xref: /openbmc/linux/net/mpls/af_mpls.c (revision 5a170e9e)
1 #include <linux/types.h>
2 #include <linux/skbuff.h>
3 #include <linux/socket.h>
4 #include <linux/sysctl.h>
5 #include <linux/net.h>
6 #include <linux/module.h>
7 #include <linux/if_arp.h>
8 #include <linux/ipv6.h>
9 #include <linux/mpls.h>
10 #include <linux/netconf.h>
11 #include <linux/nospec.h>
12 #include <linux/vmalloc.h>
13 #include <linux/percpu.h>
14 #include <net/ip.h>
15 #include <net/dst.h>
16 #include <net/sock.h>
17 #include <net/arp.h>
18 #include <net/ip_fib.h>
19 #include <net/netevent.h>
20 #include <net/ip_tunnels.h>
21 #include <net/netns/generic.h>
22 #if IS_ENABLED(CONFIG_IPV6)
23 #include <net/ipv6.h>
24 #endif
25 #include <net/addrconf.h>
26 #include <net/nexthop.h>
27 #include "internal.h"
28 
29 /* max memory we will use for mpls_route */
30 #define MAX_MPLS_ROUTE_MEM	4096
31 
32 /* Maximum number of labels to look ahead at when selecting a path of
33  * a multipath route
34  */
35 #define MAX_MP_SELECT_LABELS 4
36 
37 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
38 
39 static int zero = 0;
40 static int one = 1;
41 static int label_limit = (1 << 20) - 1;
42 static int ttl_max = 255;
43 
44 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
45 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
46 {
47 	return sizeof(struct mpls_shim_hdr);
48 }
49 
50 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
51 	.encap_hlen	= ipgre_mpls_encap_hlen,
52 };
53 
54 static int ipgre_tunnel_encap_add_mpls_ops(void)
55 {
56 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
57 }
58 
59 static void ipgre_tunnel_encap_del_mpls_ops(void)
60 {
61 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
62 }
63 #else
64 static int ipgre_tunnel_encap_add_mpls_ops(void)
65 {
66 	return 0;
67 }
68 
69 static void ipgre_tunnel_encap_del_mpls_ops(void)
70 {
71 }
72 #endif
73 
74 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
75 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
76 		       unsigned int nlm_flags);
77 
78 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
79 {
80 	struct mpls_route *rt = NULL;
81 
82 	if (index < net->mpls.platform_labels) {
83 		struct mpls_route __rcu **platform_label =
84 			rcu_dereference(net->mpls.platform_label);
85 		rt = rcu_dereference(platform_label[index]);
86 	}
87 	return rt;
88 }
89 
90 bool mpls_output_possible(const struct net_device *dev)
91 {
92 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
93 }
94 EXPORT_SYMBOL_GPL(mpls_output_possible);
95 
96 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
97 {
98 	return (u8 *)nh + rt->rt_via_offset;
99 }
100 
101 static const u8 *mpls_nh_via(const struct mpls_route *rt,
102 			     const struct mpls_nh *nh)
103 {
104 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
105 }
106 
107 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
108 {
109 	/* The size of the layer 2.5 labels to be added for this route */
110 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
111 }
112 
113 unsigned int mpls_dev_mtu(const struct net_device *dev)
114 {
115 	/* The amount of data the layer 2 frame can hold */
116 	return dev->mtu;
117 }
118 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
119 
120 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
121 {
122 	if (skb->len <= mtu)
123 		return false;
124 
125 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
126 		return false;
127 
128 	return true;
129 }
130 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
131 
132 void mpls_stats_inc_outucastpkts(struct net_device *dev,
133 				 const struct sk_buff *skb)
134 {
135 	struct mpls_dev *mdev;
136 
137 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
138 		mdev = mpls_dev_get(dev);
139 		if (mdev)
140 			MPLS_INC_STATS_LEN(mdev, skb->len,
141 					   tx_packets,
142 					   tx_bytes);
143 	} else if (skb->protocol == htons(ETH_P_IP)) {
144 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
145 #if IS_ENABLED(CONFIG_IPV6)
146 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
147 		struct inet6_dev *in6dev = __in6_dev_get(dev);
148 
149 		if (in6dev)
150 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
151 					 IPSTATS_MIB_OUT, skb->len);
152 #endif
153 	}
154 }
155 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
156 
157 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
158 {
159 	struct mpls_entry_decoded dec;
160 	unsigned int mpls_hdr_len = 0;
161 	struct mpls_shim_hdr *hdr;
162 	bool eli_seen = false;
163 	int label_index;
164 	u32 hash = 0;
165 
166 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
167 	     label_index++) {
168 		mpls_hdr_len += sizeof(*hdr);
169 		if (!pskb_may_pull(skb, mpls_hdr_len))
170 			break;
171 
172 		/* Read and decode the current label */
173 		hdr = mpls_hdr(skb) + label_index;
174 		dec = mpls_entry_decode(hdr);
175 
176 		/* RFC6790 - reserved labels MUST NOT be used as keys
177 		 * for the load-balancing function
178 		 */
179 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
180 			hash = jhash_1word(dec.label, hash);
181 
182 			/* The entropy label follows the entropy label
183 			 * indicator, so this means that the entropy
184 			 * label was just added to the hash - no need to
185 			 * go any deeper either in the label stack or in the
186 			 * payload
187 			 */
188 			if (eli_seen)
189 				break;
190 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
191 			eli_seen = true;
192 		}
193 
194 		if (!dec.bos)
195 			continue;
196 
197 		/* found bottom label; does skb have room for a header? */
198 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
199 			const struct iphdr *v4hdr;
200 
201 			v4hdr = (const struct iphdr *)(hdr + 1);
202 			if (v4hdr->version == 4) {
203 				hash = jhash_3words(ntohl(v4hdr->saddr),
204 						    ntohl(v4hdr->daddr),
205 						    v4hdr->protocol, hash);
206 			} else if (v4hdr->version == 6 &&
207 				   pskb_may_pull(skb, mpls_hdr_len +
208 						 sizeof(struct ipv6hdr))) {
209 				const struct ipv6hdr *v6hdr;
210 
211 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
212 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
213 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
214 				hash = jhash_1word(v6hdr->nexthdr, hash);
215 			}
216 		}
217 
218 		break;
219 	}
220 
221 	return hash;
222 }
223 
224 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
225 {
226 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
227 }
228 
229 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
230  * a next hop (nh->nh_flags) are modified by netdev event handlers.
231  * Since those fields can change at any moment, use READ_ONCE to
232  * access both.
233  */
234 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
235 					     struct sk_buff *skb)
236 {
237 	u32 hash = 0;
238 	int nh_index = 0;
239 	int n = 0;
240 	u8 alive;
241 
242 	/* No need to look further into packet if there's only
243 	 * one path
244 	 */
245 	if (rt->rt_nhn == 1)
246 		return rt->rt_nh;
247 
248 	alive = READ_ONCE(rt->rt_nhn_alive);
249 	if (alive == 0)
250 		return NULL;
251 
252 	hash = mpls_multipath_hash(rt, skb);
253 	nh_index = hash % alive;
254 	if (alive == rt->rt_nhn)
255 		goto out;
256 	for_nexthops(rt) {
257 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
258 
259 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
260 			continue;
261 		if (n == nh_index)
262 			return nh;
263 		n++;
264 	} endfor_nexthops(rt);
265 
266 out:
267 	return mpls_get_nexthop(rt, nh_index);
268 }
269 
270 static bool mpls_egress(struct net *net, struct mpls_route *rt,
271 			struct sk_buff *skb, struct mpls_entry_decoded dec)
272 {
273 	enum mpls_payload_type payload_type;
274 	bool success = false;
275 
276 	/* The IPv4 code below accesses through the IPv4 header
277 	 * checksum, which is 12 bytes into the packet.
278 	 * The IPv6 code below accesses through the IPv6 hop limit
279 	 * which is 8 bytes into the packet.
280 	 *
281 	 * For all supported cases there should always be at least 12
282 	 * bytes of packet data present.  The IPv4 header is 20 bytes
283 	 * without options and the IPv6 header is always 40 bytes
284 	 * long.
285 	 */
286 	if (!pskb_may_pull(skb, 12))
287 		return false;
288 
289 	payload_type = rt->rt_payload_type;
290 	if (payload_type == MPT_UNSPEC)
291 		payload_type = ip_hdr(skb)->version;
292 
293 	switch (payload_type) {
294 	case MPT_IPV4: {
295 		struct iphdr *hdr4 = ip_hdr(skb);
296 		u8 new_ttl;
297 		skb->protocol = htons(ETH_P_IP);
298 
299 		/* If propagating TTL, take the decremented TTL from
300 		 * the incoming MPLS header, otherwise decrement the
301 		 * TTL, but only if not 0 to avoid underflow.
302 		 */
303 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
304 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
305 		     net->mpls.ip_ttl_propagate))
306 			new_ttl = dec.ttl;
307 		else
308 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
309 
310 		csum_replace2(&hdr4->check,
311 			      htons(hdr4->ttl << 8),
312 			      htons(new_ttl << 8));
313 		hdr4->ttl = new_ttl;
314 		success = true;
315 		break;
316 	}
317 	case MPT_IPV6: {
318 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
319 		skb->protocol = htons(ETH_P_IPV6);
320 
321 		/* If propagating TTL, take the decremented TTL from
322 		 * the incoming MPLS header, otherwise decrement the
323 		 * hop limit, but only if not 0 to avoid underflow.
