xref: /openbmc/linux/net/mpls/af_mpls.c (revision c899710f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/types.h>
3 #include <linux/skbuff.h>
4 #include <linux/socket.h>
5 #include <linux/sysctl.h>
6 #include <linux/net.h>
7 #include <linux/module.h>
8 #include <linux/if_arp.h>
9 #include <linux/ipv6.h>
10 #include <linux/mpls.h>
11 #include <linux/netconf.h>
12 #include <linux/nospec.h>
13 #include <linux/vmalloc.h>
14 #include <linux/percpu.h>
15 #include <net/gso.h>
16 #include <net/ip.h>
17 #include <net/dst.h>
18 #include <net/sock.h>
19 #include <net/arp.h>
20 #include <net/ip_fib.h>
21 #include <net/netevent.h>
22 #include <net/ip_tunnels.h>
23 #include <net/netns/generic.h>
24 #if IS_ENABLED(CONFIG_IPV6)
25 #include <net/ipv6.h>
26 #endif
27 #include <net/ipv6_stubs.h>
28 #include <net/rtnh.h>
29 #include "internal.h"
30 
31 /* max memory we will use for mpls_route */
32 #define MAX_MPLS_ROUTE_MEM	4096
33 
34 /* Maximum number of labels to look ahead at when selecting a path of
35  * a multipath route
36  */
37 #define MAX_MP_SELECT_LABELS 4
38 
39 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
40 
41 static int label_limit = (1 << 20) - 1;
42 static int ttl_max = 255;
43 
44 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
45 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
46 {
47 	return sizeof(struct mpls_shim_hdr);
48 }
49 
50 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
51 	.encap_hlen	= ipgre_mpls_encap_hlen,
52 };
53 
54 static int ipgre_tunnel_encap_add_mpls_ops(void)
55 {
56 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
57 }
58 
59 static void ipgre_tunnel_encap_del_mpls_ops(void)
60 {
61 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
62 }
63 #else
64 static int ipgre_tunnel_encap_add_mpls_ops(void)
65 {
66 	return 0;
67 }
68 
69 static void ipgre_tunnel_encap_del_mpls_ops(void)
70 {
71 }
72 #endif
73 
74 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
75 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
76 		       unsigned int nlm_flags);
77 
78 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
79 {
80 	struct mpls_route *rt = NULL;
81 
82 	if (index < net->mpls.platform_labels) {
83 		struct mpls_route __rcu **platform_label =
84 			rcu_dereference(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 const 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 	const struct mpls_nh *nh;
348 	struct mpls_route *rt;
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 	skb_dst_drop(skb);
382 
383 	/* Read and decode the label */
384 	hdr = mpls_hdr(skb);
385 	dec = mpls_entry_decode(hdr);
386 
387 	rt = mpls_route_input_rcu(net, dec.label);
388 	if (!rt) {
389 		MPLS_INC_STATS(mdev, rx_noroute);
390 		goto drop;
391 	}
392 
393 	nh = mpls_select_multipath(rt, skb);
394 	if (!nh)
395 		goto err;
396 
397 	/* Pop the label */
398 	skb_pull(skb, sizeof(*hdr));
399 	skb_reset_network_header(skb);
400 
401 	skb_orphan(skb);
402 
403 	if (skb_warn_if_lro(skb))
404 		goto err;
405 
406 	skb_forward_csum(skb);
407 
408 	/* Verify ttl is valid */
409 	if (dec.ttl <= 1)
410 		goto err;
411 
412 	/* Find the output device */
413 	out_dev = 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 	dec.ttl -= 1;
435 	if (unlikely(!new_header_size && dec.bos)) {
436 		/* Penultimate hop popping */
437 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
438 			goto err;
439 	} else {
440 		bool bos;
441 		int i;
442 		skb_push(skb, new_header_size);
443 		skb_reset_network_header(skb);
444 		/* Push the new labels */
445 		hdr = mpls_hdr(skb);
446 		bos = dec.bos;
447 		for (i = nh->nh_labels - 1; i >= 0; i--) {
448 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
449 						   dec.ttl, 0, bos);
450 			bos = false;
451 		}
452 	}
453 
454 	mpls_stats_inc_outucastpkts(out_dev, skb);
455 
456 	/* If via wasn't specified then send out using device address */
457 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
458 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
459 				 out_dev->dev_addr, skb);
460 	else
461 		err = neigh_xmit(nh->nh_via_table, out_dev,
462 				 mpls_nh_via(rt, nh), skb);
463 	if (err)
464 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
465 				    __func__, err);
466 	return 0;
467 
468 tx_err:
469 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
470 	if (out_mdev)
471 		MPLS_INC_STATS(out_mdev, tx_errors);
472 	goto drop;
473 err:
474 	MPLS_INC_STATS(mdev, rx_errors);
475 drop:
476 	kfree_skb(skb);
477 	return NET_RX_DROP;
478 }
479 
480 static struct packet_type mpls_packet_type __read_mostly = {
481 	.type = cpu_to_be16(ETH_P_MPLS_UC),
482 	.func = mpls_forward,
483 };
484 
485 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
486 	[RTA_DST]		= { .type = NLA_U32 },
487 	[RTA_OIF]		= { .type = NLA_U32 },
488 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
489 };
490 
491 struct mpls_route_config {
492 	u32			rc_protocol;
493 	u32			rc_ifindex;
494 	u8			rc_via_table;
495 	u8			rc_via_alen;
496 	u8			rc_via[MAX_VIA_ALEN];
497 	u32			rc_label;
498 	u8			rc_ttl_propagate;
499 	u8			rc_output_labels;
500 	u32			rc_output_label[MAX_NEW_LABELS];
501 	u32			rc_nlflags;
502 	enum mpls_payload_type	rc_payload_type;
503 	struct nl_info		rc_nlinfo;
504 	struct rtnexthop	*rc_mp;
505 	int			rc_mp_len;
506 };
507 
508 /* all nexthops within a route have the same size based on max
509  * number of labels and max via length for a hop
510  */
511 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
512 {
513 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
514 	struct mpls_route *rt;
515 	size_t size;
516 
517 	size = sizeof(*rt) + num_nh * nh_size;
518 	if (size > MAX_MPLS_ROUTE_MEM)
519 		return ERR_PTR(-EINVAL);
520 
521 	rt = kzalloc(size, GFP_KERNEL);
522 	if (!rt)
523 		return ERR_PTR(-ENOMEM);
524 
525 	rt->rt_nhn = num_nh;
526 	rt->rt_nhn_alive = num_nh;
527 	rt->rt_nh_size = nh_size;
528 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
529 
530 	return rt;
531 }
532 
533 static void mpls_rt_free(struct mpls_route *rt)
534 {
535 	if (rt)
536 		kfree_rcu(rt, rt_rcu);
537 }
538 
539 static void mpls_notify_route(struct net *net, unsigned index,
540 			      struct mpls_route *old, struct mpls_route *new,
541 			      const struct nl_info *info)
542 {
543 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
544 	unsigned portid = info ? info->portid : 0;
545 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
546 	struct mpls_route *rt = new ? new : old;
547 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
548 	/* Ignore reserved labels for now */
549 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
550 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
551 }
552 
553 static void mpls_route_update(struct net *net, unsigned index,
554 			      struct mpls_route *new,
555 			      const struct nl_info *info)
556 {
557 	struct mpls_route __rcu **platform_label;
558 	struct mpls_route *rt;
559 
560 	ASSERT_RTNL();
561 
562 	platform_label = rtnl_dereference(net->mpls.platform_label);
563 	rt = rtnl_dereference(platform_label[index]);
564 	rcu_assign_pointer(platform_label[index], new);
565 
566 	mpls_notify_route(net, index, rt, new, info);
567 
568 	/* If we removed a route free it now */
569 	mpls_rt_free(rt);
570 }
571 
572 static unsigned find_free_label(struct net *net)
573 {
