1 #include <linux/types.h>
2 #include <linux/skbuff.h>
3 #include <linux/socket.h>
4 #include <linux/sysctl.h>
5 #include <linux/net.h>
6 #include <linux/module.h>
7 #include <linux/if_arp.h>
8 #include <linux/ipv6.h>
9 #include <linux/mpls.h>
10 #include <linux/nospec.h>
11 #include <linux/vmalloc.h>
12 #include <net/ip.h>
13 #include <net/dst.h>
14 #include <net/sock.h>
15 #include <net/arp.h>
16 #include <net/ip_fib.h>
17 #include <net/netevent.h>
18 #include <net/netns/generic.h>
19 #if IS_ENABLED(CONFIG_IPV6)
20 #include <net/ipv6.h>
21 #include <net/addrconf.h>
22 #endif
23 #include <net/nexthop.h>
24 #include "internal.h"
25
26 /* Maximum number of labels to look ahead at when selecting a path of
27 * a multipath route
28 */
29 #define MAX_MP_SELECT_LABELS 4
30
31 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
32
33 static int zero = 0;
34 static int label_limit = (1 << 20) - 1;
35
36 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
37 struct nlmsghdr *nlh, struct net *net, u32 portid,
38 unsigned int nlm_flags);
39
mpls_route_input_rcu(struct net * net,unsigned index)40 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
41 {
42 struct mpls_route *rt = NULL;
43
44 if (index < net->mpls.platform_labels) {
45 struct mpls_route __rcu **platform_label =
46 rcu_dereference(net->mpls.platform_label);
47 rt = rcu_dereference(platform_label[index]);
48 }
49 return rt;
50 }
51
mpls_dev_get(const struct net_device * dev)52 static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
53 {
54 return rcu_dereference_rtnl(dev->mpls_ptr);
55 }
56
mpls_output_possible(const struct net_device * dev)57 bool mpls_output_possible(const struct net_device *dev)
58 {
59 return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
60 }
61 EXPORT_SYMBOL_GPL(mpls_output_possible);
62
__mpls_nh_via(struct mpls_route * rt,struct mpls_nh * nh)63 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
64 {
65 u8 *nh0_via = PTR_ALIGN((u8 *)&rt->rt_nh[rt->rt_nhn], VIA_ALEN_ALIGN);
66 int nh_index = nh - rt->rt_nh;
67
68 return nh0_via + rt->rt_max_alen * nh_index;
69 }
70
mpls_nh_via(const struct mpls_route * rt,const struct mpls_nh * nh)71 static const u8 *mpls_nh_via(const struct mpls_route *rt,
72 const struct mpls_nh *nh)
73 {
74 return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
75 }
76
mpls_nh_header_size(const struct mpls_nh * nh)77 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
78 {
79 /* The size of the layer 2.5 labels to be added for this route */
80 return nh->nh_labels * sizeof(struct mpls_shim_hdr);
81 }
82
mpls_dev_mtu(const struct net_device * dev)83 unsigned int mpls_dev_mtu(const struct net_device *dev)
84 {
85 /* The amount of data the layer 2 frame can hold */
86 return dev->mtu;
87 }
88 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
89
mpls_pkt_too_big(const struct sk_buff * skb,unsigned int mtu)90 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
91 {
92 if (skb->len <= mtu)
93 return false;
94
95 if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
96 return false;
97
98 return true;
99 }
100 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
101
mpls_multipath_hash(struct mpls_route * rt,struct sk_buff * skb)102 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
103 {
104 struct mpls_entry_decoded dec;
105 unsigned int mpls_hdr_len = 0;
106 struct mpls_shim_hdr *hdr;
107 bool eli_seen = false;
108 int label_index;
109 u32 hash = 0;
110
111 for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
112 label_index++) {
113 mpls_hdr_len += sizeof(*hdr);
114 if (!pskb_may_pull(skb, mpls_hdr_len))
115 break;
116
117 /* Read and decode the current label */
118 hdr = mpls_hdr(skb) + label_index;
119 dec = mpls_entry_decode(hdr);
120
121 /* RFC6790 - reserved labels MUST NOT be used as keys
122 * for the load-balancing function
123 */
124 if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
125 hash = jhash_1word(dec.label, hash);
126
127 /* The entropy label follows the entropy label
128 * indicator, so this means that the entropy
129 * label was just added to the hash - no need to
130 * go any deeper either in the label stack or in the
131 * payload
132 */
133 if (eli_seen)
134 break;
135 } else if (dec.label == MPLS_LABEL_ENTROPY) {
136 eli_seen = true;
137 }
138
139 if (!dec.bos)
140 continue;
141
142 /* found bottom label; does skb have room for a header? */
143 if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
144 const struct iphdr *v4hdr;
145
146 v4hdr = (const struct iphdr *)(hdr + 1);
147 if (v4hdr->version == 4) {
148 hash = jhash_3words(ntohl(v4hdr->saddr),
149 ntohl(v4hdr->daddr),
150 v4hdr->protocol, hash);
151 } else if (v4hdr->version == 6 &&
152 pskb_may_pull(skb, mpls_hdr_len +
153 sizeof(struct ipv6hdr))) {
154 const struct ipv6hdr *v6hdr;
155
156 v6hdr = (const struct ipv6hdr *)(hdr + 1);
157 hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
158 hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
159 hash = jhash_1word(v6hdr->nexthdr, hash);
160 }
161 }
162
163 break;
164 }
165
166 return hash;
167 }
168
mpls_select_multipath(struct mpls_route * rt,struct sk_buff * skb)169 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
170 struct sk_buff *skb)
171 {
172 int alive = ACCESS_ONCE(rt->rt_nhn_alive);
173 u32 hash = 0;
174 int nh_index = 0;
175 int n = 0;
176
177 /* No need to look further into packet if there's only
178 * one path
179 */
180 if (rt->rt_nhn == 1)
181 goto out;
182
183 if (alive <= 0)
184 return NULL;
185
186 hash = mpls_multipath_hash(rt, skb);
187 nh_index = hash % alive;
188 if (alive == rt->rt_nhn)
189 goto out;
190 for_nexthops(rt) {
191 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
192 continue;
193 if (n == nh_index)
194 return nh;
195 n++;
196 } endfor_nexthops(rt);
197
198 out:
199 return &rt->rt_nh[nh_index];
200 }
201
mpls_egress(struct mpls_route * rt,struct sk_buff * skb,struct mpls_entry_decoded dec)202 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
203 struct mpls_entry_decoded dec)
204 {
205 enum mpls_payload_type payload_type;
206 bool success = false;
207
208 /* The IPv4 code below accesses through the IPv4 header
209 * checksum, which is 12 bytes into the packet.
210 * The IPv6 code below accesses through the IPv6 hop limit
211 * which is 8 bytes into the packet.
212 *
213 * For all supported cases there should always be at least 12
214 * bytes of packet data present. The IPv4 header is 20 bytes
215 * without options and the IPv6 header is always 40 bytes
216 * long.