324 		 */
325 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
326 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
327 		     net->mpls.ip_ttl_propagate))
328 			hdr6->hop_limit = dec.ttl;
329 		else if (hdr6->hop_limit)
330 			hdr6->hop_limit = hdr6->hop_limit - 1;
331 		success = true;
332 		break;
333 	}
334 	case MPT_UNSPEC:
335 		/* Should have decided which protocol it is by now */
336 		break;
337 	}
338 
339 	return success;
340 }
341 
342 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
343 			struct packet_type *pt, struct net_device *orig_dev)
344 {
345 	struct net *net = dev_net(dev);
346 	struct mpls_shim_hdr *hdr;
347 	struct mpls_route *rt;
348 	struct mpls_nh *nh;
349 	struct mpls_entry_decoded dec;
350 	struct net_device *out_dev;
351 	struct mpls_dev *out_mdev;
352 	struct mpls_dev *mdev;
353 	unsigned int hh_len;
354 	unsigned int new_header_size;
355 	unsigned int mtu;
356 	int err;
357 
358 	/* Careful this entire function runs inside of an rcu critical section */
359 
360 	mdev = mpls_dev_get(dev);
361 	if (!mdev)
362 		goto drop;
363 
364 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
365 			   rx_bytes);
366 
367 	if (!mdev->input_enabled) {
368 		MPLS_INC_STATS(mdev, rx_dropped);
369 		goto drop;
370 	}
371 
372 	if (skb->pkt_type != PACKET_HOST)
373 		goto err;
374 
375 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
376 		goto err;
377 
378 	if (!pskb_may_pull(skb, sizeof(*hdr)))
379 		goto err;
380 
381 	/* Read and decode the label */
382 	hdr = mpls_hdr(skb);
383 	dec = mpls_entry_decode(hdr);
384 
385 	rt = mpls_route_input_rcu(net, dec.label);
386 	if (!rt) {
387 		MPLS_INC_STATS(mdev, rx_noroute);
388 		goto drop;
389 	}
390 
391 	nh = mpls_select_multipath(rt, skb);
392 	if (!nh)
393 		goto err;
394 
395 	/* Pop the label */
396 	skb_pull(skb, sizeof(*hdr));
397 	skb_reset_network_header(skb);
398 
399 	skb_orphan(skb);
400 
401 	if (skb_warn_if_lro(skb))
402 		goto err;
403 
404 	skb_forward_csum(skb);
405 
406 	/* Verify ttl is valid */
407 	if (dec.ttl <= 1)
408 		goto err;
409 	dec.ttl -= 1;
410 
411 	/* Find the output device */
412 	out_dev = rcu_dereference(nh->nh_dev);
413 	if (!mpls_output_possible(out_dev))
414 		goto tx_err;
415 
416 	/* Verify the destination can hold the packet */
417 	new_header_size = mpls_nh_header_size(nh);
418 	mtu = mpls_dev_mtu(out_dev);
419 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
420 		goto tx_err;
421 
422 	hh_len = LL_RESERVED_SPACE(out_dev);
423 	if (!out_dev->header_ops)
424 		hh_len = 0;
425 
426 	/* Ensure there is enough space for the headers in the skb */
427 	if (skb_cow(skb, hh_len + new_header_size))
428 		goto tx_err;
429 
430 	skb->dev = out_dev;
431 	skb->protocol = htons(ETH_P_MPLS_UC);
432 
433 	if (unlikely(!new_header_size && dec.bos)) {
434 		/* Penultimate hop popping */
435 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
436 			goto err;
437 	} else {
438 		bool bos;
439 		int i;
440 		skb_push(skb, new_header_size);
441 		skb_reset_network_header(skb);
442 		/* Push the new labels */
443 		hdr = mpls_hdr(skb);
444 		bos = dec.bos;
445 		for (i = nh->nh_labels - 1; i >= 0; i--) {
446 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
447 						   dec.ttl, 0, bos);
448 			bos = false;
449 		}
450 	}
451 
452 	mpls_stats_inc_outucastpkts(out_dev, skb);
453 
454 	/* If via wasn't specified then send out using device address */
455 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
456 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
457 				 out_dev->dev_addr, skb);
458 	else
459 		err = neigh_xmit(nh->nh_via_table, out_dev,
460 				 mpls_nh_via(rt, nh), skb);
461 	if (err)
462 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
463 				    __func__, err);
464 	return 0;
465 
466 tx_err:
467 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
468 	if (out_mdev)
469 		MPLS_INC_STATS(out_mdev, tx_errors);
470 	goto drop;
471 err:
472 	MPLS_INC_STATS(mdev, rx_errors);
473 drop:
474 	kfree_skb(skb);
475 	return NET_RX_DROP;
476 }
477 
478 static struct packet_type mpls_packet_type __read_mostly = {
479 	.type = cpu_to_be16(ETH_P_MPLS_UC),
480 	.func = mpls_forward,
481 };
482 
483 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
484 	[RTA_DST]		= { .type = NLA_U32 },
485 	[RTA_OIF]		= { .type = NLA_U32 },
486 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
487 };
488 
489 struct mpls_route_config {
490 	u32			rc_protocol;
491 	u32			rc_ifindex;
492 	u8			rc_via_table;
493 	u8			rc_via_alen;
494 	u8			rc_via[MAX_VIA_ALEN];
495 	u32			rc_label;
496 	u8			rc_ttl_propagate;
497 	u8			rc_output_labels;
498 	u32			rc_output_label[MAX_NEW_LABELS];
499 	u32			rc_nlflags;
500 	enum mpls_payload_type	rc_payload_type;
501 	struct nl_info		rc_nlinfo;
502 	struct rtnexthop	*rc_mp;
503 	int			rc_mp_len;
504 };
505 
506 /* all nexthops within a route have the same size based on max
507  * number of labels and max via length for a hop
508  */
509 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
510 {
511 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
512 	struct mpls_route *rt;
513 	size_t size;
514 
515 	size = sizeof(*rt) + num_nh * nh_size;
516 	if (size > MAX_MPLS_ROUTE_MEM)
517 		return ERR_PTR(-EINVAL);
518 
519 	rt = kzalloc(size, GFP_KERNEL);
520 	if (!rt)
521 		return ERR_PTR(-ENOMEM);
522 
523 	rt->rt_nhn = num_nh;
524 	rt->rt_nhn_alive = num_nh;
525 	rt->rt_nh_size = nh_size;
526 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
527 
528 	return rt;
529 }
530 
531 static void mpls_rt_free(struct mpls_route *rt)
532 {
533 	if (rt)
534 		kfree_rcu(rt, rt_rcu);
535 }
536 
537 static void mpls_notify_route(struct net *net, unsigned index,
538 			      struct mpls_route *old, struct mpls_route *new,
539 			      const struct nl_info *info)
540 {
541 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
542 	unsigned portid = info ? info->portid : 0;
543 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
544 	struct mpls_route *rt = new ? new : old;
545 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
546 	/* Ignore reserved labels for now */
547 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
548 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
549 }
550 
551 static void mpls_route_update(struct net *net, unsigned index,
552 			      struct mpls_route *new,
553 			      const struct nl_info *info)
554 {
555 	struct mpls_route __rcu **platform_label;
556 	struct mpls_route *rt;
557 
558 	ASSERT_RTNL();
559 
560 	platform_label = rtnl_dereference(net->mpls.platform_label);
561 	rt = rtnl_dereference(platform_label[index]);
562 	rcu_assign_pointer(platform_label[index], new);
563 
564 	mpls_notify_route(net, index, rt, new, info);
565 
566 	/* If we removed a route free it now */
567 	mpls_rt_free(rt);
568 }
569 
570 static unsigned find_free_label(struct net *net)
571 {
572 	struct mpls_route __rcu **platform_label;
573 	size_t platform_labels;
574 	unsigned index;
575 
576 	platform_label = rtnl_dereference(net->mpls.platform_label);
577 	platform_labels = net->mpls.platform_labels;
578 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
579 	     index++) {
580 		if (!rtnl_dereference(platform_label[index]))
581 			return index;
582 	}
583 	return LABEL_NOT_SPECIFIED;
584 }
585 
586 #if IS_ENABLED(CONFIG_INET)
587 static struct net_device *inet_fib_lookup_dev(struct net *net,
588 					      const void *addr)
589 {
590 	struct net_device *dev;
591 	struct rtable *rt;
592 	struct in_addr daddr;
593 
594 	memcpy(&daddr, addr, sizeof(struct in_addr));
595 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
596 	if (IS_ERR(rt))
597 		return ERR_CAST(rt);
598 
599 	dev = rt->dst.dev;
600 	dev_hold(dev);
601 
602 	ip_rt_put(rt);
603 
604 	return dev;
605 }
606 #else
607 static struct net_device *inet_fib_lookup_dev(struct net *net,
608 					      const void *addr)
609 {
610 	return ERR_PTR(-EAFNOSUPPORT);
611 }
612 #endif
613 
614 #if IS_ENABLED(CONFIG_IPV6)
615 static struct net_device *inet6_fib_lookup_dev(struct net *net,
616 					       const void *addr)
617 {
618 	struct net_device *dev;
619 	struct dst_entry *dst;
620 	struct flowi6 fl6;
621 	int err;
622 
623 	if (!ipv6_stub)
624 		return ERR_PTR(-EAFNOSUPPORT);
625 
626 	memset(&fl6, 0, sizeof(fl6));
627 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
628 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
629 	if (err)
630 		return ERR_PTR(err);
631 
632 	dev = dst->dev;
633 	dev_hold(dev);
634 	dst_release(dst);
635 
636 	return dev;
637 }
638 #else
639 static struct net_device *inet6_fib_lookup_dev(struct net *net,
640 					       const void *addr)
641 {
642 	return ERR_PTR(-EAFNOSUPPORT);
643 }
644 #endif
645 
646 static struct net_device *find_outdev(struct net *net,
647 				      struct mpls_route *rt,
648 				      struct mpls_nh *nh, int oif)
649 {
650 	struct net_device *dev = NULL;
651 
652 	if (!oif) {
653 		switch (nh->nh_via_table) {
654 		case NEIGH_ARP_TABLE:
655 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
656 			break;
657 		case NEIGH_ND_TABLE:
658 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
659 			break;
660 		case NEIGH_LINK_TABLE:
661 			break;
662 		}
663 	} else {
664 		dev = dev_get_by_index(net, oif);