574 	struct mpls_route __rcu **platform_label;
575 	size_t platform_labels;
576 	unsigned index;
577 
578 	platform_label = rtnl_dereference(net->mpls.platform_label);
579 	platform_labels = net->mpls.platform_labels;
580 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
581 	     index++) {
582 		if (!rtnl_dereference(platform_label[index]))
583 			return index;
584 	}
585 	return LABEL_NOT_SPECIFIED;
586 }
587 
588 #if IS_ENABLED(CONFIG_INET)
589 static struct net_device *inet_fib_lookup_dev(struct net *net,
590 					      const void *addr)
591 {
592 	struct net_device *dev;
593 	struct rtable *rt;
594 	struct in_addr daddr;
595 
596 	memcpy(&daddr, addr, sizeof(struct in_addr));
597 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
598 	if (IS_ERR(rt))
599 		return ERR_CAST(rt);
600 
601 	dev = rt->dst.dev;
602 	dev_hold(dev);
603 
604 	ip_rt_put(rt);
605 
606 	return dev;
607 }
608 #else
609 static struct net_device *inet_fib_lookup_dev(struct net *net,
610 					      const void *addr)
611 {
612 	return ERR_PTR(-EAFNOSUPPORT);
613 }
614 #endif
615 
616 #if IS_ENABLED(CONFIG_IPV6)
617 static struct net_device *inet6_fib_lookup_dev(struct net *net,
618 					       const void *addr)
619 {
620 	struct net_device *dev;
621 	struct dst_entry *dst;
622 	struct flowi6 fl6;
623 
624 	if (!ipv6_stub)
625 		return ERR_PTR(-EAFNOSUPPORT);
626 
627 	memset(&fl6, 0, sizeof(fl6));
628 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
629 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
630 	if (IS_ERR(dst))
631 		return ERR_CAST(dst);
632 
633 	dev = dst->dev;
634 	dev_hold(dev);
635 	dst_release(dst);
636 
637 	return dev;
638 }
639 #else
640 static struct net_device *inet6_fib_lookup_dev(struct net *net,
641 					       const void *addr)
642 {
643 	return ERR_PTR(-EAFNOSUPPORT);
644 }
645 #endif
646 
647 static struct net_device *find_outdev(struct net *net,
648 				      struct mpls_route *rt,
649 				      struct mpls_nh *nh, int oif)
650 {
651 	struct net_device *dev = NULL;
652 
653 	if (!oif) {
654 		switch (nh->nh_via_table) {
655 		case NEIGH_ARP_TABLE:
656 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
657 			break;
658 		case NEIGH_ND_TABLE:
659 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
660 			break;
661 		case NEIGH_LINK_TABLE:
662 			break;
663 		}
664 	} else {
665 		dev = dev_get_by_index(net, oif);
666 	}
667 
668 	if (!dev)
669 		return ERR_PTR(-ENODEV);
670 
671 	if (IS_ERR(dev))
672 		return dev;
673 
674 	/* The caller is holding rtnl anyways, so release the dev reference */
675 	dev_put(dev);
676 
677 	return dev;
678 }
679 
680 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
681 			      struct mpls_nh *nh, int oif)
682 {
683 	struct net_device *dev = NULL;
684 	int err = -ENODEV;
685 
686 	dev = find_outdev(net, rt, nh, oif);
687 	if (IS_ERR(dev)) {
688 		err = PTR_ERR(dev);
689 		dev = NULL;
690 		goto errout;
691 	}
692 
693 	/* Ensure this is a supported device */
694 	err = -EINVAL;
695 	if (!mpls_dev_get(dev))
696 		goto errout;
697 
698 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
699 	    (dev->addr_len != nh->nh_via_alen))
700 		goto errout;
701 
702 	nh->nh_dev = dev;
703 
704 	if (!(dev->flags & IFF_UP)) {
705 		nh->nh_flags |= RTNH_F_DEAD;
706 	} else {
707 		unsigned int flags;
708 
709 		flags = dev_get_flags(dev);
710 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
711 			nh->nh_flags |= RTNH_F_LINKDOWN;
712 	}
713 
714 	return 0;
715 
716 errout:
717 	return err;
718 }
719 
720 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
721 		       u8 via_addr[], struct netlink_ext_ack *extack)
722 {
723 	struct rtvia *via = nla_data(nla);
724 	int err = -EINVAL;
725 	int alen;
726 
727 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
728 		NL_SET_ERR_MSG_ATTR(extack, nla,
729 				    "Invalid attribute length for RTA_VIA");
730 		goto errout;
731 	}
732 	alen = nla_len(nla) -
733 			offsetof(struct rtvia, rtvia_addr);
734 	if (alen > MAX_VIA_ALEN) {
735 		NL_SET_ERR_MSG_ATTR(extack, nla,
736 				    "Invalid address length for RTA_VIA");
737 		goto errout;
738 	}
739 
740 	/* Validate the address family */
741 	switch (via->rtvia_family) {
742 	case AF_PACKET:
743 		*via_table = NEIGH_LINK_TABLE;
744 		break;
745 	case AF_INET:
746 		*via_table = NEIGH_ARP_TABLE;
747 		if (alen != 4)
748 			goto errout;
749 		break;
750 	case AF_INET6:
751 		*via_table = NEIGH_ND_TABLE;
752 		if (alen != 16)
753 			goto errout;
754 		break;
755 	default:
756 		/* Unsupported address family */
757 		goto errout;
758 	}
759 
760 	memcpy(via_addr, via->rtvia_addr, alen);
761 	*via_alen = alen;
762 	err = 0;
763 
764 errout:
765 	return err;
766 }
767 
768 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
769 				  struct mpls_route *rt)
770 {
771 	struct net *net = cfg->rc_nlinfo.nl_net;
772 	struct mpls_nh *nh = rt->rt_nh;
773 	int err;
774 	int i;
775 
776 	if (!nh)
777 		return -ENOMEM;
778 
779 	nh->nh_labels = cfg->rc_output_labels;
780 	for (i = 0; i < nh->nh_labels; i++)
781 		nh->nh_label[i] = cfg->rc_output_label[i];
782 
783 	nh->nh_via_table = cfg->rc_via_table;
784 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
785 	nh->nh_via_alen = cfg->rc_via_alen;
786 
787 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
788 	if (err)
789 		goto errout;
790 
791 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
792 		rt->rt_nhn_alive--;
793 
794 	return 0;
795 
796 errout:
797 	return err;
798 }
799 
800 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
801 			 struct mpls_nh *nh, int oif, struct nlattr *via,
802 			 struct nlattr *newdst, u8 max_labels,
803 			 struct netlink_ext_ack *extack)
804 {
805 	int err = -ENOMEM;
806 
807 	if (!nh)
808 		goto errout;
809 
810 	if (newdst) {
811 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
812 				     nh->nh_label, extack);
813 		if (err)
814 			goto errout;
815 	}
816 
817 	if (via) {
818 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
819 				  __mpls_nh_via(rt, nh), extack);
820 		if (err)
821 			goto errout;
822 	} else {
823 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
824 	}
825 
826 	err = mpls_nh_assign_dev(net, rt, nh, oif);
827 	if (err)
828 		goto errout;
829 
830 	return 0;
831 
832 errout:
833 	return err;
834 }
835 
836 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
837 			      u8 cfg_via_alen, u8 *max_via_alen,
838 			      u8 *max_labels)
839 {
840 	int remaining = len;
841 	u8 nhs = 0;
842 
843 	*max_via_alen = 0;
844 	*max_labels = 0;
845 
846 	while (rtnh_ok(rtnh, remaining)) {
847 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
848 		int attrlen;
849 		u8 n_labels = 0;
850 
851 		attrlen = rtnh_attrlen(rtnh);
852 		nla = nla_find(attrs, attrlen, RTA_VIA);
853 		if (nla && nla_len(nla) >=
854 		    offsetof(struct rtvia, rtvia_addr)) {
855 			int via_alen = nla_len(nla) -
856 				offsetof(struct rtvia, rtvia_addr);
857 
858 			if (via_alen <= MAX_VIA_ALEN)
859 				*max_via_alen = max_t(u16, *max_via_alen,
860 						      via_alen);
861 		}
862 
863 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
864 		if (nla &&
865 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
866 				   NULL, NULL) != 0)
867 			return 0;
868 
869 		*max_labels = max_t(u8, *max_labels, n_labels);
870 
871 		/* number of nexthops is tracked by a u8.
