xref: /openbmc/linux/net/mpls/af_mpls.c (revision 002dff36)
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 
621 	if (!ipv6_stub)
622 		return ERR_PTR(-EAFNOSUPPORT);
623 
624 	memset(&fl6, 0, sizeof(fl6));
625 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
626 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
627 	if (IS_ERR(dst))
628 		return ERR_CAST(dst);
629 
630 	dev = dst->dev;
631 	dev_hold(dev);
632 	dst_release(dst);
633 
634 	return dev;
635 }
636 #else
637 static struct net_device *inet6_fib_lookup_dev(struct net *net,
638 					       const void *addr)
639 {
640 	return ERR_PTR(-EAFNOSUPPORT);
641 }
642 #endif
643 
644 static struct net_device *find_outdev(struct net *net,
645 				      struct mpls_route *rt,
646 				      struct mpls_nh *nh, int oif)
647 {
648 	struct net_device *dev = NULL;
649 
650 	if (!oif) {
651 		switch (nh->nh_via_table) {
652 		case NEIGH_ARP_TABLE:
653 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
654 			break;
655 		case NEIGH_ND_TABLE:
656 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
657 			break;
658 		case NEIGH_LINK_TABLE:
659 			break;
660 		}
661 	} else {
662 		dev = dev_get_by_index(net, oif);
663 	}
664 
665 	if (!dev)
666 		return ERR_PTR(-ENODEV);
667 
668 	if (IS_ERR(dev))
669 		return dev;
670 
671 	/* The caller is holding rtnl anyways, so release the dev reference */
672 	dev_put(dev);
673 
674 	return dev;
675 }
676 
677 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
678 			      struct mpls_nh *nh, int oif)
679 {
680 	struct net_device *dev = NULL;
681 	int err = -ENODEV;
682 
683 	dev = find_outdev(net, rt, nh, oif);
684 	if (IS_ERR(dev)) {
685 		err = PTR_ERR(dev);
686 		dev = NULL;
687 		goto errout;
688 	}
689 
690 	/* Ensure this is a supported device */
691 	err = -EINVAL;
692 	if (!mpls_dev_get(dev))
693 		goto errout;
694 
695 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
696 	    (dev->addr_len != nh->nh_via_alen))
697 		goto errout;
698 
699 	RCU_INIT_POINTER(nh->nh_dev, dev);
700 
701 	if (!(dev->flags & IFF_UP)) {
702 		nh->nh_flags |= RTNH_F_DEAD;
703 	} else {
704 		unsigned int flags;
705 
706 		flags = dev_get_flags(dev);
707 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
708 			nh->nh_flags |= RTNH_F_LINKDOWN;
709 	}
710 
711 	return 0;
712 
713 errout:
714 	return err;
715 }
716 
717 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
718 		       u8 via_addr[], struct netlink_ext_ack *extack)
719 {
720 	struct rtvia *via = nla_data(nla);
721 	int err = -EINVAL;
722 	int alen;
723 
724 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
725 		NL_SET_ERR_MSG_ATTR(extack, nla,
726 				    "Invalid attribute length for RTA_VIA");
727 		goto errout;
728 	}
729 	alen = nla_len(nla) -
730 			offsetof(struct rtvia, rtvia_addr);
731 	if (alen > MAX_VIA_ALEN) {
732 		NL_SET_ERR_MSG_ATTR(extack, nla,
733 				    "Invalid address length for RTA_VIA");
734 		goto errout;
735 	}
736 
737 	/* Validate the address family */
738 	switch (via->rtvia_family) {
739 	case AF_PACKET:
740 		*via_table = NEIGH_LINK_TABLE;
741 		break;
742 	case AF_INET:
743 		*via_table = NEIGH_ARP_TABLE;
744 		if (alen != 4)
745 			goto errout;
746 		break;
747 	case AF_INET6:
748 		*via_table = NEIGH_ND_TABLE;
749 		if (alen != 16)
750 			goto errout;
751 		break;
752 	default:
753 		/* Unsupported address family */
754 		goto errout;
755 	}
756 
757 	memcpy(via_addr, via->rtvia_addr, alen);
758 	*via_alen = alen;
759 	err = 0;
760 
761 errout:
762 	return err;
763 }
764 
765 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
766 				  struct mpls_route *rt)
767 {
768 	struct net *net = cfg->rc_nlinfo.nl_net;
769 	struct mpls_nh *nh = rt->rt_nh;
770 	int err;
771 	int i;
772 
773 	if (!nh)
774 		return -ENOMEM;
775 
776 	nh->nh_labels = cfg->rc_output_labels;
777 	for (i = 0; i < nh->nh_labels; i++)
778 		nh->nh_label[i] = cfg->rc_output_label[i];
779 
780 	nh->nh_via_table = cfg->rc_via_table;
781 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
782 	nh->nh_via_alen = cfg->rc_via_alen;
783 
784 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
785 	if (err)
786 		goto errout;
787 
788 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
789 		rt->rt_nhn_alive--;
790 
791 	return 0;
792 
793 errout:
794 	return err;
795 }
796 
797 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
798 			 struct mpls_nh *nh, int oif, struct nlattr *via,
799 			 struct nlattr *newdst, u8 max_labels,
800 			 struct netlink_ext_ack *extack)
801 {
802 	int err = -ENOMEM;
803 
804 	if (!nh)
805 		goto errout;
806 
807 	if (newdst) {
808 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
809 				     nh->nh_label, extack);
810 		if (err)
811 			goto errout;
812 	}
813 
814 	if (via) {
815 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
816 				  __mpls_nh_via(rt, nh), extack);
817 		if (err)
818 			goto errout;
819 	} else {
820 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
821 	}
822 
823 	err = mpls_nh_assign_dev(net, rt, nh, oif);
824 	if (err)
825 		goto errout;
826 
827 	return 0;
828 
829 errout:
830 	return err;
831 }
832 
833 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
834 			      u8 cfg_via_alen, u8 *max_via_alen,
835 			      u8 *max_labels)
836 {
837 	int remaining = len;
838 	u8 nhs = 0;
839 
840 	*max_via_alen = 0;
841 	*max_labels = 0;
842 
843 	while (rtnh_ok(rtnh, remaining)) {
844 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
845 		int attrlen;
846 		u8 n_labels = 0;
847 
848 		attrlen = rtnh_attrlen(rtnh);
849 		nla = nla_find(attrs, attrlen, RTA_VIA);
850 		if (nla && nla_len(nla) >=
851 		    offsetof(struct rtvia, rtvia_addr)) {
852 			int via_alen = nla_len(nla) -
853 				offsetof(struct rtvia, rtvia_addr);
854 
855 			if (via_alen <= MAX_VIA_ALEN)
856 				*max_via_alen = max_t(u16, *max_via_alen,
857 						      via_alen);
858 		}
859 
860 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
861 		if (nla &&
862 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
863 				   NULL, NULL) != 0)
864 			return 0;
865 
866 		*max_labels = max_t(u8, *max_labels, n_labels);
867 
868 		/* number of nexthops is tracked by a u8.
869 		 * Check for overflow.