665 	}
666 
667 	if (!dev)
668 		return ERR_PTR(-ENODEV);
669 
670 	if (IS_ERR(dev))
671 		return dev;
672 
673 	/* The caller is holding rtnl anyways, so release the dev reference */
674 	dev_put(dev);
675 
676 	return dev;
677 }
678 
679 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
680 			      struct mpls_nh *nh, int oif)
681 {
682 	struct net_device *dev = NULL;
683 	int err = -ENODEV;
684 
685 	dev = find_outdev(net, rt, nh, oif);
686 	if (IS_ERR(dev)) {
687 		err = PTR_ERR(dev);
688 		dev = NULL;
689 		goto errout;
690 	}
691 
692 	/* Ensure this is a supported device */
693 	err = -EINVAL;
694 	if (!mpls_dev_get(dev))
695 		goto errout;
696 
697 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
698 	    (dev->addr_len != nh->nh_via_alen))
699 		goto errout;
700 
701 	RCU_INIT_POINTER(nh->nh_dev, dev);
702 
703 	if (!(dev->flags & IFF_UP)) {
704 		nh->nh_flags |= RTNH_F_DEAD;
705 	} else {
706 		unsigned int flags;
707 
708 		flags = dev_get_flags(dev);
709 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
710 			nh->nh_flags |= RTNH_F_LINKDOWN;
711 	}
712 
713 	return 0;
714 
715 errout:
716 	return err;
717 }
718 
719 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
720 		       u8 via_addr[], struct netlink_ext_ack *extack)
721 {
722 	struct rtvia *via = nla_data(nla);
723 	int err = -EINVAL;
724 	int alen;
725 
726 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
727 		NL_SET_ERR_MSG_ATTR(extack, nla,
728 				    "Invalid attribute length for RTA_VIA");
729 		goto errout;
730 	}
731 	alen = nla_len(nla) -
732 			offsetof(struct rtvia, rtvia_addr);
733 	if (alen > MAX_VIA_ALEN) {
734 		NL_SET_ERR_MSG_ATTR(extack, nla,
735 				    "Invalid address length for RTA_VIA");
736 		goto errout;
737 	}
738 
739 	/* Validate the address family */
740 	switch (via->rtvia_family) {
741 	case AF_PACKET:
742 		*via_table = NEIGH_LINK_TABLE;
743 		break;
744 	case AF_INET:
745 		*via_table = NEIGH_ARP_TABLE;
746 		if (alen != 4)
747 			goto errout;
748 		break;
749 	case AF_INET6:
750 		*via_table = NEIGH_ND_TABLE;
751 		if (alen != 16)
752 			goto errout;
753 		break;
754 	default:
755 		/* Unsupported address family */
756 		goto errout;
757 	}
758 
759 	memcpy(via_addr, via->rtvia_addr, alen);
760 	*via_alen = alen;
761 	err = 0;
762 
763 errout:
764 	return err;
765 }
766 
767 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
768 				  struct mpls_route *rt)
769 {
770 	struct net *net = cfg->rc_nlinfo.nl_net;
771 	struct mpls_nh *nh = rt->rt_nh;
772 	int err;
773 	int i;
774 
775 	if (!nh)
776 		return -ENOMEM;
777 
778 	nh->nh_labels = cfg->rc_output_labels;
779 	for (i = 0; i < nh->nh_labels; i++)
780 		nh->nh_label[i] = cfg->rc_output_label[i];
781 
782 	nh->nh_via_table = cfg->rc_via_table;
783 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
784 	nh->nh_via_alen = cfg->rc_via_alen;
785 
786 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
787 	if (err)
788 		goto errout;
789 
790 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
791 		rt->rt_nhn_alive--;
792 
793 	return 0;
794 
795 errout:
796 	return err;
797 }
798 
799 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
800 			 struct mpls_nh *nh, int oif, struct nlattr *via,
801 			 struct nlattr *newdst, u8 max_labels,
802 			 struct netlink_ext_ack *extack)
803 {
804 	int err = -ENOMEM;
805 
806 	if (!nh)
807 		goto errout;
808 
809 	if (newdst) {
810 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
811 				     nh->nh_label, extack);
812 		if (err)
813 			goto errout;
814 	}
815 
816 	if (via) {
817 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
818 				  __mpls_nh_via(rt, nh), extack);
819 		if (err)
820 			goto errout;
821 	} else {
822 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
823 	}
824 
825 	err = mpls_nh_assign_dev(net, rt, nh, oif);
826 	if (err)
827 		goto errout;
828 
829 	return 0;
830 
831 errout:
832 	return err;
833 }
834 
835 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
836 			      u8 cfg_via_alen, u8 *max_via_alen,
837 			      u8 *max_labels)
838 {
839 	int remaining = len;
840 	u8 nhs = 0;
841 
842 	*max_via_alen = 0;
843 	*max_labels = 0;
844 
845 	while (rtnh_ok(rtnh, remaining)) {
846 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
847 		int attrlen;
848 		u8 n_labels = 0;
849 
850 		attrlen = rtnh_attrlen(rtnh);
851 		nla = nla_find(attrs, attrlen, RTA_VIA);
852 		if (nla && nla_len(nla) >=
853 		    offsetof(struct rtvia, rtvia_addr)) {
854 			int via_alen = nla_len(nla) -
855 				offsetof(struct rtvia, rtvia_addr);
856 
857 			if (via_alen <= MAX_VIA_ALEN)
858 				*max_via_alen = max_t(u16, *max_via_alen,
859 						      via_alen);
860 		}
861 
862 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
863 		if (nla &&
864 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
865 				   NULL, NULL) != 0)
866 			return 0;
867 
868 		*max_labels = max_t(u8, *max_labels, n_labels);
869 
870 		/* number of nexthops is tracked by a u8.
871 		 * Check for overflow.
872 		 */
873 		if (nhs == 255)
874 			return 0;
875 		nhs++;
876 
877 		rtnh = rtnh_next(rtnh, &remaining);
878 	}
879 
880 	/* leftover implies invalid nexthop configuration, discard it */
881 	return remaining > 0 ? 0 : nhs;
882 }
883 
884 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
885 			       struct mpls_route *rt, u8 max_labels,
886 			       struct netlink_ext_ack *extack)
887 {
888 	struct rtnexthop *rtnh = cfg->rc_mp;
889 	struct nlattr *nla_via, *nla_newdst;
890 	int remaining = cfg->rc_mp_len;
891 	int err = 0;
892 	u8 nhs = 0;
893 
894 	change_nexthops(rt) {
895 		int attrlen;
896 
897 		nla_via = NULL;
898 		nla_newdst = NULL;
899 
900 		err = -EINVAL;
901 		if (!rtnh_ok(rtnh, remaining))
902 			goto errout;
903 
904 		/* neither weighted multipath nor any flags
905 		 * are supported
906 		 */
907 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
908 			goto errout;
909 
910 		attrlen = rtnh_attrlen(rtnh);
911 		if (attrlen > 0) {
912 			struct nlattr *attrs = rtnh_attrs(rtnh);
913 
914 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
915 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
916 		}
917 
918 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
919 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
920 				    max_labels, extack);
921 		if (err)
922 			goto errout;
923 
924 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
925 			rt->rt_nhn_alive--;
926 
927 		rtnh = rtnh_next(rtnh, &remaining);
928 		nhs++;
929 	} endfor_nexthops(rt);
930 
931 	rt->rt_nhn = nhs;
932 
933 	return 0;
934 
935 errout:
936 	return err;
937 }
938 
939 static bool mpls_label_ok(struct net *net, unsigned int *index,
940 			  struct netlink_ext_ack *extack)
941 {
942 	bool is_ok = true;
943 
944 	/* Reserved labels may not be set */
945 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
946 		NL_SET_ERR_MSG(extack,
947 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
948 		is_ok = false;
949 	}
950 
951 	/* The full 20 bit range may not be supported. */
952 	if (is_ok && *index >= net->mpls.platform_labels) {
953 		NL_SET_ERR_MSG(extack,
954 			       "Label >= configured maximum in platform_labels");
955 		is_ok = false;
956 	}
957 
958 	*index = array_index_nospec(*index, net->mpls.platform_labels);
959 	return is_ok;
960 }
961 
962 static int mpls_route_add(struct mpls_route_config *cfg,
963 			  struct netlink_ext_ack *extack)
964 {
965 	struct mpls_route __rcu **platform_label;
966 	struct net *net = cfg->rc_nlinfo.nl_net;
967 	struct mpls_route *rt, *old;
968 	int err = -EINVAL;
969 	u8 max_via_alen;
970 	unsigned index;
971 	u8 max_labels;
972 	u8 nhs;
973 
974 	index = cfg->rc_label;
975 
976 	/* If a label was not specified during insert pick one */
977 	if ((index == LABEL_NOT_SPECIFIED) &&
978 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
979 		index = find_free_label(net);
980 	}
981 
982 	if (!mpls_label_ok(net, &index, extack))
983 		goto errout;
984 
985 	/* Append makes no sense with mpls */
986 	err = -EOPNOTSUPP;
987 	if (cfg->rc_nlflags & NLM_F_APPEND) {
988 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
989 		goto errout;
990 	}
991 
992 	err = -EEXIST;
993 	platform_label = rtnl_dereference(net->mpls.platform_label);
994 	old = rtnl_dereference(platform_label[index]);
995 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
996 		goto errout;
997 
998 	err = -EEXIST;
999 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1000 		goto errout;
1001 
1002 	err = -ENOENT;
1003 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1004 		goto errout;
1005 
1006 	err = -EINVAL;
1007 	if (cfg->rc_mp) {
1008 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1009 					  cfg->rc_via_alen, &max_via_alen,
1010 					  &max_labels);
1011 	} else {
1012 		max_via_alen = cfg->rc_via_alen;
1013 		max_labels = cfg->rc_output_labels;
1014 		nhs = 1;
1015 	}
1016 
1017 	if (nhs == 0) {
1018 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1019 		goto errout;
1020 	}
1021 
1022 	err = -ENOMEM;
1023 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1024 	if (IS_ERR(rt)) {
1025 		err = PTR_ERR(rt);
1026 		goto errout;
1027 	}
1028 
1029 	rt->rt_protocol = cfg->rc_protocol;
1030 	rt->rt_payload_type = cfg->rc_payload_type;