872 		 * Check for overflow.
873 		 */
874 		if (nhs == 255)
875 			return 0;
876 		nhs++;
877 
878 		rtnh = rtnh_next(rtnh, &remaining);
879 	}
880 
881 	/* leftover implies invalid nexthop configuration, discard it */
882 	return remaining > 0 ? 0 : nhs;
883 }
884 
885 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
886 			       struct mpls_route *rt, u8 max_labels,
887 			       struct netlink_ext_ack *extack)
888 {
889 	struct rtnexthop *rtnh = cfg->rc_mp;
890 	struct nlattr *nla_via, *nla_newdst;
891 	int remaining = cfg->rc_mp_len;
892 	int err = 0;
893 	u8 nhs = 0;
894 
895 	change_nexthops(rt) {
896 		int attrlen;
897 
898 		nla_via = NULL;
899 		nla_newdst = NULL;
900 
901 		err = -EINVAL;
902 		if (!rtnh_ok(rtnh, remaining))
903 			goto errout;
904 
905 		/* neither weighted multipath nor any flags
906 		 * are supported
907 		 */
908 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
909 			goto errout;
910 
911 		attrlen = rtnh_attrlen(rtnh);
912 		if (attrlen > 0) {
913 			struct nlattr *attrs = rtnh_attrs(rtnh);
914 
915 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
916 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
917 		}
918 
919 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
920 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
921 				    max_labels, extack);
922 		if (err)
923 			goto errout;
924 
925 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
926 			rt->rt_nhn_alive--;
927 
928 		rtnh = rtnh_next(rtnh, &remaining);
929 		nhs++;
930 	} endfor_nexthops(rt);
931 
932 	rt->rt_nhn = nhs;
933 
934 	return 0;
935 
936 errout:
937 	return err;
938 }
939 
940 static bool mpls_label_ok(struct net *net, unsigned int *index,
941 			  struct netlink_ext_ack *extack)
942 {
943 	bool is_ok = true;
944 
945 	/* Reserved labels may not be set */
946 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
947 		NL_SET_ERR_MSG(extack,
948 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
949 		is_ok = false;
950 	}
951 
952 	/* The full 20 bit range may not be supported. */
953 	if (is_ok && *index >= net->mpls.platform_labels) {
954 		NL_SET_ERR_MSG(extack,
955 			       "Label >= configured maximum in platform_labels");
956 		is_ok = false;
957 	}
958 
959 	*index = array_index_nospec(*index, net->mpls.platform_labels);
960 	return is_ok;
961 }
962 
963 static int mpls_route_add(struct mpls_route_config *cfg,
964 			  struct netlink_ext_ack *extack)
965 {
966 	struct mpls_route __rcu **platform_label;
967 	struct net *net = cfg->rc_nlinfo.nl_net;
968 	struct mpls_route *rt, *old;
969 	int err = -EINVAL;
970 	u8 max_via_alen;
971 	unsigned index;
972 	u8 max_labels;
973 	u8 nhs;
974 
975 	index = cfg->rc_label;
976 
977 	/* If a label was not specified during insert pick one */
978 	if ((index == LABEL_NOT_SPECIFIED) &&
979 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
980 		index = find_free_label(net);
981 	}
982 
983 	if (!mpls_label_ok(net, &index, extack))
984 		goto errout;
985 
986 	/* Append makes no sense with mpls */
987 	err = -EOPNOTSUPP;
988 	if (cfg->rc_nlflags & NLM_F_APPEND) {
989 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
990 		goto errout;
991 	}
992 
993 	err = -EEXIST;
994 	platform_label = rtnl_dereference(net->mpls.platform_label);
995 	old = rtnl_dereference(platform_label[index]);
996 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
997 		goto errout;
998 
999 	err = -EEXIST;
1000 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1001 		goto errout;
1002 
1003 	err = -ENOENT;
1004 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1005 		goto errout;
1006 
1007 	err = -EINVAL;
1008 	if (cfg->rc_mp) {
1009 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1010 					  cfg->rc_via_alen, &max_via_alen,
1011 					  &max_labels);
1012 	} else {
1013 		max_via_alen = cfg->rc_via_alen;
1014 		max_labels = cfg->rc_output_labels;
1015 		nhs = 1;
1016 	}
1017 
1018 	if (nhs == 0) {
1019 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1020 		goto errout;
1021 	}
1022 
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_sz(net, path, table,
1423 					      ARRAY_SIZE(mpls_dev_table));
1424 	if (!mdev->sysctl)
1425 		goto free;
1426 
1427 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1428 	return 0;
1429 
1430 free:
1431 	kfree(table);
1432 out:
1433 	mdev->sysctl = NULL;
1434 	return -ENOBUFS;
1435 }
1436 
1437 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1438 				       struct mpls_dev *mdev)
1439 {
1440 	struct net *net = dev_net(dev);
1441 	struct ctl_table *table;
1442 
1443 	if (!mdev->sysctl)
1444 		return;
1445 
1446 	table = mdev->sysctl->ctl_table_arg;
1447 	unregister_net_sysctl_table(mdev->sysctl);
1448 	kfree(table);
1449 
1450 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1451 }
1452 
1453 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1454 {
1455 	struct mpls_dev *mdev;
1456 	int err = -ENOMEM;
1457 	int i;
1458 
1459 	ASSERT_RTNL();
1460 
1461 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1462 	if (!mdev)
1463 		return ERR_PTR(err);
1464 
1465 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1466 	if (!mdev->stats)
1467 		goto free;
1468 
1469 	for_each_possible_cpu(i) {
1470 		struct mpls_pcpu_stats *mpls_stats;
1471 
1472 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1473 		u64_stats_init(&mpls_stats->syncp);
1474 	}
1475 
1476 	mdev->dev = dev;
1477 
1478 	err = mpls_dev_sysctl_register(dev, mdev);
1479 	if (err)
1480 		goto free;
1481 
1482 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1483 
1484 	return mdev;
1485 
1486 free:
1487 	free_percpu(mdev->stats);
1488 	kfree(mdev);
1489 	return ERR_PTR(err);
1490 }
1491 
1492 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1493 {
1494 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1495 
1496 	free_percpu(mdev->stats);
1497 	kfree(mdev);
1498 }
1499 
1500 static int mpls_ifdown(struct net_device *dev, int event)
1501 {
1502 	struct mpls_route __rcu **platform_label;
1503 	struct net *net = dev_net(dev);
1504 	unsigned index;
1505 
1506 	platform_label = rtnl_dereference(net->mpls.platform_label);
1507 	for (index = 0; index < net->mpls.platform_labels; index++) {
1508 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1509 		bool nh_del = false;
1510 		u8 alive = 0;
1511 
1512 		if (!rt)
1513 			continue;
1514 
1515 		if (event == NETDEV_UNREGISTER) {
1516 			u8 deleted = 0;
1517 
1518 			for_nexthops(rt) {
1519 				if (!nh->nh_dev || nh->nh_dev == dev)
1520 					deleted++;
1521 				if (nh->nh_dev == dev)
