xref: /linux/net/mpls/af_mpls.c (revision b8d312aa075f33282565467662c4628dae0a2aff)
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/ip.h>
16 #include <net/dst.h>
17 #include <net/sock.h>
18 #include <net/arp.h>
19 #include <net/ip_fib.h>
20 #include <net/netevent.h>
21 #include <net/ip_tunnels.h>
22 #include <net/netns/generic.h>
23 #if IS_ENABLED(CONFIG_IPV6)
24 #include <net/ipv6.h>
25 #endif
26 #include <net/ipv6_stubs.h>
27 #include <net/rtnh.h>
28 #include "internal.h"
29 
30 /* max memory we will use for mpls_route */
31 #define MAX_MPLS_ROUTE_MEM	4096
32 
33 /* Maximum number of labels to look ahead at when selecting a path of
34  * a multipath route
35  */
36 #define MAX_MP_SELECT_LABELS 4
37 
38 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
39 
40 static int label_limit = (1 << 20) - 1;
41 static int ttl_max = 255;
42 
43 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
44 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
45 {
46 	return sizeof(struct mpls_shim_hdr);
47 }
48 
49 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
50 	.encap_hlen	= ipgre_mpls_encap_hlen,
51 };
52 
53 static int ipgre_tunnel_encap_add_mpls_ops(void)
54 {
55 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
56 }
57 
58 static void ipgre_tunnel_encap_del_mpls_ops(void)
59 {
60 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
61 }
62 #else
63 static int ipgre_tunnel_encap_add_mpls_ops(void)
64 {
65 	return 0;
66 }
67 
68 static void ipgre_tunnel_encap_del_mpls_ops(void)
69 {
70 }
71 #endif
72 
73 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
74 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
75 		       unsigned int nlm_flags);
76 
77 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
78 {
79 	struct mpls_route *rt = NULL;
80 
81 	if (index < net->mpls.platform_labels) {
82 		struct mpls_route __rcu **platform_label =
83 			rcu_dereference(net->mpls.platform_label);
84 		rt = rcu_dereference(platform_label[index]);
85 	}
86 	return rt;
87 }
88 
89 bool mpls_output_possible(const struct net_device *dev)
90 {
91 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
92 }
93 EXPORT_SYMBOL_GPL(mpls_output_possible);
94 
95 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
96 {
97 	return (u8 *)nh + rt->rt_via_offset;
98 }
99 
100 static const u8 *mpls_nh_via(const struct mpls_route *rt,
101 			     const struct mpls_nh *nh)
102 {
103 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
104 }
105 
106 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
107 {
108 	/* The size of the layer 2.5 labels to be added for this route */
109 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
110 }
111 
112 unsigned int mpls_dev_mtu(const struct net_device *dev)
113 {
114 	/* The amount of data the layer 2 frame can hold */
115 	return dev->mtu;
116 }
117 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
118 
119 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
120 {
121 	if (skb->len <= mtu)
122 		return false;
123 
124 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
125 		return false;
126 
127 	return true;
128 }
129 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
130 
131 void mpls_stats_inc_outucastpkts(struct net_device *dev,
132 				 const struct sk_buff *skb)
133 {
134 	struct mpls_dev *mdev;
135 
136 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
137 		mdev = mpls_dev_get(dev);
138 		if (mdev)
139 			MPLS_INC_STATS_LEN(mdev, skb->len,
140 					   tx_packets,
141 					   tx_bytes);
142 	} else if (skb->protocol == htons(ETH_P_IP)) {
143 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
144 #if IS_ENABLED(CONFIG_IPV6)
145 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
146 		struct inet6_dev *in6dev = __in6_dev_get(dev);
147 
148 		if (in6dev)
149 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
150 					 IPSTATS_MIB_OUT, skb->len);
151 #endif
152 	}
153 }
154 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
155 
156 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
157 {
158 	struct mpls_entry_decoded dec;
159 	unsigned int mpls_hdr_len = 0;
160 	struct mpls_shim_hdr *hdr;
161 	bool eli_seen = false;
162 	int label_index;
163 	u32 hash = 0;
164 
165 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
166 	     label_index++) {
167 		mpls_hdr_len += sizeof(*hdr);
168 		if (!pskb_may_pull(skb, mpls_hdr_len))
169 			break;
170 
171 		/* Read and decode the current label */
172 		hdr = mpls_hdr(skb) + label_index;
173 		dec = mpls_entry_decode(hdr);
174 
175 		/* RFC6790 - reserved labels MUST NOT be used as keys
176 		 * for the load-balancing function
177 		 */
178 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
179 			hash = jhash_1word(dec.label, hash);
180 
181 			/* The entropy label follows the entropy label
182 			 * indicator, so this means that the entropy
183 			 * label was just added to the hash - no need to
184 			 * go any deeper either in the label stack or in the
185 			 * payload
186 			 */
187 			if (eli_seen)
188 				break;
189 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
190 			eli_seen = true;
191 		}
192 
193 		if (!dec.bos)
194 			continue;
195 
196 		/* found bottom label; does skb have room for a header? */
197 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
198 			const struct iphdr *v4hdr;
199 
200 			v4hdr = (const struct iphdr *)(hdr + 1);
201 			if (v4hdr->version == 4) {
202 				hash = jhash_3words(ntohl(v4hdr->saddr),
203 						    ntohl(v4hdr->daddr),
204 						    v4hdr->protocol, hash);
205 			} else if (v4hdr->version == 6 &&
206 				   pskb_may_pull(skb, mpls_hdr_len +
207 						 sizeof(struct ipv6hdr))) {
208 				const struct ipv6hdr *v6hdr;
209 
210 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
211 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
212 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
213 				hash = jhash_1word(v6hdr->nexthdr, hash);
214 			}
215 		}
216 
217 		break;
218 	}
219 
220 	return hash;
221 }
222 
223 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
224 {
225 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
226 }
227 
228 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
229  * a next hop (nh->nh_flags) are modified by netdev event handlers.
230  * Since those fields can change at any moment, use READ_ONCE to
231  * access both.
232  */
233 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
234 					     struct sk_buff *skb)
235 {
236 	u32 hash = 0;
237 	int nh_index = 0;
238 	int n = 0;
239 	u8 alive;
240 
241 	/* No need to look further into packet if there's only
242 	 * one path
243 	 */
244 	if (rt->rt_nhn == 1)
245 		return rt->rt_nh;
246 
247 	alive = READ_ONCE(rt->rt_nhn_alive);
248 	if (alive == 0)
249 		return NULL;
250 
251 	hash = mpls_multipath_hash(rt, skb);
252 	nh_index = hash % alive;
253 	if (alive == rt->rt_nhn)
254 		goto out;
255 	for_nexthops(rt) {
256 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
257 
258 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
259 			continue;
260 		if (n == nh_index)
261 			return nh;
262 		n++;
263 	} endfor_nexthops(rt);
264 
265 out:
266 	return mpls_get_nexthop(rt, nh_index);
267 }
268 
269 static bool mpls_egress(struct net *net, struct mpls_route *rt,
270 			struct sk_buff *skb, struct mpls_entry_decoded dec)
271 {
272 	enum mpls_payload_type payload_type;
273 	bool success = false;
274 
275 	/* The IPv4 code below accesses through the IPv4 header
276 	 * checksum, which is 12 bytes into the packet.
277 	 * The IPv6 code below accesses through the IPv6 hop limit
278 	 * which is 8 bytes into the packet.
279 	 *
280 	 * For all supported cases there should always be at least 12
281 	 * bytes of packet data present.  The IPv4 header is 20 bytes
282 	 * without options and the IPv6 header is always 40 bytes
283 	 * long.
284 	 */
285 	if (!pskb_may_pull(skb, 12))
286 		return false;
287 
288 	payload_type = rt->rt_payload_type;
289 	if (payload_type == MPT_UNSPEC)
290 		payload_type = ip_hdr(skb)->version;
291 
292 	switch (payload_type) {
293 	case MPT_IPV4: {
294 		struct iphdr *hdr4 = ip_hdr(skb);
295 		u8 new_ttl;
296 		skb->protocol = htons(ETH_P_IP);
297 
298 		/* If propagating TTL, take the decremented TTL from
299 		 * the incoming MPLS header, otherwise decrement the
300 		 * TTL, but only if not 0 to avoid underflow.
301 		 */
302 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
303 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
304 		     net->mpls.ip_ttl_propagate))
305 			new_ttl = dec.ttl;
306 		else
307 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
308 
309 		csum_replace2(&hdr4->check,
310 			      htons(hdr4->ttl << 8),
311 			      htons(new_ttl << 8));
312 		hdr4->ttl = new_ttl;
313 		success = true;
314 		break;
315 	}
316 	case MPT_IPV6: {
317 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
318 		skb->protocol = htons(ETH_P_IPV6);
319 
320 		/* If propagating TTL, take the decremented TTL from
321 		 * the incoming MPLS header, otherwise decrement the
322 		 * hop limit, but only if not 0 to avoid underflow.
