xref: /openbmc/linux/net/mpls/af_mpls.c (revision d9fd5a71)
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 	skb_dst_drop(skb);
381 
382 	/* Read and decode the label */
383 	hdr = mpls_hdr(skb);
384 	dec = mpls_entry_decode(hdr);
385 
386 	rt = mpls_route_input_rcu(net, dec.label);
387 	if (!rt) {
388 		MPLS_INC_STATS(mdev, rx_noroute);
389 		goto drop;
390 	}
391 
392 	nh = mpls_select_multipath(rt, skb);
393 	if (!nh)
394 		goto err;
395 
396 	/* Pop the label */
397 	skb_pull(skb, sizeof(*hdr));
398 	skb_reset_network_header(skb);
399 
400 	skb_orphan(skb);
401 
402 	if (skb_warn_if_lro(skb))
403 		goto err;
404 
405 	skb_forward_csum(skb);
406 
407 	/* Verify ttl is valid */
408 	if (dec.ttl <= 1)
409 		goto err;
410 	dec.ttl -= 1;
411 
412 	/* Find the output device */
413 	out_dev = rcu_dereference(nh->nh_dev);
414 	if (!mpls_output_possible(out_dev))
415 		goto tx_err;
416 
417 	/* Verify the destination can hold the packet */
418 	new_header_size = mpls_nh_header_size(nh);
419 	mtu = mpls_dev_mtu(out_dev);
420 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
421 		goto tx_err;
422 
423 	hh_len = LL_RESERVED_SPACE(out_dev);
424 	if (!out_dev->header_ops)
425 		hh_len = 0;
426 
427 	/* Ensure there is enough space for the headers in the skb */
428 	if (skb_cow(skb, hh_len + new_header_size))
429 		goto tx_err;
430 
431 	skb->dev = out_dev;
432 	skb->protocol = htons(ETH_P_MPLS_UC);
433 
434 	if (unlikely(!new_header_size && dec.bos)) {
435 		/* Penultimate hop popping */
436 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
437 			goto err;
438 	} else {
439 		bool bos;
440 		int i;
441 		skb_push(skb, new_header_size);
442 		skb_reset_network_header(skb);
443 		/* Push the new labels */
444 		hdr = mpls_hdr(skb);
445 		bos = dec.bos;
446 		for (i = nh->nh_labels - 1; i >= 0; i--) {
447 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
448 						   dec.ttl, 0, bos);
449 			bos = false;
450 		}
451 	}
452 
453 	mpls_stats_inc_outucastpkts(out_dev, skb);
454 
455 	/* If via wasn't specified then send out using device address */
456 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
457 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
458 				 out_dev->dev_addr, skb);
459 	else
460 		err = neigh_xmit(nh->nh_via_table, out_dev,
461 				 mpls_nh_via(rt, nh), skb);
462 	if (err)
463 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
464 				    __func__, err);
465 	return 0;
466 
467 tx_err:
468 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
469 	if (out_mdev)
470 		MPLS_INC_STATS(out_mdev, tx_errors);
471 	goto drop;
472 err:
473 	MPLS_INC_STATS(mdev, rx_errors);
474 drop:
475 	kfree_skb(skb);
476 	return NET_RX_DROP;
477 }
478 
479 static struct packet_type mpls_packet_type __read_mostly = {
480 	.type = cpu_to_be16(ETH_P_MPLS_UC),
481 	.func = mpls_forward,
482 };
483 
484 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
485 	[RTA_DST]		= { .type = NLA_U32 },
486 	[RTA_OIF]		= { .type = NLA_U32 },
487 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
488 };
489 
490 struct mpls_route_config {
491 	u32			rc_protocol;
492 	u32			rc_ifindex;
493 	u8			rc_via_table;
494 	u8			rc_via_alen;
495 	u8			rc_via[MAX_VIA_ALEN];
496 	u32			rc_label;
497 	u8			rc_ttl_propagate;
498 	u8			rc_output_labels;
499 	u32			rc_output_label[MAX_NEW_LABELS];
500 	u32			rc_nlflags;
501 	enum mpls_payload_type	rc_payload_type;
502 	struct nl_info		rc_nlinfo;
503 	struct rtnexthop	*rc_mp;
504 	int			rc_mp_len;
505 };
506 
507 /* all nexthops within a route have the same size based on max
508  * number of labels and max via length for a hop
509  */
510 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
511 {
512 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
513 	struct mpls_route *rt;
514 	size_t size;
515 
516 	size = sizeof(*rt) + num_nh * nh_size;
517 	if (size > MAX_MPLS_ROUTE_MEM)
518 		return ERR_PTR(-EINVAL);
519 
520 	rt = kzalloc(size, GFP_KERNEL);
521 	if (!rt)
522 		return ERR_PTR(-ENOMEM);
523 
524 	rt->rt_nhn = num_nh;
525 	rt->rt_nhn_alive = num_nh;
526 	rt->rt_nh_size = nh_size;
527 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
528 
529 	return rt;
530 }
531 
532 static void mpls_rt_free(struct mpls_route *rt)
533 {
534 	if (rt)
535 		kfree_rcu(rt, rt_rcu);
536 }
537 
538 static void mpls_notify_route(struct net *net, unsigned index,
539 			      struct mpls_route *old, struct mpls_route *new,
540 			      const struct nl_info *info)
541 {
542 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
543 	unsigned portid = info ? info->portid : 0;
544 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
545 	struct mpls_route *rt = new ? new : old;
546 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
547 	/* Ignore reserved labels for now */
548 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
549 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
550 }
551 
552 static void mpls_route_update(struct net *net, unsigned index,
553 			      struct mpls_route *new,
554 			      const struct nl_info *info)
555 {
556 	struct mpls_route __rcu **platform_label;
557 	struct mpls_route *rt;
558 
559 	ASSERT_RTNL();
560 
561 	platform_label = rtnl_dereference(net->mpls.platform_label);
562 	rt = rtnl_dereference(platform_label[index]);
563 	rcu_assign_pointer(platform_label[index], new);
564 
565 	mpls_notify_route(net, index, rt, new, info);
566 
567 	/* If we removed a route free it now */
568 	mpls_rt_free(rt);
569 }
570 
571 static unsigned find_free_label(struct net *net)
572 {
573 	struct mpls_route __rcu **platform_label;
574 	size_t platform_labels;
575 	unsigned index;
576 
577 	platform_label = rtnl_dereference(net->mpls.platform_label);
578 	platform_labels = net->mpls.platform_labels;
579 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
580 	     index++) {
581 		if (!rtnl_dereference(platform_label[index]))
582 			return index;
583 	}
584 	return LABEL_NOT_SPECIFIED;
585 }
586 
587 #if IS_ENABLED(CONFIG_INET)
588 static struct net_device *inet_fib_lookup_dev(struct net *net,
589 					      const void *addr)
590 {
591 	struct net_device *dev;
592 	struct rtable *rt;
593 	struct in_addr daddr;
594 
595 	memcpy(&daddr, addr, sizeof(struct in_addr));
596 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
