xref: /openbmc/linux/net/mpls/af_mpls.c (revision bc5aa3a0)
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_validate_mtu(skb, mtu))
95 		return false;
96 
97 	return true;
98 }
99 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
100 
101 static u32 mpls_multipath_hash(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 	u32 hash = 0;
109 
110 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS && !bos;
111 	     label_index++) {
112 		if (!pskb_may_pull(skb, sizeof(*hdr) * label_index))
113 			break;
114 
115 		/* Read and decode the current label */
116 		hdr = mpls_hdr(skb) + label_index;
117 		dec = mpls_entry_decode(hdr);
118 
119 		/* RFC6790 - reserved labels MUST NOT be used as keys
120 		 * for the load-balancing function
121 		 */
122 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
123 			hash = jhash_1word(dec.label, hash);
124 
125 			/* The entropy label follows the entropy label
126 			 * indicator, so this means that the entropy
127 			 * label was just added to the hash - no need to
128 			 * go any deeper either in the label stack or in the
129 			 * payload
130 			 */
131 			if (eli_seen)
132 				break;
133 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
134 			eli_seen = true;
135 		}
136 
137 		bos = dec.bos;
138 		if (bos && pskb_may_pull(skb, sizeof(*hdr) * label_index +
139 					 sizeof(struct iphdr))) {
140 			const struct iphdr *v4hdr;
141 
142 			v4hdr = (const struct iphdr *)(mpls_hdr(skb) +
143 						       label_index);
144 			if (v4hdr->version == 4) {
145 				hash = jhash_3words(ntohl(v4hdr->saddr),
146 						    ntohl(v4hdr->daddr),
147 						    v4hdr->protocol, hash);
148 			} else if (v4hdr->version == 6 &&
149 				pskb_may_pull(skb, sizeof(*hdr) * label_index +
150 					      sizeof(struct ipv6hdr))) {
151 				const struct ipv6hdr *v6hdr;
152 
153 				v6hdr = (const struct ipv6hdr *)(mpls_hdr(skb) +
154 								label_index);
155 
156 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
157 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
158 				hash = jhash_1word(v6hdr->nexthdr, hash);
159 			}
160 		}
161 	}
162 
163 	return hash;
164 }
165 
166 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
167 					     struct sk_buff *skb, bool bos)
168 {
169 	int alive = ACCESS_ONCE(rt->rt_nhn_alive);
170 	u32 hash = 0;
171 	int nh_index = 0;
172 	int n = 0;
173 
174 	/* No need to look further into packet if there's only
175 	 * one path
176 	 */
177 	if (rt->rt_nhn == 1)
178 		goto out;
179 
180 	if (alive <= 0)
181 		return NULL;
182 
183 	hash = mpls_multipath_hash(rt, skb, bos);
184 	nh_index = hash % alive;
185 	if (alive == rt->rt_nhn)
186 		goto out;
187 	for_nexthops(rt) {
188 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
189 			continue;
190 		if (n == nh_index)
191 			return nh;
192 		n++;
193 	} endfor_nexthops(rt);
194 
195 out:
196 	return &rt->rt_nh[nh_index];
197 }
198 
199 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
200 			struct mpls_entry_decoded dec)
201 {
202 	enum mpls_payload_type payload_type;
203 	bool success = false;
204 
205 	/* The IPv4 code below accesses through the IPv4 header
206 	 * checksum, which is 12 bytes into the packet.
207 	 * The IPv6 code below accesses through the IPv6 hop limit
208 	 * which is 8 bytes into the packet.
209 	 *
210 	 * For all supported cases there should always be at least 12
211 	 * bytes of packet data present.  The IPv4 header is 20 bytes
212 	 * without options and the IPv6 header is always 40 bytes
213 	 * long.
214 	 */
215 	if (!pskb_may_pull(skb, 12))
216 		return false;
217 
218 	payload_type = rt->rt_payload_type;
219 	if (payload_type == MPT_UNSPEC)
220 		payload_type = ip_hdr(skb)->version;
221 
222 	switch (payload_type) {
223 	case MPT_IPV4: {
224 		struct iphdr *hdr4 = ip_hdr(skb);
225 		skb->protocol = htons(ETH_P_IP);
226 		csum_replace2(&hdr4->check,
227 			      htons(hdr4->ttl << 8),
228 			      htons(dec.ttl << 8));
229 		hdr4->ttl = dec.ttl;
230 		success = true;
231 		break;
232 	}
233 	case MPT_IPV6: {
234 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
235 		skb->protocol = htons(ETH_P_IPV6);
236 		hdr6->hop_limit = dec.ttl;
237 		success = true;
238 		break;
239 	}
240 	case MPT_UNSPEC:
241 		break;
242 	}
243 
244 	return success;
245 }
246 
247 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
248 			struct packet_type *pt, struct net_device *orig_dev)
249 {
250 	struct net *net = dev_net(dev);
251 	struct mpls_shim_hdr *hdr;
252 	struct mpls_route *rt;
253 	struct mpls_nh *nh;
254 	struct mpls_entry_decoded dec;
255 	struct net_device *out_dev;
256 	struct mpls_dev *mdev;
257 	unsigned int hh_len;
258 	unsigned int new_header_size;
259 	unsigned int mtu;
260 	int err;
261 
262 	/* Careful this entire function runs inside of an rcu critical section */
263 
264 	mdev = mpls_dev_get(dev);
265 	if (!mdev || !mdev->input_enabled)
266 		goto drop;
267 
268 	if (skb->pkt_type != PACKET_HOST)
269 		goto drop;
270 
271 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
272 		goto drop;
273 
274 	if (!pskb_may_pull(skb, sizeof(*hdr)))
275 		goto drop;
276 
277 	/* Read and decode the label */
278 	hdr = mpls_hdr(skb);
279 	dec = mpls_entry_decode(hdr);
280 
281 	/* Pop the label */
282 	skb_pull(skb, sizeof(*hdr));
283 	skb_reset_network_header(skb);
284 
285 	skb_orphan(skb);
286 
287 	rt = mpls_route_input_rcu(net, dec.label);
288 	if (!rt)
289 		goto drop;
290 
291 	nh = mpls_select_multipath(rt, skb, dec.bos);
292 	if (!nh)
293 		goto drop;
294 
295 	/* Find the output device */
296 	out_dev = rcu_dereference(nh->nh_dev);
