projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
vrf: ndo_add|del_slave drop unnecessary checks
[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 static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb)
101 {
102 vrf_dev->stats.tx_errors++;
103 kfree_skb(skb);
104 }
105
106 /* note: already called with rcu_read_lock */
107 static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb)
108 {
109 struct sk_buff *skb = *pskb;
110
111 if (is_ip_rx_frame(skb)) {
112 struct net_device *dev = vrf_master_get_rcu(skb->dev);
113 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
114
115 u64_stats_update_begin(&dstats->syncp);
116 dstats->rx_pkts++;
117 dstats->rx_bytes += skb->len;
118 u64_stats_update_end(&dstats->syncp);
119
120 skb->dev = dev;
121
122 return RX_HANDLER_ANOTHER;
123 }
124 return RX_HANDLER_PASS;
125 }
126
127 static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev,
128 struct rtnl_link_stats64 *stats)
129 {
130 int i;
131
132 for_each_possible_cpu(i) {
133 const struct pcpu_dstats *dstats;
134 u64 tbytes, tpkts, tdrops, rbytes, rpkts;
135 unsigned int start;
136
137 dstats = per_cpu_ptr(dev->dstats, i);
138 do {
139 start = u64_stats_fetch_begin_irq(&dstats->syncp);
140 tbytes = dstats->tx_bytes;
141 tpkts = dstats->tx_pkts;
142 tdrops = dstats->tx_drps;
143 rbytes = dstats->rx_bytes;
144 rpkts = dstats->rx_pkts;
145 } while (u64_stats_fetch_retry_irq(&dstats->syncp, start));
146 stats->tx_bytes += tbytes;
147 stats->tx_packets += tpkts;
148 stats->tx_dropped += tdrops;
149 stats->rx_bytes += rbytes;
150 stats->rx_packets += rpkts;
151 }
152 return stats;
153 }
154
155 static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb,
156 struct net_device *dev)
157 {
158 vrf_tx_error(dev, skb);
159 return NET_XMIT_DROP;
160 }
161
162 static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4,
163 struct net_device *vrf_dev)
164 {
165 struct rtable *rt;
166 int err = 1;
167
168 rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL);
169 if (IS_ERR(rt))
170 goto out;
171
172 /* TO-DO: what about broadcast ? */
173 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
174 ip_rt_put(rt);
175 goto out;
176 }
177
178 skb_dst_drop(skb);
179 skb_dst_set(skb, &rt->dst);
180 err = 0;
181 out:
182 return err;
183 }
184
185 static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb,
186 struct net_device *vrf_dev)
187 {
188 struct iphdr *ip4h = ip_hdr(skb);
189 int ret = NET_XMIT_DROP;
190 struct flowi4 fl4 = {
191 /* needed to match OIF rule */
192 .flowi4_oif = vrf_dev->ifindex,
193 .flowi4_iif = LOOPBACK_IFINDEX,
194 .flowi4_tos = RT_TOS(ip4h->tos),
195 .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_VRFSRC,
196 .daddr = ip4h->daddr,
197 };
198
199 if (vrf_send_v4_prep(skb, &fl4, vrf_dev))
200 goto err;
201
202 if (!ip4h->saddr) {
203 ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0,
204 RT_SCOPE_LINK);
205 }
206
207 ret = ip_local_out(skb);
208 if (unlikely(net_xmit_eval(ret)))
209 vrf_dev->stats.tx_errors++;
210 else
211 ret = NET_XMIT_SUCCESS;
212
213 out:
214 return ret;
215 err:
216 vrf_tx_error(vrf_dev, skb);
217 goto out;
218 }
219
220 static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev)
221 {
222 switch (skb->protocol) {
223 case htons(ETH_P_IP):
224 return vrf_process_v4_outbound(skb, dev);
225 case htons(ETH_P_IPV6):
226 return vrf_process_v6_outbound(skb, dev);
227 default:
228 vrf_tx_error(dev, skb);
229 return NET_XMIT_DROP;
230 }
231 }
232
233 static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev)
234 {
235 netdev_tx_t ret = is_ip_tx_frame(skb, dev);
236
237 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
238 struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats);
239