217 */
218 if (!pskb_may_pull(skb, 12))
219 return false;
220
221 payload_type = rt->rt_payload_type;
222 if (payload_type == MPT_UNSPEC)
223 payload_type = ip_hdr(skb)->version;
224
225 switch (payload_type) {
226 case MPT_IPV4: {
227 struct iphdr *hdr4 = ip_hdr(skb);
228 skb->protocol = htons(ETH_P_IP);
229 csum_replace2(&hdr4->check,
230 htons(hdr4->ttl << 8),
231 htons(dec.ttl << 8));
232 hdr4->ttl = dec.ttl;
233 success = true;
234 break;
235 }
236 case MPT_IPV6: {
237 struct ipv6hdr *hdr6 = ipv6_hdr(skb);
238 skb->protocol = htons(ETH_P_IPV6);
239 hdr6->hop_limit = dec.ttl;
240 success = true;
241 break;
242 }
243 case MPT_UNSPEC:
244 break;
245 }
246
247 return success;
248 }
249
mpls_forward(struct sk_buff * skb,struct net_device * dev,struct packet_type * pt,struct net_device * orig_dev)250 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
251 struct packet_type *pt, struct net_device *orig_dev)
252 {
253 struct net *net = dev_net(dev);
254 struct mpls_shim_hdr *hdr;
255 struct mpls_route *rt;
256 struct mpls_nh *nh;
257 struct mpls_entry_decoded dec;
258 struct net_device *out_dev;
259 struct mpls_dev *mdev;
260 unsigned int hh_len;
261 unsigned int new_header_size;
262 unsigned int mtu;
263 int err;
264
265 /* Careful this entire function runs inside of an rcu critical section */
266
267 mdev = mpls_dev_get(dev);
268 if (!mdev || !mdev->input_enabled)
269 goto drop;
270
271 if (skb->pkt_type != PACKET_HOST)
272 goto drop;
273
274 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
275 goto drop;
276
277 if (!pskb_may_pull(skb, sizeof(*hdr)))
278 goto drop;
279
280 /* Read and decode the label */
281 hdr = mpls_hdr(skb);
282 dec = mpls_entry_decode(hdr);
283
284 rt = mpls_route_input_rcu(net, dec.label);
285 if (!rt)
286 goto drop;
287
288 nh = mpls_select_multipath(rt, skb);
289 if (!nh)
290 goto drop;
291
292 /* Find the output device */
293 out_dev = rcu_dereference(nh->nh_dev);
294 if (!mpls_output_possible(out_dev))
295 goto drop;
296
297 /* Pop the label */
298 skb_pull(skb, sizeof(*hdr));
299 skb_reset_network_header(skb);
300
301 skb_orphan(skb);
302
303 if (skb_warn_if_lro(skb))
304 goto drop;
305
306 skb_forward_csum(skb);
307
308 /* Verify ttl is valid */
309 if (dec.ttl <= 1)
310 goto drop;
311 dec.ttl -= 1;
312
313 /* Verify the destination can hold the packet */
314 new_header_size = mpls_nh_header_size(nh);
315 mtu = mpls_dev_mtu(out_dev);
316 if (mpls_pkt_too_big(skb, mtu - new_header_size))
317 goto drop;
318
319 hh_len = LL_RESERVED_SPACE(out_dev);
320 if (!out_dev->header_ops)
321 hh_len = 0;
322
323 /* Ensure there is enough space for the headers in the skb */
324 if (skb_cow(skb, hh_len + new_header_size))
325 goto drop;
326
327 skb->dev = out_dev;
328 skb->protocol = htons(ETH_P_MPLS_UC);
329
330 if (unlikely(!new_header_size && dec.bos)) {
331 /* Penultimate hop popping */
332 if (!mpls_egress(rt, skb, dec))
333 goto drop;
334 } else {
335 bool bos;
336 int i;
337 skb_push(skb, new_header_size);
338 skb_reset_network_header(skb);
339 /* Push the new labels */
340 hdr = mpls_hdr(skb);
341 bos = dec.bos;
342 for (i = nh->nh_labels - 1; i >= 0; i--) {
343 hdr[i] = mpls_entry_encode(nh->nh_label[i],
344 dec.ttl, 0, bos);
345 bos = false;
346 }
347 }
348
349 /* If via wasn't specified then send out using device address */
350 if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
351 err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
352 out_dev->dev_addr, skb);
353 else
354 err = neigh_xmit(nh->nh_via_table, out_dev,
355 mpls_nh_via(rt, nh), skb);
356 if (err)
357 net_dbg_ratelimited("%s: packet transmission failed: %d\n",
358 __func__, err);
359 return 0;
360
361 drop:
362 kfree_skb(skb);
363 return NET_RX_DROP;
364 }
365
366 static struct packet_type mpls_packet_type __read_mostly = {
367 .type = cpu_to_be16(ETH_P_MPLS_UC),
368 .func = mpls_forward,
369 };
370
371 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
372 [RTA_DST] = { .type = NLA_U32 },
373 [RTA_OIF] = { .type = NLA_U32 },
374 };
375
376 struct mpls_route_config {
377 u32 rc_protocol;
378 u32 rc_ifindex;
379 u8 rc_via_table;
380 u8 rc_via_alen;
381 u8 rc_via[MAX_VIA_ALEN];
382 u32 rc_label;
383 u8 rc_output_labels;
384 u32 rc_output_label[MAX_NEW_LABELS];
385 u32 rc_nlflags;
386 enum mpls_payload_type rc_payload_type;
387 struct nl_info rc_nlinfo;
388 struct rtnexthop *rc_mp;
389 int rc_mp_len;
390 };
391
mpls_rt_alloc(int num_nh,u8 max_alen)392 static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
393 {
394 u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
395 struct mpls_route *rt;
396
397 rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
398 VIA_ALEN_ALIGN) +
399 num_nh * max_alen_aligned,
400 GFP_KERNEL);
401 if (rt) {
402 rt->rt_nhn = num_nh;
403 rt->rt_nhn_alive = num_nh;
404 rt->rt_max_alen = max_alen_aligned;
405 }
406
407 return rt;
408 }
409
mpls_rt_free(struct mpls_route * rt)410 static void mpls_rt_free(struct mpls_route *rt)
411 {
412 if (rt)
413 kfree_rcu(rt, rt_rcu);
414 }
415
mpls_notify_route(struct net * net,unsigned index,struct mpls_route * old,struct mpls_route * new,const struct nl_info * info)416 static void mpls_notify_route(struct net *net, unsigned index,
417 struct mpls_route *old, struct mpls_route *new,
418 const struct nl_info *info)
419 {
420 struct nlmsghdr *nlh = info ? info->nlh : NULL;
421 unsigned portid = info ? info->portid : 0;
422 int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
423 struct mpls_route *rt = new ? new : old;
424 unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
425 /* Ignore reserved labels for now */
426 if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
427 rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
428 }
429
mpls_route_update(struct net * net,unsigned index,struct mpls_route * new,const struct nl_info * info)430 static void mpls_route_update(struct net *net, unsigned index,