870 		 */
871 		if (nhs == 255)
872 			return 0;
873 		nhs++;
874 
875 		rtnh = rtnh_next(rtnh, &remaining);
876 	}
877 
878 	/* leftover implies invalid nexthop configuration, discard it */
879 	return remaining > 0 ? 0 : nhs;
880 }
881 
882 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
883 			       struct mpls_route *rt, u8 max_labels,
884 			       struct netlink_ext_ack *extack)
885 {
886 	struct rtnexthop *rtnh = cfg->rc_mp;
887 	struct nlattr *nla_via, *nla_newdst;
888 	int remaining = cfg->rc_mp_len;
889 	int err = 0;
890 	u8 nhs = 0;
891 
892 	change_nexthops(rt) {
893 		int attrlen;
894 
895 		nla_via = NULL;
896 		nla_newdst = NULL;
897 
898 		err = -EINVAL;
899 		if (!rtnh_ok(rtnh, remaining))
900 			goto errout;
901 
902 		/* neither weighted multipath nor any flags
903 		 * are supported
904 		 */
905 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
906 			goto errout;
907 
908 		attrlen = rtnh_attrlen(rtnh);
909 		if (attrlen > 0) {
910 			struct nlattr *attrs = rtnh_attrs(rtnh);
911 
912 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
913 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
914 		}
915 
916 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
917 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
918 				    max_labels, extack);
919 		if (err)
920 			goto errout;
921 
922 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
923 			rt->rt_nhn_alive--;
924 
925 		rtnh = rtnh_next(rtnh, &remaining);
926 		nhs++;
927 	} endfor_nexthops(rt);
928 
929 	rt->rt_nhn = nhs;
930 
931 	return 0;
932 
933 errout:
934 	return err;
935 }
936 
937 static bool mpls_label_ok(struct net *net, unsigned int *index,
938 			  struct netlink_ext_ack *extack)
939 {
940 	bool is_ok = true;
941 
942 	/* Reserved labels may not be set */
943 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
944 		NL_SET_ERR_MSG(extack,
945 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
946 		is_ok = false;
947 	}
948 
949 	/* The full 20 bit range may not be supported. */
950 	if (is_ok && *index >= net->mpls.platform_labels) {
951 		NL_SET_ERR_MSG(extack,
952 			       "Label >= configured maximum in platform_labels");
953 		is_ok = false;
954 	}
955 
956 	*index = array_index_nospec(*index, net->mpls.platform_labels);
957 	return is_ok;
958 }
959 
960 static int mpls_route_add(struct mpls_route_config *cfg,
961 			  struct netlink_ext_ack *extack)
962 {
963 	struct mpls_route __rcu **platform_label;
964 	struct net *net = cfg->rc_nlinfo.nl_net;
965 	struct mpls_route *rt, *old;
966 	int err = -EINVAL;
967 	u8 max_via_alen;
968 	unsigned index;
969 	u8 max_labels;
970 	u8 nhs;
971 
972 	index = cfg->rc_label;
973 
974 	/* If a label was not specified during insert pick one */
975 	if ((index == LABEL_NOT_SPECIFIED) &&
976 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
977 		index = find_free_label(net);
978 	}
979 
980 	if (!mpls_label_ok(net, &index, extack))
981 		goto errout;
982 
983 	/* Append makes no sense with mpls */
984 	err = -EOPNOTSUPP;
985 	if (cfg->rc_nlflags & NLM_F_APPEND) {
986 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
987 		goto errout;
988 	}
989 
990 	err = -EEXIST;
991 	platform_label = rtnl_dereference(net->mpls.platform_label);
992 	old = rtnl_dereference(platform_label[index]);
993 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
994 		goto errout;
995 
996 	err = -EEXIST;
997 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
998 		goto errout;
999 
1000 	err = -ENOENT;
1001 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1002 		goto errout;
1003 
1004 	err = -EINVAL;
1005 	if (cfg->rc_mp) {
1006 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1007 					  cfg->rc_via_alen, &max_via_alen,
1008 					  &max_labels);
1009 	} else {
1010 		max_via_alen = cfg->rc_via_alen;
1011 		max_labels = cfg->rc_output_labels;
1012 		nhs = 1;
1013 	}
1014 
1015 	if (nhs == 0) {
1016 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1017 		goto errout;
1018 	}
1019 
1020 	err = -ENOMEM;
1021 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1022 	if (IS_ERR(rt)) {
1023 		err = PTR_ERR(rt);
1024 		goto errout;
1025 	}
1026 
1027 	rt->rt_protocol = cfg->rc_protocol;
1028 	rt->rt_payload_type = cfg->rc_payload_type;
1029 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1030 
1031 	if (cfg->rc_mp)
1032 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1033 	else
1034 		err = mpls_nh_build_from_cfg(cfg, rt);
1035 	if (err)
1036 		goto freert;
1037 
1038 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1039 
1040 	return 0;
1041 
1042 freert:
1043 	mpls_rt_free(rt);
1044 errout:
1045 	return err;
1046 }
1047 
1048 static int mpls_route_del(struct mpls_route_config *cfg,
1049 			  struct netlink_ext_ack *extack)
1050 {
1051 	struct net *net = cfg->rc_nlinfo.nl_net;
1052 	unsigned index;
1053 	int err = -EINVAL;
1054 
1055 	index = cfg->rc_label;
1056 
1057 	if (!mpls_label_ok(net, &index, extack))
1058 		goto errout;
1059 
1060 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1061 
1062 	err = 0;
1063 errout:
1064 	return err;
1065 }
1066 
1067 static void mpls_get_stats(struct mpls_dev *mdev,
1068 			   struct mpls_link_stats *stats)
1069 {
1070 	struct mpls_pcpu_stats *p;
1071 	int i;
1072 
1073 	memset(stats, 0, sizeof(*stats));
1074 
1075 	for_each_possible_cpu(i) {
1076 		struct mpls_link_stats local;
1077 		unsigned int start;
1078 
1079 		p = per_cpu_ptr(mdev->stats, i);
1080 		do {
1081 			start = u64_stats_fetch_begin(&p->syncp);
1082 			local = p->stats;
1083 		} while (u64_stats_fetch_retry(&p->syncp, start));
1084 
1085 		stats->rx_packets	+= local.rx_packets;
1086 		stats->rx_bytes		+= local.rx_bytes;
1087 		stats->tx_packets	+= local.tx_packets;
1088 		stats->tx_bytes		+= local.tx_bytes;
1089 		stats->rx_errors	+= local.rx_errors;
1090 		stats->tx_errors	+= local.tx_errors;
1091 		stats->rx_dropped	+= local.rx_dropped;
1092 		stats->tx_dropped	+= local.tx_dropped;
1093 		stats->rx_noroute	+= local.rx_noroute;
1094 	}
1095 }
1096 
1097 static int mpls_fill_stats_af(struct sk_buff *skb,
1098 			      const struct net_device *dev)
1099 {
1100 	struct mpls_link_stats *stats;
1101 	struct mpls_dev *mdev;
1102 	struct nlattr *nla;
1103 
1104 	mdev = mpls_dev_get(dev);
1105 	if (!mdev)
1106 		return -ENODATA;
1107 
1108 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1109 				sizeof(struct mpls_link_stats),
1110 				MPLS_STATS_UNSPEC);
1111 	if (!nla)
1112 		return -EMSGSIZE;
1113 
1114 	stats = nla_data(nla);
1115 	mpls_get_stats(mdev, stats);
1116 
1117 	return 0;
1118 }
1119 
1120 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1121 {
1122 	struct mpls_dev *mdev;
1123 
1124 	mdev = mpls_dev_get(dev);
1125 	if (!mdev)
1126 		return 0;
1127 
1128 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1129 }
1130 
1131 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1132 				     u32 portid, u32 seq, int event,
1133 				     unsigned int flags, int type)
1134 {
1135 	struct nlmsghdr  *nlh;
1136 	struct netconfmsg *ncm;
1137 	bool all = false;
1138 
1139 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1140 			flags);
1141 	if (!nlh)
1142 		return -EMSGSIZE;
1143 
1144 	if (type == NETCONFA_ALL)
1145 		all = true;
1146 
1147 	ncm = nlmsg_data(nlh);
1148 	ncm->ncm_family = AF_MPLS;
1149 
