xref: /linux/net/mpls/af_mpls.c (revision bc7ebc569e8cc768342dfb01af1a26c7fbef513e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/types.h>
3 #include <linux/skbuff.h>
4 #include <linux/socket.h>
5 #include <linux/sysctl.h>
6 #include <linux/net.h>
7 #include <linux/module.h>
8 #include <linux/if_arp.h>
9 #include <linux/ipv6.h>
10 #include <linux/mpls.h>
11 #include <linux/netconf.h>
12 #include <linux/nospec.h>
13 #include <linux/vmalloc.h>
14 #include <linux/percpu.h>
15 #include <net/gso.h>
16 #include <net/ip.h>
17 #include <net/dst.h>
18 #include <net/sock.h>
19 #include <net/arp.h>
20 #include <net/ip_fib.h>
21 #include <net/netevent.h>
22 #include <net/ip_tunnels.h>
23 #include <net/netns/generic.h>
24 #if IS_ENABLED(CONFIG_IPV6)
25 #include <net/ipv6.h>
26 #endif
27 #include <net/ipv6_stubs.h>
28 #include <net/rtnh.h>
29 #include "internal.h"
30 
31 /* max memory we will use for mpls_route */
32 #define MAX_MPLS_ROUTE_MEM	4096
33 
34 /* Maximum number of labels to look ahead at when selecting a path of
35  * a multipath route
36  */
37 #define MAX_MP_SELECT_LABELS 4
38 
39 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
40 
41 static int label_limit = (1 << 20) - 1;
42 static int ttl_max = 255;
43 
44 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
45 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
46 {
47 	return sizeof(struct mpls_shim_hdr);
48 }
49 
50 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
51 	.encap_hlen	= ipgre_mpls_encap_hlen,
52 };
53 
54 static int ipgre_tunnel_encap_add_mpls_ops(void)
55 {
56 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
57 }
58 
59 static void ipgre_tunnel_encap_del_mpls_ops(void)
60 {
61 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
62 }
63 #else
64 static int ipgre_tunnel_encap_add_mpls_ops(void)
65 {
66 	return 0;
67 }
68 
69 static void ipgre_tunnel_encap_del_mpls_ops(void)
70 {
71 }
72 #endif
73 
74 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
75 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
76 		       unsigned int nlm_flags);
77 
78 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
79 {
80 	struct mpls_route *rt = NULL;
81 
82 	if (index < net->mpls.platform_labels) {
83 		struct mpls_route __rcu **platform_label =
84 			rcu_dereference_rtnl(net->mpls.platform_label);
85 		rt = rcu_dereference_rtnl(platform_label[index]);
86 	}
87 	return rt;
88 }
89 
90 bool mpls_output_possible(const struct net_device *dev)
91 {
92 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
93 }
94 EXPORT_SYMBOL_GPL(mpls_output_possible);
95 
96 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
97 {
98 	return (u8 *)nh + rt->rt_via_offset;
99 }
100 
101 static const u8 *mpls_nh_via(const struct mpls_route *rt,
102 			     const struct mpls_nh *nh)
103 {
104 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
105 }
106 
107 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
108 {
109 	/* The size of the layer 2.5 labels to be added for this route */
110 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
111 }
112 
113 unsigned int mpls_dev_mtu(const struct net_device *dev)
114 {
115 	/* The amount of data the layer 2 frame can hold */
116 	return dev->mtu;
117 }
118 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
119 
120 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
121 {
122 	if (skb->len <= mtu)
123 		return false;
124 
125 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
126 		return false;
127 
128 	return true;
129 }
130 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
131 
132 void mpls_stats_inc_outucastpkts(struct net *net,
133 				 struct net_device *dev,
134 				 const struct sk_buff *skb)
135 {
136 	struct mpls_dev *mdev;
137 
138 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
139 		mdev = mpls_dev_get(dev);
140 		if (mdev)
141 			MPLS_INC_STATS_LEN(mdev, skb->len,
142 					   tx_packets,
143 					   tx_bytes);
144 	} else if (skb->protocol == htons(ETH_P_IP)) {
145 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
146 #if IS_ENABLED(CONFIG_IPV6)
147 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
148 		struct inet6_dev *in6dev = in6_dev_rcu(dev);
149 
150 		if (in6dev)
151 			IP6_UPD_PO_STATS(net, in6dev,
152 					 IPSTATS_MIB_OUT, skb->len);
153 #endif
154 	}
155 }
156 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
157 
158 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
159 {
160 	struct mpls_entry_decoded dec;
161 	unsigned int mpls_hdr_len = 0;
162 	struct mpls_shim_hdr *hdr;
163 	bool eli_seen = false;
164 	int label_index;
165 	u32 hash = 0;
166 
167 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
168 	     label_index++) {
169 		mpls_hdr_len += sizeof(*hdr);
170 		if (!pskb_may_pull(skb, mpls_hdr_len))
171 			break;
172 
173 		/* Read and decode the current label */
174 		hdr = mpls_hdr(skb) + label_index;
175 		dec = mpls_entry_decode(hdr);
176 
177 		/* RFC6790 - reserved labels MUST NOT be used as keys
178 		 * for the load-balancing function
179 		 */
180 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
181 			hash = jhash_1word(dec.label, hash);
182 
183 			/* The entropy label follows the entropy label
184 			 * indicator, so this means that the entropy
185 			 * label was just added to the hash - no need to
186 			 * go any deeper either in the label stack or in the
187 			 * payload
188 			 */
189 			if (eli_seen)
190 				break;
191 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
192 			eli_seen = true;
193 		}
194 
195 		if (!dec.bos)
196 			continue;
197 
198 		/* found bottom label; does skb have room for a header? */
199 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
200 			const struct iphdr *v4hdr;
201 
202 			v4hdr = (const struct iphdr *)(hdr + 1);
203 			if (v4hdr->version == 4) {
204 				hash = jhash_3words(ntohl(v4hdr->saddr),
205 						    ntohl(v4hdr->daddr),
206 						    v4hdr->protocol, hash);
207 			} else if (v4hdr->version == 6 &&
208 				   pskb_may_pull(skb, mpls_hdr_len +
209 						 sizeof(struct ipv6hdr))) {
210 				const struct ipv6hdr *v6hdr;
211 
212 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
213 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
214 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
215 				hash = jhash_1word(v6hdr->nexthdr, hash);
216 			}
217 		}
218 
219 		break;
220 	}
221 
222 	return hash;
223 }
224 
225 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
226 {
227 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
228 }
229 
230 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
231  * a next hop (nh->nh_flags) are modified by netdev event handlers.
232  * Since those fields can change at any moment, use READ_ONCE to
233  * access both.
234  */
235 static const struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
236 						   struct sk_buff *skb)
237 {
238 	u32 hash = 0;
239 	int nh_index = 0;
240 	int n = 0;
241 	u8 alive;
242 
243 	/* No need to look further into packet if there's only
244 	 * one path
245 	 */
246 	if (rt->rt_nhn == 1)
247 		return rt->rt_nh;
248 
249 	alive = READ_ONCE(rt->rt_nhn_alive);
250 	if (alive == 0)
251 		return NULL;
252 
253 	hash = mpls_multipath_hash(rt, skb);
254 	nh_index = hash % alive;
255 	if (alive == rt->rt_nhn)
256 		goto out;
257 	for_nexthops(rt) {
258 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
259 
260 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
261 			continue;
262 		if (n == nh_index)
263 			return nh;
264 		n++;
265 	} endfor_nexthops(rt);
266 
267 out:
268 	return mpls_get_nexthop(rt, nh_index);
269 }
270 
271 static bool mpls_egress(struct net *net, struct mpls_route *rt,
272 			struct sk_buff *skb, struct mpls_entry_decoded dec)
273 {
274 	enum mpls_payload_type payload_type;
275 	bool success = false;
276 
277 	/* The IPv4 code below accesses through the IPv4 header
278 	 * checksum, which is 12 bytes into the packet.
279 	 * The IPv6 code below accesses through the IPv6 hop limit
280 	 * which is 8 bytes into the packet.
281 	 *
282 	 * For all supported cases there should always be at least 12
283 	 * bytes of packet data present.  The IPv4 header is 20 bytes
284 	 * without options and the IPv6 header is always 40 bytes
285 	 * long.
286 	 */
287 	if (!pskb_may_pull(skb, 12))
288 		return false;
289 
290 	payload_type = rt->rt_payload_type;
291 	if (payload_type == MPT_UNSPEC)
292 		payload_type = ip_hdr(skb)->version;
293 
294 	switch (payload_type) {
295 	case MPT_IPV4: {
296 		struct iphdr *hdr4 = ip_hdr(skb);
297 		u8 new_ttl;
298 		skb->protocol = htons(ETH_P_IP);
299 
300 		/* If propagating TTL, take the decremented TTL from
301 		 * the incoming MPLS header, otherwise decrement the
302 		 * TTL, but only if not 0 to avoid underflow.
303 		 */
304 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
305 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
306 		     net->mpls.ip_ttl_propagate))
307 			new_ttl = dec.ttl;
308 		else
309 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
310 
311 		csum_replace2(&hdr4->check,
312 			      htons(hdr4->ttl << 8),
313 			      htons(new_ttl << 8));
314 		hdr4->ttl = new_ttl;
315 		success = true;
316 		break;
317 	}
318 	case MPT_IPV6: {
319 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
320 		skb->protocol = htons(ETH_P_IPV6);
321 
322 		/* If propagating TTL, take the decremented TTL from
323 		 * the incoming MPLS header, otherwise decrement the
324 		 * hop limit, but only if not 0 to avoid underflow.