323 		 */
324 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
325 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
326 		     net->mpls.ip_ttl_propagate))
327 			hdr6->hop_limit = dec.ttl;
328 		else if (hdr6->hop_limit)
329 			hdr6->hop_limit = hdr6->hop_limit - 1;
330 		success = true;
331 		break;
332 	}
333 	case MPT_UNSPEC:
334 		/* Should have decided which protocol it is by now */
335 		break;
336 	}
337 
338 	return success;
339 }
340 
341 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
342 			struct packet_type *pt, struct net_device *orig_dev)
343 {
344 	struct net *net = dev_net(dev);
345 	struct mpls_shim_hdr *hdr;
346 	struct mpls_route *rt;
347 	struct mpls_nh *nh;
348 	struct mpls_entry_decoded dec;
349 	struct net_device *out_dev;
350 	struct mpls_dev *out_mdev;
351 	struct mpls_dev *mdev;
352 	unsigned int hh_len;
353 	unsigned int new_header_size;
354 	unsigned int mtu;
355 	int err;
356 
357 	/* Careful this entire function runs inside of an rcu critical section */
358 
359 	mdev = mpls_dev_get(dev);
360 	if (!mdev)
361 		goto drop;
362 
363 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
364 			   rx_bytes);
365 
366 	if (!mdev->input_enabled) {
367 		MPLS_INC_STATS(mdev, rx_dropped);
368 		goto drop;
369 	}
370 
371 	if (skb->pkt_type != PACKET_HOST)
372 		goto err;
373 
374 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
375 		goto err;
376 
377 	if (!pskb_may_pull(skb, sizeof(*hdr)))
378 		goto err;
379 
380 	/* Read and decode the label */
381 	hdr = mpls_hdr(skb);
382 	dec = mpls_entry_decode(hdr);
383 
384 	rt = mpls_route_input_rcu(net, dec.label);
385 	if (!rt) {
386 		MPLS_INC_STATS(mdev, rx_noroute);
387 		goto drop;
388 	}
389 
390 	nh = mpls_select_multipath(rt, skb);
391 	if (!nh)
392 		goto err;
393 
394 	/* Pop the label */
395 	skb_pull(skb, sizeof(*hdr));
396 	skb_reset_network_header(skb);
397 
398 	skb_orphan(skb);
399 
400 	if (skb_warn_if_lro(skb))
401 		goto err;
402 
403 	skb_forward_csum(skb);
404 
405 	/* Verify ttl is valid */
406 	if (dec.ttl <= 1)
407 		goto err;
408 	dec.ttl -= 1;
409 
410 	/* Find the output device */
411 	out_dev = rcu_dereference(nh->nh_dev);
412 	if (!mpls_output_possible(out_dev))
413 		goto tx_err;
414 
415 	/* Verify the destination can hold the packet */
416 	new_header_size = mpls_nh_header_size(nh);
417 	mtu = mpls_dev_mtu(out_dev);
418 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
419 		goto tx_err;
420 
421 	hh_len = LL_RESERVED_SPACE(out_dev);
422 	if (!out_dev->header_ops)
423 		hh_len = 0;
424 
425 	/* Ensure there is enough space for the headers in the skb */
426 	if (skb_cow(skb, hh_len + new_header_size))
427 		goto tx_err;
428 
429 	skb->dev = out_dev;
430 	skb->protocol = htons(ETH_P_MPLS_UC);
431 
432 	if (unlikely(!new_header_size && dec.bos)) {
433 		/* Penultimate hop popping */
434 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
435 			goto err;
436 	} else {
437 		bool bos;
438 		int i;
439 		skb_push(skb, new_header_size);
440 		skb_reset_network_header(skb);
441 		/* Push the new labels */
442 		hdr = mpls_hdr(skb);
443 		bos = dec.bos;
444 		for (i = nh->nh_labels - 1; i >= 0; i--) {
445 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
446 						   dec.ttl, 0, bos);
447 			bos = false;
448 		}
449 	}
450 
451 	mpls_stats_inc_outucastpkts(out_dev, skb);
452 
453 	/* If via wasn't specified then send out using device address */
454 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
455 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
456 				 out_dev->dev_addr, skb);
457 	else
458 		err = neigh_xmit(nh->nh_via_table, out_dev,
459 				 mpls_nh_via(rt, nh), skb);
460 	if (err)
461 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
462 				    __func__, err);
463 	return 0;
464 
465 tx_err:
466 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
467 	if (out_mdev)
468 		MPLS_INC_STATS(out_mdev, tx_errors);
469 	goto drop;
470 err:
471 	MPLS_INC_STATS(mdev, rx_errors);
472 drop:
473 	kfree_skb(skb);
474 	return NET_RX_DROP;
475 }
476 
477 static struct packet_type mpls_packet_type __read_mostly = {
478 	.type = cpu_to_be16(ETH_P_MPLS_UC),
479 	.func = mpls_forward,
480 };
481 
482 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
483 	[RTA_DST]		= { .type = NLA_U32 },
484 	[RTA_OIF]		= { .type = NLA_U32 },
485 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
486 };
487 
488 struct mpls_route_config {
489 	u32			rc_protocol;
490 	u32			rc_ifindex;
491 	u8			rc_via_table;
492 	u8			rc_via_alen;
493 	u8			rc_via[MAX_VIA_ALEN];
494 	u32			rc_label;
495 	u8			rc_ttl_propagate;
496 	u8			rc_output_labels;
497 	u32			rc_output_label[MAX_NEW_LABELS];
498 	u32			rc_nlflags;
499 	enum mpls_payload_type	rc_payload_type;
500 	struct nl_info		rc_nlinfo;
501 	struct rtnexthop	*rc_mp;
502 	int			rc_mp_len;
503 };
504 
505 /* all nexthops within a route have the same size based on max
506  * number of labels and max via length for a hop
507  */
508 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
509 {
510 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
511 	struct mpls_route *rt;
512 	size_t size;
513 
514 	size = sizeof(*rt) + num_nh * nh_size;
515 	if (size > MAX_MPLS_ROUTE_MEM)
516 		return ERR_PTR(-EINVAL);
517 
518 	rt = kzalloc(size, GFP_KERNEL);
519 	if (!rt)
520 		return ERR_PTR(-ENOMEM);
521 
522 	rt->rt_nhn = num_nh;
523 	rt->rt_nhn_alive = num_nh;
524 	rt->rt_nh_size = nh_size;
525 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
526 
527 	return rt;
528 }
529 
530 static void mpls_rt_free(struct mpls_route *rt)
531 {
532 	if (rt)
533 		kfree_rcu(rt, rt_rcu);
534 }
535 
536 static void mpls_notify_route(struct net *net, unsigned index,
537 			      struct mpls_route *old, struct mpls_route *new,
538 			      const struct nl_info *info)
539 {
540 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
541 	unsigned portid = info ? info->portid : 0;
542 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
543 	struct mpls_route *rt = new ? new : old;
544 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
545 	/* Ignore reserved labels for now */
546 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
547 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
548 }
549 
550 static void mpls_route_update(struct net *net, unsigned index,
551 			      struct mpls_route *new,
552 			      const struct nl_info *info)
553 {
554 	struct mpls_route __rcu **platform_label;
555 	struct mpls_route *rt;
556 
557 	ASSERT_RTNL();
558 
559 	platform_label = rtnl_dereference(net->mpls.platform_label);
560 	rt = rtnl_dereference(platform_label[index]);
561 	rcu_assign_pointer(platform_label[index], new);
562 
563 	mpls_notify_route(net, index, rt, new, info);
564 
565 	/* If we removed a route free it now */
566 	mpls_rt_free(rt);
567 }
568 
569 static unsigned find_free_label(struct net *net)
570 {
571 	struct mpls_route __rcu **platform_label;
572 	size_t platform_labels;
573 	unsigned index;
574 
575 	platform_label = rtnl_dereference(net->mpls.platform_label);
576 	platform_labels = net->mpls.platform_labels;
577 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
578 	     index++) {
579 		if (!rtnl_dereference(platform_label[index]))
580 			return index;
581 	}
582 	return LABEL_NOT_SPECIFIED;
583 }
584 
585 #if IS_ENABLED(CONFIG_INET)
586 static struct net_device *inet_fib_lookup_dev(struct net *net,
587 					      const void *addr)
588 {
589 	struct net_device *dev;
590 	struct rtable *rt;
591 	struct in_addr daddr;
592 
593 	memcpy(&daddr, addr, sizeof(struct in_addr));
594 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
595 	if (IS_ERR(rt))
596 		return ERR_CAST(rt);
597 
598 	dev = rt->dst.dev;
599 	dev_hold(dev);
600 
601 	ip_rt_put(rt);
602 
603 	return dev;
604 }
605 #else
606 static struct net_device *inet_fib_lookup_dev(struct net *net,
607 					      const void *addr)
608 {
609 	return ERR_PTR(-EAFNOSUPPORT);
610 }
611 #endif
612 
613 #if IS_ENABLED(CONFIG_IPV6)
614 static struct net_device *inet6_fib_lookup_dev(struct net *net,
615 					       const void *addr)
616 {
617 	struct net_device *dev;
618 	struct dst_entry *dst;
619 	struct flowi6 fl6;
620 	int err;
621 
622 	if (!ipv6_stub)
623 		return ERR_PTR(-EAFNOSUPPORT);
624 
625 	memset(&fl6, 0, sizeof(fl6));
626 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
627 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
628 	if (err)
629 		return ERR_PTR(err);
630 
631 	dev = dst->dev;
632 	dev_hold(dev);
633 	dst_release(dst);
634 
635 	return dev;
636 }
637 #else
638 static struct net_device *inet6_fib_lookup_dev(struct net *net,
639 					       const void *addr)
640 {
641 	return ERR_PTR(-EAFNOSUPPORT);
642 }
643 #endif
644 
645 static struct net_device *find_outdev(struct net *net,
646 				      struct mpls_route *rt,
647 				      struct mpls_nh *nh, int oif)
648 {
649 	struct net_device *dev = NULL;
650 
651 	if (!oif) {
652 		switch (nh->nh_via_table) {
653 		case NEIGH_ARP_TABLE:
654 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
655 			break;
656 		case NEIGH_ND_TABLE:
657 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
658 			break;
659 		case NEIGH_LINK_TABLE:
660 			break;
661 		}
662 	} else {
663 		dev = dev_get_by_index(net, oif);
664 	}
665 
666 	if (!dev)
667 		return ERR_PTR(-ENODEV);
668 
669 	if (IS_ERR(dev))
670 		return dev;
671 
672 	/* The caller is holding rtnl anyways, so release the dev reference */
673 	dev_put(dev);
674 
675 	return dev;
676 }
677 
678 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
679 			      struct mpls_nh *nh, int oif)
