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> | |
8f58336d | 30 | #include <net/arp.h> |
193125db DA |
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 | ||
39 | #define DRV_NAME "vrf" | |
40 | #define DRV_VERSION "1.0" | |
41 | ||
42 | #define vrf_is_slave(dev) ((dev)->flags & IFF_SLAVE) | |
43 | ||
44 | #define vrf_master_get_rcu(dev) \ | |
45 | ((struct net_device *)rcu_dereference(dev->rx_handler_data)) | |
46 | ||
47 | struct pcpu_dstats { | |
48 | u64 tx_pkts; | |
49 | u64 tx_bytes; | |
50 | u64 tx_drps; | |
51 | u64 rx_pkts; | |
52 | u64 rx_bytes; | |
53 | struct u64_stats_sync syncp; | |
54 | }; | |
55 | ||
56 | static struct dst_entry *vrf_ip_check(struct dst_entry *dst, u32 cookie) | |
57 | { | |
58 | return dst; | |
59 | } | |
60 | ||
61 | static int vrf_ip_local_out(struct sk_buff *skb) | |
62 | { | |
63 | return ip_local_out(skb); | |
64 | } | |
65 | ||
66 | static unsigned int vrf_v4_mtu(const struct dst_entry *dst) | |
67 | { | |
68 | /* TO-DO: return max ethernet size? */ | |
69 | return dst->dev->mtu; | |
70 | } | |
71 | ||
72 | static void vrf_dst_destroy(struct dst_entry *dst) | |
73 | { | |
74 | /* our dst lives forever - or until the device is closed */ | |
75 | } | |
76 | ||
77 | static unsigned int vrf_default_advmss(const struct dst_entry *dst) | |
78 | { | |
79 | return 65535 - 40; | |
80 | } | |
81 | ||
82 | static struct dst_ops vrf_dst_ops = { | |
83 | .family = AF_INET, | |
84 | .local_out = vrf_ip_local_out, | |
85 | .check = vrf_ip_check, | |
86 | .mtu = vrf_v4_mtu, | |
87 | .destroy = vrf_dst_destroy, | |
88 | .default_advmss = vrf_default_advmss, | |
89 | }; | |
90 | ||
91 | static bool is_ip_rx_frame(struct sk_buff *skb) | |
92 | { | |
93 | switch (skb->protocol) { | |
94 | case htons(ETH_P_IP): | |
95 | case htons(ETH_P_IPV6): | |
96 | return true; | |
97 | } | |
98 | return false; | |
99 | } | |
100 | ||
57b8efa1 NA |
101 | static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) |
102 | { | |
103 | vrf_dev->stats.tx_errors++; | |
104 | kfree_skb(skb); | |
105 | } | |
106 | ||
193125db DA |
107 | /* note: already called with rcu_read_lock */ |
108 | static rx_handler_result_t vrf_handle_frame(struct sk_buff **pskb) | |
109 | { | |
110 | struct sk_buff *skb = *pskb; | |
111 | ||
112 | if (is_ip_rx_frame(skb)) { | |
113 | struct net_device *dev = vrf_master_get_rcu(skb->dev); | |
114 | struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); | |
115 | ||
116 | u64_stats_update_begin(&dstats->syncp); | |
117 | dstats->rx_pkts++; | |
118 | dstats->rx_bytes += skb->len; | |
119 | u64_stats_update_end(&dstats->syncp); | |
120 | ||
121 | skb->dev = dev; | |
122 | ||
123 | return RX_HANDLER_ANOTHER; | |
124 | } | |
125 | return RX_HANDLER_PASS; | |
126 | } | |
127 | ||
128 | static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, | |
129 | struct rtnl_link_stats64 *stats) | |
130 | { | |
131 | int i; | |
132 | ||
133 | for_each_possible_cpu(i) { | |
134 | const struct pcpu_dstats *dstats; | |
135 | u64 tbytes, tpkts, tdrops, rbytes, rpkts; | |
136 | unsigned int start; | |
137 | ||
138 | dstats = per_cpu_ptr(dev->dstats, i); | |
139 | do { | |
140 | start = u64_stats_fetch_begin_irq(&dstats->syncp); | |
141 | tbytes = dstats->tx_bytes; | |
142 | tpkts = dstats->tx_pkts; | |
143 | tdrops = dstats->tx_drps; | |
144 | rbytes = dstats->rx_bytes; | |
145 | rpkts = dstats->rx_pkts; | |
146 | } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); | |
147 | stats->tx_bytes += tbytes; | |
148 | stats->tx_packets += tpkts; | |
149 | stats->tx_dropped += tdrops; | |