325 		 */
326 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
327 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
328 		     net->mpls.ip_ttl_propagate))
329 			hdr6->hop_limit = dec.ttl;
330 		else if (hdr6->hop_limit)
331 			hdr6->hop_limit = hdr6->hop_limit - 1;
332 		success = true;
333 		break;
334 	}
335 	case MPT_UNSPEC:
336 		/* Should have decided which protocol it is by now */
337 		break;
338 	}
339 
340 	return success;
341 }
342 
343 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
344 			struct packet_type *pt, struct net_device *orig_dev)
345 {
346 	struct net *net = dev_net_rcu(dev);
347 	struct mpls_shim_hdr *hdr;
348 	const struct mpls_nh *nh;
349 	struct mpls_route *rt;
350 	struct mpls_entry_decoded dec;
351 	struct net_device *out_dev;
352 	struct mpls_dev *out_mdev;
353 	struct mpls_dev *mdev;
354 	unsigned int hh_len;
355 	unsigned int new_header_size;
356 	unsigned int mtu;
357 	int err;
358 
359 	/* Careful this entire function runs inside of an rcu critical section */
360 
361 	mdev = mpls_dev_get(dev);
362 	if (!mdev)
363 		goto drop;
364 
365 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
366 			   rx_bytes);
367 
368 	if (!mdev->input_enabled) {
369 		MPLS_INC_STATS(mdev, rx_dropped);
370 		goto drop;
371 	}
372 
373 	if (skb->pkt_type != PACKET_HOST)
374 		goto err;
375 
376 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
377 		goto err;
378 
379 	if (!pskb_may_pull(skb, sizeof(*hdr)))
380 		goto err;
381 
382 	skb_dst_drop(skb);
383 
384 	/* Read and decode the label */
385 	hdr = mpls_hdr(skb);
386 	dec = mpls_entry_decode(hdr);
387 
388 	rt = mpls_route_input_rcu(net, dec.label);
389 	if (!rt) {
390 		MPLS_INC_STATS(mdev, rx_noroute);
391 		goto drop;
392 	}
393 
394 	nh = mpls_select_multipath(rt, skb);
395 	if (!nh)
396 		goto err;
397 
398 	/* Pop the label */
399 	skb_pull(skb, sizeof(*hdr));
400 	skb_reset_network_header(skb);
401 
402 	skb_orphan(skb);
403 
404 	if (skb_warn_if_lro(skb))
405 		goto err;
406 
407 	skb_forward_csum(skb);
408 
409 	/* Verify ttl is valid */
410 	if (dec.ttl <= 1)
411 		goto err;
412 
413 	/* Find the output device */
414 	out_dev = nh->nh_dev;
415 	if (!mpls_output_possible(out_dev))
416 		goto tx_err;
417 
418 	/* Verify the destination can hold the packet */
419 	new_header_size = mpls_nh_header_size(nh);
420 	mtu = mpls_dev_mtu(out_dev);
421 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
422 		goto tx_err;
423 
424 	hh_len = LL_RESERVED_SPACE(out_dev);
425 	if (!out_dev->header_ops)
426 		hh_len = 0;
427 
428 	/* Ensure there is enough space for the headers in the skb */
429 	if (skb_cow(skb, hh_len + new_header_size))
430 		goto tx_err;
431 
432 	skb->dev = out_dev;
433 	skb->protocol = htons(ETH_P_MPLS_UC);
434 
435 	dec.ttl -= 1;
436 	if (unlikely(!new_header_size && dec.bos)) {
437 		/* Penultimate hop popping */
438 		if (!mpls_egress(net, rt, skb, dec))
439 			goto err;
440 	} else {
441 		bool bos;
442 		int i;
443 		skb_push(skb, new_header_size);
444 		skb_reset_network_header(skb);
445 		/* Push the new labels */
446 		hdr = mpls_hdr(skb);
447 		bos = dec.bos;
448 		for (i = nh->nh_labels - 1; i >= 0; i--) {
449 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
450 						   dec.ttl, 0, bos);
451 			bos = false;
452 		}
453 	}
454 
455 	mpls_stats_inc_outucastpkts(net, out_dev, skb);
456 
457 	/* If via wasn't specified then send out using device address */
458 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
459 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
460 				 out_dev->dev_addr, skb);
461 	else
462 		err = neigh_xmit(nh->nh_via_table, out_dev,
463 				 mpls_nh_via(rt, nh), skb);
464 	if (err)
465 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
466 				    __func__, err);
467 	return 0;
468 
469 tx_err:
470 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
471 	if (out_mdev)
472 		MPLS_INC_STATS(out_mdev, tx_errors);
473 	goto drop;
474 err:
475 	MPLS_INC_STATS(mdev, rx_errors);
476 drop:
477 	kfree_skb(skb);
478 	return NET_RX_DROP;
479 }
480 
481 static struct packet_type mpls_packet_type __read_mostly = {
482 	.type = cpu_to_be16(ETH_P_MPLS_UC),
483 	.func = mpls_forward,
484 };
485 
486 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
487 	[RTA_DST]		= { .type = NLA_U32 },
488 	[RTA_OIF]		= { .type = NLA_U32 },
489 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
490 };
491 
492 struct mpls_route_config {
493 	u32			rc_protocol;
494 	u32			rc_ifindex;
495 	u8			rc_via_table;
496 	u8			rc_via_alen;
497 	u8			rc_via[MAX_VIA_ALEN];
498 	u32			rc_label;
499 	u8			rc_ttl_propagate;
500 	u8			rc_output_labels;
501 	u32			rc_output_label[MAX_NEW_LABELS];
502 	u32			rc_nlflags;
503 	enum mpls_payload_type	rc_payload_type;
504 	struct nl_info		rc_nlinfo;
505 	struct rtnexthop	*rc_mp;
506 	int			rc_mp_len;
507 };
508 
509 /* all nexthops within a route have the same size based on max
510  * number of labels and max via length for a hop
511  */
512 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
513 {
514 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
515 	struct mpls_route *rt;
516 	size_t size;
517 
518 	size = sizeof(*rt) + num_nh * nh_size;
519 	if (size > MAX_MPLS_ROUTE_MEM)
520 		return ERR_PTR(-EINVAL);
521 
522 	rt = kzalloc(size, GFP_KERNEL);
523 	if (!rt)
524 		return ERR_PTR(-ENOMEM);
525 
526 	rt->rt_nhn = num_nh;
527 	rt->rt_nhn_alive = num_nh;
528 	rt->rt_nh_size = nh_size;
529 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
530 
531 	return rt;
532 }
533 
534 static void mpls_rt_free_rcu(struct rcu_head *head)
535 {
536 	struct mpls_route *rt;
537 
538 	rt = container_of(head, struct mpls_route, rt_rcu);
539 
540 	change_nexthops(rt) {
541 		netdev_put(nh->nh_dev, &nh->nh_dev_tracker);
542 	} endfor_nexthops(rt);
543 
544 	kfree(rt);
545 }
546 
547 static void mpls_rt_free(struct mpls_route *rt)
548 {
549 	if (rt)
550 		call_rcu(&rt->rt_rcu, mpls_rt_free_rcu);
551 }
552 
553 static void mpls_notify_route(struct net *net, unsigned index,
554 			      struct mpls_route *old, struct mpls_route *new,
555 			      const struct nl_info *info)
556 {
557 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
558 	unsigned portid = info ? info->portid : 0;
559 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
560 	struct mpls_route *rt = new ? new : old;
561 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
562 	/* Ignore reserved labels for now */
563 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
564 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
565 }
566 
567 static void mpls_route_update(struct net *net, unsigned index,
568 			      struct mpls_route *new,
569 			      const struct nl_info *info)
570 {
571 	struct mpls_route __rcu **platform_label;
572 	struct mpls_route *rt;
573 
574 	ASSERT_RTNL();
575 
576 	platform_label = rtnl_dereference(net->mpls.platform_label);
577 	rt = rtnl_dereference(platform_label[index]);
578 	rcu_assign_pointer(platform_label[index], new);
579 
580 	mpls_notify_route(net, index, rt, new, info);
581 
582 	/* If we removed a route free it now */
583 	mpls_rt_free(rt);
584 }
585 
586 static unsigned find_free_label(struct net *net)
587 {
588 	struct mpls_route __rcu **platform_label;
589 	size_t platform_labels;
590 	unsigned index;
591 
592 	platform_label = rtnl_dereference(net->mpls.platform_label);
593 	platform_labels = net->mpls.platform_labels;
594 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
595 	     index++) {
596 		if (!rtnl_dereference(platform_label[index]))
597 			return index;
598 	}
599 	return LABEL_NOT_SPECIFIED;
600 }
601 
602 #if IS_ENABLED(CONFIG_INET)
603 static struct net_device *inet_fib_lookup_dev(struct net *net,
604 					      struct mpls_nh *nh,
605 					      const void *addr)
606 {
607 	struct net_device *dev;
608 	struct rtable *rt;
609 	struct in_addr daddr;
610 
611 	memcpy(&daddr, addr, sizeof(struct in_addr));
612 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0, RT_SCOPE_UNIVERSE);
613 	if (IS_ERR(rt))
614 		return ERR_CAST(rt);
615 
616 	dev = rt->dst.dev;
617 	netdev_hold(dev, &nh->nh_dev_tracker, GFP_KERNEL);
618 	ip_rt_put(rt);
619 
620 	return dev;
621 }
622 #else
623 static struct net_device *inet_fib_lookup_dev(struct net *net,
624 					      struct mpls_nh *nh,
625 					      const void *addr)
626 {
627 	return ERR_PTR(-EAFNOSUPPORT);
628 }
629 #endif
630 
631 #if IS_ENABLED(CONFIG_IPV6)
632 static struct net_device *inet6_fib_lookup_dev(struct net *net,
633 					       struct mpls_nh *nh,
634 					       const void *addr)
635 {
636 	struct net_device *dev;
637 	struct dst_entry *dst;
638 	struct flowi6 fl6;
639 
640 	if (!ipv6_stub)
641 		return ERR_PTR(-EAFNOSUPPORT);
642 
643 	memset(&fl6, 0, sizeof(fl6));
644 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
645 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
646 	if (IS_ERR(dst))
647 		return ERR_CAST(dst);
648 
649 	dev = dst->dev;
650 	netdev_hold(dev, &nh->nh_dev_tracker, GFP_KERNEL);
651 	dst_release(dst);
652 
653 	return dev;
654 }
655 #else
656 static struct net_device *inet6_fib_lookup_dev(struct net *net,
657 					       struct mpls_nh *nh,
658 					       const void *addr)
659 {
660 	return ERR_PTR(-EAFNOSUPPORT);
661 }
662 #endif
663 
664 static struct net_device *find_outdev(struct net *net,
665 				      struct mpls_route *rt,
666 				      struct mpls_nh *nh, int oif)
667 {
668 	struct net_device *dev = NULL;
669 
670 	if (!oif) {
671 		switch (nh->nh_via_table) {
672 		case NEIGH_ARP_TABLE:
673 			dev = inet_fib_lookup_dev(net, nh, mpls_nh_via(rt, nh));
