]>
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> | |
35402e31 | 33 | #include <net/ip6_fib.h> |
193125db | 34 | #include <net/ip6_route.h> |
193125db DA |
35 | #include <net/route.h> |
36 | #include <net/addrconf.h> | |
ee15ee5d | 37 | #include <net/l3mdev.h> |
193125db | 38 | |
8cbb512c DA |
39 | #define RT_FL_TOS(oldflp4) \ |
40 | ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) | |
41 | ||
193125db DA |
42 | #define DRV_NAME "vrf" |
43 | #define DRV_VERSION "1.0" | |
44 | ||
ec539514 | 45 | struct net_vrf { |
b0e95ccd DA |
46 | struct rtable __rcu *rth; |
47 | struct rt6_info __rcu *rt6; | |
ec539514 DA |
48 | u32 tb_id; |
49 | }; | |
50 | ||
193125db DA |
51 | struct pcpu_dstats { |
52 | u64 tx_pkts; | |
53 | u64 tx_bytes; | |
54 | u64 tx_drps; | |
55 | u64 rx_pkts; | |
56 | u64 rx_bytes; | |
57 | struct u64_stats_sync syncp; | |
58 | }; | |
59 | ||
57b8efa1 NA |
60 | static void vrf_tx_error(struct net_device *vrf_dev, struct sk_buff *skb) |
61 | { | |
62 | vrf_dev->stats.tx_errors++; | |
63 | kfree_skb(skb); | |
64 | } | |
65 | ||
193125db DA |
66 | static struct rtnl_link_stats64 *vrf_get_stats64(struct net_device *dev, |
67 | struct rtnl_link_stats64 *stats) | |
68 | { | |
69 | int i; | |
70 | ||
71 | for_each_possible_cpu(i) { | |
72 | const struct pcpu_dstats *dstats; | |
73 | u64 tbytes, tpkts, tdrops, rbytes, rpkts; | |
74 | unsigned int start; | |
75 | ||
76 | dstats = per_cpu_ptr(dev->dstats, i); | |
77 | do { | |
78 | start = u64_stats_fetch_begin_irq(&dstats->syncp); | |
79 | tbytes = dstats->tx_bytes; | |
80 | tpkts = dstats->tx_pkts; | |
81 | tdrops = dstats->tx_drps; | |
82 | rbytes = dstats->rx_bytes; | |
83 | rpkts = dstats->rx_pkts; | |
84 | } while (u64_stats_fetch_retry_irq(&dstats->syncp, start)); | |
85 | stats->tx_bytes += tbytes; | |
86 | stats->tx_packets += tpkts; | |
87 | stats->tx_dropped += tdrops; | |
88 | stats->rx_bytes += rbytes; | |
89 | stats->rx_packets += rpkts; | |
90 | } | |
91 | return stats; | |
92 | } | |
93 | ||
35402e31 DA |
94 | #if IS_ENABLED(CONFIG_IPV6) |
95 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, | |
96 | struct net_device *dev) | |
97 | { | |
98 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
99 | struct net *net = dev_net(skb->dev); | |
100 | struct flowi6 fl6 = { | |
101 | /* needed to match OIF rule */ | |
102 | .flowi6_oif = dev->ifindex, | |
103 | .flowi6_iif = LOOPBACK_IFINDEX, | |
104 | .daddr = iph->daddr, | |
105 | .saddr = iph->saddr, | |
106 | .flowlabel = ip6_flowinfo(iph), | |
107 | .flowi6_mark = skb->mark, | |
108 | .flowi6_proto = iph->nexthdr, | |
109 | .flowi6_flags = FLOWI_FLAG_L3MDEV_SRC | FLOWI_FLAG_SKIP_NH_OIF, | |
110 | }; | |
111 | int ret = NET_XMIT_DROP; | |
112 | struct dst_entry *dst; | |
113 | struct dst_entry *dst_null = &net->ipv6.ip6_null_entry->dst; | |
114 | ||
115 | dst = ip6_route_output(net, NULL, &fl6); | |
116 | if (dst == dst_null) | |
117 | goto err; | |
118 | ||
119 | skb_dst_drop(skb); | |
120 | skb_dst_set(skb, dst); | |
121 | ||
911a66fb DA |
122 | /* strip the ethernet header added for pass through VRF device */ |
123 | __skb_pull(skb, skb_network_offset(skb)); | |
124 | ||
35402e31 DA |
125 | ret = ip6_local_out(net, skb->sk, skb); |
126 | if (unlikely(net_xmit_eval(ret))) | |
127 | dev->stats.tx_errors++; | |
128 | else | |
129 | ret = NET_XMIT_SUCCESS; | |
130 | ||
131 | return ret; | |
132 | err: | |
133 | vrf_tx_error(dev, skb); | |
134 | return NET_XMIT_DROP; | |
135 | } | |
136 | #else | |
193125db DA |
