Commit | Line | Data |
---|---|---|
193125db DA |
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 | ||
57b8efa1 NA |
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 | ||
193125db DA |
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 | { | |
57b8efa1 NA |
158 | vrf_tx_error(dev, skb); |
159 | return NET_XMIT_DROP; | |
193125db DA |
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: | |
57b8efa1 | 216 | vrf_tx_error(vrf_dev, skb); |
193125db DA |
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: | |
57b8efa1 | 228 | vrf_tx_error(dev, skb); |
193125db DA |
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 | ||
3a4a27d3 | 275 | dst_destroy(dst); |
193125db DA |
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 | { | |
193125db | 343 | list_del(&slave->list); |
193125db DA |
344 | } |
345 | ||
346 | static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave) | |
347 | { | |
193125db | 348 | list_add(&slave->list, &queue->all_slaves); |
193125db DA |
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); | |
193125db DA |
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; | |
193125db DA |
363 | vrf_ptr->ifindex = dev->ifindex; |
364 | vrf_ptr->tb_id = vrf->tb_id; | |
365 | ||
193125db DA |
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); | |
15df5e71 | 372 | goto out_fail; |
193125db DA |
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; | |
15df5e71 | 380 | __vrf_insert_slave(queue, slave); |
193125db DA |
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); | |
193125db DA |
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(dev) || netif_is_vrf(port_dev) || | |
397 | vrf_is_slave(port_dev)) | |
398 | return -EINVAL; | |
399 | ||
400 | return do_vrf_add_slave(dev, port_dev); | |
401 | } | |
402 | ||
403 | /* inverse of do_vrf_add_slave */ | |
404 | static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
405 | { | |
406 | struct net_vrf_dev *vrf_ptr = rtnl_dereference(port_dev->vrf_ptr); | |
407 | struct net_vrf *vrf = netdev_priv(dev); | |
408 | struct slave_queue *queue = &vrf->queue; | |
409 | struct slave *slave; | |
410 | ||
411 | RCU_INIT_POINTER(port_dev->vrf_ptr, NULL); | |
412 | ||
413 | netdev_upper_dev_unlink(port_dev, dev); | |
414 | port_dev->flags &= ~IFF_SLAVE; | |
415 | ||
416 | netdev_rx_handler_unregister(port_dev); | |
417 | ||
418 | /* after netdev_rx_handler_unregister for synchronize_rcu */ | |
419 | kfree(vrf_ptr); | |
420 | ||
421 | cycle_netdev(port_dev); | |
422 | ||
423 | slave = __vrf_find_slave_dev(queue, port_dev); | |
424 | if (slave) | |
425 | __vrf_remove_slave(queue, slave); | |
426 | ||
427 | kfree(slave); | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
432 | static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
433 | { | |
434 | if (!netif_is_vrf(dev)) | |
435 | return -EINVAL; | |
436 | ||
437 | return do_vrf_del_slave(dev, port_dev); | |
438 | } | |
439 | ||
440 | static void vrf_dev_uninit(struct net_device *dev) | |
441 | { | |
442 | struct net_vrf *vrf = netdev_priv(dev); | |
443 | struct slave_queue *queue = &vrf->queue; | |
444 | struct list_head *head = &queue->all_slaves; | |
445 | struct slave *slave, *next; | |
446 | ||
447 | vrf_rtable_destroy(vrf); | |
448 | ||
449 | list_for_each_entry_safe(slave, next, head, list) | |
450 | vrf_del_slave(dev, slave->dev); | |
451 | ||
3a4a27d3 | 452 | free_percpu(dev->dstats); |
193125db DA |
453 | dev->dstats = NULL; |
454 | } | |
455 | ||
