xref: /openbmc/linux/net/mpls/af_mpls.c (revision 2572f00d)
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/vmalloc.h>
11 #include <net/ip.h>
12 #include <net/dst.h>
13 #include <net/sock.h>
14 #include <net/arp.h>
15 #include <net/ip_fib.h>
16 #include <net/netevent.h>
17 #include <net/netns/generic.h>
18 #if IS_ENABLED(CONFIG_IPV6)
19 #include <net/ipv6.h>
20 #include <net/addrconf.h>
21 #endif
22 #include <net/nexthop.h>
23 #include "internal.h"
24 
25 /* Maximum number of labels to look ahead at when selecting a path of
26  * a multipath route
27  */
28 #define MAX_MP_SELECT_LABELS 4
29 
30 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
31 
32 static int zero = 0;
33 static int label_limit = (1 << 20) - 1;
34 
35 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
36 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
37 		       unsigned int nlm_flags);
38 
39 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
40 {
41 	struct mpls_route *rt = NULL;
42 
43 	if (index < net->mpls.platform_labels) {
44 		struct mpls_route __rcu **platform_label =
45 			rcu_dereference(net->mpls.platform_label);
46 		rt = rcu_dereference(platform_label[index]);
47 	}
48 	return rt;
49 }
50 
51 static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
52 {
53 	return rcu_dereference_rtnl(dev->mpls_ptr);
54 }
55 
56 bool mpls_output_possible(const struct net_device *dev)
57 {
58 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
59 }
60 EXPORT_SYMBOL_GPL(mpls_output_possible);
61 
62 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
63 {
64 	u8 *nh0_via = PTR_ALIGN((u8 *)&rt->rt_nh[rt->rt_nhn], VIA_ALEN_ALIGN);
65 	int nh_index = nh - rt->rt_nh;
66 
67 	return nh0_via + rt->rt_max_alen * nh_index;
68 }
69 
70 static const u8 *mpls_nh_via(const struct mpls_route *rt,
71 			     const struct mpls_nh *nh)
72 {
73 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
74 }
75 
76 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
77 {
78 	/* The size of the layer 2.5 labels to be added for this route */
79 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
80 }
81 
82 unsigned int mpls_dev_mtu(const struct net_device *dev)
83 {
84 	/* The amount of data the layer 2 frame can hold */
85 	return dev->mtu;
86 }
87 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
88 
89 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
90 {
91 	if (skb->len <= mtu)
92 		return false;
93 
94 	if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
95 		return false;
96 
97 	return true;
98 }
99 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
100 
101 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
102 					     struct sk_buff *skb, bool bos)
103 {
104 	struct mpls_entry_decoded dec;
105 	struct mpls_shim_hdr *hdr;
106 	bool eli_seen = false;
107 	int label_index;
108 	int nh_index = 0;
109 	u32 hash = 0;
110 
111 	/* No need to look further into packet if there's only
112 	 * one path
113 	 */
114 	if (rt->rt_nhn == 1)
115 		goto out;
116 
117 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS && !bos;
118 	     label_index++) {
119 		if (!pskb_may_pull(skb, sizeof(*hdr) * label_index))
120 			break;
121 
122 		/* Read and decode the current label */
123 		hdr = mpls_hdr(skb) + label_index;
124 		dec = mpls_entry_decode(hdr);
125 
126 		/* RFC6790 - reserved labels MUST NOT be used as keys
127 		 * for the load-balancing function
128 		 */
129 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
130 			hash = jhash_1word(dec.label, hash);
131 
132 			/* The entropy label follows the entropy label
133 			 * indicator, so this means that the entropy
134 			 * label was just added to the hash - no need to
135 			 * go any deeper either in the label stack or in the
136 			 * payload
137 			 */
138 			if (eli_seen)
139 				break;
140 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
141 			eli_seen = true;
142 		}
143 
144 		bos = dec.bos;
145 		if (bos && pskb_may_pull(skb, sizeof(*hdr) * label_index +
146 					 sizeof(struct iphdr))) {
147 			const struct iphdr *v4hdr;
148 
149 			v4hdr = (const struct iphdr *)(mpls_hdr(skb) +
150 						       label_index);
151 			if (v4hdr->version == 4) {
152 				hash = jhash_3words(ntohl(v4hdr->saddr),
153 						    ntohl(v4hdr->daddr),
154 						    v4hdr->protocol, hash);
155 			} else if (v4hdr->version == 6 &&
156 				pskb_may_pull(skb, sizeof(*hdr) * label_index +
157 					      sizeof(struct ipv6hdr))) {
158 				const struct ipv6hdr *v6hdr;
159 
160 				v6hdr = (const struct ipv6hdr *)(mpls_hdr(skb) +
161 								label_index);
162 
163 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
164 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
165 				hash = jhash_1word(v6hdr->nexthdr, hash);
166 			}
167 		}
168 	}
169 
170 	nh_index = hash % rt->rt_nhn;
171 out:
172 	return &rt->rt_nh[nh_index];
173 }
174 
175 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
176 			struct mpls_entry_decoded dec)
177 {
178 	enum mpls_payload_type payload_type;
179 	bool success = false;
180 
181 	/* The IPv4 code below accesses through the IPv4 header
182 	 * checksum, which is 12 bytes into the packet.
183 	 * The IPv6 code below accesses through the IPv6 hop limit
184 	 * which is 8 bytes into the packet.
185 	 *
186 	 * For all supported cases there should always be at least 12
187 	 * bytes of packet data present.  The IPv4 header is 20 bytes
188 	 * without options and the IPv6 header is always 40 bytes
189 	 * long.