1031 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1032 
1033 	if (cfg->rc_mp)
1034 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1035 	else
1036 		err = mpls_nh_build_from_cfg(cfg, rt);
1037 	if (err)
1038 		goto freert;
1039 
1040 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1041 
1042 	return 0;
1043 
1044 freert:
1045 	mpls_rt_free(rt);
1046 errout:
1047 	return err;
1048 }
1049 
1050 static int mpls_route_del(struct mpls_route_config *cfg,
1051 			  struct netlink_ext_ack *extack)
1052 {
1053 	struct net *net = cfg->rc_nlinfo.nl_net;
1054 	unsigned index;
1055 	int err = -EINVAL;
1056 
1057 	index = cfg->rc_label;
1058 
1059 	if (!mpls_label_ok(net, &index, extack))
1060 		goto errout;
1061 
1062 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1063 
1064 	err = 0;
1065 errout:
1066 	return err;
1067 }
1068 
1069 static void mpls_get_stats(struct mpls_dev *mdev,
1070 			   struct mpls_link_stats *stats)
1071 {
1072 	struct mpls_pcpu_stats *p;
1073 	int i;
1074 
1075 	memset(stats, 0, sizeof(*stats));
1076 
1077 	for_each_possible_cpu(i) {
1078 		struct mpls_link_stats local;
1079 		unsigned int start;
1080 
1081 		p = per_cpu_ptr(mdev->stats, i);
1082 		do {
1083 			start = u64_stats_fetch_begin(&p->syncp);
1084 			local = p->stats;
1085 		} while (u64_stats_fetch_retry(&p->syncp, start));
1086 
1087 		stats->rx_packets	+= local.rx_packets;
1088 		stats->rx_bytes		+= local.rx_bytes;
1089 		stats->tx_packets	+= local.tx_packets;
1090 		stats->tx_bytes		+= local.tx_bytes;
1091 		stats->rx_errors	+= local.rx_errors;
1092 		stats->tx_errors	+= local.tx_errors;
1093 		stats->rx_dropped	+= local.rx_dropped;
1094 		stats->tx_dropped	+= local.tx_dropped;
1095 		stats->rx_noroute	+= local.rx_noroute;
1096 	}
1097 }
1098 
1099 static int mpls_fill_stats_af(struct sk_buff *skb,
1100 			      const struct net_device *dev)
1101 {
1102 	struct mpls_link_stats *stats;
1103 	struct mpls_dev *mdev;
1104 	struct nlattr *nla;
1105 
1106 	mdev = mpls_dev_get(dev);
1107 	if (!mdev)
1108 		return -ENODATA;
1109 
1110 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1111 				sizeof(struct mpls_link_stats),
1112 				MPLS_STATS_UNSPEC);
1113 	if (!nla)
1114 		return -EMSGSIZE;
1115 
1116 	stats = nla_data(nla);
1117 	mpls_get_stats(mdev, stats);
1118 
1119 	return 0;
1120 }
1121 
1122 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1123 {
1124 	struct mpls_dev *mdev;
1125 
1126 	mdev = mpls_dev_get(dev);
1127 	if (!mdev)
1128 		return 0;
1129 
1130 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1131 }
1132 
1133 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1134 				     u32 portid, u32 seq, int event,
1135 				     unsigned int flags, int type)
1136 {
1137 	struct nlmsghdr  *nlh;
1138 	struct netconfmsg *ncm;
1139 	bool all = false;
1140 
1141 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1142 			flags);
1143 	if (!nlh)
1144 		return -EMSGSIZE;
1145 
1146 	if (type == NETCONFA_ALL)
1147 		all = true;
1148 
1149 	ncm = nlmsg_data(nlh);
1150 	ncm->ncm_family = AF_MPLS;
1151 
1152 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1153 		goto nla_put_failure;
1154 
1155 	if ((all || type == NETCONFA_INPUT) &&
1156 	    nla_put_s32(skb, NETCONFA_INPUT,
1157 			mdev->input_enabled) < 0)
1158 		goto nla_put_failure;
1159 
1160 	nlmsg_end(skb, nlh);
1161 	return 0;
1162 
1163 nla_put_failure:
1164 	nlmsg_cancel(skb, nlh);
1165 	return -EMSGSIZE;
1166 }
1167 
1168 static int mpls_netconf_msgsize_devconf(int type)
1169 {
1170 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1171 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1172 	bool all = false;
1173 
1174 	if (type == NETCONFA_ALL)
1175 		all = true;
1176 
1177 	if (all || type == NETCONFA_INPUT)
1178 		size += nla_total_size(4);
1179 
1180 	return size;
1181 }
1182 
1183 static void mpls_netconf_notify_devconf(struct net *net, int event,
1184 					int type, struct mpls_dev *mdev)
1185 {
1186 	struct sk_buff *skb;
1187 	int err = -ENOBUFS;
1188 
1189 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1190 	if (!skb)
1191 		goto errout;
1192 
1193 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1194 	if (err < 0) {
1195 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1196 		WARN_ON(err == -EMSGSIZE);
1197 		kfree_skb(skb);
1198 		goto errout;
1199 	}
1200 
1201 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1202 	return;
1203 errout:
1204 	if (err < 0)
1205 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1206 }
1207 
1208 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1209 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1210 };
1211 
1212 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1213 				    struct nlmsghdr *nlh,
1214 				    struct netlink_ext_ack *extack)
1215 {
1216 	struct net *net = sock_net(in_skb->sk);
1217 	struct nlattr *tb[NETCONFA_MAX + 1];
1218 	struct netconfmsg *ncm;
1219 	struct net_device *dev;
1220 	struct mpls_dev *mdev;
1221 	struct sk_buff *skb;
1222 	int ifindex;
1223 	int err;
1224 
1225 	err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1226 			  devconf_mpls_policy, extack);
1227 	if (err < 0)
1228 		goto errout;
1229 
1230 	err = -EINVAL;
1231 	if (!tb[NETCONFA_IFINDEX])
1232 		goto errout;
1233 
1234 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1235 	dev = __dev_get_by_index(net, ifindex);
1236 	if (!dev)
1237 		goto errout;
1238 
1239 	mdev = mpls_dev_get(dev);
1240 	if (!mdev)
1241 		goto errout;
1242 
1243 	err = -ENOBUFS;
1244 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1245 	if (!skb)
1246 		goto errout;
1247 
1248 	err = mpls_netconf_fill_devconf(skb, mdev,
1249 					NETLINK_CB(in_skb).portid,
1250 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1251 					NETCONFA_ALL);
1252 	if (err < 0) {
1253 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1254 		WARN_ON(err == -EMSGSIZE);
1255 		kfree_skb(skb);
1256 		goto errout;
1257 	}
1258 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1259 errout:
1260 	return err;
1261 }
1262 
1263 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1264 				     struct netlink_callback *cb)
1265 {
1266 	const struct nlmsghdr *nlh = cb->nlh;
1267 	struct net *net = sock_net(skb->sk);
1268 	struct hlist_head *head;
1269 	struct net_device *dev;
1270 	struct mpls_dev *mdev;
1271 	int idx, s_idx;
1272 	int h, s_h;
1273 
1274 	if (cb->strict_check) {
1275 		struct netlink_ext_ack *extack = cb->extack;
1276 		struct netconfmsg *ncm;
1277 
1278 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1279 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1280 			return -EINVAL;
1281 		}
1282 
1283 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1284 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1285 			return -EINVAL;
1286 		}
1287 	}
1288 
1289 	s_h = cb->args[0];
1290 	s_idx = idx = cb->args[1];
1291 
1292 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1293 		idx = 0;
1294 		head = &net->dev_index_head[h];
1295 		rcu_read_lock();
1296 		cb->seq = net->dev_base_seq;
1297 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1298 			if (idx < s_idx)
1299 				goto cont;
1300 			mdev = mpls_dev_get(dev);
1301 			if (!mdev)
1302 				goto cont;
1303 			if (mpls_netconf_fill_devconf(skb, mdev,
1304 						      NETLINK_CB(cb->skb).portid,
1305 						      nlh->nlmsg_seq,
1306 						      RTM_NEWNETCONF,
1307 						      NLM_F_MULTI,
1308 						      NETCONFA_ALL) < 0) {
1309 				rcu_read_unlock();
1310 				goto done;
1311 			}
1312 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1313 cont:
1314 			idx++;
1315 		}
1316 		rcu_read_unlock();
1317 	}
1318 done:
1319 	cb->args[0] = h;
1320 	cb->args[1] = idx;
1321 
1322 	return skb->len;
1323 }
1324 
1325 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1326 	(&((struct mpls_dev *)0)->field)
1327 
1328 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1329 			  void __user *buffer,
1330 			  size_t *lenp, loff_t *ppos)
1331 {
1332 	int oval = *(int *)ctl->data;
1333 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1334 
1335 	if (write) {
1336 		struct mpls_dev *mdev = ctl->extra1;
1337 		int i = (int *)ctl->data - (int *)mdev;
1338 		struct net *net = ctl->extra2;
1339 		int val = *(int *)ctl->data;
1340 
1341 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1342 		    val != oval) {
1343 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1344 						    NETCONFA_INPUT, mdev);
1345 		}
1346 	}
1347 
1348 	return ret;
1349 }
1350 
1351 static const struct ctl_table mpls_dev_table[] = {
1352 	{
1353 		.procname	= "input",
1354 		.maxlen		= sizeof(int),
1355 		.mode		= 0644,
1356 		.proc_handler	= mpls_conf_proc,
1357 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1358 	},
1359 	{ }
1360 };
1361 
1362 static int mpls_dev_sysctl_register(struct net_device *dev,
1363 				    struct mpls_dev *mdev)
1364 {
1365 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1366 	struct net *net = dev_net(dev);
1367 	struct ctl_table *table;
1368 	int i;
1369 
1370 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1371 	if (!table)
1372 		goto out;
1373 
1374 	/* Table data contains only offsets relative to the base of
1375 	 * the mdev at this point, so make them absolute.