1522 					nh_del = true;
1523 			} endfor_nexthops(rt);
1524 
1525 			/* if there are no more nexthops, delete the route */
1526 			if (deleted == rt->rt_nhn) {
1527 				mpls_route_update(net, index, NULL, NULL);
1528 				continue;
1529 			}
1530 
1531 			if (nh_del) {
1532 				size_t size = sizeof(*rt) + rt->rt_nhn *
1533 					rt->rt_nh_size;
1534 				struct mpls_route *orig = rt;
1535 
1536 				rt = kmemdup(orig, size, GFP_KERNEL);
1537 				if (!rt)
1538 					return -ENOMEM;
1539 			}
1540 		}
1541 
1542 		change_nexthops(rt) {
1543 			unsigned int nh_flags = nh->nh_flags;
1544 
1545 			if (nh->nh_dev != dev)
1546 				goto next;
1547 
1548 			switch (event) {
1549 			case NETDEV_DOWN:
1550 			case NETDEV_UNREGISTER:
1551 				nh_flags |= RTNH_F_DEAD;
1552 				fallthrough;
1553 			case NETDEV_CHANGE:
1554 				nh_flags |= RTNH_F_LINKDOWN;
1555 				break;
1556 			}
1557 			if (event == NETDEV_UNREGISTER)
1558 				nh->nh_dev = NULL;
1559 
1560 			if (nh->nh_flags != nh_flags)
1561 				WRITE_ONCE(nh->nh_flags, nh_flags);
1562 next:
1563 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1564 				alive++;
1565 		} endfor_nexthops(rt);
1566 
1567 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1568 
1569 		if (nh_del)
1570 			mpls_route_update(net, index, rt, NULL);
1571 	}
1572 
1573 	return 0;
1574 }
1575 
1576 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1577 {
1578 	struct mpls_route __rcu **platform_label;
1579 	struct net *net = dev_net(dev);
1580 	unsigned index;
1581 	u8 alive;
1582 
1583 	platform_label = rtnl_dereference(net->mpls.platform_label);
1584 	for (index = 0; index < net->mpls.platform_labels; index++) {
1585 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1586 
1587 		if (!rt)
1588 			continue;
1589 
1590 		alive = 0;
1591 		change_nexthops(rt) {
1592 			unsigned int nh_flags = nh->nh_flags;
1593 
1594 			if (!(nh_flags & flags)) {
1595 				alive++;
1596 				continue;
1597 			}
1598 			if (nh->nh_dev != dev)
1599 				continue;
1600 			alive++;
1601 			nh_flags &= ~flags;
1602 			WRITE_ONCE(nh->nh_flags, nh_flags);
1603 		} endfor_nexthops(rt);
1604 
1605 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1606 	}
1607 }
1608 
1609 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1610 			   void *ptr)
1611 {
1612 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1613 	struct mpls_dev *mdev;
1614 	unsigned int flags;
1615 	int err;
1616 
1617 	if (event == NETDEV_REGISTER) {
1618 		mdev = mpls_add_dev(dev);
1619 		if (IS_ERR(mdev))
1620 			return notifier_from_errno(PTR_ERR(mdev));
1621 
1622 		return NOTIFY_OK;
1623 	}
1624 
1625 	mdev = mpls_dev_get(dev);
1626 	if (!mdev)
1627 		return NOTIFY_OK;
1628 
1629 	switch (event) {
1630 
1631 	case NETDEV_DOWN:
1632 		err = mpls_ifdown(dev, event);
1633 		if (err)
1634 			return notifier_from_errno(err);
1635 		break;
1636 	case NETDEV_UP:
1637 		flags = dev_get_flags(dev);
1638 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1639 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1640 		else
1641 			mpls_ifup(dev, RTNH_F_DEAD);
1642 		break;
1643 	case NETDEV_CHANGE:
1644 		flags = dev_get_flags(dev);
1645 		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1646 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1647 		} else {
1648 			err = mpls_ifdown(dev, event);
1649 			if (err)
1650 				return notifier_from_errno(err);
1651 		}
1652 		break;
1653 	case NETDEV_UNREGISTER:
1654 		err = mpls_ifdown(dev, event);
1655 		if (err)
1656 			return notifier_from_errno(err);
1657 		mdev = mpls_dev_get(dev);
1658 		if (mdev) {
1659 			mpls_dev_sysctl_unregister(dev, mdev);
1660 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1661 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1662 		}
1663 		break;
1664 	case NETDEV_CHANGENAME:
1665 		mdev = mpls_dev_get(dev);
1666 		if (mdev) {
1667 			mpls_dev_sysctl_unregister(dev, mdev);
1668 			err = mpls_dev_sysctl_register(dev, mdev);
1669 			if (err)
1670 				return notifier_from_errno(err);
1671 		}
1672 		break;
1673 	}
1674 	return NOTIFY_OK;
1675 }
1676 
1677 static struct notifier_block mpls_dev_notifier = {
1678 	.notifier_call = mpls_dev_notify,
1679 };
1680 
1681 static int nla_put_via(struct sk_buff *skb,
1682 		       u8 table, const void *addr, int alen)
1683 {
1684 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1685 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1686 	};
1687 	struct nlattr *nla;
1688 	struct rtvia *via;
1689 	int family = AF_UNSPEC;
1690 
1691 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1692 	if (!nla)
1693 		return -EMSGSIZE;
1694 
1695 	if (table <= NEIGH_NR_TABLES)
1696 		family = table_to_family[table];
1697 
1698 	via = nla_data(nla);
1699 	via->rtvia_family = family;
1700 	memcpy(via->rtvia_addr, addr, alen);
1701 	return 0;
1702 }
1703 
1704 int nla_put_labels(struct sk_buff *skb, int attrtype,
1705 		   u8 labels, const u32 label[])
1706 {
1707 	struct nlattr *nla;
1708 	struct mpls_shim_hdr *nla_label;
1709 	bool bos;
1710 	int i;
1711 	nla = nla_reserve(skb, attrtype, labels*4);
1712 	if (!nla)
1713 		return -EMSGSIZE;
1714 
1715 	nla_label = nla_data(nla);
1716 	bos = true;
1717 	for (i = labels - 1; i >= 0; i--) {
1718 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1719 		bos = false;
1720 	}
1721 
1722 	return 0;
1723 }
1724 EXPORT_SYMBOL_GPL(nla_put_labels);
1725 
1726 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1727 		   u32 label[], struct netlink_ext_ack *extack)
1728 {
1729 	unsigned len = nla_len(nla);
1730 	struct mpls_shim_hdr *nla_label;
1731 	u8 nla_labels;
1732 	bool bos;
1733 	int i;
1734 
1735 	/* len needs to be an even multiple of 4 (the label size). Number
1736 	 * of labels is a u8 so check for overflow.
1737 	 */
1738 	if (len & 3 || len / 4 > 255) {
1739 		NL_SET_ERR_MSG_ATTR(extack, nla,
1740 				    "Invalid length for labels attribute");
1741 		return -EINVAL;
1742 	}
1743 
1744 	/* Limit the number of new labels allowed */
1745 	nla_labels = len/4;
1746 	if (nla_labels > max_labels) {
1747 		NL_SET_ERR_MSG(extack, "Too many labels");
1748 		return -EINVAL;
1749 	}
1750 
1751 	/* when label == NULL, caller wants number of labels */
1752 	if (!label)
1753 		goto out;
1754 
1755 	nla_label = nla_data(nla);
1756 	bos = true;
1757 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1758 		struct mpls_entry_decoded dec;
1759 		dec = mpls_entry_decode(nla_label + i);
1760 
1761 		/* Ensure the bottom of stack flag is properly set
1762 		 * and ttl and tc are both clear.