190 	 */
191 	if (!pskb_may_pull(skb, 12))
192 		return false;
193 
194 	payload_type = rt->rt_payload_type;
195 	if (payload_type == MPT_UNSPEC)
196 		payload_type = ip_hdr(skb)->version;
197 
198 	switch (payload_type) {
199 	case MPT_IPV4: {
200 		struct iphdr *hdr4 = ip_hdr(skb);
201 		skb->protocol = htons(ETH_P_IP);
202 		csum_replace2(&hdr4->check,
203 			      htons(hdr4->ttl << 8),
204 			      htons(dec.ttl << 8));
205 		hdr4->ttl = dec.ttl;
206 		success = true;
207 		break;
208 	}
209 	case MPT_IPV6: {
210 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
211 		skb->protocol = htons(ETH_P_IPV6);
212 		hdr6->hop_limit = dec.ttl;
213 		success = true;
214 		break;
215 	}
216 	case MPT_UNSPEC:
217 		break;
218 	}
219 
220 	return success;
221 }
222 
223 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
224 			struct packet_type *pt, struct net_device *orig_dev)
225 {
226 	struct net *net = dev_net(dev);
227 	struct mpls_shim_hdr *hdr;
228 	struct mpls_route *rt;
229 	struct mpls_nh *nh;
230 	struct mpls_entry_decoded dec;
231 	struct net_device *out_dev;
232 	struct mpls_dev *mdev;
233 	unsigned int hh_len;
234 	unsigned int new_header_size;
235 	unsigned int mtu;
236 	int err;
237 
238 	/* Careful this entire function runs inside of an rcu critical section */
239 
240 	mdev = mpls_dev_get(dev);
241 	if (!mdev || !mdev->input_enabled)
242 		goto drop;
243 
244 	if (skb->pkt_type != PACKET_HOST)
245 		goto drop;
246 
247 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
248 		goto drop;
249 
250 	if (!pskb_may_pull(skb, sizeof(*hdr)))
251 		goto drop;
252 
253 	/* Read and decode the label */
254 	hdr = mpls_hdr(skb);
255 	dec = mpls_entry_decode(hdr);
256 
257 	/* Pop the label */
258 	skb_pull(skb, sizeof(*hdr));
259 	skb_reset_network_header(skb);
260 
261 	skb_orphan(skb);
262 
263 	rt = mpls_route_input_rcu(net, dec.label);
264 	if (!rt)
265 		goto drop;
266 
267 	nh = mpls_select_multipath(rt, skb, dec.bos);
268 	if (!nh)
269 		goto drop;
270 
271 	/* Find the output device */
272 	out_dev = rcu_dereference(nh->nh_dev);
273 	if (!mpls_output_possible(out_dev))
274 		goto drop;
275 
276 	if (skb_warn_if_lro(skb))
277 		goto drop;
278 
279 	skb_forward_csum(skb);
280 
281 	/* Verify ttl is valid */
282 	if (dec.ttl <= 1)
283 		goto drop;
284 	dec.ttl -= 1;
285 
286 	/* Verify the destination can hold the packet */
287 	new_header_size = mpls_nh_header_size(nh);
288 	mtu = mpls_dev_mtu(out_dev);
289 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
290 		goto drop;
291 
292 	hh_len = LL_RESERVED_SPACE(out_dev);
293 	if (!out_dev->header_ops)
294 		hh_len = 0;
295 
296 	/* Ensure there is enough space for the headers in the skb */
297 	if (skb_cow(skb, hh_len + new_header_size))
298 		goto drop;
299 
300 	skb->dev = out_dev;
301 	skb->protocol = htons(ETH_P_MPLS_UC);
302 
303 	if (unlikely(!new_header_size && dec.bos)) {
304 		/* Penultimate hop popping */
305 		if (!mpls_egress(rt, skb, dec))
306 			goto drop;
307 	} else {
308 		bool bos;
309 		int i;
310 		skb_push(skb, new_header_size);
311 		skb_reset_network_header(skb);
312 		/* Push the new labels */
313 		hdr = mpls_hdr(skb);
314 		bos = dec.bos;
315 		for (i = nh->nh_labels - 1; i >= 0; i--) {
316 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
317 						   dec.ttl, 0, bos);
318 			bos = false;
319 		}
320 	}
321 
322 	/* If via wasn't specified then send out using device address */
323 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
324 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
325 				 out_dev->dev_addr, skb);
326 	else
327 		err = neigh_xmit(nh->nh_via_table, out_dev,
328 				 mpls_nh_via(rt, nh), skb);
329 	if (err)
330 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
331 				    __func__, err);
332 	return 0;
333 
334 drop:
335 	kfree_skb(skb);
336 	return NET_RX_DROP;
337 }
338 
339 static struct packet_type mpls_packet_type __read_mostly = {
340 	.type = cpu_to_be16(ETH_P_MPLS_UC),
341 	.func = mpls_forward,
342 };
343 
344 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
345 	[RTA_DST]		= { .type = NLA_U32 },
346 	[RTA_OIF]		= { .type = NLA_U32 },
347 };
348 
349 struct mpls_route_config {
350 	u32			rc_protocol;
351 	u32			rc_ifindex;
352 	u8			rc_via_table;
353 	u8			rc_via_alen;
354 	u8			rc_via[MAX_VIA_ALEN];
355 	u32			rc_label;
356 	u8			rc_output_labels;
357 	u32			rc_output_label[MAX_NEW_LABELS];
358 	u32			rc_nlflags;
359 	enum mpls_payload_type	rc_payload_type;
360 	struct nl_info		rc_nlinfo;
361 	struct rtnexthop	*rc_mp;
362 	int			rc_mp_len;
363 };
364 
365 static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
366 {
367 	u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
368 	struct mpls_route *rt;
369 
370 	rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
371 			   VIA_ALEN_ALIGN) +
372 		     num_nh * max_alen_aligned,
373 		     GFP_KERNEL);
374 	if (rt) {
375 		rt->rt_nhn = num_nh;
376 		rt->rt_max_alen = max_alen_aligned;
377 	}
378 
379 	return rt;
380 }
381 
382 static void mpls_rt_free(struct mpls_route *rt)
383 {
384 	if (rt)
385 		kfree_rcu(rt, rt_rcu);
386 }
387 
388 static void mpls_notify_route(struct net *net, unsigned index,
389 			      struct mpls_route *old, struct mpls_route *new,
390 			      const struct nl_info *info)
391 {
392 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
393 	unsigned portid = info ? info->portid : 0;
394 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
395 	struct mpls_route *rt = new ? new : old;
396 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