597 	if (IS_ERR(rt))
598 		return ERR_CAST(rt);
599 
600 	dev = rt->dst.dev;
601 	dev_hold(dev);
602 
603 	ip_rt_put(rt);
604 
605 	return dev;
606 }
607 #else
608 static struct net_device *inet_fib_lookup_dev(struct net *net,
609 					      const void *addr)
610 {
611 	return ERR_PTR(-EAFNOSUPPORT);
612 }
613 #endif
614 
615 #if IS_ENABLED(CONFIG_IPV6)
616 static struct net_device *inet6_fib_lookup_dev(struct net *net,
617 					       const void *addr)
618 {
619 	struct net_device *dev;
620 	struct dst_entry *dst;
621 	struct flowi6 fl6;
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 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
629 	if (IS_ERR(dst))
630 		return ERR_CAST(dst);
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_valid_get_req(struct sk_buff *skb,
1213 				      const struct nlmsghdr *nlh,
1214 				      struct nlattr **tb,
1215 				      struct netlink_ext_ack *extack)
1216 {
1217 	int i, err;
1218 
1219 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1220 		NL_SET_ERR_MSG_MOD(extack,
1221 				   "Invalid header for netconf get request");
1222 		return -EINVAL;
1223 	}
1224 
1225 	if (!netlink_strict_get_check(skb))
1226 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1227 					      tb, NETCONFA_MAX,
1228 					      devconf_mpls_policy, extack);
1229 
1230 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1231 					    tb, NETCONFA_MAX,
1232 					    devconf_mpls_policy, extack);
1233 	if (err)
1234 		return err;
1235 
1236 	for (i = 0; i <= NETCONFA_MAX; i++) {
1237 		if (!tb[i])
1238 			continue;
1239 
1240 		switch (i) {
1241 		case NETCONFA_IFINDEX:
1242 			break;
1243 		default:
1244 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1245 			return -EINVAL;
1246 		}
1247 	}
1248 
1249 	return 0;
1250 }
1251 
1252 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1253 				    struct nlmsghdr *nlh,
1254 				    struct netlink_ext_ack *extack)
1255 {
1256 	struct net *net = sock_net(in_skb->sk);
1257 	struct nlattr *tb[NETCONFA_MAX + 1];
1258 	struct net_device *dev;
1259 	struct mpls_dev *mdev;
1260 	struct sk_buff *skb;
1261 	int ifindex;
1262 	int err;
1263 
1264 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1265 	if (err < 0)
1266 		goto errout;
1267 
1268 	err = -EINVAL;
1269 	if (!tb[NETCONFA_IFINDEX])
1270 		goto errout;
1271 
1272 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1273 	dev = __dev_get_by_index(net, ifindex);
1274 	if (!dev)
1275 		goto errout;
1276 
1277 	mdev = mpls_dev_get(dev);
1278 	if (!mdev)
1279 		goto errout;
1280 
1281 	err = -ENOBUFS;
1282 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1283 	if (!skb)
1284 		goto errout;
1285 
1286 	err = mpls_netconf_fill_devconf(skb, mdev,
1287 					NETLINK_CB(in_skb).portid,
1288 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1289 					NETCONFA_ALL);
1290 	if (err < 0) {
1291 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1292 		WARN_ON(err == -EMSGSIZE);
1293 		kfree_skb(skb);
1294 		goto errout;
1295 	}
1296 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1297 errout:
1298 	return err;
1299 }
1300 
1301 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1302 				     struct netlink_callback *cb)
1303 {
1304 	const struct nlmsghdr *nlh = cb->nlh;
1305 	struct net *net = sock_net(skb->sk);
1306 	struct hlist_head *head;
1307 	struct net_device *dev;
1308 	struct mpls_dev *mdev;
1309 	int idx, s_idx;
1310 	int h, s_h;
1311 
1312 	if (cb->strict_check) {
1313 		struct netlink_ext_ack *extack = cb->extack;
1314 		struct netconfmsg *ncm;
1315 
1316 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1317 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1318 			return -EINVAL;
1319 		}
1320 
1321 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1322 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1323 			return -EINVAL;
1324 		}
1325 	}
1326 
1327 	s_h = cb->args[0];
1328 	s_idx = idx = cb->args[1];
1329 
1330 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1331 		idx = 0;
1332 		head = &net->dev_index_head[h];
1333 		rcu_read_lock();
1334 		cb->seq = net->dev_base_seq;
1335 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1336 			if (idx < s_idx)
1337 				goto cont;
1338 			mdev = mpls_dev_get(dev);
1339 			if (!mdev)
1340 				goto cont;
1341 			if (mpls_netconf_fill_devconf(skb, mdev,
1342 						      NETLINK_CB(cb->skb).portid,
1343 						      nlh->nlmsg_seq,
1344 						      RTM_NEWNETCONF,
1345 						      NLM_F_MULTI,
1346 						      NETCONFA_ALL) < 0) {
1347 				rcu_read_unlock();
1348 				goto done;
1349 			}
1350 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1351 cont:
1352 			idx++;
1353 		}
1354 		rcu_read_unlock();
1355 	}
1356 done:
1357 	cb->args[0] = h;
1358 	cb->args[1] = idx;
1359 
1360 	return skb->len;
1361 }
1362 
1363 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1364 	(&((struct mpls_dev *)0)->field)
1365 
1366 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1367 			  void *buffer, size_t *lenp, loff_t *ppos)
1368 {
1369 	int oval = *(int *)ctl->data;
1370 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1371 
1372 	if (write) {
1373 		struct mpls_dev *mdev = ctl->extra1;
1374 		int i = (int *)ctl->data - (int *)mdev;
1375 		struct net *net = ctl->extra2;
1376 		int val = *(int *)ctl->data;
1377 
1378 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1379 		    val != oval) {
1380 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1381 						    NETCONFA_INPUT, mdev);
1382 		}
1383 	}
1384 
1385 	return ret;
1386 }
1387 
1388 static const struct ctl_table mpls_dev_table[] = {
1389 	{
1390 		.procname	= "input",
1391 		.maxlen		= sizeof(int),
1392 		.mode		= 0644,
1393 		.proc_handler	= mpls_conf_proc,
1394 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1395 	},
1396 	{ }
1397 };
1398 
1399 static int mpls_dev_sysctl_register(struct net_device *dev,
1400 				    struct mpls_dev *mdev)
1401 {
1402 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1403 	struct net *net = dev_net(dev);
1404 	struct ctl_table *table;
1405 	int i;
1406 
1407 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1408 	if (!table)
1409 		goto out;
1410 
1411 	/* Table data contains only offsets relative to the base of
1412 	 * the mdev at this point, so make them absolute.