1763 		 */
1764 		if (dec.ttl) {
1765 			NL_SET_ERR_MSG_ATTR(extack, nla,
1766 					    "TTL in label must be 0");
1767 			return -EINVAL;
1768 		}
1769 
1770 		if (dec.tc) {
1771 			NL_SET_ERR_MSG_ATTR(extack, nla,
1772 					    "Traffic class in label must be 0");
1773 			return -EINVAL;
1774 		}
1775 
1776 		if (dec.bos != bos) {
1777 			NL_SET_BAD_ATTR(extack, nla);
1778 			if (bos) {
1779 				NL_SET_ERR_MSG(extack,
1780 					       "BOS bit must be set in first label");
1781 			} else {
1782 				NL_SET_ERR_MSG(extack,
1783 					       "BOS bit can only be set in first label");
1784 			}
1785 			return -EINVAL;
1786 		}
1787 
1788 		switch (dec.label) {
1789 		case MPLS_LABEL_IMPLNULL:
1790 			/* RFC3032: This is a label that an LSR may
1791 			 * assign and distribute, but which never
1792 			 * actually appears in the encapsulation.
1793 			 */
1794 			NL_SET_ERR_MSG_ATTR(extack, nla,
1795 					    "Implicit NULL Label (3) can not be used in encapsulation");
1796 			return -EINVAL;
1797 		}
1798 
1799 		label[i] = dec.label;
1800 	}
1801 out:
1802 	*labels = nla_labels;
1803 	return 0;
1804 }
1805 EXPORT_SYMBOL_GPL(nla_get_labels);
1806 
1807 static int rtm_to_route_config(struct sk_buff *skb,
1808 			       struct nlmsghdr *nlh,
1809 			       struct mpls_route_config *cfg,
1810 			       struct netlink_ext_ack *extack)
1811 {
1812 	struct rtmsg *rtm;
1813 	struct nlattr *tb[RTA_MAX+1];
1814 	int index;
1815 	int err;
1816 
1817 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1818 				     rtm_mpls_policy, extack);
1819 	if (err < 0)
1820 		goto errout;
1821 
1822 	err = -EINVAL;
1823 	rtm = nlmsg_data(nlh);
1824 
1825 	if (rtm->rtm_family != AF_MPLS) {
1826 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1827 		goto errout;
1828 	}
1829 	if (rtm->rtm_dst_len != 20) {
1830 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1831 		goto errout;
1832 	}
1833 	if (rtm->rtm_src_len != 0) {
1834 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1835 		goto errout;
1836 	}
1837 	if (rtm->rtm_tos != 0) {
1838 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1839 		goto errout;
1840 	}
1841 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1842 		NL_SET_ERR_MSG(extack,
1843 			       "MPLS only supports the main route table");
1844 		goto errout;
1845 	}
1846 	/* Any value is acceptable for rtm_protocol */
1847 
1848 	/* As mpls uses destination specific addresses
1849 	 * (or source specific address in the case of multicast)
1850 	 * all addresses have universal scope.
1851 	 */
1852 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1853 		NL_SET_ERR_MSG(extack,
1854 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1855 		goto errout;
1856 	}
1857 	if (rtm->rtm_type != RTN_UNICAST) {
1858 		NL_SET_ERR_MSG(extack,
1859 			       "Invalid route type - MPLS only supports UNICAST");
1860 		goto errout;
1861 	}
1862 	if (rtm->rtm_flags != 0) {
1863 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1864 		goto errout;
1865 	}
1866 
1867 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1868 	cfg->rc_protocol	= rtm->rtm_protocol;
1869 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1870 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1871 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1872 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1873 	cfg->rc_nlinfo.nlh	= nlh;
1874 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1875 
1876 	for (index = 0; index <= RTA_MAX; index++) {
1877 		struct nlattr *nla = tb[index];
1878 		if (!nla)
1879 			continue;
1880 
1881 		switch (index) {
1882 		case RTA_OIF:
1883 			cfg->rc_ifindex = nla_get_u32(nla);
1884 			break;
1885 		case RTA_NEWDST:
1886 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1887 					   &cfg->rc_output_labels,
1888 					   cfg->rc_output_label, extack))
1889 				goto errout;
1890 			break;
1891 		case RTA_DST:
1892 		{
1893 			u8 label_count;
1894 			if (nla_get_labels(nla, 1, &label_count,
1895 					   &cfg->rc_label, extack))
1896 				goto errout;
1897 
1898 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1899 					   &cfg->rc_label, extack))
1900 				goto errout;
1901 			break;
1902 		}
1903 		case RTA_GATEWAY:
1904 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1905 			goto errout;
1906 		case RTA_VIA:
1907 		{
1908 			if (nla_get_via(nla, &cfg->rc_via_alen,
1909 					&cfg->rc_via_table, cfg->rc_via,
1910 					extack))
1911 				goto errout;
1912 			break;
1913 		}
1914 		case RTA_MULTIPATH:
1915 		{
1916 			cfg->rc_mp = nla_data(nla);
1917 			cfg->rc_mp_len = nla_len(nla);
1918 			break;
1919 		}
1920 		case RTA_TTL_PROPAGATE:
1921 		{
1922 			u8 ttl_propagate = nla_get_u8(nla);
1923 
1924 			if (ttl_propagate > 1) {
1925 				NL_SET_ERR_MSG_ATTR(extack, nla,
1926 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1927 				goto errout;
1928 			}
1929 			cfg->rc_ttl_propagate = ttl_propagate ?
1930 				MPLS_TTL_PROP_ENABLED :
1931 				MPLS_TTL_PROP_DISABLED;
1932 			break;
1933 		}
1934 		default:
1935 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1936 			/* Unsupported attribute */
1937 			goto errout;
1938 		}
1939 	}
1940 
1941 	err = 0;
1942 errout:
1943 	return err;
1944 }
1945 
1946 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1947 			     struct netlink_ext_ack *extack)
1948 {
1949 	struct mpls_route_config *cfg;
1950 	int err;
1951 
1952 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1953 	if (!cfg)
1954 		return -ENOMEM;
1955 
1956 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1957 	if (err < 0)
1958 		goto out;
1959 
1960 	err = mpls_route_del(cfg, extack);
1961 out:
1962 	kfree(cfg);
1963 
1964 	return err;
1965 }
1966 
1967 
1968 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1969 			     struct netlink_ext_ack *extack)
1970 {
1971 	struct mpls_route_config *cfg;
1972 	int err;
1973 
1974 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1975 	if (!cfg)
1976 		return -ENOMEM;
1977 
1978 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1979 	if (err < 0)
1980 		goto out;
1981 
1982 	err = mpls_route_add(cfg, extack);
1983 out:
1984 	kfree(cfg);
1985 
1986 	return err;
1987 }
1988 
1989 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1990 			   u32 label, struct mpls_route *rt, int flags)
1991 {
1992 	struct net_device *dev;
1993 	struct nlmsghdr *nlh;
1994 	struct rtmsg *rtm;
1995 
1996 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1997 	if (nlh == NULL)
1998 		return -EMSGSIZE;
1999 
2000 	rtm = nlmsg_data(nlh);
2001 	rtm->rtm_family = AF_MPLS;
2002 	rtm->rtm_dst_len = 20;
2003 	rtm->rtm_src_len = 0;
2004 	rtm->rtm_tos = 0;
2005 	rtm->rtm_table = RT_TABLE_MAIN;
2006 	rtm->rtm_protocol = rt->rt_protocol;
2007 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
2008 	rtm->rtm_type = RTN_UNICAST;
2009 	rtm->rtm_flags = 0;
2010 
2011 	if (nla_put_labels(skb, RTA_DST, 1, &label))
2012 		goto nla_put_failure;
2013 
2014 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
2015 		bool ttl_propagate =
2016 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2017 
2018 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2019 			       ttl_propagate))
2020 			goto nla_put_failure;
2021 	}
2022 	if (rt->rt_nhn == 1) {
2023 		const struct mpls_nh *nh = rt->rt_nh;
2024 
2025 		if (nh->nh_labels &&
2026 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2027 				   nh->nh_label))
2028 			goto nla_put_failure;