297 	if (!mpls_output_possible(out_dev))
298 		goto drop;
299 
300 	if (skb_warn_if_lro(skb))
301 		goto drop;
302 
303 	skb_forward_csum(skb);
304 
305 	/* Verify ttl is valid */
306 	if (dec.ttl <= 1)
307 		goto drop;
308 	dec.ttl -= 1;
309 
310 	/* Verify the destination can hold the packet */
311 	new_header_size = mpls_nh_header_size(nh);
312 	mtu = mpls_dev_mtu(out_dev);
313 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
314 		goto drop;
315 
316 	hh_len = LL_RESERVED_SPACE(out_dev);
317 	if (!out_dev->header_ops)
318 		hh_len = 0;
319 
320 	/* Ensure there is enough space for the headers in the skb */
321 	if (skb_cow(skb, hh_len + new_header_size))
322 		goto drop;
323 
324 	skb->dev = out_dev;
325 	skb->protocol = htons(ETH_P_MPLS_UC);
326 
327 	if (unlikely(!new_header_size && dec.bos)) {
328 		/* Penultimate hop popping */
329 		if (!mpls_egress(rt, skb, dec))
330 			goto drop;
331 	} else {
332 		bool bos;
333 		int i;
334 		skb_push(skb, new_header_size);
335 		skb_reset_network_header(skb);
336 		/* Push the new labels */
337 		hdr = mpls_hdr(skb);
338 		bos = dec.bos;
339 		for (i = nh->nh_labels - 1; i >= 0; i--) {
340 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
341 						   dec.ttl, 0, bos);
342 			bos = false;
343 		}
344 	}
345 
346 	/* If via wasn't specified then send out using device address */
347 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
348 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
349 				 out_dev->dev_addr, skb);
350 	else
351 		err = neigh_xmit(nh->nh_via_table, out_dev,
352 				 mpls_nh_via(rt, nh), skb);
353 	if (err)
354 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
355 				    __func__, err);
356 	return 0;
357 
358 drop:
359 	kfree_skb(skb);
360 	return NET_RX_DROP;
361 }
362 
363 static struct packet_type mpls_packet_type __read_mostly = {
364 	.type = cpu_to_be16(ETH_P_MPLS_UC),
365 	.func = mpls_forward,
366 };
367 
368 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
369 	[RTA_DST]		= { .type = NLA_U32 },
370 	[RTA_OIF]		= { .type = NLA_U32 },
371 };
372 
373 struct mpls_route_config {
374 	u32			rc_protocol;
375 	u32			rc_ifindex;
376 	u8			rc_via_table;
377 	u8			rc_via_alen;
378 	u8			rc_via[MAX_VIA_ALEN];
379 	u32			rc_label;
380 	u8			rc_output_labels;
381 	u32			rc_output_label[MAX_NEW_LABELS];
382 	u32			rc_nlflags;
383 	enum mpls_payload_type	rc_payload_type;
384 	struct nl_info		rc_nlinfo;
385 	struct rtnexthop	*rc_mp;
386 	int			rc_mp_len;
387 };
388 
389 static struct mpls_route *mpls_rt_alloc(int num_nh, u8 max_alen)
390 {
391 	u8 max_alen_aligned = ALIGN(max_alen, VIA_ALEN_ALIGN);
392 	struct mpls_route *rt;
393 
394 	rt = kzalloc(ALIGN(sizeof(*rt) + num_nh * sizeof(*rt->rt_nh),
395 			   VIA_ALEN_ALIGN) +
396 		     num_nh * max_alen_aligned,
397 		     GFP_KERNEL);
398 	if (rt) {
399 		rt->rt_nhn = num_nh;
400 		rt->rt_nhn_alive = num_nh;
401 		rt->rt_max_alen = max_alen_aligned;
402 	}
403 
404 	return rt;
405 }
406 
407 static void mpls_rt_free(struct mpls_route *rt)
408 {
409 	if (rt)
410 		kfree_rcu(rt, rt_rcu);
411 }
412 
413 static void mpls_notify_route(struct net *net, unsigned index,
414 			      struct mpls_route *old, struct mpls_route *new,
415 			      const struct nl_info *info)
416 {
417 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
418 	unsigned portid = info ? info->portid : 0;
419 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
420 	struct mpls_route *rt = new ? new : old;
421 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
422 	/* Ignore reserved labels for now */
423 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
424 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
425 }
426 
427 static void mpls_route_update(struct net *net, unsigned index,
428 			      struct mpls_route *new,
429 			      const struct nl_info *info)
430 {
431 	struct mpls_route __rcu **platform_label;
432 	struct mpls_route *rt;
433 
434 	ASSERT_RTNL();
435 
436 	platform_label = rtnl_dereference(net->mpls.platform_label);
437 	rt = rtnl_dereference(platform_label[index]);
438 	rcu_assign_pointer(platform_label[index], new);
439 
440 	mpls_notify_route(net, index, rt, new, info);
441 
442 	/* If we removed a route free it now */
443 	mpls_rt_free(rt);
444 }
445 
446 static unsigned find_free_label(struct net *net)
447 {
448 	struct mpls_route __rcu **platform_label;
449 	size_t platform_labels;
450 	unsigned index;
451 
452 	platform_label = rtnl_dereference(net->mpls.platform_label);
453 	platform_labels = net->mpls.platform_labels;
454 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
455 	     index++) {
456 		if (!rtnl_dereference(platform_label[index]))
457 			return index;
458 	}
459 	return LABEL_NOT_SPECIFIED;
460 }
461 
462 #if IS_ENABLED(CONFIG_INET)
463 static struct net_device *inet_fib_lookup_dev(struct net *net,
464 					      const void *addr)
465 {
466 	struct net_device *dev;
467 	struct rtable *rt;
468 	struct in_addr daddr;
469 
470 	memcpy(&daddr, addr, sizeof(struct in_addr));
471 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
472 	if (IS_ERR(rt))
473 		return ERR_CAST(rt);
474 
475 	dev = rt->dst.dev;
476 	dev_hold(dev);
477 
478 	ip_rt_put(rt);
479 
480 	return dev;
481 }
482 #else
483 static struct net_device *inet_fib_lookup_dev(struct net *net,
484 					      const void *addr)
485 {
486 	return ERR_PTR(-EAFNOSUPPORT);
487 }
488 #endif
489 
490 #if IS_ENABLED(CONFIG_IPV6)
491 static struct net_device *inet6_fib_lookup_dev(struct net *net,
492 					       const void *addr)
493 {
494 	struct net_device *dev;
495 	struct dst_entry *dst;
