xref: /linux/net/mpls/af_mpls.c (revision 2697b79a469b68e3ad3640f55284359c1396278d)
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, RT_SCOPE_UNIVERSE);
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 			READ_ONCE(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 {
1307 		unsigned long ifindex;
1308 	} *ctx = (void *)cb->ctx;
1309 	struct net_device *dev;
1310 	struct mpls_dev *mdev;
1311 	int err = 0;
1312 
1313 	if (cb->strict_check) {
1314 		struct netlink_ext_ack *extack = cb->extack;
1315 		struct netconfmsg *ncm;
1316 
1317 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1318 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1319 			return -EINVAL;
1320 		}
1321 
1322 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1323 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1324 			return -EINVAL;
1325 		}
1326 	}
1327 
1328 	rcu_read_lock();
1329 	for_each_netdev_dump(net, dev, ctx->ifindex) {
1330 		mdev = mpls_dev_get(dev);
1331 		if (!mdev)
1332 			continue;
1333 		err = mpls_netconf_fill_devconf(skb, mdev,
1334 						NETLINK_CB(cb->skb).portid,
1335 						nlh->nlmsg_seq,
1336 						RTM_NEWNETCONF,
1337 						NLM_F_MULTI,
1338 						NETCONFA_ALL);
1339 		if (err < 0)
1340 			break;
1341 	}
1342 	rcu_read_unlock();
1343 
1344 	return err;
1345 }
1346 
1347 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1348 	(&((struct mpls_dev *)0)->field)
1349 
1350 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1351 			  void *buffer, size_t *lenp, loff_t *ppos)
1352 {
1353 	int oval = *(int *)ctl->data;
1354 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1355 
1356 	if (write) {
1357 		struct mpls_dev *mdev = ctl->extra1;
1358 		int i = (int *)ctl->data - (int *)mdev;
1359 		struct net *net = ctl->extra2;
1360 		int val = *(int *)ctl->data;
1361 
1362 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1363 		    val != oval) {
1364 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1365 						    NETCONFA_INPUT, mdev);
1366 		}
1367 	}
1368 
1369 	return ret;
1370 }
1371 
1372 static const struct ctl_table mpls_dev_table[] = {
1373 	{
1374 		.procname	= "input",
1375 		.maxlen		= sizeof(int),
1376 		.mode		= 0644,
1377 		.proc_handler	= mpls_conf_proc,
1378 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1379 	},
1380 };
1381 
1382 static int mpls_dev_sysctl_register(struct net_device *dev,
1383 				    struct mpls_dev *mdev)
1384 {
1385 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1386 	size_t table_size = ARRAY_SIZE(mpls_dev_table);
1387 	struct net *net = dev_net(dev);
1388 	struct ctl_table *table;
1389 	int i;
1390 
1391 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1392 	if (!table)
1393 		goto out;
1394 
1395 	/* Table data contains only offsets relative to the base of
1396 	 * the mdev at this point, so make them absolute.
1397 	 */
1398 	for (i = 0; i < table_size; i++) {
1399 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1400 		table[i].extra1 = mdev;
1401 		table[i].extra2 = net;
1402 	}
1403 
1404 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1405 
1406 	mdev->sysctl = register_net_sysctl_sz(net, path, table, table_size);
1407 	if (!mdev->sysctl)
1408 		goto free;
1409 
1410 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1411 	return 0;
1412 
1413 free:
1414 	kfree(table);
1415 out:
1416 	mdev->sysctl = NULL;
1417 	return -ENOBUFS;
1418 }
1419 
1420 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1421 				       struct mpls_dev *mdev)
1422 {
1423 	struct net *net = dev_net(dev);
1424 	const struct ctl_table *table;
1425 
1426 	if (!mdev->sysctl)
1427 		return;
1428 
1429 	table = mdev->sysctl->ctl_table_arg;
1430 	unregister_net_sysctl_table(mdev->sysctl);
1431 	kfree(table);
1432 