2029 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2030 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2031 				nh->nh_via_alen))
2032 			goto nla_put_failure;
2033 		dev = nh->nh_dev;
2034 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2035 			goto nla_put_failure;
2036 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2037 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2038 		if (nh->nh_flags & RTNH_F_DEAD)
2039 			rtm->rtm_flags |= RTNH_F_DEAD;
2040 	} else {
2041 		struct rtnexthop *rtnh;
2042 		struct nlattr *mp;
2043 		u8 linkdown = 0;
2044 		u8 dead = 0;
2045 
2046 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2047 		if (!mp)
2048 			goto nla_put_failure;
2049 
2050 		for_nexthops(rt) {
2051 			dev = nh->nh_dev;
2052 			if (!dev)
2053 				continue;
2054 
2055 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2056 			if (!rtnh)
2057 				goto nla_put_failure;
2058 
2059 			rtnh->rtnh_ifindex = dev->ifindex;
2060 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2061 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2062 				linkdown++;
2063 			}
2064 			if (nh->nh_flags & RTNH_F_DEAD) {
2065 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2066 				dead++;
2067 			}
2068 
2069 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2070 							    nh->nh_labels,
2071 							    nh->nh_label))
2072 				goto nla_put_failure;
2073 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2074 			    nla_put_via(skb, nh->nh_via_table,
2075 					mpls_nh_via(rt, nh),
2076 					nh->nh_via_alen))
2077 				goto nla_put_failure;
2078 
2079 			/* length of rtnetlink header + attributes */
2080 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2081 		} endfor_nexthops(rt);
2082 
2083 		if (linkdown == rt->rt_nhn)
2084 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2085 		if (dead == rt->rt_nhn)
2086 			rtm->rtm_flags |= RTNH_F_DEAD;
2087 
2088 		nla_nest_end(skb, mp);
2089 	}
2090 
2091 	nlmsg_end(skb, nlh);
2092 	return 0;
2093 
2094 nla_put_failure:
2095 	nlmsg_cancel(skb, nlh);
2096 	return -EMSGSIZE;
2097 }
2098 
2099 #if IS_ENABLED(CONFIG_INET)
2100 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2101 				   struct fib_dump_filter *filter,
2102 				   struct netlink_callback *cb)
2103 {
2104 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2105 }
2106 #else
2107 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2108 				   struct fib_dump_filter *filter,
2109 				   struct netlink_callback *cb)
2110 {
2111 	struct netlink_ext_ack *extack = cb->extack;
2112 	struct nlattr *tb[RTA_MAX + 1];
2113 	struct rtmsg *rtm;
2114 	int err, i;
2115 
2116 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2117 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2118 		return -EINVAL;
2119 	}
2120 
2121 	rtm = nlmsg_data(nlh);
2122 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2123 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2124 	    rtm->rtm_flags) {
2125 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2126 		return -EINVAL;
2127 	}
2128 
2129 	if (rtm->rtm_protocol) {
2130 		filter->protocol = rtm->rtm_protocol;
2131 		filter->filter_set = 1;
2132 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2133 	}
2134 
2135 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2136 					    rtm_mpls_policy, extack);
2137 	if (err < 0)
2138 		return err;
2139 
2140 	for (i = 0; i <= RTA_MAX; ++i) {
2141 		int ifindex;
2142 
2143 		if (i == RTA_OIF) {
2144 			ifindex = nla_get_u32(tb[i]);
2145 			filter->dev = __dev_get_by_index(net, ifindex);
2146 			if (!filter->dev)
2147 				return -ENODEV;
2148 			filter->filter_set = 1;
2149 		} else if (tb[i]) {
2150 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2151 			return -EINVAL;
2152 		}
2153 	}
2154 
2155 	return 0;
2156 }
2157 #endif
2158 
2159 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2160 			     const struct net_device *dev)
2161 {
2162 	if (rt->rt_nhn == 1) {
2163 		struct mpls_nh *nh = rt->rt_nh;
2164 
2165 		if (nh->nh_dev == dev)
2166 			return true;
2167 	} else {
2168 		for_nexthops(rt) {
2169 			if (nh->nh_dev == dev)
2170 				return true;
2171 		} endfor_nexthops(rt);
2172 	}
2173 
2174 	return false;
2175 }
2176 
2177 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2178 {
2179 	const struct nlmsghdr *nlh = cb->nlh;
2180 	struct net *net = sock_net(skb->sk);
2181 	struct mpls_route __rcu **platform_label;
2182 	struct fib_dump_filter filter = {};
2183 	unsigned int flags = NLM_F_MULTI;
2184 	size_t platform_labels;
2185 	unsigned int index;
2186 
2187 	ASSERT_RTNL();
2188 
2189 	if (cb->strict_check) {
2190 		int err;
2191 
2192 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2193 		if (err < 0)
2194 			return err;
2195 
2196 		/* for MPLS, there is only 1 table with fixed type and flags.
2197 		 * If either are set in the filter then return nothing.
2198 		 */
2199 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2200 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2201 		     filter.flags)
2202 			return skb->len;
2203 	}
2204 
2205 	index = cb->args[0];
2206 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2207 		index = MPLS_LABEL_FIRST_UNRESERVED;
2208 
2209 	platform_label = rtnl_dereference(net->mpls.platform_label);
2210 	platform_labels = net->mpls.platform_labels;
2211 
2212 	if (filter.filter_set)
2213 		flags |= NLM_F_DUMP_FILTERED;
2214 
2215 	for (; index < platform_labels; index++) {
2216 		struct mpls_route *rt;
2217 
2218 		rt = rtnl_dereference(platform_label[index]);
2219 		if (!rt)
2220 			continue;
2221 
2222 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2223 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2224 			continue;
2225 
2226 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2227 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2228 				    index, rt, flags) < 0)
2229 			break;
2230 	}
2231 	cb->args[0] = index;
2232 
2233 	return skb->len;
2234 }
2235 
2236 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2237 {
2238 	size_t payload =
2239 		NLMSG_ALIGN(sizeof(struct rtmsg))
2240 		+ nla_total_size(4)			/* RTA_DST */
2241 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2242 
2243 	if (rt->rt_nhn == 1) {
2244 		struct mpls_nh *nh = rt->rt_nh;
2245 
2246 		if (nh->nh_dev)
2247 			payload += nla_total_size(4); /* RTA_OIF */
2248 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2249 			payload += nla_total_size(2 + nh->nh_via_alen);
2250 		if (nh->nh_labels) /* RTA_NEWDST */
2251 			payload += nla_total_size(nh->nh_labels * 4);
2252 	} else {
2253 		/* each nexthop is packed in an attribute */
2254 		size_t nhsize = 0;
2255 
2256 		for_nexthops(rt) {
2257 			if (!nh->nh_dev)
2258 				continue;
2259 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2260 			/* RTA_VIA */
2261 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2262 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2263 			if (nh->nh_labels)
2264 				nhsize += nla_total_size(nh->nh_labels * 4);
2265 		} endfor_nexthops(rt);
2266 		/* nested attribute */
2267 		payload += nla_total_size(nhsize);
2268 	}
2269 
2270 	return payload;
2271 }
2272 
2273 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2274 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2275 		       unsigned int nlm_flags)
2276 {
2277 	struct sk_buff *skb;
2278 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2279 	int err = -ENOBUFS;
2280 
2281 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2282 	if (skb == NULL)