674 			break;
675 		case NEIGH_ND_TABLE:
676 			dev = inet6_fib_lookup_dev(net, nh, mpls_nh_via(rt, nh));
677 			break;
678 		case NEIGH_LINK_TABLE:
679 			break;
680 		}
681 	} else {
682 		dev = netdev_get_by_index(net, oif,
683 					  &nh->nh_dev_tracker, GFP_KERNEL);
684 	}
685 
686 	if (!dev)
687 		return ERR_PTR(-ENODEV);
688 
689 	if (IS_ERR(dev))
690 		return dev;
691 
692 	nh->nh_dev = dev;
693 
694 	return dev;
695 }
696 
697 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
698 			      struct mpls_nh *nh, int oif)
699 {
700 	struct net_device *dev = NULL;
701 	int err = -ENODEV;
702 
703 	dev = find_outdev(net, rt, nh, oif);
704 	if (IS_ERR(dev)) {
705 		err = PTR_ERR(dev);
706 		goto errout;
707 	}
708 
709 	/* Ensure this is a supported device */
710 	err = -EINVAL;
711 	if (!mpls_dev_get(dev))
712 		goto errout_put;
713 
714 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
715 	    (dev->addr_len != nh->nh_via_alen))
716 		goto errout_put;
717 
718 	if (!(dev->flags & IFF_UP)) {
719 		nh->nh_flags |= RTNH_F_DEAD;
720 	} else {
721 		unsigned int flags;
722 
723 		flags = netif_get_flags(dev);
724 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
725 			nh->nh_flags |= RTNH_F_LINKDOWN;
726 	}
727 
728 	return 0;
729 
730 errout_put:
731 	netdev_put(nh->nh_dev, &nh->nh_dev_tracker);
732 	nh->nh_dev = NULL;
733 errout:
734 	return err;
735 }
736 
737 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
738 		       u8 via_addr[], struct netlink_ext_ack *extack)
739 {
740 	struct rtvia *via = nla_data(nla);
741 	int err = -EINVAL;
742 	int alen;
743 
744 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
745 		NL_SET_ERR_MSG_ATTR(extack, nla,
746 				    "Invalid attribute length for RTA_VIA");
747 		goto errout;
748 	}
749 	alen = nla_len(nla) -
750 			offsetof(struct rtvia, rtvia_addr);
751 	if (alen > MAX_VIA_ALEN) {
752 		NL_SET_ERR_MSG_ATTR(extack, nla,
753 				    "Invalid address length for RTA_VIA");
754 		goto errout;
755 	}
756 
757 	/* Validate the address family */
758 	switch (via->rtvia_family) {
759 	case AF_PACKET:
760 		*via_table = NEIGH_LINK_TABLE;
761 		break;
762 	case AF_INET:
763 		*via_table = NEIGH_ARP_TABLE;
764 		if (alen != 4)
765 			goto errout;
766 		break;
767 	case AF_INET6:
768 		*via_table = NEIGH_ND_TABLE;
769 		if (alen != 16)
770 			goto errout;
771 		break;
772 	default:
773 		/* Unsupported address family */
774 		goto errout;
775 	}
776 
777 	memcpy(via_addr, via->rtvia_addr, alen);
778 	*via_alen = alen;
779 	err = 0;
780 
781 errout:
782 	return err;
783 }
784 
785 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
786 				  struct mpls_route *rt)
787 {
788 	struct net *net = cfg->rc_nlinfo.nl_net;
789 	struct mpls_nh *nh = rt->rt_nh;
790 	int err;
791 	int i;
792 
793 	if (!nh)
794 		return -ENOMEM;
795 
796 	nh->nh_labels = cfg->rc_output_labels;
797 	for (i = 0; i < nh->nh_labels; i++)
798 		nh->nh_label[i] = cfg->rc_output_label[i];
799 
800 	nh->nh_via_table = cfg->rc_via_table;
801 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
802 	nh->nh_via_alen = cfg->rc_via_alen;
803 
804 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
805 	if (err)
806 		goto errout;
807 
808 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
809 		rt->rt_nhn_alive--;
810 
811 	return 0;
812 
813 errout:
814 	return err;
815 }
816 
817 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
818 			 struct mpls_nh *nh, int oif, struct nlattr *via,
819 			 struct nlattr *newdst, u8 max_labels,
820 			 struct netlink_ext_ack *extack)
821 {
822 	int err = -ENOMEM;
823 
824 	if (!nh)
825 		goto errout;
826 
827 	if (newdst) {
828 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
829 				     nh->nh_label, extack);
830 		if (err)
831 			goto errout;
832 	}
833 
834 	if (via) {
835 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
836 				  __mpls_nh_via(rt, nh), extack);
837 		if (err)
838 			goto errout;
839 	} else {
840 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
841 	}
842 
843 	err = mpls_nh_assign_dev(net, rt, nh, oif);
844 	if (err)
845 		goto errout;
846 
847 	return 0;
848 
849 errout:
850 	return err;
851 }
852 
853 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
854 			      u8 cfg_via_alen, u8 *max_via_alen,
855 			      u8 *max_labels)
856 {
857 	int remaining = len;
858 	u8 nhs = 0;
859 
860 	*max_via_alen = 0;
861 	*max_labels = 0;
862 
863 	while (rtnh_ok(rtnh, remaining)) {
864 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
865 		int attrlen;
866 		u8 n_labels = 0;
867 
868 		attrlen = rtnh_attrlen(rtnh);
869 		nla = nla_find(attrs, attrlen, RTA_VIA);
870 		if (nla && nla_len(nla) >=
871 		    offsetof(struct rtvia, rtvia_addr)) {
872 			int via_alen = nla_len(nla) -
873 				offsetof(struct rtvia, rtvia_addr);
874 
875 			if (via_alen <= MAX_VIA_ALEN)
876 				*max_via_alen = max_t(u16, *max_via_alen,
877 						      via_alen);
878 		}
879 
880 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
881 		if (nla &&
882 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
883 				   NULL, NULL) != 0)
884 			return 0;
885 
886 		*max_labels = max_t(u8, *max_labels, n_labels);
887 
888 		/* number of nexthops is tracked by a u8.
889 		 * Check for overflow.
890 		 */
891 		if (nhs == 255)
892 			return 0;
893 		nhs++;
894 
895 		rtnh = rtnh_next(rtnh, &remaining);
896 	}
897 
898 	/* leftover implies invalid nexthop configuration, discard it */
899 	return remaining > 0 ? 0 : nhs;
900 }
901 
902 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
903 			       struct mpls_route *rt, u8 max_labels,
904 			       struct netlink_ext_ack *extack)
905 {
906 	struct rtnexthop *rtnh = cfg->rc_mp;
907 	struct nlattr *nla_via, *nla_newdst;
908 	int remaining = cfg->rc_mp_len;
909 	int err = 0;
910 
911 	rt->rt_nhn = 0;
912 
913 	change_nexthops(rt) {
914 		int attrlen;
915 
916 		nla_via = NULL;
917 		nla_newdst = NULL;
918 
919 		err = -EINVAL;
920 		if (!rtnh_ok(rtnh, remaining))
921 			goto errout;
922 
923 		/* neither weighted multipath nor any flags
924 		 * are supported
925 		 */
926 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
927 			goto errout;
928 
929 		attrlen = rtnh_attrlen(rtnh);
930 		if (attrlen > 0) {
931 			struct nlattr *attrs = rtnh_attrs(rtnh);
932 
933 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
934 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
935 		}
936 
937 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
938 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
939 				    max_labels, extack);
940 		if (err)
941 			goto errout;
942 
943 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
944 			rt->rt_nhn_alive--;
945 
946 		rtnh = rtnh_next(rtnh, &remaining);
947 		rt->rt_nhn++;
948 	} endfor_nexthops(rt);
949 
950 	return 0;
951 
952 errout:
953 	return err;
954 }
955 
956 static bool mpls_label_ok(struct net *net, unsigned int *index,
957 			  struct netlink_ext_ack *extack)
958 {
959 	/* Reserved labels may not be set */
960 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
961 		NL_SET_ERR_MSG(extack,
962 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
963 		return false;
964 	}
965 
966 	/* The full 20 bit range may not be supported. */
967 	if (*index >= net->mpls.platform_labels) {
968 		NL_SET_ERR_MSG(extack,
969 			       "Label >= configured maximum in platform_labels");
970 		return false;
971 	}
972 
973 	*index = array_index_nospec(*index, net->mpls.platform_labels);
974 
975 	return true;
976 }
977 
978 static int mpls_route_add(struct mpls_route_config *cfg,
979 			  struct netlink_ext_ack *extack)
980 {
981 	struct mpls_route __rcu **platform_label;
982 	struct net *net = cfg->rc_nlinfo.nl_net;
983 	struct mpls_route *rt, *old;
984 	int err = -EINVAL;
985 	u8 max_via_alen;
986 	unsigned index;
987 	u8 max_labels;
988 	u8 nhs;
989 
990 	index = cfg->rc_label;
991 
992 	/* If a label was not specified during insert pick one */
993 	if ((index == LABEL_NOT_SPECIFIED) &&
994 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
995 		index = find_free_label(net);
996 	}
997 
998 	if (!mpls_label_ok(net, &index, extack))
999 		goto errout;
1000 
1001 	/* Append makes no sense with mpls */
1002 	err = -EOPNOTSUPP;
1003 	if (cfg->rc_nlflags & NLM_F_APPEND) {
1004 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
1005 		goto errout;
1006 	}
1007 
1008 	err = -EEXIST;
1009 	platform_label = rtnl_dereference(net->mpls.platform_label);
1010 	old = rtnl_dereference(platform_label[index]);
1011 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
1012 		goto errout;
1013 
1014 	err = -EEXIST;
1015 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1016 		goto errout;
1017 
1018 	err = -ENOENT;
1019 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1020 		goto errout;
1021 
1022 	err = -EINVAL;
1023 	if (cfg->rc_mp) {
1024 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1025 					  cfg->rc_via_alen, &max_via_alen,
1026 					  &max_labels);
1027 	} else {
1028 		max_via_alen = cfg->rc_via_alen;
1029 		max_labels = cfg->rc_output_labels;
1030 		nhs = 1;
1031 	}
1032 
1033 	if (nhs == 0) {
1034 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1035 		goto errout;
1036 	}
1037 
1038 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1039 	if (IS_ERR(rt)) {
1040 		err = PTR_ERR(rt);
1041 		goto errout;
1042 	}
1043 
1044 	rt->rt_protocol = cfg->rc_protocol;
1045 	rt->rt_payload_type = cfg->rc_payload_type;
1046 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1047 
1048 	if (cfg->rc_mp)
1049 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1050 	else