1413 	 */
1414 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1415 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1416 		table[i].extra1 = mdev;
1417 		table[i].extra2 = net;
1418 	}
1419 
1420 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1421 
1422 	mdev->sysctl = register_net_sysctl(net, path, table);
1423 	if (!mdev->sysctl)
1424 		goto free;
1425 
1426 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1427 	return 0;
1428 
1429 free:
1430 	kfree(table);
1431 out:
1432 	return -ENOBUFS;
1433 }
1434 
1435 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1436 				       struct mpls_dev *mdev)
1437 {
1438 	struct net *net = dev_net(dev);
1439 	struct ctl_table *table;
1440 
1441 	table = mdev->sysctl->ctl_table_arg;
1442 	unregister_net_sysctl_table(mdev->sysctl);
1443 	kfree(table);
1444 
1445 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1446 }
1447 
1448 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1449 {
1450 	struct mpls_dev *mdev;
1451 	int err = -ENOMEM;
1452 	int i;
1453 
1454 	ASSERT_RTNL();
1455 
1456 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1457 	if (!mdev)
1458 		return ERR_PTR(err);
1459 
1460 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1461 	if (!mdev->stats)
1462 		goto free;
1463 
1464 	for_each_possible_cpu(i) {
1465 		struct mpls_pcpu_stats *mpls_stats;
1466 
1467 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1468 		u64_stats_init(&mpls_stats->syncp);
1469 	}
1470 
1471 	mdev->dev = dev;
1472 
1473 	err = mpls_dev_sysctl_register(dev, mdev);
1474 	if (err)
1475 		goto free;
1476 
1477 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1478 
1479 	return mdev;
1480 
1481 free:
1482 	free_percpu(mdev->stats);
1483 	kfree(mdev);
1484 	return ERR_PTR(err);
1485 }
1486 
1487 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1488 {
1489 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1490 
1491 	free_percpu(mdev->stats);
1492 	kfree(mdev);
1493 }
1494 
1495 static void mpls_ifdown(struct net_device *dev, int event)
1496 {
1497 	struct mpls_route __rcu **platform_label;
1498 	struct net *net = dev_net(dev);
1499 	u8 alive, deleted;
1500 	unsigned index;
1501 
1502 	platform_label = rtnl_dereference(net->mpls.platform_label);
1503 	for (index = 0; index < net->mpls.platform_labels; index++) {
1504 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1505 
1506 		if (!rt)
1507 			continue;
1508 
1509 		alive = 0;
1510 		deleted = 0;
1511 		change_nexthops(rt) {
1512 			unsigned int nh_flags = nh->nh_flags;
1513 
1514 			if (rtnl_dereference(nh->nh_dev) != dev)
1515 				goto next;
1516 
1517 			switch (event) {
1518 			case NETDEV_DOWN:
1519 			case NETDEV_UNREGISTER:
1520 				nh_flags |= RTNH_F_DEAD;
1521 				fallthrough;
1522 			case NETDEV_CHANGE:
1523 				nh_flags |= RTNH_F_LINKDOWN;
1524 				break;
1525 			}
1526 			if (event == NETDEV_UNREGISTER)
1527 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1528 
1529 			if (nh->nh_flags != nh_flags)
1530 				WRITE_ONCE(nh->nh_flags, nh_flags);
1531 next:
1532 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1533 				alive++;
1534 			if (!rtnl_dereference(nh->nh_dev))
1535 				deleted++;
1536 		} endfor_nexthops(rt);
1537 
1538 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1539 
1540 		/* if there are no more nexthops, delete the route */
1541 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1542 			mpls_route_update(net, index, NULL, NULL);
1543 	}
1544 }
1545 
1546 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1547 {
1548 	struct mpls_route __rcu **platform_label;
1549 	struct net *net = dev_net(dev);
1550 	unsigned index;
1551 	u8 alive;
1552 
1553 	platform_label = rtnl_dereference(net->mpls.platform_label);
1554 	for (index = 0; index < net->mpls.platform_labels; index++) {
1555 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1556 
1557 		if (!rt)
1558 			continue;
1559 
1560 		alive = 0;
1561 		change_nexthops(rt) {
1562 			unsigned int nh_flags = nh->nh_flags;
1563 			struct net_device *nh_dev =
1564 				rtnl_dereference(nh->nh_dev);
1565 
1566 			if (!(nh_flags & flags)) {
1567 				alive++;
1568 				continue;
1569 			}
1570 			if (nh_dev != dev)
1571 				continue;
1572 			alive++;
1573 			nh_flags &= ~flags;
1574 			WRITE_ONCE(nh->nh_flags, nh_flags);
1575 		} endfor_nexthops(rt);
1576 
1577 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1578 	}
1579 }
1580 
1581 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1582 			   void *ptr)
1583 {
1584 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1585 	struct mpls_dev *mdev;
1586 	unsigned int flags;
1587 
1588 	if (event == NETDEV_REGISTER) {
1589 		mdev = mpls_add_dev(dev);
1590 		if (IS_ERR(mdev))
1591 			return notifier_from_errno(PTR_ERR(mdev));
1592 
1593 		return NOTIFY_OK;
1594 	}
1595 
1596 	mdev = mpls_dev_get(dev);
1597 	if (!mdev)
1598 		return NOTIFY_OK;
1599 
1600 	switch (event) {
1601 	case NETDEV_DOWN:
1602 		mpls_ifdown(dev, event);
1603 		break;
1604 	case NETDEV_UP:
1605 		flags = dev_get_flags(dev);
1606 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1607 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1608 		else
1609 			mpls_ifup(dev, RTNH_F_DEAD);
1610 		break;
1611 	case NETDEV_CHANGE:
1612 		flags = dev_get_flags(dev);
1613 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1614 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1615 		else
1616 			mpls_ifdown(dev, event);
1617 		break;
1618 	case NETDEV_UNREGISTER:
1619 		mpls_ifdown(dev, event);
1620 		mdev = mpls_dev_get(dev);
1621 		if (mdev) {
1622 			mpls_dev_sysctl_unregister(dev, mdev);
1623 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1624 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1625 		}
1626 		break;
1627 	case NETDEV_CHANGENAME:
1628 		mdev = mpls_dev_get(dev);
1629 		if (mdev) {
1630 			int err;
1631 
1632 			mpls_dev_sysctl_unregister(dev, mdev);
1633 			err = mpls_dev_sysctl_register(dev, mdev);
1634 			if (err)
1635 				return notifier_from_errno(err);
1636 		}
1637 		break;
1638 	}
1639 	return NOTIFY_OK;
1640 }
1641 
1642 static struct notifier_block mpls_dev_notifier = {
1643 	.notifier_call = mpls_dev_notify,
1644 };
1645 
1646 static int nla_put_via(struct sk_buff *skb,
1647 		       u8 table, const void *addr, int alen)
1648 {
1649 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1650 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1651 	};
1652 	struct nlattr *nla;
1653 	struct rtvia *via;
1654 	int family = AF_UNSPEC;
1655 
1656 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1657 	if (!nla)
1658 		return -EMSGSIZE;
1659 
1660 	if (table <= NEIGH_NR_TABLES)
1661 		family = table_to_family[table];
1662 
1663 	via = nla_data(nla);
1664 	via->rtvia_family = family;
1665 	memcpy(via->rtvia_addr, addr, alen);
1666 	return 0;
1667 }
1668 
1669 int nla_put_labels(struct sk_buff *skb, int attrtype,
1670 		   u8 labels, const u32 label[])
1671 {
1672 	struct nlattr *nla;
1673 	struct mpls_shim_hdr *nla_label;
1674 	bool bos;
1675 	int i;
1676 	nla = nla_reserve(skb, attrtype, labels*4);
1677 	if (!nla)
1678 		return -EMSGSIZE;
1679 
1680 	nla_label = nla_data(nla);
1681 	bos = true;
1682 	for (i = labels - 1; i >= 0; i--) {
1683 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1684 		bos = false;
1685 	}
1686 
1687 	return 0;
1688 }
1689 EXPORT_SYMBOL_GPL(nla_put_labels);
1690 
1691 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1692 		   u32 label[], struct netlink_ext_ack *extack)
1693 {
1694 	unsigned len = nla_len(nla);
1695 	struct mpls_shim_hdr *nla_label;
1696 	u8 nla_labels;
1697 	bool bos;
1698 	int i;
1699 
1700 	/* len needs to be an even multiple of 4 (the label size). Number
1701 	 * of labels is a u8 so check for overflow.