1376 	 */
1377 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1378 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1379 		table[i].extra1 = mdev;
1380 		table[i].extra2 = net;
1381 	}
1382 
1383 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1384 
1385 	mdev->sysctl = register_net_sysctl(net, path, table);
1386 	if (!mdev->sysctl)
1387 		goto free;
1388 
1389 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1390 	return 0;
1391 
1392 free:
1393 	kfree(table);
1394 out:
1395 	return -ENOBUFS;
1396 }
1397 
1398 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1399 				       struct mpls_dev *mdev)
1400 {
1401 	struct net *net = dev_net(dev);
1402 	struct ctl_table *table;
1403 
1404 	table = mdev->sysctl->ctl_table_arg;
1405 	unregister_net_sysctl_table(mdev->sysctl);
1406 	kfree(table);
1407 
1408 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1409 }
1410 
1411 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1412 {
1413 	struct mpls_dev *mdev;
1414 	int err = -ENOMEM;
1415 	int i;
1416 
1417 	ASSERT_RTNL();
1418 
1419 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1420 	if (!mdev)
1421 		return ERR_PTR(err);
1422 
1423 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1424 	if (!mdev->stats)
1425 		goto free;
1426 
1427 	for_each_possible_cpu(i) {
1428 		struct mpls_pcpu_stats *mpls_stats;
1429 
1430 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1431 		u64_stats_init(&mpls_stats->syncp);
1432 	}
1433 
1434 	mdev->dev = dev;
1435 
1436 	err = mpls_dev_sysctl_register(dev, mdev);
1437 	if (err)
1438 		goto free;
1439 
1440 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1441 
1442 	return mdev;
1443 
1444 free:
1445 	free_percpu(mdev->stats);
1446 	kfree(mdev);
1447 	return ERR_PTR(err);
1448 }
1449 
1450 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1451 {
1452 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1453 
1454 	free_percpu(mdev->stats);
1455 	kfree(mdev);
1456 }
1457 
1458 static void mpls_ifdown(struct net_device *dev, int event)
1459 {
1460 	struct mpls_route __rcu **platform_label;
1461 	struct net *net = dev_net(dev);
1462 	u8 alive, deleted;
1463 	unsigned index;
1464 
1465 	platform_label = rtnl_dereference(net->mpls.platform_label);
1466 	for (index = 0; index < net->mpls.platform_labels; index++) {
1467 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1468 
1469 		if (!rt)
1470 			continue;
1471 
1472 		alive = 0;
1473 		deleted = 0;
1474 		change_nexthops(rt) {
1475 			unsigned int nh_flags = nh->nh_flags;
1476 
1477 			if (rtnl_dereference(nh->nh_dev) != dev)
1478 				goto next;
1479 
1480 			switch (event) {
1481 			case NETDEV_DOWN:
1482 			case NETDEV_UNREGISTER:
1483 				nh_flags |= RTNH_F_DEAD;
1484 				/* fall through */
1485 			case NETDEV_CHANGE:
1486 				nh_flags |= RTNH_F_LINKDOWN;
1487 				break;
1488 			}
1489 			if (event == NETDEV_UNREGISTER)
1490 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1491 
1492 			if (nh->nh_flags != nh_flags)
1493 				WRITE_ONCE(nh->nh_flags, nh_flags);
1494 next:
1495 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1496 				alive++;
1497 			if (!rtnl_dereference(nh->nh_dev))
1498 				deleted++;
1499 		} endfor_nexthops(rt);
1500 
1501 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1502 
1503 		/* if there are no more nexthops, delete the route */
1504 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1505 			mpls_route_update(net, index, NULL, NULL);
1506 	}
1507 }
1508 
1509 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1510 {
1511 	struct mpls_route __rcu **platform_label;
1512 	struct net *net = dev_net(dev);
1513 	unsigned index;
1514 	u8 alive;
1515 
1516 	platform_label = rtnl_dereference(net->mpls.platform_label);
1517 	for (index = 0; index < net->mpls.platform_labels; index++) {
1518 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1519 
1520 		if (!rt)
1521 			continue;
1522 
1523 		alive = 0;
1524 		change_nexthops(rt) {
1525 			unsigned int nh_flags = nh->nh_flags;
1526 			struct net_device *nh_dev =
1527 				rtnl_dereference(nh->nh_dev);
1528 
1529 			if (!(nh_flags & flags)) {
1530 				alive++;
1531 				continue;
1532 			}
1533 			if (nh_dev != dev)
1534 				continue;
1535 			alive++;
1536 			nh_flags &= ~flags;
1537 			WRITE_ONCE(nh->nh_flags, nh_flags);
1538 		} endfor_nexthops(rt);
1539 
1540 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1541 	}
1542 }
1543 
1544 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1545 			   void *ptr)
1546 {
1547 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1548 	struct mpls_dev *mdev;
1549 	unsigned int flags;
1550 
1551 	if (event == NETDEV_REGISTER) {
1552 
1553 		/* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1554 		 * IPIP devices
1555 		 */
1556 		if (dev->type == ARPHRD_ETHER ||
1557 		    dev->type == ARPHRD_LOOPBACK ||
1558 		    dev->type == ARPHRD_IPGRE ||
1559 		    dev->type == ARPHRD_IP6GRE ||
1560 		    dev->type == ARPHRD_SIT ||
1561 		    dev->type == ARPHRD_TUNNEL) {
1562 			mdev = mpls_add_dev(dev);
1563 			if (IS_ERR(mdev))
1564 				return notifier_from_errno(PTR_ERR(mdev));
1565 		}
1566 		return NOTIFY_OK;
1567 	}
1568 
1569 	mdev = mpls_dev_get(dev);
1570 	if (!mdev)
1571 		return NOTIFY_OK;
1572 
1573 	switch (event) {
1574 	case NETDEV_DOWN:
1575 		mpls_ifdown(dev, event);
1576 		break;
1577 	case NETDEV_UP:
1578 		flags = dev_get_flags(dev);
1579 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1580 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1581 		else
1582 			mpls_ifup(dev, RTNH_F_DEAD);
1583 		break;
1584 	case NETDEV_CHANGE:
1585 		flags = dev_get_flags(dev);
1586 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1587 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1588 		else
1589 			mpls_ifdown(dev, event);
1590 		break;
1591 	case NETDEV_UNREGISTER:
1592 		mpls_ifdown(dev, event);
1593 		mdev = mpls_dev_get(dev);
1594 		if (mdev) {
1595 			mpls_dev_sysctl_unregister(dev, mdev);
1596 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1597 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1598 		}
1599 		break;
1600 	case NETDEV_CHANGENAME:
1601 		mdev = mpls_dev_get(dev);
1602 		if (mdev) {
1603 			int err;
1604 
1605 			mpls_dev_sysctl_unregister(dev, mdev);
1606 			err = mpls_dev_sysctl_register(dev, mdev);
1607 			if (err)
1608 				return notifier_from_errno(err);
1609 		}
1610 		break;
1611 	}
1612 	return NOTIFY_OK;
1613 }
1614 
1615 static struct notifier_block mpls_dev_notifier = {
1616 	.notifier_call = mpls_dev_notify,
1617 };
1618 
1619 static int nla_put_via(struct sk_buff *skb,
1620 		       u8 table, const void *addr, int alen)
1621 {
1622 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1623 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1624 	};
1625 	struct nlattr *nla;
1626 	struct rtvia *via;
1627 	int family = AF_UNSPEC;
1628 
1629 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1630 	if (!nla)
1631 		return -EMSGSIZE;
1632 
1633 	if (table <= NEIGH_NR_TABLES)
1634 		family = table_to_family[table];
1635 
1636 	via = nla_data(nla);
1637 	via->rtvia_family = family;
1638 	memcpy(via->rtvia_addr, addr, alen);
1639 	return 0;
1640 }
1641 
1642 int nla_put_labels(struct sk_buff *skb, int attrtype,
1643 		   u8 labels, const u32 label[])
1644 {
1645 	struct nlattr *nla;
1646 	struct mpls_shim_hdr *nla_label;
1647 	bool bos;
1648 	int i;
1649 	nla = nla_reserve(skb, attrtype, labels*4);
1650 	if (!nla)
1651 		return -EMSGSIZE;
1652 
1653 	nla_label = nla_data(nla);
1654 	bos = true;
1655 	for (i = labels - 1; i >= 0; i--) {
1656 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1657 		bos = false;
1658 	}
1659 
1660 	return 0;
1661 }
1662 EXPORT_SYMBOL_GPL(nla_put_labels);
1663 
1664 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1665 		   u32 label[], struct netlink_ext_ack *extack)
1666 {
1667 	unsigned len = nla_len(nla);
1668 	struct mpls_shim_hdr *nla_label;
1669 	u8 nla_labels;
1670 	bool bos;
1671 	int i;
1672 
1673 	/* len needs to be an even multiple of 4 (the label size). Number
1674 	 * of labels is a u8 so check for overflow.
1675 	 */
1676 	if (len & 3 || len / 4 > 255) {
1677 		NL_SET_ERR_MSG_ATTR(extack, nla,
1678 				    "Invalid length for labels attribute");
1679 		return -EINVAL;
1680 	}
1681 
1682 	/* Limit the number of new labels allowed */
1683 	nla_labels = len/4;
1684 	if (nla_labels > max_labels) {
1685 		NL_SET_ERR_MSG(extack, "Too many labels");
1686 		return -EINVAL;
1687 	}
1688 
1689 	/* when label == NULL, caller wants number of labels */
1690 	if (!label)
1691 		goto out;
1692 
1693 	nla_label = nla_data(nla);
1694 	bos = true;
1695 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1696 		struct mpls_entry_decoded dec;
1697 		dec = mpls_entry_decode(nla_label + i);
1698 
1699 		/* Ensure the bottom of stack flag is properly set
1700 		 * and ttl and tc are both clear.