680 {
681 	struct net_device *dev = NULL;
682 	int err = -ENODEV;
683 
684 	dev = find_outdev(net, rt, nh, oif);
685 	if (IS_ERR(dev)) {
686 		err = PTR_ERR(dev);
687 		dev = NULL;
688 		goto errout;
689 	}
690 
691 	/* Ensure this is a supported device */
692 	err = -EINVAL;
693 	if (!mpls_dev_get(dev))
694 		goto errout;
695 
696 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
697 	    (dev->addr_len != nh->nh_via_alen))
698 		goto errout;
699 
700 	RCU_INIT_POINTER(nh->nh_dev, dev);
701 
702 	if (!(dev->flags & IFF_UP)) {
703 		nh->nh_flags |= RTNH_F_DEAD;
704 	} else {
705 		unsigned int flags;
706 
707 		flags = dev_get_flags(dev);
708 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
709 			nh->nh_flags |= RTNH_F_LINKDOWN;
710 	}
711 
712 	return 0;
713 
714 errout:
715 	return err;
716 }
717 
718 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
719 		       u8 via_addr[], struct netlink_ext_ack *extack)
720 {
721 	struct rtvia *via = nla_data(nla);
722 	int err = -EINVAL;
723 	int alen;
724 
725 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
726 		NL_SET_ERR_MSG_ATTR(extack, nla,
727 				    "Invalid attribute length for RTA_VIA");
728 		goto errout;
729 	}
730 	alen = nla_len(nla) -
731 			offsetof(struct rtvia, rtvia_addr);
732 	if (alen > MAX_VIA_ALEN) {
733 		NL_SET_ERR_MSG_ATTR(extack, nla,
734 				    "Invalid address length for RTA_VIA");
735 		goto errout;
736 	}
737 
738 	/* Validate the address family */
739 	switch (via->rtvia_family) {
740 	case AF_PACKET:
741 		*via_table = NEIGH_LINK_TABLE;
742 		break;
743 	case AF_INET:
744 		*via_table = NEIGH_ARP_TABLE;
745 		if (alen != 4)
746 			goto errout;
747 		break;
748 	case AF_INET6:
749 		*via_table = NEIGH_ND_TABLE;
750 		if (alen != 16)
751 			goto errout;
752 		break;
753 	default:
754 		/* Unsupported address family */
755 		goto errout;
756 	}
757 
758 	memcpy(via_addr, via->rtvia_addr, alen);
759 	*via_alen = alen;
760 	err = 0;
761 
762 errout:
763 	return err;
764 }
765 
766 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
767 				  struct mpls_route *rt)
768 {
769 	struct net *net = cfg->rc_nlinfo.nl_net;
770 	struct mpls_nh *nh = rt->rt_nh;
771 	int err;
772 	int i;
773 
774 	if (!nh)
775 		return -ENOMEM;
776 
777 	nh->nh_labels = cfg->rc_output_labels;
778 	for (i = 0; i < nh->nh_labels; i++)
779 		nh->nh_label[i] = cfg->rc_output_label[i];
780 
781 	nh->nh_via_table = cfg->rc_via_table;
782 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
783 	nh->nh_via_alen = cfg->rc_via_alen;
784 
785 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
786 	if (err)
787 		goto errout;
788 
789 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
790 		rt->rt_nhn_alive--;
791 
792 	return 0;
793 
794 errout:
795 	return err;
796 }
797 
798 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
799 			 struct mpls_nh *nh, int oif, struct nlattr *via,
800 			 struct nlattr *newdst, u8 max_labels,
801 			 struct netlink_ext_ack *extack)
802 {
803 	int err = -ENOMEM;
804 
805 	if (!nh)
806 		goto errout;
807 
808 	if (newdst) {
809 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
810 				     nh->nh_label, extack);
811 		if (err)
812 			goto errout;
813 	}
814 
815 	if (via) {
816 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
817 				  __mpls_nh_via(rt, nh), extack);
818 		if (err)
819 			goto errout;
820 	} else {
821 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
822 	}
823 
824 	err = mpls_nh_assign_dev(net, rt, nh, oif);
825 	if (err)
826 		goto errout;
827 
828 	return 0;
829 
830 errout:
831 	return err;
832 }
833 
834 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
835 			      u8 cfg_via_alen, u8 *max_via_alen,
836 			      u8 *max_labels)
837 {
838 	int remaining = len;
839 	u8 nhs = 0;
840 
841 	*max_via_alen = 0;
842 	*max_labels = 0;
843 
844 	while (rtnh_ok(rtnh, remaining)) {
845 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
846 		int attrlen;
847 		u8 n_labels = 0;
848 
849 		attrlen = rtnh_attrlen(rtnh);
850 		nla = nla_find(attrs, attrlen, RTA_VIA);
851 		if (nla && nla_len(nla) >=
852 		    offsetof(struct rtvia, rtvia_addr)) {
853 			int via_alen = nla_len(nla) -
854 				offsetof(struct rtvia, rtvia_addr);
855 
856 			if (via_alen <= MAX_VIA_ALEN)
857 				*max_via_alen = max_t(u16, *max_via_alen,
858 						      via_alen);
859 		}
860 
861 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
862 		if (nla &&
863 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
864 				   NULL, NULL) != 0)
865 			return 0;
866 
867 		*max_labels = max_t(u8, *max_labels, n_labels);
868 
869 		/* number of nexthops is tracked by a u8.
870 		 * Check for overflow.
871 		 */
872 		if (nhs == 255)
873 			return 0;
874 		nhs++;
875 
876 		rtnh = rtnh_next(rtnh, &remaining);
877 	}
878 
879 	/* leftover implies invalid nexthop configuration, discard it */
880 	return remaining > 0 ? 0 : nhs;
881 }
882 
883 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
884 			       struct mpls_route *rt, u8 max_labels,
885 			       struct netlink_ext_ack *extack)
886 {
887 	struct rtnexthop *rtnh = cfg->rc_mp;
888 	struct nlattr *nla_via, *nla_newdst;
889 	int remaining = cfg->rc_mp_len;
890 	int err = 0;
891 	u8 nhs = 0;
892 
893 	change_nexthops(rt) {
894 		int attrlen;
895 
896 		nla_via = NULL;
897 		nla_newdst = NULL;
898 
899 		err = -EINVAL;
900 		if (!rtnh_ok(rtnh, remaining))
901 			goto errout;
902 
903 		/* neither weighted multipath nor any flags
904 		 * are supported
905 		 */
906 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
907 			goto errout;
908 
909 		attrlen = rtnh_attrlen(rtnh);
910 		if (attrlen > 0) {
911 			struct nlattr *attrs = rtnh_attrs(rtnh);
912 
913 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
914 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
915 		}
916 
917 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
918 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
919 				    max_labels, extack);
920 		if (err)
921 			goto errout;
922 
923 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
924 			rt->rt_nhn_alive--;
925 
926 		rtnh = rtnh_next(rtnh, &remaining);
927 		nhs++;
928 	} endfor_nexthops(rt);
929 
930 	rt->rt_nhn = nhs;
931 
932 	return 0;
933 
934 errout:
935 	return err;
936 }
937 
938 static bool mpls_label_ok(struct net *net, unsigned int *index,
939 			  struct netlink_ext_ack *extack)
940 {
941 	bool is_ok = true;
942 
943 	/* Reserved labels may not be set */
944 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
945 		NL_SET_ERR_MSG(extack,
946 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
947 		is_ok = false;
948 	}
949 
950 	/* The full 20 bit range may not be supported. */
951 	if (is_ok && *index >= net->mpls.platform_labels) {
952 		NL_SET_ERR_MSG(extack,
953 			       "Label >= configured maximum in platform_labels");
954 		is_ok = false;
955 	}
956 
957 	*index = array_index_nospec(*index, net->mpls.platform_labels);
958 	return is_ok;
959 }
960 
961 static int mpls_route_add(struct mpls_route_config *cfg,
962 			  struct netlink_ext_ack *extack)
963 {
964 	struct mpls_route __rcu **platform_label;
965 	struct net *net = cfg->rc_nlinfo.nl_net;
966 	struct mpls_route *rt, *old;
967 	int err = -EINVAL;
968 	u8 max_via_alen;
969 	unsigned index;
970 	u8 max_labels;
971 	u8 nhs;
972 
973 	index = cfg->rc_label;
974 
975 	/* If a label was not specified during insert pick one */
976 	if ((index == LABEL_NOT_SPECIFIED) &&
977 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
978 		index = find_free_label(net);
979 	}
980 
981 	if (!mpls_label_ok(net, &index, extack))
982 		goto errout;
983 
984 	/* Append makes no sense with mpls */
985 	err = -EOPNOTSUPP;
986 	if (cfg->rc_nlflags & NLM_F_APPEND) {
987 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
988 		goto errout;
989 	}
990 
991 	err = -EEXIST;
992 	platform_label = rtnl_dereference(net->mpls.platform_label);
993 	old = rtnl_dereference(platform_label[index]);
994 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
995 		goto errout;
996 
997 	err = -EEXIST;
998 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
999 		goto errout;
1000 
1001 	err = -ENOENT;
1002 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1003 		goto errout;
1004 
1005 	err = -EINVAL;
1006 	if (cfg->rc_mp) {
1007 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1008 					  cfg->rc_via_alen, &max_via_alen,
1009 					  &max_labels);
1010 	} else {
1011 		max_via_alen = cfg->rc_via_alen;
1012 		max_labels = cfg->rc_output_labels;
1013 		nhs = 1;
1014 	}
1015 
1016 	if (nhs == 0) {
1017 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1018 		goto errout;
1019 	}
1020 
1021 	err = -ENOMEM;
1022 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1023 	if (IS_ERR(rt)) {
1024 		err = PTR_ERR(rt);
1025 		goto errout;
1026 	}
1027 
1028 	rt->rt_protocol = cfg->rc_protocol;
1029 	rt->rt_payload_type = cfg->rc_payload_type;
1030 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1031 
1032 	if (cfg->rc_mp)
1033 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1034 	else
1035 		err = mpls_nh_build_from_cfg(cfg, rt);
1036 	if (err)
1037 		goto freert;
1038 
1039 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1040 
1041 	return 0;
1042 
1043 freert:
1044 	mpls_rt_free(rt);
1045 errout:
1046 	return err;
1047 }
1048 
1049 static int mpls_route_del(struct mpls_route_config *cfg,
1050 			  struct netlink_ext_ack *extack)
1051 {
1052 	struct net *net = cfg->rc_nlinfo.nl_net;
1053 	unsigned index;
1054 	int err = -EINVAL;
1055 
1056 	index = cfg->rc_label;