150 | stats->rx_bytes += rbytes; | |
151 | stats->rx_packets += rpkts; | |
152 | } | |
153 | return stats; | |
154 | } | |
155 | ||
156 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, | |
157 | struct net_device *dev) | |
158 | { | |
57b8efa1 NA |
159 | vrf_tx_error(dev, skb); |
160 | return NET_XMIT_DROP; | |
193125db DA |
161 | } |
162 | ||
163 | static int vrf_send_v4_prep(struct sk_buff *skb, struct flowi4 *fl4, | |
164 | struct net_device *vrf_dev) | |
165 | { | |
166 | struct rtable *rt; | |
167 | int err = 1; | |
168 | ||
169 | rt = ip_route_output_flow(dev_net(vrf_dev), fl4, NULL); | |
170 | if (IS_ERR(rt)) | |
171 | goto out; | |
172 | ||
173 | /* TO-DO: what about broadcast ? */ | |
174 | if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { | |
175 | ip_rt_put(rt); | |
176 | goto out; | |
177 | } | |
178 | ||
179 | skb_dst_drop(skb); | |
180 | skb_dst_set(skb, &rt->dst); | |
181 | err = 0; | |
182 | out: | |
183 | return err; | |
184 | } | |
185 | ||
186 | static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, | |
187 | struct net_device *vrf_dev) | |
188 | { | |
189 | struct iphdr *ip4h = ip_hdr(skb); | |
190 | int ret = NET_XMIT_DROP; | |
191 | struct flowi4 fl4 = { | |
192 | /* needed to match OIF rule */ | |
193 | .flowi4_oif = vrf_dev->ifindex, | |
194 | .flowi4_iif = LOOPBACK_IFINDEX, | |
195 | .flowi4_tos = RT_TOS(ip4h->tos), | |
196 | .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_VRFSRC, | |
197 | .daddr = ip4h->daddr, | |
198 | }; | |
199 | ||
200 | if (vrf_send_v4_prep(skb, &fl4, vrf_dev)) | |
201 | goto err; | |
202 | ||
203 | if (!ip4h->saddr) { | |
204 | ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, | |
205 | RT_SCOPE_LINK); | |
206 | } | |
207 | ||
208 | ret = ip_local_out(skb); | |
209 | if (unlikely(net_xmit_eval(ret))) | |
210 | vrf_dev->stats.tx_errors++; | |
211 | else | |
212 | ret = NET_XMIT_SUCCESS; | |
213 | ||
214 | out: | |
215 | return ret; | |
216 | err: | |
57b8efa1 | 217 | vrf_tx_error(vrf_dev, skb); |
193125db DA |
218 | goto out; |
219 | } | |
220 | ||
221 | static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) | |
222 | { | |
8f58336d DA |
223 | /* strip the ethernet header added for pass through VRF device */ |
224 | __skb_pull(skb, skb_network_offset(skb)); | |
225 | ||
193125db DA |
226 | switch (skb->protocol) { |
227 | case htons(ETH_P_IP): | |
228 | return vrf_process_v4_outbound(skb, dev); | |
229 | case htons(ETH_P_IPV6): | |
230 | return vrf_process_v6_outbound(skb, dev); | |
231 | default: | |
57b8efa1 | 232 | vrf_tx_error(dev, skb); |
193125db DA |
233 | return NET_XMIT_DROP; |
234 | } | |
235 | } | |
236 | ||
237 | static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) | |
238 | { | |
239 | netdev_tx_t ret = is_ip_tx_frame(skb, dev); | |
240 | ||
241 | if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { | |
242 | struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); | |
243 | ||
244 | u64_stats_update_begin(&dstats->syncp); | |
245 | dstats->tx_pkts++; | |
246 | dstats->tx_bytes += skb->len; | |
247 | u64_stats_update_end(&dstats->syncp); | |
248 | } else { | |
249 | this_cpu_inc(dev->dstats->tx_drps); | |
250 | } | |
251 | ||
252 | return ret; | |
253 | } | |
254 | ||
8f58336d DA |
255 | /* modelled after ip_finish_output2 */ |
256 | static int vrf_finish_output(struct sock *sk, struct sk_buff *skb) | |
193125db | 257 | { |
8f58336d DA |
258 | struct dst_entry *dst = skb_dst(skb); |
259 | struct rtable *rt = (struct rtable *)dst; | |
260 | struct net_device *dev = dst->dev; | |
