
86 .destroy = vrf_dst_destroy,
87 .default_advmss = vrf_default_advmss,
88 };
89
90 static bool is_ip_rx_frame(struct sk_buff *skb)
91 {
92 switch (skb->protocol) {
93 case htons(ETH_P_IP):
94 case htons(ETH_P_IPV6):
95 return true;
96 }
97 return false;
98 }
99
100 /* note: already called with rcu_read_lock */
101 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
102 {
103 struct sk_buff *skb = *pskb;
104
105 if (is_ip_rx_frame(skb)) {
106 struct net_device *dev = vrf_master_get_rcu(skb->dev);
107 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
108
109 u64_stats_update_begin(&dstats->syncp);
110 dstats->rx_pkts++;
111 dstats->rx_bytes += skb->len;
112 u64_stats_update_end(&dstats->syncp);
113
114 skb->dev = dev;
115
116 return RX_HANDLER_ANOTHER;
117 }
118 return RX_HANDLER_PASS;
119 }
120
121 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
122 struct rtnl_link_stats64 *stats)
123 {
124 int i;
125
126 for_each_possible_cpu(i) {
127 const struct pcpu_dstats *dstats;
128 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
129 unsigned int start;
130
131 dstats = per_cpu_ptr(dev->dstats, i);
132 do {
133 start = u64_stats_fetch_begin_irq(&dstats->syncp);
134 tbytes = dstats->tx_bytes;
135 tpkts = dstats->tx_pkts;
136 tdrops = dstats->tx_drps;
137 rbytes = dstats->rx_bytes;
138 rpkts = dstats->rx_pkts;
139 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
140 stats->tx_bytes += tbytes;
141 stats->tx_packets += tpkts;
142 stats->tx_dropped += tdrops;
143 stats->rx_bytes += rbytes;
144 stats->rx_packets += rpkts;
145 }
146 return stats;
147 }
148
149 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
150 struct net_device *dev)
151 {
152 return 0;
153 }
154
155 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
156 struct net_device *vrf_dev)
157 {
158 struct rtable *rt;
159 int err = 1;
160
161 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
162 if (IS_ERR(rt))
163 goto out;
164
165 /* TO-DO: what about broadcast ? */
166 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
167 ip_rt_put(rt);
168 goto out;
169 }
170
171 skb_dst_drop(skb);
172 skb_dst_set(skb, &rt->dst);
173 err = 0;
174 out:
175 return err;
176 }
177
178 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
179 struct net_device *vrf_dev)
180 {
181 struct iphdr *ip4h = ip_hdr(skb);
182 int ret = NET_XMIT_DROP;
183 struct flowi4 fl4 = {
184 /* needed to match OIF rule */
185 .flowi4_oif = vrf_dev->ifindex,
186 .flowi4_iif = LOOPBACK_IFINDEX,
187 .flowi4_tos = RT_TOS(ip4h->tos),
188 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_VRFSRC,
189 .daddr = ip4h->daddr,
190 };
191
192 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
193 goto err;
194
195 if (!ip4h->saddr) {
196 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
197 RT_SCOPE_LINK);
198 }
199
200 ret = ip_local_out(skb);
201 if (unlikely(net_xmit_eval(ret)))
202 vrf_dev->stats.tx_errors++;
203 else
204 ret = NET_XMIT_SUCCESS;
205
206 out:
207 return ret;
208 err:
209 vrf_dev->stats.tx_errors++;
210 kfree_skb(skb);
211 goto out;
212 }
213
214 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
215 {
216 switch (skb->protocol) {
217 case htons(ETH_P_IP):
218 return vrf_process_v4_outbound(skb, dev);
219 case htons(ETH_P_IPV6):
220 return vrf_process_v6_outbound(skb, dev);
221 default:
222 return NET_XMIT_DROP;
223 }
224 }
225
226 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
227 {
228 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
229
230 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
231 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
232
233 u64_stats_update_begin(&dstats->syncp);
234 dstats->tx_pkts++;
235 dstats->tx_bytes += skb->len;
236 u64_stats_update_end(&dstats->syncp);
237 } else {