1057 
1058 	if (!mpls_label_ok(net, &index, extack))
1059 		goto errout;
1060 
1061 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1062 
1063 	err = 0;
1064 errout:
1065 	return err;
1066 }
1067 
1068 static void mpls_get_stats(struct mpls_dev *mdev,
1069 			   struct mpls_link_stats *stats)
1070 {
1071 	struct mpls_pcpu_stats *p;
1072 	int i;
1073 
1074 	memset(stats, 0, sizeof(*stats));
1075 
1076 	for_each_possible_cpu(i) {
1077 		struct mpls_link_stats local;
1078 		unsigned int start;
1079 
1080 		p = per_cpu_ptr(mdev->stats, i);
1081 		do {
1082 			start = u64_stats_fetch_begin(&p->syncp);
1083 			local = p->stats;
1084 		} while (u64_stats_fetch_retry(&p->syncp, start));
1085 
1086 		stats->rx_packets	+= local.rx_packets;
1087 		stats->rx_bytes		+= local.rx_bytes;
1088 		stats->tx_packets	+= local.tx_packets;
1089 		stats->tx_bytes		+= local.tx_bytes;
1090 		stats->rx_errors	+= local.rx_errors;
1091 		stats->tx_errors	+= local.tx_errors;
1092 		stats->rx_dropped	+= local.rx_dropped;
1093 		stats->tx_dropped	+= local.tx_dropped;
1094 		stats->rx_noroute	+= local.rx_noroute;
1095 	}
1096 }
1097 
1098 static int mpls_fill_stats_af(struct sk_buff *skb,
1099 			      const struct net_device *dev)
1100 {
1101 	struct mpls_link_stats *stats;
1102 	struct mpls_dev *mdev;
1103 	struct nlattr *nla;
1104 
1105 	mdev = mpls_dev_get(dev);
1106 	if (!mdev)
1107 		return -ENODATA;
1108 
1109 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1110 				sizeof(struct mpls_link_stats),
1111 				MPLS_STATS_UNSPEC);
1112 	if (!nla)
1113 		return -EMSGSIZE;
1114 
1115 	stats = nla_data(nla);
1116 	mpls_get_stats(mdev, stats);
1117 
1118 	return 0;
1119 }
1120 
1121 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1122 {
1123 	struct mpls_dev *mdev;
1124 
1125 	mdev = mpls_dev_get(dev);
1126 	if (!mdev)
1127 		return 0;
1128 
1129 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1130 }
1131 
1132 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1133 				     u32 portid, u32 seq, int event,
1134 				     unsigned int flags, int type)
1135 {
1136 	struct nlmsghdr  *nlh;
1137 	struct netconfmsg *ncm;
1138 	bool all = false;
1139 
1140 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1141 			flags);
1142 	if (!nlh)
1143 		return -EMSGSIZE;
1144 
1145 	if (type == NETCONFA_ALL)
1146 		all = true;
1147 
1148 	ncm = nlmsg_data(nlh);
1149 	ncm->ncm_family = AF_MPLS;
1150 
1151 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1152 		goto nla_put_failure;
1153 
1154 	if ((all || type == NETCONFA_INPUT) &&
1155 	    nla_put_s32(skb, NETCONFA_INPUT,
1156 			mdev->input_enabled) < 0)
1157 		goto nla_put_failure;
1158 
1159 	nlmsg_end(skb, nlh);
1160 	return 0;
1161 
1162 nla_put_failure:
1163 	nlmsg_cancel(skb, nlh);
1164 	return -EMSGSIZE;
1165 }
1166 
1167 static int mpls_netconf_msgsize_devconf(int type)
1168 {
1169 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1170 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1171 	bool all = false;
1172 
1173 	if (type == NETCONFA_ALL)
1174 		all = true;
1175 
1176 	if (all || type == NETCONFA_INPUT)
1177 		size += nla_total_size(4);
1178 
1179 	return size;
1180 }
1181 
1182 static void mpls_netconf_notify_devconf(struct net *net, int event,
1183 					int type, struct mpls_dev *mdev)
1184 {
1185 	struct sk_buff *skb;
1186 	int err = -ENOBUFS;
1187 
1188 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1189 	if (!skb)
1190 		goto errout;
1191 
1192 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1193 	if (err < 0) {
1194 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1195 		WARN_ON(err == -EMSGSIZE);
1196 		kfree_skb(skb);
1197 		goto errout;
1198 	}
1199 
1200 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1201 	return;
1202 errout:
1203 	if (err < 0)
1204 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1205 }
1206 
1207 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1208 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1209 };
1210 
1211 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1212 				      const struct nlmsghdr *nlh,
1213 				      struct nlattr **tb,
1214 				      struct netlink_ext_ack *extack)
1215 {
1216 	int i, err;
1217 
1218 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1219 		NL_SET_ERR_MSG_MOD(extack,
1220 				   "Invalid header for netconf get request");
1221 		return -EINVAL;
1222 	}
1223 
1224 	if (!netlink_strict_get_check(skb))
1225 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1226 					      tb, NETCONFA_MAX,
1227 					      devconf_mpls_policy, extack);
1228 
1229 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1230 					    tb, NETCONFA_MAX,
1231 					    devconf_mpls_policy, extack);
1232 	if (err)
1233 		return err;
1234 
1235 	for (i = 0; i <= NETCONFA_MAX; i++) {
1236 		if (!tb[i])
1237 			continue;
1238 
1239 		switch (i) {
1240 		case NETCONFA_IFINDEX:
1241 			break;
1242 		default:
1243 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1244 			return -EINVAL;
1245 		}
1246 	}
1247 
1248 	return 0;
1249 }
1250 
1251 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1252 				    struct nlmsghdr *nlh,
1253 				    struct netlink_ext_ack *extack)
1254 {
1255 	struct net *net = sock_net(in_skb->sk);
1256 	struct nlattr *tb[NETCONFA_MAX + 1];
1257 	struct net_device *dev;
1258 	struct mpls_dev *mdev;
1259 	struct sk_buff *skb;
1260 	int ifindex;
1261 	int err;
1262 
1263 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1264 	if (err < 0)
1265 		goto errout;
1266 
1267 	err = -EINVAL;
1268 	if (!tb[NETCONFA_IFINDEX])
1269 		goto errout;
1270 
1271 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1272 	dev = __dev_get_by_index(net, ifindex);
1273 	if (!dev)
1274 		goto errout;
1275 
1276 	mdev = mpls_dev_get(dev);
1277 	if (!mdev)
1278 		goto errout;
1279 
1280 	err = -ENOBUFS;
1281 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1282 	if (!skb)
1283 		goto errout;
1284 
1285 	err = mpls_netconf_fill_devconf(skb, mdev,
1286 					NETLINK_CB(in_skb).portid,
1287 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1288 					NETCONFA_ALL);
1289 	if (err < 0) {
1290 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1291 		WARN_ON(err == -EMSGSIZE);
1292 		kfree_skb(skb);
1293 		goto errout;
1294 	}
1295 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1296 errout:
1297 	return err;
1298 }
1299 
1300 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1301 				     struct netlink_callback *cb)
1302 {
1303 	const struct nlmsghdr *nlh = cb->nlh;
1304 	struct net *net = sock_net(skb->sk);
1305 	struct hlist_head *head;
1306 	struct net_device *dev;
1307 	struct mpls_dev *mdev;
1308 	int idx, s_idx;
1309 	int h, s_h;
1310 
1311 	if (cb->strict_check) {
1312 		struct netlink_ext_ack *extack = cb->extack;
1313 		struct netconfmsg *ncm;
1314 
1315 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1316 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1317 			return -EINVAL;
1318 		}
1319 
1320 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1321 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1322 			return -EINVAL;
1323 		}
1324 	}
1325 
1326 	s_h = cb->args[0];
1327 	s_idx = idx = cb->args[1];
1328 
1329 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1330 		idx = 0;
1331 		head = &net->dev_index_head[h];
1332 		rcu_read_lock();
1333 		cb->seq = net->dev_base_seq;
1334 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1335 			if (idx < s_idx)
1336 				goto cont;
1337 			mdev = mpls_dev_get(dev);
1338 			if (!mdev)
1339 				goto cont;
1340 			if (mpls_netconf_fill_devconf(skb, mdev,
1341 						      NETLINK_CB(cb->skb).portid,
1342 						      nlh->nlmsg_seq,
1343 						      RTM_NEWNETCONF,
1344 						      NLM_F_MULTI,
1345 						      NETCONFA_ALL) < 0) {
1346 				rcu_read_unlock();
1347 				goto done;
1348 			}
1349 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1350 cont:
1351 			idx++;
1352 		}
1353 		rcu_read_unlock();
1354 	}
1355 done:
1356 	cb->args[0] = h;
1357 	cb->args[1] = idx;
1358 
1359 	return skb->len;
1360 }
1361 
1362 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1363 	(&((struct mpls_dev *)0)->field)
1364 
1365 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1366 			  void __user *buffer,
1367 			  size_t *lenp, loff_t *ppos)
1368 {
1369 	int oval = *(int *)ctl->data;
1370 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1371 
1372 	if (write) {
1373 		struct mpls_dev *mdev = ctl->extra1;
1374 		int i = (int *)ctl->data - (int *)mdev;
1375 		struct net *net = ctl->extra2;
1376 		int val = *(int *)ctl->data;
1377 
1378 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1379 		    val != oval) {
1380 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1381 						    NETCONFA_INPUT, mdev);
1382 		}
1383 	}
1384 
1385 	return ret;
1386 }
1387 
1388 static const struct ctl_table mpls_dev_table[] = {
1389 	{
1390 		.procname	= "input",
1391 		.maxlen		= sizeof(int),
1392 		.mode		= 0644,
1393 		.proc_handler	= mpls_conf_proc,
1394 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1395 	},
1396 	{ }
1397 };
1398 
1399 static int mpls_dev_sysctl_register(struct net_device *dev,
1400 				    struct mpls_dev *mdev)
1401 {
1402 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1403 	struct net *net = dev_net(dev);
1404 	struct ctl_table *table;
1405 	int i;
1406 
1407 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1408 	if (!table)
1409 		goto out;
1410 
1411 	/* Table data contains only offsets relative to the base of
1412 	 * the mdev at this point, so make them absolute.