261 | unsigned int hh_len = LL_RESERVED_SPACE(dev); | |
262 | struct neighbour *neigh; | |
263 | u32 nexthop; | |
264 | int ret = -EINVAL; | |
265 | ||
266 | /* Be paranoid, rather than too clever. */ | |
267 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { | |
268 | struct sk_buff *skb2; | |
269 | ||
270 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
271 | if (!skb2) { | |
272 | ret = -ENOMEM; | |
273 | goto err; | |
274 | } | |
275 | if (skb->sk) | |
276 | skb_set_owner_w(skb2, skb->sk); | |
277 | ||
278 | consume_skb(skb); | |
279 | skb = skb2; | |
280 | } | |
281 | ||
282 | rcu_read_lock_bh(); | |
283 | ||
284 | nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); | |
285 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); | |
286 | if (unlikely(!neigh)) | |
287 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
288 | if (!IS_ERR(neigh)) | |
289 | ret = dst_neigh_output(dst, neigh, skb); | |
290 | ||
291 | rcu_read_unlock_bh(); | |
292 | err: | |
293 | if (unlikely(ret < 0)) | |
294 | vrf_tx_error(skb->dev, skb); | |
295 | return ret; | |
193125db DA |
296 | } |
297 | ||
298 | static int vrf_output(struct sock *sk, struct sk_buff *skb) | |
299 | { | |
300 | struct net_device *dev = skb_dst(skb)->dev; | |
301 | ||
302 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); | |
303 | ||
304 | skb->dev = dev; | |
305 | skb->protocol = htons(ETH_P_IP); | |
306 | ||
307 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb, | |
308 | NULL, dev, | |
8f58336d | 309 | vrf_finish_output, |
193125db DA |
310 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
311 | } | |
312 | ||
313 | static void vrf_rtable_destroy(struct net_vrf *vrf) | |
314 | { | |
315 | struct dst_entry *dst = (struct dst_entry *)vrf->rth; | |
316 | ||
3a4a27d3 | 317 | dst_destroy(dst); |
193125db DA |
318 | vrf->rth = NULL; |
319 | } | |
320 | ||
321 | static struct rtable *vrf_rtable_create(struct net_device *dev) | |
322 | { | |
323 | struct rtable *rth; | |
324 | ||
325 | rth = dst_alloc(&vrf_dst_ops, dev, 2, | |
326 | DST_OBSOLETE_NONE, | |
327 | (DST_HOST | DST_NOPOLICY | DST_NOXFRM)); | |
328 | if (rth) { | |
329 | rth->dst.output = vrf_output; | |
330 | rth->rt_genid = rt_genid_ipv4(dev_net(dev)); | |
331 | rth->rt_flags = 0; | |
332 | rth->rt_type = RTN_UNICAST; | |
333 | rth->rt_is_input = 0; | |
334 | rth->rt_iif = 0; | |
335 | rth->rt_pmtu = 0; | |
336 | rth->rt_gateway = 0; | |
337 | rth->rt_uses_gateway = 0; | |
338 | INIT_LIST_HEAD(&rth->rt_uncached); | |
339 | rth->rt_uncached_list = NULL; | |
193125db DA |
340 | } |
341 | ||
342 | return rth; | |
343 | } | |
344 | ||
345 | /**************************** device handling ********************/ | |
346 | ||
347 | /* cycle interface to flush neighbor cache and move routes across tables */ | |
348 | static void cycle_netdev(struct net_device *dev) | |
349 | { | |
350 | unsigned int flags = dev->flags; | |
351 | int ret; | |
352 | ||
353 | if (!netif_running(dev)) | |
354 | return; | |
355 | ||
356 | ret = dev_change_flags(dev, flags & ~IFF_UP); | |
357 | if (ret >= 0) | |
358 | ret = dev_change_flags(dev, flags); | |
359 | ||
360 | if (ret < 0) { | |
361 | netdev_err(dev, | |
362 | "Failed to cycle device %s; route tables might be wrong!\n", | |
363 | dev->name); | |
364 | } | |
365 | } | |
366 | ||
367 | static struct slave *__vrf_find_slave_dev(struct slave_queue *queue, | |
368 | struct net_device *dev) | |
369 | { | |
370 | struct list_head *head = &queue->all_slaves; | |
371 | struct slave *slave; | |
372 | ||
373 | list_for_each_entry(slave, head, list) { | |