1150 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1151 		goto nla_put_failure;
1152 
1153 	if ((all || type == NETCONFA_INPUT) &&
1154 	    nla_put_s32(skb, NETCONFA_INPUT,
1155 			mdev->input_enabled) < 0)
1156 		goto nla_put_failure;
1157 
1158 	nlmsg_end(skb, nlh);
1159 	return 0;
1160 
1161 nla_put_failure:
1162 	nlmsg_cancel(skb, nlh);
1163 	return -EMSGSIZE;
1164 }
1165 
1166 static int mpls_netconf_msgsize_devconf(int type)
1167 {
1168 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1169 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1170 	bool all = false;
1171 
1172 	if (type == NETCONFA_ALL)
1173 		all = true;
1174 
1175 	if (all || type == NETCONFA_INPUT)
1176 		size += nla_total_size(4);
1177 
1178 	return size;
1179 }
1180 
1181 static void mpls_netconf_notify_devconf(struct net *net, int event,
1182 					int type, struct mpls_dev *mdev)
1183 {
1184 	struct sk_buff *skb;
1185 	int err = -ENOBUFS;
1186 
1187 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1188 	if (!skb)
1189 		goto errout;
1190 
1191 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1192 	if (err < 0) {
1193 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1194 		WARN_ON(err == -EMSGSIZE);
1195 		kfree_skb(skb);
1196 		goto errout;
1197 	}
1198 
1199 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1200 	return;
1201 errout:
1202 	if (err < 0)
1203 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1204 }
1205 
1206 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1207 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1208 };
1209 
1210 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1211 				      const struct nlmsghdr *nlh,
1212 				      struct nlattr **tb,
1213 				      struct netlink_ext_ack *extack)
1214 {
1215 	int i, err;
1216 
1217 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1218 		NL_SET_ERR_MSG_MOD(extack,
1219 				   "Invalid header for netconf get request");
1220 		return -EINVAL;
1221 	}
1222 
1223 	if (!netlink_strict_get_check(skb))
1224 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1225 					      tb, NETCONFA_MAX,
1226 					      devconf_mpls_policy, extack);
1227 
1228 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1229 					    tb, NETCONFA_MAX,
1230 					    devconf_mpls_policy, extack);
1231 	if (err)
1232 		return err;
1233 
1234 	for (i = 0; i <= NETCONFA_MAX; i++) {
1235 		if (!tb[i])
1236 			continue;
1237 
1238 		switch (i) {
1239 		case NETCONFA_IFINDEX:
1240 			break;
1241 		default:
1242 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1243 			return -EINVAL;
1244 		}
1245 	}
1246 
1247 	return 0;
1248 }
1249 
1250 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1251 				    struct nlmsghdr *nlh,
1252 				    struct netlink_ext_ack *extack)
1253 {
1254 	struct net *net = sock_net(in_skb->sk);
1255 	struct nlattr *tb[NETCONFA_MAX + 1];
1256 	struct net_device *dev;
1257 	struct mpls_dev *mdev;
1258 	struct sk_buff *skb;
1259 	int ifindex;
1260 	int err;
1261 
1262 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1263 	if (err < 0)
1264 		goto errout;
1265 
1266 	err = -EINVAL;
1267 	if (!tb[NETCONFA_IFINDEX])
1268 		goto errout;
1269 
1270 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1271 	dev = __dev_get_by_index(net, ifindex);
1272 	if (!dev)
1273 		goto errout;
1274 
1275 	mdev = mpls_dev_get(dev);
1276 	if (!mdev)
1277 		goto errout;
1278 
1279 	err = -ENOBUFS;
1280 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1281 	if (!skb)
1282 		goto errout;
1283 
1284 	err = mpls_netconf_fill_devconf(skb, mdev,
1285 					NETLINK_CB(in_skb).portid,
1286 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1287 					NETCONFA_ALL);
1288 	if (err < 0) {
1289 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1290 		WARN_ON(err == -EMSGSIZE);
1291 		kfree_skb(skb);
1292 		goto errout;
1293 	}
1294 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1295 errout:
1296 	return err;
1297 }
1298 
1299 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1300 				     struct netlink_callback *cb)
1301 {
1302 	const struct nlmsghdr *nlh = cb->nlh;
1303 	struct net *net = sock_net(skb->sk);
1304 	struct hlist_head *head;
1305 	struct net_device *dev;
1306 	struct mpls_dev *mdev;
1307 	int idx, s_idx;
1308 	int h, s_h;
1309 
1310 	if (cb->strict_check) {
1311 		struct netlink_ext_ack *extack = cb->extack;
1312 		struct netconfmsg *ncm;
1313 
1314 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1315 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1316 			return -EINVAL;
1317 		}
1318 
1319 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1320 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1321 			return -EINVAL;
1322 		}
1323 	}
1324 
1325 	s_h = cb->args[0];
1326 	s_idx = idx = cb->args[1];
1327 
1328 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1329 		idx = 0;
1330 		head = &net->dev_index_head[h];
1331 		rcu_read_lock();
1332 		cb->seq = net->dev_base_seq;
1333 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1334 			if (idx < s_idx)
1335 				goto cont;
1336 			mdev = mpls_dev_get(dev);
1337 			if (!mdev)
1338 				goto cont;
1339 			if (mpls_netconf_fill_devconf(skb, mdev,
1340 						      NETLINK_CB(cb->skb).portid,
1341 						      nlh->nlmsg_seq,
1342 						      RTM_NEWNETCONF,
1343 						      NLM_F_MULTI,
1344 						      NETCONFA_ALL) < 0) {
1345 				rcu_read_unlock();
1346 				goto done;
1347 			}
1348 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1349 cont:
1350 			idx++;
1351 		}
1352 		rcu_read_unlock();
1353 	}
1354 done:
1355 	cb->args[0] = h;
1356 	cb->args[1] = idx;
1357 
1358 	return skb->len;
1359 }
1360 
1361 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1362 	(&((struct mpls_dev *)0)->field)
1363 
1364 static int mpls_conf_proc(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 
1397 static int mpls_dev_sysctl_register(struct net_device *dev,
1398 				    struct mpls_dev *mdev)
1399 {
1400 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
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 < ARRAY_SIZE(mpls_dev_table); 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(net, path, table);
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 	return -ENOBUFS;
1431 }
1432 
1433 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1434 				       struct mpls_dev *mdev)
1435 {
1436 	struct net *net = dev_net(dev);
1437 	struct ctl_table *table;
1438 
1439 	table = mdev->sysctl->ctl_table_arg;
1440 	unregister_net_sysctl_table(mdev->sysctl);
1441 	kfree(table);
1442 
1443 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1444 }
1445 
1446 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1447 {
1448 	struct mpls_dev *mdev;
1449 	int err = -ENOMEM;
1450 	int i;
1451 
1452 	ASSERT_RTNL();
1453 
1454 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1455 	if (!mdev)
1456 		return ERR_PTR(err);
1457 
1458 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1459 	if (!mdev->stats)
1460 		goto free;
1461 
1462 	for_each_possible_cpu(i) {
1463 		struct mpls_pcpu_stats *mpls_stats;
1464 
1465 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1466 		u64_stats_init(&mpls_stats->syncp);
1467 	}
1468 
1469 	mdev->dev = dev;
1470 
1471 	err = mpls_dev_sysctl_register(dev, mdev);
1472 	if (err)
1473 		goto free;
1474 