1413 	 */
1414 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1415 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1416 		table[i].extra1 = mdev;
1417 		table[i].extra2 = net;
1418 	}
1419 
1420 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1421 
1422 	mdev->sysctl = register_net_sysctl(net, path, table);
1423 	if (!mdev->sysctl)
1424 		goto free;
1425 
1426 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1427 	return 0;
1428 
1429 free:
1430 	kfree(table);
1431 out:
1432 	return -ENOBUFS;
1433 }
1434 
1435 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1436 				       struct mpls_dev *mdev)
1437 {
1438 	struct net *net = dev_net(dev);
1439 	struct ctl_table *table;
1440 
1441 	table = mdev->sysctl->ctl_table_arg;
1442 	unregister_net_sysctl_table(mdev->sysctl);
1443 	kfree(table);
1444 
1445 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1446 }
1447 
1448 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1449 {
1450 	struct mpls_dev *mdev;
1451 	int err = -ENOMEM;
1452 	int i;
1453 
1454 	ASSERT_RTNL();
1455 
1456 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1457 	if (!mdev)
1458 		return ERR_PTR(err);
1459 
1460 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1461 	if (!mdev->stats)
1462 		goto free;
1463 
1464 	for_each_possible_cpu(i) {
1465 		struct mpls_pcpu_stats *mpls_stats;
1466 
1467 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1468 		u64_stats_init(&mpls_stats->syncp);
1469 	}
1470 
1471 	mdev->dev = dev;
1472 
1473 	err = mpls_dev_sysctl_register(dev, mdev);
1474 	if (err)
1475 		goto free;
1476 
1477 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1478 
1479 	return mdev;
1480 
1481 free:
1482 	free_percpu(mdev->stats);
1483 	kfree(mdev);
1484 	return ERR_PTR(err);
1485 }
1486 
1487 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1488 {
1489 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1490 
1491 	free_percpu(mdev->stats);
1492 	kfree(mdev);
1493 }
1494 
1495 static void mpls_ifdown(struct net_device *dev, int event)
1496 {
1497 	struct mpls_route __rcu **platform_label;
1498 	struct net *net = dev_net(dev);
1499 	u8 alive, deleted;
1500 	unsigned index;
1501 
1502 	platform_label = rtnl_dereference(net->mpls.platform_label);
1503 	for (index = 0; index < net->mpls.platform_labels; index++) {
1504 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1505 
1506 		if (!rt)
1507 			continue;
1508 
1509 		alive = 0;
1510 		deleted = 0;
1511 		change_nexthops(rt) {
1512 			unsigned int nh_flags = nh->nh_flags;
1513 
1514 			if (rtnl_dereference(nh->nh_dev) != dev)
1515 				goto next;
1516 
1517 			switch (event) {
1518 			case NETDEV_DOWN:
1519 			case NETDEV_UNREGISTER:
1520 				nh_flags |= RTNH_F_DEAD;
1521 				/* fall through */
1522 			case NETDEV_CHANGE:
1523 				nh_flags |= RTNH_F_LINKDOWN;
1524 				break;
1525 			}
1526 			if (event == NETDEV_UNREGISTER)
1527 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1528 
1529 			if (nh->nh_flags != nh_flags)
1530 				WRITE_ONCE(nh->nh_flags, nh_flags);
1531 next:
1532 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1533 				alive++;
1534 			if (!rtnl_dereference(nh->nh_dev))
1535 				deleted++;
1536 		} endfor_nexthops(rt);
1537 
1538 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1539 
1540 		/* if there are no more nexthops, delete the route */
1541 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1542 			mpls_route_update(net, index, NULL, NULL);
1543 	}
1544 }
1545 
1546 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1547 {
1548 	struct mpls_route __rcu **platform_label;
1549 	struct net *net = dev_net(dev);
1550 	unsigned index;
1551 	u8 alive;
1552 
1553 	platform_label = rtnl_dereference(net->mpls.platform_label);
1554 	for (index = 0; index < net->mpls.platform_labels; index++) {
1555 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1556 
1557 		if (!rt)
1558 			continue;
1559 
1560 		alive = 0;
1561 		change_nexthops(rt) {
1562 			unsigned int nh_flags = nh->nh_flags;
1563 			struct net_device *nh_dev =
1564 				rtnl_dereference(nh->nh_dev);
1565 
1566 			if (!(nh_flags & flags)) {
1567 				alive++;
1568 				continue;
1569 			}
1570 			if (nh_dev != dev)
1571 				continue;
1572 			alive++;
1573 			nh_flags &= ~flags;
1574 			WRITE_ONCE(nh->nh_flags, nh_flags);
1575 		} endfor_nexthops(rt);
1576 
1577 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1578 	}
1579 }
1580 
1581 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1582 			   void *ptr)
1583 {
1584 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1585 	struct mpls_dev *mdev;
1586 	unsigned int flags;
1587 
1588 	if (event == NETDEV_REGISTER) {
1589 
1590 		/* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1591 		 * IPIP devices
1592 		 */
1593 		if (dev->type == ARPHRD_ETHER ||
1594 		    dev->type == ARPHRD_LOOPBACK ||
1595 		    dev->type == ARPHRD_IPGRE ||
1596 		    dev->type == ARPHRD_IP6GRE ||
1597 		    dev->type == ARPHRD_SIT ||
1598 		    dev->type == ARPHRD_TUNNEL) {
1599 			mdev = mpls_add_dev(dev);
1600 			if (IS_ERR(mdev))
1601 				return notifier_from_errno(PTR_ERR(mdev));
1602 		}
1603 		return NOTIFY_OK;
1604 	}
1605 
1606 	mdev = mpls_dev_get(dev);
1607 	if (!mdev)
1608 		return NOTIFY_OK;
1609 
1610 	switch (event) {
1611 	case NETDEV_DOWN:
1612 		mpls_ifdown(dev, event);
1613 		break;
1614 	case NETDEV_UP:
1615 		flags = dev_get_flags(dev);
1616 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1617 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1618 		else
1619 			mpls_ifup(dev, RTNH_F_DEAD);
1620 		break;
1621 	case NETDEV_CHANGE:
1622 		flags = dev_get_flags(dev);
1623 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1624 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1625 		else
1626 			mpls_ifdown(dev, event);
1627 		break;
1628 	case NETDEV_UNREGISTER:
1629 		mpls_ifdown(dev, event);
1630 		mdev = mpls_dev_get(dev);
1631 		if (mdev) {
1632 			mpls_dev_sysctl_unregister(dev, mdev);
1633 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1634 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1635 		}
1636 		break;
1637 	case NETDEV_CHANGENAME:
1638 		mdev = mpls_dev_get(dev);
1639 		if (mdev) {
1640 			int err;
1641 
1642 			mpls_dev_sysctl_unregister(dev, mdev);
1643 			err = mpls_dev_sysctl_register(dev, mdev);
1644 			if (err)
1645 				return notifier_from_errno(err);
1646 		}
1647 		break;
1648 	}
1649 	return NOTIFY_OK;
1650 }
1651 
1652 static struct notifier_block mpls_dev_notifier = {
1653 	.notifier_call = mpls_dev_notify,
1654 };
1655 
1656 static int nla_put_via(struct sk_buff *skb,
1657 		       u8 table, const void *addr, int alen)
1658 {
1659 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1660 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1661 	};
1662 	struct nlattr *nla;
1663 	struct rtvia *via;
1664 	int family = AF_UNSPEC;
1665 
1666 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1667 	if (!nla)
1668 		return -EMSGSIZE;
1669 
1670 	if (table <= NEIGH_NR_TABLES)
1671 		family = table_to_family[table];
1672 
1673 	via = nla_data(nla);
1674 	via->rtvia_family = family;
1675 	memcpy(via->rtvia_addr, addr, alen);
1676 	return 0;
1677 }
1678 
1679 int nla_put_labels(struct sk_buff *skb, int attrtype,
1680 		   u8 labels, const u32 label[])
1681 {
1682 	struct nlattr *nla;
1683 	struct mpls_shim_hdr *nla_label;
1684 	bool bos;
1685 	int i;
1686 	nla = nla_reserve(skb, attrtype, labels*4);
1687 	if (!nla)
1688 		return -EMSGSIZE;
1689 
1690 	nla_label = nla_data(nla);
1691 	bos = true;
1692 	for (i = labels - 1; i >= 0; i--) {
1693 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1694 		bos = false;
1695 	}
1696 
1697 	return 0;
1698 }
1699 EXPORT_SYMBOL_GPL(nla_put_labels);
1700 
1701 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1702 		   u32 label[], struct netlink_ext_ack *extack)
1703 {
1704 	unsigned len = nla_len(nla);
1705 	struct mpls_shim_hdr *nla_label;
1706 	u8 nla_labels;
1707 	bool bos;
1708 	int i;
1709 
1710 	/* len needs to be an even multiple of 4 (the label size). Number
1711 	 * of labels is a u8 so check for overflow.
1712 	 */
1713 	if (len & 3 || len / 4 > 255) {
1714 		NL_SET_ERR_MSG_ATTR(extack, nla,
1715 				    "Invalid length for labels attribute");
1716 		return -EINVAL;
1717 	}
1718 
1719 	/* Limit the number of new labels allowed */
1720 	nla_labels = len/4;
1721 	if (nla_labels > max_labels) {
1722 		NL_SET_ERR_MSG(extack, "Too many labels");
1723 		return -EINVAL;
1724 	}
1725 
1726 	/* when label == NULL, caller wants number of labels */
1727 	if (!label)
1728 		goto out;
1729 
1730 	nla_label = nla_data(nla);
1731 	bos = true;
1732 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1733 		struct mpls_entry_decoded dec;
1734 		dec = mpls_entry_decode(nla_label + i);
1735 
1736 		/* Ensure the bottom of stack flag is properly set
1737 		 * and ttl and tc are both clear.
1738 		 */
1739 		if (dec.ttl) {
1740 			NL_SET_ERR_MSG_ATTR(extack, nla,
1741 					    "TTL in label must be 0");
1742 			return -EINVAL;
1743 		}
1744 
1745 		if (dec.tc) {
1746 			NL_SET_ERR_MSG_ATTR(extack, nla,
1747 					    "Traffic class in label must be 0");
1748 			return -EINVAL;
1749 		}
1750 
1751 		if (dec.bos != bos) {
1752 			NL_SET_BAD_ATTR(extack, nla);
1753 			if (bos) {
1754 				NL_SET_ERR_MSG(extack,
1755 					       "BOS bit must be set in first label");
1756 			} else {
1757 				NL_SET_ERR_MSG(extack,
1758 					       "BOS bit can only be set in first label");
1759 			}
1760 			return -EINVAL;
1761 		}
1762 
1763 		switch (dec.label) {
1764 		case MPLS_LABEL_IMPLNULL:
1765 			/* RFC3032: This is a label that an LSR may
1766 			 * assign and distribute, but which never
1767 			 * actually appears in the encapsulation.