137 | static netdev_tx_t vrf_process_v6_outbound(struct sk_buff *skb, |
138 | struct net_device *dev) | |
139 | { | |
57b8efa1 NA |
140 | vrf_tx_error(dev, skb); |
141 | return NET_XMIT_DROP; | |
193125db | 142 | } |
35402e31 | 143 | #endif |
193125db | 144 | |
193125db DA |
145 | static netdev_tx_t vrf_process_v4_outbound(struct sk_buff *skb, |
146 | struct net_device *vrf_dev) | |
147 | { | |
148 | struct iphdr *ip4h = ip_hdr(skb); | |
149 | int ret = NET_XMIT_DROP; | |
150 | struct flowi4 fl4 = { | |
151 | /* needed to match OIF rule */ | |
152 | .flowi4_oif = vrf_dev->ifindex, | |
153 | .flowi4_iif = LOOPBACK_IFINDEX, | |
154 | .flowi4_tos = RT_TOS(ip4h->tos), | |
6e2895a8 | 155 | .flowi4_flags = FLOWI_FLAG_ANYSRC | FLOWI_FLAG_L3MDEV_SRC | |
58189ca7 | 156 | FLOWI_FLAG_SKIP_NH_OIF, |
193125db DA |
157 | .daddr = ip4h->daddr, |
158 | }; | |
911a66fb DA |
159 | struct net *net = dev_net(vrf_dev); |
160 | struct rtable *rt; | |
161 | ||
162 | rt = ip_route_output_flow(net, &fl4, NULL); | |
163 | if (IS_ERR(rt)) | |
164 | goto err; | |
193125db | 165 | |
911a66fb DA |
166 | if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { |
167 | ip_rt_put(rt); | |
193125db | 168 | goto err; |
911a66fb DA |
169 | } |
170 | ||
171 | skb_dst_drop(skb); | |
172 | skb_dst_set(skb, &rt->dst); | |
173 | ||
174 | /* strip the ethernet header added for pass through VRF device */ | |
175 | __skb_pull(skb, skb_network_offset(skb)); | |
193125db DA |
176 | |
177 | if (!ip4h->saddr) { | |
178 | ip4h->saddr = inet_select_addr(skb_dst(skb)->dev, 0, | |
179 | RT_SCOPE_LINK); | |
180 | } | |
181 | ||
33224b16 | 182 | ret = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); |
193125db DA |
183 | if (unlikely(net_xmit_eval(ret))) |
184 | vrf_dev->stats.tx_errors++; | |
185 | else | |
186 | ret = NET_XMIT_SUCCESS; | |
187 | ||
188 | out: | |
189 | return ret; | |
190 | err: | |
57b8efa1 | 191 | vrf_tx_error(vrf_dev, skb); |
193125db DA |
192 | goto out; |
193 | } | |
194 | ||
195 | static netdev_tx_t is_ip_tx_frame(struct sk_buff *skb, struct net_device *dev) | |
196 | { | |
197 | switch (skb->protocol) { | |
198 | case htons(ETH_P_IP): | |
199 | return vrf_process_v4_outbound(skb, dev); | |
200 | case htons(ETH_P_IPV6): | |
201 | return vrf_process_v6_outbound(skb, dev); | |
202 | default: | |
57b8efa1 | 203 | vrf_tx_error(dev, skb); |
193125db DA |
204 | return NET_XMIT_DROP; |
205 | } | |
206 | } | |
207 | ||
208 | static netdev_tx_t vrf_xmit(struct sk_buff *skb, struct net_device *dev) | |
209 | { | |
210 | netdev_tx_t ret = is_ip_tx_frame(skb, dev); | |
211 | ||
212 | if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { | |
213 | struct pcpu_dstats *dstats = this_cpu_ptr(dev->dstats); | |
214 | ||
215 | u64_stats_update_begin(&dstats->syncp); | |
216 | dstats->tx_pkts++; | |
217 | dstats->tx_bytes += skb->len; | |
218 | u64_stats_update_end(&dstats->syncp); | |
219 | } else { | |
220 | this_cpu_inc(dev->dstats->tx_drps); | |
221 | } | |
222 | ||
223 | return ret; | |
224 | } | |
225 | ||
35402e31 | 226 | #if IS_ENABLED(CONFIG_IPV6) |
35402e31 DA |
227 | /* modelled after ip6_finish_output2 */ |
228 | static int vrf_finish_output6(struct net *net, struct sock *sk, | |
229 | struct sk_buff *skb) | |
230 | { | |
231 | struct dst_entry *dst = skb_dst(skb); | |
232 | struct net_device *dev = dst->dev; | |
233 | struct neighbour *neigh; | |
234 | struct in6_addr *nexthop; | |
235 | int ret; | |
236 | ||
237 | skb->protocol = htons(ETH_P_IPV6); | |
238 | skb->dev = dev; | |
239 | ||
240 | rcu_read_lock_bh(); | |