1702 	 */
1703 	if (len & 3 || len / 4 > 255) {
1704 		NL_SET_ERR_MSG_ATTR(extack, nla,
1705 				    "Invalid length for labels attribute");
1706 		return -EINVAL;
1707 	}
1708 
1709 	/* Limit the number of new labels allowed */
1710 	nla_labels = len/4;
1711 	if (nla_labels > max_labels) {
1712 		NL_SET_ERR_MSG(extack, "Too many labels");
1713 		return -EINVAL;
1714 	}
1715 
1716 	/* when label == NULL, caller wants number of labels */
1717 	if (!label)
1718 		goto out;
1719 
1720 	nla_label = nla_data(nla);
1721 	bos = true;
1722 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1723 		struct mpls_entry_decoded dec;
1724 		dec = mpls_entry_decode(nla_label + i);
1725 
1726 		/* Ensure the bottom of stack flag is properly set
1727 		 * and ttl and tc are both clear.
1728 		 */
1729 		if (dec.ttl) {
1730 			NL_SET_ERR_MSG_ATTR(extack, nla,
1731 					    "TTL in label must be 0");
1732 			return -EINVAL;
1733 		}
1734 
1735 		if (dec.tc) {
1736 			NL_SET_ERR_MSG_ATTR(extack, nla,
1737 					    "Traffic class in label must be 0");
1738 			return -EINVAL;
1739 		}
1740 
1741 		if (dec.bos != bos) {
1742 			NL_SET_BAD_ATTR(extack, nla);
1743 			if (bos) {
1744 				NL_SET_ERR_MSG(extack,
1745 					       "BOS bit must be set in first label");
1746 			} else {
1747 				NL_SET_ERR_MSG(extack,
1748 					       "BOS bit can only be set in first label");
1749 			}
1750 			return -EINVAL;
1751 		}
1752 
1753 		switch (dec.label) {
1754 		case MPLS_LABEL_IMPLNULL:
1755 			/* RFC3032: This is a label that an LSR may
1756 			 * assign and distribute, but which never
1757 			 * actually appears in the encapsulation.
1758 			 */
1759 			NL_SET_ERR_MSG_ATTR(extack, nla,
1760 					    "Implicit NULL Label (3) can not be used in encapsulation");
1761 			return -EINVAL;
1762 		}
1763 
1764 		label[i] = dec.label;
1765 	}
1766 out:
1767 	*labels = nla_labels;
1768 	return 0;
1769 }
1770 EXPORT_SYMBOL_GPL(nla_get_labels);
1771 
1772 static int rtm_to_route_config(struct sk_buff *skb,
1773 			       struct nlmsghdr *nlh,
1774 			       struct mpls_route_config *cfg,
1775 			       struct netlink_ext_ack *extack)
1776 {
1777 	struct rtmsg *rtm;
1778 	struct nlattr *tb[RTA_MAX+1];
1779 	int index;
1780 	int err;
1781 
1782 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1783 				     rtm_mpls_policy, extack);
1784 	if (err < 0)
1785 		goto errout;
1786 
1787 	err = -EINVAL;
1788 	rtm = nlmsg_data(nlh);
1789 
1790 	if (rtm->rtm_family != AF_MPLS) {
1791 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1792 		goto errout;
1793 	}
1794 	if (rtm->rtm_dst_len != 20) {
1795 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1796 		goto errout;
1797 	}
1798 	if (rtm->rtm_src_len != 0) {
1799 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1800 		goto errout;
1801 	}
1802 	if (rtm->rtm_tos != 0) {
1803 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1804 		goto errout;
1805 	}
1806 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1807 		NL_SET_ERR_MSG(extack,
1808 			       "MPLS only supports the main route table");
1809 		goto errout;
1810 	}
1811 	/* Any value is acceptable for rtm_protocol */
1812 
1813 	/* As mpls uses destination specific addresses
1814 	 * (or source specific address in the case of multicast)
1815 	 * all addresses have universal scope.
1816 	 */
1817 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1818 		NL_SET_ERR_MSG(extack,
1819 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1820 		goto errout;
1821 	}
1822 	if (rtm->rtm_type != RTN_UNICAST) {
1823 		NL_SET_ERR_MSG(extack,
1824 			       "Invalid route type - MPLS only supports UNICAST");
1825 		goto errout;
1826 	}
1827 	if (rtm->rtm_flags != 0) {
1828 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1829 		goto errout;
1830 	}
1831 
1832 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1833 	cfg->rc_protocol	= rtm->rtm_protocol;
1834 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1835 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1836 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1837 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1838 	cfg->rc_nlinfo.nlh	= nlh;
1839 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1840 
1841 	for (index = 0; index <= RTA_MAX; index++) {
1842 		struct nlattr *nla = tb[index];
1843 		if (!nla)
1844 			continue;
1845 
1846 		switch (index) {
1847 		case RTA_OIF:
1848 			cfg->rc_ifindex = nla_get_u32(nla);
1849 			break;
1850 		case RTA_NEWDST:
1851 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1852 					   &cfg->rc_output_labels,
1853 					   cfg->rc_output_label, extack))
1854 				goto errout;
1855 			break;
1856 		case RTA_DST:
1857 		{
1858 			u8 label_count;
1859 			if (nla_get_labels(nla, 1, &label_count,
1860 					   &cfg->rc_label, extack))
1861 				goto errout;
1862 
1863 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1864 					   &cfg->rc_label, extack))
1865 				goto errout;
1866 			break;
1867 		}
1868 		case RTA_GATEWAY:
1869 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1870 			goto errout;
1871 		case RTA_VIA:
1872 		{
1873 			if (nla_get_via(nla, &cfg->rc_via_alen,
1874 					&cfg->rc_via_table, cfg->rc_via,
1875 					extack))
1876 				goto errout;
1877 			break;
1878 		}
1879 		case RTA_MULTIPATH:
1880 		{
1881 			cfg->rc_mp = nla_data(nla);
1882 			cfg->rc_mp_len = nla_len(nla);
1883 			break;
1884 		}
1885 		case RTA_TTL_PROPAGATE:
1886 		{
1887 			u8 ttl_propagate = nla_get_u8(nla);
1888 
1889 			if (ttl_propagate > 1) {
1890 				NL_SET_ERR_MSG_ATTR(extack, nla,
1891 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1892 				goto errout;
1893 			}
1894 			cfg->rc_ttl_propagate = ttl_propagate ?