1768 			 */
1769 			NL_SET_ERR_MSG_ATTR(extack, nla,
1770 					    "Implicit NULL Label (3) can not be used in encapsulation");
1771 			return -EINVAL;
1772 		}
1773 
1774 		label[i] = dec.label;
1775 	}
1776 out:
1777 	*labels = nla_labels;
1778 	return 0;
1779 }
1780 EXPORT_SYMBOL_GPL(nla_get_labels);
1781 
1782 static int rtm_to_route_config(struct sk_buff *skb,
1783 			       struct nlmsghdr *nlh,
1784 			       struct mpls_route_config *cfg,
1785 			       struct netlink_ext_ack *extack)
1786 {
1787 	struct rtmsg *rtm;
1788 	struct nlattr *tb[RTA_MAX+1];
1789 	int index;
1790 	int err;
1791 
1792 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1793 				     rtm_mpls_policy, extack);
1794 	if (err < 0)
1795 		goto errout;
1796 
1797 	err = -EINVAL;
1798 	rtm = nlmsg_data(nlh);
1799 
1800 	if (rtm->rtm_family != AF_MPLS) {
1801 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1802 		goto errout;
1803 	}
1804 	if (rtm->rtm_dst_len != 20) {
1805 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1806 		goto errout;
1807 	}
1808 	if (rtm->rtm_src_len != 0) {
1809 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1810 		goto errout;
1811 	}
1812 	if (rtm->rtm_tos != 0) {
1813 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1814 		goto errout;
1815 	}
1816 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1817 		NL_SET_ERR_MSG(extack,
1818 			       "MPLS only supports the main route table");
1819 		goto errout;
1820 	}
1821 	/* Any value is acceptable for rtm_protocol */
1822 
1823 	/* As mpls uses destination specific addresses
1824 	 * (or source specific address in the case of multicast)
1825 	 * all addresses have universal scope.
1826 	 */
1827 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1828 		NL_SET_ERR_MSG(extack,
1829 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1830 		goto errout;
1831 	}
1832 	if (rtm->rtm_type != RTN_UNICAST) {
1833 		NL_SET_ERR_MSG(extack,
1834 			       "Invalid route type - MPLS only supports UNICAST");
1835 		goto errout;
1836 	}
1837 	if (rtm->rtm_flags != 0) {
1838 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1839 		goto errout;
1840 	}
1841 
1842 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1843 	cfg->rc_protocol	= rtm->rtm_protocol;
1844 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1845 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1846 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1847 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1848 	cfg->rc_nlinfo.nlh	= nlh;
1849 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1850 
1851 	for (index = 0; index <= RTA_MAX; index++) {
1852 		struct nlattr *nla = tb[index];
1853 		if (!nla)
1854 			continue;
1855 
1856 		switch (index) {
1857 		case RTA_OIF:
1858 			cfg->rc_ifindex = nla_get_u32(nla);
1859 			break;
1860 		case RTA_NEWDST:
1861 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1862 					   &cfg->rc_output_labels,
1863 					   cfg->rc_output_label, extack))
1864 				goto errout;
1865 			break;
1866 		case RTA_DST:
1867 		{
1868 			u8 label_count;
1869 			if (nla_get_labels(nla, 1, &label_count,
1870 					   &cfg->rc_label, extack))
1871 				goto errout;
1872 
1873 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1874 					   &cfg->rc_label, extack))
1875 				goto errout;
1876 			break;
1877 		}
1878 		case RTA_GATEWAY:
1879 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1880 			goto errout;
1881 		case RTA_VIA:
1882 		{
1883 			if (nla_get_via(nla, &cfg->rc_via_alen,
1884 					&cfg->rc_via_table, cfg->rc_via,
1885 					extack))
1886 				goto errout;
1887 			break;
1888 		}
1889 		case RTA_MULTIPATH:
1890 		{
1891 			cfg->rc_mp = nla_data(nla);
1892 			cfg->rc_mp_len = nla_len(nla);
1893 			break;
1894 		}
1895 		case RTA_TTL_PROPAGATE:
1896 		{
1897 			u8 ttl_propagate = nla_get_u8(nla);
1898 
1899 			if (ttl_propagate > 1) {
1900 				NL_SET_ERR_MSG_ATTR(extack, nla,
1901 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1902 				goto errout;
1903 			}
1904 			cfg->rc_ttl_propagate = ttl_propagate ?
1905 				MPLS_TTL_PROP_ENABLED :
1906 				MPLS_TTL_PROP_DISABLED;
1907 			break;
1908 		}
1909 		default:
1910 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1911 			/* Unsupported attribute */
1912 			goto errout;
1913 		}
1914 	}
1915 
1916 	err = 0;
1917 errout:
1918 	return err;
1919 }
1920 
1921 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1922 			     struct netlink_ext_ack *extack)
1923 {
1924 	struct mpls_route_config *cfg;
1925 	int err;
1926 
1927 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1928 	if (!cfg)
1929 		return -ENOMEM;
1930 
1931 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1932 	if (err < 0)
1933 		goto out;
1934 
1935 	err = mpls_route_del(cfg, extack);
1936 out:
1937 	kfree(cfg);
1938 
1939 	return err;
1940 }
1941 
1942 
1943 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1944 			     struct netlink_ext_ack *extack)
1945 {
1946 	struct mpls_route_config *cfg;
1947 	int err;
1948 
1949 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1950 	if (!cfg)
1951 		return -ENOMEM;
1952 
1953 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1954 	if (err < 0)
1955 		goto out;
1956 
1957 	err = mpls_route_add(cfg, extack);
1958 out:
1959 	kfree(cfg);
1960 
1961 	return err;
1962 }
1963 
1964 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1965 			   u32 label, struct mpls_route *rt, int flags)
1966 {
1967 	struct net_device *dev;
1968 	struct nlmsghdr *nlh;
1969 	struct rtmsg *rtm;
1970 
1971 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1972 	if (nlh == NULL)
1973 		return -EMSGSIZE;
1974 
1975 	rtm = nlmsg_data(nlh);
1976 	rtm->rtm_family = AF_MPLS;
1977 	rtm->rtm_dst_len = 20;
1978 	rtm->rtm_src_len = 0;
1979 	rtm->rtm_tos = 0;
1980 	rtm->rtm_table = RT_TABLE_MAIN;
1981 	rtm->rtm_protocol = rt->rt_protocol;
1982 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1983 	rtm->rtm_type = RTN_UNICAST;
1984 	rtm->rtm_flags = 0;
1985 
1986 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1987 		goto nla_put_failure;
1988 
1989 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1990 		bool ttl_propagate =
1991 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1992 
1993 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1994 			       ttl_propagate))
1995 			goto nla_put_failure;
1996 	}
1997 	if (rt->rt_nhn == 1) {
1998 		const struct mpls_nh *nh = rt->rt_nh;
1999 
2000 		if (nh->nh_labels &&
2001 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2002 				   nh->nh_label))
2003 			goto nla_put_failure;
2004 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2005 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2006 				nh->nh_via_alen))
2007 			goto nla_put_failure;
2008 		dev = rtnl_dereference(nh->nh_dev);
2009 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2010 			goto nla_put_failure;
2011 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2012 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2013 		if (nh->nh_flags & RTNH_F_DEAD)
2014 			rtm->rtm_flags |= RTNH_F_DEAD;
2015 	} else {
2016 		struct rtnexthop *rtnh;
2017 		struct nlattr *mp;
2018 		u8 linkdown = 0;
2019 		u8 dead = 0;
2020 
2021 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2022 		if (!mp)
2023 			goto nla_put_failure;
2024 
2025 		for_nexthops(rt) {
2026 			dev = rtnl_dereference(nh->nh_dev);
2027 			if (!dev)
2028 				continue;
2029 
2030 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2031 			if (!rtnh)
2032 				goto nla_put_failure;
2033 
2034 			rtnh->rtnh_ifindex = dev->ifindex;
2035 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2036 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2037 				linkdown++;
2038 			}
2039 			if (nh->nh_flags & RTNH_F_DEAD) {
2040 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2041 				dead++;
2042 			}
2043 
2044 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2045 							    nh->nh_labels,
2046 							    nh->nh_label))
2047 				goto nla_put_failure;
2048 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2049 			    nla_put_via(skb, nh->nh_via_table,
2050 					mpls_nh_via(rt, nh),
2051 					nh->nh_via_alen))
2052 				goto nla_put_failure;
2053 
2054 			/* length of rtnetlink header + attributes */
2055 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2056 		} endfor_nexthops(rt);
2057 
2058 		if (linkdown == rt->rt_nhn)
2059 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2060 		if (dead == rt->rt_nhn)
2061 			rtm->rtm_flags |= RTNH_F_DEAD;
2062 
2063 		nla_nest_end(skb, mp);
2064 	}
2065 
2066 	nlmsg_end(skb, nlh);
2067 	return 0;
2068 
2069 nla_put_failure:
2070 	nlmsg_cancel(skb, nlh);
2071 	return -EMSGSIZE;
2072 }
2073 
2074 #if IS_ENABLED(CONFIG_INET)
2075 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2076 				   struct fib_dump_filter *filter,
2077 				   struct netlink_callback *cb)
2078 {
2079 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2080 }
2081 #else
2082 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2083 				   struct fib_dump_filter *filter,
2084 				   struct netlink_callback *cb)
2085 {
2086 	struct netlink_ext_ack *extack = cb->extack;
2087 	struct nlattr *tb[RTA_MAX + 1];
2088 	struct rtmsg *rtm;
2089 	int err, i;
2090 
2091 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2092 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2093 		return -EINVAL;
2094 	}
2095 
2096 	rtm = nlmsg_data(nlh);
2097 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2098 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2099 	    rtm->rtm_flags) {
2100 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2101 		return -EINVAL;
2102 	}
2103 
2104 	if (rtm->rtm_protocol) {
2105 		filter->protocol = rtm->rtm_protocol;
2106 		filter->filter_set = 1;
2107 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2108 	}
2109 
2110 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2111 					    rtm_mpls_policy, extack);
2112 	if (err < 0)
2113 		return err;
2114 
2115 	for (i = 0; i <= RTA_MAX; ++i) {
2116 		int ifindex;
2117 
2118 		if (i == RTA_OIF) {
2119 			ifindex = nla_get_u32(tb[i]);
2120 			filter->dev = __dev_get_by_index(net, ifindex);
2121 			if (!filter->dev)
2122 				return -ENODEV;
2123 			filter->filter_set = 1;
2124 		} else if (tb[i]) {
2125 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2126 			return -EINVAL;
2127 		}
2128 	}
2129 
2130 	return 0;
2131 }
2132 #endif
2133 
2134 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2135 			     const struct net_device *dev)
2136 {
2137 	struct net_device *nh_dev;
2138 
2139 	if (rt->rt_nhn == 1) {
2140 		struct mpls_nh *nh = rt->rt_nh;
2141 
2142 		nh_dev = rtnl_dereference(nh->nh_dev);
2143 		if (dev == nh_dev)
2144 			return true;
2145 	} else {
2146 		for_nexthops(rt) {
2147 			nh_dev = rtnl_dereference(nh->nh_dev);
2148 			if (nh_dev == dev)
2149 				return true;
2150 		} endfor_nexthops(rt);
2151 	}
2152 
2153 	return false;
2154 }
2155 
2156 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2157 {
2158 	const struct nlmsghdr *nlh = cb->nlh;
2159 	struct net *net = sock_net(skb->sk);
2160 	struct mpls_route __rcu **platform_label;
2161 	struct fib_dump_filter filter = {};
2162 	unsigned int flags = NLM_F_MULTI;
2163 	size_t platform_labels;
2164 	unsigned int index;
2165 
2166 	ASSERT_RTNL();
2167 
2168 	if (cb->strict_check) {
2169 		int err;
2170 
2171 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2172 		if (err < 0)
2173 			return err;
2174 
2175 		/* for MPLS, there is only 1 table with fixed type and flags.