397 	/* Ignore reserved labels for now */
398 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
399 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
400 }
401 
402 static void mpls_route_update(struct net *net, unsigned index,
403 			      struct mpls_route *new,
404 			      const struct nl_info *info)
405 {
406 	struct mpls_route __rcu **platform_label;
407 	struct mpls_route *rt;
408 
409 	ASSERT_RTNL();
410 
411 	platform_label = rtnl_dereference(net->mpls.platform_label);
412 	rt = rtnl_dereference(platform_label[index]);
413 	rcu_assign_pointer(platform_label[index], new);
414 
415 	mpls_notify_route(net, index, rt, new, info);
416 
417 	/* If we removed a route free it now */
418 	mpls_rt_free(rt);
419 }
420 
421 static unsigned find_free_label(struct net *net)
422 {
423 	struct mpls_route __rcu **platform_label;
424 	size_t platform_labels;
425 	unsigned index;
426 
427 	platform_label = rtnl_dereference(net->mpls.platform_label);
428 	platform_labels = net->mpls.platform_labels;
429 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
430 	     index++) {
431 		if (!rtnl_dereference(platform_label[index]))
432 			return index;
433 	}
434 	return LABEL_NOT_SPECIFIED;
435 }
436 
437 #if IS_ENABLED(CONFIG_INET)
438 static struct net_device *inet_fib_lookup_dev(struct net *net,
439 					      const void *addr)
440 {
441 	struct net_device *dev;
442 	struct rtable *rt;
443 	struct in_addr daddr;
444 
445 	memcpy(&daddr, addr, sizeof(struct in_addr));
446 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
447 	if (IS_ERR(rt))
448 		return ERR_CAST(rt);
449 
450 	dev = rt->dst.dev;
451 	dev_hold(dev);
452 
453 	ip_rt_put(rt);
454 
455 	return dev;
456 }
457 #else
458 static struct net_device *inet_fib_lookup_dev(struct net *net,
459 					      const void *addr)
460 {
461 	return ERR_PTR(-EAFNOSUPPORT);
462 }
463 #endif
464 
465 #if IS_ENABLED(CONFIG_IPV6)
466 static struct net_device *inet6_fib_lookup_dev(struct net *net,
467 					       const void *addr)
468 {
469 	struct net_device *dev;
470 	struct dst_entry *dst;
471 	struct flowi6 fl6;
472 	int err;
473 
474 	if (!ipv6_stub)
475 		return ERR_PTR(-EAFNOSUPPORT);
476 
477 	memset(&fl6, 0, sizeof(fl6));
478 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
479 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
480 	if (err)
481 		return ERR_PTR(err);
482 
483 	dev = dst->dev;
484 	dev_hold(dev);
485 	dst_release(dst);
486 
487 	return dev;
488 }
489 #else
490 static struct net_device *inet6_fib_lookup_dev(struct net *net,
491 					       const void *addr)
492 {
493 	return ERR_PTR(-EAFNOSUPPORT);
494 }
495 #endif
496 
497 static struct net_device *find_outdev(struct net *net,
498 				      struct mpls_route *rt,
499 				      struct mpls_nh *nh, int oif)
500 {
501 	struct net_device *dev = NULL;
502 
503 	if (!oif) {
504 		switch (nh->nh_via_table) {
505 		case NEIGH_ARP_TABLE:
506 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
507 			break;
508 		case NEIGH_ND_TABLE:
509 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
510 			break;
511 		case NEIGH_LINK_TABLE:
512 			break;
513 		}
514 	} else {
515 		dev = dev_get_by_index(net, oif);
516 	}
517 
518 	if (!dev)
519 		return ERR_PTR(-ENODEV);
520 
521 	/* The caller is holding rtnl anyways, so release the dev reference */
522 	dev_put(dev);
523 
524 	return dev;
525 }
526 
527 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
528 			      struct mpls_nh *nh, int oif)
529 {
530 	struct net_device *dev = NULL;
531 	int err = -ENODEV;
532 
533 	dev = find_outdev(net, rt, nh, oif);
534 	if (IS_ERR(dev)) {
535 		err = PTR_ERR(dev);
536 		dev = NULL;
537 		goto errout;
538 	}
539 
540 	/* Ensure this is a supported device */
541 	err = -EINVAL;
542 	if (!mpls_dev_get(dev))
543 		goto errout;
544 
545 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
546 	    (dev->addr_len != nh->nh_via_alen))
547 		goto errout;
548 
549 	RCU_INIT_POINTER(nh->nh_dev, dev);
550 
551 	return 0;
552 
553 errout:
554 	return err;
555 }
556 
557 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
558 				  struct mpls_route *rt)
559 {
560 	struct net *net = cfg->rc_nlinfo.nl_net;
561 	struct mpls_nh *nh = rt->rt_nh;
562 	int err;
563 	int i;
564 
565 	if (!nh)
566 		return -ENOMEM;
567 
568 	err = -EINVAL;
569 	/* Ensure only a supported number of labels are present */
570 	if (cfg->rc_output_labels > MAX_NEW_LABELS)
571 		goto errout;
572 
573 	nh->nh_labels = cfg->rc_output_labels;
574 	for (i = 0; i < nh->nh_labels; i++)
575 		nh->nh_label[i] = cfg->rc_output_label[i];
576 
577 	nh->nh_via_table = cfg->rc_via_table;
578 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
579 	nh->nh_via_alen = cfg->rc_via_alen;
580 
581 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
582 	if (err)
583 		goto errout;
584 
585 	return 0;
586 
587 errout:
588 	return err;
589 }
590 
591 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
592 			 struct mpls_nh *nh, int oif,
593 			 struct nlattr *via, struct nlattr *newdst)
594 {
595 	int err = -ENOMEM;
596 
597 	if (!nh)
598 		goto errout;
599 
600 	if (newdst) {
601 		err = nla_get_labels(newdst, MAX_NEW_LABELS,
602 				     &nh->nh_labels, nh->nh_label);
603 		if (err)
604 			goto errout;
605 	}
606 
607 	if (via) {
608 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
609 				  __mpls_nh_via(rt, nh));
610 		if (err)
611 			goto errout;
612 	} else {
613 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
614 	}
615 
616 	err = mpls_nh_assign_dev(net, rt, nh, oif);
617 	if (err)
618 		goto errout;
619 