1433 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1434 }
1435 
1436 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1437 {
1438 	struct mpls_dev *mdev;
1439 	int err = -ENOMEM;
1440 	int i;
1441 
1442 	ASSERT_RTNL();
1443 
1444 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1445 	if (!mdev)
1446 		return ERR_PTR(err);
1447 
1448 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1449 	if (!mdev->stats)
1450 		goto free;
1451 
1452 	for_each_possible_cpu(i) {
1453 		struct mpls_pcpu_stats *mpls_stats;
1454 
1455 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1456 		u64_stats_init(&mpls_stats->syncp);
1457 	}
1458 
1459 	mdev->dev = dev;
1460 
1461 	err = mpls_dev_sysctl_register(dev, mdev);
1462 	if (err)
1463 		goto free;
1464 
1465 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1466 
1467 	return mdev;
1468 
1469 free:
1470 	free_percpu(mdev->stats);
1471 	kfree(mdev);
1472 	return ERR_PTR(err);
1473 }
1474 
1475 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1476 {
1477 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1478 
1479 	free_percpu(mdev->stats);
1480 	kfree(mdev);
1481 }
1482 
1483 static int mpls_ifdown(struct net_device *dev, int event)
1484 {
1485 	struct mpls_route __rcu **platform_label;
1486 	struct net *net = dev_net(dev);
1487 	unsigned index;
1488 
1489 	platform_label = rtnl_dereference(net->mpls.platform_label);
1490 	for (index = 0; index < net->mpls.platform_labels; index++) {
1491 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1492 		bool nh_del = false;
1493 		u8 alive = 0;
1494 
1495 		if (!rt)
1496 			continue;
1497 
1498 		if (event == NETDEV_UNREGISTER) {
1499 			u8 deleted = 0;
1500 
1501 			for_nexthops(rt) {
1502 				if (!nh->nh_dev || nh->nh_dev == dev)
1503 					deleted++;
1504 				if (nh->nh_dev == dev)
1505 					nh_del = true;
1506 			} endfor_nexthops(rt);
1507 
1508 			/* if there are no more nexthops, delete the route */
1509 			if (deleted == rt->rt_nhn) {
1510 				mpls_route_update(net, index, NULL, NULL);
1511 				continue;
1512 			}
1513 
1514 			if (nh_del) {
1515 				size_t size = sizeof(*rt) + rt->rt_nhn *
1516 					rt->rt_nh_size;
1517 				struct mpls_route *orig = rt;
1518 
1519 				rt = kmemdup(orig, size, GFP_KERNEL);
1520 				if (!rt)
1521 					return -ENOMEM;
1522 			}
1523 		}
1524 
1525 		change_nexthops(rt) {
1526 			unsigned int nh_flags = nh->nh_flags;
1527 
1528 			if (nh->nh_dev != dev)
1529 				goto next;
1530 
1531 			switch (event) {
1532 			case NETDEV_DOWN:
1533 			case NETDEV_UNREGISTER:
1534 				nh_flags |= RTNH_F_DEAD;
1535 				fallthrough;
1536 			case NETDEV_CHANGE:
1537 				nh_flags |= RTNH_F_LINKDOWN;
1538 				break;
1539 			}
1540 			if (event == NETDEV_UNREGISTER)
1541 				nh->nh_dev = NULL;
1542 
1543 			if (nh->nh_flags != nh_flags)
1544 				WRITE_ONCE(nh->nh_flags, nh_flags);
1545 next:
1546 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1547 				alive++;
1548 		} endfor_nexthops(rt);
1549 
1550 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1551 
1552 		if (nh_del)
1553 			mpls_route_update(net, index, rt, NULL);
1554 	}
1555 
1556 	return 0;
1557 }
1558 
1559 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1560 {
1561 	struct mpls_route __rcu **platform_label;
1562 	struct net *net = dev_net(dev);
1563 	unsigned index;
1564 	u8 alive;
1565 
1566 	platform_label = rtnl_dereference(net->mpls.platform_label);
1567 	for (index = 0; index < net->mpls.platform_labels; index++) {
1568 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1569 
1570 		if (!rt)
1571 			continue;
1572 
1573 		alive = 0;
1574 		change_nexthops(rt) {
1575 			unsigned int nh_flags = nh->nh_flags;
1576 
1577 			if (!(nh_flags & flags)) {
1578 				alive++;
1579 				continue;
1580 			}
1581 			if (nh->nh_dev != dev)