1051 		err = mpls_nh_build_from_cfg(cfg, rt);
1052 	if (err)
1053 		goto freert;
1054 
1055 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1056 
1057 	return 0;
1058 
1059 freert:
1060 	mpls_rt_free(rt);
1061 errout:
1062 	return err;
1063 }
1064 
1065 static int mpls_route_del(struct mpls_route_config *cfg,
1066 			  struct netlink_ext_ack *extack)
1067 {
1068 	struct net *net = cfg->rc_nlinfo.nl_net;
1069 	unsigned index;
1070 	int err = -EINVAL;
1071 
1072 	index = cfg->rc_label;
1073 
1074 	if (!mpls_label_ok(net, &index, extack))
1075 		goto errout;
1076 
1077 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1078 
1079 	err = 0;
1080 errout:
1081 	return err;
1082 }
1083 
1084 static void mpls_get_stats(struct mpls_dev *mdev,
1085 			   struct mpls_link_stats *stats)
1086 {
1087 	struct mpls_pcpu_stats *p;
1088 	int i;
1089 
1090 	memset(stats, 0, sizeof(*stats));
1091 
1092 	for_each_possible_cpu(i) {
1093 		struct mpls_link_stats local;
1094 		unsigned int start;
1095 
1096 		p = per_cpu_ptr(mdev->stats, i);
1097 		do {
1098 			start = u64_stats_fetch_begin(&p->syncp);
1099 			local = p->stats;
1100 		} while (u64_stats_fetch_retry(&p->syncp, start));
1101 
1102 		stats->rx_packets	+= local.rx_packets;
1103 		stats->rx_bytes		+= local.rx_bytes;
1104 		stats->tx_packets	+= local.tx_packets;
1105 		stats->tx_bytes		+= local.tx_bytes;
1106 		stats->rx_errors	+= local.rx_errors;
1107 		stats->tx_errors	+= local.tx_errors;
1108 		stats->rx_dropped	+= local.rx_dropped;
1109 		stats->tx_dropped	+= local.tx_dropped;
1110 		stats->rx_noroute	+= local.rx_noroute;
1111 	}
1112 }
1113 
1114 static int mpls_fill_stats_af(struct sk_buff *skb,
1115 			      const struct net_device *dev)
1116 {
1117 	struct mpls_link_stats *stats;
1118 	struct mpls_dev *mdev;
1119 	struct nlattr *nla;
1120 
1121 	mdev = mpls_dev_get(dev);
1122 	if (!mdev)
1123 		return -ENODATA;
1124 
1125 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1126 				sizeof(struct mpls_link_stats),
1127 				MPLS_STATS_UNSPEC);
1128 	if (!nla)
1129 		return -EMSGSIZE;
1130 
1131 	stats = nla_data(nla);
1132 	mpls_get_stats(mdev, stats);
1133 
1134 	return 0;
1135 }
1136 
1137 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1138 {
1139 	struct mpls_dev *mdev;
1140 
1141 	mdev = mpls_dev_get(dev);
1142 	if (!mdev)
1143 		return 0;
1144 
1145 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1146 }
1147 
1148 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1149 				     u32 portid, u32 seq, int event,
1150 				     unsigned int flags, int type)
1151 {
1152 	struct nlmsghdr  *nlh;
1153 	struct netconfmsg *ncm;
1154 	bool all = false;
1155 
1156 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1157 			flags);
1158 	if (!nlh)
1159 		return -EMSGSIZE;
1160 
1161 	if (type == NETCONFA_ALL)
1162 		all = true;
1163 
1164 	ncm = nlmsg_data(nlh);
1165 	ncm->ncm_family = AF_MPLS;
1166 
1167 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1168 		goto nla_put_failure;
1169 
1170 	if ((all || type == NETCONFA_INPUT) &&
1171 	    nla_put_s32(skb, NETCONFA_INPUT,
1172 			READ_ONCE(mdev->input_enabled)) < 0)
1173 		goto nla_put_failure;
1174 
1175 	nlmsg_end(skb, nlh);
1176 	return 0;
1177 
1178 nla_put_failure:
1179 	nlmsg_cancel(skb, nlh);
1180 	return -EMSGSIZE;
1181 }
1182 
1183 static int mpls_netconf_msgsize_devconf(int type)
1184 {
1185 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1186 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1187 	bool all = false;
1188 
1189 	if (type == NETCONFA_ALL)
1190 		all = true;
1191 
1192 	if (all || type == NETCONFA_INPUT)
1193 		size += nla_total_size(4);
1194 
1195 	return size;
1196 }
1197 
1198 static void mpls_netconf_notify_devconf(struct net *net, int event,
1199 					int type, struct mpls_dev *mdev)
1200 {
1201 	struct sk_buff *skb;
1202 	int err = -ENOBUFS;
1203 
1204 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1205 	if (!skb)
1206 		goto errout;
1207 
1208 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1209 	if (err < 0) {
1210 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1211 		WARN_ON(err == -EMSGSIZE);
1212 		kfree_skb(skb);
1213 		goto errout;
1214 	}
1215 
1216 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1217 	return;
1218 errout:
1219 	rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1220 }
1221 
1222 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1223 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1224 };
1225 
1226 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1227 				      const struct nlmsghdr *nlh,
1228 				      struct nlattr **tb,
1229 				      struct netlink_ext_ack *extack)
1230 {
1231 	int i, err;
1232 
1233 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1234 		NL_SET_ERR_MSG_MOD(extack,
1235 				   "Invalid header for netconf get request");
1236 		return -EINVAL;
1237 	}
1238 
1239 	if (!netlink_strict_get_check(skb))
1240 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1241 					      tb, NETCONFA_MAX,
1242 					      devconf_mpls_policy, extack);
1243 
1244 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1245 					    tb, NETCONFA_MAX,
1246 					    devconf_mpls_policy, extack);
1247 	if (err)
1248 		return err;
1249 
1250 	for (i = 0; i <= NETCONFA_MAX; i++) {
1251 		if (!tb[i])
1252 			continue;
1253 
1254 		switch (i) {
1255 		case NETCONFA_IFINDEX:
1256 			break;
1257 		default:
1258 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1259 			return -EINVAL;
1260 		}
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1267 				    struct nlmsghdr *nlh,
1268 				    struct netlink_ext_ack *extack)
1269 {
1270 	struct net *net = sock_net(in_skb->sk);
1271 	struct nlattr *tb[NETCONFA_MAX + 1];
1272 	struct net_device *dev;
1273 	struct mpls_dev *mdev;
1274 	struct sk_buff *skb;
1275 	int ifindex;
1276 	int err;
1277 
1278 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1279 	if (err < 0)
1280 		goto errout;
1281 
1282 	err = -EINVAL;
1283 	if (!tb[NETCONFA_IFINDEX])
1284 		goto errout;
1285 
1286 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1287 	dev = __dev_get_by_index(net, ifindex);
1288 	if (!dev)
1289 		goto errout;
1290 
1291 	mdev = mpls_dev_get(dev);
1292 	if (!mdev)
1293 		goto errout;
1294 
1295 	err = -ENOBUFS;
1296 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1297 	if (!skb)
1298 		goto errout;
1299 
1300 	err = mpls_netconf_fill_devconf(skb, mdev,
1301 					NETLINK_CB(in_skb).portid,
1302 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1303 					NETCONFA_ALL);
1304 	if (err < 0) {
1305 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1306 		WARN_ON(err == -EMSGSIZE);
1307 		kfree_skb(skb);
1308 		goto errout;
1309 	}
1310 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1311 errout:
1312 	return err;
1313 }
1314 
1315 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1316 				     struct netlink_callback *cb)
1317 {
1318 	const struct nlmsghdr *nlh = cb->nlh;
1319 	struct net *net = sock_net(skb->sk);
1320 	struct {
1321 		unsigned long ifindex;
1322 	} *ctx = (void *)cb->ctx;
1323 	struct net_device *dev;
1324 	struct mpls_dev *mdev;
1325 	int err = 0;
1326 
1327 	if (cb->strict_check) {
1328 		struct netlink_ext_ack *extack = cb->extack;
1329 		struct netconfmsg *ncm;
1330 
1331 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1332 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1333 			return -EINVAL;
1334 		}
1335 
1336 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1337 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1338 			return -EINVAL;
1339 		}
1340 	}
1341 
1342 	rcu_read_lock();
1343 	for_each_netdev_dump(net, dev, ctx->ifindex) {
1344 		mdev = mpls_dev_get(dev);
1345 		if (!mdev)
1346 			continue;
1347 		err = mpls_netconf_fill_devconf(skb, mdev,
1348 						NETLINK_CB(cb->skb).portid,
1349 						nlh->nlmsg_seq,
1350 						RTM_NEWNETCONF,
1351 						NLM_F_MULTI,
1352 						NETCONFA_ALL);
1353 		if (err < 0)
1354 			break;
1355 	}
1356 	rcu_read_unlock();
1357 
1358 	return err;
1359 }
1360 
1361 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1362 	(&((struct mpls_dev *)0)->field)
1363 
1364 static int mpls_conf_proc(const struct ctl_table *ctl, int write,
1365 			  void *buffer, size_t *lenp, loff_t *ppos)
1366 {
1367 	int oval = *(int *)ctl->data;
1368 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1369 
1370 	if (write) {
1371 		struct mpls_dev *mdev = ctl->extra1;
1372 		int i = (int *)ctl->data - (int *)mdev;
1373 		struct net *net = ctl->extra2;
1374 		int val = *(int *)ctl->data;
1375 
1376 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1377 		    val != oval) {
1378 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1379 						    NETCONFA_INPUT, mdev);
1380 		}
1381 	}
1382 
1383 	return ret;
1384 }
1385 
1386 static const struct ctl_table mpls_dev_table[] = {
1387 	{
1388 		.procname	= "input",
1389 		.maxlen		= sizeof(int),
1390 		.mode		= 0644,
1391 		.proc_handler	= mpls_conf_proc,
1392 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1393 	},
1394 };
1395 
1396 static int mpls_dev_sysctl_register(struct net_device *dev,
1397 				    struct mpls_dev *mdev)
1398 {
1399 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1400 	size_t table_size = ARRAY_SIZE(mpls_dev_table);
1401 	struct net *net = dev_net(dev);
1402 	struct ctl_table *table;
1403 	int i;
1404 
1405 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1406 	if (!table)
1407 		goto out;
1408 
1409 	/* Table data contains only offsets relative to the base of
1410 	 * the mdev at this point, so make them absolute.