241 | nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); | |
242 | neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); | |
243 | if (unlikely(!neigh)) | |
244 | neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); | |
245 | if (!IS_ERR(neigh)) { | |
246 | ret = dst_neigh_output(dst, neigh, skb); | |
247 | rcu_read_unlock_bh(); | |
248 | return ret; | |
249 | } | |
250 | rcu_read_unlock_bh(); | |
251 | ||
252 | IP6_INC_STATS(dev_net(dst->dev), | |
253 | ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); | |
254 | kfree_skb(skb); | |
255 | return -EINVAL; | |
256 | } | |
257 | ||
258 | /* modelled after ip6_output */ | |
259 | static int vrf_output6(struct net *net, struct sock *sk, struct sk_buff *skb) | |
260 | { | |
261 | return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, | |
262 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
263 | vrf_finish_output6, | |
264 | !(IP6CB(skb)->flags & IP6SKB_REROUTED)); | |
265 | } | |
266 | ||
b0e95ccd | 267 | /* holding rtnl */ |
9ab179d8 | 268 | static void vrf_rt6_release(struct net_vrf *vrf) |
35402e31 | 269 | { |
b0e95ccd DA |
270 | struct rt6_info *rt6 = rtnl_dereference(vrf->rt6); |
271 | ||
3d9dc408 | 272 | rcu_assign_pointer(vrf->rt6, NULL); |
b0e95ccd DA |
273 | |
274 | if (rt6) | |
275 | dst_release(&rt6->dst); | |
35402e31 DA |
276 | } |
277 | ||
278 | static int vrf_rt6_create(struct net_device *dev) | |
279 | { | |
280 | struct net_vrf *vrf = netdev_priv(dev); | |
9ab179d8 | 281 | struct net *net = dev_net(dev); |
b3b4663c | 282 | struct fib6_table *rt6i_table; |
3d9dc408 | 283 | struct rt6_info *rt6; |
35402e31 DA |
284 | int rc = -ENOMEM; |
285 | ||
b3b4663c DA |
286 | rt6i_table = fib6_new_table(net, vrf->tb_id); |
287 | if (!rt6i_table) | |
288 | goto out; | |
289 | ||
3d9dc408 DM |
290 | rt6 = ip6_dst_alloc(net, dev, |
291 | DST_HOST | DST_NOPOLICY | DST_NOXFRM | DST_NOCACHE); | |
35402e31 DA |
292 | if (!rt6) |
293 | goto out; | |
294 | ||
9ab179d8 | 295 | dst_hold(&rt6->dst); |
b3b4663c DA |
296 | |
297 | rt6->rt6i_table = rt6i_table; | |
298 | rt6->dst.output = vrf_output6; | |
b0e95ccd DA |
299 | rcu_assign_pointer(vrf->rt6, rt6); |
300 | ||
35402e31 DA |
301 | rc = 0; |
302 | out: | |
303 | return rc; | |
304 | } | |
305 | #else | |
9ab179d8 | 306 | static void vrf_rt6_release(struct net_vrf *vrf) |
35402e31 DA |
307 | { |
308 | } | |
309 | ||
310 | static int vrf_rt6_create(struct net_device *dev) | |
311 | { | |
312 | return 0; | |
313 | } | |
314 | #endif | |
315 | ||
8f58336d | 316 | /* modelled after ip_finish_output2 */ |
0c4b51f0 | 317 | static int vrf_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
193125db | 318 | { |
8f58336d DA |
319 | struct dst_entry *dst = skb_dst(skb); |
320 | struct rtable *rt = (struct rtable *)dst; | |
321 | struct net_device *dev = dst->dev; | |
322 | unsigned int hh_len = LL_RESERVED_SPACE(dev); | |
323 | struct neighbour *neigh; | |
324 | u32 nexthop; | |
325 | int ret = -EINVAL; | |
326 | ||
327 | /* Be paranoid, rather than too clever. */ | |
328 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { | |
329 | struct sk_buff *skb2; | |
330 | ||
331 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
332 | if (!skb2) { | |
333 | ret = -ENOMEM; | |
334 | goto err; | |
335 | } | |
336 | if (skb->sk) | |
337 | skb_set_owner_w(skb2, skb->sk); | |
338 | ||
339 | consume_skb(skb); | |
340 | skb = skb2; | |
341 | } | |
342 | ||
343 | rcu_read_lock_bh(); | |
344 | ||
345 | nexthop = (__force u32)rt_nexthop(rt, ip_hdr(skb)->daddr); | |