456 | static int vrf_dev_init(struct net_device *dev) | |
457 | { | |
458 | struct net_vrf *vrf = netdev_priv(dev); | |
459 | ||
460 | INIT_LIST_HEAD(&vrf->queue.all_slaves); | |
461 | ||
462 | dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); | |
463 | if (!dev->dstats) | |
464 | goto out_nomem; | |
465 | ||
466 | /* create the default dst which points back to us */ | |
467 | vrf->rth = vrf_rtable_create(dev); | |
468 | if (!vrf->rth) | |
469 | goto out_stats; | |
470 | ||
471 | dev->flags = IFF_MASTER | IFF_NOARP; | |
472 | ||
473 | return 0; | |
474 | ||
475 | out_stats: | |
476 | free_percpu(dev->dstats); | |
477 | dev->dstats = NULL; | |
478 | out_nomem: | |
479 | return -ENOMEM; | |
480 | } | |
481 | ||
482 | static const struct net_device_ops vrf_netdev_ops = { | |
483 | .ndo_init = vrf_dev_init, | |
484 | .ndo_uninit = vrf_dev_uninit, | |
485 | .ndo_start_xmit = vrf_xmit, | |
486 | .ndo_get_stats64 = vrf_get_stats64, | |
487 | .ndo_add_slave = vrf_add_slave, | |
488 | .ndo_del_slave = vrf_del_slave, | |
489 | }; | |
490 | ||
491 | static void vrf_get_drvinfo(struct net_device *dev, | |
492 | struct ethtool_drvinfo *info) | |
493 | { | |
494 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); | |
495 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
496 | } | |
497 | ||
498 | static const struct ethtool_ops vrf_ethtool_ops = { | |
499 | .get_drvinfo = vrf_get_drvinfo, | |
500 | }; | |
501 | ||
502 | static void vrf_setup(struct net_device *dev) | |
503 | { | |
504 | ether_setup(dev); | |
505 | ||
506 | /* Initialize the device structure. */ | |
507 | dev->netdev_ops = &vrf_netdev_ops; | |
508 | dev->ethtool_ops = &vrf_ethtool_ops; | |
509 | dev->destructor = free_netdev; | |
510 | ||
511 | /* Fill in device structure with ethernet-generic values. */ | |
512 | eth_hw_addr_random(dev); | |
513 | ||
514 | /* don't acquire vrf device's netif_tx_lock when transmitting */ | |
515 | dev->features |= NETIF_F_LLTX; | |
516 | ||
517 | /* don't allow vrf devices to change network namespaces. */ | |
518 | dev->features |= NETIF_F_NETNS_LOCAL; | |
519 | } | |
520 | ||
521 | static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) | |
522 | { | |
523 | if (tb[IFLA_ADDRESS]) { | |
524 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
525 | return -EINVAL; | |
526 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
527 | return -EADDRNOTAVAIL; | |
528 | } | |
529 | return 0; | |
530 | } | |
531 | ||
532 | static void vrf_dellink(struct net_device *dev, struct list_head *head) | |
533 | { | |
534 | struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr); | |
535 | ||
536 | RCU_INIT_POINTER(dev->vrf_ptr, NULL); | |
537 | kfree_rcu(vrf_ptr, rcu); | |
538 | unregister_netdevice_queue(dev, head); | |
539 | } | |
540 | ||
541 | static int vrf_newlink(struct net *src_net, struct net_device *dev, | |
542 | struct nlattr *tb[], struct nlattr *data[]) | |
543 | { | |
544 | struct net_vrf *vrf = netdev_priv(dev); | |
545 | struct net_vrf_dev *vrf_ptr; | |
546 | int err; | |
547 | ||
548 | if (!data || !data[IFLA_VRF_TABLE]) | |
549 | return -EINVAL; | |
550 | ||
551 | vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); | |
552 | ||
553 | dev->priv_flags |= IFF_VRF_MASTER; | |
554 | ||
555 | err = -ENOMEM; | |
556 | vrf_ptr = kmalloc(sizeof(*dev->vrf_ptr), GFP_KERNEL); | |
557 | if (!vrf_ptr) | |
558 | goto out_fail; | |
559 | ||
560 | vrf_ptr->ifindex = dev->ifindex; | |
561 | vrf_ptr->tb_id = vrf->tb_id; | |
562 | ||