1411 	 */
1412 	for (i = 0; i < table_size; i++) {
1413 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1414 		table[i].extra1 = mdev;
1415 		table[i].extra2 = net;
1416 	}
1417 
1418 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1419 
1420 	mdev->sysctl = register_net_sysctl_sz(net, path, table, table_size);
1421 	if (!mdev->sysctl)
1422 		goto free;
1423 
1424 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1425 	return 0;
1426 
1427 free:
1428 	kfree(table);
1429 out:
1430 	mdev->sysctl = NULL;
1431 	return -ENOBUFS;
1432 }
1433 
1434 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1435 				       struct mpls_dev *mdev)
1436 {
1437 	struct net *net = dev_net(dev);
1438 	const struct ctl_table *table;
1439 
1440 	if (!mdev->sysctl)
1441 		return;
1442 
1443 	table = mdev->sysctl->ctl_table_arg;
1444 	unregister_net_sysctl_table(mdev->sysctl);
1445 	kfree(table);
1446 
1447 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1448 }
1449 
1450 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1451 {
1452 	struct mpls_dev *mdev;
1453 	int err = -ENOMEM;
1454 	int i;
1455 
1456 	ASSERT_RTNL();
1457 
1458 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1459 	if (!mdev)
1460 		return ERR_PTR(err);
1461 
1462 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1463 	if (!mdev->stats)
1464 		goto free;
1465 
1466 	for_each_possible_cpu(i) {
1467 		struct mpls_pcpu_stats *mpls_stats;
1468 
1469 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1470 		u64_stats_init(&mpls_stats->syncp);
1471 	}
1472 
1473 	mdev->dev = dev;
1474 
1475 	err = mpls_dev_sysctl_register(dev, mdev);
1476 	if (err)
1477 		goto free;
1478 
1479 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1480 
1481 	return mdev;
1482 
1483 free:
1484 	free_percpu(mdev->stats);
1485 	kfree(mdev);
1486 	return ERR_PTR(err);
1487 }
1488 
1489 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1490 {
1491 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1492 
1493 	free_percpu(mdev->stats);
1494 	kfree(mdev);
1495 }
1496 
1497 static int mpls_ifdown(struct net_device *dev, int event)
1498 {
1499 	struct mpls_route __rcu **platform_label;
1500 	struct net *net = dev_net(dev);
1501 	unsigned index;
1502 
1503 	platform_label = rtnl_dereference(net->mpls.platform_label);
1504 	for (index = 0; index < net->mpls.platform_labels; index++) {
1505 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1506 		bool nh_del = false;
1507 		u8 alive = 0;
1508 
1509 		if (!rt)
1510 			continue;
1511 
1512 		if (event == NETDEV_UNREGISTER) {
1513 			u8 deleted = 0;
1514 
1515 			for_nexthops(rt) {
1516 				if (!nh->nh_dev || nh->nh_dev == dev)
1517 					deleted++;
1518 				if (nh->nh_dev == dev)
1519 					nh_del = true;
1520 			} endfor_nexthops(rt);
1521 
1522 			/* if there are no more nexthops, delete the route */
1523 			if (deleted == rt->rt_nhn) {
1524 				mpls_route_update(net, index, NULL, NULL);
1525 				continue;
1526 			}
1527 
1528 			if (nh_del) {
1529 				size_t size = sizeof(*rt) + rt->rt_nhn *
1530 					rt->rt_nh_size;
1531 				struct mpls_route *orig = rt;
1532 
1533 				rt = kmemdup(orig, size, GFP_KERNEL);
1534 				if (!rt)
1535 					return -ENOMEM;
1536 			}
1537 		}
1538 
1539 		change_nexthops(rt) {
1540 			unsigned int nh_flags = nh->nh_flags;
1541 
1542 			if (nh->nh_dev != dev) {
1543 				if (nh_del)
1544 					netdev_hold(nh->nh_dev, &nh->nh_dev_tracker,
1545 						    GFP_KERNEL);
1546 				goto next;
1547 			}
1548 
1549 			switch (event) {
1550 			case NETDEV_DOWN:
1551 			case NETDEV_UNREGISTER:
1552 				nh_flags |= RTNH_F_DEAD;
1553 				fallthrough;
1554 			case NETDEV_CHANGE:
1555 				nh_flags |= RTNH_F_LINKDOWN;
1556 				break;
1557 			}
1558 			if (event == NETDEV_UNREGISTER)
1559 				nh->nh_dev = NULL;
1560 
1561 			if (nh->nh_flags != nh_flags)
1562 				WRITE_ONCE(nh->nh_flags, nh_flags);
1563 next:
1564 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1565 				alive++;
1566 		} endfor_nexthops(rt);
1567 
1568 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1569 
1570 		if (nh_del)
1571 			mpls_route_update(net, index, rt, NULL);
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1578 {
1579 	struct mpls_route __rcu **platform_label;
1580 	struct net *net = dev_net(dev);
1581 	unsigned index;
1582 	u8 alive;
1583 
1584 	platform_label = rtnl_dereference(net->mpls.platform_label);
1585 	for (index = 0; index < net->mpls.platform_labels; index++) {
1586 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1587 
1588 		if (!rt)
1589 			continue;
1590 
1591 		alive = 0;
1592 		change_nexthops(rt) {
1593 			unsigned int nh_flags = nh->nh_flags;
1594 
1595 			if (!(nh_flags & flags)) {
1596 				alive++;
1597 				continue;
1598 			}
1599 			if (nh->nh_dev != dev)
1600 				continue;
1601 			alive++;
1602 			nh_flags &= ~flags;
1603 			WRITE_ONCE(nh->nh_flags, nh_flags);
1604 		} endfor_nexthops(rt);
1605 
1606 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1607 	}
1608 }
1609 
1610 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1611 			   void *ptr)
1612 {
1613 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1614 	struct mpls_dev *mdev;
1615 	unsigned int flags;
1616 	int err;
1617 
1618 	if (event == NETDEV_REGISTER) {
1619 		mdev = mpls_add_dev(dev);
1620 		if (IS_ERR(mdev)) {
1621 			err = PTR_ERR(mdev);
1622 			goto err;
1623 		}
1624 
1625 		goto out;
1626 	}
1627 
1628 	mdev = mpls_dev_get(dev);
1629 	if (!mdev)
1630 		goto out;
1631 
1632 	switch (event) {
1633 
1634 	case NETDEV_DOWN:
1635 		err = mpls_ifdown(dev, event);
1636 		if (err)
1637 			goto err;
1638 		break;
1639 	case NETDEV_UP:
1640 		flags = netif_get_flags(dev);
1641 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1642 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1643 		else
1644 			mpls_ifup(dev, RTNH_F_DEAD);
1645 		break;
1646 	case NETDEV_CHANGE:
1647 		flags = netif_get_flags(dev);
1648 		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1649 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1650 		} else {
1651 			err = mpls_ifdown(dev, event);
1652 			if (err)
1653 				goto err;
1654 		}
1655 		break;
1656 	case NETDEV_UNREGISTER:
1657 		err = mpls_ifdown(dev, event);
1658 		if (err)
1659 			goto err;
1660 
1661 		mdev = mpls_dev_get(dev);
1662 		if (mdev) {
1663 			mpls_dev_sysctl_unregister(dev, mdev);
1664 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1665 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1666 		}
1667 		break;
1668 	case NETDEV_CHANGENAME:
1669 		mdev = mpls_dev_get(dev);
1670 		if (mdev) {
1671 			mpls_dev_sysctl_unregister(dev, mdev);
1672 			err = mpls_dev_sysctl_register(dev, mdev);
1673 			if (err)
1674 				goto err;
1675 		}
1676 		break;
1677 	}
1678 
1679 out:
1680 	return NOTIFY_OK;
1681 
1682 err:
1683 	return notifier_from_errno(err);
1684 }
1685 
1686 static struct notifier_block mpls_dev_notifier = {
1687 	.notifier_call = mpls_dev_notify,
1688 };
1689 
1690 static int nla_put_via(struct sk_buff *skb,
1691 		       u8 table, const void *addr, int alen)
1692 {
1693 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1694 		AF_INET, AF_INET6, AF_PACKET,
1695 	};
1696 	struct nlattr *nla;
1697 	struct rtvia *via;
1698 	int family = AF_UNSPEC;
1699 
1700 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1701 	if (!nla)
1702 		return -EMSGSIZE;
1703 
1704 	if (table <= NEIGH_NR_TABLES)
1705 		family = table_to_family[table];
1706 
1707 	via = nla_data(nla);
1708 	via->rtvia_family = family;
1709 	memcpy(via->rtvia_addr, addr, alen);
1710 	return 0;
1711 }
1712 
1713 int nla_put_labels(struct sk_buff *skb, int attrtype,
1714 		   u8 labels, const u32 label[])
1715 {
1716 	struct nlattr *nla;
1717 	struct mpls_shim_hdr *nla_label;
1718 	bool bos;
1719 	int i;
1720 	nla = nla_reserve(skb, attrtype, labels*4);
1721 	if (!nla)
1722 		return -EMSGSIZE;
1723 
1724 	nla_label = nla_data(nla);
1725 	bos = true;
1726 	for (i = labels - 1; i >= 0; i--) {
1727 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1728 		bos = false;
1729 	}
1730 
1731 	return 0;
1732 }
1733 EXPORT_SYMBOL_GPL(nla_put_labels);
1734 
1735 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1736 		   u32 label[], struct netlink_ext_ack *extack)
1737 {
1738 	unsigned len = nla_len(nla);
1739 	struct mpls_shim_hdr *nla_label;
1740 	u8 nla_labels;
1741 	bool bos;
1742 	int i;
1743 
1744 	/* len needs to be an even multiple of 4 (the label size). Number
1745 	 * of labels is a u8 so check for overflow.
1746 	 */
1747 	if (len & 3 || len / 4 > 255) {
1748 		NL_SET_ERR_MSG_ATTR(extack, nla,
1749 				    "Invalid length for labels attribute");
1750 		return -EINVAL;
1751 	}
1752 
1753 	/* Limit the number of new labels allowed */
1754 	nla_labels = len/4;
1755 	if (nla_labels > max_labels) {
1756 		NL_SET_ERR_MSG(extack, "Too many labels");
1757 		return -EINVAL;
1758 	}
1759 
1760 	/* when label == NULL, caller wants number of labels */
1761 	if (!label)
1762 		goto out;
1763 
1764 	nla_label = nla_data(nla);
1765 	bos = true;
1766 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1767 		struct mpls_entry_decoded dec;
1768 		dec = mpls_entry_decode(nla_label + i);
1769 
1770 		/* Ensure the bottom of stack flag is properly set
1771 		 * and ttl and tc are both clear.
1772 		 */
1773 		if (dec.ttl) {
1774 			NL_SET_ERR_MSG_ATTR(extack, nla,
1775 					    "TTL in label must be 0");
1776 			return -EINVAL;
1777 		}
1778 
1779 		if (dec.tc) {
1780 			NL_SET_ERR_MSG_ATTR(extack, nla,
1781 					    "Traffic class in label must be 0");
1782 			return -EINVAL;
1783 		}
1784 
1785 		if (dec.bos != bos) {
1786 			NL_SET_BAD_ATTR(extack, nla);
1787 			if (bos) {
1788 				NL_SET_ERR_MSG(extack,
1789 					       "BOS bit must be set in first label");
1790 			} else {
1791 				NL_SET_ERR_MSG(extack,
1792 					       "BOS bit can only be set in first label");
1793 			}
1794 			return -EINVAL;
1795 		}
1796 
1797 		switch (dec.label) {
1798 		case MPLS_LABEL_IMPLNULL:
1799 			/* RFC3032: This is a label that an LSR may
1800 			 * assign and distribute, but which never
1801 			 * actually appears in the encapsulation.