1895 				MPLS_TTL_PROP_ENABLED :
1896 				MPLS_TTL_PROP_DISABLED;
1897 			break;
1898 		}
1899 		default:
1900 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1901 			/* Unsupported attribute */
1902 			goto errout;
1903 		}
1904 	}
1905 
1906 	err = 0;
1907 errout:
1908 	return err;
1909 }
1910 
1911 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1912 			     struct netlink_ext_ack *extack)
1913 {
1914 	struct mpls_route_config *cfg;
1915 	int err;
1916 
1917 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1918 	if (!cfg)
1919 		return -ENOMEM;
1920 
1921 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1922 	if (err < 0)
1923 		goto out;
1924 
1925 	err = mpls_route_del(cfg, extack);
1926 out:
1927 	kfree(cfg);
1928 
1929 	return err;
1930 }
1931 
1932 
1933 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1934 			     struct netlink_ext_ack *extack)
1935 {
1936 	struct mpls_route_config *cfg;
1937 	int err;
1938 
1939 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1940 	if (!cfg)
1941 		return -ENOMEM;
1942 
1943 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1944 	if (err < 0)
1945 		goto out;
1946 
1947 	err = mpls_route_add(cfg, extack);
1948 out:
1949 	kfree(cfg);
1950 
1951 	return err;
1952 }
1953 
1954 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1955 			   u32 label, struct mpls_route *rt, int flags)
1956 {
1957 	struct net_device *dev;
1958 	struct nlmsghdr *nlh;
1959 	struct rtmsg *rtm;
1960 
1961 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1962 	if (nlh == NULL)
1963 		return -EMSGSIZE;
1964 
1965 	rtm = nlmsg_data(nlh);
1966 	rtm->rtm_family = AF_MPLS;
1967 	rtm->rtm_dst_len = 20;
1968 	rtm->rtm_src_len = 0;
1969 	rtm->rtm_tos = 0;
1970 	rtm->rtm_table = RT_TABLE_MAIN;
1971 	rtm->rtm_protocol = rt->rt_protocol;
1972 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1973 	rtm->rtm_type = RTN_UNICAST;
1974 	rtm->rtm_flags = 0;
1975 
1976 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1977 		goto nla_put_failure;
1978 
1979 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1980 		bool ttl_propagate =
1981 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1982 
1983 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1984 			       ttl_propagate))
1985 			goto nla_put_failure;
1986 	}
1987 	if (rt->rt_nhn == 1) {
1988 		const struct mpls_nh *nh = rt->rt_nh;
1989 
1990 		if (nh->nh_labels &&
1991 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1992 				   nh->nh_label))
1993 			goto nla_put_failure;
1994 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1995 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1996 				nh->nh_via_alen))
1997 			goto nla_put_failure;
1998 		dev = rtnl_dereference(nh->nh_dev);
1999 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2000 			goto nla_put_failure;
2001 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2002 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2003 		if (nh->nh_flags & RTNH_F_DEAD)
2004 			rtm->rtm_flags |= RTNH_F_DEAD;
2005 	} else {
2006 		struct rtnexthop *rtnh;
2007 		struct nlattr *mp;
2008 		u8 linkdown = 0;
2009 		u8 dead = 0;
2010 
2011 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2012 		if (!mp)
2013 			goto nla_put_failure;
2014 
2015 		for_nexthops(rt) {
2016 			dev = rtnl_dereference(nh->nh_dev);
2017 			if (!dev)
2018 				continue;
2019 
2020 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2021 			if (!rtnh)
2022 				goto nla_put_failure;
2023 
2024 			rtnh->rtnh_ifindex = dev->ifindex;
2025 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2026 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2027 				linkdown++;
2028 			}
2029 			if (nh->nh_flags & RTNH_F_DEAD) {
2030 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2031 				dead++;
2032 			}
2033 
2034 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2035 							    nh->nh_labels,
2036 							    nh->nh_label))
2037 				goto nla_put_failure;
2038 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2039 			    nla_put_via(skb, nh->nh_via_table,
2040 					mpls_nh_via(rt, nh),
2041 					nh->nh_via_alen))
2042 				goto nla_put_failure;
2043 
2044 			/* length of rtnetlink header + attributes */
2045 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2046 		} endfor_nexthops(rt);
2047 
2048 		if (linkdown == rt->rt_nhn)
2049 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2050 		if (dead == rt->rt_nhn)
2051 			rtm->rtm_flags |= RTNH_F_DEAD;
2052 
2053 		nla_nest_end(skb, mp);
2054 	}
2055 
2056 	nlmsg_end(skb, nlh);
2057 	return 0;
2058 
2059 nla_put_failure:
2060 	nlmsg_cancel(skb, nlh);
2061 	return -EMSGSIZE;
2062 }
2063 
2064 #if IS_ENABLED(CONFIG_INET)
2065 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2066 				   struct fib_dump_filter *filter,
2067 				   struct netlink_callback *cb)
2068 {
2069 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2070 }
2071 #else
2072 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2073 				   struct fib_dump_filter *filter,
2074 				   struct netlink_callback *cb)
2075 {
2076 	struct netlink_ext_ack *extack = cb->extack;
2077 	struct nlattr *tb[RTA_MAX + 1];
2078 	struct rtmsg *rtm;
2079 	int err, i;
2080 
2081 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2082 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2083 		return -EINVAL;
2084 	}
2085 
2086 	rtm = nlmsg_data(nlh);
2087 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2088 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2089 	    rtm->rtm_flags) {
2090 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2091 		return -EINVAL;
2092 	}
2093 
2094 	if (rtm->rtm_protocol) {
2095 		filter->protocol = rtm->rtm_protocol;
2096 		filter->filter_set = 1;
2097 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2098 	}
2099 
2100 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2101 					    rtm_mpls_policy, extack);
2102 	if (err < 0)
2103 		return err;
2104 
2105 	for (i = 0; i <= RTA_MAX; ++i) {
2106 		int ifindex;
2107 
2108 		if (i == RTA_OIF) {
2109 			ifindex = nla_get_u32(tb[i]);
2110 			filter->dev = __dev_get_by_index(net, ifindex);
2111 			if (!filter->dev)
2112 				return -ENODEV;
2113 			filter->filter_set = 1;
2114 		} else if (tb[i]) {
2115 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2116 			return -EINVAL;
2117 		}
2118 	}
2119 
2120 	return 0;
2121 }
2122 #endif
2123 
2124 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2125 			     const struct net_device *dev)
2126 {
2127 	struct net_device *nh_dev;
2128 
2129 	if (rt->rt_nhn == 1) {
2130 		struct mpls_nh *nh = rt->rt_nh;
2131 
2132 		nh_dev = rtnl_dereference(nh->nh_dev);
2133 		if (dev == nh_dev)
2134 			return true;
2135 	} else {
2136 		for_nexthops(rt) {
2137 			nh_dev = rtnl_dereference(nh->nh_dev);
2138 			if (nh_dev == dev)
2139 				return true;
2140 		} endfor_nexthops(rt);
2141 	}
2142 
2143 	return false;
2144 }
2145 
2146 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2147 {
2148 	const struct nlmsghdr *nlh = cb->nlh;
2149 	struct net *net = sock_net(skb->sk);
2150 	struct mpls_route __rcu **platform_label;
2151 	struct fib_dump_filter filter = {};
2152 	unsigned int flags = NLM_F_MULTI;
2153 	size_t platform_labels;
2154 	unsigned int index;
2155 
2156 	ASSERT_RTNL();
2157 
2158 	if (cb->strict_check) {
2159 		int err;
2160 
2161 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2162 		if (err < 0)
2163 			return err;
2164 
2165 		/* for MPLS, there is only 1 table with fixed type and flags.