431 struct mpls_route *new,
432 const struct nl_info *info)
433 {
434 struct mpls_route __rcu **platform_label;
435 struct mpls_route *rt;
436
437 ASSERT_RTNL();
438
439 platform_label = rtnl_dereference(net->mpls.platform_label);
440 rt = rtnl_dereference(platform_label[index]);
441 rcu_assign_pointer(platform_label[index], new);
442
443 mpls_notify_route(net, index, rt, new, info);
444
445 /* If we removed a route free it now */
446 mpls_rt_free(rt);
447 }
448
find_free_label(struct net * net)449 static unsigned find_free_label(struct net *net)
450 {
451 struct mpls_route __rcu **platform_label;
452 size_t platform_labels;
453 unsigned index;
454
455 platform_label = rtnl_dereference(net->mpls.platform_label);
456 platform_labels = net->mpls.platform_labels;
457 for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
458 index++) {
459 if (!rtnl_dereference(platform_label[index]))
460 return index;
461 }
462 return LABEL_NOT_SPECIFIED;
463 }
464
465 #if IS_ENABLED(CONFIG_INET)
inet_fib_lookup_dev(struct net * net,const void * addr)466 static struct net_device *inet_fib_lookup_dev(struct net *net,
467 const void *addr)
468 {
469 struct net_device *dev;
470 struct rtable *rt;
471 struct in_addr daddr;
472
473 memcpy(&daddr, addr, sizeof(struct in_addr));
474 rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
475 if (IS_ERR(rt))
476 return ERR_CAST(rt);
477
478 dev = rt->dst.dev;
479 dev_hold(dev);
480
481 ip_rt_put(rt);
482
483 return dev;
484 }
485 #else
inet_fib_lookup_dev(struct net * net,const void * addr)486 static struct net_device *inet_fib_lookup_dev(struct net *net,
487 const void *addr)
488 {
489 return ERR_PTR(-EAFNOSUPPORT);
490 }
491 #endif
492
493 #if IS_ENABLED(CONFIG_IPV6)
inet6_fib_lookup_dev(struct net * net,const void * addr)494 static struct net_device *inet6_fib_lookup_dev(struct net *net,
495 const void *addr)
496 {
497 struct net_device *dev;
498 struct dst_entry *dst;
499 struct flowi6 fl6;
500 int err;
501
502 if (!ipv6_stub)
503 return ERR_PTR(-EAFNOSUPPORT);
504
505 memset(&fl6, 0, sizeof(fl6));
506 memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
507 err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
508 if (err)
509 return ERR_PTR(err);
510
511 dev = dst->dev;
512 dev_hold(dev);
513 dst_release(dst);
514
515 return dev;
516 }
517 #else
inet6_fib_lookup_dev(struct net * net,const void * addr)518 static struct net_device *inet6_fib_lookup_dev(struct net *net,
519 const void *addr)
520 {
521 return ERR_PTR(-EAFNOSUPPORT);
522 }
523 #endif
524
find_outdev(struct net * net,struct mpls_route * rt,struct mpls_nh * nh,int oif)525 static struct net_device *find_outdev(struct net *net,
526 struct mpls_route *rt,
527 struct mpls_nh *nh, int oif)
528 {
529 struct net_device *dev = NULL;
530
531 if (!oif) {
532 switch (nh->nh_via_table) {
533 case NEIGH_ARP_TABLE:
534 dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
535 break;
536 case NEIGH_ND_TABLE:
537 dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
538 break;
539 case NEIGH_LINK_TABLE:
540 break;
541 }
542 } else {
543 dev = dev_get_by_index(net, oif);
544 }
545
546 if (!dev)
547 return ERR_PTR(-ENODEV);
548
549 if (IS_ERR(dev))
550 return dev;
551
552 /* The caller is holding rtnl anyways, so release the dev reference */
553 dev_put(dev);
554
555 return dev;
556 }
557
mpls_nh_assign_dev(struct net * net,struct mpls_route * rt,struct mpls_nh * nh,int oif)558 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
559 struct mpls_nh *nh, int oif)
560 {
561 struct net_device *dev = NULL;
562 int err = -ENODEV;
563
564 dev = find_outdev(net, rt, nh, oif);
565 if (IS_ERR(dev)) {
566 err = PTR_ERR(dev);
567 dev = NULL;
568 goto errout;
569 }
570
571 /* Ensure this is a supported device */
572 err = -EINVAL;
573 if (!mpls_dev_get(dev))
574 goto errout;
575
576 if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
577 (dev->addr_len != nh->nh_via_alen))
578 goto errout;
579
580 RCU_INIT_POINTER(nh->nh_dev, dev);
581
582 if (!(dev->flags & IFF_UP)) {
583 nh->nh_flags |= RTNH_F_DEAD;
584 } else {
585 unsigned int flags;
586
587 flags = dev_get_flags(dev);
588 if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
589 nh->nh_flags |= RTNH_F_LINKDOWN;
590 }
591
592 return 0;
593
594 errout:
595 return err;
596 }
597
mpls_nh_build_from_cfg(struct mpls_route_config * cfg,struct mpls_route * rt)598 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
599 struct mpls_route *rt)
600 {
601 struct net *net = cfg->rc_nlinfo.nl_net;
602 struct mpls_nh *nh = rt->rt_nh;
603 int err;
604 int i;
605
606 if (!nh)
607 return -ENOMEM;
608
609 err = -EINVAL;
610 /* Ensure only a supported number of labels are present */
611 if (cfg->rc_output_labels > MAX_NEW_LABELS)
612 goto errout;
613
614 nh->nh_labels = cfg->rc_output_labels;
615 for (i = 0; i < nh->nh_labels; i++)
616 nh->nh_label[i] = cfg->rc_output_label[i];
617
618 nh->nh_via_table = cfg->rc_via_table;
619 memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
620 nh->nh_via_alen = cfg->rc_via_alen;
621
622 err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
623 if (err)
624 goto errout;
625
626 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
627 rt->rt_nhn_alive--;
628
629 return 0;
630
631 errout:
632 return err;
633 }
634
mpls_nh_build(struct net * net,struct mpls_route * rt,struct mpls_nh * nh,int oif,struct nlattr * via,struct nlattr * newdst)635 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
636 struct mpls_nh *nh, int oif, struct nlattr *via,
637 struct nlattr *newdst)
638 {
639 int err = -ENOMEM;
640
641 if (!nh)
642 goto errout;
643
644 if (newdst) {
645 err = nla_get_labels(newdst, MAX_NEW_LABELS,
646 &nh->nh_labels, nh->nh_label);
647 if (err)
648 goto errout;
649 }
650
651 if (via) {
652 err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
653 __mpls_nh_via(rt, nh));
654 if (err)
655 goto errout;
656 } else {
657 nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