346 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); | |
347 | if (unlikely(!neigh)) | |
348 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
349 | if (!IS_ERR(neigh)) | |
350 | ret = dst_neigh_output(dst, neigh, skb); | |
351 | ||
352 | rcu_read_unlock_bh(); | |
353 | err: | |
354 | if (unlikely(ret < 0)) | |
355 | vrf_tx_error(skb->dev, skb); | |
356 | return ret; | |
193125db DA |
357 | } |
358 | ||
ede2059d | 359 | static int vrf_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
193125db DA |
360 | { |
361 | struct net_device *dev = skb_dst(skb)->dev; | |
362 | ||
29a26a56 | 363 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
193125db DA |
364 | |
365 | skb->dev = dev; | |
366 | skb->protocol = htons(ETH_P_IP); | |
367 | ||
29a26a56 EB |
368 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
369 | net, sk, skb, NULL, dev, | |
8f58336d | 370 | vrf_finish_output, |
193125db DA |
371 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
372 | } | |
373 | ||
b0e95ccd | 374 | /* holding rtnl */ |
9ab179d8 | 375 | static void vrf_rtable_release(struct net_vrf *vrf) |
193125db | 376 | { |
b0e95ccd DA |
377 | struct rtable *rth = rtnl_dereference(vrf->rth); |
378 | ||
3d9dc408 | 379 | rcu_assign_pointer(vrf->rth, NULL); |
193125db | 380 | |
b0e95ccd DA |
381 | if (rth) |
382 | dst_release(&rth->dst); | |
193125db DA |
383 | } |
384 | ||
b0e95ccd | 385 | static int vrf_rtable_create(struct net_device *dev) |
193125db | 386 | { |
b7503e0c | 387 | struct net_vrf *vrf = netdev_priv(dev); |
3d9dc408 | 388 | struct rtable *rth; |
193125db | 389 | |
b3b4663c | 390 | if (!fib_new_table(dev_net(dev), vrf->tb_id)) |
b0e95ccd | 391 | return -ENOMEM; |
b3b4663c | 392 | |
9ab179d8 | 393 | rth = rt_dst_alloc(dev, 0, RTN_UNICAST, 1, 1, 0); |
b0e95ccd DA |
394 | if (!rth) |
395 | return -ENOMEM; | |
193125db | 396 | |
b0e95ccd DA |
397 | rth->dst.output = vrf_output; |
398 | rth->rt_table_id = vrf->tb_id; | |
399 | ||
400 | rcu_assign_pointer(vrf->rth, rth); | |
401 | ||
402 | return 0; | |
193125db DA |
403 | } |
404 | ||
405 | /**************************** device handling ********************/ | |
406 | ||
407 | /* cycle interface to flush neighbor cache and move routes across tables */ | |
408 | static void cycle_netdev(struct net_device *dev) | |
409 | { | |
410 | unsigned int flags = dev->flags; | |
411 | int ret; | |
412 | ||
413 | if (!netif_running(dev)) | |
414 | return; | |
415 | ||
416 | ret = dev_change_flags(dev, flags & ~IFF_UP); | |
417 | if (ret >= 0) | |
418 | ret = dev_change_flags(dev, flags); | |
419 | ||
420 | if (ret < 0) { | |
421 | netdev_err(dev, | |
422 | "Failed to cycle device %s; route tables might be wrong!\n", | |
423 | dev->name); | |
424 | } | |
425 | } | |
426 | ||
193125db DA |
427 | static int do_vrf_add_slave(struct net_device *dev, struct net_device *port_dev) |
428 | { | |
bad53162 | 429 | int ret; |
193125db | 430 | |
29bf24af | 431 | ret = netdev_master_upper_dev_link(port_dev, dev, NULL, NULL); |
193125db | 432 | if (ret < 0) |
74b20582 | 433 | return ret; |
193125db | 434 | |
fee6d4c7 | 435 | port_dev->priv_flags |= IFF_L3MDEV_SLAVE; |
193125db DA |
436 | cycle_netdev(port_dev); |
437 | ||
438 | return 0; | |
193125db DA |
439 | } |
440 | ||
441 | static int vrf_add_slave(struct net_device *dev, struct net_device *port_dev) | |
442 | { | |
fee6d4c7 | 443 | if (netif_is_l3_master(port_dev) || netif_is_l3_slave(port_dev)) |
193125db DA |
444 | return -EINVAL; |
445 | ||
446 | return do_vrf_add_slave(dev, port_dev); | |