1701 		 */
1702 		if (dec.ttl) {
1703 			NL_SET_ERR_MSG_ATTR(extack, nla,
1704 					    "TTL in label must be 0");
1705 			return -EINVAL;
1706 		}
1707 
1708 		if (dec.tc) {
1709 			NL_SET_ERR_MSG_ATTR(extack, nla,
1710 					    "Traffic class in label must be 0");
1711 			return -EINVAL;
1712 		}
1713 
1714 		if (dec.bos != bos) {
1715 			NL_SET_BAD_ATTR(extack, nla);
1716 			if (bos) {
1717 				NL_SET_ERR_MSG(extack,
1718 					       "BOS bit must be set in first label");
1719 			} else {
1720 				NL_SET_ERR_MSG(extack,
1721 					       "BOS bit can only be set in first label");
1722 			}
1723 			return -EINVAL;
1724 		}
1725 
1726 		switch (dec.label) {
1727 		case MPLS_LABEL_IMPLNULL:
1728 			/* RFC3032: This is a label that an LSR may
1729 			 * assign and distribute, but which never
1730 			 * actually appears in the encapsulation.
1731 			 */
1732 			NL_SET_ERR_MSG_ATTR(extack, nla,
1733 					    "Implicit NULL Label (3) can not be used in encapsulation");
1734 			return -EINVAL;
1735 		}
1736 
1737 		label[i] = dec.label;
1738 	}
1739 out:
1740 	*labels = nla_labels;
1741 	return 0;
1742 }
1743 EXPORT_SYMBOL_GPL(nla_get_labels);
1744 
1745 static int rtm_to_route_config(struct sk_buff *skb,
1746 			       struct nlmsghdr *nlh,
1747 			       struct mpls_route_config *cfg,
1748 			       struct netlink_ext_ack *extack)
1749 {
1750 	struct rtmsg *rtm;
1751 	struct nlattr *tb[RTA_MAX+1];
1752 	int index;
1753 	int err;
1754 
1755 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy,
1756 			  extack);
1757 	if (err < 0)
1758 		goto errout;
1759 
1760 	err = -EINVAL;
1761 	rtm = nlmsg_data(nlh);
1762 
1763 	if (rtm->rtm_family != AF_MPLS) {
1764 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1765 		goto errout;
1766 	}
1767 	if (rtm->rtm_dst_len != 20) {
1768 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1769 		goto errout;
1770 	}
1771 	if (rtm->rtm_src_len != 0) {
1772 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1773 		goto errout;
1774 	}
1775 	if (rtm->rtm_tos != 0) {
1776 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1777 		goto errout;
1778 	}
1779 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1780 		NL_SET_ERR_MSG(extack,
1781 			       "MPLS only supports the main route table");
1782 		goto errout;
1783 	}
1784 	/* Any value is acceptable for rtm_protocol */
1785 
1786 	/* As mpls uses destination specific addresses
1787 	 * (or source specific address in the case of multicast)
1788 	 * all addresses have universal scope.
1789 	 */
1790 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1791 		NL_SET_ERR_MSG(extack,
1792 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1793 		goto errout;
1794 	}
1795 	if (rtm->rtm_type != RTN_UNICAST) {
1796 		NL_SET_ERR_MSG(extack,
1797 			       "Invalid route type - MPLS only supports UNICAST");
1798 		goto errout;
1799 	}
1800 	if (rtm->rtm_flags != 0) {
1801 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1802 		goto errout;
1803 	}
1804 
1805 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1806 	cfg->rc_protocol	= rtm->rtm_protocol;
1807 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1808 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1809 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1810 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1811 	cfg->rc_nlinfo.nlh	= nlh;
1812 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1813 
1814 	for (index = 0; index <= RTA_MAX; index++) {
1815 		struct nlattr *nla = tb[index];
1816 		if (!nla)
1817 			continue;
1818 
1819 		switch (index) {
1820 		case RTA_OIF:
1821 			cfg->rc_ifindex = nla_get_u32(nla);
1822 			break;
1823 		case RTA_NEWDST:
1824 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1825 					   &cfg->rc_output_labels,
1826 					   cfg->rc_output_label, extack))
1827 				goto errout;
1828 			break;
1829 		case RTA_DST:
1830 		{
1831 			u8 label_count;
1832 			if (nla_get_labels(nla, 1, &label_count,
1833 					   &cfg->rc_label, extack))
1834 				goto errout;
1835 
1836 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1837 					   &cfg->rc_label, extack))
1838 				goto errout;
1839 			break;
1840 		}
1841 		case RTA_VIA:
1842 		{
1843 			if (nla_get_via(nla, &cfg->rc_via_alen,
1844 					&cfg->rc_via_table, cfg->rc_via,
1845 					extack))
1846 				goto errout;
1847 			break;
1848 		}
1849 		case RTA_MULTIPATH:
1850 		{
1851 			cfg->rc_mp = nla_data(nla);
1852 			cfg->rc_mp_len = nla_len(nla);
1853 			break;
1854 		}
1855 		case RTA_TTL_PROPAGATE:
1856 		{
1857 			u8 ttl_propagate = nla_get_u8(nla);
1858 
1859 			if (ttl_propagate > 1) {
1860 				NL_SET_ERR_MSG_ATTR(extack, nla,
1861 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1862 				goto errout;
1863 			}
1864 			cfg->rc_ttl_propagate = ttl_propagate ?
1865 				MPLS_TTL_PROP_ENABLED :
1866 				MPLS_TTL_PROP_DISABLED;
1867 			break;
1868 		}
1869 		default:
1870 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1871 			/* Unsupported attribute */
1872 			goto errout;
1873 		}
1874 	}
1875 
1876 	err = 0;
1877 errout:
1878 	return err;
1879 }
1880 
1881 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1882 			     struct netlink_ext_ack *extack)
1883 {
1884 	struct mpls_route_config *cfg;
1885 	int err;
1886 
1887 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1888 	if (!cfg)
1889 		return -ENOMEM;
1890 
1891 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1892 	if (err < 0)
1893 		goto out;
1894 
1895 	err = mpls_route_del(cfg, extack);
1896 out:
1897 	kfree(cfg);
1898 
1899 	return err;
1900 }
1901 
1902 
1903 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1904 			     struct netlink_ext_ack *extack)
1905 {
1906 	struct mpls_route_config *cfg;
1907 	int err;
1908 
1909 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1910 	if (!cfg)
1911 		return -ENOMEM;
1912 
1913 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1914 	if (err < 0)
1915 		goto out;
1916 
1917 	err = mpls_route_add(cfg, extack);
1918 out:
1919 	kfree(cfg);
1920 
1921 	return err;
1922 }
1923 
1924 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1925 			   u32 label, struct mpls_route *rt, int flags)
1926 {
1927 	struct net_device *dev;
1928 	struct nlmsghdr *nlh;
1929 	struct rtmsg *rtm;
1930 
1931 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1932 	if (nlh == NULL)
1933 		return -EMSGSIZE;
1934 
1935 	rtm = nlmsg_data(nlh);
1936 	rtm->rtm_family = AF_MPLS;
1937 	rtm->rtm_dst_len = 20;
1938 	rtm->rtm_src_len = 0;
1939 	rtm->rtm_tos = 0;
1940 	rtm->rtm_table = RT_TABLE_MAIN;
1941 	rtm->rtm_protocol = rt->rt_protocol;
1942 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1943 	rtm->rtm_type = RTN_UNICAST;
1944 	rtm->rtm_flags = 0;
1945 
1946 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1947 		goto nla_put_failure;
1948 
1949 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1950 		bool ttl_propagate =
1951 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1952 
1953 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1954 			       ttl_propagate))
1955 			goto nla_put_failure;
1956 	}
1957 	if (rt->rt_nhn == 1) {
1958 		const struct mpls_nh *nh = rt->rt_nh;
1959 
1960 		if (nh->nh_labels &&
1961 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1962 				   nh->nh_label))
1963 			goto nla_put_failure;
1964 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1965 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1966 				nh->nh_via_alen))
1967 			goto nla_put_failure;
1968 		dev = rtnl_dereference(nh->nh_dev);
1969 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1970 			goto nla_put_failure;
1971 		if (nh->nh_flags & RTNH_F_LINKDOWN)
1972 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1973 		if (nh->nh_flags & RTNH_F_DEAD)
1974 			rtm->rtm_flags |= RTNH_F_DEAD;
1975 	} else {
1976 		struct rtnexthop *rtnh;
1977 		struct nlattr *mp;
1978 		u8 linkdown = 0;
1979 		u8 dead = 0;
1980 
1981 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1982 		if (!mp)
1983 			goto nla_put_failure;
1984 
1985 		for_nexthops(rt) {
1986 			dev = rtnl_dereference(nh->nh_dev);
1987 			if (!dev)
1988 				continue;
1989 
1990 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1991 			if (!rtnh)
1992 				goto nla_put_failure;
1993 
1994 			rtnh->rtnh_ifindex = dev->ifindex;
1995 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1996 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1997 				linkdown++;
1998 			}
1999 			if (nh->nh_flags & RTNH_F_DEAD) {
2000 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2001 				dead++;
2002 			}
2003 
2004 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2005 							    nh->nh_labels,
2006 							    nh->nh_label))
2007 				goto nla_put_failure;
2008 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2009 			    nla_put_via(skb, nh->nh_via_table,
2010 					mpls_nh_via(rt, nh),
2011 					nh->nh_via_alen))
2012 				goto nla_put_failure;
2013 
2014 			/* length of rtnetlink header + attributes */
2015 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2016 		} endfor_nexthops(rt);
2017 
2018 		if (linkdown == rt->rt_nhn)
2019 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2020 		if (dead == rt->rt_nhn)
2021 			rtm->rtm_flags |= RTNH_F_DEAD;
2022 
2023 		nla_nest_end(skb, mp);
2024 	}
2025 
2026 	nlmsg_end(skb, nlh);
2027 	return 0;
2028 
2029 nla_put_failure:
2030 	nlmsg_cancel(skb, nlh);
2031 	return -EMSGSIZE;
2032 }
2033 
2034 #if IS_ENABLED(CONFIG_INET)
2035 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2036 				   struct fib_dump_filter *filter,
2037 				   struct netlink_callback *cb)
2038 {
2039 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2040 }
2041 #else
2042 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2043 				   struct fib_dump_filter *filter,
2044 				   struct netlink_callback *cb)
2045 {
2046 	struct netlink_ext_ack *extack = cb->extack;
2047 	struct nlattr *tb[RTA_MAX + 1];
2048 	struct rtmsg *rtm;
2049 	int err, i;
2050 
2051 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2052 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2053 		return -EINVAL;
2054 	}
2055 
2056 	rtm = nlmsg_data(nlh);
2057 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2058 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2059 	    rtm->rtm_flags) {
2060 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2061 		return -EINVAL;
2062 	}
2063 
2064 	if (rtm->rtm_protocol) {
2065 		filter->protocol = rtm->rtm_protocol;
2066 		filter->filter_set = 1;
2067 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2068 	}
2069 
2070 	err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2071 				 rtm_mpls_policy, extack);
2072 	if (err < 0)
2073 		return err;
2074 
2075 	for (i = 0; i <= RTA_MAX; ++i) {
2076 		int ifindex;
2077 
2078 		if (i == RTA_OIF) {
2079 			ifindex = nla_get_u32(tb[i]);
2080 			filter->dev = __dev_get_by_index(net, ifindex);
2081 			if (!filter->dev)
2082 				return -ENODEV;
2083 			filter->filter_set = 1;
2084 		} else if (tb[i]) {
2085 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2086 			return -EINVAL;
2087 		}
2088 	}
2089 
2090 	return 0;
2091 }
2092 #endif
2093 
2094 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2095 			     const struct net_device *dev)
2096 {
2097 	struct net_device *nh_dev;
2098 
2099 	if (rt->rt_nhn == 1) {
2100 		struct mpls_nh *nh = rt->rt_nh;
2101 
2102 		nh_dev = rtnl_dereference(nh->nh_dev);
2103 		if (dev == nh_dev)
2104 			return true;
2105 	} else {
2106 		for_nexthops(rt) {
2107 			nh_dev = rtnl_dereference(nh->nh_dev);
2108 			if (nh_dev == dev)
2109 				return true;
2110 		} endfor_nexthops(rt);
2111 	}
2112 
2113 	return false;
2114 }
2115 
2116 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2117 {
2118 	const struct nlmsghdr *nlh = cb->nlh;
2119 	struct net *net = sock_net(skb->sk);
2120 	struct mpls_route __rcu **platform_label;
2121 	struct fib_dump_filter filter = {};
2122 	unsigned int flags = NLM_F_MULTI;
2123 	size_t platform_labels;
2124 	unsigned int index;
2125 
2126 	ASSERT_RTNL();
2127 
2128 	if (cb->strict_check) {
2129 		int err;
2130 
2131 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2132 		if (err < 0)
2133 			return err;
2134 
2135 		/* for MPLS, there is only 1 table with fixed type and flags.