2283 		goto errout;
2284 
2285 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2286 	if (err < 0) {
2287 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2288 		WARN_ON(err == -EMSGSIZE);
2289 		kfree_skb(skb);
2290 		goto errout;
2291 	}
2292 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2293 
2294 	return;
2295 errout:
2296 	if (err < 0)
2297 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2298 }
2299 
2300 static int mpls_valid_getroute_req(struct sk_buff *skb,
2301 				   const struct nlmsghdr *nlh,
2302 				   struct nlattr **tb,
2303 				   struct netlink_ext_ack *extack)
2304 {
2305 	struct rtmsg *rtm;
2306 	int i, err;
2307 
2308 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2309 		NL_SET_ERR_MSG_MOD(extack,
2310 				   "Invalid header for get route request");
2311 		return -EINVAL;
2312 	}
2313 
2314 	if (!netlink_strict_get_check(skb))
2315 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2316 					      rtm_mpls_policy, extack);
2317 
2318 	rtm = nlmsg_data(nlh);
2319 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2320 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2321 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2322 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2323 		return -EINVAL;
2324 	}
2325 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2326 		NL_SET_ERR_MSG_MOD(extack,
2327 				   "Invalid flags for get route request");
2328 		return -EINVAL;
2329 	}
2330 
2331 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2332 					    rtm_mpls_policy, extack);
2333 	if (err)
2334 		return err;
2335 
2336 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2337 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2338 		return -EINVAL;
2339 	}
2340 
2341 	for (i = 0; i <= RTA_MAX; i++) {
2342 		if (!tb[i])
2343 			continue;
2344 
2345 		switch (i) {
2346 		case RTA_DST:
2347 		case RTA_NEWDST:
2348 			break;
2349 		default:
2350 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2351 			return -EINVAL;
2352 		}
2353 	}
2354 
2355 	return 0;
2356 }
2357 
2358 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2359 			 struct netlink_ext_ack *extack)
2360 {
2361 	struct net *net = sock_net(in_skb->sk);
2362 	u32 portid = NETLINK_CB(in_skb).portid;
2363 	u32 in_label = LABEL_NOT_SPECIFIED;
2364 	struct nlattr *tb[RTA_MAX + 1];
2365 	u32 labels[MAX_NEW_LABELS];
2366 	struct mpls_shim_hdr *hdr;
2367 	unsigned int hdr_size = 0;
2368 	const struct mpls_nh *nh;
2369 	struct net_device *dev;
2370 	struct mpls_route *rt;
2371 	struct rtmsg *rtm, *r;
2372 	struct nlmsghdr *nlh;
2373 	struct sk_buff *skb;
2374 	u8 n_labels;
2375 	int err;
2376 
2377 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2378 	if (err < 0)
2379 		goto errout;
2380 
2381 	rtm = nlmsg_data(in_nlh);
2382 
2383 	if (tb[RTA_DST]) {
2384 		u8 label_count;
2385 
2386 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2387 				   &in_label, extack)) {
2388 			err = -EINVAL;
2389 			goto errout;
2390 		}
2391 
2392 		if (!mpls_label_ok(net, &in_label, extack)) {
2393 			err = -EINVAL;
2394 			goto errout;
2395 		}
2396 	}
2397 
2398 	rt = mpls_route_input_rcu(net, in_label);
2399 	if (!rt) {
2400 		err = -ENETUNREACH;
2401 		goto errout;
2402 	}
2403 
2404 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2405 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2406 		if (!skb) {
2407 			err = -ENOBUFS;
2408 			goto errout;
2409 		}
2410 
2411 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2412 				      RTM_NEWROUTE, in_label, rt, 0);
2413 		if (err < 0) {
2414 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2415 			WARN_ON(err == -EMSGSIZE);
2416 			goto errout_free;
2417 		}
2418 
2419 		return rtnl_unicast(skb, net, portid);
2420 	}
2421 
2422 	if (tb[RTA_NEWDST]) {
2423 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2424 				   labels, extack) != 0) {
2425 			err = -EINVAL;
2426 			goto errout;
2427 		}
2428 
2429 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2430 	}
2431 
2432 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2433 	if (!skb) {
2434 		err = -ENOBUFS;
2435 		goto errout;
2436 	}
2437 
2438 	skb->protocol = htons(ETH_P_MPLS_UC);
2439 
2440 	if (hdr_size) {
2441 		bool bos;
2442 		int i;
2443 
2444 		if (skb_cow(skb, hdr_size)) {
2445 			err = -ENOBUFS;
2446 			goto errout_free;
2447 		}
2448 
2449 		skb_reserve(skb, hdr_size);
2450 		skb_push(skb, hdr_size);
2451 		skb_reset_network_header(skb);
2452 
2453 		/* Push new labels */
2454 		hdr = mpls_hdr(skb);
2455 		bos = true;
2456 		for (i = n_labels - 1; i >= 0; i--) {
2457 			hdr[i] = mpls_entry_encode(labels[i],
2458 						   1, 0, bos);
2459 			bos = false;
2460 		}
2461 	}
2462 
2463 	nh = mpls_select_multipath(rt, skb);
2464 	if (!nh) {
2465 		err = -ENETUNREACH;
2466 		goto errout_free;
2467 	}
2468 
2469 	if (hdr_size) {
2470 		skb_pull(skb, hdr_size);
2471 		skb_reset_network_header(skb);
2472 	}
2473 
2474 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2475 			RTM_NEWROUTE, sizeof(*r), 0);
2476 	if (!nlh) {
2477 		err = -EMSGSIZE;
2478 		goto errout_free;
2479 	}
2480 
2481 	r = nlmsg_data(nlh);
2482 	r->rtm_family	 = AF_MPLS;
2483 	r->rtm_dst_len	= 20;
2484 	r->rtm_src_len	= 0;
2485 	r->rtm_table	= RT_TABLE_MAIN;
2486 	r->rtm_type	= RTN_UNICAST;
2487 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2488 	r->rtm_protocol = rt->rt_protocol;
2489 	r->rtm_flags	= 0;
2490 
2491 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2492 		goto nla_put_failure;
2493 
2494 	if (nh->nh_labels &&
2495 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2496 			   nh->nh_label))
2497 		goto nla_put_failure;
2498 
2499 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2500 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2501 			nh->nh_via_alen))
2502 		goto nla_put_failure;
2503 	dev = nh->nh_dev;
2504 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2505 		goto nla_put_failure;
2506 
2507 	nlmsg_end(skb, nlh);
2508 
2509 	err = rtnl_unicast(skb, net, portid);
2510 errout:
2511 	return err;
2512 
2513 nla_put_failure:
2514 	nlmsg_cancel(skb, nlh);
2515 	err = -EMSGSIZE;
2516 errout_free:
2517 	kfree_skb(skb);
2518 	return err;
2519 }
2520 
2521 static int resize_platform_label_table(struct net *net, size_t limit)
2522 {
2523 	size_t size = sizeof(struct mpls_route *) * limit;
2524 	size_t old_limit;
2525 	size_t cp_size;
2526 	struct mpls_route __rcu **labels = NULL, **old;
2527 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2528 	unsigned index;
2529 
2530 	if (size) {
2531 		labels = kvzalloc(size, GFP_KERNEL);
2532 		if (!labels)
2533 			goto nolabels;
2534 	}
2535 
2536 	/* In case the predefined labels need to be populated */
2537 	if (limit > MPLS_LABEL_IPV4NULL) {
2538 		struct net_device *lo = net->loopback_dev;
2539 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2540 		if (IS_ERR(rt0))
2541 			goto nort0;
2542 		rt0->rt_nh->nh_dev = lo;
2543 		rt0->rt_protocol = RTPROT_KERNEL;
2544 		rt0->rt_payload_type = MPT_IPV4;
2545 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2546 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2547 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2548 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2549 		       lo->addr_len);
2550 	}
2551 	if (limit > MPLS_LABEL_IPV6NULL) {
2552 		struct net_device *lo = net->loopback_dev;
2553 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2554 		if (IS_ERR(rt2))