1475 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1476 
1477 	return mdev;
1478 
1479 free:
1480 	free_percpu(mdev->stats);
1481 	kfree(mdev);
1482 	return ERR_PTR(err);
1483 }
1484 
1485 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1486 {
1487 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1488 
1489 	free_percpu(mdev->stats);
1490 	kfree(mdev);
1491 }
1492 
1493 static void mpls_ifdown(struct net_device *dev, int event)
1494 {
1495 	struct mpls_route __rcu **platform_label;
1496 	struct net *net = dev_net(dev);
1497 	u8 alive, deleted;
1498 	unsigned index;
1499 
1500 	platform_label = rtnl_dereference(net->mpls.platform_label);
1501 	for (index = 0; index < net->mpls.platform_labels; index++) {
1502 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1503 
1504 		if (!rt)
1505 			continue;
1506 
1507 		alive = 0;
1508 		deleted = 0;
1509 		change_nexthops(rt) {
1510 			unsigned int nh_flags = nh->nh_flags;
1511 
1512 			if (rtnl_dereference(nh->nh_dev) != dev)
1513 				goto next;
1514 
1515 			switch (event) {
1516 			case NETDEV_DOWN:
1517 			case NETDEV_UNREGISTER:
1518 				nh_flags |= RTNH_F_DEAD;
1519 				/* fall through */
1520 			case NETDEV_CHANGE:
1521 				nh_flags |= RTNH_F_LINKDOWN;
1522 				break;
1523 			}
1524 			if (event == NETDEV_UNREGISTER)
1525 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1526 
1527 			if (nh->nh_flags != nh_flags)
1528 				WRITE_ONCE(nh->nh_flags, nh_flags);
1529 next:
1530 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1531 				alive++;
1532 			if (!rtnl_dereference(nh->nh_dev))
1533 				deleted++;
1534 		} endfor_nexthops(rt);
1535 
1536 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1537 
1538 		/* if there are no more nexthops, delete the route */
1539 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1540 			mpls_route_update(net, index, NULL, NULL);
1541 	}
1542 }
1543 
1544 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1545 {
1546 	struct mpls_route __rcu **platform_label;
1547 	struct net *net = dev_net(dev);
1548 	unsigned index;
1549 	u8 alive;
1550 
1551 	platform_label = rtnl_dereference(net->mpls.platform_label);
1552 	for (index = 0; index < net->mpls.platform_labels; index++) {
1553 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1554 
1555 		if (!rt)
1556 			continue;
1557 
1558 		alive = 0;
1559 		change_nexthops(rt) {
1560 			unsigned int nh_flags = nh->nh_flags;
1561 			struct net_device *nh_dev =
1562 				rtnl_dereference(nh->nh_dev);
1563 
1564 			if (!(nh_flags & flags)) {
1565 				alive++;
1566 				continue;
1567 			}
1568 			if (nh_dev != dev)
1569 				continue;
1570 			alive++;
1571 			nh_flags &= ~flags;
1572 			WRITE_ONCE(nh->nh_flags, nh_flags);
1573 		} endfor_nexthops(rt);
1574 
1575 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1576 	}
1577 }
1578 
1579 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1580 			   void *ptr)
1581 {
1582 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1583 	struct mpls_dev *mdev;
1584 	unsigned int flags;
1585 
1586 	if (event == NETDEV_REGISTER) {
1587 
1588 		/* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1589 		 * IPIP devices
1590 		 */
1591 		if (dev->type == ARPHRD_ETHER ||
1592 		    dev->type == ARPHRD_LOOPBACK ||
1593 		    dev->type == ARPHRD_IPGRE ||
1594 		    dev->type == ARPHRD_IP6GRE ||
1595 		    dev->type == ARPHRD_SIT ||
1596 		    dev->type == ARPHRD_TUNNEL ||
1597 		    dev->type == ARPHRD_TUNNEL6) {
1598 			mdev = mpls_add_dev(dev);
1599 			if (IS_ERR(mdev))
1600 				return notifier_from_errno(PTR_ERR(mdev));
1601 		}
1602 		return NOTIFY_OK;
1603 	}
1604 
1605 	mdev = mpls_dev_get(dev);
1606 	if (!mdev)
1607 		return NOTIFY_OK;
1608 
1609 	switch (event) {
1610 	case NETDEV_DOWN:
1611 		mpls_ifdown(dev, event);
1612 		break;
1613 	case NETDEV_UP:
1614 		flags = dev_get_flags(dev);
1615 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1616 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1617 		else
1618 			mpls_ifup(dev, RTNH_F_DEAD);
1619 		break;
1620 	case NETDEV_CHANGE:
1621 		flags = dev_get_flags(dev);
1622 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1623 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1624 		else
1625 			mpls_ifdown(dev, event);
1626 		break;
1627 	case NETDEV_UNREGISTER:
1628 		mpls_ifdown(dev, event);
1629 		mdev = mpls_dev_get(dev);
1630 		if (mdev) {
1631 			mpls_dev_sysctl_unregister(dev, mdev);
1632 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1633 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1634 		}
1635 		break;
1636 	case NETDEV_CHANGENAME:
1637 		mdev = mpls_dev_get(dev);
1638 		if (mdev) {
1639 			int err;
1640 
1641 			mpls_dev_sysctl_unregister(dev, mdev);
1642 			err = mpls_dev_sysctl_register(dev, mdev);
1643 			if (err)
1644 				return notifier_from_errno(err);
1645 		}
1646 		break;
1647 	}
1648 	return NOTIFY_OK;
1649 }
1650 
1651 static struct notifier_block mpls_dev_notifier = {
1652 	.notifier_call = mpls_dev_notify,
1653 };
1654 
1655 static int nla_put_via(struct sk_buff *skb,
1656 		       u8 table, const void *addr, int alen)
1657 {
1658 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1659 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1660 	};
1661 	struct nlattr *nla;
1662 	struct rtvia *via;
1663 	int family = AF_UNSPEC;
1664 
1665 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1666 	if (!nla)
1667 		return -EMSGSIZE;
1668 
1669 	if (table <= NEIGH_NR_TABLES)
1670 		family = table_to_family[table];
1671 
1672 	via = nla_data(nla);
1673 	via->rtvia_family = family;
1674 	memcpy(via->rtvia_addr, addr, alen);
1675 	return 0;
1676 }
1677 
1678 int nla_put_labels(struct sk_buff *skb, int attrtype,
1679 		   u8 labels, const u32 label[])
1680 {
1681 	struct nlattr *nla;
1682 	struct mpls_shim_hdr *nla_label;
1683 	bool bos;
1684 	int i;
1685 	nla = nla_reserve(skb, attrtype, labels*4);
1686 	if (!nla)
1687 		return -EMSGSIZE;
1688 
1689 	nla_label = nla_data(nla);
1690 	bos = true;
1691 	for (i = labels - 1; i >= 0; i--) {
1692 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1693 		bos = false;
1694 	}
1695 
1696 	return 0;
1697 }
1698 EXPORT_SYMBOL_GPL(nla_put_labels);
1699 
1700 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1701 		   u32 label[], struct netlink_ext_ack *extack)
1702 {
1703 	unsigned len = nla_len(nla);
1704 	struct mpls_shim_hdr *nla_label;
1705 	u8 nla_labels;
1706 	bool bos;
1707 	int i;
1708 
1709 	/* len needs to be an even multiple of 4 (the label size). Number
1710 	 * of labels is a u8 so check for overflow.
1711 	 */
1712 	if (len & 3 || len / 4 > 255) {
1713 		NL_SET_ERR_MSG_ATTR(extack, nla,
1714 				    "Invalid length for labels attribute");
1715 		return -EINVAL;
1716 	}
1717 
1718 	/* Limit the number of new labels allowed */
1719 	nla_labels = len/4;
1720 	if (nla_labels > max_labels) {
1721 		NL_SET_ERR_MSG(extack, "Too many labels");
1722 		return -EINVAL;
1723 	}
1724 
1725 	/* when label == NULL, caller wants number of labels */
1726 	if (!label)
1727 		goto out;
1728 
1729 	nla_label = nla_data(nla);
1730 	bos = true;
1731 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1732 		struct mpls_entry_decoded dec;
1733 		dec = mpls_entry_decode(nla_label + i);
1734 
1735 		/* Ensure the bottom of stack flag is properly set
1736 		 * and ttl and tc are both clear.