447 | } | |
448 | ||
449 | /* inverse of do_vrf_add_slave */ | |
450 | static int do_vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
451 | { | |
193125db | 452 | netdev_upper_dev_unlink(port_dev, dev); |
fee6d4c7 | 453 | port_dev->priv_flags &= ~IFF_L3MDEV_SLAVE; |
193125db | 454 | |
193125db DA |
455 | cycle_netdev(port_dev); |
456 | ||
193125db DA |
457 | return 0; |
458 | } | |
459 | ||
460 | static int vrf_del_slave(struct net_device *dev, struct net_device *port_dev) | |
461 | { | |
193125db DA |
462 | return do_vrf_del_slave(dev, port_dev); |
463 | } | |
464 | ||
465 | static void vrf_dev_uninit(struct net_device *dev) | |
466 | { | |
467 | struct net_vrf *vrf = netdev_priv(dev); | |
bad53162 NA |
468 | struct net_device *port_dev; |
469 | struct list_head *iter; | |
193125db | 470 | |
9ab179d8 DA |
471 | vrf_rtable_release(vrf); |
472 | vrf_rt6_release(vrf); | |
193125db | 473 | |
bad53162 NA |
474 | netdev_for_each_lower_dev(dev, port_dev, iter) |
475 | vrf_del_slave(dev, port_dev); | |
193125db | 476 | |
3a4a27d3 | 477 | free_percpu(dev->dstats); |
193125db DA |
478 | dev->dstats = NULL; |
479 | } | |
480 | ||
481 | static int vrf_dev_init(struct net_device *dev) | |
482 | { | |
483 | struct net_vrf *vrf = netdev_priv(dev); | |
484 | ||
193125db DA |
485 | dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); |
486 | if (!dev->dstats) | |
487 | goto out_nomem; | |
488 | ||
489 | /* create the default dst which points back to us */ | |
b0e95ccd | 490 | if (vrf_rtable_create(dev) != 0) |
193125db DA |
491 | goto out_stats; |
492 | ||
35402e31 DA |
493 | if (vrf_rt6_create(dev) != 0) |
494 | goto out_rth; | |
495 | ||
193125db DA |
496 | dev->flags = IFF_MASTER | IFF_NOARP; |
497 | ||
b87ab6b8 DA |
498 | /* MTU is irrelevant for VRF device; set to 64k similar to lo */ |
499 | dev->mtu = 64 * 1024; | |
500 | ||
501 | /* similarly, oper state is irrelevant; set to up to avoid confusion */ | |
502 | dev->operstate = IF_OPER_UP; | |
503 | ||
193125db DA |
504 | return 0; |
505 | ||
35402e31 | 506 | out_rth: |
9ab179d8 | 507 | vrf_rtable_release(vrf); |
193125db DA |
508 | out_stats: |
509 | free_percpu(dev->dstats); | |
510 | dev->dstats = NULL; | |
511 | out_nomem: | |
512 | return -ENOMEM; | |
513 | } | |
514 | ||
515 | static const struct net_device_ops vrf_netdev_ops = { | |
516 | .ndo_init = vrf_dev_init, | |
517 | .ndo_uninit = vrf_dev_uninit, | |
518 | .ndo_start_xmit = vrf_xmit, | |
519 | .ndo_get_stats64 = vrf_get_stats64, | |
520 | .ndo_add_slave = vrf_add_slave, | |
521 | .ndo_del_slave = vrf_del_slave, | |
522 | }; | |
523 | ||
ee15ee5d DA |
524 | static u32 vrf_fib_table(const struct net_device *dev) |
525 | { | |
526 | struct net_vrf *vrf = netdev_priv(dev); | |
527 | ||
528 | return vrf->tb_id; | |
529 | } | |
530 | ||
531 | static struct rtable *vrf_get_rtable(const struct net_device *dev, | |
532 | const struct flowi4 *fl4) | |
533 | { | |
534 | struct rtable *rth = NULL; | |
535 | ||
6e2895a8 | 536 | if (!(fl4->flowi4_flags & FLOWI_FLAG_L3MDEV_SRC)) { |
ee15ee5d DA |
537 | struct net_vrf *vrf = netdev_priv(dev); |
538 | ||
b0e95ccd DA |
539 | rcu_read_lock(); |
540 | ||
541 | rth = rcu_dereference(vrf->rth); | |
542 | if (likely(rth)) | |
543 | dst_hold(&rth->dst); | |
544 | ||
545 | rcu_read_unlock(); | |
ee15ee5d DA |
546 | } |
547 | ||
548 | return rth; | |
549 | } | |
550 | ||
8cbb512c | 551 | /* called under rcu_read_lock */ |
b5bdacf3 | 552 | static int vrf_get_saddr(struct net_device *dev, struct flowi4 *fl4) |