374 | if (slave->dev == dev) | |
375 | return slave; | |
376 | } | |
377 | ||
378 | return NULL; | |
379 | } | |
380 | ||
381 | /* inverse of __vrf_insert_slave */ | |
382 | static void __vrf_remove_slave(struct slave_queue *queue, struct slave *slave) | |
383 | { | |
193125db | 384 | list_del(&slave->list); |
193125db DA |
385 | } |
386 | ||
387 | static void __vrf_insert_slave(struct slave_queue *queue, struct slave *slave) | |
388 | { | |
193125db | 389 | list_add(&slave->list, &queue->all_slaves); |
193125db DA |
390 | } |
391 | ||
392 | static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) | |
393 | { | |
394 | struct net_vrf_dev *vrf_ptr = kmalloc(sizeof(*vrf_ptr), GFP_KERNEL); | |
395 | struct slave *slave = kzalloc(sizeof(*slave), GFP_KERNEL); | |
193125db DA |
396 | struct net_vrf *vrf = netdev_priv(dev); |
397 | struct slave_queue *queue = &vrf->queue; | |
398 | int ret = -ENOMEM; | |
399 | ||
400 | if (!slave || !vrf_ptr) | |
401 | goto out_fail; | |
402 | ||
403 | slave->dev = port_dev; | |
193125db DA |
404 | vrf_ptr->ifindex = dev->ifindex; |
405 | vrf_ptr->tb_id = vrf->tb_id; | |
406 | ||
193125db DA |
407 | /* register the packet handler for slave ports */ |
408 | ret = netdev_rx_handler_register(port_dev, vrf_handle_frame, dev); | |
409 | if (ret) { | |
410 | netdev_err(port_dev, | |
411 | "Device %s failed to register rx_handler\n", | |
412 | port_dev->name); | |
15df5e71 | 413 | goto out_fail; |
193125db DA |
414 | } |
415 | ||
416 | ret = netdev_master_upper_dev_link(port_dev, dev); | |
417 | if (ret < 0) | |
418 | goto out_unregister; | |
419 | ||
420 | port_dev->flags |= IFF_SLAVE; | |
15df5e71 | 421 | __vrf_insert_slave(queue, slave); |
193125db DA |
422 | rcu_assign_pointer(port_dev->vrf_ptr, vrf_ptr); |
423 | cycle_netdev(port_dev); | |
424 | ||
425 | return 0; | |
426 | ||
427 | out_unregister: | |
428 | netdev_rx_handler_unregister(port_dev); | |
193125db DA |
429 | out_fail: |
430 | kfree(vrf_ptr); | |
431 | kfree(slave); | |
432 | return ret; | |
433 | } | |
434 | ||
435 | static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) | |
436 | { | |
2640b39c | 437 | if (netif_is_vrf(port_dev) || vrf_is_slave(port_dev)) |
193125db DA |
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_dev *vrf_ptr = rtnl_dereference(port_dev->vrf_ptr); | |
447 | struct net_vrf *vrf = netdev_priv(dev); | |
448 | struct slave_queue *queue = &vrf->queue; | |
449 | struct slave *slave; | |
450 | ||
451 | RCU_INIT_POINTER(port_dev->vrf_ptr, NULL); | |
452 | ||
453 | netdev_upper_dev_unlink(port_dev, dev); | |
454 | port_dev->flags &= ~IFF_SLAVE; | |
455 | ||
456 | netdev_rx_handler_unregister(port_dev); | |
457 | ||
458 | /* after netdev_rx_handler_unregister for synchronize_rcu */ | |
459 | kfree(vrf_ptr); | |
460 | ||
461 | cycle_netdev(port_dev); | |
462 | ||
463 | slave = __vrf_find_slave_dev(queue, port_dev); | |
464 | if (slave) | |
465 | __vrf_remove_slave(queue, slave); | |
466 | ||
467 | kfree(slave); | |
468 | ||
469 | return 0; | |
470 | } | |
471 | ||
472 | static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
473 | { | |
193125db DA |
474 | return do_vrf_del_slave(dev, port_dev); |
475 | } | |
476 | ||
477 | static void vrf_dev_uninit(struct net_device *dev) | |
478 | { | |
479 | struct net_vrf *vrf = netdev_priv(dev); | |
480 | struct slave_queue *queue = &vrf->queue; | |
481 | struct list_head *head = &queue->all_slaves; | |
482 | struct slave *slave, *next; | |
483 | ||
484 | vrf_rtable_destroy(vrf); | |
485 | ||
486 | list_for_each_entry_safe(slave, next, head, list) | |