1737 		 */
1738 		if (dec.ttl) {
1739 			NL_SET_ERR_MSG_ATTR(extack, nla,
1740 					    "TTL in label must be 0");
1741 			return -EINVAL;
1742 		}
1743 
1744 		if (dec.tc) {
1745 			NL_SET_ERR_MSG_ATTR(extack, nla,
1746 					    "Traffic class in label must be 0");
1747 			return -EINVAL;
1748 		}
1749 
1750 		if (dec.bos != bos) {
1751 			NL_SET_BAD_ATTR(extack, nla);
1752 			if (bos) {
1753 				NL_SET_ERR_MSG(extack,
1754 					       "BOS bit must be set in first label");
1755 			} else {
1756 				NL_SET_ERR_MSG(extack,
1757 					       "BOS bit can only be set in first label");
1758 			}
1759 			return -EINVAL;
1760 		}
1761 
1762 		switch (dec.label) {
1763 		case MPLS_LABEL_IMPLNULL:
1764 			/* RFC3032: This is a label that an LSR may
1765 			 * assign and distribute, but which never
1766 			 * actually appears in the encapsulation.
1767 			 */
1768 			NL_SET_ERR_MSG_ATTR(extack, nla,
1769 					    "Implicit NULL Label (3) can not be used in encapsulation");
1770 			return -EINVAL;
1771 		}
1772 
1773 		label[i] = dec.label;
1774 	}
1775 out:
1776 	*labels = nla_labels;
1777 	return 0;
1778 }
1779 EXPORT_SYMBOL_GPL(nla_get_labels);
1780 
1781 static int rtm_to_route_config(struct sk_buff *skb,
1782 			       struct nlmsghdr *nlh,
1783 			       struct mpls_route_config *cfg,
1784 			       struct netlink_ext_ack *extack)
1785 {
1786 	struct rtmsg *rtm;
1787 	struct nlattr *tb[RTA_MAX+1];
1788 	int index;
1789 	int err;
1790 
1791 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1792 				     rtm_mpls_policy, extack);
1793 	if (err < 0)
1794 		goto errout;
1795 
1796 	err = -EINVAL;
1797 	rtm = nlmsg_data(nlh);
1798 
1799 	if (rtm->rtm_family != AF_MPLS) {
1800 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1801 		goto errout;
1802 	}
1803 	if (rtm->rtm_dst_len != 20) {
1804 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1805 		goto errout;
1806 	}
1807 	if (rtm->rtm_src_len != 0) {
1808 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1809 		goto errout;
1810 	}
1811 	if (rtm->rtm_tos != 0) {
1812 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1813 		goto errout;
1814 	}
1815 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1816 		NL_SET_ERR_MSG(extack,
1817 			       "MPLS only supports the main route table");
1818 		goto errout;
1819 	}
1820 	/* Any value is acceptable for rtm_protocol */
1821 
1822 	/* As mpls uses destination specific addresses
1823 	 * (or source specific address in the case of multicast)
1824 	 * all addresses have universal scope.
1825 	 */
1826 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1827 		NL_SET_ERR_MSG(extack,
1828 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1829 		goto errout;
1830 	}
1831 	if (rtm->rtm_type != RTN_UNICAST) {
1832 		NL_SET_ERR_MSG(extack,
1833 			       "Invalid route type - MPLS only supports UNICAST");
1834 		goto errout;
1835 	}
1836 	if (rtm->rtm_flags != 0) {
1837 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1838 		goto errout;
1839 	}
1840 
1841 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1842 	cfg->rc_protocol	= rtm->rtm_protocol;
1843 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1844 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1845 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1846 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1847 	cfg->rc_nlinfo.nlh	= nlh;
1848 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1849 
1850 	for (index = 0; index <= RTA_MAX; index++) {
1851 		struct nlattr *nla = tb[index];
1852 		if (!nla)
1853 			continue;
1854 
1855 		switch (index) {
1856 		case RTA_OIF:
1857 			cfg->rc_ifindex = nla_get_u32(nla);
1858 			break;
1859 		case RTA_NEWDST:
1860 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1861 					   &cfg->rc_output_labels,
1862 					   cfg->rc_output_label, extack))
1863 				goto errout;
1864 			break;
1865 		case RTA_DST:
1866 		{
1867 			u8 label_count;
1868 			if (nla_get_labels(nla, 1, &label_count,
1869 					   &cfg->rc_label, extack))
1870 				goto errout;
1871 
1872 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1873 					   &cfg->rc_label, extack))
1874 				goto errout;
1875 			break;
1876 		}
1877 		case RTA_GATEWAY:
1878 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1879 			goto errout;
1880 		case RTA_VIA:
1881 		{
1882 			if (nla_get_via(nla, &cfg->rc_via_alen,
1883 					&cfg->rc_via_table, cfg->rc_via,
1884 					extack))
1885 				goto errout;
1886 			break;
1887 		}
1888 		case RTA_MULTIPATH:
1889 		{
1890 			cfg->rc_mp = nla_data(nla);
1891 			cfg->rc_mp_len = nla_len(nla);
1892 			break;
1893 		}
1894 		case RTA_TTL_PROPAGATE:
1895 		{
1896 			u8 ttl_propagate = nla_get_u8(nla);
1897 
1898 			if (ttl_propagate > 1) {
1899 				NL_SET_ERR_MSG_ATTR(extack, nla,
1900 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1901 				goto errout;
1902 			}
1903 			cfg->rc_ttl_propagate = ttl_propagate ?
1904 				MPLS_TTL_PROP_ENABLED :
1905 				MPLS_TTL_PROP_DISABLED;
1906 			break;
1907 		}
1908 		default:
1909 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1910 			/* Unsupported attribute */
1911 			goto errout;
1912 		}
1913 	}
1914 
1915 	err = 0;
1916 errout:
1917 	return err;
1918 }
1919 
1920 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1921 			     struct netlink_ext_ack *extack)
1922 {
1923 	struct mpls_route_config *cfg;
1924 	int err;
1925 
1926 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1927 	if (!cfg)
1928 		return -ENOMEM;
1929 
1930 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1931 	if (err < 0)
1932 		goto out;
1933 
1934 	err = mpls_route_del(cfg, extack);
1935 out:
1936 	kfree(cfg);
1937 
1938 	return err;
1939 }
1940 
1941 
1942 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1943 			     struct netlink_ext_ack *extack)
1944 {
1945 	struct mpls_route_config *cfg;
1946 	int err;
1947 
1948 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1949 	if (!cfg)
1950 		return -ENOMEM;
1951 
1952 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1953 	if (err < 0)
1954 		goto out;
1955 
1956 	err = mpls_route_add(cfg, extack);
1957 out:
1958 	kfree(cfg);
1959 
1960 	return err;
1961 }
1962 
1963 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1964 			   u32 label, struct mpls_route *rt, int flags)
1965 {
1966 	struct net_device *dev;
1967 	struct nlmsghdr *nlh;
1968 	struct rtmsg *rtm;
1969 
1970 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1971 	if (nlh == NULL)
1972 		return -EMSGSIZE;
1973 
1974 	rtm = nlmsg_data(nlh);
1975 	rtm->rtm_family = AF_MPLS;
1976 	rtm->rtm_dst_len = 20;
1977 	rtm->rtm_src_len = 0;
1978 	rtm->rtm_tos = 0;
1979 	rtm->rtm_table = RT_TABLE_MAIN;
1980 	rtm->rtm_protocol = rt->rt_protocol;
1981 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1982 	rtm->rtm_type = RTN_UNICAST;
1983 	rtm->rtm_flags = 0;
1984 
1985 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1986 		goto nla_put_failure;
1987 
1988 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1989 		bool ttl_propagate =
1990 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1991 
1992 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1993 			       ttl_propagate))
1994 			goto nla_put_failure;
1995 	}
1996 	if (rt->rt_nhn == 1) {
1997 		const struct mpls_nh *nh = rt->rt_nh;
1998 
1999 		if (nh->nh_labels &&
2000 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2001 				   nh->nh_label))
2002 			goto nla_put_failure;
2003 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2004 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2005 				nh->nh_via_alen))