1582 				continue;
1583 			alive++;
1584 			nh_flags &= ~flags;
1585 			WRITE_ONCE(nh->nh_flags, nh_flags);
1586 		} endfor_nexthops(rt);
1587 
1588 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1589 	}
1590 }
1591 
1592 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1593 			   void *ptr)
1594 {
1595 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1596 	struct mpls_dev *mdev;
1597 	unsigned int flags;
1598 	int err;
1599 
1600 	if (event == NETDEV_REGISTER) {
1601 		mdev = mpls_add_dev(dev);
1602 		if (IS_ERR(mdev))
1603 			return notifier_from_errno(PTR_ERR(mdev));
1604 
1605 		return NOTIFY_OK;
1606 	}
1607 
1608 	mdev = mpls_dev_get(dev);
1609 	if (!mdev)
1610 		return NOTIFY_OK;
1611 
1612 	switch (event) {
1613 
1614 	case NETDEV_DOWN:
1615 		err = mpls_ifdown(dev, event);
1616 		if (err)
1617 			return notifier_from_errno(err);
1618 		break;
1619 	case NETDEV_UP:
1620 		flags = dev_get_flags(dev);
1621 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1622 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1623 		else
1624 			mpls_ifup(dev, RTNH_F_DEAD);
1625 		break;
1626 	case NETDEV_CHANGE:
1627 		flags = dev_get_flags(dev);
1628 		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1629 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1630 		} else {
1631 			err = mpls_ifdown(dev, event);
1632 			if (err)
1633 				return notifier_from_errno(err);
1634 		}
1635 		break;
1636 	case NETDEV_UNREGISTER:
1637 		err = mpls_ifdown(dev, event);
1638 		if (err)
1639 			return notifier_from_errno(err);
1640 		mdev = mpls_dev_get(dev);
1641 		if (mdev) {
1642 			mpls_dev_sysctl_unregister(dev, mdev);
1643 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1644 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1645 		}
1646 		break;
1647 	case NETDEV_CHANGENAME:
1648 		mdev = mpls_dev_get(dev);
1649 		if (mdev) {
1650 			mpls_dev_sysctl_unregister(dev, mdev);
1651 			err = mpls_dev_sysctl_register(dev, mdev);
1652 			if (err)
1653 				return notifier_from_errno(err);
1654 		}
1655 		break;
1656 	}
1657 	return NOTIFY_OK;
1658 }
1659 
1660 static struct notifier_block mpls_dev_notifier = {
1661 	.notifier_call = mpls_dev_notify,
1662 };
1663 
1664 static int nla_put_via(struct sk_buff *skb,
1665 		       u8 table, const void *addr, int alen)
1666 {
1667 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1668 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1669 	};
1670 	struct nlattr *nla;
1671 	struct rtvia *via;
1672 	int family = AF_UNSPEC;
1673 
1674 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1675 	if (!nla)
1676 		return -EMSGSIZE;
1677 
1678 	if (table <= NEIGH_NR_TABLES)
1679 		family = table_to_family[table];
1680 
1681 	via = nla_data(nla);
1682 	via->rtvia_family = family;
1683 	memcpy(via->rtvia_addr, addr, alen);
1684 	return 0;
1685 }
1686 
1687 int nla_put_labels(struct sk_buff *skb, int attrtype,
1688 		   u8 labels, const u32 label[])
1689 {
1690 	struct nlattr *nla;
1691 	struct mpls_shim_hdr *nla_label;
1692 	bool bos;
1693 	int i;
1694 	nla = nla_reserve(skb, attrtype, labels*4);
1695 	if (!nla)
1696 		return -EMSGSIZE;
1697 
1698 	nla_label = nla_data(nla);
1699 	bos = true;
1700 	for (i = labels - 1; i >= 0; i--) {
1701 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1702 		bos = false;
1703 	}
1704 
1705 	return 0;
1706 }
1707 EXPORT_SYMBOL_GPL(nla_put_labels);
1708 
1709 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1710 		   u32 label[], struct netlink_ext_ack *extack)
1711 {
1712 	unsigned len = nla_len(nla);
1713 	struct mpls_shim_hdr *nla_label;
1714 	u8 nla_labels;
1715 	bool bos;
1716 	int i;
1717 
1718 	/* len needs to be an even multiple of 4 (the label size). Number
1719 	 * of labels is a u8 so check for overflow.