658 }
659
660 err = mpls_nh_assign_dev(net, rt, nh, oif);
661 if (err)
662 goto errout;
663
664 return 0;
665
666 errout:
667 return err;
668 }
669
mpls_count_nexthops(struct rtnexthop * rtnh,int len,u8 cfg_via_alen,u8 * max_via_alen)670 static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
671 u8 cfg_via_alen, u8 *max_via_alen)
672 {
673 int nhs = 0;
674 int remaining = len;
675
676 if (!rtnh) {
677 *max_via_alen = cfg_via_alen;
678 return 1;
679 }
680
681 *max_via_alen = 0;
682
683 while (rtnh_ok(rtnh, remaining)) {
684 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
685 int attrlen;
686
687 attrlen = rtnh_attrlen(rtnh);
688 nla = nla_find(attrs, attrlen, RTA_VIA);
689 if (nla && nla_len(nla) >=
690 offsetof(struct rtvia, rtvia_addr)) {
691 int via_alen = nla_len(nla) -
692 offsetof(struct rtvia, rtvia_addr);
693
694 if (via_alen <= MAX_VIA_ALEN)
695 *max_via_alen = max_t(u16, *max_via_alen,
696 via_alen);
697 }
698
699 nhs++;
700 rtnh = rtnh_next(rtnh, &remaining);
701 }
702
703 /* leftover implies invalid nexthop configuration, discard it */
704 return remaining > 0 ? 0 : nhs;
705 }
706
mpls_nh_build_multi(struct mpls_route_config * cfg,struct mpls_route * rt)707 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
708 struct mpls_route *rt)
709 {
710 struct rtnexthop *rtnh = cfg->rc_mp;
711 struct nlattr *nla_via, *nla_newdst;
712 int remaining = cfg->rc_mp_len;
713 int nhs = 0;
714 int err = 0;
715
716 change_nexthops(rt) {
717 int attrlen;
718
719 nla_via = NULL;
720 nla_newdst = NULL;
721
722 err = -EINVAL;
723 if (!rtnh_ok(rtnh, remaining))
724 goto errout;
725
726 /* neither weighted multipath nor any flags
727 * are supported
728 */
729 if (rtnh->rtnh_hops || rtnh->rtnh_flags)
730 goto errout;
731
732 attrlen = rtnh_attrlen(rtnh);
733 if (attrlen > 0) {
734 struct nlattr *attrs = rtnh_attrs(rtnh);
735
736 nla_via = nla_find(attrs, attrlen, RTA_VIA);
737 nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
738 }
739
740 err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
741 rtnh->rtnh_ifindex, nla_via, nla_newdst);
742 if (err)
743 goto errout;
744
745 if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
746 rt->rt_nhn_alive--;
747
748 rtnh = rtnh_next(rtnh, &remaining);
749 nhs++;
750 } endfor_nexthops(rt);
751
752 rt->rt_nhn = nhs;
753
754 return 0;
755
756 errout:
757 return err;
758 }
759
mpls_label_ok(struct net * net,unsigned int * index)760 static bool mpls_label_ok(struct net *net, unsigned int *index)
761 {
762 bool is_ok = true;
763
764 /* Reserved labels may not be set */
765 if (*index < MPLS_LABEL_FIRST_UNRESERVED)
766 is_ok = false;
767
768 /* The full 20 bit range may not be supported. */
769 if (is_ok && *index >= net->mpls.platform_labels)
770 is_ok = false;
771
772 *index = array_index_nospec(*index, net->mpls.platform_labels);
773 return is_ok;
774 }
775
mpls_route_add(struct mpls_route_config * cfg)776 static int mpls_route_add(struct mpls_route_config *cfg)
777 {
778 struct mpls_route __rcu **platform_label;
779 struct net *net = cfg->rc_nlinfo.nl_net;
780 struct mpls_route *rt, *old;
781 int err = -EINVAL;
782 u8 max_via_alen;
783 unsigned index;
784 int nhs;
785
786 index = cfg->rc_label;
787
788 /* If a label was not specified during insert pick one */
789 if ((index == LABEL_NOT_SPECIFIED) &&
790 (cfg->rc_nlflags & NLM_F_CREATE)) {
791 index = find_free_label(net);
792 }
793
794 if (!mpls_label_ok(net, &index))
795 goto errout;
796
797 /* Append makes no sense with mpls */
798 err = -EOPNOTSUPP;
799 if (cfg->rc_nlflags & NLM_F_APPEND)
800 goto errout;
801
802 err = -EEXIST;
803 platform_label = rtnl_dereference(net->mpls.platform_label);
804 old = rtnl_dereference(platform_label[index]);
805 if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
806 goto errout;
807
808 err = -EEXIST;
809 if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
810 goto errout;
811
812 err = -ENOENT;
813 if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
814 goto errout;
815
816 err = -EINVAL;
817 nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
818 cfg->rc_via_alen, &max_via_alen);
819 if (nhs == 0)
820 goto errout;
821
822 err = -ENOMEM;
823 rt = mpls_rt_alloc(nhs, max_via_alen);
824 if (!rt)
825 goto errout;
826
827 rt->rt_protocol = cfg->rc_protocol;
828 rt->rt_payload_type = cfg->rc_payload_type;
829
830 if (cfg->rc_mp)
831 err = mpls_nh_build_multi(cfg, rt);
832 else
833 err = mpls_nh_build_from_cfg(cfg, rt);
834 if (err)
835 goto freert;
836
837 mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
838
839 return 0;
840
841 freert:
842 mpls_rt_free(rt);
843 errout:
844 return err;
845 }
846
mpls_route_del(struct mpls_route_config * cfg)847 static int mpls_route_del(struct mpls_route_config *cfg)
848 {
849 struct net *net = cfg->rc_nlinfo.nl_net;
850 unsigned index;
851 int err = -EINVAL;
852
853 index = cfg->rc_label;
854
855 if (!mpls_label_ok(net, &index))
856 goto errout;
857
858 mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
859
860 err = 0;
861 errout:
862 return err;
863 }
864
865 #define MPLS_PERDEV_SYSCTL_OFFSET(field) \
866 (&((struct mpls_dev *)0)->field)
867
868 static const struct ctl_table mpls_dev_table[] = {
869 {
870 .procname = "input",
871 .maxlen = sizeof(int),
872 .mode = 0644,
873 .proc_handler = proc_dointvec,
874 .data = MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
875 },
876 { }
877 };
878
mpls_dev_sysctl_register(struct net_device * dev,struct mpls_dev * mdev)879 static int mpls_dev_sysctl_register(struct net_device *dev,
880 struct mpls_dev *mdev)
881 {
882 char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
883 struct ctl_table *table;
884 int i;
885
886 table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
887 if (!table)
888 goto out;
889
890 /* Table data contains only offsets relative to the base of
891 * the mdev at this point, so make them absolute.