8cbb512c DA |
553 | { |
554 | struct fib_result res = { .tclassid = 0 }; | |
555 | struct net *net = dev_net(dev); | |
556 | u32 orig_tos = fl4->flowi4_tos; | |
557 | u8 flags = fl4->flowi4_flags; | |
558 | u8 scope = fl4->flowi4_scope; | |
559 | u8 tos = RT_FL_TOS(fl4); | |
b5bdacf3 | 560 | int rc; |
8cbb512c DA |
561 | |
562 | if (unlikely(!fl4->daddr)) | |
b5bdacf3 | 563 | return 0; |
8cbb512c DA |
564 | |
565 | fl4->flowi4_flags |= FLOWI_FLAG_SKIP_NH_OIF; | |
566 | fl4->flowi4_iif = LOOPBACK_IFINDEX; | |
1ff23bee DA |
567 | /* make sure oif is set to VRF device for lookup */ |
568 | fl4->flowi4_oif = dev->ifindex; | |
8cbb512c DA |
569 | fl4->flowi4_tos = tos & IPTOS_RT_MASK; |
570 | fl4->flowi4_scope = ((tos & RTO_ONLINK) ? | |
571 | RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); | |
572 | ||
b5bdacf3 DA |
573 | rc = fib_lookup(net, fl4, &res, 0); |
574 | if (!rc) { | |
8cbb512c DA |
575 | if (res.type == RTN_LOCAL) |
576 | fl4->saddr = res.fi->fib_prefsrc ? : fl4->daddr; | |
577 | else | |
578 | fib_select_path(net, &res, fl4, -1); | |
579 | } | |
580 | ||
581 | fl4->flowi4_flags = flags; | |
582 | fl4->flowi4_tos = orig_tos; | |
583 | fl4->flowi4_scope = scope; | |
b5bdacf3 DA |
584 | |
585 | return rc; | |
8cbb512c DA |
586 | } |
587 | ||
74b20582 DA |
588 | #if IS_ENABLED(CONFIG_IPV6) |
589 | /* neighbor handling is done with actual device; do not want | |
590 | * to flip skb->dev for those ndisc packets. This really fails | |
591 | * for multiple next protocols (e.g., NEXTHDR_HOP). But it is | |
592 | * a start. | |
593 | */ | |
594 | static bool ipv6_ndisc_frame(const struct sk_buff *skb) | |
595 | { | |
596 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
597 | bool rc = false; | |
598 | ||
599 | if (iph->nexthdr == NEXTHDR_ICMP) { | |
600 | const struct icmp6hdr *icmph; | |
601 | struct icmp6hdr _icmph; | |
602 | ||
603 | icmph = skb_header_pointer(skb, sizeof(*iph), | |
604 | sizeof(_icmph), &_icmph); | |
605 | if (!icmph) | |
606 | goto out; | |
607 | ||
608 | switch (icmph->icmp6_type) { | |
609 | case NDISC_ROUTER_SOLICITATION: | |
610 | case NDISC_ROUTER_ADVERTISEMENT: | |
611 | case NDISC_NEIGHBOUR_SOLICITATION: | |
612 | case NDISC_NEIGHBOUR_ADVERTISEMENT: | |
613 | case NDISC_REDIRECT: | |
614 | rc = true; | |
615 | break; | |
616 | } | |
617 | } | |
618 | ||
619 | out: | |
620 | return rc; | |
621 | } | |
622 | ||
623 | static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, | |
624 | struct sk_buff *skb) | |
625 | { | |
626 | /* if packet is NDISC keep the ingress interface */ | |
627 | if (!ipv6_ndisc_frame(skb)) { | |
628 | skb->dev = vrf_dev; | |
629 | skb->skb_iif = vrf_dev->ifindex; | |
630 | ||
631 | skb_push(skb, skb->mac_len); | |
632 | dev_queue_xmit_nit(skb, vrf_dev); | |
633 | skb_pull(skb, skb->mac_len); | |
634 | ||
635 | IP6CB(skb)->flags |= IP6SKB_L3SLAVE; | |
636 | } | |
637 | ||
638 | return skb; | |
639 | } | |
640 | ||
641 | #else | |
642 | static struct sk_buff *vrf_ip6_rcv(struct net_device *vrf_dev, | |
643 | struct sk_buff *skb) | |
644 | { | |
645 | return skb; | |
646 | } | |
647 | #endif | |
648 | ||
649 | static struct sk_buff *vrf_ip_rcv(struct net_device *vrf_dev, | |
650 | struct sk_buff *skb) | |
651 | { | |
652 | skb->dev = vrf_dev; | |
653 | skb->skb_iif = vrf_dev->ifindex; | |
654 | ||
655 | skb_push(skb, skb->mac_len); | |
656 | dev_queue_xmit_nit(skb, vrf_dev); | |
657 | skb_pull(skb, skb->mac_len); | |
658 | ||
659 | return skb; | |
660 | } | |
661 | ||
662 | /* called with rcu lock held */ | |