496 	struct flowi6 fl6;
497 	int err;
498 
499 	if (!ipv6_stub)
500 		return ERR_PTR(-EAFNOSUPPORT);
501 
502 	memset(&fl6, 0, sizeof(fl6));
503 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
504 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
505 	if (err)
506 		return ERR_PTR(err);
507 
508 	dev = dst->dev;
509 	dev_hold(dev);
510 	dst_release(dst);
511 
512 	return dev;
513 }
514 #else
515 static struct net_device *inet6_fib_lookup_dev(struct net *net,
516 					       const void *addr)
517 {
518 	return ERR_PTR(-EAFNOSUPPORT);
519 }
520 #endif
521 
522 static struct net_device *find_outdev(struct net *net,
523 				      struct mpls_route *rt,
524 				      struct mpls_nh *nh, int oif)
525 {
526 	struct net_device *dev = NULL;
527 
528 	if (!oif) {
529 		switch (nh->nh_via_table) {
530 		case NEIGH_ARP_TABLE:
531 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
532 			break;
533 		case NEIGH_ND_TABLE:
534 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
535 			break;
536 		case NEIGH_LINK_TABLE:
537 			break;
538 		}
539 	} else {
540 		dev = dev_get_by_index(net, oif);
541 	}
542 
543 	if (!dev)
544 		return ERR_PTR(-ENODEV);
545 
546 	if (IS_ERR(dev))
547 		return dev;
548 
549 	/* The caller is holding rtnl anyways, so release the dev reference */
550 	dev_put(dev);
551 
552 	return dev;
553 }
554 
555 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
556 			      struct mpls_nh *nh, int oif)
557 {
558 	struct net_device *dev = NULL;
559 	int err = -ENODEV;
560 
561 	dev = find_outdev(net, rt, nh, oif);
562 	if (IS_ERR(dev)) {
563 		err = PTR_ERR(dev);
564 		dev = NULL;
565 		goto errout;
566 	}
567 
568 	/* Ensure this is a supported device */
569 	err = -EINVAL;
570 	if (!mpls_dev_get(dev))
571 		goto errout;
572 
573 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
574 	    (dev->addr_len != nh->nh_via_alen))
575 		goto errout;
576 
577 	RCU_INIT_POINTER(nh->nh_dev, dev);
578 
579 	if (!(dev->flags & IFF_UP)) {
580 		nh->nh_flags |= RTNH_F_DEAD;
581 	} else {
582 		unsigned int flags;
583 
584 		flags = dev_get_flags(dev);
585 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
586 			nh->nh_flags |= RTNH_F_LINKDOWN;
587 	}
588 
589 	return 0;
590 
591 errout:
592 	return err;
593 }
594 
595 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
596 				  struct mpls_route *rt)
597 {
598 	struct net *net = cfg->rc_nlinfo.nl_net;
599 	struct mpls_nh *nh = rt->rt_nh;
600 	int err;
601 	int i;
602 
603 	if (!nh)
604 		return -ENOMEM;
605 
606 	err = -EINVAL;
607 	/* Ensure only a supported number of labels are present */
608 	if (cfg->rc_output_labels > MAX_NEW_LABELS)
609 		goto errout;
610 
611 	nh->nh_labels = cfg->rc_output_labels;
612 	for (i = 0; i < nh->nh_labels; i++)
613 		nh->nh_label[i] = cfg->rc_output_label[i];
614 
615 	nh->nh_via_table = cfg->rc_via_table;
616 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
617 	nh->nh_via_alen = cfg->rc_via_alen;
618 
619 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
620 	if (err)
621 		goto errout;
622 
623 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
624 		rt->rt_nhn_alive--;
625 
626 	return 0;
627 
628 errout:
629 	return err;
630 }
631 
632 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
633 			 struct mpls_nh *nh, int oif, struct nlattr *via,
634 			 struct nlattr *newdst)
635 {
636 	int err = -ENOMEM;
637 
638 	if (!nh)
639 		goto errout;
640 
641 	if (newdst) {
642 		err = nla_get_labels(newdst, MAX_NEW_LABELS,
643 				     &nh->nh_labels, nh->nh_label);
644 		if (err)
645 			goto errout;
646 	}
647 
648 	if (via) {
649 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
650 				  __mpls_nh_via(rt, nh));
651 		if (err)
652 			goto errout;
653 	} else {
654 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
655 	}
656 
657 	err = mpls_nh_assign_dev(net, rt, nh, oif);
658 	if (err)
659 		goto errout;
660 
661 	return 0;
662 
663 errout:
664 	return err;
665 }
666 
667 static int mpls_count_nexthops(struct rtnexthop *rtnh, int len,
668 			       u8 cfg_via_alen, u8 *max_via_alen)
669 {
670 	int nhs = 0;
671 	int remaining = len;
672 
673 	if (!rtnh) {
674 		*max_via_alen = cfg_via_alen;
675 		return 1;
676 	}
677 
678 	*max_via_alen = 0;
679 
680 	while (rtnh_ok(rtnh, remaining)) {
681 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
682 		int attrlen;
683 
684 		attrlen = rtnh_attrlen(rtnh);
685 		nla = nla_find(attrs, attrlen, RTA_VIA);
686 		if (nla && nla_len(nla) >=
687 		    offsetof(struct rtvia, rtvia_addr)) {
688 			int via_alen = nla_len(nla) -
689 				offsetof(struct rtvia, rtvia_addr);
690 
691 			if (via_alen <= MAX_VIA_ALEN)
692 				*max_via_alen = max_t(u16, *max_via_alen,
693 						      via_alen);
694 		}
695 
696 		nhs++;
697 		rtnh = rtnh_next(rtnh, &remaining);
698 	}
699 
700 	/* leftover implies invalid nexthop configuration, discard it */
701 	return remaining > 0 ? 0 : nhs;
702 }
703 
704 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
705 			       struct mpls_route *rt)
706 {
707 	struct rtnexthop *rtnh = cfg->rc_mp;
708 	struct nlattr *nla_via, *nla_newdst;
709 	int remaining = cfg->rc_mp_len;
710 	int nhs = 0;
711 	int err = 0;
712 
713 	change_nexthops(rt) {
714 		int attrlen;
715 
716 		nla_via = NULL;
717 		nla_newdst = NULL;
718 
719 		err = -EINVAL;
720 		if (!rtnh_ok(rtnh, remaining))
721 			goto errout;
722 
723 		/* neither weighted multipath nor any flags
724 		 * are supported
725 		 */
726 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
727 			goto errout;
728 
729 		attrlen = rtnh_attrlen(rtnh);