1802 			 */
1803 			NL_SET_ERR_MSG_ATTR(extack, nla,
1804 					    "Implicit NULL Label (3) can not be used in encapsulation");
1805 			return -EINVAL;
1806 		}
1807 
1808 		label[i] = dec.label;
1809 	}
1810 out:
1811 	*labels = nla_labels;
1812 	return 0;
1813 }
1814 EXPORT_SYMBOL_GPL(nla_get_labels);
1815 
1816 static int rtm_to_route_config(struct sk_buff *skb,
1817 			       struct nlmsghdr *nlh,
1818 			       struct mpls_route_config *cfg,
1819 			       struct netlink_ext_ack *extack)
1820 {
1821 	struct rtmsg *rtm;
1822 	struct nlattr *tb[RTA_MAX+1];
1823 	int index;
1824 	int err;
1825 
1826 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1827 				     rtm_mpls_policy, extack);
1828 	if (err < 0)
1829 		goto errout;
1830 
1831 	err = -EINVAL;
1832 	rtm = nlmsg_data(nlh);
1833 
1834 	if (rtm->rtm_family != AF_MPLS) {
1835 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1836 		goto errout;
1837 	}
1838 	if (rtm->rtm_dst_len != 20) {
1839 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1840 		goto errout;
1841 	}
1842 	if (rtm->rtm_src_len != 0) {
1843 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1844 		goto errout;
1845 	}
1846 	if (rtm->rtm_tos != 0) {
1847 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1848 		goto errout;
1849 	}
1850 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1851 		NL_SET_ERR_MSG(extack,
1852 			       "MPLS only supports the main route table");
1853 		goto errout;
1854 	}
1855 	/* Any value is acceptable for rtm_protocol */
1856 
1857 	/* As mpls uses destination specific addresses
1858 	 * (or source specific address in the case of multicast)
1859 	 * all addresses have universal scope.
1860 	 */
1861 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1862 		NL_SET_ERR_MSG(extack,
1863 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1864 		goto errout;
1865 	}
1866 	if (rtm->rtm_type != RTN_UNICAST) {
1867 		NL_SET_ERR_MSG(extack,
1868 			       "Invalid route type - MPLS only supports UNICAST");
1869 		goto errout;
1870 	}
1871 	if (rtm->rtm_flags != 0) {
1872 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1873 		goto errout;
1874 	}
1875 
1876 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1877 	cfg->rc_protocol	= rtm->rtm_protocol;
1878 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1879 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1880 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1881 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1882 	cfg->rc_nlinfo.nlh	= nlh;
1883 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1884 
1885 	for (index = 0; index <= RTA_MAX; index++) {
1886 		struct nlattr *nla = tb[index];
1887 		if (!nla)
1888 			continue;
1889 
1890 		switch (index) {
1891 		case RTA_OIF:
1892 			cfg->rc_ifindex = nla_get_u32(nla);
1893 			break;
1894 		case RTA_NEWDST:
1895 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1896 					   &cfg->rc_output_labels,
1897 					   cfg->rc_output_label, extack))
1898 				goto errout;
1899 			break;
1900 		case RTA_DST:
1901 		{
1902 			u8 label_count;
1903 			if (nla_get_labels(nla, 1, &label_count,
1904 					   &cfg->rc_label, extack))
1905 				goto errout;
1906 
1907 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1908 					   &cfg->rc_label, extack))
1909 				goto errout;
1910 			break;
1911 		}
1912 		case RTA_GATEWAY:
1913 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1914 			goto errout;
1915 		case RTA_VIA:
1916 		{
1917 			if (nla_get_via(nla, &cfg->rc_via_alen,
1918 					&cfg->rc_via_table, cfg->rc_via,
1919 					extack))
1920 				goto errout;
1921 			break;
1922 		}
1923 		case RTA_MULTIPATH:
1924 		{
1925 			cfg->rc_mp = nla_data(nla);
1926 			cfg->rc_mp_len = nla_len(nla);
1927 			break;
1928 		}
1929 		case RTA_TTL_PROPAGATE:
1930 		{
1931 			u8 ttl_propagate = nla_get_u8(nla);
1932 
1933 			if (ttl_propagate > 1) {
1934 				NL_SET_ERR_MSG_ATTR(extack, nla,
1935 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1936 				goto errout;
1937 			}
1938 			cfg->rc_ttl_propagate = ttl_propagate ?
1939 				MPLS_TTL_PROP_ENABLED :
1940 				MPLS_TTL_PROP_DISABLED;
1941 			break;
1942 		}
1943 		default:
1944 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1945 			/* Unsupported attribute */
1946 			goto errout;
1947 		}
1948 	}
1949 
1950 	err = 0;
1951 errout:
1952 	return err;
1953 }
1954 
1955 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1956 			     struct netlink_ext_ack *extack)
1957 {
1958 	struct mpls_route_config *cfg;
1959 	int err;
1960 
1961 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1962 	if (!cfg)
1963 		return -ENOMEM;
1964 
1965 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1966 	if (err < 0)
1967 		goto out;
1968 
1969 	err = mpls_route_del(cfg, extack);
1970 out:
1971 	kfree(cfg);
1972 
1973 	return err;
1974 }
1975 
1976 
1977 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1978 			     struct netlink_ext_ack *extack)
1979 {
1980 	struct mpls_route_config *cfg;
1981 	int err;
1982 
1983 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1984 	if (!cfg)
1985 		return -ENOMEM;
1986 
1987 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1988 	if (err < 0)
1989 		goto out;
1990 
1991 	err = mpls_route_add(cfg, extack);
1992 out:
1993 	kfree(cfg);
1994 
1995 	return err;
1996 }
1997 
1998 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1999 			   u32 label, struct mpls_route *rt, int flags)
2000 {
2001 	struct net_device *dev;
2002 	struct nlmsghdr *nlh;
2003 	struct rtmsg *rtm;
2004 
2005 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
2006 	if (nlh == NULL)
2007 		return -EMSGSIZE;
2008 
2009 	rtm = nlmsg_data(nlh);
2010 	rtm->rtm_family = AF_MPLS;
2011 	rtm->rtm_dst_len = 20;
2012 	rtm->rtm_src_len = 0;
2013 	rtm->rtm_tos = 0;
2014 	rtm->rtm_table = RT_TABLE_MAIN;
2015 	rtm->rtm_protocol = rt->rt_protocol;
2016 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
2017 	rtm->rtm_type = RTN_UNICAST;
2018 	rtm->rtm_flags = 0;
2019 
2020 	if (nla_put_labels(skb, RTA_DST, 1, &label))
2021 		goto nla_put_failure;
2022 
2023 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
2024 		bool ttl_propagate =
2025 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2026 
2027 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2028 			       ttl_propagate))
2029 			goto nla_put_failure;
2030 	}
2031 	if (rt->rt_nhn == 1) {
2032 		const struct mpls_nh *nh = rt->rt_nh;
2033 
2034 		if (nh->nh_labels &&
2035 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2036 				   nh->nh_label))
2037 			goto nla_put_failure;
2038 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2039 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2040 				nh->nh_via_alen))
2041 			goto nla_put_failure;
2042 		dev = nh->nh_dev;
2043 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2044 			goto nla_put_failure;
2045 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2046 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2047 		if (nh->nh_flags & RTNH_F_DEAD)
2048 			rtm->rtm_flags |= RTNH_F_DEAD;
2049 	} else {
2050 		struct rtnexthop *rtnh;
2051 		struct nlattr *mp;
2052 		u8 linkdown = 0;
2053 		u8 dead = 0;
2054 
2055 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2056 		if (!mp)
2057 			goto nla_put_failure;
2058 
2059 		for_nexthops(rt) {
2060 			dev = nh->nh_dev;
2061 			if (!dev)
2062 				continue;
2063 
2064 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2065 			if (!rtnh)
2066 				goto nla_put_failure;
2067 
2068 			rtnh->rtnh_ifindex = dev->ifindex;
2069 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2070 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2071 				linkdown++;
2072 			}
2073 			if (nh->nh_flags & RTNH_F_DEAD) {
2074 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2075 				dead++;
2076 			}
2077 
2078 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2079 							    nh->nh_labels,
2080 							    nh->nh_label))
2081 				goto nla_put_failure;
2082 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2083 			    nla_put_via(skb, nh->nh_via_table,
2084 					mpls_nh_via(rt, nh),
2085 					nh->nh_via_alen))
2086 				goto nla_put_failure;
2087 
2088 			/* length of rtnetlink header + attributes */
2089 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2090 		} endfor_nexthops(rt);
2091 
2092 		if (linkdown == rt->rt_nhn)
2093 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2094 		if (dead == rt->rt_nhn)
2095 			rtm->rtm_flags |= RTNH_F_DEAD;
2096 
2097 		nla_nest_end(skb, mp);
2098 	}
2099 
2100 	nlmsg_end(skb, nlh);
2101 	return 0;
2102 
2103 nla_put_failure:
2104 	nlmsg_cancel(skb, nlh);
2105 	return -EMSGSIZE;
2106 }
2107 
2108 #if IS_ENABLED(CONFIG_INET)
2109 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2110 				   struct fib_dump_filter *filter,
2111 				   struct netlink_callback *cb)
2112 {
2113 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2114 }
2115 #else
2116 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2117 				   struct fib_dump_filter *filter,
2118 				   struct netlink_callback *cb)
2119 {
2120 	struct netlink_ext_ack *extack = cb->extack;
2121 	struct nlattr *tb[RTA_MAX + 1];
2122 	struct rtmsg *rtm;
2123 	int err, i;
2124 
2125 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
2126 	if (!rtm) {
2127 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2128 		return -EINVAL;
2129 	}
2130 
2131 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2132 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2133 	    rtm->rtm_flags) {
2134 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2135 		return -EINVAL;
2136 	}
2137 
2138 	if (rtm->rtm_protocol) {
2139 		filter->protocol = rtm->rtm_protocol;
2140 		filter->filter_set = 1;
2141 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2142 	}
2143 
2144 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2145 					    rtm_mpls_policy, extack);
2146 	if (err < 0)
2147 		return err;
2148 
2149 	for (i = 0; i <= RTA_MAX; ++i) {
2150 		int ifindex;
2151 
2152 		if (i == RTA_OIF) {
2153 			ifindex = nla_get_u32(tb[i]);
2154 			filter->dev = __dev_get_by_index(net, ifindex);
2155 			if (!filter->dev)
2156 				return -ENODEV;
2157 			filter->filter_set = 1;
2158 		} else if (tb[i]) {
2159 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2160 			return -EINVAL;
2161 		}
2162 	}
2163 
2164 	return 0;
2165 }
2166 #endif
2167 
2168 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2169 			     const struct net_device *dev)
2170 {
2171 	if (rt->rt_nhn == 1) {
2172 		struct mpls_nh *nh = rt->rt_nh;
2173 
2174 		if (nh->nh_dev == dev)
2175 			return true;
2176 	} else {
2177 		for_nexthops(rt) {
2178 			if (nh->nh_dev == dev)
2179 				return true;
2180 		} endfor_nexthops(rt);
2181 	}
2182 
2183 	return false;
2184 }
2185 
2186 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2187 {
2188 	const struct nlmsghdr *nlh = cb->nlh;
2189 	struct net *net = sock_net(skb->sk);
2190 	struct mpls_route __rcu **platform_label;
2191 	struct fib_dump_filter filter = {
2192 		.rtnl_held = true,
2193 	};
2194 	unsigned int flags = NLM_F_MULTI;
2195 	size_t platform_labels;
2196 	unsigned int index;
2197 
2198 	ASSERT_RTNL();
2199 
2200 	if (cb->strict_check) {
2201 		int err;
2202 
2203 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2204 		if (err < 0)
2205 			return err;
2206 
2207 		/* for MPLS, there is only 1 table with fixed type and flags.