563 | err = register_netdevice(dev); | |
564 | if (err < 0) | |
565 | goto out_fail; | |
566 | ||
567 | rcu_assign_pointer(dev->vrf_ptr, vrf_ptr); | |
568 | ||
569 | return 0; | |
570 | ||
571 | out_fail: | |
572 | kfree(vrf_ptr); | |
573 | free_netdev(dev); | |
574 | return err; | |
575 | } | |
576 | ||
577 | static size_t vrf_nl_getsize(const struct net_device *dev) | |
578 | { | |
579 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ | |
580 | } | |
581 | ||
582 | static int vrf_fillinfo(struct sk_buff *skb, | |
583 | const struct net_device *dev) | |
584 | { | |
585 | struct net_vrf *vrf = netdev_priv(dev); | |
586 | ||
587 | return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); | |
588 | } | |
589 | ||
590 | static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { | |
591 | [IFLA_VRF_TABLE] = { .type = NLA_U32 }, | |
592 | }; | |
593 | ||
594 | static struct rtnl_link_ops vrf_link_ops __read_mostly = { | |
595 | .kind = DRV_NAME, | |
596 | .priv_size = sizeof(struct net_vrf), | |
597 | ||
598 | .get_size = vrf_nl_getsize, | |
599 | .policy = vrf_nl_policy, | |
600 | .validate = vrf_validate, | |
601 | .fill_info = vrf_fillinfo, | |
602 | ||
603 | .newlink = vrf_newlink, | |
604 | .dellink = vrf_dellink, | |
605 | .setup = vrf_setup, | |
606 | .maxtype = IFLA_VRF_MAX, | |
607 | }; | |
608 | ||
609 | static int vrf_device_event(struct notifier_block *unused, | |
610 | unsigned long event, void *ptr) | |
611 | { | |
612 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
613 | ||
614 | /* only care about unregister events to drop slave references */ | |
615 | if (event == NETDEV_UNREGISTER) { | |
616 | struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr); | |
617 | struct net_device *vrf_dev; | |
618 | ||
619 | if (!vrf_ptr || netif_is_vrf(dev)) | |
620 | goto out; | |
621 | ||
58aa9087 NA |
622 | vrf_dev = netdev_master_upper_dev_get(dev); |
623 | vrf_del_slave(vrf_dev, dev); | |
193125db DA |
624 | } |
625 | out: | |
626 | return NOTIFY_DONE; | |
627 | } | |
628 | ||
629 | static struct notifier_block vrf_notifier_block __read_mostly = { | |
630 | .notifier_call = vrf_device_event, | |
631 | }; | |
632 | ||
633 | static int __init vrf_init_module(void) | |
634 | { | |
635 | int rc; | |
636 | ||
637 | vrf_dst_ops.kmem_cachep = | |
638 | kmem_cache_create("vrf_ip_dst_cache", | |
639 | sizeof(struct rtable), 0, | |
e367da02 | 640 | SLAB_HWCACHE_ALIGN, |
193125db DA |
641 | NULL); |
642 | ||
643 | if (!vrf_dst_ops.kmem_cachep) | |
644 | return -ENOMEM; | |
645 | ||
646 | register_netdevice_notifier(&vrf_notifier_block); | |
647 | ||
648 | rc = rtnl_link_register(&vrf_link_ops); | |
649 | if (rc < 0) | |
650 | goto error; | |
651 | ||
652 | return 0; | |
653 | ||
654 | error: | |
655 | unregister_netdevice_notifier(&vrf_notifier_block); | |
656 | kmem_cache_destroy(vrf_dst_ops.kmem_cachep); | |
657 | return rc; | |
658 | } | |
659 | ||
660 | static void __exit vrf_cleanup_module(void) | |
661 | { | |
662 | rtnl_link_unregister(&vrf_link_ops); | |
663 | unregister_netdevice_notifier(&vrf_notifier_block); | |
664 | kmem_cache_destroy(vrf_dst_ops.kmem_cachep); | |
665 | } | |
666 | ||
667 | module_init(vrf_init_module); | |
668 | module_exit(vrf_cleanup_module); | |
669 | MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); | |
670 | MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); | |
671 | MODULE_LICENSE("GPL"); | |
672 | MODULE_ALIAS_RTNL_LINK(DRV_NAME); | |
673 | MODULE_VERSION(DRV_VERSION); |