240 u64_stats_update_begin(&dstats->syncp);
241 dstats->tx_pkts++;
242 dstats->tx_bytes += skb->len;
243 u64_stats_update_end(&dstats->syncp);
244 } else {
245 this_cpu_inc(dev->dstats->tx_drps);
246 }
247
248 return ret;
249 }
250
251 static netdev_tx_t vrf_finish(struct sock *sk, struct sk_buff *skb)
252 {
253 return dev_queue_xmit(skb);
254 }
255
256 static int vrf_output(struct sock *sk, struct sk_buff *skb)
257 {
258 struct net_device *dev = skb_dst(skb)->dev;
259
260 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
261
262 skb->dev = dev;
263 skb->protocol = htons(ETH_P_IP);
264
265 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb,
266 NULL, dev,
267 vrf_finish,
268 !(IPCB(skb)->flags & IPSKB_REROUTED));
269 }
270
271 static void vrf_rtable_destroy(struct net_vrf *vrf)
272 {
273 struct dst_entry *dst = (struct dst_entry *)vrf->rth;
274
275 dst_destroy(dst);
276 vrf->rth = NULL;
277 }
278
279 static struct rtable *vrf_rtable_create(struct net_device *dev)
280 {
281 struct rtable *rth;
282
283 rth = dst_alloc(&vrf_dst_ops, dev, 2,
284 DST_OBSOLETE_NONE,
285 (DST_HOST | DST_NOPOLICY | DST_NOXFRM));
286 if (rth) {
287 rth->dst.output = vrf_output;
288 rth->rt_genid = rt_genid_ipv4(dev_net(dev));
289 rth->rt_flags = 0;
290 rth->rt_type = RTN_UNICAST;
291 rth->rt_is_input = 0;
292 rth->rt_iif = 0;
293 rth->rt_pmtu = 0;
294 rth->rt_gateway = 0;
295 rth->rt_uses_gateway = 0;
296 INIT_LIST_HEAD(&rth->rt_uncached);
297 rth->rt_uncached_list = NULL;
298 rth->rt_lwtstate = NULL;
299 }
300
301 return rth;
302 }
303
304 /**************************** device handling ********************/
305
306 /* cycle interface to flush neighbor cache and move routes across tables */
307 static void cycle_netdev(struct net_device *dev)
308 {
309 unsigned int flags = dev->flags;
310 int ret;
311
312 if (!netif_running(dev))
313 return;
314
315 ret = dev_change_flags(dev, flags & ~IFF_UP);
316 if (ret >= 0)
317 ret = dev_change_flags(dev, flags);
318
319 if (ret < 0) {
320 netdev_err(dev,
321 "Failed to cycle device %s; route tables might be wrong!\n",
322 dev->name);
323 }
324 }
325
326 static struct slave *__vrf_find_slave_dev(struct slave_queue *queue,
327 struct net_device *dev)
328 {
329 struct list_head *head = &queue->all_slaves;
330 struct slave *slave;
331
332 list_for_each_entry(slave, head, list) {
333 if (slave->dev == dev)
334 return slave;
335 }
336
337 return NULL;
338 }
339
340 /* inverse of __vrf_insert_slave */
341 static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave)
342 {
343 list_del(&slave->list);
344 }
345
346 static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave)
347 {
348 list_add(&slave->list, &queue->all_slaves);
349 }
350
351 static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
352 {
353 struct net_vrf_dev *vrf_ptr = kmalloc(sizeof(*vrf_ptr), GFP_KERNEL);
354 struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL);
355 struct net_vrf *vrf = netdev_priv(dev);
356 struct slave_queue *queue = &vrf->queue;
357 int ret = -ENOMEM;
358
359 if (!slave || !vrf_ptr)
360 goto out_fail;
361
362 slave->dev = port_dev;
363 vrf_ptr->ifindex = dev->ifindex;
364 vrf_ptr->tb_id = vrf->tb_id;
365
366 /* register the packet handler for slave ports */
367 ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev);
368 if (ret) {
369 netdev_err(port_dev,
370 "Device %s failed to register rx_handler\n",
371 port_dev->name);
372 goto out_fail;
373 }
374
375 ret = netdev_master_upper_dev_link(port_dev, dev);
376 if (ret < 0)
377 goto out_unregister;
378
379 port_dev->flags |= IFF_SLAVE;
380 __vrf_insert_slave(queue, slave);
381 rcu_assign_pointer(port_dev->vrf_ptr, vrf_ptr);
382 cycle_netdev(port_dev);
383
384 return 0;
385
386 out_unregister:
387 netdev_rx_handler_unregister(port_dev);