620 	return 0;
621 
622 errout:
623 	return err;
624 }
625 
626 static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
627 			       u8 cfg_via_alen, u8 *max_via_alen)
628 {
629 	int nhs = 0;
630 	int remaining = len;
631 
632 	if (!rtnh) {
633 		*max_via_alen = cfg_via_alen;
634 		return 1;
635 	}
636 
637 	*max_via_alen = 0;
638 
639 	while (rtnh_ok(rtnh, remaining)) {
640 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
641 		int attrlen;
642 
643 		attrlen = rtnh_attrlen(rtnh);
644 		nla = nla_find(attrs, attrlen, RTA_VIA);
645 		if (nla && nla_len(nla) >=
646 		    offsetof(struct rtvia, rtvia_addr)) {
647 			int via_alen = nla_len(nla) -
648 				offsetof(struct rtvia, rtvia_addr);
649 
650 			if (via_alen <= MAX_VIA_ALEN)
651 				*max_via_alen = max_t(u16, *max_via_alen,
652 						      via_alen);
653 		}
654 
655 		nhs++;
656 		rtnh = rtnh_next(rtnh, &remaining);
657 	}
658 
659 	/* leftover implies invalid nexthop configuration, discard it */
660 	return remaining > 0 ? 0 : nhs;
661 }
662 
663 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
664 			       struct mpls_route *rt)
665 {
666 	struct rtnexthop *rtnh = cfg->rc_mp;
667 	struct nlattr *nla_via, *nla_newdst;
668 	int remaining = cfg->rc_mp_len;
669 	int nhs = 0;
670 	int err = 0;
671 
672 	change_nexthops(rt) {
673 		int attrlen;
674 
675 		nla_via = NULL;
676 		nla_newdst = NULL;
677 
678 		err = -EINVAL;
679 		if (!rtnh_ok(rtnh, remaining))
680 			goto errout;
681 
682 		/* neither weighted multipath nor any flags
683 		 * are supported
684 		 */
685 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
686 			goto errout;
687 
688 		attrlen = rtnh_attrlen(rtnh);
689 		if (attrlen > 0) {
690 			struct nlattr *attrs = rtnh_attrs(rtnh);
691 
692 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
693 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
694 		}
695 
696 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
697 				    rtnh->rtnh_ifindex, nla_via,
698 				    nla_newdst);
699 		if (err)
700 			goto errout;
701 
702 		rtnh = rtnh_next(rtnh, &remaining);
703 		nhs++;
704 	} endfor_nexthops(rt);
705 
706 	rt->rt_nhn = nhs;
707 
708 	return 0;
709 
710 errout:
711 	return err;
712 }
713 
714 static int mpls_route_add(struct mpls_route_config *cfg)
715 {
716 	struct mpls_route __rcu **platform_label;
717 	struct net *net = cfg->rc_nlinfo.nl_net;
718 	struct mpls_route *rt, *old;
719 	int err = -EINVAL;
720 	u8 max_via_alen;
721 	unsigned index;
722 	int nhs;
723 
724 	index = cfg->rc_label;
725 
726 	/* If a label was not specified during insert pick one */
727 	if ((index == LABEL_NOT_SPECIFIED) &&
728 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
729 		index = find_free_label(net);
730 	}
731 
732 	/* Reserved labels may not be set */
733 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
734 		goto errout;
735 
736 	/* The full 20 bit range may not be supported. */
737 	if (index >= net->mpls.platform_labels)
738 		goto errout;
739 
740 	/* Append makes no sense with mpls */
741 	err = -EOPNOTSUPP;
742 	if (cfg->rc_nlflags & NLM_F_APPEND)
743 		goto errout;
744 
745 	err = -EEXIST;
746 	platform_label = rtnl_dereference(net->mpls.platform_label);
747 	old = rtnl_dereference(platform_label[index]);
748 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
749 		goto errout;
750 
751 	err = -EEXIST;
752 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
753 		goto errout;
754 
755 	err = -ENOENT;
756 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
757 		goto errout;
758 
759 	err = -EINVAL;
760 	nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
761 				  cfg->rc_via_alen, &max_via_alen);
762 	if (nhs == 0)
763 		goto errout;
764 
765 	err = -ENOMEM;
766 	rt = mpls_rt_alloc(nhs, max_via_alen);
767 	if (!rt)
768 		goto errout;
769 
770 	rt->rt_protocol = cfg->rc_protocol;
771 	rt->rt_payload_type = cfg->rc_payload_type;
772 
773 	if (cfg->rc_mp)
774 		err = mpls_nh_build_multi(cfg, rt);
775 	else
776 		err = mpls_nh_build_from_cfg(cfg, rt);
777 	if (err)
778 		goto freert;
779 
780 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
781 
782 	return 0;
783 
784 freert:
785 	mpls_rt_free(rt);
786 errout:
787 	return err;
788 }
789 
790 static int mpls_route_del(struct mpls_route_config *cfg)
791 {
792 	struct net *net = cfg->rc_nlinfo.nl_net;
793 	unsigned index;
794 	int err = -EINVAL;
795 
796 	index = cfg->rc_label;
797 
798 	/* Reserved labels may not be removed */
799 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
800 		goto errout;
801 
802 	/* The full 20 bit range may not be supported */
803 	if (index >= net->mpls.platform_labels)
804 		goto errout;
805 
806 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
807 
808 	err = 0;
809 errout:
810 	return err;
811 }
812 
813 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
814 	(&((struct mpls_dev *)0)->field)
815 
816 static const struct ctl_table mpls_dev_table[] = {
817 	{
818 		.procname	= "input",
819 		.maxlen		= sizeof(int),
820 		.mode		= 0644,
821 		.proc_handler	= proc_dointvec,
822 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
823 	},
824 	{ }
825 };
826 
827 static int mpls_dev_sysctl_register(struct net_device *dev,
828 				    struct mpls_dev *mdev)
829 {
830 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
831 	struct ctl_table *table;
832 	int i;
833 
834 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
835 	if (!table)
836 		goto out;
837 
838 	/* Table data contains only offsets relative to the base of
839 	 * the mdev at this point, so make them absolute.