730 		if (attrlen > 0) {
731 			struct nlattr *attrs = rtnh_attrs(rtnh);
732 
733 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
734 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
735 		}
736 
737 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
738 				    rtnh->rtnh_ifindex, nla_via, nla_newdst);
739 		if (err)
740 			goto errout;
741 
742 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
743 			rt->rt_nhn_alive--;
744 
745 		rtnh = rtnh_next(rtnh, &remaining);
746 		nhs++;
747 	} endfor_nexthops(rt);
748 
749 	rt->rt_nhn = nhs;
750 
751 	return 0;
752 
753 errout:
754 	return err;
755 }
756 
757 static int mpls_route_add(struct mpls_route_config *cfg)
758 {
759 	struct mpls_route __rcu **platform_label;
760 	struct net *net = cfg->rc_nlinfo.nl_net;
761 	struct mpls_route *rt, *old;
762 	int err = -EINVAL;
763 	u8 max_via_alen;
764 	unsigned index;
765 	int nhs;
766 
767 	index = cfg->rc_label;
768 
769 	/* If a label was not specified during insert pick one */
770 	if ((index == LABEL_NOT_SPECIFIED) &&
771 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
772 		index = find_free_label(net);
773 	}
774 
775 	/* Reserved labels may not be set */
776 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
777 		goto errout;
778 
779 	/* The full 20 bit range may not be supported. */
780 	if (index >= net->mpls.platform_labels)
781 		goto errout;
782 
783 	/* Append makes no sense with mpls */
784 	err = -EOPNOTSUPP;
785 	if (cfg->rc_nlflags & NLM_F_APPEND)
786 		goto errout;
787 
788 	err = -EEXIST;
789 	platform_label = rtnl_dereference(net->mpls.platform_label);
790 	old = rtnl_dereference(platform_label[index]);
791 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
792 		goto errout;
793 
794 	err = -EEXIST;
795 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
796 		goto errout;
797 
798 	err = -ENOENT;
799 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
800 		goto errout;
801 
802 	err = -EINVAL;
803 	nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
804 				  cfg->rc_via_alen, &max_via_alen);
805 	if (nhs == 0)
806 		goto errout;
807 
808 	err = -ENOMEM;
809 	rt = mpls_rt_alloc(nhs, max_via_alen);
810 	if (!rt)
811 		goto errout;
812 
813 	rt->rt_protocol = cfg->rc_protocol;
814 	rt->rt_payload_type = cfg->rc_payload_type;
815 
816 	if (cfg->rc_mp)
817 		err = mpls_nh_build_multi(cfg, rt);
818 	else
819 		err = mpls_nh_build_from_cfg(cfg, rt);
820 	if (err)
821 		goto freert;
822 
823 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
824 
825 	return 0;
826 
827 freert:
828 	mpls_rt_free(rt);
829 errout:
830 	return err;
831 }
832 
833 static int mpls_route_del(struct mpls_route_config *cfg)
834 {
835 	struct net *net = cfg->rc_nlinfo.nl_net;
836 	unsigned index;
837 	int err = -EINVAL;
838 
839 	index = cfg->rc_label;
840 
841 	/* Reserved labels may not be removed */
842 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
843 		goto errout;
844 
845 	/* The full 20 bit range may not be supported */
846 	if (index >= net->mpls.platform_labels)
847 		goto errout;
848 
849 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
850 
851 	err = 0;
852 errout:
853 	return err;
854 }
855 
856 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
857 	(&((struct mpls_dev *)0)->field)
858 
859 static const struct ctl_table mpls_dev_table[] = {
860 	{
861 		.procname	= "input",
862 		.maxlen		= sizeof(int),
863 		.mode		= 0644,
864 		.proc_handler	= proc_dointvec,
865 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
866 	},
867 	{ }
868 };
869 
870 static int mpls_dev_sysctl_register(struct net_device *dev,
871 				    struct mpls_dev *mdev)
872 {
873 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
874 	struct ctl_table *table;
875 	int i;
876 
877 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
878 	if (!table)
879 		goto out;
880 
881 	/* Table data contains only offsets relative to the base of
882 	 * the mdev at this point, so make them absolute.
883 	 */
884 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
885 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
886 
887 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
888 
889 	mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
890 	if (!mdev->sysctl)
891 		goto free;
892 
893 	return 0;
894 
895 free:
896 	kfree(table);
897 out:
898 	return -ENOBUFS;
899 }
900 
901 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
902 {
903 	struct ctl_table *table;
904 
905 	table = mdev->sysctl->ctl_table_arg;
906 	unregister_net_sysctl_table(mdev->sysctl);
907 	kfree(table);
908 }
909 
910 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
911 {
912 	struct mpls_dev *mdev;
913 	int err = -ENOMEM;
914 
915 	ASSERT_RTNL();
916 
917 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
918 	if (!mdev)
919 		return ERR_PTR(err);
920 
921 	err = mpls_dev_sysctl_register(dev, mdev);
922 	if (err)
923 		goto free;
924 
925 	rcu_assign_pointer(dev->mpls_ptr, mdev);
926 
927 	return mdev;
928 
929 free:
930 	kfree(mdev);
931 	return ERR_PTR(err);
932 }
933 
934 static void mpls_ifdown(struct net_device *dev, int event)
935 {
936 	struct mpls_route __rcu **platform_label;
937 	struct net *net = dev_net(dev);
938 	unsigned index;
939 
940 	platform_label = rtnl_dereference(net->mpls.platform_label);
941 	for (index = 0; index < net->mpls.platform_labels; index++) {
942 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
943 
944 		if (!rt)
945 			continue;
946 
947 		change_nexthops(rt) {
948 			if (rtnl_dereference(nh->nh_dev) != dev)
949 				continue;
950 			switch (event) {