487 | vrf_del_slave(dev, slave->dev); | |
488 | ||
3a4a27d3 | 489 | free_percpu(dev->dstats); |
193125db DA |
490 | dev->dstats = NULL; |
491 | } | |
492 | ||
493 | static int vrf_dev_init(struct net_device *dev) | |
494 | { | |
495 | struct net_vrf *vrf = netdev_priv(dev); | |
496 | ||
497 | INIT_LIST_HEAD(&vrf->queue.all_slaves); | |
498 | ||
499 | dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); | |
500 | if (!dev->dstats) | |
501 | goto out_nomem; | |
502 | ||
503 | /* create the default dst which points back to us */ | |
504 | vrf->rth = vrf_rtable_create(dev); | |
505 | if (!vrf->rth) | |
506 | goto out_stats; | |
507 | ||
508 | dev->flags = IFF_MASTER | IFF_NOARP; | |
509 | ||
510 | return 0; | |
511 | ||
512 | out_stats: | |
513 | free_percpu(dev->dstats); | |
514 | dev->dstats = NULL; | |
515 | out_nomem: | |
516 | return -ENOMEM; | |
517 | } | |
518 | ||
519 | static const struct net_device_ops vrf_netdev_ops = { | |
520 | .ndo_init = vrf_dev_init, | |
521 | .ndo_uninit = vrf_dev_uninit, | |
522 | .ndo_start_xmit = vrf_xmit, | |
523 | .ndo_get_stats64 = vrf_get_stats64, | |
524 | .ndo_add_slave = vrf_add_slave, | |
525 | .ndo_del_slave = vrf_del_slave, | |
526 | }; | |
527 | ||
528 | static void vrf_get_drvinfo(struct net_device *dev, | |
529 | struct ethtool_drvinfo *info) | |
530 | { | |
531 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); | |
532 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
533 | } | |
534 | ||
535 | static const struct ethtool_ops vrf_ethtool_ops = { | |
536 | .get_drvinfo = vrf_get_drvinfo, | |
537 | }; | |
538 | ||
539 | static void vrf_setup(struct net_device *dev) | |
540 | { | |
541 | ether_setup(dev); | |
542 | ||
543 | /* Initialize the device structure. */ | |
544 | dev->netdev_ops = &vrf_netdev_ops; | |
545 | dev->ethtool_ops = &vrf_ethtool_ops; | |
546 | dev->destructor = free_netdev; | |
547 | ||
548 | /* Fill in device structure with ethernet-generic values. */ | |
549 | eth_hw_addr_random(dev); | |
550 | ||
551 | /* don't acquire vrf device's netif_tx_lock when transmitting */ | |
552 | dev->features |= NETIF_F_LLTX; | |
553 | ||
554 | /* don't allow vrf devices to change network namespaces. */ | |
555 | dev->features |= NETIF_F_NETNS_LOCAL; | |
556 | } | |
557 | ||
558 | static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) | |
559 | { | |
560 | if (tb[IFLA_ADDRESS]) { | |
561 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
562 | return -EINVAL; | |
563 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
564 | return -EADDRNOTAVAIL; | |
565 | } | |
566 | return 0; | |
567 | } | |
568 | ||
569 | static void vrf_dellink(struct net_device *dev, struct list_head *head) | |
570 | { | |
571 | struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr); | |
572 | ||
573 | RCU_INIT_POINTER(dev->vrf_ptr, NULL); | |
574 | kfree_rcu(vrf_ptr, rcu); | |
575 | unregister_netdevice_queue(dev, head); | |
576 | } | |
577 | ||
578 | static int vrf_newlink(struct net *src_net, struct net_device *dev, | |
579 | struct nlattr *tb[], struct nlattr *data[]) | |
580 | { | |
581 | struct net_vrf *vrf = netdev_priv(dev); | |
582 | struct net_vrf_dev *vrf_ptr; | |
583 | int err; | |
584 | ||
585 | if (!data || !data[IFLA_VRF_TABLE]) | |
586 | return -EINVAL; | |
587 | ||
588 | vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); | |
589 | ||
590 | dev->priv_flags |= IFF_VRF_MASTER; | |
591 | ||
592 | err = -ENOMEM; | |
593 | vrf_ptr = kmalloc(sizeof(*dev->vrf_ptr), GFP_KERNEL); | |
594 | if (!vrf_ptr) | |
595 | goto out_fail; | |
596 | ||