2176 		 * If either are set in the filter then return nothing.
2177 		 */
2178 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2179 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2180 		     filter.flags)
2181 			return skb->len;
2182 	}
2183 
2184 	index = cb->args[0];
2185 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2186 		index = MPLS_LABEL_FIRST_UNRESERVED;
2187 
2188 	platform_label = rtnl_dereference(net->mpls.platform_label);
2189 	platform_labels = net->mpls.platform_labels;
2190 
2191 	if (filter.filter_set)
2192 		flags |= NLM_F_DUMP_FILTERED;
2193 
2194 	for (; index < platform_labels; index++) {
2195 		struct mpls_route *rt;
2196 
2197 		rt = rtnl_dereference(platform_label[index]);
2198 		if (!rt)
2199 			continue;
2200 
2201 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2202 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2203 			continue;
2204 
2205 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2206 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2207 				    index, rt, flags) < 0)
2208 			break;
2209 	}
2210 	cb->args[0] = index;
2211 
2212 	return skb->len;
2213 }
2214 
2215 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2216 {
2217 	size_t payload =
2218 		NLMSG_ALIGN(sizeof(struct rtmsg))
2219 		+ nla_total_size(4)			/* RTA_DST */
2220 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2221 
2222 	if (rt->rt_nhn == 1) {
2223 		struct mpls_nh *nh = rt->rt_nh;
2224 
2225 		if (nh->nh_dev)
2226 			payload += nla_total_size(4); /* RTA_OIF */
2227 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2228 			payload += nla_total_size(2 + nh->nh_via_alen);
2229 		if (nh->nh_labels) /* RTA_NEWDST */
2230 			payload += nla_total_size(nh->nh_labels * 4);
2231 	} else {
2232 		/* each nexthop is packed in an attribute */
2233 		size_t nhsize = 0;
2234 
2235 		for_nexthops(rt) {
2236 			if (!rtnl_dereference(nh->nh_dev))
2237 				continue;
2238 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2239 			/* RTA_VIA */
2240 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2241 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2242 			if (nh->nh_labels)
2243 				nhsize += nla_total_size(nh->nh_labels * 4);
2244 		} endfor_nexthops(rt);
2245 		/* nested attribute */
2246 		payload += nla_total_size(nhsize);
2247 	}
2248 
2249 	return payload;
2250 }
2251 
2252 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2253 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2254 		       unsigned int nlm_flags)
2255 {
2256 	struct sk_buff *skb;
2257 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2258 	int err = -ENOBUFS;
2259 
2260 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2261 	if (skb == NULL)
2262 		goto errout;
2263 
2264 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2265 	if (err < 0) {
2266 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2267 		WARN_ON(err == -EMSGSIZE);
2268 		kfree_skb(skb);
2269 		goto errout;
2270 	}
2271 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2272 
2273 	return;
2274 errout:
2275 	if (err < 0)
2276 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2277 }
2278 
2279 static int mpls_valid_getroute_req(struct sk_buff *skb,
2280 				   const struct nlmsghdr *nlh,
2281 				   struct nlattr **tb,
2282 				   struct netlink_ext_ack *extack)
2283 {
2284 	struct rtmsg *rtm;
2285 	int i, err;
2286 
2287 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2288 		NL_SET_ERR_MSG_MOD(extack,
2289 				   "Invalid header for get route request");
2290 		return -EINVAL;
2291 	}
2292 
2293 	if (!netlink_strict_get_check(skb))
2294 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2295 					      rtm_mpls_policy, extack);
2296 
2297 	rtm = nlmsg_data(nlh);
2298 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2299 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2300 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2301 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2302 		return -EINVAL;
2303 	}
2304 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2305 		NL_SET_ERR_MSG_MOD(extack,
2306 				   "Invalid flags for get route request");
2307 		return -EINVAL;
2308 	}
2309 
2310 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2311 					    rtm_mpls_policy, extack);
2312 	if (err)
2313 		return err;
2314 
2315 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2316 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2317 		return -EINVAL;
2318 	}
2319 
2320 	for (i = 0; i <= RTA_MAX; i++) {
2321 		if (!tb[i])
2322 			continue;
2323 
2324 		switch (i) {
2325 		case RTA_DST:
2326 		case RTA_NEWDST:
2327 			break;
2328 		default:
2329 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2330 			return -EINVAL;
2331 		}
2332 	}
2333 
2334 	return 0;
2335 }
2336 
2337 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2338 			 struct netlink_ext_ack *extack)
2339 {
2340 	struct net *net = sock_net(in_skb->sk);
2341 	u32 portid = NETLINK_CB(in_skb).portid;
2342 	u32 in_label = LABEL_NOT_SPECIFIED;
2343 	struct nlattr *tb[RTA_MAX + 1];
2344 	u32 labels[MAX_NEW_LABELS];
2345 	struct mpls_shim_hdr *hdr;
2346 	unsigned int hdr_size = 0;
2347 	struct net_device *dev;
2348 	struct mpls_route *rt;
2349 	struct rtmsg *rtm, *r;
2350 	struct nlmsghdr *nlh;
2351 	struct sk_buff *skb;
2352 	struct mpls_nh *nh;
2353 	u8 n_labels;
2354 	int err;
2355 
2356 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2357 	if (err < 0)
2358 		goto errout;
2359 
2360 	rtm = nlmsg_data(in_nlh);
2361 
2362 	if (tb[RTA_DST]) {
2363 		u8 label_count;
2364 
2365 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2366 				   &in_label, extack)) {
2367 			err = -EINVAL;
2368 			goto errout;
2369 		}
2370 
2371 		if (!mpls_label_ok(net, &in_label, extack)) {
2372 			err = -EINVAL;
2373 			goto errout;
2374 		}
2375 	}
2376 
2377 	rt = mpls_route_input_rcu(net, in_label);
2378 	if (!rt) {
2379 		err = -ENETUNREACH;
2380 		goto errout;
2381 	}
2382 
2383 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2384 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2385 		if (!skb) {
2386 			err = -ENOBUFS;
2387 			goto errout;
2388 		}
2389 
2390 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2391 				      RTM_NEWROUTE, in_label, rt, 0);
2392 		if (err < 0) {
2393 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2394 			WARN_ON(err == -EMSGSIZE);
2395 			goto errout_free;
2396 		}
2397 
2398 		return rtnl_unicast(skb, net, portid);
2399 	}
2400 
2401 	if (tb[RTA_NEWDST]) {
2402 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2403 				   labels, extack) != 0) {
2404 			err = -EINVAL;
2405 			goto errout;
2406 		}
2407 
2408 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2409 	}
2410 
2411 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2412 	if (!skb) {
2413 		err = -ENOBUFS;
2414 		goto errout;
2415 	}
2416 
2417 	skb->protocol = htons(ETH_P_MPLS_UC);
2418 
2419 	if (hdr_size) {
2420 		bool bos;
2421 		int i;
2422 
2423 		if (skb_cow(skb, hdr_size)) {
2424 			err = -ENOBUFS;
2425 			goto errout_free;
2426 		}
2427 
2428 		skb_reserve(skb, hdr_size);
2429 		skb_push(skb, hdr_size);
2430 		skb_reset_network_header(skb);
2431 
2432 		/* Push new labels */
2433 		hdr = mpls_hdr(skb);
2434 		bos = true;
2435 		for (i = n_labels - 1; i >= 0; i--) {
2436 			hdr[i] = mpls_entry_encode(labels[i],
2437 						   1, 0, bos);
2438 			bos = false;
2439 		}
2440 	}
2441 
2442 	nh = mpls_select_multipath(rt, skb);
2443 	if (!nh) {
2444 		err = -ENETUNREACH;
2445 		goto errout_free;
2446 	}
2447 
2448 	if (hdr_size) {
2449 		skb_pull(skb, hdr_size);
2450 		skb_reset_network_header(skb);
2451 	}
2452 
2453 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2454 			RTM_NEWROUTE, sizeof(*r), 0);
2455 	if (!nlh) {
2456 		err = -EMSGSIZE;
2457 		goto errout_free;
2458 	}
2459 
2460 	r = nlmsg_data(nlh);
2461 	r->rtm_family	 = AF_MPLS;
2462 	r->rtm_dst_len	= 20;
2463 	r->rtm_src_len	= 0;
2464 	r->rtm_table	= RT_TABLE_MAIN;
2465 	r->rtm_type	= RTN_UNICAST;
2466 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2467 	r->rtm_protocol = rt->rt_protocol;
2468 	r->rtm_flags	= 0;
2469 
2470 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2471 		goto nla_put_failure;
2472 
2473 	if (nh->nh_labels &&
2474 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2475 			   nh->nh_label))
2476 		goto nla_put_failure;
2477 
2478 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2479 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2480 			nh->nh_via_alen))
2481 		goto nla_put_failure;
2482 	dev = rtnl_dereference(nh->nh_dev);
2483 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2484 		goto nla_put_failure;
2485 
2486 	nlmsg_end(skb, nlh);
2487 
2488 	err = rtnl_unicast(skb, net, portid);
2489 errout:
2490 	return err;
2491 
2492 nla_put_failure:
2493 	nlmsg_cancel(skb, nlh);
2494 	err = -EMSGSIZE;
2495 errout_free:
2496 	kfree_skb(skb);
2497 	return err;
2498 }
2499 
2500 static int resize_platform_label_table(struct net *net, size_t limit)
2501 {
2502 	size_t size = sizeof(struct mpls_route *) * limit;
2503 	size_t old_limit;
2504 	size_t cp_size;
2505 	struct mpls_route __rcu **labels = NULL, **old;
2506 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2507 	unsigned index;
2508 
2509 	if (size) {
2510 		labels = kvzalloc(size, GFP_KERNEL);
2511 		if (!labels)
2512 			goto nolabels;
2513 	}
2514 
2515 	/* In case the predefined labels need to be populated */
2516 	if (limit > MPLS_LABEL_IPV4NULL) {
2517 		struct net_device *lo = net->loopback_dev;
2518 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2519 		if (IS_ERR(rt0))
2520 			goto nort0;
2521 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2522 		rt0->rt_protocol = RTPROT_KERNEL;
2523 		rt0->rt_payload_type = MPT_IPV4;
2524 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2525 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2526 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2527 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2528 		       lo->addr_len);
2529 	}
2530 	if (limit > MPLS_LABEL_IPV6NULL) {
2531 		struct net_device *lo = net->loopback_dev;
2532 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2533 		if (IS_ERR(rt2))
2534 			goto nort2;
2535 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2536 		rt2->rt_protocol = RTPROT_KERNEL;
2537 		rt2->rt_payload_type = MPT_IPV6;
2538 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2539 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2540 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2541 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2542 		       lo->addr_len);
2543 	}
2544 
2545 	rtnl_lock();
2546 	/* Remember the original table */
2547 	old = rtnl_dereference(net->mpls.platform_label);
2548 	old_limit = net->mpls.platform_labels;
2549 
2550 	/* Free any labels beyond the new table */
2551 	for (index = limit; index < old_limit; index++)
2552 		mpls_route_update(net, index, NULL, NULL);
2553 
2554 	/* Copy over the old labels */
2555 	cp_size = size;
2556 	if (old_limit < limit)
2557 		cp_size = old_limit * sizeof(struct mpls_route *);
2558 
2559 	memcpy(labels, old, cp_size);
2560 
2561 	/* If needed set the predefined labels */
2562 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2563 	    (limit > MPLS_LABEL_IPV6NULL)) {
2564 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2565 		rt2 = NULL;
2566 	}
2567 
2568 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2569 	    (limit > MPLS_LABEL_IPV4NULL)) {
2570 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2571 		rt0 = NULL;
2572 	}
2573 
2574 	/* Update the global pointers */
2575 	net->mpls.platform_labels = limit;
2576 	rcu_assign_pointer(net->mpls.platform_label, labels);
2577 
2578 	rtnl_unlock();
2579 
2580 	mpls_rt_free(rt2);
2581 	mpls_rt_free(rt0);
2582 
2583 	if (old) {
2584 		synchronize_rcu();
2585 		kvfree(old);
2586 	}
2587 	return 0;
2588 
2589 nort2:
2590 	mpls_rt_free(rt0);
2591 nort0:
2592 	kvfree(labels);
2593 nolabels:
2594 	return -ENOMEM;
2595 }
2596 
2597 static int mpls_platform_labels(struct ctl_table *table, int write,
2598 				void __user *buffer, size_t *lenp, loff_t *ppos)
2599 {
2600 	struct net *net = table->data;
2601 	int platform_labels = net->mpls.platform_labels;
2602 	int ret;
2603 	struct ctl_table tmp = {
2604 		.procname	= table->procname,
2605 		.data		= &platform_labels,
2606 		.maxlen		= sizeof(int),
2607 		.mode		= table->mode,
2608 		.extra1		= SYSCTL_ZERO,
2609 		.extra2		= &label_limit,
2610 	};
2611 
2612 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2613 
2614 	if (write && ret == 0)
2615 		ret = resize_platform_label_table(net, platform_labels);
2616 
2617 	return ret;
2618 }
2619 
2620 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2621 	(&((struct net *)0)->field)
2622 
2623 static const struct ctl_table mpls_table[] = {
2624 	{
2625 		.procname	= "platform_labels",
2626 		.data		= NULL,
2627 		.maxlen		= sizeof(int),
2628 		.mode		= 0644,
2629 		.proc_handler	= mpls_platform_labels,
2630 	},
2631 	{
2632 		.procname	= "ip_ttl_propagate",
2633 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2634 		.maxlen		= sizeof(int),
2635 		.mode		= 0644,
2636 		.proc_handler	= proc_dointvec_minmax,
2637 		.extra1		= SYSCTL_ZERO,
2638 		.extra2		= SYSCTL_ONE,
2639 	},
2640 	{
2641 		.procname	= "default_ttl",
2642 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2643 		.maxlen		= sizeof(int),
2644 		.mode		= 0644,
2645 		.proc_handler	= proc_dointvec_minmax,
2646 		.extra1		= SYSCTL_ONE,
2647 		.extra2		= &ttl_max,
2648 	},
2649 	{ }
2650 };
2651 
2652 static int mpls_net_init(struct net *net)
2653 {
2654 	struct ctl_table *table;
2655 	int i;
2656 
2657 	net->mpls.platform_labels = 0;
2658 	net->mpls.platform_label = NULL;
2659 	net->mpls.ip_ttl_propagate = 1;
2660 	net->mpls.default_ttl = 255;
2661 
2662 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2663 	if (table == NULL)
2664 		return -ENOMEM;
2665 
2666 	/* Table data contains only offsets relative to the base of
2667 	 * the mdev at this point, so make them absolute.
2668 	 */
2669 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2670 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2671 
2672 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2673 	if (net->mpls.ctl == NULL) {
2674 		kfree(table);
2675 		return -ENOMEM;
2676 	}
2677 
2678 	return 0;
2679 }
2680 
2681 static void mpls_net_exit(struct net *net)
2682 {
2683 	struct mpls_route __rcu **platform_label;
2684 	size_t platform_labels;
2685 	struct ctl_table *table;
2686 	unsigned int index;
2687 
2688 	table = net->mpls.ctl->ctl_table_arg;
2689 	unregister_net_sysctl_table(net->mpls.ctl);
2690 	kfree(table);
2691 
2692 	/* An rcu grace period has passed since there was a device in
2693 	 * the network namespace (and thus the last in flight packet)
2694 	 * left this network namespace.  This is because
2695 	 * unregister_netdevice_many and netdev_run_todo has completed
2696 	 * for each network device that was in this network namespace.
2697 	 *
2698 	 * As such no additional rcu synchronization is necessary when
2699 	 * freeing the platform_label table.
2700 	 */
2701 	rtnl_lock();
2702 	platform_label = rtnl_dereference(net->mpls.platform_label);
2703 	platform_labels = net->mpls.platform_labels;
2704 	for (index = 0; index < platform_labels; index++) {
2705 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2706 		RCU_INIT_POINTER(platform_label[index], NULL);
2707 		mpls_notify_route(net, index, rt, NULL, NULL);
2708 		mpls_rt_free(rt);
2709 	}
2710 	rtnl_unlock();
2711 
2712 	kvfree(platform_label);
2713 }
2714 
2715 static struct pernet_operations mpls_net_ops = {
2716 	.init = mpls_net_init,
2717 	.exit = mpls_net_exit,
2718 };
2719 
2720 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2721 	.family		   = AF_MPLS,
2722 	.fill_stats_af	   = mpls_fill_stats_af,
2723 	.get_stats_af_size = mpls_get_stats_af_size,
2724 };
2725 
2726 static int __init mpls_init(void)
2727 {
2728 	int err;
2729 
2730 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2731 
2732 	err = register_pernet_subsys(&mpls_net_ops);
2733 	if (err)
2734 		goto out;
2735 
2736 	err = register_netdevice_notifier(&mpls_dev_notifier);
2737 	if (err)
2738 		goto out_unregister_pernet;
2739 
2740 	dev_add_pack(&mpls_packet_type);
2741 
2742 	rtnl_af_register(&mpls_af_ops);
2743 
2744 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2745 			     mpls_rtm_newroute, NULL, 0);
2746 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2747 			     mpls_rtm_delroute, NULL, 0);
2748 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2749 			     mpls_getroute, mpls_dump_routes, 0);
2750 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2751 			     mpls_netconf_get_devconf,
2752 			     mpls_netconf_dump_devconf, 0);
2753 	err = ipgre_tunnel_encap_add_mpls_ops();
2754 	if (err)
2755 		pr_err("Can't add mpls over gre tunnel ops\n");
2756 
2757 	err = 0;
2758 out:
2759 	return err;
2760 
2761 out_unregister_pernet:
2762 	unregister_pernet_subsys(&mpls_net_ops);
2763 	goto out;
2764 }
2765 module_init(mpls_init);
2766 
2767 static void __exit mpls_exit(void)
2768 {
2769 	rtnl_unregister_all(PF_MPLS);
2770 	rtnl_af_unregister(&mpls_af_ops);
2771 	dev_remove_pack(&mpls_packet_type);
2772 	unregister_netdevice_notifier(&mpls_dev_notifier);
2773 	unregister_pernet_subsys(&mpls_net_ops);
2774 	ipgre_tunnel_encap_del_mpls_ops();
2775 }
2776 module_exit(mpls_exit);
2777 
2778 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2779 MODULE_LICENSE("GPL v2");
2780 MODULE_ALIAS_NETPROTO(PF_MPLS);
2781