1720 	 */
1721 	if (len & 3 || len / 4 > 255) {
1722 		NL_SET_ERR_MSG_ATTR(extack, nla,
1723 				    "Invalid length for labels attribute");
1724 		return -EINVAL;
1725 	}
1726 
1727 	/* Limit the number of new labels allowed */
1728 	nla_labels = len/4;
1729 	if (nla_labels > max_labels) {
1730 		NL_SET_ERR_MSG(extack, "Too many labels");
1731 		return -EINVAL;
1732 	}
1733 
1734 	/* when label == NULL, caller wants number of labels */
1735 	if (!label)
1736 		goto out;
1737 
1738 	nla_label = nla_data(nla);
1739 	bos = true;
1740 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1741 		struct mpls_entry_decoded dec;
1742 		dec = mpls_entry_decode(nla_label + i);
1743 
1744 		/* Ensure the bottom of stack flag is properly set
1745 		 * and ttl and tc are both clear.
1746 		 */
1747 		if (dec.ttl) {
1748 			NL_SET_ERR_MSG_ATTR(extack, nla,
1749 					    "TTL in label must be 0");
1750 			return -EINVAL;
1751 		}
1752 
1753 		if (dec.tc) {
1754 			NL_SET_ERR_MSG_ATTR(extack, nla,
1755 					    "Traffic class in label must be 0");
1756 			return -EINVAL;
1757 		}
1758 
1759 		if (dec.bos != bos) {
1760 			NL_SET_BAD_ATTR(extack, nla);
1761 			if (bos) {
1762 				NL_SET_ERR_MSG(extack,
1763 					       "BOS bit must be set in first label");
1764 			} else {
1765 				NL_SET_ERR_MSG(extack,
1766 					       "BOS bit can only be set in first label");
1767 			}
1768 			return -EINVAL;
1769 		}
1770 
1771 		switch (dec.label) {
1772 		case MPLS_LABEL_IMPLNULL:
1773 			/* RFC3032: This is a label that an LSR may
1774 			 * assign and distribute, but which never
1775 			 * actually appears in the encapsulation.
1776 			 */
1777 			NL_SET_ERR_MSG_ATTR(extack, nla,
1778 					    "Implicit NULL Label (3) can not be used in encapsulation");
1779 			return -EINVAL;
1780 		}
1781 
1782 		label[i] = dec.label;
1783 	}
1784 out:
1785 	*labels = nla_labels;
1786 	return 0;
1787 }
1788 EXPORT_SYMBOL_GPL(nla_get_labels);
1789 
1790 static int rtm_to_route_config(struct sk_buff *skb,
1791 			       struct nlmsghdr *nlh,
1792 			       struct mpls_route_config *cfg,
1793 			       struct netlink_ext_ack *extack)
1794 {
1795 	struct rtmsg *rtm;
1796 	struct nlattr *tb[RTA_MAX+1];
1797 	int index;
1798 	int err;
1799 
1800 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1801 				     rtm_mpls_policy, extack);
1802 	if (err < 0)
1803 		goto errout;
1804 
1805 	err = -EINVAL;
1806 	rtm = nlmsg_data(nlh);
1807 
1808 	if (rtm->rtm_family != AF_MPLS) {
1809 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1810 		goto errout;
1811 	}
1812 	if (rtm->rtm_dst_len != 20) {
1813 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1814 		goto errout;
1815 	}
1816 	if (rtm->rtm_src_len != 0) {
1817 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1818 		goto errout;
1819 	}
1820 	if (rtm->rtm_tos != 0) {
1821 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1822 		goto errout;
1823 	}
1824 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1825 		NL_SET_ERR_MSG(extack,
1826 			       "MPLS only supports the main route table");
1827 		goto errout;
1828 	}
1829 	/* Any value is acceptable for rtm_protocol */
1830 
1831 	/* As mpls uses destination specific addresses
1832 	 * (or source specific address in the case of multicast)
1833 	 * all addresses have universal scope.