597 | vrf_ptr->ifindex = dev->ifindex; | |
598 | vrf_ptr->tb_id = vrf->tb_id; | |
599 | ||
600 | err = register_netdevice(dev); | |
601 | if (err < 0) | |
602 | goto out_fail; | |
603 | ||
604 | rcu_assign_pointer(dev->vrf_ptr, vrf_ptr); | |
605 | ||
606 | return 0; | |
607 | ||
608 | out_fail: | |
609 | kfree(vrf_ptr); | |
610 | free_netdev(dev); | |
611 | return err; | |
612 | } | |
613 | ||
614 | static size_t vrf_nl_getsize(const struct net_device *dev) | |
615 | { | |
616 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ | |
617 | } | |
618 | ||
619 | static int vrf_fillinfo(struct sk_buff *skb, | |
620 | const struct net_device *dev) | |
621 | { | |
622 | struct net_vrf *vrf = netdev_priv(dev); | |
623 | ||
624 | return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); | |
625 | } | |
626 | ||
627 | static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { | |
628 | [IFLA_VRF_TABLE] = { .type = NLA_U32 }, | |
629 | }; | |
630 | ||
631 | static struct rtnl_link_ops vrf_link_ops __read_mostly = { | |
632 | .kind = DRV_NAME, | |
633 | .priv_size = sizeof(struct net_vrf), | |
634 | ||
635 | .get_size = vrf_nl_getsize, | |
636 | .policy = vrf_nl_policy, | |
637 | .validate = vrf_validate, | |
638 | .fill_info = vrf_fillinfo, | |
639 | ||
640 | .newlink = vrf_newlink, | |
641 | .dellink = vrf_dellink, | |
642 | .setup = vrf_setup, | |
643 | .maxtype = IFLA_VRF_MAX, | |
644 | }; | |
645 | ||
646 | static int vrf_device_event(struct notifier_block *unused, | |
647 | unsigned long event, void *ptr) | |
648 | { | |
649 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
650 | ||
651 | /* only care about unregister events to drop slave references */ | |
652 | if (event == NETDEV_UNREGISTER) { | |
653 | struct net_vrf_dev *vrf_ptr = rtnl_dereference(dev->vrf_ptr); | |
654 | struct net_device *vrf_dev; | |
655 | ||
656 | if (!vrf_ptr || netif_is_vrf(dev)) | |
657 | goto out; | |
658 | ||
58aa9087 NA |
659 | vrf_dev = netdev_master_upper_dev_get(dev); |
660 | vrf_del_slave(vrf_dev, dev); | |
193125db DA |
661 | } |
662 | out: | |
663 | return NOTIFY_DONE; | |
664 | } | |
665 | ||
666 | static struct notifier_block vrf_notifier_block __read_mostly = { | |
667 | .notifier_call = vrf_device_event, | |
668 | }; | |
669 | ||
670 | static int __init vrf_init_module(void) | |
671 | { | |
672 | int rc; | |
673 | ||
674 | vrf_dst_ops.kmem_cachep = | |
675 | kmem_cache_create("vrf_ip_dst_cache", | |
676 | sizeof(struct rtable), 0, | |
e367da02 | 677 | SLAB_HWCACHE_ALIGN, |
193125db DA |
678 | NULL); |
679 | ||
680 | if (!vrf_dst_ops.kmem_cachep) | |
681 | return -ENOMEM; | |
682 | ||
683 | register_netdevice_notifier(&vrf_notifier_block); | |
684 | ||
685 | rc = rtnl_link_register(&vrf_link_ops); | |
686 | if (rc < 0) | |
687 | goto error; | |
688 | ||
689 | return 0; | |
690 | ||
691 | error: | |
692 | unregister_netdevice_notifier(&vrf_notifier_block); | |
693 | kmem_cache_destroy(vrf_dst_ops.kmem_cachep); | |
694 | return rc; | |
695 | } | |
696 | ||
697 | static void __exit vrf_cleanup_module(void) | |
698 | { | |
699 | rtnl_link_unregister(&vrf_link_ops); | |
700 | unregister_netdevice_notifier(&vrf_notifier_block); | |
701 | kmem_cache_destroy(vrf_dst_ops.kmem_cachep); | |
702 | } | |
703 | ||
704 | module_init(vrf_init_module); | |
705 | module_exit(vrf_cleanup_module); | |
706 | MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); | |
707 | MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); | |
708 | MODULE_LICENSE("GPL"); | |
709 | MODULE_ALIAS_RTNL_LINK(DRV_NAME); | |
710 | MODULE_VERSION(DRV_VERSION); |