2006 			goto nla_put_failure;
2007 		dev = rtnl_dereference(nh->nh_dev);
2008 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2009 			goto nla_put_failure;
2010 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2011 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2012 		if (nh->nh_flags & RTNH_F_DEAD)
2013 			rtm->rtm_flags |= RTNH_F_DEAD;
2014 	} else {
2015 		struct rtnexthop *rtnh;
2016 		struct nlattr *mp;
2017 		u8 linkdown = 0;
2018 		u8 dead = 0;
2019 
2020 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2021 		if (!mp)
2022 			goto nla_put_failure;
2023 
2024 		for_nexthops(rt) {
2025 			dev = rtnl_dereference(nh->nh_dev);
2026 			if (!dev)
2027 				continue;
2028 
2029 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2030 			if (!rtnh)
2031 				goto nla_put_failure;
2032 
2033 			rtnh->rtnh_ifindex = dev->ifindex;
2034 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2035 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2036 				linkdown++;
2037 			}
2038 			if (nh->nh_flags & RTNH_F_DEAD) {
2039 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2040 				dead++;
2041 			}
2042 
2043 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2044 							    nh->nh_labels,
2045 							    nh->nh_label))
2046 				goto nla_put_failure;
2047 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2048 			    nla_put_via(skb, nh->nh_via_table,
2049 					mpls_nh_via(rt, nh),
2050 					nh->nh_via_alen))
2051 				goto nla_put_failure;
2052 
2053 			/* length of rtnetlink header + attributes */
2054 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2055 		} endfor_nexthops(rt);
2056 
2057 		if (linkdown == rt->rt_nhn)
2058 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2059 		if (dead == rt->rt_nhn)
2060 			rtm->rtm_flags |= RTNH_F_DEAD;
2061 
2062 		nla_nest_end(skb, mp);
2063 	}
2064 
2065 	nlmsg_end(skb, nlh);
2066 	return 0;
2067 
2068 nla_put_failure:
2069 	nlmsg_cancel(skb, nlh);
2070 	return -EMSGSIZE;
2071 }
2072 
2073 #if IS_ENABLED(CONFIG_INET)
2074 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2075 				   struct fib_dump_filter *filter,
2076 				   struct netlink_callback *cb)
2077 {
2078 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2079 }
2080 #else
2081 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2082 				   struct fib_dump_filter *filter,
2083 				   struct netlink_callback *cb)
2084 {
2085 	struct netlink_ext_ack *extack = cb->extack;
2086 	struct nlattr *tb[RTA_MAX + 1];
2087 	struct rtmsg *rtm;
2088 	int err, i;
2089 
2090 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2091 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2092 		return -EINVAL;
2093 	}
2094 
2095 	rtm = nlmsg_data(nlh);
2096 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2097 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2098 	    rtm->rtm_flags) {
2099 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2100 		return -EINVAL;
2101 	}
2102 
2103 	if (rtm->rtm_protocol) {
2104 		filter->protocol = rtm->rtm_protocol;
2105 		filter->filter_set = 1;
2106 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2107 	}
2108 
2109 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2110 					    rtm_mpls_policy, extack);
2111 	if (err < 0)
2112 		return err;
2113 
2114 	for (i = 0; i <= RTA_MAX; ++i) {
2115 		int ifindex;
2116 
2117 		if (i == RTA_OIF) {
2118 			ifindex = nla_get_u32(tb[i]);
2119 			filter->dev = __dev_get_by_index(net, ifindex);
2120 			if (!filter->dev)
2121 				return -ENODEV;
2122 			filter->filter_set = 1;
2123 		} else if (tb[i]) {
2124 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2125 			return -EINVAL;
2126 		}
2127 	}
2128 
2129 	return 0;
2130 }
2131 #endif
2132 
2133 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2134 			     const struct net_device *dev)
2135 {
2136 	struct net_device *nh_dev;
2137 
2138 	if (rt->rt_nhn == 1) {
2139 		struct mpls_nh *nh = rt->rt_nh;
2140 
2141 		nh_dev = rtnl_dereference(nh->nh_dev);
2142 		if (dev == nh_dev)
2143 			return true;
2144 	} else {
2145 		for_nexthops(rt) {
2146 			nh_dev = rtnl_dereference(nh->nh_dev);
2147 			if (nh_dev == dev)
2148 				return true;
2149 		} endfor_nexthops(rt);
2150 	}
2151 
2152 	return false;
2153 }
2154 
2155 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2156 {
2157 	const struct nlmsghdr *nlh = cb->nlh;
2158 	struct net *net = sock_net(skb->sk);
2159 	struct mpls_route __rcu **platform_label;
2160 	struct fib_dump_filter filter = {};
2161 	unsigned int flags = NLM_F_MULTI;
2162 	size_t platform_labels;
2163 	unsigned int index;
2164 
2165 	ASSERT_RTNL();
2166 
2167 	if (cb->strict_check) {
2168 		int err;
2169 
2170 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2171 		if (err < 0)
2172 			return err;
2173 
2174 		/* for MPLS, there is only 1 table with fixed type and flags.
2175 		 * If either are set in the filter then return nothing.
2176 		 */
2177 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2178 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2179 		     filter.flags)
2180 			return skb->len;
2181 	}
2182 
2183 	index = cb->args[0];
2184 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2185 		index = MPLS_LABEL_FIRST_UNRESERVED;
2186 
2187 	platform_label = rtnl_dereference(net->mpls.platform_label);
2188 	platform_labels = net->mpls.platform_labels;
2189 
2190 	if (filter.filter_set)
2191 		flags |= NLM_F_DUMP_FILTERED;
2192 
2193 	for (; index < platform_labels; index++) {
2194 		struct mpls_route *rt;
2195 
2196 		rt = rtnl_dereference(platform_label[index]);
2197 		if (!rt)
2198 			continue;
2199 
2200 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2201 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2202 			continue;
2203 
2204 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2205 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2206 				    index, rt, flags) < 0)
2207 			break;
2208 	}
2209 	cb->args[0] = index;
2210 
2211 	return skb->len;
2212 }
2213 
2214 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2215 {
2216 	size_t payload =
2217 		NLMSG_ALIGN(sizeof(struct rtmsg))
2218 		+ nla_total_size(4)			/* RTA_DST */
2219 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2220 
2221 	if (rt->rt_nhn == 1) {
2222 		struct mpls_nh *nh = rt->rt_nh;
2223 
2224 		if (nh->nh_dev)
2225 			payload += nla_total_size(4); /* RTA_OIF */
2226 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2227 			payload += nla_total_size(2 + nh->nh_via_alen);
2228 		if (nh->nh_labels) /* RTA_NEWDST */
2229 			payload += nla_total_size(nh->nh_labels * 4);
2230 	} else {
2231 		/* each nexthop is packed in an attribute */
2232 		size_t nhsize = 0;
2233 
2234 		for_nexthops(rt) {
2235 			if (!rtnl_dereference(nh->nh_dev))
2236 				continue;
2237 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2238 			/* RTA_VIA */
2239 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2240 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2241 			if (nh->nh_labels)
2242 				nhsize += nla_total_size(nh->nh_labels * 4);
2243 		} endfor_nexthops(rt);
2244 		/* nested attribute */
2245 		payload += nla_total_size(nhsize);
2246 	}
2247 
2248 	return payload;
2249 }
2250 
2251 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2252 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2253 		       unsigned int nlm_flags)
2254 {
2255 	struct sk_buff *skb;
2256 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2257 	int err = -ENOBUFS;