951 			case NETDEV_DOWN:
952 			case NETDEV_UNREGISTER:
953 				nh->nh_flags |= RTNH_F_DEAD;
954 				/* fall through */
955 			case NETDEV_CHANGE:
956 				nh->nh_flags |= RTNH_F_LINKDOWN;
957 				ACCESS_ONCE(rt->rt_nhn_alive) = rt->rt_nhn_alive - 1;
958 				break;
959 			}
960 			if (event == NETDEV_UNREGISTER)
961 				RCU_INIT_POINTER(nh->nh_dev, NULL);
962 		} endfor_nexthops(rt);
963 	}
964 
965 
966 	return;
967 }
968 
969 static void mpls_ifup(struct net_device *dev, unsigned int nh_flags)
970 {
971 	struct mpls_route __rcu **platform_label;
972 	struct net *net = dev_net(dev);
973 	unsigned index;
974 	int alive;
975 
976 	platform_label = rtnl_dereference(net->mpls.platform_label);
977 	for (index = 0; index < net->mpls.platform_labels; index++) {
978 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
979 
980 		if (!rt)
981 			continue;
982 
983 		alive = 0;
984 		change_nexthops(rt) {
985 			struct net_device *nh_dev =
986 				rtnl_dereference(nh->nh_dev);
987 
988 			if (!(nh->nh_flags & nh_flags)) {
989 				alive++;
990 				continue;
991 			}
992 			if (nh_dev != dev)
993 				continue;
994 			alive++;
995 			nh->nh_flags &= ~nh_flags;
996 		} endfor_nexthops(rt);
997 
998 		ACCESS_ONCE(rt->rt_nhn_alive) = alive;
999 	}
1000 
1001 	return;
1002 }
1003 
1004 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1005 			   void *ptr)
1006 {
1007 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1008 	struct mpls_dev *mdev;
1009 	unsigned int flags;
1010 
1011 	if (event == NETDEV_REGISTER) {
1012 		/* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1013 		if (dev->type == ARPHRD_ETHER ||
1014 		    dev->type == ARPHRD_LOOPBACK ||
1015 		    dev->type == ARPHRD_IPGRE ||
1016 		    dev->type == ARPHRD_SIT ||
1017 		    dev->type == ARPHRD_TUNNEL) {
1018 			mdev = mpls_add_dev(dev);
1019 			if (IS_ERR(mdev))
1020 				return notifier_from_errno(PTR_ERR(mdev));
1021 		}
1022 		return NOTIFY_OK;
1023 	}
1024 
1025 	mdev = mpls_dev_get(dev);
1026 	if (!mdev)
1027 		return NOTIFY_OK;
1028 
1029 	switch (event) {
1030 	case NETDEV_DOWN:
1031 		mpls_ifdown(dev, event);
1032 		break;
1033 	case NETDEV_UP:
1034 		flags = dev_get_flags(dev);
1035 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1036 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1037 		else
1038 			mpls_ifup(dev, RTNH_F_DEAD);
1039 		break;
1040 	case NETDEV_CHANGE:
1041 		flags = dev_get_flags(dev);
1042 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1043 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1044 		else
1045 			mpls_ifdown(dev, event);
1046 		break;
1047 	case NETDEV_UNREGISTER:
1048 		mpls_ifdown(dev, event);
1049 		mdev = mpls_dev_get(dev);
1050 		if (mdev) {
1051 			mpls_dev_sysctl_unregister(mdev);
1052 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1053 			kfree_rcu(mdev, rcu);
1054 		}
1055 		break;
1056 	case NETDEV_CHANGENAME:
1057 		mdev = mpls_dev_get(dev);
1058 		if (mdev) {
1059 			int err;
1060 
1061 			mpls_dev_sysctl_unregister(mdev);
1062 			err = mpls_dev_sysctl_register(dev, mdev);
1063 			if (err)
1064 				return notifier_from_errno(err);
1065 		}
1066 		break;
1067 	}
1068 	return NOTIFY_OK;
1069 }
1070 
1071 static struct notifier_block mpls_dev_notifier = {
1072 	.notifier_call = mpls_dev_notify,
1073 };
1074 
1075 static int nla_put_via(struct sk_buff *skb,
1076 		       u8 table, const void *addr, int alen)
1077 {
1078 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1079 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1080 	};
1081 	struct nlattr *nla;
1082 	struct rtvia *via;
1083 	int family = AF_UNSPEC;
1084 
1085 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1086 	if (!nla)
1087 		return -EMSGSIZE;
1088 
1089 	if (table <= NEIGH_NR_TABLES)
1090 		family = table_to_family[table];
1091 
1092 	via = nla_data(nla);
1093 	via->rtvia_family = family;
1094 	memcpy(via->rtvia_addr, addr, alen);
1095 	return 0;
1096 }
1097 
1098 int nla_put_labels(struct sk_buff *skb, int attrtype,
1099 		   u8 labels, const u32 label[])
1100 {
1101 	struct nlattr *nla;
1102 	struct mpls_shim_hdr *nla_label;
1103 	bool bos;
1104 	int i;
1105 	nla = nla_reserve(skb, attrtype, labels*4);
1106 	if (!nla)
1107 		return -EMSGSIZE;
1108 
1109 	nla_label = nla_data(nla);
1110 	bos = true;
1111 	for (i = labels - 1; i >= 0; i--) {
1112 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1113 		bos = false;
1114 	}
1115 
1116 	return 0;
1117 }
1118 EXPORT_SYMBOL_GPL(nla_put_labels);
1119 
1120 int nla_get_labels(const struct nlattr *nla,
1121 		   u32 max_labels, u8 *labels, u32 label[])
1122 {
1123 	unsigned len = nla_len(nla);
1124 	unsigned nla_labels;
1125 	struct mpls_shim_hdr *nla_label;
1126 	bool bos;
1127 	int i;
1128 
1129 	/* len needs to be an even multiple of 4 (the label size) */
1130 	if (len & 3)
1131 		return -EINVAL;
1132 
1133 	/* Limit the number of new labels allowed */
1134 	nla_labels = len/4;
1135 	if (nla_labels > max_labels)
1136 		return -EINVAL;
1137 
1138 	nla_label = nla_data(nla);
1139 	bos = true;
1140 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1141 		struct mpls_entry_decoded dec;
1142 		dec = mpls_entry_decode(nla_label + i);
1143 
1144 		/* Ensure the bottom of stack flag is properly set
1145 		 * and ttl and tc are both clear.
1146 		 */
1147 		if ((dec.bos != bos) || dec.ttl || dec.tc)
1148 			return -EINVAL;
1149 
1150 		switch (dec.label) {
1151 		case MPLS_LABEL_IMPLNULL:
1152 			/* RFC3032: This is a label that an LSR may
1153 			 * assign and distribute, but which never
1154 			 * actually appears in the encapsulation.