1834 	 */
1835 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1836 		NL_SET_ERR_MSG(extack,
1837 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1838 		goto errout;
1839 	}
1840 	if (rtm->rtm_type != RTN_UNICAST) {
1841 		NL_SET_ERR_MSG(extack,
1842 			       "Invalid route type - MPLS only supports UNICAST");
1843 		goto errout;
1844 	}
1845 	if (rtm->rtm_flags != 0) {
1846 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1847 		goto errout;
1848 	}
1849 
1850 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1851 	cfg->rc_protocol	= rtm->rtm_protocol;
1852 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1853 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1854 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1855 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1856 	cfg->rc_nlinfo.nlh	= nlh;
1857 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1858 
1859 	for (index = 0; index <= RTA_MAX; index++) {
1860 		struct nlattr *nla = tb[index];
1861 		if (!nla)
1862 			continue;
1863 
1864 		switch (index) {
1865 		case RTA_OIF:
1866 			cfg->rc_ifindex = nla_get_u32(nla);
1867 			break;
1868 		case RTA_NEWDST:
1869 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1870 					   &cfg->rc_output_labels,
1871 					   cfg->rc_output_label, extack))
1872 				goto errout;
1873 			break;
1874 		case RTA_DST:
1875 		{
1876 			u8 label_count;
1877 			if (nla_get_labels(nla, 1, &label_count,
1878 					   &cfg->rc_label, extack))
1879 				goto errout;
1880 
1881 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1882 					   &cfg->rc_label, extack))
1883 				goto errout;
1884 			break;
1885 		}
1886 		case RTA_GATEWAY:
1887 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1888 			goto errout;
1889 		case RTA_VIA:
1890 		{
1891 			if (nla_get_via(nla, &cfg->rc_via_alen,
1892 					&cfg->rc_via_table, cfg->rc_via,
1893 					extack))
1894 				goto errout;
1895 			break;
1896 		}
1897 		case RTA_MULTIPATH:
1898 		{
1899 			cfg->rc_mp = nla_data(nla);
1900 			cfg->rc_mp_len = nla_len(nla);
1901 			break;
1902 		}
1903 		case RTA_TTL_PROPAGATE:
1904 		{
1905 			u8 ttl_propagate = nla_get_u8(nla);
1906 
1907 			if (ttl_propagate > 1) {
1908 				NL_SET_ERR_MSG_ATTR(extack, nla,
1909 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1910 				goto errout;
1911 			}
1912 			cfg->rc_ttl_propagate = ttl_propagate ?
1913 				MPLS_TTL_PROP_ENABLED :
1914 				MPLS_TTL_PROP_DISABLED;
1915 			break;
1916 		}
1917 		default:
1918 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1919 			/* Unsupported attribute */
1920 			goto errout;
1921 		}
1922 	}
1923 
1924 	err = 0;
1925 errout:
1926 	return err;
1927 }
1928 
1929 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1930 			     struct netlink_ext_ack *extack)
1931 {
1932 	struct mpls_route_config *cfg;
1933 	int err;
1934 
1935 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1936 	if (!cfg)
1937 		return -ENOMEM;
1938 
1939 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1940 	if (err < 0)
1941 		goto out;
1942 
1943 	err = mpls_route_del(cfg, extack);
1944 out:
1945 	kfree(cfg);
1946 
1947 	return err;
1948 }
1949 
1950 
1951 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1952 			     struct netlink_ext_ack *extack)
1953 {
1954 	struct mpls_route_config *cfg;
1955 	int err;
1956 
1957 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1958 	if (!cfg)
1959 		return -ENOMEM;
1960 
1961 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1962 	if (err < 0)
1963 		goto out;
1964 
1965 	err = mpls_route_add(cfg, extack);
1966 out:
1967 	kfree(cfg);
1968 
1969 	return err;
1970 }
1971 
1972 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1973 			   u32 label, struct mpls_route *rt, int flags)
1974 {
1975 	struct net_device *dev;
1976 	struct nlmsghdr *nlh;
1977 	struct rtmsg *rtm;
1978 
1979 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1980 	if (nlh == NULL)
1981 		return -EMSGSIZE;
1982 
1983 	rtm = nlmsg_data(nlh);
1984 	rtm->rtm_family = AF_MPLS;
1985 	rtm->rtm_dst_len = 20;
1986 	rtm->rtm_src_len = 0;
1987 	rtm->rtm_tos = 0;
1988 	rtm->rtm_table = RT_TABLE_MAIN;
1989 	rtm->rtm_protocol = rt->rt_protocol;
1990 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1991 	rtm->rtm_type = RTN_UNICAST;
1992 	rtm->rtm_flags = 0;
1993 
1994 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1995 		goto nla_put_failure;
1996 