2555 			goto nort2;
2556 		rt2->rt_nh->nh_dev = lo;
2557 		rt2->rt_protocol = RTPROT_KERNEL;
2558 		rt2->rt_payload_type = MPT_IPV6;
2559 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2560 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2561 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2562 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2563 		       lo->addr_len);
2564 	}
2565 
2566 	rtnl_lock();
2567 	/* Remember the original table */
2568 	old = rtnl_dereference(net->mpls.platform_label);
2569 	old_limit = net->mpls.platform_labels;
2570 
2571 	/* Free any labels beyond the new table */
2572 	for (index = limit; index < old_limit; index++)
2573 		mpls_route_update(net, index, NULL, NULL);
2574 
2575 	/* Copy over the old labels */
2576 	cp_size = size;
2577 	if (old_limit < limit)
2578 		cp_size = old_limit * sizeof(struct mpls_route *);
2579 
2580 	memcpy(labels, old, cp_size);
2581 
2582 	/* If needed set the predefined labels */
2583 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2584 	    (limit > MPLS_LABEL_IPV6NULL)) {
2585 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2586 		rt2 = NULL;
2587 	}
2588 
2589 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2590 	    (limit > MPLS_LABEL_IPV4NULL)) {
2591 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2592 		rt0 = NULL;
2593 	}
2594 
2595 	/* Update the global pointers */
2596 	net->mpls.platform_labels = limit;
2597 	rcu_assign_pointer(net->mpls.platform_label, labels);
2598 
2599 	rtnl_unlock();
2600 
2601 	mpls_rt_free(rt2);
2602 	mpls_rt_free(rt0);
2603 
2604 	if (old) {
2605 		synchronize_rcu();
2606 		kvfree(old);
2607 	}
2608 	return 0;
2609 
2610 nort2:
2611 	mpls_rt_free(rt0);
2612 nort0:
2613 	kvfree(labels);
2614 nolabels:
2615 	return -ENOMEM;
2616 }
2617 
2618 static int mpls_platform_labels(struct ctl_table *table, int write,
2619 				void *buffer, size_t *lenp, loff_t *ppos)
2620 {
2621 	struct net *net = table->data;
2622 	int platform_labels = net->mpls.platform_labels;
2623 	int ret;
2624 	struct ctl_table tmp = {
2625 		.procname	= table->procname,
2626 		.data		= &platform_labels,
2627 		.maxlen		= sizeof(int),
2628 		.mode		= table->mode,
2629 		.extra1		= SYSCTL_ZERO,
2630 		.extra2		= &label_limit,
2631 	};
2632 
2633 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2634 
2635 	if (write && ret == 0)
2636 		ret = resize_platform_label_table(net, platform_labels);
2637 
2638 	return ret;
2639 }
2640 
2641 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2642 	(&((struct net *)0)->field)
2643 
2644 static const struct ctl_table mpls_table[] = {
2645 	{
2646 		.procname	= "platform_labels",
2647 		.data		= NULL,
2648 		.maxlen		= sizeof(int),
2649 		.mode		= 0644,
2650 		.proc_handler	= mpls_platform_labels,
2651 	},
2652 	{
2653 		.procname	= "ip_ttl_propagate",
2654 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2655 		.maxlen		= sizeof(int),
2656 		.mode		= 0644,
2657 		.proc_handler	= proc_dointvec_minmax,
2658 		.extra1		= SYSCTL_ZERO,
2659 		.extra2		= SYSCTL_ONE,
2660 	},
2661 	{
2662 		.procname	= "default_ttl",
2663 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2664 		.maxlen		= sizeof(int),
2665 		.mode		= 0644,
2666 		.proc_handler	= proc_dointvec_minmax,
2667 		.extra1		= SYSCTL_ONE,
2668 		.extra2		= &ttl_max,
2669 	},
2670 	{ }
2671 };
2672 
2673 static int mpls_net_init(struct net *net)
2674 {
2675 	struct ctl_table *table;
2676 	int i;
2677 
2678 	net->mpls.platform_labels = 0;
2679 	net->mpls.platform_label = NULL;
2680 	net->mpls.ip_ttl_propagate = 1;
2681 	net->mpls.default_ttl = 255;
2682 
2683 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2684 	if (table == NULL)
2685 		return -ENOMEM;
2686 
2687 	/* Table data contains only offsets relative to the base of
2688 	 * the mdev at this point, so make them absolute.
2689 	 */
2690 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2691 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2692 
2693 	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2694 					       ARRAY_SIZE(mpls_table));
2695 	if (net->mpls.ctl == NULL) {
2696 		kfree(table);
2697 		return -ENOMEM;
2698 	}
2699 
2700 	return 0;
2701 }
2702 
2703 static void mpls_net_exit(struct net *net)
2704 {
2705 	struct mpls_route __rcu **platform_label;
2706 	size_t platform_labels;
2707 	struct ctl_table *table;
2708 	unsigned int index;
2709 
2710 	table = net->mpls.ctl->ctl_table_arg;
2711 	unregister_net_sysctl_table(net->mpls.ctl);
2712 	kfree(table);
2713 
2714 	/* An rcu grace period has passed since there was a device in
2715 	 * the network namespace (and thus the last in flight packet)
2716 	 * left this network namespace.  This is because
2717 	 * unregister_netdevice_many and netdev_run_todo has completed
2718 	 * for each network device that was in this network namespace.
2719 	 *
2720 	 * As such no additional rcu synchronization is necessary when
2721 	 * freeing the platform_label table.
2722 	 */
2723 	rtnl_lock();
2724 	platform_label = rtnl_dereference(net->mpls.platform_label);
2725 	platform_labels = net->mpls.platform_labels;
2726 	for (index = 0; index < platform_labels; index++) {
2727 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2728 		RCU_INIT_POINTER(platform_label[index], NULL);
2729 		mpls_notify_route(net, index, rt, NULL, NULL);
2730 		mpls_rt_free(rt);
2731 	}
2732 	rtnl_unlock();
2733 
2734 	kvfree(platform_label);
2735 }
2736 
2737 static struct pernet_operations mpls_net_ops = {
2738 	.init = mpls_net_init,
2739 	.exit = mpls_net_exit,
2740 };
2741 
2742 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2743 	.family		   = AF_MPLS,
2744 	.fill_stats_af	   = mpls_fill_stats_af,
2745 	.get_stats_af_size = mpls_get_stats_af_size,
2746 };
2747 
2748 static int __init mpls_init(void)
2749 {
2750 	int err;
2751 
2752 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2753 
2754 	err = register_pernet_subsys(&mpls_net_ops);
2755 	if (err)
2756 		goto out;
2757 
2758 	err = register_netdevice_notifier(&mpls_dev_notifier);
2759 	if (err)
2760 		goto out_unregister_pernet;
2761 
2762 	dev_add_pack(&mpls_packet_type);
2763 
2764 	rtnl_af_register(&mpls_af_ops);
2765 
2766 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2767 			     mpls_rtm_newroute, NULL, 0);
2768 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2769 			     mpls_rtm_delroute, NULL, 0);
2770 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2771 			     mpls_getroute, mpls_dump_routes, 0);
2772 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2773 			     mpls_netconf_get_devconf,
2774 			     mpls_netconf_dump_devconf, 0);
2775 	err = ipgre_tunnel_encap_add_mpls_ops();
2776 	if (err)
2777 		pr_err("Can't add mpls over gre tunnel ops\n");
2778 
2779 	err = 0;
2780 out:
2781 	return err;
2782 
2783 out_unregister_pernet:
2784 	unregister_pernet_subsys(&mpls_net_ops);
2785 	goto out;
2786 }
2787 module_init(mpls_init);
2788 
2789 static void __exit mpls_exit(void)
2790 {
2791 	rtnl_unregister_all(PF_MPLS);
2792 	rtnl_af_unregister(&mpls_af_ops);
2793 	dev_remove_pack(&mpls_packet_type);
2794 	unregister_netdevice_notifier(&mpls_dev_notifier);
2795 	unregister_pernet_subsys(&mpls_net_ops);
2796 	ipgre_tunnel_encap_del_mpls_ops();
2797 }
2798 module_exit(mpls_exit);
2799 
2800 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2801 MODULE_LICENSE("GPL v2");
2802 MODULE_ALIAS_NETPROTO(PF_MPLS);
2803