2166 		 * If either are set in the filter then return nothing.
2167 		 */
2168 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2169 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2170 		     filter.flags)
2171 			return skb->len;
2172 	}
2173 
2174 	index = cb->args[0];
2175 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2176 		index = MPLS_LABEL_FIRST_UNRESERVED;
2177 
2178 	platform_label = rtnl_dereference(net->mpls.platform_label);
2179 	platform_labels = net->mpls.platform_labels;
2180 
2181 	if (filter.filter_set)
2182 		flags |= NLM_F_DUMP_FILTERED;
2183 
2184 	for (; index < platform_labels; index++) {
2185 		struct mpls_route *rt;
2186 
2187 		rt = rtnl_dereference(platform_label[index]);
2188 		if (!rt)
2189 			continue;
2190 
2191 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2192 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2193 			continue;
2194 
2195 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2196 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2197 				    index, rt, flags) < 0)
2198 			break;
2199 	}
2200 	cb->args[0] = index;
2201 
2202 	return skb->len;
2203 }
2204 
2205 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2206 {
2207 	size_t payload =
2208 		NLMSG_ALIGN(sizeof(struct rtmsg))
2209 		+ nla_total_size(4)			/* RTA_DST */
2210 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2211 
2212 	if (rt->rt_nhn == 1) {
2213 		struct mpls_nh *nh = rt->rt_nh;
2214 
2215 		if (nh->nh_dev)
2216 			payload += nla_total_size(4); /* RTA_OIF */
2217 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2218 			payload += nla_total_size(2 + nh->nh_via_alen);
2219 		if (nh->nh_labels) /* RTA_NEWDST */
2220 			payload += nla_total_size(nh->nh_labels * 4);
2221 	} else {
2222 		/* each nexthop is packed in an attribute */
2223 		size_t nhsize = 0;
2224 
2225 		for_nexthops(rt) {
2226 			if (!rtnl_dereference(nh->nh_dev))
2227 				continue;
2228 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2229 			/* RTA_VIA */
2230 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2231 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2232 			if (nh->nh_labels)
2233 				nhsize += nla_total_size(nh->nh_labels * 4);
2234 		} endfor_nexthops(rt);
2235 		/* nested attribute */
2236 		payload += nla_total_size(nhsize);
2237 	}
2238 
2239 	return payload;
2240 }
2241 
2242 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2243 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2244 		       unsigned int nlm_flags)
2245 {
2246 	struct sk_buff *skb;
2247 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2248 	int err = -ENOBUFS;
2249 
2250 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2251 	if (skb == NULL)
2252 		goto errout;
2253 
2254 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2255 	if (err < 0) {
2256 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2257 		WARN_ON(err == -EMSGSIZE);
2258 		kfree_skb(skb);
2259 		goto errout;
2260 	}
2261 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2262 
2263 	return;
2264 errout:
2265 	if (err < 0)
2266 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2267 }
2268 
2269 static int mpls_valid_getroute_req(struct sk_buff *skb,
2270 				   const struct nlmsghdr *nlh,
2271 				   struct nlattr **tb,
2272 				   struct netlink_ext_ack *extack)
2273 {
2274 	struct rtmsg *rtm;
2275 	int i, err;
2276 
2277 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2278 		NL_SET_ERR_MSG_MOD(extack,
2279 				   "Invalid header for get route request");
2280 		return -EINVAL;
2281 	}
2282 
2283 	if (!netlink_strict_get_check(skb))
2284 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2285 					      rtm_mpls_policy, extack);
2286 
2287 	rtm = nlmsg_data(nlh);
2288 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2289 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2290 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2291 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2292 		return -EINVAL;
2293 	}
2294 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2295 		NL_SET_ERR_MSG_MOD(extack,
2296 				   "Invalid flags for get route request");
2297 		return -EINVAL;
2298 	}
2299 
2300 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2301 					    rtm_mpls_policy, extack);
2302 	if (err)
2303 		return err;
2304 
2305 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2306 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2307 		return -EINVAL;
2308 	}
2309 
2310 	for (i = 0; i <= RTA_MAX; i++) {
2311 		if (!tb[i])
2312 			continue;
2313 
2314 		switch (i) {
2315 		case RTA_DST:
2316 		case RTA_NEWDST:
2317 			break;
2318 		default:
2319 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2320 			return -EINVAL;
2321 		}
2322 	}
2323 
2324 	return 0;
2325 }
2326 
2327 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2328 			 struct netlink_ext_ack *extack)
2329 {
2330 	struct net *net = sock_net(in_skb->sk);
2331 	u32 portid = NETLINK_CB(in_skb).portid;
2332 	u32 in_label = LABEL_NOT_SPECIFIED;
2333 	struct nlattr *tb[RTA_MAX + 1];
2334 	u32 labels[MAX_NEW_LABELS];
2335 	struct mpls_shim_hdr *hdr;
2336 	unsigned int hdr_size = 0;
2337 	struct net_device *dev;
2338 	struct mpls_route *rt;
2339 	struct rtmsg *rtm, *r;
2340 	struct nlmsghdr *nlh;
2341 	struct sk_buff *skb;
2342 	struct mpls_nh *nh;
2343 	u8 n_labels;
2344 	int err;
2345 
2346 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2347 	if (err < 0)
2348 		goto errout;
2349 
2350 	rtm = nlmsg_data(in_nlh);
2351 
2352 	if (tb[RTA_DST]) {
2353 		u8 label_count;
2354 
2355 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2356 				   &in_label, extack)) {
2357 			err = -EINVAL;
2358 			goto errout;
2359 		}
2360 
2361 		if (!mpls_label_ok(net, &in_label, extack)) {
2362 			err = -EINVAL;
2363 			goto errout;
2364 		}
2365 	}
2366 
2367 	rt = mpls_route_input_rcu(net, in_label);