238 this_cpu_inc(dev->dstats->tx_drps);
239 }
240
241 return ret;
242 }
243
244 static netdev_tx_t vrf_finish(struct sock *sk, struct sk_buff *skb)
245 {
246 return dev_queue_xmit(skb);
247 }
248
249 static int vrf_output(struct sock *sk, struct sk_buff *skb)
250 {
251 struct net_device *dev = skb_dst(skb)->dev;
252
253 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
254
255 skb->dev = dev;
256 skb->protocol = htons(ETH_P_IP);
257
258 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb,
259 NULL, dev,
260 vrf_finish,
261 !(IPCB(skb)->flags & IPSKB_REROUTED));
262 }
263
264 static void vrf_rtable_destroy(struct net_vrf *vrf)
265 {
266 struct dst_entry *dst = (struct dst_entry *)vrf->rth;
267
268 dst_destroy(dst);
269 vrf->rth = NULL;
270 }
271
272 static struct rtable *vrf_rtable_create(struct net_device *dev)
273 {
274 struct rtable *rth;
275
276 rth = dst_alloc(&vrf_dst_ops, dev, 2,
277 DST_OBSOLETE_NONE,
278 (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
279 if (rth) {
280 rth->dst.output = vrf_output;
281 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
282 rth->rt_flags = 0;
283 rth->rt_type = RTN_UNICAST;
284 rth->rt_is_input = 0;
285 rth->rt_iif = 0;
286 rth->rt_pmtu = 0;
287 rth->rt_gateway = 0;
288 rth->rt_uses_gateway = 0;
289 INIT_LIST_HEAD(&rth->rt_uncached);
290 rth->rt_uncached_list = NULL;
291 rth->rt_lwtstate = NULL;
292 }
293
294 return rth;
295 }
296
297 /**************************** device handling ********************/
298
299 /* cycle interface to flush neighbor cache and move routes across tables */
300 static void cycle_netdev(struct net_device *dev)
301 {
302 unsigned int flags = dev->flags;
303 int ret;
304
305 if (!netif_running(dev))
306 return;
307
308 ret = dev_change_flags(dev, flags & ~IFF_UP);
309 if (ret >= 0)
310 ret = dev_change_flags(dev, flags);
311
312 if (ret < 0) {
313 netdev_err(dev,
314 "Failed to cycle device %s; route tables might be wrong!\n",
315 dev->name);
316 }
317 }
318
319 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
320 struct net_device *dev)
321 {
322 struct list_head *head = &queue->all_slaves;
323 struct slave *slave;
324
325 list_for_each_entry(slave, head, list) {
326 if (slave->dev == dev)
327 return slave;
328 }
329
330 return NULL;
331 }
332
333 /* inverse of __vrf_insert_slave */
334 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
335 {
336 list_del(&slave->list);
337 }
338
339 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
340 {
341 list_add(&slave->list, &queue->all_slaves);
342 }
343
344 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
345 {
346 struct net_vrf_dev *vrf_ptr = kmalloc(sizeof(*vrf_ptr), GFP_KERNEL);
347 struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
348 struct slave *duplicate_slave;
349 struct net_vrf *vrf = netdev_priv(dev);
350 struct slave_queue *queue = &vrf->queue;
351 int ret = -ENOMEM;
352
353 if (!slave || !vrf_ptr)
354 goto out_fail;
355
356 slave->dev = port_dev;
357
358 vrf_ptr->ifindex = dev->ifindex;
359 vrf_ptr->tb_id = vrf->tb_id;
360
361 duplicate_slave = __vrf_find_slave_dev(queue, port_dev);
362 if (duplicate_slave) {
363 ret = -EBUSY;
364 goto out_fail;
365 }
366
367 __vrf_insert_slave(queue, slave);
368
369 /* register the packet handler for slave ports */
370 ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
371 if (ret) {
372 netdev_err(port_dev,
373 "Device %s failed to register rx_handler\n",
374 port_dev->name);
375 goto out_remove;
376 }
377
378 ret = netdev_master_upper_dev_link(port_dev, dev);
379 if (ret < 0)
380 goto out_unregister;
381
382 port_dev->flags |= IFF_SLAVE;
383
384 rcu_assign_pointer(port_dev->vrf_ptr, vrf_ptr);
385 cycle_netdev(port_dev);
386
387 return 0;
388
389 out_unregister:
390 netdev_rx_handler_unregister(port_dev);
391 out_remove:
392 __vrf_remove_slave(queue, slave);
393 out_fail:
394 kfree(vrf_ptr);
395 kfree(slave);
396 return ret;
397 }
398
399 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
400 {
401 if (!netif_is_vrf(dev) || netif_is_vrf(port_dev) ||
402 vrf_is_slave(port_dev))
403 return -EINVAL;
404
405 return do_vrf_add_slave(dev, port_dev);
406 }
407
408 /* inverse of do_vrf_add_slave */
409 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
410 {
411 struct net_vrf_dev *vrf_ptr = rtnl_dereference(port_dev->vrf_ptr);
412 struct net_vrf *vrf = netdev_priv(dev);
413 struct slave_queue *queue = &vrf->queue;
414 struct slave *slave;
415
416 RCU_INIT_POINTER(port_dev->vrf_ptr, NULL);
417
418 netdev_upper_dev_unlink(port_dev, dev);
419 port_dev->flags &= ~IFF_SLAVE;
420
421 netdev_rx_handler_unregister(port_dev);
422
423 /* after netdev_rx_handler_unregister for synchronize_rcu */
424 kfree(vrf_ptr);
425
426 cycle_netdev(port_dev);
427
428 slave = __vrf_find_slave_dev(queue, port_dev);
429 if (slave)
430 __vrf_remove_slave(queue, slave);
431
432 kfree(slave);
433
434 return 0;
435 }
436
437 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
438 {
439 if (!netif_is_vrf(dev))
440 return -EINVAL;
441
442 return do_vrf_del_slave(dev, port_dev);
443 }
444
445 static void vrf_dev_uninit(struct net_device *dev)
446 {
447 struct net_vrf *vrf = netdev_priv(dev);
448 struct slave_queue *queue = &vrf->queue;
449 struct list_head *head = &queue->all_slaves;
450 struct slave *slave, *next;
451
452 vrf_rtable_destroy(vrf);
453
454 list_for_each_entry_safe(slave, next, head, list)
455 vrf_del_slave(dev, slave->dev);
456
457 free_percpu(dev->dstats);
458 dev->dstats = NULL;
459 }
460
461 static int vrf_dev_init(struct net_device *dev)
462 {
463 struct net_vrf *vrf = netdev_priv(dev);
464
465 INIT_LIST_HEAD(&vrf->queue.all_slaves);
466
467 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
468 if (!dev->dstats)
469 goto out_nomem;
470
471 /* create the default dst which points back to us */
472 vrf->rth = vrf_rtable_create(dev);
473 if (!vrf->rth)
474 goto out_stats;
475
476 dev->flags = IFF_MASTER | IFF_NOARP;
477
478 return 0;
479
480 out_stats:
481 free_percpu(dev->dstats);
482 dev->dstats = NULL;
483 out_nomem:
484 return -ENOMEM;
485 }
486
487 static const struct net_device_ops vrf_netdev_ops = {
488 .ndo_init = vrf_dev_init,
489 .ndo_uninit = vrf_dev_uninit,
490 .ndo_start_xmit = vrf_xmit,
491 .ndo_get_stats64 = vrf_get_stats64,
492 .ndo_add_slave = vrf_add_slave,
493 .ndo_del_slave = vrf_del_slave,
494 };
495
496 static void vrf_get_drvinfo(struct net_device *dev,
497 struct ethtool_drvinfo *info)
498 {
499 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
500 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
501 }
502
503 static const struct ethtool_ops vrf_ethtool_ops = {
504 .get_drvinfo = vrf_get_drvinfo,
505 };
506
507 static void vrf_setup(struct net_device *dev)
508 {
509 ether_setup(dev);
510
511 /* Initialize the device structure. */
512 dev->netdev_ops = &vrf_netdev_ops;
513 dev->ethtool_ops = &vrf_ethtool_ops;
514 dev->destructor = free_netdev;
515
516 /* Fill in device structure with ethernet-generic values. */
517 eth_hw_addr_random(dev);
518
519 /* don't acquire vrf device's netif_tx_lock when transmitting */
520 dev->features |= NETIF_F_LLTX;
521
522 /* don't allow vrf devices to change network namespaces. */
523 dev->features |= NETIF_F_NETNS_LOCAL;
524 }
525
526 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
527 {
528 if (tb[IFLA_ADDRESS]) {
529 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
530 return -EINVAL;
531 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
532 return -EADDRNOTAVAIL;
533 }
534 return 0;
535 }
536
537 static void vrf_dellink(struct net_device *dev, struct list_head *head)