840 	 */
841 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
842 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
843 
844 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
845 
846 	mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
847 	if (!mdev->sysctl)
848 		goto free;
849 
850 	return 0;
851 
852 free:
853 	kfree(table);
854 out:
855 	return -ENOBUFS;
856 }
857 
858 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
859 {
860 	struct ctl_table *table;
861 
862 	table = mdev->sysctl->ctl_table_arg;
863 	unregister_net_sysctl_table(mdev->sysctl);
864 	kfree(table);
865 }
866 
867 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
868 {
869 	struct mpls_dev *mdev;
870 	int err = -ENOMEM;
871 
872 	ASSERT_RTNL();
873 
874 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
875 	if (!mdev)
876 		return ERR_PTR(err);
877 
878 	err = mpls_dev_sysctl_register(dev, mdev);
879 	if (err)
880 		goto free;
881 
882 	rcu_assign_pointer(dev->mpls_ptr, mdev);
883 
884 	return mdev;
885 
886 free:
887 	kfree(mdev);
888 	return ERR_PTR(err);
889 }
890 
891 static void mpls_ifdown(struct net_device *dev)
892 {
893 	struct mpls_route __rcu **platform_label;
894 	struct net *net = dev_net(dev);
895 	struct mpls_dev *mdev;
896 	unsigned index;
897 
898 	platform_label = rtnl_dereference(net->mpls.platform_label);
899 	for (index = 0; index < net->mpls.platform_labels; index++) {
900 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
901 		if (!rt)
902 			continue;
903 		for_nexthops(rt) {
904 			if (rtnl_dereference(nh->nh_dev) != dev)
905 				continue;
906 			nh->nh_dev = NULL;
907 		} endfor_nexthops(rt);
908 	}
909 
910 	mdev = mpls_dev_get(dev);
911 	if (!mdev)
912 		return;
913 
914 	mpls_dev_sysctl_unregister(mdev);
915 
916 	RCU_INIT_POINTER(dev->mpls_ptr, NULL);
917 
918 	kfree_rcu(mdev, rcu);
919 }
920 
921 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
922 			   void *ptr)
923 {
924 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
925 	struct mpls_dev *mdev;
926 
927 	switch(event) {
928 	case NETDEV_REGISTER:
929 		/* For now just support ethernet devices */
930 		if ((dev->type == ARPHRD_ETHER) ||
931 		    (dev->type == ARPHRD_LOOPBACK)) {
932 			mdev = mpls_add_dev(dev);
933 			if (IS_ERR(mdev))
934 				return notifier_from_errno(PTR_ERR(mdev));
935 		}
936 		break;
937 
938 	case NETDEV_UNREGISTER:
939 		mpls_ifdown(dev);
940 		break;
941 	case NETDEV_CHANGENAME:
942 		mdev = mpls_dev_get(dev);
943 		if (mdev) {
944 			int err;
945 
946 			mpls_dev_sysctl_unregister(mdev);
947 			err = mpls_dev_sysctl_register(dev, mdev);
948 			if (err)
949 				return notifier_from_errno(err);
950 		}
951 		break;
952 	}
953 	return NOTIFY_OK;
954 }
955 
956 static struct notifier_block mpls_dev_notifier = {
957 	.notifier_call = mpls_dev_notify,
958 };
959 
960 static int nla_put_via(struct sk_buff *skb,
961 		       u8 table, const void *addr, int alen)
962 {
963 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
964 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
965 	};
966 	struct nlattr *nla;
967 	struct rtvia *via;
968 	int family = AF_UNSPEC;
969 
970 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
971 	if (!nla)
972 		return -EMSGSIZE;
973 
974 	if (table <= NEIGH_NR_TABLES)
975 		family = table_to_family[table];
976 
977 	via = nla_data(nla);
978 	via->rtvia_family = family;
979 	memcpy(via->rtvia_addr, addr, alen);
980 	return 0;
981 }
982 
983 int nla_put_labels(struct sk_buff *skb, int attrtype,
984 		   u8 labels, const u32 label[])
985 {
986 	struct nlattr *nla;
987 	struct mpls_shim_hdr *nla_label;
988 	bool bos;
989 	int i;
990 	nla = nla_reserve(skb, attrtype, labels*4);
991 	if (!nla)
992 		return -EMSGSIZE;
993 
994 	nla_label = nla_data(nla);
995 	bos = true;
996 	for (i = labels - 1; i >= 0; i--) {
997 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
998 		bos = false;
999 	}
1000 
1001 	return 0;
1002 }
1003 EXPORT_SYMBOL_GPL(nla_put_labels);
1004 
1005 int nla_get_labels(const struct nlattr *nla,
1006 		   u32 max_labels, u8 *labels, u32 label[])
1007 {
1008 	unsigned len = nla_len(nla);
1009 	unsigned nla_labels;
1010 	struct mpls_shim_hdr *nla_label;
1011 	bool bos;
1012 	int i;
1013 
1014 	/* len needs to be an even multiple of 4 (the label size) */
1015 	if (len & 3)
1016 		return -EINVAL;
1017 
1018 	/* Limit the number of new labels allowed */
1019 	nla_labels = len/4;
1020 	if (nla_labels > max_labels)
1021 		return -EINVAL;
1022 
1023 	nla_label = nla_data(nla);
1024 	bos = true;
1025 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1026 		struct mpls_entry_decoded dec;
1027 		dec = mpls_entry_decode(nla_label + i);
1028 
1029 		/* Ensure the bottom of stack flag is properly set
1030 		 * and ttl and tc are both clear.
1031 		 */
1032 		if ((dec.bos != bos) || dec.ttl || dec.tc)
1033 			return -EINVAL;
1034 
1035 		switch (dec.label) {
1036 		case MPLS_LABEL_IMPLNULL:
1037 			/* RFC3032: This is a label that an LSR may
1038 			 * assign and distribute, but which never
1039 			 * actually appears in the encapsulation.