2208 		 * If either are set in the filter then return nothing.
2209 		 */
2210 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2211 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2212 		     filter.flags)
2213 			return skb->len;
2214 	}
2215 
2216 	index = cb->args[0];
2217 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2218 		index = MPLS_LABEL_FIRST_UNRESERVED;
2219 
2220 	platform_label = rtnl_dereference(net->mpls.platform_label);
2221 	platform_labels = net->mpls.platform_labels;
2222 
2223 	if (filter.filter_set)
2224 		flags |= NLM_F_DUMP_FILTERED;
2225 
2226 	for (; index < platform_labels; index++) {
2227 		struct mpls_route *rt;
2228 
2229 		rt = rtnl_dereference(platform_label[index]);
2230 		if (!rt)
2231 			continue;
2232 
2233 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2234 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2235 			continue;
2236 
2237 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2238 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2239 				    index, rt, flags) < 0)
2240 			break;
2241 	}
2242 	cb->args[0] = index;
2243 
2244 	return skb->len;
2245 }
2246 
2247 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2248 {
2249 	size_t payload =
2250 		NLMSG_ALIGN(sizeof(struct rtmsg))
2251 		+ nla_total_size(4)			/* RTA_DST */
2252 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2253 
2254 	if (rt->rt_nhn == 1) {
2255 		struct mpls_nh *nh = rt->rt_nh;
2256 
2257 		if (nh->nh_dev)
2258 			payload += nla_total_size(4); /* RTA_OIF */
2259 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2260 			payload += nla_total_size(2 + nh->nh_via_alen);
2261 		if (nh->nh_labels) /* RTA_NEWDST */
2262 			payload += nla_total_size(nh->nh_labels * 4);
2263 	} else {
2264 		/* each nexthop is packed in an attribute */
2265 		size_t nhsize = 0;
2266 
2267 		for_nexthops(rt) {
2268 			if (!nh->nh_dev)
2269 				continue;
2270 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2271 			/* RTA_VIA */
2272 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2273 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2274 			if (nh->nh_labels)
2275 				nhsize += nla_total_size(nh->nh_labels * 4);
2276 		} endfor_nexthops(rt);
2277 		/* nested attribute */
2278 		payload += nla_total_size(nhsize);
2279 	}
2280 
2281 	return payload;
2282 }
2283 
2284 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2285 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2286 		       unsigned int nlm_flags)
2287 {
2288 	struct sk_buff *skb;
2289 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2290 	int err = -ENOBUFS;
2291 
2292 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2293 	if (skb == NULL)
2294 		goto errout;
2295 
2296 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2297 	if (err < 0) {
2298 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2299 		WARN_ON(err == -EMSGSIZE);
2300 		kfree_skb(skb);
2301 		goto errout;
2302 	}
2303 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2304 
2305 	return;
2306 errout:
2307 	rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2308 }
2309 
2310 static int mpls_valid_getroute_req(struct sk_buff *skb,
2311 				   const struct nlmsghdr *nlh,
2312 				   struct nlattr **tb,
2313 				   struct netlink_ext_ack *extack)
2314 {
2315 	struct rtmsg *rtm;
2316 	int i, err;
2317 
2318 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
2319 	if (!rtm) {
2320 		NL_SET_ERR_MSG_MOD(extack,
2321 				   "Invalid header for get route request");
2322 		return -EINVAL;
2323 	}
2324 
2325 	if (!netlink_strict_get_check(skb))
2326 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2327 					      rtm_mpls_policy, extack);
2328 
2329 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2330 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2331 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2332 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2333 		return -EINVAL;
2334 	}
2335 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2336 		NL_SET_ERR_MSG_MOD(extack,
2337 				   "Invalid flags for get route request");
2338 		return -EINVAL;
2339 	}
2340 
2341 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2342 					    rtm_mpls_policy, extack);
2343 	if (err)
2344 		return err;
2345 
2346 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2347 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2348 		return -EINVAL;
2349 	}
2350 
2351 	for (i = 0; i <= RTA_MAX; i++) {
2352 		if (!tb[i])
2353 			continue;
2354 
2355 		switch (i) {
2356 		case RTA_DST:
2357 		case RTA_NEWDST:
2358 			break;
2359 		default:
2360 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2361 			return -EINVAL;
2362 		}
2363 	}
2364 
2365 	return 0;
2366 }
2367 
2368 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2369 			 struct netlink_ext_ack *extack)
2370 {
2371 	struct net *net = sock_net(in_skb->sk);
2372 	u32 portid = NETLINK_CB(in_skb).portid;
2373 	u32 in_label = LABEL_NOT_SPECIFIED;
2374 	struct nlattr *tb[RTA_MAX + 1];
2375 	u32 labels[MAX_NEW_LABELS];
2376 	struct mpls_shim_hdr *hdr;
2377 	unsigned int hdr_size = 0;
2378 	const struct mpls_nh *nh;
2379 	struct net_device *dev;
2380 	struct mpls_route *rt;
2381 	struct rtmsg *rtm, *r;
2382 	struct nlmsghdr *nlh;
2383 	struct sk_buff *skb;
2384 	u8 n_labels;
2385 	int err;
2386 
2387 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2388 	if (err < 0)
2389 		goto errout;
2390 
2391 	rtm = nlmsg_data(in_nlh);
2392 
2393 	if (tb[RTA_DST]) {
2394 		u8 label_count;
2395 
2396 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2397 				   &in_label, extack)) {
2398 			err = -EINVAL;
2399 			goto errout;
2400 		}
2401 
2402 		if (!mpls_label_ok(net, &in_label, extack)) {
2403 			err = -EINVAL;
2404 			goto errout;
2405 		}
2406 	}
2407 
2408 	rt = mpls_route_input_rcu(net, in_label);
2409 	if (!rt) {
2410 		err = -ENETUNREACH;
2411 		goto errout;
2412 	}
2413 
2414 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2415 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2416 		if (!skb) {
2417 			err = -ENOBUFS;
2418 			goto errout;
2419 		}
2420 
2421 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2422 				      RTM_NEWROUTE, in_label, rt, 0);
2423 		if (err < 0) {
2424 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2425 			WARN_ON(err == -EMSGSIZE);
2426 			goto errout_free;
2427 		}
2428 
2429 		return rtnl_unicast(skb, net, portid);
2430 	}
2431 
2432 	if (tb[RTA_NEWDST]) {
2433 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2434 				   labels, extack) != 0) {
2435 			err = -EINVAL;
2436 			goto errout;
2437 		}
2438 
2439 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2440 	}
2441 
2442 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2443 	if (!skb) {
2444 		err = -ENOBUFS;
2445 		goto errout;
2446 	}
2447 
2448 	skb->protocol = htons(ETH_P_MPLS_UC);
2449 
2450 	if (hdr_size) {
2451 		bool bos;
2452 		int i;
2453 
2454 		if (skb_cow(skb, hdr_size)) {
2455 			err = -ENOBUFS;
2456 			goto errout_free;
2457 		}
2458 
2459 		skb_reserve(skb, hdr_size);
2460 		skb_push(skb, hdr_size);
2461 		skb_reset_network_header(skb);
2462 
2463 		/* Push new labels */
2464 		hdr = mpls_hdr(skb);
2465 		bos = true;
2466 		for (i = n_labels - 1; i >= 0; i--) {
2467 			hdr[i] = mpls_entry_encode(labels[i],
2468 						   1, 0, bos);
2469 			bos = false;
2470 		}
2471 	}
2472 
2473 	nh = mpls_select_multipath(rt, skb);
2474 	if (!nh) {
2475 		err = -ENETUNREACH;
2476 		goto errout_free;
2477 	}
2478 
2479 	if (hdr_size) {
2480 		skb_pull(skb, hdr_size);
2481 		skb_reset_network_header(skb);
2482 	}
2483 
2484 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2485 			RTM_NEWROUTE, sizeof(*r), 0);
2486 	if (!nlh) {
2487 		err = -EMSGSIZE;
2488 		goto errout_free;
2489 	}
2490 
2491 	r = nlmsg_data(nlh);
2492 	r->rtm_family	 = AF_MPLS;
2493 	r->rtm_dst_len	= 20;
2494 	r->rtm_src_len	= 0;
2495 	r->rtm_table	= RT_TABLE_MAIN;
2496 	r->rtm_type	= RTN_UNICAST;
2497 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2498 	r->rtm_protocol = rt->rt_protocol;
2499 	r->rtm_flags	= 0;
2500 
2501 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2502 		goto nla_put_failure;
2503 
2504 	if (nh->nh_labels &&
2505 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2506 			   nh->nh_label))
2507 		goto nla_put_failure;
2508 
2509 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2510 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2511 			nh->nh_via_alen))
2512 		goto nla_put_failure;
2513 	dev = nh->nh_dev;
2514 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2515 		goto nla_put_failure;
2516 
2517 	nlmsg_end(skb, nlh);
2518 
2519 	err = rtnl_unicast(skb, net, portid);
2520 errout:
2521 	return err;
2522 
2523 nla_put_failure:
2524 	nlmsg_cancel(skb, nlh);
2525 	err = -EMSGSIZE;
2526 errout_free:
2527 	kfree_skb(skb);
2528 	return err;
2529 }
2530 
2531 static int resize_platform_label_table(struct net *net, size_t limit)
2532 {
2533 	size_t size = sizeof(struct mpls_route *) * limit;
2534 	size_t old_limit;
2535 	size_t cp_size;
2536 	struct mpls_route __rcu **labels = NULL, **old;