1155 			 */
1156 			return -EINVAL;
1157 		}
1158 
1159 		label[i] = dec.label;
1160 	}
1161 	*labels = nla_labels;
1162 	return 0;
1163 }
1164 EXPORT_SYMBOL_GPL(nla_get_labels);
1165 
1166 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1167 		u8 *via_table, u8 via_addr[])
1168 {
1169 	struct rtvia *via = nla_data(nla);
1170 	int err = -EINVAL;
1171 	int alen;
1172 
1173 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1174 		goto errout;
1175 	alen = nla_len(nla) -
1176 			offsetof(struct rtvia, rtvia_addr);
1177 	if (alen > MAX_VIA_ALEN)
1178 		goto errout;
1179 
1180 	/* Validate the address family */
1181 	switch (via->rtvia_family) {
1182 	case AF_PACKET:
1183 		*via_table = NEIGH_LINK_TABLE;
1184 		break;
1185 	case AF_INET:
1186 		*via_table = NEIGH_ARP_TABLE;
1187 		if (alen != 4)
1188 			goto errout;
1189 		break;
1190 	case AF_INET6:
1191 		*via_table = NEIGH_ND_TABLE;
1192 		if (alen != 16)
1193 			goto errout;
1194 		break;
1195 	default:
1196 		/* Unsupported address family */
1197 		goto errout;
1198 	}
1199 
1200 	memcpy(via_addr, via->rtvia_addr, alen);
1201 	*via_alen = alen;
1202 	err = 0;
1203 
1204 errout:
1205 	return err;
1206 }
1207 
1208 static int rtm_to_route_config(struct sk_buff *skb,  struct nlmsghdr *nlh,
1209 			       struct mpls_route_config *cfg)
1210 {
1211 	struct rtmsg *rtm;
1212 	struct nlattr *tb[RTA_MAX+1];
1213 	int index;
1214 	int err;
1215 
1216 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1217 	if (err < 0)
1218 		goto errout;
1219 
1220 	err = -EINVAL;
1221 	rtm = nlmsg_data(nlh);
1222 	memset(cfg, 0, sizeof(*cfg));
1223 
1224 	if (rtm->rtm_family != AF_MPLS)
1225 		goto errout;
1226 	if (rtm->rtm_dst_len != 20)
1227 		goto errout;
1228 	if (rtm->rtm_src_len != 0)
1229 		goto errout;
1230 	if (rtm->rtm_tos != 0)
1231 		goto errout;
1232 	if (rtm->rtm_table != RT_TABLE_MAIN)
1233 		goto errout;
1234 	/* Any value is acceptable for rtm_protocol */
1235 
1236 	/* As mpls uses destination specific addresses
1237 	 * (or source specific address in the case of multicast)
1238 	 * all addresses have universal scope.
1239 	 */
1240 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1241 		goto errout;
1242 	if (rtm->rtm_type != RTN_UNICAST)
1243 		goto errout;
1244 	if (rtm->rtm_flags != 0)
1245 		goto errout;
1246 
1247 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1248 	cfg->rc_protocol	= rtm->rtm_protocol;
1249 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1250 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1251 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1252 	cfg->rc_nlinfo.nlh	= nlh;
1253 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1254 
1255 	for (index = 0; index <= RTA_MAX; index++) {
1256 		struct nlattr *nla = tb[index];
1257 		if (!nla)
1258 			continue;
1259 
1260 		switch(index) {
1261 		case RTA_OIF:
1262 			cfg->rc_ifindex = nla_get_u32(nla);
1263 			break;
1264 		case RTA_NEWDST:
1265 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1266 					   &cfg->rc_output_labels,
1267 					   cfg->rc_output_label))
1268 				goto errout;
1269 			break;
1270 		case RTA_DST:
1271 		{
1272 			u8 label_count;
1273 			if (nla_get_labels(nla, 1, &label_count,
1274 					   &cfg->rc_label))
1275 				goto errout;
1276 
1277 			/* Reserved labels may not be set */
1278 			if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1279 				goto errout;
1280 
1281 			break;
1282 		}
1283 		case RTA_VIA:
1284 		{
1285 			if (nla_get_via(nla, &cfg->rc_via_alen,
1286 					&cfg->rc_via_table, cfg->rc_via))
1287 				goto errout;
1288 			break;
1289 		}
1290 		case RTA_MULTIPATH:
1291 		{
1292 			cfg->rc_mp = nla_data(nla);
1293 			cfg->rc_mp_len = nla_len(nla);
1294 			break;
1295 		}
1296 		default:
1297 			/* Unsupported attribute */
1298 			goto errout;
1299 		}
1300 	}
1301 
1302 	err = 0;
1303 errout:
1304 	return err;
1305 }
1306 
1307 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1308 {
1309 	struct mpls_route_config cfg;
1310 	int err;
1311 
1312 	err = rtm_to_route_config(skb, nlh, &cfg);
1313 	if (err < 0)
1314 		return err;
1315 
1316 	return mpls_route_del(&cfg);
1317 }
1318 
1319 
1320 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1321 {
1322 	struct mpls_route_config cfg;
1323 	int err;
1324 
1325 	err = rtm_to_route_config(skb, nlh, &cfg);
1326 	if (err < 0)
1327 		return err;
1328 
1329 	return mpls_route_add(&cfg);
1330 }
1331 
1332 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1333 			   u32 label, struct mpls_route *rt, int flags)
1334 {
1335 	struct net_device *dev;
1336 	struct nlmsghdr *nlh;
1337 	struct rtmsg *rtm;
1338 
1339 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1340 	if (nlh == NULL)
1341 		return -EMSGSIZE;
1342 
1343 	rtm = nlmsg_data(nlh);
1344 	rtm->rtm_family = AF_MPLS;
1345 	rtm->rtm_dst_len = 20;
1346 	rtm->rtm_src_len = 0;
1347 	rtm->rtm_tos = 0;
1348 	rtm->rtm_table = RT_TABLE_MAIN;
1349 	rtm->rtm_protocol = rt->rt_protocol;
1350 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1351 	rtm->rtm_type = RTN_UNICAST;
1352 	rtm->rtm_flags = 0;
1353 
1354 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1355 		goto nla_put_failure;
1356 	if (rt->rt_nhn == 1) {
1357 		const struct mpls_nh *nh = rt->rt_nh;
1358 
1359 		if (nh->nh_labels &&
1360 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1361 				   nh->nh_label))
1362 			goto nla_put_failure;
1363 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1364 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1365 				nh->nh_via_alen))
1366 			goto nla_put_failure;
1367 		dev = rtnl_dereference(nh->nh_dev);
1368 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1369 			goto nla_put_failure;