1040 			 */
1041 			return -EINVAL;
1042 		}
1043 
1044 		label[i] = dec.label;
1045 	}
1046 	*labels = nla_labels;
1047 	return 0;
1048 }
1049 EXPORT_SYMBOL_GPL(nla_get_labels);
1050 
1051 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1052 		u8 *via_table, u8 via_addr[])
1053 {
1054 	struct rtvia *via = nla_data(nla);
1055 	int err = -EINVAL;
1056 	int alen;
1057 
1058 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1059 		goto errout;
1060 	alen = nla_len(nla) -
1061 			offsetof(struct rtvia, rtvia_addr);
1062 	if (alen > MAX_VIA_ALEN)
1063 		goto errout;
1064 
1065 	/* Validate the address family */
1066 	switch (via->rtvia_family) {
1067 	case AF_PACKET:
1068 		*via_table = NEIGH_LINK_TABLE;
1069 		break;
1070 	case AF_INET:
1071 		*via_table = NEIGH_ARP_TABLE;
1072 		if (alen != 4)
1073 			goto errout;
1074 		break;
1075 	case AF_INET6:
1076 		*via_table = NEIGH_ND_TABLE;
1077 		if (alen != 16)
1078 			goto errout;
1079 		break;
1080 	default:
1081 		/* Unsupported address family */
1082 		goto errout;
1083 	}
1084 
1085 	memcpy(via_addr, via->rtvia_addr, alen);
1086 	*via_alen = alen;
1087 	err = 0;
1088 
1089 errout:
1090 	return err;
1091 }
1092 
1093 static int rtm_to_route_config(struct sk_buff *skb,  struct nlmsghdr *nlh,
1094 			       struct mpls_route_config *cfg)
1095 {
1096 	struct rtmsg *rtm;
1097 	struct nlattr *tb[RTA_MAX+1];
1098 	int index;
1099 	int err;
1100 
1101 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1102 	if (err < 0)
1103 		goto errout;
1104 
1105 	err = -EINVAL;
1106 	rtm = nlmsg_data(nlh);
1107 	memset(cfg, 0, sizeof(*cfg));
1108 
1109 	if (rtm->rtm_family != AF_MPLS)
1110 		goto errout;
1111 	if (rtm->rtm_dst_len != 20)
1112 		goto errout;
1113 	if (rtm->rtm_src_len != 0)
1114 		goto errout;
1115 	if (rtm->rtm_tos != 0)
1116 		goto errout;
1117 	if (rtm->rtm_table != RT_TABLE_MAIN)
1118 		goto errout;
1119 	/* Any value is acceptable for rtm_protocol */
1120 
1121 	/* As mpls uses destination specific addresses
1122 	 * (or source specific address in the case of multicast)
1123 	 * all addresses have universal scope.
1124 	 */
1125 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1126 		goto errout;
1127 	if (rtm->rtm_type != RTN_UNICAST)
1128 		goto errout;
1129 	if (rtm->rtm_flags != 0)
1130 		goto errout;
1131 
1132 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1133 	cfg->rc_protocol	= rtm->rtm_protocol;
1134 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1135 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1136 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1137 	cfg->rc_nlinfo.nlh	= nlh;
1138 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1139 
1140 	for (index = 0; index <= RTA_MAX; index++) {
1141 		struct nlattr *nla = tb[index];
1142 		if (!nla)
1143 			continue;
1144 
1145 		switch(index) {
1146 		case RTA_OIF:
1147 			cfg->rc_ifindex = nla_get_u32(nla);
1148 			break;
1149 		case RTA_NEWDST:
1150 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1151 					   &cfg->rc_output_labels,
1152 					   cfg->rc_output_label))
1153 				goto errout;
1154 			break;
1155 		case RTA_DST:
1156 		{
1157 			u8 label_count;
1158 			if (nla_get_labels(nla, 1, &label_count,
1159 					   &cfg->rc_label))
1160 				goto errout;
1161 
1162 			/* Reserved labels may not be set */
1163 			if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1164 				goto errout;
1165 
1166 			break;
1167 		}
1168 		case RTA_VIA:
1169 		{
1170 			if (nla_get_via(nla, &cfg->rc_via_alen,
1171 					&cfg->rc_via_table, cfg->rc_via))
1172 				goto errout;
1173 			break;
1174 		}
1175 		case RTA_MULTIPATH:
1176 		{
1177 			cfg->rc_mp = nla_data(nla);
1178 			cfg->rc_mp_len = nla_len(nla);
1179 			break;
1180 		}
1181 		default:
1182 			/* Unsupported attribute */
1183 			goto errout;
1184 		}
1185 	}
1186 
1187 	err = 0;
1188 errout:
1189 	return err;
1190 }
1191 
1192 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1193 {
1194 	struct mpls_route_config cfg;
1195 	int err;
1196 
1197 	err = rtm_to_route_config(skb, nlh, &cfg);
1198 	if (err < 0)
1199 		return err;
1200 
1201 	return mpls_route_del(&cfg);
1202 }
1203 
1204 
1205 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1206 {
1207 	struct mpls_route_config cfg;
1208 	int err;
1209 
1210 	err = rtm_to_route_config(skb, nlh, &cfg);
1211 	if (err < 0)
1212 		return err;
1213 
1214 	return mpls_route_add(&cfg);
1215 }
1216 
1217 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1218 			   u32 label, struct mpls_route *rt, int flags)
1219 {
1220 	struct net_device *dev;
1221 	struct nlmsghdr *nlh;
1222 	struct rtmsg *rtm;
1223 
1224 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1225 	if (nlh == NULL)
1226 		return -EMSGSIZE;
1227 
1228 	rtm = nlmsg_data(nlh);
1229 	rtm->rtm_family = AF_MPLS;
1230 	rtm->rtm_dst_len = 20;
1231 	rtm->rtm_src_len = 0;
1232 	rtm->rtm_tos = 0;
1233 	rtm->rtm_table = RT_TABLE_MAIN;
1234 	rtm->rtm_protocol = rt->rt_protocol;
1235 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1236 	rtm->rtm_type = RTN_UNICAST;
1237 	rtm->rtm_flags = 0;
1238 
1239 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1240 		goto nla_put_failure;
1241 	if (rt->rt_nhn == 1) {
1242 		const struct mpls_nh *nh = rt->rt_nh;
1243 
1244 		if (nh->nh_labels &&
1245 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1246 				   nh->nh_label))
1247 			goto nla_put_failure;
1248 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1249 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1250 				nh->nh_via_alen))