388 out_fail:
389 kfree(vrf_ptr);
390 kfree(slave);
391 return ret;
392 }
393
394 static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev)
395 {
396 if (netif_is_vrf(port_dev) || vrf_is_slave(port_dev))
397 return -EINVAL;
398
399 return do_vrf_add_slave(dev, port_dev);
400 }
401
402 /* inverse of do_vrf_add_slave */
403 static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
404 {
405 struct net_vrf_dev *vrf_ptr = rtnl_dereference(port_dev->vrf_ptr);
406 struct net_vrf *vrf = netdev_priv(dev);
407 struct slave_queue *queue = &vrf->queue;
408 struct slave *slave;
409
410 RCU_INIT_POINTER(port_dev->vrf_ptr, NULL);
411
412 netdev_upper_dev_unlink(port_dev, dev);
413 port_dev->flags &= ~IFF_SLAVE;
414
415 netdev_rx_handler_unregister(port_dev);
416
417 /* after netdev_rx_handler_unregister for synchronize_rcu */
418 kfree(vrf_ptr);
419
420 cycle_netdev(port_dev);
421
422 slave = __vrf_find_slave_dev(queue, port_dev);
423 if (slave)
424 __vrf_remove_slave(queue, slave);
425
426 kfree(slave);
427
428 return 0;
429 }
430
431 static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev)
432 {
433 return do_vrf_del_slave(dev, port_dev);
434 }
435
436 static void vrf_dev_uninit(struct net_device *dev)
437 {
438 struct net_vrf *vrf = netdev_priv(dev);
439 struct slave_queue *queue = &vrf->queue;
440 struct list_head *head = &queue->all_slaves;
441 struct slave *slave, *next;
442
443 vrf_rtable_destroy(vrf);
444
445 list_for_each_entry_safe(slave, next, head, list)
446 vrf_del_slave(dev, slave->dev);
447
448 free_percpu(dev->dstats);
449 dev->dstats = NULL;
450 }
451
452 static int vrf_dev_init(struct net_device *dev)
453 {
454 struct net_vrf *vrf = netdev_priv(dev);
455
456 INIT_LIST_HEAD(&vrf->queue.all_slaves);
457
458 dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
459 if (!dev->dstats)
460 goto out_nomem;
461
462 /* create the default dst which points back to us */
463 vrf->rth = vrf_rtable_create(dev);
464 if (!vrf->rth)
465 goto out_stats;
466
467 dev->flags = IFF_MASTER | IFF_NOARP;
468
469 return 0;
470
471 out_stats:
472 free_percpu(dev->dstats);
473 dev->dstats = NULL;
474 out_nomem:
475 return -ENOMEM;
476 }
477
478 static const struct net_device_ops vrf_netdev_ops = {
479 .ndo_init = vrf_dev_init,
480 .ndo_uninit = vrf_dev_uninit,
481 .ndo_start_xmit = vrf_xmit,
482 .ndo_get_stats64 = vrf_get_stats64,
483 .ndo_add_slave = vrf_add_slave,
484 .ndo_del_slave = vrf_del_slave,
485 };
486
487 static void vrf_get_drvinfo(struct net_device *dev,
488 struct ethtool_drvinfo *info)
489 {
490 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
491 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
492 }
493
494 static const struct ethtool_ops vrf_ethtool_ops = {
495 .get_drvinfo = vrf_get_drvinfo,
496 };
497
498 static void vrf_setup(struct net_device *dev)
499 {
500 ether_setup(dev);
501
502 /* Initialize the device structure. */
503 dev->netdev_ops = &vrf_netdev_ops;
504 dev->ethtool_ops = &vrf_ethtool_ops;
505 dev->destructor = free_netdev;
506
507 /* Fill in device structure with ethernet-generic values. */
508 eth_hw_addr_random(dev);
509
510 /* don't acquire vrf device's netif_tx_lock when transmitting */
511 dev->features |= NETIF_F_LLTX;
512
513 /* don't allow vrf devices to change network namespaces. */
514 dev->features |= NETIF_F_NETNS_LOCAL;
515 }
516
517 static int vrf_validate(struct nlattr *tb[], struct nlattr *data[])
518 {
519 if (tb[IFLA_ADDRESS]) {
520 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
521 return -EINVAL;
522 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
523 return -EADDRNOTAVAIL;
524 }
525 return 0;
526 }
527
528 static void vrf_dellink(struct net_device *dev, struct list_head *head)
529 {