892 */
893 for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
894 table[i].data = (char *)mdev + (uintptr_t)table[i].data;
895
896 snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
897
898 mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
899 if (!mdev->sysctl)
900 goto free;
901
902 return 0;
903
904 free:
905 kfree(table);
906 out:
907 return -ENOBUFS;
908 }
909
mpls_dev_sysctl_unregister(struct mpls_dev * mdev)910 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
911 {
912 struct ctl_table *table;
913
914 table = mdev->sysctl->ctl_table_arg;
915 unregister_net_sysctl_table(mdev->sysctl);
916 kfree(table);
917 }
918
mpls_add_dev(struct net_device * dev)919 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
920 {
921 struct mpls_dev *mdev;
922 int err = -ENOMEM;
923
924 ASSERT_RTNL();
925
926 mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
927 if (!mdev)
928 return ERR_PTR(err);
929
930 err = mpls_dev_sysctl_register(dev, mdev);
931 if (err)
932 goto free;
933
934 rcu_assign_pointer(dev->mpls_ptr, mdev);
935
936 return mdev;
937
938 free:
939 kfree(mdev);
940 return ERR_PTR(err);
941 }
942
mpls_ifdown(struct net_device * dev,int event)943 static void mpls_ifdown(struct net_device *dev, int event)
944 {
945 struct mpls_route __rcu **platform_label;
946 struct net *net = dev_net(dev);
947 unsigned int nh_flags = RTNH_F_DEAD | RTNH_F_LINKDOWN;
948 unsigned int alive;
949 unsigned index;
950
951 platform_label = rtnl_dereference(net->mpls.platform_label);
952 for (index = 0; index < net->mpls.platform_labels; index++) {
953 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
954
955 if (!rt)
956 continue;
957
958 alive = 0;
959 change_nexthops(rt) {
960 if (rtnl_dereference(nh->nh_dev) != dev)
961 goto next;
962
963 switch (event) {
964 case NETDEV_DOWN:
965 case NETDEV_UNREGISTER:
966 nh->nh_flags |= RTNH_F_DEAD;
967 /* fall through */
968 case NETDEV_CHANGE:
969 nh->nh_flags |= RTNH_F_LINKDOWN;
970 break;
971 }
972 if (event == NETDEV_UNREGISTER)
973 RCU_INIT_POINTER(nh->nh_dev, NULL);
974 next:
975 if (!(nh->nh_flags & nh_flags))
976 alive++;
977 } endfor_nexthops(rt);
978
979 WRITE_ONCE(rt->rt_nhn_alive, alive);
980 }
981 }
982
mpls_ifup(struct net_device * dev,unsigned int nh_flags)983 static void mpls_ifup(struct net_device *dev, unsigned int nh_flags)
984 {
985 struct mpls_route __rcu **platform_label;
986 struct net *net = dev_net(dev);
987 unsigned index;
988 int alive;
989
990 platform_label = rtnl_dereference(net->mpls.platform_label);
991 for (index = 0; index < net->mpls.platform_labels; index++) {
992 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
993
994 if (!rt)
995 continue;
996
997 alive = 0;
998 change_nexthops(rt) {
999 struct net_device *nh_dev =
1000 rtnl_dereference(nh->nh_dev);
1001
1002 if (!(nh->nh_flags & nh_flags)) {
1003 alive++;
1004 continue;
1005 }
1006 if (nh_dev != dev)
1007 continue;
1008 alive++;
1009 nh->nh_flags &= ~nh_flags;
1010 } endfor_nexthops(rt);
1011
1012 ACCESS_ONCE(rt->rt_nhn_alive) = alive;
1013 }
1014 }
1015
mpls_dev_notify(struct notifier_block * this,unsigned long event,void * ptr)1016 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1017 void *ptr)
1018 {
1019 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1020 struct mpls_dev *mdev;
1021 unsigned int flags;
1022
1023 if (event == NETDEV_REGISTER) {
1024 /* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1025 if (dev->type == ARPHRD_ETHER ||
1026 dev->type == ARPHRD_LOOPBACK ||
1027 dev->type == ARPHRD_IPGRE ||
1028 dev->type == ARPHRD_SIT ||
1029 dev->type == ARPHRD_TUNNEL) {
1030 mdev = mpls_add_dev(dev);
1031 if (IS_ERR(mdev))
1032 return notifier_from_errno(PTR_ERR(mdev));
1033 }
1034 return NOTIFY_OK;
1035 }
1036
1037 mdev = mpls_dev_get(dev);
1038 if (!mdev)
1039 return NOTIFY_OK;
1040
1041 switch (event) {
1042 case NETDEV_DOWN:
1043 mpls_ifdown(dev, event);
1044 break;
1045 case NETDEV_UP:
1046 flags = dev_get_flags(dev);
1047 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1048 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1049 else
1050 mpls_ifup(dev, RTNH_F_DEAD);
1051 break;
1052 case NETDEV_CHANGE:
1053 flags = dev_get_flags(dev);
1054 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1055 mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1056 else
1057 mpls_ifdown(dev, event);
1058 break;
1059 case NETDEV_UNREGISTER:
1060 mpls_ifdown(dev, event);
1061 mdev = mpls_dev_get(dev);
1062 if (mdev) {
1063 mpls_dev_sysctl_unregister(mdev);
1064 RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1065 kfree_rcu(mdev, rcu);
1066 }
1067 break;
1068 case NETDEV_CHANGENAME:
1069 mdev = mpls_dev_get(dev);
1070 if (mdev) {
1071 int err;
1072
1073 mpls_dev_sysctl_unregister(mdev);
1074 err = mpls_dev_sysctl_register(dev, mdev);
1075 if (err)
1076 return notifier_from_errno(err);
1077 }
1078 break;
1079 }
1080 return NOTIFY_OK;
1081 }
1082
1083 static struct notifier_block mpls_dev_notifier = {
1084 .notifier_call = mpls_dev_notify,
1085 };
1086
nla_put_via(struct sk_buff * skb,u8 table,const void * addr,int alen)1087 static int nla_put_via(struct sk_buff *skb,
1088 u8 table, const void *addr, int alen)
1089 {
1090 static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1091 AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1092 };
1093 struct nlattr *nla;
1094 struct rtvia *via;
1095 int family = AF_UNSPEC;
1096
1097 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1098 if (!nla)
1099 return -EMSGSIZE;
1100
1101 if (table <= NEIGH_NR_TABLES)
1102 family = table_to_family[table];
1103
1104 via = nla_data(nla);
1105 via->rtvia_family = family;
1106 memcpy(via->rtvia_addr, addr, alen);
1107 return 0;
1108 }
1109
nla_put_labels(struct sk_buff * skb,int attrtype,u8 labels,const u32 label[])1110 int nla_put_labels(struct sk_buff *skb, int attrtype,
1111 u8 labels, const u32 label[])
1112 {
1113 struct nlattr *nla;
1114 struct mpls_shim_hdr *nla_label;
1115 bool bos;
1116 int i;
1117 nla = nla_reserve(skb, attrtype, labels*4);
1118 if (!nla)
1119 return -EMSGSIZE;
1120
1121 nla_label = nla_data(nla);
1122 bos = true;
1123 for (i = labels - 1; i >= 0; i--) {
1124 nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1125 bos = false;
1126 }
1127
1128 return 0;
1129 }
1130 EXPORT_SYMBOL_GPL(nla_put_labels);
1131
nla_get_labels(const struct nlattr * nla,u32 max_labels,u8 * labels,u32 label[])1132 int nla_get_labels(const struct nlattr *nla,
1133 u32 max_labels, u8 *labels, u32 label[])
1134 {
1135 unsigned len = nla_len(nla);
1136 unsigned nla_labels;
1137 struct mpls_shim_hdr *nla_label;
1138 bool bos;
1139 int i;
1140
1141 /* len needs to be an even multiple of 4 (the label size) */
1142 if (len & 3)
1143 return -EINVAL;
1144
1145 /* Limit the number of new labels allowed */
1146 nla_labels = len/4;
1147 if (nla_labels > max_labels)
1148 return -EINVAL;
1149
1150 nla_label = nla_data(nla);
1151 bos = true;
1152 for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1153 struct mpls_entry_decoded dec;
1154 dec = mpls_entry_decode(nla_label + i);
1155
1156 /* Ensure the bottom of stack flag is properly set
1157 * and ttl and tc are both clear.