1370 		if (nh->nh_flags & RTNH_F_LINKDOWN)
1371 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1372 		if (nh->nh_flags & RTNH_F_DEAD)
1373 			rtm->rtm_flags |= RTNH_F_DEAD;
1374 	} else {
1375 		struct rtnexthop *rtnh;
1376 		struct nlattr *mp;
1377 		int dead = 0;
1378 		int linkdown = 0;
1379 
1380 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1381 		if (!mp)
1382 			goto nla_put_failure;
1383 
1384 		for_nexthops(rt) {
1385 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1386 			if (!rtnh)
1387 				goto nla_put_failure;
1388 
1389 			dev = rtnl_dereference(nh->nh_dev);
1390 			if (dev)
1391 				rtnh->rtnh_ifindex = dev->ifindex;
1392 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1393 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1394 				linkdown++;
1395 			}
1396 			if (nh->nh_flags & RTNH_F_DEAD) {
1397 				rtnh->rtnh_flags |= RTNH_F_DEAD;
1398 				dead++;
1399 			}
1400 
1401 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1402 							    nh->nh_labels,
1403 							    nh->nh_label))
1404 				goto nla_put_failure;
1405 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1406 			    nla_put_via(skb, nh->nh_via_table,
1407 					mpls_nh_via(rt, nh),
1408 					nh->nh_via_alen))
1409 				goto nla_put_failure;
1410 
1411 			/* length of rtnetlink header + attributes */
1412 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1413 		} endfor_nexthops(rt);
1414 
1415 		if (linkdown == rt->rt_nhn)
1416 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1417 		if (dead == rt->rt_nhn)
1418 			rtm->rtm_flags |= RTNH_F_DEAD;
1419 
1420 		nla_nest_end(skb, mp);
1421 	}
1422 
1423 	nlmsg_end(skb, nlh);
1424 	return 0;
1425 
1426 nla_put_failure:
1427 	nlmsg_cancel(skb, nlh);
1428 	return -EMSGSIZE;
1429 }
1430 
1431 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1432 {
1433 	struct net *net = sock_net(skb->sk);
1434 	struct mpls_route __rcu **platform_label;
1435 	size_t platform_labels;
1436 	unsigned int index;
1437 
1438 	ASSERT_RTNL();
1439 
1440 	index = cb->args[0];
1441 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
1442 		index = MPLS_LABEL_FIRST_UNRESERVED;
1443 
1444 	platform_label = rtnl_dereference(net->mpls.platform_label);
1445 	platform_labels = net->mpls.platform_labels;
1446 	for (; index < platform_labels; index++) {
1447 		struct mpls_route *rt;
1448 		rt = rtnl_dereference(platform_label[index]);
1449 		if (!rt)
1450 			continue;
1451 
1452 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1453 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1454 				    index, rt, NLM_F_MULTI) < 0)
1455 			break;
1456 	}
1457 	cb->args[0] = index;
1458 
1459 	return skb->len;
1460 }
1461 
1462 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1463 {
1464 	size_t payload =
1465 		NLMSG_ALIGN(sizeof(struct rtmsg))
1466 		+ nla_total_size(4);			/* RTA_DST */
1467 
1468 	if (rt->rt_nhn == 1) {
1469 		struct mpls_nh *nh = rt->rt_nh;
1470 
1471 		if (nh->nh_dev)
1472 			payload += nla_total_size(4); /* RTA_OIF */
1473 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1474 			payload += nla_total_size(2 + nh->nh_via_alen);
1475 		if (nh->nh_labels) /* RTA_NEWDST */
1476 			payload += nla_total_size(nh->nh_labels * 4);
1477 	} else {
1478 		/* each nexthop is packed in an attribute */
1479 		size_t nhsize = 0;
1480 
1481 		for_nexthops(rt) {
1482 			nhsize += nla_total_size(sizeof(struct rtnexthop));
1483 			/* RTA_VIA */
1484 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1485 				nhsize += nla_total_size(2 + nh->nh_via_alen);
1486 			if (nh->nh_labels)
1487 				nhsize += nla_total_size(nh->nh_labels * 4);
1488 		} endfor_nexthops(rt);
1489 		/* nested attribute */
1490 		payload += nla_total_size(nhsize);
1491 	}
1492 
1493 	return payload;
1494 }
1495 
1496 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1497 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
1498 		       unsigned int nlm_flags)
1499 {
1500 	struct sk_buff *skb;
1501 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
1502 	int err = -ENOBUFS;
1503 
1504 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1505 	if (skb == NULL)
1506 		goto errout;
1507 
1508 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1509 	if (err < 0) {
1510 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1511 		WARN_ON(err == -EMSGSIZE);
1512 		kfree_skb(skb);
1513 		goto errout;
1514 	}
1515 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1516 
1517 	return;
1518 errout:
1519 	if (err < 0)
1520 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1521 }
1522 
1523 static int resize_platform_label_table(struct net *net, size_t limit)
1524 {
1525 	size_t size = sizeof(struct mpls_route *) * limit;
1526 	size_t old_limit;
1527 	size_t cp_size;
1528 	struct mpls_route __rcu **labels = NULL, **old;
1529 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
1530 	unsigned index;
1531 
1532 	if (size) {
1533 		labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
1534 		if (!labels)
1535 			labels = vzalloc(size);
1536 
1537 		if (!labels)
1538 			goto nolabels;
1539 	}
1540 
1541 	/* In case the predefined labels need to be populated */
1542 	if (limit > MPLS_LABEL_IPV4NULL) {
1543 		struct net_device *lo = net->loopback_dev;
1544 		rt0 = mpls_rt_alloc(1, lo->addr_len);
1545 		if (!rt0)
1546 			goto nort0;
1547 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
1548 		rt0->rt_protocol = RTPROT_KERNEL;
1549 		rt0->rt_payload_type = MPT_IPV4;
1550 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1551 		rt0->rt_nh->nh_via_alen = lo->addr_len;
1552 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
1553 		       lo->addr_len);
1554 	}
1555 	if (limit > MPLS_LABEL_IPV6NULL) {