2258 
2259 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2260 	if (skb == NULL)
2261 		goto errout;
2262 
2263 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2264 	if (err < 0) {
2265 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2266 		WARN_ON(err == -EMSGSIZE);
2267 		kfree_skb(skb);
2268 		goto errout;
2269 	}
2270 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2271 
2272 	return;
2273 errout:
2274 	if (err < 0)
2275 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2276 }
2277 
2278 static int mpls_valid_getroute_req(struct sk_buff *skb,
2279 				   const struct nlmsghdr *nlh,
2280 				   struct nlattr **tb,
2281 				   struct netlink_ext_ack *extack)
2282 {
2283 	struct rtmsg *rtm;
2284 	int i, err;
2285 
2286 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2287 		NL_SET_ERR_MSG_MOD(extack,
2288 				   "Invalid header for get route request");
2289 		return -EINVAL;
2290 	}
2291 
2292 	if (!netlink_strict_get_check(skb))
2293 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2294 					      rtm_mpls_policy, extack);
2295 
2296 	rtm = nlmsg_data(nlh);
2297 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2298 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2299 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2300 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2301 		return -EINVAL;
2302 	}
2303 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2304 		NL_SET_ERR_MSG_MOD(extack,
2305 				   "Invalid flags for get route request");
2306 		return -EINVAL;
2307 	}
2308 
2309 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2310 					    rtm_mpls_policy, extack);
2311 	if (err)
2312 		return err;
2313 
2314 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2315 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2316 		return -EINVAL;
2317 	}
2318 
2319 	for (i = 0; i <= RTA_MAX; i++) {
2320 		if (!tb[i])
2321 			continue;
2322 
2323 		switch (i) {
2324 		case RTA_DST:
2325 		case RTA_NEWDST:
2326 			break;
2327 		default:
2328 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2329 			return -EINVAL;
2330 		}
2331 	}
2332 
2333 	return 0;
2334 }
2335 
2336 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2337 			 struct netlink_ext_ack *extack)
2338 {
2339 	struct net *net = sock_net(in_skb->sk);
2340 	u32 portid = NETLINK_CB(in_skb).portid;
2341 	u32 in_label = LABEL_NOT_SPECIFIED;
2342 	struct nlattr *tb[RTA_MAX + 1];
2343 	u32 labels[MAX_NEW_LABELS];
2344 	struct mpls_shim_hdr *hdr;
2345 	unsigned int hdr_size = 0;
2346 	struct net_device *dev;
2347 	struct mpls_route *rt;
2348 	struct rtmsg *rtm, *r;
2349 	struct nlmsghdr *nlh;
2350 	struct sk_buff *skb;
2351 	struct mpls_nh *nh;
2352 	u8 n_labels;
2353 	int err;
2354 
2355 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2356 	if (err < 0)
2357 		goto errout;
2358 
2359 	rtm = nlmsg_data(in_nlh);
2360 
2361 	if (tb[RTA_DST]) {
2362 		u8 label_count;
2363 
2364 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2365 				   &in_label, extack)) {
2366 			err = -EINVAL;
2367 			goto errout;
2368 		}
2369 
2370 		if (!mpls_label_ok(net, &in_label, extack)) {
2371 			err = -EINVAL;
2372 			goto errout;
2373 		}
2374 	}
2375 
2376 	rt = mpls_route_input_rcu(net, in_label);
2377 	if (!rt) {
2378 		err = -ENETUNREACH;
2379 		goto errout;
2380 	}
2381 
2382 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2383 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2384 		if (!skb) {
2385 			err = -ENOBUFS;
2386 			goto errout;
2387 		}
2388 
2389 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2390 				      RTM_NEWROUTE, in_label, rt, 0);
2391 		if (err < 0) {
2392 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2393 			WARN_ON(err == -EMSGSIZE);
2394 			goto errout_free;
2395 		}
2396 
2397 		return rtnl_unicast(skb, net, portid);
2398 	}
2399 
2400 	if (tb[RTA_NEWDST]) {
2401 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2402 				   labels, extack) != 0) {
2403 			err = -EINVAL;
2404 			goto errout;
2405 		}
2406 
2407 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2408 	}
2409 
2410 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2411 	if (!skb) {
2412 		err = -ENOBUFS;
2413 		goto errout;
2414 	}
2415 
2416 	skb->protocol = htons(ETH_P_MPLS_UC);
2417 
2418 	if (hdr_size) {
2419 		bool bos;
2420 		int i;
2421 
2422 		if (skb_cow(skb, hdr_size)) {
2423 			err = -ENOBUFS;
2424 			goto errout_free;
2425 		}
2426 
2427 		skb_reserve(skb, hdr_size);
2428 		skb_push(skb, hdr_size);
2429 		skb_reset_network_header(skb);
2430 
2431 		/* Push new labels */
2432 		hdr = mpls_hdr(skb);
2433 		bos = true;
2434 		for (i = n_labels - 1; i >= 0; i--) {
2435 			hdr[i] = mpls_entry_encode(labels[i],
2436 						   1, 0, bos);
2437 			bos = false;
2438 		}
2439 	}
2440 
2441 	nh = mpls_select_multipath(rt, skb);
2442 	if (!nh) {
2443 		err = -ENETUNREACH;
2444 		goto errout_free;
2445 	}
2446 
2447 	if (hdr_size) {
2448 		skb_pull(skb, hdr_size);
2449 		skb_reset_network_header(skb);
2450 	}
2451 
2452 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2453 			RTM_NEWROUTE, sizeof(*r), 0);
2454 	if (!nlh) {
2455 		err = -EMSGSIZE;
2456 		goto errout_free;
2457 	}
2458 
2459 	r = nlmsg_data(nlh);
2460 	r->rtm_family	 = AF_MPLS;
2461 	r->rtm_dst_len	= 20;
2462 	r->rtm_src_len	= 0;
2463 	r->rtm_table	= RT_TABLE_MAIN;
2464 	r->rtm_type	= RTN_UNICAST;
2465 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2466 	r->rtm_protocol = rt->rt_protocol;
2467 	r->rtm_flags	= 0;
2468 
2469 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2470 		goto nla_put_failure;
2471 
2472 	if (nh->nh_labels &&
2473 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2474 			   nh->nh_label))
2475 		goto nla_put_failure;
2476 
2477 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2478 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2479 			nh->nh_via_alen))
2480 		goto nla_put_failure;
2481 	dev = rtnl_dereference(nh->nh_dev);
2482 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2483 		goto nla_put_failure;
2484 
2485 	nlmsg_end(skb, nlh);
2486 
2487 	err = rtnl_unicast(skb, net, portid);
2488 errout:
2489 	return err;
2490 
2491 nla_put_failure:
2492 	nlmsg_cancel(skb, nlh);
2493 	err = -EMSGSIZE;
2494 errout_free:
2495 	kfree_skb(skb);
2496 	return err;
2497 }
2498 
2499 static int resize_platform_label_table(struct net *net, size_t limit)
2500 {
2501 	size_t size = sizeof(struct mpls_route *) * limit;
2502 	size_t old_limit;
2503 	size_t cp_size;
2504 	struct mpls_route __rcu **labels = NULL, **old;
2505 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2506 	unsigned index;
2507 
2508 	if (size) {
2509 		labels = kvzalloc(size, GFP_KERNEL);
2510 		if (!labels)
2511 			goto nolabels;
2512 	}
2513 
2514 	/* In case the predefined labels need to be populated */
2515 	if (limit > MPLS_LABEL_IPV4NULL) {
2516 		struct net_device *lo = net->loopback_dev;
2517 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2518 		if (IS_ERR(rt0))
2519 			goto nort0;
2520 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2521 		rt0->rt_protocol = RTPROT_KERNEL;
2522 		rt0->rt_payload_type = MPT_IPV4;
2523 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2524 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2525 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2526 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2527 		       lo->addr_len);
2528 	}
2529 	if (limit > MPLS_LABEL_IPV6NULL) {
2530 		struct net_device *lo = net->loopback_dev;