1158 */
1159 if ((dec.bos != bos) || dec.ttl || dec.tc)
1160 return -EINVAL;
1161
1162 switch (dec.label) {
1163 case MPLS_LABEL_IMPLNULL:
1164 /* RFC3032: This is a label that an LSR may
1165 * assign and distribute, but which never
1166 * actually appears in the encapsulation.
1167 */
1168 return -EINVAL;
1169 }
1170
1171 label[i] = dec.label;
1172 }
1173 *labels = nla_labels;
1174 return 0;
1175 }
1176 EXPORT_SYMBOL_GPL(nla_get_labels);
1177
nla_get_via(const struct nlattr * nla,u8 * via_alen,u8 * via_table,u8 via_addr[])1178 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1179 u8 *via_table, u8 via_addr[])
1180 {
1181 struct rtvia *via = nla_data(nla);
1182 int err = -EINVAL;
1183 int alen;
1184
1185 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1186 goto errout;
1187 alen = nla_len(nla) -
1188 offsetof(struct rtvia, rtvia_addr);
1189 if (alen > MAX_VIA_ALEN)
1190 goto errout;
1191
1192 /* Validate the address family */
1193 switch (via->rtvia_family) {
1194 case AF_PACKET:
1195 *via_table = NEIGH_LINK_TABLE;
1196 break;
1197 case AF_INET:
1198 *via_table = NEIGH_ARP_TABLE;
1199 if (alen != 4)
1200 goto errout;
1201 break;
1202 case AF_INET6:
1203 *via_table = NEIGH_ND_TABLE;
1204 if (alen != 16)
1205 goto errout;
1206 break;
1207 default:
1208 /* Unsupported address family */
1209 goto errout;
1210 }
1211
1212 memcpy(via_addr, via->rtvia_addr, alen);
1213 *via_alen = alen;
1214 err = 0;
1215
1216 errout:
1217 return err;
1218 }
1219
rtm_to_route_config(struct sk_buff * skb,struct nlmsghdr * nlh,struct mpls_route_config * cfg)1220 static int rtm_to_route_config(struct sk_buff *skb, struct nlmsghdr *nlh,
1221 struct mpls_route_config *cfg)
1222 {
1223 struct rtmsg *rtm;
1224 struct nlattr *tb[RTA_MAX+1];
1225 int index;
1226 int err;
1227
1228 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1229 if (err < 0)
1230 goto errout;
1231
1232 err = -EINVAL;
1233 rtm = nlmsg_data(nlh);
1234 memset(cfg, 0, sizeof(*cfg));
1235
1236 if (rtm->rtm_family != AF_MPLS)
1237 goto errout;
1238 if (rtm->rtm_dst_len != 20)
1239 goto errout;
1240 if (rtm->rtm_src_len != 0)
1241 goto errout;
1242 if (rtm->rtm_tos != 0)
1243 goto errout;
1244 if (rtm->rtm_table != RT_TABLE_MAIN)
1245 goto errout;
1246 /* Any value is acceptable for rtm_protocol */
1247
1248 /* As mpls uses destination specific addresses
1249 * (or source specific address in the case of multicast)
1250 * all addresses have universal scope.
1251 */
1252 if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1253 goto errout;
1254 if (rtm->rtm_type != RTN_UNICAST)
1255 goto errout;
1256 if (rtm->rtm_flags != 0)
1257 goto errout;
1258
1259 cfg->rc_label = LABEL_NOT_SPECIFIED;
1260 cfg->rc_protocol = rtm->rtm_protocol;
1261 cfg->rc_via_table = MPLS_NEIGH_TABLE_UNSPEC;
1262 cfg->rc_nlflags = nlh->nlmsg_flags;
1263 cfg->rc_nlinfo.portid = NETLINK_CB(skb).portid;
1264 cfg->rc_nlinfo.nlh = nlh;
1265 cfg->rc_nlinfo.nl_net = sock_net(skb->sk);
1266
1267 for (index = 0; index <= RTA_MAX; index++) {
1268 struct nlattr *nla = tb[index];
1269 if (!nla)
1270 continue;
1271
1272 switch (index) {
1273 case RTA_OIF:
1274 cfg->rc_ifindex = nla_get_u32(nla);
1275 break;
1276 case RTA_NEWDST:
1277 if (nla_get_labels(nla, MAX_NEW_LABELS,
1278 &cfg->rc_output_labels,
1279 cfg->rc_output_label))
1280 goto errout;
1281 break;
1282 case RTA_DST:
1283 {
1284 u8 label_count;
1285 if (nla_get_labels(nla, 1, &label_count,
1286 &cfg->rc_label))
1287 goto errout;
1288
1289 if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1290 &cfg->rc_label))
1291 goto errout;
1292 break;
1293 }
1294 case RTA_VIA:
1295 {
1296 if (nla_get_via(nla, &cfg->rc_via_alen,
1297 &cfg->rc_via_table, cfg->rc_via))
1298 goto errout;
1299 break;
1300 }
1301 case RTA_MULTIPATH:
1302 {
1303 cfg->rc_mp = nla_data(nla);
1304 cfg->rc_mp_len = nla_len(nla);
1305 break;
1306 }
1307 default:
1308 /* Unsupported attribute */
1309 goto errout;
1310 }
1311 }
1312
1313 err = 0;
1314 errout:
1315 return err;
1316 }
1317
mpls_rtm_delroute(struct sk_buff * skb,struct nlmsghdr * nlh)1318 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1319 {
1320 struct mpls_route_config cfg;
1321 int err;
1322
1323 err = rtm_to_route_config(skb, nlh, &cfg);
1324 if (err < 0)
1325 return err;
1326
1327 return mpls_route_del(&cfg);
1328 }
1329
1330
mpls_rtm_newroute(struct sk_buff * skb,struct nlmsghdr * nlh)1331 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1332 {
1333 struct mpls_route_config cfg;
1334 int err;
1335
1336 err = rtm_to_route_config(skb, nlh, &cfg);
1337 if (err < 0)
1338 return err;
1339
1340 return mpls_route_add(&cfg);
1341 }
1342
mpls_dump_route(struct sk_buff * skb,u32 portid,u32 seq,int event,u32 label,struct mpls_route * rt,int flags)1343 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1344 u32 label, struct mpls_route *rt, int flags)
1345 {
1346 struct net_device *dev;
1347 struct nlmsghdr *nlh;
1348 struct rtmsg *rtm;
1349
1350 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1351 if (nlh == NULL)
1352 return -EMSGSIZE;
1353
1354 rtm = nlmsg_data(nlh);
1355 rtm->rtm_family = AF_MPLS;
1356 rtm->rtm_dst_len = 20;
1357 rtm->rtm_src_len = 0;
1358 rtm->rtm_tos = 0;
1359 rtm->rtm_table = RT_TABLE_MAIN;
1360 rtm->rtm_protocol = rt->rt_protocol;
1361 rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1362 rtm->rtm_type = RTN_UNICAST;
1363 rtm->rtm_flags = 0;
1364
1365 if (nla_put_labels(skb, RTA_DST, 1, &label))
1366 goto nla_put_failure;
1367 if (rt->rt_nhn == 1) {
1368 const struct mpls_nh *nh = rt->rt_nh;
1369
1370 if (nh->nh_labels &&
1371 nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1372 nh->nh_label))
1373 goto nla_put_failure;
1374 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1375 nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1376 nh->nh_via_alen))