2136 		 * If either are set in the filter then return nothing.
2137 		 */
2138 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2139 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2140 		     filter.flags)
2141 			return skb->len;
2142 	}
2143 
2144 	index = cb->args[0];
2145 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2146 		index = MPLS_LABEL_FIRST_UNRESERVED;
2147 
2148 	platform_label = rtnl_dereference(net->mpls.platform_label);
2149 	platform_labels = net->mpls.platform_labels;
2150 
2151 	if (filter.filter_set)
2152 		flags |= NLM_F_DUMP_FILTERED;
2153 
2154 	for (; index < platform_labels; index++) {
2155 		struct mpls_route *rt;
2156 
2157 		rt = rtnl_dereference(platform_label[index]);
2158 		if (!rt)
2159 			continue;
2160 
2161 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2162 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2163 			continue;
2164 
2165 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2166 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2167 				    index, rt, flags) < 0)
2168 			break;
2169 	}
2170 	cb->args[0] = index;
2171 
2172 	return skb->len;
2173 }
2174 
2175 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2176 {
2177 	size_t payload =
2178 		NLMSG_ALIGN(sizeof(struct rtmsg))
2179 		+ nla_total_size(4)			/* RTA_DST */
2180 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2181 
2182 	if (rt->rt_nhn == 1) {
2183 		struct mpls_nh *nh = rt->rt_nh;
2184 
2185 		if (nh->nh_dev)
2186 			payload += nla_total_size(4); /* RTA_OIF */
2187 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2188 			payload += nla_total_size(2 + nh->nh_via_alen);
2189 		if (nh->nh_labels) /* RTA_NEWDST */
2190 			payload += nla_total_size(nh->nh_labels * 4);
2191 	} else {
2192 		/* each nexthop is packed in an attribute */
2193 		size_t nhsize = 0;
2194 
2195 		for_nexthops(rt) {
2196 			if (!rtnl_dereference(nh->nh_dev))
2197 				continue;
2198 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2199 			/* RTA_VIA */
2200 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2201 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2202 			if (nh->nh_labels)
2203 				nhsize += nla_total_size(nh->nh_labels * 4);
2204 		} endfor_nexthops(rt);
2205 		/* nested attribute */
2206 		payload += nla_total_size(nhsize);
2207 	}
2208 
2209 	return payload;
2210 }
2211 
2212 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2213 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2214 		       unsigned int nlm_flags)
2215 {
2216 	struct sk_buff *skb;
2217 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2218 	int err = -ENOBUFS;
2219 
2220 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2221 	if (skb == NULL)
2222 		goto errout;
2223 
2224 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2225 	if (err < 0) {
2226 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2227 		WARN_ON(err == -EMSGSIZE);
2228 		kfree_skb(skb);
2229 		goto errout;
2230 	}
2231 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2232 
2233 	return;
2234 errout:
2235 	if (err < 0)
2236 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2237 }
2238 
2239 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2240 			 struct netlink_ext_ack *extack)
2241 {
2242 	struct net *net = sock_net(in_skb->sk);
2243 	u32 portid = NETLINK_CB(in_skb).portid;
2244 	u32 in_label = LABEL_NOT_SPECIFIED;
2245 	struct nlattr *tb[RTA_MAX + 1];
2246 	u32 labels[MAX_NEW_LABELS];
2247 	struct mpls_shim_hdr *hdr;
2248 	unsigned int hdr_size = 0;
2249 	struct net_device *dev;
2250 	struct mpls_route *rt;
2251 	struct rtmsg *rtm, *r;
2252 	struct nlmsghdr *nlh;
2253 	struct sk_buff *skb;
2254 	struct mpls_nh *nh;
2255 	u8 n_labels;
2256 	int err;
2257 
2258 	err = nlmsg_parse(in_nlh, sizeof(*rtm), tb, RTA_MAX,
2259 			  rtm_mpls_policy, extack);
2260 	if (err < 0)
2261 		goto errout;
2262 
2263 	rtm = nlmsg_data(in_nlh);
2264 
2265 	if (tb[RTA_DST]) {
2266 		u8 label_count;
2267 
2268 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2269 				   &in_label, extack)) {
2270 			err = -EINVAL;
2271 			goto errout;
2272 		}
2273 
2274 		if (!mpls_label_ok(net, &in_label, extack)) {
2275 			err = -EINVAL;
2276 			goto errout;
2277 		}
2278 	}
2279 
2280 	rt = mpls_route_input_rcu(net, in_label);
2281 	if (!rt) {
2282 		err = -ENETUNREACH;
2283 		goto errout;
2284 	}
2285 
2286 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2287 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2288 		if (!skb) {
2289 			err = -ENOBUFS;
2290 			goto errout;
2291 		}
2292 
2293 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2294 				      RTM_NEWROUTE, in_label, rt, 0);
2295 		if (err < 0) {
2296 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2297 			WARN_ON(err == -EMSGSIZE);
2298 			goto errout_free;
2299 		}
2300 
2301 		return rtnl_unicast(skb, net, portid);
2302 	}
2303 
2304 	if (tb[RTA_NEWDST]) {
2305 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2306 				   labels, extack) != 0) {
2307 			err = -EINVAL;
2308 			goto errout;
2309 		}
2310 
2311 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2312 	}
2313 
2314 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2315 	if (!skb) {
2316 		err = -ENOBUFS;
2317 		goto errout;
2318 	}
2319 
2320 	skb->protocol = htons(ETH_P_MPLS_UC);
2321 
2322 	if (hdr_size) {
2323 		bool bos;
2324 		int i;
2325 
2326 		if (skb_cow(skb, hdr_size)) {
2327 			err = -ENOBUFS;
2328 			goto errout_free;
2329 		}
2330 
2331 		skb_reserve(skb, hdr_size);
2332 		skb_push(skb, hdr_size);
2333 		skb_reset_network_header(skb);
2334 
2335 		/* Push new labels */
2336 		hdr = mpls_hdr(skb);
2337 		bos = true;
2338 		for (i = n_labels - 1; i >= 0; i--) {
2339 			hdr[i] = mpls_entry_encode(labels[i],
2340 						   1, 0, bos);
2341 			bos = false;
2342 		}
2343 	}
2344 
2345 	nh = mpls_select_multipath(rt, skb);
2346 	if (!nh) {
2347 		err = -ENETUNREACH;
2348 		goto errout_free;
2349 	}
2350 
2351 	if (hdr_size) {
2352 		skb_pull(skb, hdr_size);
2353 		skb_reset_network_header(skb);
2354 	}
2355 
2356 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2357 			RTM_NEWROUTE, sizeof(*r), 0);
2358 	if (!nlh) {
2359 		err = -EMSGSIZE;
2360 		goto errout_free;
2361 	}
2362 
2363 	r = nlmsg_data(nlh);
2364 	r->rtm_family	 = AF_MPLS;
2365 	r->rtm_dst_len	= 20;
2366 	r->rtm_src_len	= 0;
2367 	r->rtm_table	= RT_TABLE_MAIN;
2368 	r->rtm_type	= RTN_UNICAST;
2369 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2370 	r->rtm_protocol = rt->rt_protocol;
2371 	r->rtm_flags	= 0;
2372 
2373 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2374 		goto nla_put_failure;
2375 
2376 	if (nh->nh_labels &&
2377 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2378 			   nh->nh_label))
2379 		goto nla_put_failure;
2380 
2381 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2382 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2383 			nh->nh_via_alen))
2384 		goto nla_put_failure;
2385 	dev = rtnl_dereference(nh->nh_dev);
2386 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2387 		goto nla_put_failure;
2388 
2389 	nlmsg_end(skb, nlh);
2390 
2391 	err = rtnl_unicast(skb, net, portid);
2392 errout:
2393 	return err;
2394 
2395 nla_put_failure:
2396 	nlmsg_cancel(skb, nlh);
2397 	err = -EMSGSIZE;
2398 errout_free:
2399 	kfree_skb(skb);
2400 	return err;
2401 }
2402 
2403 static int resize_platform_label_table(struct net *net, size_t limit)
2404 {
2405 	size_t size = sizeof(struct mpls_route *) * limit;
2406 	size_t old_limit;
2407 	size_t cp_size;
2408 	struct mpls_route __rcu **labels = NULL, **old;
2409 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2410 	unsigned index;
2411 
2412 	if (size) {
2413 		labels = kvzalloc(size, GFP_KERNEL);
2414 		if (!labels)
2415 			goto nolabels;
2416 	}
2417 
2418 	/* In case the predefined labels need to be populated */
2419 	if (limit > MPLS_LABEL_IPV4NULL) {
2420 		struct net_device *lo = net->loopback_dev;
2421 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2422 		if (IS_ERR(rt0))