538 {
539 struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr);
540
541 RCU_INIT_POINTER(dev->vrf_ptr, NULL);
542 kfree_rcu(vrf_ptr, rcu);
543 unregister_netdevice_queue(dev, head);
544 }
545
546 static int vrf_newlink(struct net *src_net, struct net_device *dev,
547 struct nlattr *tb[], struct nlattr *data[])
548 {
549 struct net_vrf *vrf = netdev_priv(dev);
550 struct net_vrf_dev *vrf_ptr;
551 int err;
552
553 if (!data || !data[IFLA_VRF_TABLE])
554 return -EINVAL;
555
556 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
557
558 dev->priv_flags |= IFF_VRF_MASTER;
559
560 err = -ENOMEM;
561 vrf_ptr = kmalloc(sizeof(*dev->vrf_ptr), GFP_KERNEL);
562 if (!vrf_ptr)
563 goto out_fail;
564
565 vrf_ptr->ifindex = dev->ifindex;
566 vrf_ptr->tb_id = vrf->tb_id;
567
568 err = register_netdevice(dev);
569 if (err < 0)
570 goto out_fail;
571
572 rcu_assign_pointer(dev->vrf_ptr, vrf_ptr);
573
574 return 0;
575
576 out_fail:
577 kfree(vrf_ptr);
578 free_netdev(dev);
579 return err;
580 }
581
582 static size_t vrf_nl_getsize(const struct net_device *dev)
583 {
584 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
585 }
586
587 static int vrf_fillinfo(struct sk_buff *skb,
588 const struct net_device *dev)
589 {
590 struct net_vrf *vrf = netdev_priv(dev);
591
592 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
593 }
594
595 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
596 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
597 };
598
599 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
600 .kind = DRV_NAME,
601 .priv_size = sizeof(struct net_vrf),
602
603 .get_size = vrf_nl_getsize,
604 .policy = vrf_nl_policy,
605 .validate = vrf_validate,
606 .fill_info = vrf_fillinfo,
607
608 .newlink = vrf_newlink,
609 .dellink = vrf_dellink,
610 .setup = vrf_setup,
611 .maxtype = IFLA_VRF_MAX,
612 };
613
614 static int vrf_device_event(struct notifier_block *unused,
615 unsigned long event, void *ptr)
616 {
617 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
618
619 /* only care about unregister events to drop slave references */
620 if (event == NETDEV_UNREGISTER) {
621 struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr);
622 struct net_device *vrf_dev;
623
624 if (!vrf_ptr || netif_is_vrf(dev))
625 goto out;
626
627 vrf_dev = __dev_get_by_index(dev_net(dev), vrf_ptr->ifindex);
628 if (vrf_dev)
629 vrf_del_slave(vrf_dev, dev);
630 }
631 out:
632 return NOTIFY_DONE;
633 }
634
635 static struct notifier_block vrf_notifier_block __read_mostly = {
636 .notifier_call = vrf_device_event,
637 };
638
639 static int __init vrf_init_module(void)
640 {
641 int rc;
642
643 vrf_dst_ops.kmem_cachep =
644 kmem_cache_create("vrf_ip_dst_cache",
645 sizeof(struct rtable), 0,
646 SLAB_HWCACHE_ALIGN | SLAB_PANIC,
647 NULL);
648
649 if (!vrf_dst_ops.kmem_cachep)
650 return -ENOMEM;
651
652 register_netdevice_notifier(&vrf_notifier_block);
653
654 rc = rtnl_link_register(&vrf_link_ops);
655 if (rc < 0)
656 goto error;
657
658 return 0;
659
660 error:
661 unregister_netdevice_notifier(&vrf_notifier_block);
662 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
663 return rc;
664 }
665
666 static void __exit vrf_cleanup_module(void)
667 {
668 rtnl_link_unregister(&vrf_link_ops);
669 unregister_netdevice_notifier(&vrf_notifier_block);
670 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
671 }
672
673 module_init(vrf_init_module);
674 module_exit(vrf_cleanup_module);
675 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
676 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
677 MODULE_LICENSE("GPL");
678 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
679 MODULE_VERSION(DRV_VERSION);
Linux kernel - Deliverables
This page took 0.090418 seconds and 4 git commands to generate.