2531 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2532 		if (IS_ERR(rt2))
2533 			goto nort2;
2534 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2535 		rt2->rt_protocol = RTPROT_KERNEL;
2536 		rt2->rt_payload_type = MPT_IPV6;
2537 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2538 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2539 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2540 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2541 		       lo->addr_len);
2542 	}
2543 
2544 	rtnl_lock();
2545 	/* Remember the original table */
2546 	old = rtnl_dereference(net->mpls.platform_label);
2547 	old_limit = net->mpls.platform_labels;
2548 
2549 	/* Free any labels beyond the new table */
2550 	for (index = limit; index < old_limit; index++)
2551 		mpls_route_update(net, index, NULL, NULL);
2552 
2553 	/* Copy over the old labels */
2554 	cp_size = size;
2555 	if (old_limit < limit)
2556 		cp_size = old_limit * sizeof(struct mpls_route *);
2557 
2558 	memcpy(labels, old, cp_size);
2559 
2560 	/* If needed set the predefined labels */
2561 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2562 	    (limit > MPLS_LABEL_IPV6NULL)) {
2563 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2564 		rt2 = NULL;
2565 	}
2566 
2567 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2568 	    (limit > MPLS_LABEL_IPV4NULL)) {
2569 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2570 		rt0 = NULL;
2571 	}
2572 
2573 	/* Update the global pointers */
2574 	net->mpls.platform_labels = limit;
2575 	rcu_assign_pointer(net->mpls.platform_label, labels);
2576 
2577 	rtnl_unlock();
2578 
2579 	mpls_rt_free(rt2);
2580 	mpls_rt_free(rt0);
2581 
2582 	if (old) {
2583 		synchronize_rcu();
2584 		kvfree(old);
2585 	}
2586 	return 0;
2587 
2588 nort2:
2589 	mpls_rt_free(rt0);
2590 nort0:
2591 	kvfree(labels);
2592 nolabels:
2593 	return -ENOMEM;
2594 }
2595 
2596 static int mpls_platform_labels(struct ctl_table *table, int write,
2597 				void *buffer, size_t *lenp, loff_t *ppos)
2598 {
2599 	struct net *net = table->data;
2600 	int platform_labels = net->mpls.platform_labels;
2601 	int ret;
2602 	struct ctl_table tmp = {
2603 		.procname	= table->procname,
2604 		.data		= &platform_labels,
2605 		.maxlen		= sizeof(int),
2606 		.mode		= table->mode,
2607 		.extra1		= SYSCTL_ZERO,
2608 		.extra2		= &label_limit,
2609 	};
2610 
2611 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2612 
2613 	if (write && ret == 0)
2614 		ret = resize_platform_label_table(net, platform_labels);
2615 
2616 	return ret;
2617 }
2618 
2619 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2620 	(&((struct net *)0)->field)
2621 
2622 static const struct ctl_table mpls_table[] = {
2623 	{
2624 		.procname	= "platform_labels",
2625 		.data		= NULL,
2626 		.maxlen		= sizeof(int),
2627 		.mode		= 0644,
2628 		.proc_handler	= mpls_platform_labels,
2629 	},
2630 	{
2631 		.procname	= "ip_ttl_propagate",
2632 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2633 		.maxlen		= sizeof(int),
2634 		.mode		= 0644,
2635 		.proc_handler	= proc_dointvec_minmax,
2636 		.extra1		= SYSCTL_ZERO,
2637 		.extra2		= SYSCTL_ONE,
2638 	},
2639 	{
2640 		.procname	= "default_ttl",
2641 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2642 		.maxlen		= sizeof(int),
2643 		.mode		= 0644,
2644 		.proc_handler	= proc_dointvec_minmax,
2645 		.extra1		= SYSCTL_ONE,
2646 		.extra2		= &ttl_max,
2647 	},
2648 	{ }
2649 };
2650 
2651 static int mpls_net_init(struct net *net)
2652 {
2653 	struct ctl_table *table;
2654 	int i;
2655 
2656 	net->mpls.platform_labels = 0;
2657 	net->mpls.platform_label = NULL;
2658 	net->mpls.ip_ttl_propagate = 1;
2659 	net->mpls.default_ttl = 255;
2660 
2661 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2662 	if (table == NULL)
2663 		return -ENOMEM;
2664 
2665 	/* Table data contains only offsets relative to the base of
2666 	 * the mdev at this point, so make them absolute.
2667 	 */
2668 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2669 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2670 
2671 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2672 	if (net->mpls.ctl == NULL) {
2673 		kfree(table);
2674 		return -ENOMEM;
2675 	}
2676 
2677 	return 0;
2678 }
2679 
2680 static void mpls_net_exit(struct net *net)
2681 {
2682 	struct mpls_route __rcu **platform_label;
2683 	size_t platform_labels;
2684 	struct ctl_table *table;
2685 	unsigned int index;
2686 
2687 	table = net->mpls.ctl->ctl_table_arg;
2688 	unregister_net_sysctl_table(net->mpls.ctl);
2689 	kfree(table);
2690 
2691 	/* An rcu grace period has passed since there was a device in
2692 	 * the network namespace (and thus the last in flight packet)
2693 	 * left this network namespace.  This is because
2694 	 * unregister_netdevice_many and netdev_run_todo has completed
2695 	 * for each network device that was in this network namespace.
2696 	 *
2697 	 * As such no additional rcu synchronization is necessary when
2698 	 * freeing the platform_label table.
2699 	 */
2700 	rtnl_lock();
2701 	platform_label = rtnl_dereference(net->mpls.platform_label);
2702 	platform_labels = net->mpls.platform_labels;
2703 	for (index = 0; index < platform_labels; index++) {
2704 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2705 		RCU_INIT_POINTER(platform_label[index], NULL);
2706 		mpls_notify_route(net, index, rt, NULL, NULL);
2707 		mpls_rt_free(rt);
2708 	}
2709 	rtnl_unlock();
2710 
2711 	kvfree(platform_label);
2712 }
2713 
2714 static struct pernet_operations mpls_net_ops = {
2715 	.init = mpls_net_init,
2716 	.exit = mpls_net_exit,
2717 };
2718 
2719 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2720 	.family		   = AF_MPLS,
2721 	.fill_stats_af	   = mpls_fill_stats_af,
2722 	.get_stats_af_size = mpls_get_stats_af_size,
2723 };
2724 
2725 static int __init mpls_init(void)
2726 {
2727 	int err;
2728 
2729 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2730 
2731 	err = register_pernet_subsys(&mpls_net_ops);
2732 	if (err)
2733 		goto out;
2734 
2735 	err = register_netdevice_notifier(&mpls_dev_notifier);
2736 	if (err)
2737 		goto out_unregister_pernet;
2738 
2739 	dev_add_pack(&mpls_packet_type);
2740 
2741 	rtnl_af_register(&mpls_af_ops);
2742 
2743 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2744 			     mpls_rtm_newroute, NULL, 0);
2745 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2746 			     mpls_rtm_delroute, NULL, 0);
2747 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2748 			     mpls_getroute, mpls_dump_routes, 0);
2749 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2750 			     mpls_netconf_get_devconf,
2751 			     mpls_netconf_dump_devconf, 0);
2752 	err = ipgre_tunnel_encap_add_mpls_ops();
2753 	if (err)
2754 		pr_err("Can't add mpls over gre tunnel ops\n");
2755 
2756 	err = 0;
2757 out:
2758 	return err;
2759 
2760 out_unregister_pernet:
2761 	unregister_pernet_subsys(&mpls_net_ops);
2762 	goto out;
2763 }
2764 module_init(mpls_init);
2765 
2766 static void __exit mpls_exit(void)
2767 {
2768 	rtnl_unregister_all(PF_MPLS);
2769 	rtnl_af_unregister(&mpls_af_ops);
2770 	dev_remove_pack(&mpls_packet_type);
2771 	unregister_netdevice_notifier(&mpls_dev_notifier);
2772 	unregister_pernet_subsys(&mpls_net_ops);
2773 	ipgre_tunnel_encap_del_mpls_ops();
2774 }
2775 module_exit(mpls_exit);
2776 
2777 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2778 MODULE_LICENSE("GPL v2");
2779 MODULE_ALIAS_NETPROTO(PF_MPLS);
2780