663 | static struct sk_buff *vrf_l3_rcv(struct net_device *vrf_dev, | |
664 | struct sk_buff *skb, | |
665 | u16 proto) | |
666 | { | |
667 | switch (proto) { | |
668 | case AF_INET: | |
669 | return vrf_ip_rcv(vrf_dev, skb); | |
670 | case AF_INET6: | |
671 | return vrf_ip6_rcv(vrf_dev, skb); | |
672 | } | |
673 | ||
674 | return skb; | |
675 | } | |
676 | ||
35402e31 DA |
677 | #if IS_ENABLED(CONFIG_IPV6) |
678 | static struct dst_entry *vrf_get_rt6_dst(const struct net_device *dev, | |
679 | const struct flowi6 *fl6) | |
680 | { | |
b0e95ccd | 681 | struct dst_entry *dst = NULL; |
35402e31 DA |
682 | |
683 | if (!(fl6->flowi6_flags & FLOWI_FLAG_L3MDEV_SRC)) { | |
684 | struct net_vrf *vrf = netdev_priv(dev); | |
b0e95ccd DA |
685 | struct rt6_info *rt; |
686 | ||
687 | rcu_read_lock(); | |
688 | ||
689 | rt = rcu_dereference(vrf->rt6); | |
690 | if (likely(rt)) { | |
691 | dst = &rt->dst; | |
692 | dst_hold(dst); | |
693 | } | |
35402e31 | 694 | |
b0e95ccd | 695 | rcu_read_unlock(); |
35402e31 DA |
696 | } |
697 | ||
b0e95ccd | 698 | return dst; |
35402e31 DA |
699 | } |
700 | #endif | |
701 | ||
ee15ee5d DA |
702 | static const struct l3mdev_ops vrf_l3mdev_ops = { |
703 | .l3mdev_fib_table = vrf_fib_table, | |
704 | .l3mdev_get_rtable = vrf_get_rtable, | |
8cbb512c | 705 | .l3mdev_get_saddr = vrf_get_saddr, |
74b20582 | 706 | .l3mdev_l3_rcv = vrf_l3_rcv, |
35402e31 DA |
707 | #if IS_ENABLED(CONFIG_IPV6) |
708 | .l3mdev_get_rt6_dst = vrf_get_rt6_dst, | |
709 | #endif | |
ee15ee5d DA |
710 | }; |
711 | ||
193125db DA |
712 | static void vrf_get_drvinfo(struct net_device *dev, |
713 | struct ethtool_drvinfo *info) | |
714 | { | |
715 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); | |
716 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
717 | } | |
718 | ||
719 | static const struct ethtool_ops vrf_ethtool_ops = { | |
720 | .get_drvinfo = vrf_get_drvinfo, | |
721 | }; | |
722 | ||
723 | static void vrf_setup(struct net_device *dev) | |
724 | { | |
725 | ether_setup(dev); | |
726 | ||
727 | /* Initialize the device structure. */ | |
728 | dev->netdev_ops = &vrf_netdev_ops; | |
ee15ee5d | 729 | dev->l3mdev_ops = &vrf_l3mdev_ops; |
193125db DA |
730 | dev->ethtool_ops = &vrf_ethtool_ops; |
731 | dev->destructor = free_netdev; | |
732 | ||
733 | /* Fill in device structure with ethernet-generic values. */ | |
734 | eth_hw_addr_random(dev); | |
735 | ||
736 | /* don't acquire vrf device's netif_tx_lock when transmitting */ | |
737 | dev->features |= NETIF_F_LLTX; | |
738 | ||
739 | /* don't allow vrf devices to change network namespaces. */ | |
740 | dev->features |= NETIF_F_NETNS_LOCAL; | |
741 | } | |
742 | ||
743 | static int vrf_validate(struct nlattr *tb[], struct nlattr *data[]) | |
744 | { | |
745 | if (tb[IFLA_ADDRESS]) { | |
746 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
747 | return -EINVAL; | |
748 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
749 | return -EADDRNOTAVAIL; | |
750 | } | |
751 | return 0; | |
752 | } | |
753 | ||
754 | static void vrf_dellink(struct net_device *dev, struct list_head *head) | |
755 | { | |
193125db DA |
756 | unregister_netdevice_queue(dev, head); |
757 | } | |
758 | ||
759 | static int vrf_newlink(struct net *src_net, struct net_device *dev, | |
760 | struct nlattr *tb[], struct nlattr *data[]) | |
761 | { | |
762 | struct net_vrf *vrf = netdev_priv(dev); | |
193125db DA |
763 | |
764 | if (!data || !data[IFLA_VRF_TABLE]) | |
765 | return -EINVAL; | |
766 | ||