1377 goto nla_put_failure;
1378 dev = rtnl_dereference(nh->nh_dev);
1379 if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1380 goto nla_put_failure;
1381 if (nh->nh_flags & RTNH_F_LINKDOWN)
1382 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1383 if (nh->nh_flags & RTNH_F_DEAD)
1384 rtm->rtm_flags |= RTNH_F_DEAD;
1385 } else {
1386 struct rtnexthop *rtnh;
1387 struct nlattr *mp;
1388 int dead = 0;
1389 int linkdown = 0;
1390
1391 mp = nla_nest_start(skb, RTA_MULTIPATH);
1392 if (!mp)
1393 goto nla_put_failure;
1394
1395 for_nexthops(rt) {
1396 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1397 if (!rtnh)
1398 goto nla_put_failure;
1399
1400 dev = rtnl_dereference(nh->nh_dev);
1401 if (dev)
1402 rtnh->rtnh_ifindex = dev->ifindex;
1403 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1404 rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1405 linkdown++;
1406 }
1407 if (nh->nh_flags & RTNH_F_DEAD) {
1408 rtnh->rtnh_flags |= RTNH_F_DEAD;
1409 dead++;
1410 }
1411
1412 if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1413 nh->nh_labels,
1414 nh->nh_label))
1415 goto nla_put_failure;
1416 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1417 nla_put_via(skb, nh->nh_via_table,
1418 mpls_nh_via(rt, nh),
1419 nh->nh_via_alen))
1420 goto nla_put_failure;
1421
1422 /* length of rtnetlink header + attributes */
1423 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1424 } endfor_nexthops(rt);
1425
1426 if (linkdown == rt->rt_nhn)
1427 rtm->rtm_flags |= RTNH_F_LINKDOWN;
1428 if (dead == rt->rt_nhn)
1429 rtm->rtm_flags |= RTNH_F_DEAD;
1430
1431 nla_nest_end(skb, mp);
1432 }
1433
1434 nlmsg_end(skb, nlh);
1435 return 0;
1436
1437 nla_put_failure:
1438 nlmsg_cancel(skb, nlh);
1439 return -EMSGSIZE;
1440 }
1441
mpls_dump_routes(struct sk_buff * skb,struct netlink_callback * cb)1442 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1443 {
1444 struct net *net = sock_net(skb->sk);
1445 struct mpls_route __rcu **platform_label;
1446 size_t platform_labels;
1447 unsigned int index;
1448
1449 ASSERT_RTNL();
1450
1451 index = cb->args[0];
1452 if (index < MPLS_LABEL_FIRST_UNRESERVED)
1453 index = MPLS_LABEL_FIRST_UNRESERVED;
1454
1455 platform_label = rtnl_dereference(net->mpls.platform_label);
1456 platform_labels = net->mpls.platform_labels;
1457 for (; index < platform_labels; index++) {
1458 struct mpls_route *rt;
1459 rt = rtnl_dereference(platform_label[index]);
1460 if (!rt)
1461 continue;
1462
1463 if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1464 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1465 index, rt, NLM_F_MULTI) < 0)
1466 break;
1467 }
1468 cb->args[0] = index;
1469
1470 return skb->len;
1471 }
1472
lfib_nlmsg_size(struct mpls_route * rt)1473 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1474 {
1475 size_t payload =
1476 NLMSG_ALIGN(sizeof(struct rtmsg))
1477 + nla_total_size(4); /* RTA_DST */
1478
1479 if (rt->rt_nhn == 1) {
1480 struct mpls_nh *nh = rt->rt_nh;
1481
1482 if (nh->nh_dev)
1483 payload += nla_total_size(4); /* RTA_OIF */
1484 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1485 payload += nla_total_size(2 + nh->nh_via_alen);
1486 if (nh->nh_labels) /* RTA_NEWDST */
1487 payload += nla_total_size(nh->nh_labels * 4);
1488 } else {
1489 /* each nexthop is packed in an attribute */
1490 size_t nhsize = 0;
1491
1492 for_nexthops(rt) {
1493 nhsize += nla_total_size(sizeof(struct rtnexthop));
1494 /* RTA_VIA */
1495 if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1496 nhsize += nla_total_size(2 + nh->nh_via_alen);
1497 if (nh->nh_labels)
1498 nhsize += nla_total_size(nh->nh_labels * 4);
1499 } endfor_nexthops(rt);
1500 /* nested attribute */
1501 payload += nla_total_size(nhsize);
1502 }
1503
1504 return payload;
1505 }
1506
rtmsg_lfib(int event,u32 label,struct mpls_route * rt,struct nlmsghdr * nlh,struct net * net,u32 portid,unsigned int nlm_flags)1507 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1508 struct nlmsghdr *nlh, struct net *net, u32 portid,
1509 unsigned int nlm_flags)
1510 {
1511 struct sk_buff *skb;
1512 u32 seq = nlh ? nlh->nlmsg_seq : 0;
1513 int err = -ENOBUFS;
1514
1515 skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1516 if (skb == NULL)
1517 goto errout;
1518
1519 err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1520 if (err < 0) {
1521 /* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1522 WARN_ON(err == -EMSGSIZE);
1523 kfree_skb(skb);
1524 goto errout;
1525 }
1526 rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1527
1528 return;
1529 errout:
1530 if (err < 0)
1531 rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1532 }
1533
resize_platform_label_table(struct net * net,size_t limit)1534 static int resize_platform_label_table(struct net *net, size_t limit)
1535 {
1536 size_t size = sizeof(struct mpls_route *) * limit;
1537 size_t old_limit;
1538 size_t cp_size;
1539 struct mpls_route __rcu **labels = NULL, **old;
1540 struct mpls_route *rt0 = NULL, *rt2 = NULL;
1541 unsigned index;
1542
1543 if (size) {
1544 labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1545 if (!labels)
1546 labels = vzalloc(size);
1547
1548 if (!labels)
1549 goto nolabels;
1550 }
1551
1552 /* In case the predefined labels need to be populated */
1553 if (limit > MPLS_LABEL_IPV4NULL) {
1554 struct net_device *lo = net->loopback_dev;
1555 rt0 = mpls_rt_alloc(1, lo->addr_len);
1556 if (!rt0)
1557 goto nort0;
1558 RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1559 rt0->rt_protocol = RTPROT_KERNEL;
1560 rt0->rt_payload_type = MPT_IPV4;
1561 rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1562 rt0->rt_nh->nh_via_alen = lo->addr_len;
1563 memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1564 lo->addr_len);
1565 }