2423 			goto nort0;
2424 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2425 		rt0->rt_protocol = RTPROT_KERNEL;
2426 		rt0->rt_payload_type = MPT_IPV4;
2427 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2428 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2429 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2430 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2431 		       lo->addr_len);
2432 	}
2433 	if (limit > MPLS_LABEL_IPV6NULL) {
2434 		struct net_device *lo = net->loopback_dev;
2435 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2436 		if (IS_ERR(rt2))
2437 			goto nort2;
2438 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2439 		rt2->rt_protocol = RTPROT_KERNEL;
2440 		rt2->rt_payload_type = MPT_IPV6;
2441 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2442 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2443 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2444 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2445 		       lo->addr_len);
2446 	}
2447 
2448 	rtnl_lock();
2449 	/* Remember the original table */
2450 	old = rtnl_dereference(net->mpls.platform_label);
2451 	old_limit = net->mpls.platform_labels;
2452 
2453 	/* Free any labels beyond the new table */
2454 	for (index = limit; index < old_limit; index++)
2455 		mpls_route_update(net, index, NULL, NULL);
2456 
2457 	/* Copy over the old labels */
2458 	cp_size = size;
2459 	if (old_limit < limit)
2460 		cp_size = old_limit * sizeof(struct mpls_route *);
2461 
2462 	memcpy(labels, old, cp_size);
2463 
2464 	/* If needed set the predefined labels */
2465 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2466 	    (limit > MPLS_LABEL_IPV6NULL)) {
2467 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2468 		rt2 = NULL;
2469 	}
2470 
2471 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2472 	    (limit > MPLS_LABEL_IPV4NULL)) {
2473 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2474 		rt0 = NULL;
2475 	}
2476 
2477 	/* Update the global pointers */
2478 	net->mpls.platform_labels = limit;
2479 	rcu_assign_pointer(net->mpls.platform_label, labels);
2480 
2481 	rtnl_unlock();
2482 
2483 	mpls_rt_free(rt2);
2484 	mpls_rt_free(rt0);
2485 
2486 	if (old) {
2487 		synchronize_rcu();
2488 		kvfree(old);
2489 	}
2490 	return 0;
2491 
2492 nort2:
2493 	mpls_rt_free(rt0);
2494 nort0:
2495 	kvfree(labels);
2496 nolabels:
2497 	return -ENOMEM;
2498 }
2499 
2500 static int mpls_platform_labels(struct ctl_table *table, int write,
2501 				void __user *buffer, size_t *lenp, loff_t *ppos)
2502 {
2503 	struct net *net = table->data;
2504 	int platform_labels = net->mpls.platform_labels;
2505 	int ret;
2506 	struct ctl_table tmp = {
2507 		.procname	= table->procname,
2508 		.data		= &platform_labels,
2509 		.maxlen		= sizeof(int),
2510 		.mode		= table->mode,
2511 		.extra1		= &zero,
2512 		.extra2		= &label_limit,
2513 	};
2514 
2515 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2516 
2517 	if (write && ret == 0)
2518 		ret = resize_platform_label_table(net, platform_labels);
2519 
2520 	return ret;
2521 }
2522 
2523 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2524 	(&((struct net *)0)->field)
2525 
2526 static const struct ctl_table mpls_table[] = {
2527 	{
2528 		.procname	= "platform_labels",
2529 		.data		= NULL,
2530 		.maxlen		= sizeof(int),
2531 		.mode		= 0644,
2532 		.proc_handler	= mpls_platform_labels,
2533 	},
2534 	{
2535 		.procname	= "ip_ttl_propagate",
2536 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2537 		.maxlen		= sizeof(int),
2538 		.mode		= 0644,
2539 		.proc_handler	= proc_dointvec_minmax,
2540 		.extra1		= &zero,
2541 		.extra2		= &one,
2542 	},
2543 	{
2544 		.procname	= "default_ttl",
2545 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2546 		.maxlen		= sizeof(int),
2547 		.mode		= 0644,
2548 		.proc_handler	= proc_dointvec_minmax,
2549 		.extra1		= &one,
2550 		.extra2		= &ttl_max,
2551 	},
2552 	{ }
2553 };
2554 
2555 static int mpls_net_init(struct net *net)
2556 {
2557 	struct ctl_table *table;
2558 	int i;
2559 
2560 	net->mpls.platform_labels = 0;
2561 	net->mpls.platform_label = NULL;
2562 	net->mpls.ip_ttl_propagate = 1;
2563 	net->mpls.default_ttl = 255;
2564 
2565 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2566 	if (table == NULL)
2567 		return -ENOMEM;
2568 
2569 	/* Table data contains only offsets relative to the base of
2570 	 * the mdev at this point, so make them absolute.
2571 	 */
2572 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2573 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2574 
2575 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2576 	if (net->mpls.ctl == NULL) {
2577 		kfree(table);
2578 		return -ENOMEM;
2579 	}
2580 
2581 	return 0;
2582 }
2583 
2584 static void mpls_net_exit(struct net *net)
2585 {
2586 	struct mpls_route __rcu **platform_label;
2587 	size_t platform_labels;
2588 	struct ctl_table *table;
2589 	unsigned int index;
2590 
2591 	table = net->mpls.ctl->ctl_table_arg;
2592 	unregister_net_sysctl_table(net->mpls.ctl);
2593 	kfree(table);
2594 
2595 	/* An rcu grace period has passed since there was a device in
2596 	 * the network namespace (and thus the last in flight packet)
2597 	 * left this network namespace.  This is because
2598 	 * unregister_netdevice_many and netdev_run_todo has completed
2599 	 * for each network device that was in this network namespace.
2600 	 *
2601 	 * As such no additional rcu synchronization is necessary when
2602 	 * freeing the platform_label table.
2603 	 */
2604 	rtnl_lock();
2605 	platform_label = rtnl_dereference(net->mpls.platform_label);
2606 	platform_labels = net->mpls.platform_labels;
2607 	for (index = 0; index < platform_labels; index++) {
2608 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2609 		RCU_INIT_POINTER(platform_label[index], NULL);
2610 		mpls_notify_route(net, index, rt, NULL, NULL);
2611 		mpls_rt_free(rt);
2612 	}
2613 	rtnl_unlock();
2614 
2615 	kvfree(platform_label);
2616 }
2617 
2618 static struct pernet_operations mpls_net_ops = {
2619 	.init = mpls_net_init,
2620 	.exit = mpls_net_exit,
2621 };
2622 
2623 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2624 	.family		   = AF_MPLS,
2625 	.fill_stats_af	   = mpls_fill_stats_af,
2626 	.get_stats_af_size = mpls_get_stats_af_size,
2627 };
2628 
2629 static int __init mpls_init(void)
2630 {
2631 	int err;
2632 
2633 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2634 
2635 	err = register_pernet_subsys(&mpls_net_ops);
2636 	if (err)
2637 		goto out;
2638 
2639 	err = register_netdevice_notifier(&mpls_dev_notifier);
2640 	if (err)
2641 		goto out_unregister_pernet;
2642 
2643 	dev_add_pack(&mpls_packet_type);
2644 
2645 	rtnl_af_register(&mpls_af_ops);
2646 
2647 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2648 			     mpls_rtm_newroute, NULL, 0);
2649 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2650 			     mpls_rtm_delroute, NULL, 0);
2651 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2652 			     mpls_getroute, mpls_dump_routes, 0);
2653 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2654 			     mpls_netconf_get_devconf,
2655 			     mpls_netconf_dump_devconf, 0);
2656 	err = ipgre_tunnel_encap_add_mpls_ops();
2657 	if (err)
2658 		pr_err("Can't add mpls over gre tunnel ops\n");
2659 
2660 	err = 0;
2661 out:
2662 	return err;
2663 
2664 out_unregister_pernet:
2665 	unregister_pernet_subsys(&mpls_net_ops);
2666 	goto out;
2667 }
2668 module_init(mpls_init);
2669 
2670 static void __exit mpls_exit(void)
2671 {
2672 	rtnl_unregister_all(PF_MPLS);
2673 	rtnl_af_unregister(&mpls_af_ops);
2674 	dev_remove_pack(&mpls_packet_type);
2675 	unregister_netdevice_notifier(&mpls_dev_notifier);
2676 	unregister_pernet_subsys(&mpls_net_ops);
2677 	ipgre_tunnel_encap_del_mpls_ops();
2678 }
2679 module_exit(mpls_exit);
2680 
2681 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2682 MODULE_LICENSE("GPL v2");
2683 MODULE_ALIAS_NETPROTO(PF_MPLS);
2684