1251 			goto nla_put_failure;
1252 		dev = rtnl_dereference(nh->nh_dev);
1253 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1254 			goto nla_put_failure;
1255 	} else {
1256 		struct rtnexthop *rtnh;
1257 		struct nlattr *mp;
1258 
1259 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1260 		if (!mp)
1261 			goto nla_put_failure;
1262 
1263 		for_nexthops(rt) {
1264 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1265 			if (!rtnh)
1266 				goto nla_put_failure;
1267 
1268 			dev = rtnl_dereference(nh->nh_dev);
1269 			if (dev)
1270 				rtnh->rtnh_ifindex = dev->ifindex;
1271 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1272 							    nh->nh_labels,
1273 							    nh->nh_label))
1274 				goto nla_put_failure;
1275 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1276 			    nla_put_via(skb, nh->nh_via_table,
1277 					mpls_nh_via(rt, nh),
1278 					nh->nh_via_alen))
1279 				goto nla_put_failure;
1280 
1281 			/* length of rtnetlink header + attributes */
1282 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1283 		} endfor_nexthops(rt);
1284 
1285 		nla_nest_end(skb, mp);
1286 	}
1287 
1288 	nlmsg_end(skb, nlh);
1289 	return 0;
1290 
1291 nla_put_failure:
1292 	nlmsg_cancel(skb, nlh);
1293 	return -EMSGSIZE;
1294 }
1295 
1296 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1297 {
1298 	struct net *net = sock_net(skb->sk);
1299 	struct mpls_route __rcu **platform_label;
1300 	size_t platform_labels;
1301 	unsigned int index;
1302 
1303 	ASSERT_RTNL();
1304 
1305 	index = cb->args[0];
1306 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
1307 		index = MPLS_LABEL_FIRST_UNRESERVED;
1308 
1309 	platform_label = rtnl_dereference(net->mpls.platform_label);
1310 	platform_labels = net->mpls.platform_labels;
1311 	for (; index < platform_labels; index++) {
1312 		struct mpls_route *rt;
1313 		rt = rtnl_dereference(platform_label[index]);
1314 		if (!rt)
1315 			continue;
1316 
1317 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1318 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1319 				    index, rt, NLM_F_MULTI) < 0)
1320 			break;
1321 	}
1322 	cb->args[0] = index;
1323 
1324 	return skb->len;
1325 }
1326 
1327 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1328 {
1329 	size_t payload =
1330 		NLMSG_ALIGN(sizeof(struct rtmsg))
1331 		+ nla_total_size(4);			/* RTA_DST */
1332 
1333 	if (rt->rt_nhn == 1) {
1334 		struct mpls_nh *nh = rt->rt_nh;
1335 
1336 		if (nh->nh_dev)
1337 			payload += nla_total_size(4); /* RTA_OIF */
1338 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1339 			payload += nla_total_size(2 + nh->nh_via_alen);
1340 		if (nh->nh_labels) /* RTA_NEWDST */
1341 			payload += nla_total_size(nh->nh_labels * 4);
1342 	} else {
1343 		/* each nexthop is packed in an attribute */
1344 		size_t nhsize = 0;
1345 
1346 		for_nexthops(rt) {
1347 			nhsize += nla_total_size(sizeof(struct rtnexthop));
1348 			/* RTA_VIA */
1349 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1350 				nhsize += nla_total_size(2 + nh->nh_via_alen);
1351 			if (nh->nh_labels)
1352 				nhsize += nla_total_size(nh->nh_labels * 4);
1353 		} endfor_nexthops(rt);
1354 		/* nested attribute */
1355 		payload += nla_total_size(nhsize);
1356 	}
1357 
1358 	return payload;
1359 }
1360 
1361 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1362 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
1363 		       unsigned int nlm_flags)
1364 {
1365 	struct sk_buff *skb;
1366 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
1367 	int err = -ENOBUFS;
1368 
1369 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1370 	if (skb == NULL)
1371 		goto errout;
1372 
1373 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1374 	if (err < 0) {
1375 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1376 		WARN_ON(err == -EMSGSIZE);
1377 		kfree_skb(skb);
1378 		goto errout;
1379 	}
1380 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1381 
1382 	return;
1383 errout:
1384 	if (err < 0)
1385 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1386 }
1387 
1388 static int resize_platform_label_table(struct net *net, size_t limit)
1389 {
1390 	size_t size = sizeof(struct mpls_route *) * limit;
1391 	size_t old_limit;
1392 	size_t cp_size;
1393 	struct mpls_route __rcu **labels = NULL, **old;
1394 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
1395 	unsigned index;
1396 
1397 	if (size) {
1398 		labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1399 		if (!labels)
1400 			labels = vzalloc(size);
1401 
1402 		if (!labels)
1403 			goto nolabels;
1404 	}
1405 
1406 	/* In case the predefined labels need to be populated */
1407 	if (limit > MPLS_LABEL_IPV4NULL) {
1408 		struct net_device *lo = net->loopback_dev;
1409 		rt0 = mpls_rt_alloc(1, lo->addr_len);
1410 		if (!rt0)
1411 			goto nort0;
1412 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1413 		rt0->rt_protocol = RTPROT_KERNEL;
1414 		rt0->rt_payload_type = MPT_IPV4;
1415 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1416 		rt0->rt_nh->nh_via_alen = lo->addr_len;
1417 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1418 		       lo->addr_len);
1419 	}
1420 	if (limit > MPLS_LABEL_IPV6NULL) {
1421 		struct net_device *lo = net->loopback_dev;
1422 		rt2 = mpls_rt_alloc(1, lo->addr_len);
1423 		if (!rt2)
1424 			goto nort2;
1425 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1426 		rt2->rt_protocol = RTPROT_KERNEL;