2537 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2538 	unsigned index;
2539 
2540 	if (size) {
2541 		labels = kvzalloc(size, GFP_KERNEL);
2542 		if (!labels)
2543 			goto nolabels;
2544 	}
2545 
2546 	/* In case the predefined labels need to be populated */
2547 	if (limit > MPLS_LABEL_IPV4NULL) {
2548 		struct net_device *lo = net->loopback_dev;
2549 
2550 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2551 		if (IS_ERR(rt0))
2552 			goto nort0;
2553 
2554 		rt0->rt_nh->nh_dev = lo;
2555 		netdev_hold(lo, &rt0->rt_nh->nh_dev_tracker, GFP_KERNEL);
2556 		rt0->rt_protocol = RTPROT_KERNEL;
2557 		rt0->rt_payload_type = MPT_IPV4;
2558 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2559 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2560 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2561 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2562 		       lo->addr_len);
2563 	}
2564 	if (limit > MPLS_LABEL_IPV6NULL) {
2565 		struct net_device *lo = net->loopback_dev;
2566 
2567 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2568 		if (IS_ERR(rt2))
2569 			goto nort2;
2570 
2571 		rt2->rt_nh->nh_dev = lo;
2572 		netdev_hold(lo, &rt2->rt_nh->nh_dev_tracker, GFP_KERNEL);
2573 		rt2->rt_protocol = RTPROT_KERNEL;
2574 		rt2->rt_payload_type = MPT_IPV6;
2575 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2576 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2577 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2578 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2579 		       lo->addr_len);
2580 	}
2581 
2582 	rtnl_lock();
2583 	/* Remember the original table */
2584 	old = rtnl_dereference(net->mpls.platform_label);
2585 	old_limit = net->mpls.platform_labels;
2586 
2587 	/* Free any labels beyond the new table */
2588 	for (index = limit; index < old_limit; index++)
2589 		mpls_route_update(net, index, NULL, NULL);
2590 
2591 	/* Copy over the old labels */
2592 	cp_size = size;
2593 	if (old_limit < limit)
2594 		cp_size = old_limit * sizeof(struct mpls_route *);
2595 
2596 	memcpy(labels, old, cp_size);
2597 
2598 	/* If needed set the predefined labels */
2599 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2600 	    (limit > MPLS_LABEL_IPV6NULL)) {
2601 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2602 		rt2 = NULL;
2603 	}
2604 
2605 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2606 	    (limit > MPLS_LABEL_IPV4NULL)) {
2607 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2608 		rt0 = NULL;
2609 	}
2610 
2611 	/* Update the global pointers */
2612 	net->mpls.platform_labels = limit;
2613 	rcu_assign_pointer(net->mpls.platform_label, labels);
2614 
2615 	rtnl_unlock();
2616 
2617 	mpls_rt_free(rt2);
2618 	mpls_rt_free(rt0);
2619 
2620 	if (old) {
2621 		synchronize_rcu();
2622 		kvfree(old);
2623 	}
2624 	return 0;
2625 
2626 nort2:
2627 	mpls_rt_free(rt0);
2628 nort0:
2629 	kvfree(labels);
2630 nolabels:
2631 	return -ENOMEM;
2632 }
2633 
2634 static int mpls_platform_labels(const struct ctl_table *table, int write,
2635 				void *buffer, size_t *lenp, loff_t *ppos)
2636 {
2637 	struct net *net = table->data;
2638 	int platform_labels = net->mpls.platform_labels;
2639 	int ret;
2640 	struct ctl_table tmp = {
2641 		.procname	= table->procname,
2642 		.data		= &platform_labels,
2643 		.maxlen		= sizeof(int),
2644 		.mode		= table->mode,
2645 		.extra1		= SYSCTL_ZERO,
2646 		.extra2		= &label_limit,
2647 	};
2648 
2649 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2650 
2651 	if (write && ret == 0)
2652 		ret = resize_platform_label_table(net, platform_labels);
2653 
2654 	return ret;
2655 }
2656 
2657 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2658 	(&((struct net *)0)->field)
2659 
2660 static const struct ctl_table mpls_table[] = {
2661 	{
2662 		.procname	= "platform_labels",
2663 		.data		= NULL,
2664 		.maxlen		= sizeof(int),
2665 		.mode		= 0644,
2666 		.proc_handler	= mpls_platform_labels,
2667 	},
2668 	{
2669 		.procname	= "ip_ttl_propagate",
2670 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2671 		.maxlen		= sizeof(int),
2672 		.mode		= 0644,
2673 		.proc_handler	= proc_dointvec_minmax,
2674 		.extra1		= SYSCTL_ZERO,
2675 		.extra2		= SYSCTL_ONE,
2676 	},
2677 	{
2678 		.procname	= "default_ttl",
2679 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2680 		.maxlen		= sizeof(int),
2681 		.mode		= 0644,
2682 		.proc_handler	= proc_dointvec_minmax,
2683 		.extra1		= SYSCTL_ONE,
2684 		.extra2		= &ttl_max,
2685 	},
2686 };
2687 
2688 static int mpls_net_init(struct net *net)
2689 {
2690 	size_t table_size = ARRAY_SIZE(mpls_table);
2691 	struct ctl_table *table;
2692 	int i;
2693 
2694 	net->mpls.platform_labels = 0;
2695 	net->mpls.platform_label = NULL;
2696 	net->mpls.ip_ttl_propagate = 1;
2697 	net->mpls.default_ttl = 255;
2698 
2699 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2700 	if (table == NULL)
2701 		return -ENOMEM;
2702 
2703 	/* Table data contains only offsets relative to the base of
2704 	 * the mdev at this point, so make them absolute.
2705 	 */
2706 	for (i = 0; i < table_size; i++)
2707 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2708 
2709 	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2710 					       table_size);
2711 	if (net->mpls.ctl == NULL) {
2712 		kfree(table);
2713 		return -ENOMEM;
2714 	}
2715 
2716 	return 0;
2717 }
2718 
2719 static void mpls_net_exit(struct net *net)
2720 {
2721 	struct mpls_route __rcu **platform_label;
2722 	size_t platform_labels;
2723 	const struct ctl_table *table;
2724 	unsigned int index;
2725 
2726 	table = net->mpls.ctl->ctl_table_arg;
2727 	unregister_net_sysctl_table(net->mpls.ctl);
2728 	kfree(table);
2729 
2730 	/* An rcu grace period has passed since there was a device in
2731 	 * the network namespace (and thus the last in flight packet)
2732 	 * left this network namespace.  This is because
2733 	 * unregister_netdevice_many and netdev_run_todo has completed
2734 	 * for each network device that was in this network namespace.
2735 	 *
2736 	 * As such no additional rcu synchronization is necessary when
2737 	 * freeing the platform_label table.
2738 	 */
2739 	rtnl_lock();
2740 	platform_label = rtnl_dereference(net->mpls.platform_label);
2741 	platform_labels = net->mpls.platform_labels;
2742 	for (index = 0; index < platform_labels; index++) {
2743 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2744 		RCU_INIT_POINTER(platform_label[index], NULL);
2745 		mpls_notify_route(net, index, rt, NULL, NULL);
2746 		mpls_rt_free(rt);
2747 	}
2748 	rtnl_unlock();
2749 
2750 	kvfree(platform_label);
2751 }
2752 
2753 static struct pernet_operations mpls_net_ops = {
2754 	.init = mpls_net_init,
2755 	.exit = mpls_net_exit,
2756 };
2757 
2758 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2759 	.family		   = AF_MPLS,
2760 	.fill_stats_af	   = mpls_fill_stats_af,
2761 	.get_stats_af_size = mpls_get_stats_af_size,
2762 };
2763 
2764 static const struct rtnl_msg_handler mpls_rtnl_msg_handlers[] __initdata_or_module = {
2765 	{THIS_MODULE, PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, 0},
2766 	{THIS_MODULE, PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, 0},
2767 	{THIS_MODULE, PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes, 0},
2768 	{THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2769 	 mpls_netconf_get_devconf, mpls_netconf_dump_devconf,
2770 	 RTNL_FLAG_DUMP_UNLOCKED},
2771 };
2772 
2773 static int __init mpls_init(void)
2774 {
2775 	int err;
2776 
2777 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2778 
2779 	err = register_pernet_subsys(&mpls_net_ops);
2780 	if (err)
2781 		goto out;
2782 
2783 	err = register_netdevice_notifier(&mpls_dev_notifier);
2784 	if (err)
2785 		goto out_unregister_pernet;
2786 
2787 	dev_add_pack(&mpls_packet_type);
2788 
2789 	err = rtnl_af_register(&mpls_af_ops);
2790 	if (err)
2791 		goto out_unregister_dev_type;
2792 
2793 	err = rtnl_register_many(mpls_rtnl_msg_handlers);
2794 	if (err)
2795 		goto out_unregister_rtnl_af;
2796 
2797 	err = ipgre_tunnel_encap_add_mpls_ops();
2798 	if (err) {
2799 		pr_err("Can't add mpls over gre tunnel ops\n");
2800 		goto out_unregister_rtnl;
2801 	}
2802 
2803 	err = 0;
2804 out:
2805 	return err;
2806 
2807 out_unregister_rtnl:
2808 	rtnl_unregister_many(mpls_rtnl_msg_handlers);
2809 out_unregister_rtnl_af:
2810 	rtnl_af_unregister(&mpls_af_ops);
2811 out_unregister_dev_type:
2812 	dev_remove_pack(&mpls_packet_type);
2813 out_unregister_pernet:
2814 	unregister_pernet_subsys(&mpls_net_ops);
2815 	goto out;
2816 }
2817 module_init(mpls_init);
2818 
2819 static void __exit mpls_exit(void)
2820 {
2821 	rtnl_unregister_all(PF_MPLS);
2822 	rtnl_af_unregister(&mpls_af_ops);
2823 	dev_remove_pack(&mpls_packet_type);
2824 	unregister_netdevice_notifier(&mpls_dev_notifier);
2825 	unregister_pernet_subsys(&mpls_net_ops);
2826 	ipgre_tunnel_encap_del_mpls_ops();
2827 }
2828 module_exit(mpls_exit);
2829 
2830 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2831 MODULE_LICENSE("GPL v2");
2832 MODULE_ALIAS_NETPROTO(PF_MPLS);
2833