767 | vrf->tb_id = nla_get_u32(data[IFLA_VRF_TABLE]); | |
768 | ||
007979ea | 769 | dev->priv_flags |= IFF_L3MDEV_MASTER; |
193125db | 770 | |
7f109f7c | 771 | return register_netdevice(dev); |
193125db DA |
772 | } |
773 | ||
774 | static size_t vrf_nl_getsize(const struct net_device *dev) | |
775 | { | |
776 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_TABLE */ | |
777 | } | |
778 | ||
779 | static int vrf_fillinfo(struct sk_buff *skb, | |
780 | const struct net_device *dev) | |
781 | { | |
782 | struct net_vrf *vrf = netdev_priv(dev); | |
783 | ||
784 | return nla_put_u32(skb, IFLA_VRF_TABLE, vrf->tb_id); | |
785 | } | |
786 | ||
67eb0331 DA |
787 | static size_t vrf_get_slave_size(const struct net_device *bond_dev, |
788 | const struct net_device *slave_dev) | |
789 | { | |
790 | return nla_total_size(sizeof(u32)); /* IFLA_VRF_PORT_TABLE */ | |
791 | } | |
792 | ||
793 | static int vrf_fill_slave_info(struct sk_buff *skb, | |
794 | const struct net_device *vrf_dev, | |
795 | const struct net_device *slave_dev) | |
796 | { | |
797 | struct net_vrf *vrf = netdev_priv(vrf_dev); | |
798 | ||
799 | if (nla_put_u32(skb, IFLA_VRF_PORT_TABLE, vrf->tb_id)) | |
800 | return -EMSGSIZE; | |
801 | ||
802 | return 0; | |
803 | } | |
804 | ||
193125db DA |
805 | static const struct nla_policy vrf_nl_policy[IFLA_VRF_MAX + 1] = { |
806 | [IFLA_VRF_TABLE] = { .type = NLA_U32 }, | |
807 | }; | |
808 | ||
809 | static struct rtnl_link_ops vrf_link_ops __read_mostly = { | |
810 | .kind = DRV_NAME, | |
811 | .priv_size = sizeof(struct net_vrf), | |
812 | ||
813 | .get_size = vrf_nl_getsize, | |
814 | .policy = vrf_nl_policy, | |
815 | .validate = vrf_validate, | |
816 | .fill_info = vrf_fillinfo, | |
817 | ||
67eb0331 DA |
818 | .get_slave_size = vrf_get_slave_size, |
819 | .fill_slave_info = vrf_fill_slave_info, | |
820 | ||
193125db DA |
821 | .newlink = vrf_newlink, |
822 | .dellink = vrf_dellink, | |
823 | .setup = vrf_setup, | |
824 | .maxtype = IFLA_VRF_MAX, | |
825 | }; | |
826 | ||
827 | static int vrf_device_event(struct notifier_block *unused, | |
828 | unsigned long event, void *ptr) | |
829 | { | |
830 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
831 | ||
832 | /* only care about unregister events to drop slave references */ | |
833 | if (event == NETDEV_UNREGISTER) { | |
193125db DA |
834 | struct net_device *vrf_dev; |
835 | ||
fee6d4c7 | 836 | if (!netif_is_l3_slave(dev)) |
193125db DA |
837 | goto out; |
838 | ||
58aa9087 NA |
839 | vrf_dev = netdev_master_upper_dev_get(dev); |
840 | vrf_del_slave(vrf_dev, dev); | |
193125db DA |
841 | } |
842 | out: | |
843 | return NOTIFY_DONE; | |
844 | } | |
845 | ||
846 | static struct notifier_block vrf_notifier_block __read_mostly = { | |
847 | .notifier_call = vrf_device_event, | |
848 | }; | |
849 | ||
850 | static int __init vrf_init_module(void) | |
851 | { | |
852 | int rc; | |
853 | ||
193125db DA |
854 | register_netdevice_notifier(&vrf_notifier_block); |
855 | ||
856 | rc = rtnl_link_register(&vrf_link_ops); | |
857 | if (rc < 0) | |
858 | goto error; | |
859 | ||
860 | return 0; | |
861 | ||
862 | error: | |
863 | unregister_netdevice_notifier(&vrf_notifier_block); | |
193125db DA |
864 | return rc; |
865 | } | |
866 | ||
193125db | 867 | module_init(vrf_init_module); |
193125db DA |
868 | MODULE_AUTHOR("Shrijeet Mukherjee, David Ahern"); |
869 | MODULE_DESCRIPTION("Device driver to instantiate VRF domains"); | |
870 | MODULE_LICENSE("GPL"); | |
871 | MODULE_ALIAS_RTNL_LINK(DRV_NAME); | |
872 | MODULE_VERSION(DRV_VERSION); |