2368 	if (!rt) {
2369 		err = -ENETUNREACH;
2370 		goto errout;
2371 	}
2372 
2373 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2374 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2375 		if (!skb) {
2376 			err = -ENOBUFS;
2377 			goto errout;
2378 		}
2379 
2380 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2381 				      RTM_NEWROUTE, in_label, rt, 0);
2382 		if (err < 0) {
2383 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2384 			WARN_ON(err == -EMSGSIZE);
2385 			goto errout_free;
2386 		}
2387 
2388 		return rtnl_unicast(skb, net, portid);
2389 	}
2390 
2391 	if (tb[RTA_NEWDST]) {
2392 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2393 				   labels, extack) != 0) {
2394 			err = -EINVAL;
2395 			goto errout;
2396 		}
2397 
2398 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2399 	}
2400 
2401 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2402 	if (!skb) {
2403 		err = -ENOBUFS;
2404 		goto errout;
2405 	}
2406 
2407 	skb->protocol = htons(ETH_P_MPLS_UC);
2408 
2409 	if (hdr_size) {
2410 		bool bos;
2411 		int i;
2412 
2413 		if (skb_cow(skb, hdr_size)) {
2414 			err = -ENOBUFS;
2415 			goto errout_free;
2416 		}
2417 
2418 		skb_reserve(skb, hdr_size);
2419 		skb_push(skb, hdr_size);
2420 		skb_reset_network_header(skb);
2421 
2422 		/* Push new labels */
2423 		hdr = mpls_hdr(skb);
2424 		bos = true;
2425 		for (i = n_labels - 1; i >= 0; i--) {
2426 			hdr[i] = mpls_entry_encode(labels[i],
2427 						   1, 0, bos);
2428 			bos = false;
2429 		}
2430 	}
2431 
2432 	nh = mpls_select_multipath(rt, skb);
2433 	if (!nh) {
2434 		err = -ENETUNREACH;
2435 		goto errout_free;
2436 	}
2437 
2438 	if (hdr_size) {
2439 		skb_pull(skb, hdr_size);
2440 		skb_reset_network_header(skb);
2441 	}
2442 
2443 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2444 			RTM_NEWROUTE, sizeof(*r), 0);
2445 	if (!nlh) {
2446 		err = -EMSGSIZE;
2447 		goto errout_free;
2448 	}
2449 
2450 	r = nlmsg_data(nlh);
2451 	r->rtm_family	 = AF_MPLS;
2452 	r->rtm_dst_len	= 20;
2453 	r->rtm_src_len	= 0;
2454 	r->rtm_table	= RT_TABLE_MAIN;
2455 	r->rtm_type	= RTN_UNICAST;
2456 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2457 	r->rtm_protocol = rt->rt_protocol;
2458 	r->rtm_flags	= 0;
2459 
2460 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2461 		goto nla_put_failure;
2462 
2463 	if (nh->nh_labels &&
2464 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2465 			   nh->nh_label))
2466 		goto nla_put_failure;
2467 
2468 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2469 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2470 			nh->nh_via_alen))
2471 		goto nla_put_failure;
2472 	dev = rtnl_dereference(nh->nh_dev);
2473 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2474 		goto nla_put_failure;
2475 
2476 	nlmsg_end(skb, nlh);
2477 
2478 	err = rtnl_unicast(skb, net, portid);
2479 errout:
2480 	return err;
2481 
2482 nla_put_failure:
2483 	nlmsg_cancel(skb, nlh);
2484 	err = -EMSGSIZE;
2485 errout_free:
2486 	kfree_skb(skb);
2487 	return err;
2488 }
2489 
2490 static int resize_platform_label_table(struct net *net, size_t limit)
2491 {
2492 	size_t size = sizeof(struct mpls_route *) * limit;
2493 	size_t old_limit;
2494 	size_t cp_size;
2495 	struct mpls_route __rcu **labels = NULL, **old;
2496 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2497 	unsigned index;
2498 
2499 	if (size) {
2500 		labels = kvzalloc(size, GFP_KERNEL);
2501 		if (!labels)
2502 			goto nolabels;
2503 	}
2504 
2505 	/* In case the predefined labels need to be populated */
2506 	if (limit > MPLS_LABEL_IPV4NULL) {
2507 		struct net_device *lo = net->loopback_dev;
2508 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2509 		if (IS_ERR(rt0))
2510 			goto nort0;
2511 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2512 		rt0->rt_protocol = RTPROT_KERNEL;
2513 		rt0->rt_payload_type = MPT_IPV4;
2514 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2515 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2516 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2517 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2518 		       lo->addr_len);
2519 	}
2520 	if (limit > MPLS_LABEL_IPV6NULL) {
2521 		struct net_device *lo = net->loopback_dev;
2522 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2523 		if (IS_ERR(rt2))
2524 			goto nort2;
2525 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2526 		rt2->rt_protocol = RTPROT_KERNEL;
2527 		rt2->rt_payload_type = MPT_IPV6;
2528 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2529 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2530 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2531 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2532 		       lo->addr_len);
2533 	}
2534 
2535 	rtnl_lock();
2536 	/* Remember the original table */
2537 	old = rtnl_dereference(net->mpls.platform_label);
2538 	old_limit = net->mpls.platform_labels;
2539 
2540 	/* Free any labels beyond the new table */
2541 	for (index = limit; index < old_limit; index++)
2542 		mpls_route_update(net, index, NULL, NULL);
2543 
2544 	/* Copy over the old labels */
2545 	cp_size = size;
2546 	if (old_limit < limit)
2547 		cp_size = old_limit * sizeof(struct mpls_route *);
2548 
2549 	memcpy(labels, old, cp_size);
2550 
2551 	/* If needed set the predefined labels */
2552 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2553 	    (limit > MPLS_LABEL_IPV6NULL)) {
2554 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2555 		rt2 = NULL;
2556 	}
2557 
2558 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2559 	    (limit > MPLS_LABEL_IPV4NULL)) {
2560 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2561 		rt0 = NULL;
2562 	}
2563 
2564 	/* Update the global pointers */
2565 	net->mpls.platform_labels = limit;
2566 	rcu_assign_pointer(net->mpls.platform_label, labels);
2567 