530 struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr);
531
532 RCU_INIT_POINTER(dev->vrf_ptr, NULL);
533 kfree_rcu(vrf_ptr, rcu);
534 unregister_netdevice_queue(dev, head);
535 }
536
537 static int vrf_newlink(struct net *src_net, struct net_device *dev,
538 struct nlattr *tb[], struct nlattr *data[])
539 {
540 struct net_vrf *vrf = netdev_priv(dev);
541 struct net_vrf_dev *vrf_ptr;
542 int err;
543
544 if (!data || !data[IFLA_VRF_TABLE])
545 return -EINVAL;
546
547 vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]);
548
549 dev->priv_flags |= IFF_VRF_MASTER;
550
551 err = -ENOMEM;
552 vrf_ptr = kmalloc(sizeof(*dev->vrf_ptr), GFP_KERNEL);
553 if (!vrf_ptr)
554 goto out_fail;
555
556 vrf_ptr->ifindex = dev->ifindex;
557 vrf_ptr->tb_id = vrf->tb_id;
558
559 err = register_netdevice(dev);
560 if (err < 0)
561 goto out_fail;
562
563 rcu_assign_pointer(dev->vrf_ptr, vrf_ptr);
564
565 return 0;
566
567 out_fail:
568 kfree(vrf_ptr);
569 free_netdev(dev);
570 return err;
571 }
572
573 static size_t vrf_nl_getsize(const struct net_device *dev)
574 {
575 return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */
576 }
577
578 static int vrf_fillinfo(struct sk_buff *skb,
579 const struct net_device *dev)
580 {
581 struct net_vrf *vrf = netdev_priv(dev);
582
583 return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id);
584 }
585
586 static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = {
587 [IFLA_VRF_TABLE] = { .type = NLA_U32 },
588 };
589
590 static struct rtnl_link_ops vrf_link_ops __read_mostly = {
591 .kind = DRV_NAME,
592 .priv_size = sizeof(struct net_vrf),
593
594 .get_size = vrf_nl_getsize,
595 .policy = vrf_nl_policy,
596 .validate = vrf_validate,
597 .fill_info = vrf_fillinfo,
598
599 .newlink = vrf_newlink,
600 .dellink = vrf_dellink,
601 .setup = vrf_setup,
602 .maxtype = IFLA_VRF_MAX,
603 };
604
605 static int vrf_device_event(struct notifier_block *unused,
606 unsigned long event, void *ptr)
607 {
608 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
609
610 /* only care about unregister events to drop slave references */
611 if (event == NETDEV_UNREGISTER) {
612 struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr);
613 struct net_device *vrf_dev;
614
615 if (!vrf_ptr || netif_is_vrf(dev))
616 goto out;
617
618 vrf_dev = netdev_master_upper_dev_get(dev);
619 vrf_del_slave(vrf_dev, dev);
620 }
621 out:
622 return NOTIFY_DONE;
623 }
624
625 static struct notifier_block vrf_notifier_block __read_mostly = {
626 .notifier_call = vrf_device_event,
627 };
628
629 static int __init vrf_init_module(void)
630 {
631 int rc;
632
633 vrf_dst_ops.kmem_cachep =
634 kmem_cache_create("vrf_ip_dst_cache",
635 sizeof(struct rtable), 0,
636 SLAB_HWCACHE_ALIGN,
637 NULL);
638
639 if (!vrf_dst_ops.kmem_cachep)
640 return -ENOMEM;
641
642 register_netdevice_notifier(&vrf_notifier_block);
643
644 rc = rtnl_link_register(&vrf_link_ops);
645 if (rc < 0)
646 goto error;
647
648 return 0;
649
650 error:
651 unregister_netdevice_notifier(&vrf_notifier_block);
652 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
653 return rc;
654 }
655
656 static void __exit vrf_cleanup_module(void)
657 {
658 rtnl_link_unregister(&vrf_link_ops);
659 unregister_netdevice_notifier(&vrf_notifier_block);
660 kmem_cache_destroy(vrf_dst_ops.kmem_cachep);
661 }
662
663 module_init(vrf_init_module);
664 module_exit(vrf_cleanup_module);
665 MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern");
666 MODULE_DESCRIPTION("Device driver to instantiate VRF domains");
667 MODULE_LICENSE("GPL");
668 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
669 MODULE_VERSION(DRV_VERSION);
Linux kernel - Deliverables
This page took 0.202855 seconds and 5 git commands to generate.