1427 		rt2->rt_payload_type = MPT_IPV6;
1428 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1429 		rt2->rt_nh->nh_via_alen = lo->addr_len;
1430 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1431 		       lo->addr_len);
1432 	}
1433 
1434 	rtnl_lock();
1435 	/* Remember the original table */
1436 	old = rtnl_dereference(net->mpls.platform_label);
1437 	old_limit = net->mpls.platform_labels;
1438 
1439 	/* Free any labels beyond the new table */
1440 	for (index = limit; index < old_limit; index++)
1441 		mpls_route_update(net, index, NULL, NULL);
1442 
1443 	/* Copy over the old labels */
1444 	cp_size = size;
1445 	if (old_limit < limit)
1446 		cp_size = old_limit * sizeof(struct mpls_route *);
1447 
1448 	memcpy(labels, old, cp_size);
1449 
1450 	/* If needed set the predefined labels */
1451 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1452 	    (limit > MPLS_LABEL_IPV6NULL)) {
1453 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1454 		rt2 = NULL;
1455 	}
1456 
1457 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1458 	    (limit > MPLS_LABEL_IPV4NULL)) {
1459 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1460 		rt0 = NULL;
1461 	}
1462 
1463 	/* Update the global pointers */
1464 	net->mpls.platform_labels = limit;
1465 	rcu_assign_pointer(net->mpls.platform_label, labels);
1466 
1467 	rtnl_unlock();
1468 
1469 	mpls_rt_free(rt2);
1470 	mpls_rt_free(rt0);
1471 
1472 	if (old) {
1473 		synchronize_rcu();
1474 		kvfree(old);
1475 	}
1476 	return 0;
1477 
1478 nort2:
1479 	mpls_rt_free(rt0);
1480 nort0:
1481 	kvfree(labels);
1482 nolabels:
1483 	return -ENOMEM;
1484 }
1485 
1486 static int mpls_platform_labels(struct ctl_table *table, int write,
1487 				void __user *buffer, size_t *lenp, loff_t *ppos)
1488 {
1489 	struct net *net = table->data;
1490 	int platform_labels = net->mpls.platform_labels;
1491 	int ret;
1492 	struct ctl_table tmp = {
1493 		.procname	= table->procname,
1494 		.data		= &platform_labels,
1495 		.maxlen		= sizeof(int),
1496 		.mode		= table->mode,
1497 		.extra1		= &zero,
1498 		.extra2		= &label_limit,
1499 	};
1500 
1501 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1502 
1503 	if (write && ret == 0)
1504 		ret = resize_platform_label_table(net, platform_labels);
1505 
1506 	return ret;
1507 }
1508 
1509 static const struct ctl_table mpls_table[] = {
1510 	{
1511 		.procname	= "platform_labels",
1512 		.data		= NULL,
1513 		.maxlen		= sizeof(int),
1514 		.mode		= 0644,
1515 		.proc_handler	= mpls_platform_labels,
1516 	},
1517 	{ }
1518 };
1519 
1520 static int mpls_net_init(struct net *net)
1521 {
1522 	struct ctl_table *table;
1523 
1524 	net->mpls.platform_labels = 0;
1525 	net->mpls.platform_label = NULL;
1526 
1527 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1528 	if (table == NULL)
1529 		return -ENOMEM;
1530 
1531 	table[0].data = net;
1532 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1533 	if (net->mpls.ctl == NULL) {
1534 		kfree(table);
1535 		return -ENOMEM;
1536 	}
1537 
1538 	return 0;
1539 }
1540 
1541 static void mpls_net_exit(struct net *net)
1542 {
1543 	struct mpls_route __rcu **platform_label;
1544 	size_t platform_labels;
1545 	struct ctl_table *table;
1546 	unsigned int index;
1547 
1548 	table = net->mpls.ctl->ctl_table_arg;
1549 	unregister_net_sysctl_table(net->mpls.ctl);
1550 	kfree(table);
1551 
1552 	/* An rcu grace period has passed since there was a device in
1553 	 * the network namespace (and thus the last in flight packet)
1554 	 * left this network namespace.  This is because
1555 	 * unregister_netdevice_many and netdev_run_todo has completed
1556 	 * for each network device that was in this network namespace.
1557 	 *
1558 	 * As such no additional rcu synchronization is necessary when
1559 	 * freeing the platform_label table.
1560 	 */
1561 	rtnl_lock();
1562 	platform_label = rtnl_dereference(net->mpls.platform_label);
1563 	platform_labels = net->mpls.platform_labels;
1564 	for (index = 0; index < platform_labels; index++) {
1565 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1566 		RCU_INIT_POINTER(platform_label[index], NULL);
1567 		mpls_rt_free(rt);
1568 	}
1569 	rtnl_unlock();
1570 
1571 	kvfree(platform_label);
1572 }
1573 
1574 static struct pernet_operations mpls_net_ops = {
1575 	.init = mpls_net_init,
1576 	.exit = mpls_net_exit,
1577 };
1578 
1579 static int __init mpls_init(void)
1580 {
1581 	int err;
1582 
1583 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1584 
1585 	err = register_pernet_subsys(&mpls_net_ops);
1586 	if (err)
1587 		goto out;
1588 
1589 	err = register_netdevice_notifier(&mpls_dev_notifier);
1590 	if (err)
1591 		goto out_unregister_pernet;
1592 
1593 	dev_add_pack(&mpls_packet_type);
1594 
1595 	rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1596 	rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1597 	rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1598 	err = 0;
1599 out:
1600 	return err;
1601 
1602 out_unregister_pernet:
1603 	unregister_pernet_subsys(&mpls_net_ops);
1604 	goto out;
1605 }
1606 module_init(mpls_init);
1607 
1608 static void __exit mpls_exit(void)
1609 {
1610 	rtnl_unregister_all(PF_MPLS);
1611 	dev_remove_pack(&mpls_packet_type);
1612 	unregister_netdevice_notifier(&mpls_dev_notifier);
1613 	unregister_pernet_subsys(&mpls_net_ops);
1614 }
1615 module_exit(mpls_exit);
1616 
1617 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1618 MODULE_LICENSE("GPL v2");
1619 MODULE_ALIAS_NETPROTO(PF_MPLS);
1620