1556 		struct net_device *lo = net->loopback_dev;
1557 		rt2 = mpls_rt_alloc(1, lo->addr_len);
1558 		if (!rt2)
1559 			goto nort2;
1560 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
1561 		rt2->rt_protocol = RTPROT_KERNEL;
1562 		rt2->rt_payload_type = MPT_IPV6;
1563 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
1564 		rt2->rt_nh->nh_via_alen = lo->addr_len;
1565 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
1566 		       lo->addr_len);
1567 	}
1568 
1569 	rtnl_lock();
1570 	/* Remember the original table */
1571 	old = rtnl_dereference(net->mpls.platform_label);
1572 	old_limit = net->mpls.platform_labels;
1573 
1574 	/* Free any labels beyond the new table */
1575 	for (index = limit; index < old_limit; index++)
1576 		mpls_route_update(net, index, NULL, NULL);
1577 
1578 	/* Copy over the old labels */
1579 	cp_size = size;
1580 	if (old_limit < limit)
1581 		cp_size = old_limit * sizeof(struct mpls_route *);
1582 
1583 	memcpy(labels, old, cp_size);
1584 
1585 	/* If needed set the predefined labels */
1586 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
1587 	    (limit > MPLS_LABEL_IPV6NULL)) {
1588 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
1589 		rt2 = NULL;
1590 	}
1591 
1592 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
1593 	    (limit > MPLS_LABEL_IPV4NULL)) {
1594 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
1595 		rt0 = NULL;
1596 	}
1597 
1598 	/* Update the global pointers */
1599 	net->mpls.platform_labels = limit;
1600 	rcu_assign_pointer(net->mpls.platform_label, labels);
1601 
1602 	rtnl_unlock();
1603 
1604 	mpls_rt_free(rt2);
1605 	mpls_rt_free(rt0);
1606 
1607 	if (old) {
1608 		synchronize_rcu();
1609 		kvfree(old);
1610 	}
1611 	return 0;
1612 
1613 nort2:
1614 	mpls_rt_free(rt0);
1615 nort0:
1616 	kvfree(labels);
1617 nolabels:
1618 	return -ENOMEM;
1619 }
1620 
1621 static int mpls_platform_labels(struct ctl_table *table, int write,
1622 				void __user *buffer, size_t *lenp, loff_t *ppos)
1623 {
1624 	struct net *net = table->data;
1625 	int platform_labels = net->mpls.platform_labels;
1626 	int ret;
1627 	struct ctl_table tmp = {
1628 		.procname	= table->procname,
1629 		.data		= &platform_labels,
1630 		.maxlen		= sizeof(int),
1631 		.mode		= table->mode,
1632 		.extra1		= &zero,
1633 		.extra2		= &label_limit,
1634 	};
1635 
1636 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1637 
1638 	if (write && ret == 0)
1639 		ret = resize_platform_label_table(net, platform_labels);
1640 
1641 	return ret;
1642 }
1643 
1644 static const struct ctl_table mpls_table[] = {
1645 	{
1646 		.procname	= "platform_labels",
1647 		.data		= NULL,
1648 		.maxlen		= sizeof(int),
1649 		.mode		= 0644,
1650 		.proc_handler	= mpls_platform_labels,
1651 	},
1652 	{ }
1653 };
1654 
1655 static int mpls_net_init(struct net *net)
1656 {
1657 	struct ctl_table *table;
1658 
1659 	net->mpls.platform_labels = 0;
1660 	net->mpls.platform_label = NULL;
1661 
1662 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1663 	if (table == NULL)
1664 		return -ENOMEM;
1665 
1666 	table[0].data = net;
1667 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1668 	if (net->mpls.ctl == NULL) {
1669 		kfree(table);
1670 		return -ENOMEM;
1671 	}
1672 
1673 	return 0;
1674 }
1675 
1676 static void mpls_net_exit(struct net *net)
1677 {
1678 	struct mpls_route __rcu **platform_label;
1679 	size_t platform_labels;
1680 	struct ctl_table *table;
1681 	unsigned int index;
1682 
1683 	table = net->mpls.ctl->ctl_table_arg;
1684 	unregister_net_sysctl_table(net->mpls.ctl);
1685 	kfree(table);
1686 
1687 	/* An rcu grace period has passed since there was a device in
1688 	 * the network namespace (and thus the last in flight packet)
1689 	 * left this network namespace.  This is because
1690 	 * unregister_netdevice_many and netdev_run_todo has completed
1691 	 * for each network device that was in this network namespace.
1692 	 *
1693 	 * As such no additional rcu synchronization is necessary when
1694 	 * freeing the platform_label table.
1695 	 */
1696 	rtnl_lock();
1697 	platform_label = rtnl_dereference(net->mpls.platform_label);
1698 	platform_labels = net->mpls.platform_labels;
1699 	for (index = 0; index < platform_labels; index++) {
1700 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1701 		RCU_INIT_POINTER(platform_label[index], NULL);
1702 		mpls_rt_free(rt);
1703 	}
1704 	rtnl_unlock();
1705 
1706 	kvfree(platform_label);
1707 }
1708 
1709 static struct pernet_operations mpls_net_ops = {
1710 	.init = mpls_net_init,
1711 	.exit = mpls_net_exit,
1712 };
1713 
1714 static int __init mpls_init(void)
1715 {
1716 	int err;
1717 
1718 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1719 
1720 	err = register_pernet_subsys(&mpls_net_ops);
1721 	if (err)
1722 		goto out;
1723 
1724 	err = register_netdevice_notifier(&mpls_dev_notifier);
1725 	if (err)
1726 		goto out_unregister_pernet;
1727 
1728 	dev_add_pack(&mpls_packet_type);
1729 
1730 	rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1731 	rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1732 	rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1733 	err = 0;
1734 out:
1735 	return err;
1736 
1737 out_unregister_pernet:
1738 	unregister_pernet_subsys(&mpls_net_ops);
1739 	goto out;
1740 }
1741 module_init(mpls_init);
1742 
1743 static void __exit mpls_exit(void)
1744 {
1745 	rtnl_unregister_all(PF_MPLS);
1746 	dev_remove_pack(&mpls_packet_type);
1747 	unregister_netdevice_notifier(&mpls_dev_notifier);
1748 	unregister_pernet_subsys(&mpls_net_ops);
1749 }
1750 module_exit(mpls_exit);
1751 
1752 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1753 MODULE_LICENSE("GPL v2");
1754 MODULE_ALIAS_NETPROTO(PF_MPLS);
1755