2568 	rtnl_unlock();
2569 
2570 	mpls_rt_free(rt2);
2571 	mpls_rt_free(rt0);
2572 
2573 	if (old) {
2574 		synchronize_rcu();
2575 		kvfree(old);
2576 	}
2577 	return 0;
2578 
2579 nort2:
2580 	mpls_rt_free(rt0);
2581 nort0:
2582 	kvfree(labels);
2583 nolabels:
2584 	return -ENOMEM;
2585 }
2586 
2587 static int mpls_platform_labels(struct ctl_table *table, int write,
2588 				void *buffer, size_t *lenp, loff_t *ppos)
2589 {
2590 	struct net *net = table->data;
2591 	int platform_labels = net->mpls.platform_labels;
2592 	int ret;
2593 	struct ctl_table tmp = {
2594 		.procname	= table->procname,
2595 		.data		= &platform_labels,
2596 		.maxlen		= sizeof(int),
2597 		.mode		= table->mode,
2598 		.extra1		= SYSCTL_ZERO,
2599 		.extra2		= &label_limit,
2600 	};
2601 
2602 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2603 
2604 	if (write && ret == 0)
2605 		ret = resize_platform_label_table(net, platform_labels);
2606 
2607 	return ret;
2608 }
2609 
2610 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2611 	(&((struct net *)0)->field)
2612 
2613 static const struct ctl_table mpls_table[] = {
2614 	{
2615 		.procname	= "platform_labels",
2616 		.data		= NULL,
2617 		.maxlen		= sizeof(int),
2618 		.mode		= 0644,
2619 		.proc_handler	= mpls_platform_labels,
2620 	},
2621 	{
2622 		.procname	= "ip_ttl_propagate",
2623 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2624 		.maxlen		= sizeof(int),
2625 		.mode		= 0644,
2626 		.proc_handler	= proc_dointvec_minmax,
2627 		.extra1		= SYSCTL_ZERO,
2628 		.extra2		= SYSCTL_ONE,
2629 	},
2630 	{
2631 		.procname	= "default_ttl",
2632 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2633 		.maxlen		= sizeof(int),
2634 		.mode		= 0644,
2635 		.proc_handler	= proc_dointvec_minmax,
2636 		.extra1		= SYSCTL_ONE,
2637 		.extra2		= &ttl_max,
2638 	},
2639 	{ }
2640 };
2641 
2642 static int mpls_net_init(struct net *net)
2643 {
2644 	struct ctl_table *table;
2645 	int i;
2646 
2647 	net->mpls.platform_labels = 0;
2648 	net->mpls.platform_label = NULL;
2649 	net->mpls.ip_ttl_propagate = 1;
2650 	net->mpls.default_ttl = 255;
2651 
2652 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2653 	if (table == NULL)
2654 		return -ENOMEM;
2655 
2656 	/* Table data contains only offsets relative to the base of
2657 	 * the mdev at this point, so make them absolute.
2658 	 */
2659 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2660 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2661 
2662 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2663 	if (net->mpls.ctl == NULL) {
2664 		kfree(table);
2665 		return -ENOMEM;
2666 	}
2667 
2668 	return 0;
2669 }
2670 
2671 static void mpls_net_exit(struct net *net)
2672 {
2673 	struct mpls_route __rcu **platform_label;
2674 	size_t platform_labels;
2675 	struct ctl_table *table;
2676 	unsigned int index;
2677 
2678 	table = net->mpls.ctl->ctl_table_arg;
2679 	unregister_net_sysctl_table(net->mpls.ctl);
2680 	kfree(table);
2681 
2682 	/* An rcu grace period has passed since there was a device in
2683 	 * the network namespace (and thus the last in flight packet)
2684 	 * left this network namespace.  This is because
2685 	 * unregister_netdevice_many and netdev_run_todo has completed
2686 	 * for each network device that was in this network namespace.
2687 	 *
2688 	 * As such no additional rcu synchronization is necessary when
2689 	 * freeing the platform_label table.
2690 	 */
2691 	rtnl_lock();
2692 	platform_label = rtnl_dereference(net->mpls.platform_label);
2693 	platform_labels = net->mpls.platform_labels;
2694 	for (index = 0; index < platform_labels; index++) {
2695 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2696 		RCU_INIT_POINTER(platform_label[index], NULL);
2697 		mpls_notify_route(net, index, rt, NULL, NULL);
2698 		mpls_rt_free(rt);
2699 	}
2700 	rtnl_unlock();
2701 
2702 	kvfree(platform_label);
2703 }
2704 
2705 static struct pernet_operations mpls_net_ops = {
2706 	.init = mpls_net_init,
2707 	.exit = mpls_net_exit,
2708 };
2709 
2710 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2711 	.family		   = AF_MPLS,
2712 	.fill_stats_af	   = mpls_fill_stats_af,
2713 	.get_stats_af_size = mpls_get_stats_af_size,
2714 };
2715 
2716 static int __init mpls_init(void)
2717 {
2718 	int err;
2719 
2720 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2721 
2722 	err = register_pernet_subsys(&mpls_net_ops);
2723 	if (err)
2724 		goto out;
2725 
2726 	err = register_netdevice_notifier(&mpls_dev_notifier);
2727 	if (err)
2728 		goto out_unregister_pernet;
2729 
2730 	dev_add_pack(&mpls_packet_type);
2731 
2732 	rtnl_af_register(&mpls_af_ops);
2733 
2734 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2735 			     mpls_rtm_newroute, NULL, 0);
2736 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2737 			     mpls_rtm_delroute, NULL, 0);
2738 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2739 			     mpls_getroute, mpls_dump_routes, 0);
2740 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2741 			     mpls_netconf_get_devconf,
2742 			     mpls_netconf_dump_devconf, 0);
2743 	err = ipgre_tunnel_encap_add_mpls_ops();
2744 	if (err)
2745 		pr_err("Can't add mpls over gre tunnel ops\n");
2746 
2747 	err = 0;
2748 out:
2749 	return err;
2750 
2751 out_unregister_pernet:
2752 	unregister_pernet_subsys(&mpls_net_ops);
2753 	goto out;
2754 }
2755 module_init(mpls_init);
2756 
2757 static void __exit mpls_exit(void)
2758 {
2759 	rtnl_unregister_all(PF_MPLS);
2760 	rtnl_af_unregister(&mpls_af_ops);
2761 	dev_remove_pack(&mpls_packet_type);
2762 	unregister_netdevice_notifier(&mpls_dev_notifier);
2763 	unregister_pernet_subsys(&mpls_net_ops);
2764 	ipgre_tunnel_encap_del_mpls_ops();
2765 }
2766 module_exit(mpls_exit);
2767 
2768 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2769 MODULE_LICENSE("GPL v2");
2770 MODULE_ALIAS_NETPROTO(PF_MPLS);
2771