1566 if (limit > MPLS_LABEL_IPV6NULL) {
1567 struct net_device *lo = net->loopback_dev;
1568 rt2 = mpls_rt_alloc(1, lo->addr_len);
1569 if (!rt2)
1570 goto nort2;
1571 RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1572 rt2->rt_protocol = RTPROT_KERNEL;
1573 rt2->rt_payload_type = MPT_IPV6;
1574 rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1575 rt2->rt_nh->nh_via_alen = lo->addr_len;
1576 memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1577 lo->addr_len);
1578 }
1579
1580 rtnl_lock();
1581 /* Remember the original table */
1582 old = rtnl_dereference(net->mpls.platform_label);
1583 old_limit = net->mpls.platform_labels;
1584
1585 /* Free any labels beyond the new table */
1586 for (index = limit; index < old_limit; index++)
1587 mpls_route_update(net, index, NULL, NULL);
1588
1589 /* Copy over the old labels */
1590 cp_size = size;
1591 if (old_limit < limit)
1592 cp_size = old_limit * sizeof(struct mpls_route *);
1593
1594 memcpy(labels, old, cp_size);
1595
1596 /* If needed set the predefined labels */
1597 if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1598 (limit > MPLS_LABEL_IPV6NULL)) {
1599 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1600 rt2 = NULL;
1601 }
1602
1603 if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1604 (limit > MPLS_LABEL_IPV4NULL)) {
1605 RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1606 rt0 = NULL;
1607 }
1608
1609 /* Update the global pointers */
1610 net->mpls.platform_labels = limit;
1611 rcu_assign_pointer(net->mpls.platform_label, labels);
1612
1613 rtnl_unlock();
1614
1615 mpls_rt_free(rt2);
1616 mpls_rt_free(rt0);
1617
1618 if (old) {
1619 synchronize_rcu();
1620 kvfree(old);
1621 }
1622 return 0;
1623
1624 nort2:
1625 mpls_rt_free(rt0);
1626 nort0:
1627 kvfree(labels);
1628 nolabels:
1629 return -ENOMEM;
1630 }
1631
mpls_platform_labels(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)1632 static int mpls_platform_labels(struct ctl_table *table, int write,
1633 void __user *buffer, size_t *lenp, loff_t *ppos)
1634 {
1635 struct net *net = table->data;
1636 int platform_labels = net->mpls.platform_labels;
1637 int ret;
1638 struct ctl_table tmp = {
1639 .procname = table->procname,
1640 .data = &platform_labels,
1641 .maxlen = sizeof(int),
1642 .mode = table->mode,
1643 .extra1 = &zero,
1644 .extra2 = &label_limit,
1645 };
1646
1647 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1648
1649 if (write && ret == 0)
1650 ret = resize_platform_label_table(net, platform_labels);
1651
1652 return ret;
1653 }
1654
1655 static const struct ctl_table mpls_table[] = {
1656 {
1657 .procname = "platform_labels",
1658 .data = NULL,
1659 .maxlen = sizeof(int),
1660 .mode = 0644,
1661 .proc_handler = mpls_platform_labels,
1662 },
1663 { }
1664 };
1665
mpls_net_init(struct net * net)1666 static int mpls_net_init(struct net *net)
1667 {
1668 struct ctl_table *table;
1669
1670 net->mpls.platform_labels = 0;
1671 net->mpls.platform_label = NULL;
1672
1673 table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1674 if (table == NULL)
1675 return -ENOMEM;
1676
1677 table[0].data = net;
1678 net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1679 if (net->mpls.ctl == NULL) {
1680 kfree(table);
1681 return -ENOMEM;
1682 }
1683
1684 return 0;
1685 }
1686
mpls_net_exit(struct net * net)1687 static void mpls_net_exit(struct net *net)
1688 {
1689 struct mpls_route __rcu **platform_label;
1690 size_t platform_labels;
1691 struct ctl_table *table;
1692 unsigned int index;
1693
1694 table = net->mpls.ctl->ctl_table_arg;
1695 unregister_net_sysctl_table(net->mpls.ctl);
1696 kfree(table);
1697
1698 /* An rcu grace period has passed since there was a device in
1699 * the network namespace (and thus the last in flight packet)
1700 * left this network namespace. This is because
1701 * unregister_netdevice_many and netdev_run_todo has completed
1702 * for each network device that was in this network namespace.
1703 *
1704 * As such no additional rcu synchronization is necessary when
1705 * freeing the platform_label table.
1706 */
1707 rtnl_lock();
1708 platform_label = rtnl_dereference(net->mpls.platform_label);
1709 platform_labels = net->mpls.platform_labels;
1710 for (index = 0; index < platform_labels; index++) {
1711 struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1712 RCU_INIT_POINTER(platform_label[index], NULL);
1713 mpls_notify_route(net, index, rt, NULL, NULL);
1714 mpls_rt_free(rt);
1715 }
1716 rtnl_unlock();
1717
1718 kvfree(platform_label);
1719 }
1720
1721 static struct pernet_operations mpls_net_ops = {
1722 .init = mpls_net_init,
1723 .exit = mpls_net_exit,
1724 };
1725
mpls_init(void)1726 static int __init mpls_init(void)
1727 {
1728 int err;
1729
1730 BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1731
1732 err = register_pernet_subsys(&mpls_net_ops);
1733 if (err)
1734 goto out;
1735
1736 err = register_netdevice_notifier(&mpls_dev_notifier);
1737 if (err)
1738 goto out_unregister_pernet;
1739
1740 dev_add_pack(&mpls_packet_type);
1741
1742 rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1743 rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1744 rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1745 err = 0;
1746 out:
1747 return err;
1748
1749 out_unregister_pernet:
1750 unregister_pernet_subsys(&mpls_net_ops);
1751 goto out;
1752 }
1753 module_init(mpls_init);
1754
mpls_exit(void)1755 static void __exit mpls_exit(void)
1756 {
1757 rtnl_unregister_all(PF_MPLS);
1758 dev_remove_pack(&mpls_packet_type);
1759 unregister_netdevice_notifier(&mpls_dev_notifier);
1760 unregister_pernet_subsys(&mpls_net_ops);
1761 }
1762 module_exit(mpls_exit);
1763
1764 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1765 MODULE_LICENSE("GPL v2");
1766 MODULE_ALIAS_NETPROTO(PF_MPLS);
1767