1997 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1998 		bool ttl_propagate =
1999 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2000 
2001 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2002 			       ttl_propagate))
2003 			goto nla_put_failure;
2004 	}
2005 	if (rt->rt_nhn == 1) {
2006 		const struct mpls_nh *nh = rt->rt_nh;
2007 
2008 		if (nh->nh_labels &&
2009 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2010 				   nh->nh_label))
2011 			goto nla_put_failure;
2012 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2013 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2014 				nh->nh_via_alen))
2015 			goto nla_put_failure;
2016 		dev = nh->nh_dev;
2017 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2018 			goto nla_put_failure;
2019 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2020 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2021 		if (nh->nh_flags & RTNH_F_DEAD)
2022 			rtm->rtm_flags |= RTNH_F_DEAD;
2023 	} else {
2024 		struct rtnexthop *rtnh;
2025 		struct nlattr *mp;
2026 		u8 linkdown = 0;
2027 		u8 dead = 0;
2028 
2029 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2030 		if (!mp)
2031 			goto nla_put_failure;
2032 
2033 		for_nexthops(rt) {
2034 			dev = nh->nh_dev;
2035 			if (!dev)
2036 				continue;
2037 
2038 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2039 			if (!rtnh)
2040 				goto nla_put_failure;
2041 
2042 			rtnh->rtnh_ifindex = dev->ifindex;
2043 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2044 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2045 				linkdown++;
2046 			}
2047 			if (nh->nh_flags & RTNH_F_DEAD) {
2048 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2049 				dead++;
2050 			}
2051 
2052 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2053 							    nh->nh_labels,
2054 							    nh->nh_label))
2055 				goto nla_put_failure;
2056 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2057 			    nla_put_via(skb, nh->nh_via_table,
2058 					mpls_nh_via(rt, nh),
2059 					nh->nh_via_alen))
2060 				goto nla_put_failure;
2061 
2062 			/* length of rtnetlink header + attributes */
2063 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2064 		} endfor_nexthops(rt);
2065 
2066 		if (linkdown == rt->rt_nhn)
2067 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2068 		if (dead == rt->rt_nhn)
2069 			rtm->rtm_flags |= RTNH_F_DEAD;
2070 
2071 		nla_nest_end(skb, mp);
2072 	}
2073 
2074 	nlmsg_end(skb, nlh);
2075 	return 0;
2076 
2077 nla_put_failure:
2078 	nlmsg_cancel(skb, nlh);
2079 	return -EMSGSIZE;
2080 }
2081 
2082 #if IS_ENABLED(CONFIG_INET)
2083 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2084 				   struct fib_dump_filter *filter,
2085 				   struct netlink_callback *cb)
2086 {
2087 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2088 }
2089 #else
2090 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2091 				   struct fib_dump_filter *filter,
2092 				   struct netlink_callback *cb)
2093 {
2094 	struct netlink_ext_ack *extack = cb->extack;
2095 	struct nlattr *tb[RTA_MAX + 1];
2096 	struct rtmsg *rtm;
2097 	int err, i;
2098 
2099 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2100 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2101 		return -EINVAL;
2102 	}
2103 
2104 	rtm = nlmsg_data(nlh);
2105 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2106 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2107 	    rtm->rtm_flags) {
2108 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2109 		return -EINVAL;
2110 	}
2111 
2112 	if (rtm->rtm_protocol) {
2113 		filter->protocol = rtm->rtm_protocol;
2114 		filter->filter_set = 1;
2115 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2116 	}
2117 
2118 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2119 					    rtm_mpls_policy, extack);
2120 	if (err < 0)
2121 		return err;
2122 
2123 	for (i = 0; i <= RTA_MAX; ++i) {
2124 		int ifindex;
2125 
2126 		if (i == RTA_OIF) {
2127 			ifindex = nla_get_u32(tb[i]);
2128 			filter->dev = __dev_get_by_index(net, ifindex);
2129 			if (!filter->dev)
2130 				return -ENODEV;
2131 			filter->filter_set = 1;
2132 		} else if (tb[i]) {
2133 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2134 			return -EINVAL;
2135 		}
2136 	}
2137 
2138 	return 0;
2139 }
2140 #endif
2141 
2142 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2143 			     const struct net_device *dev)
2144 {
2145 	if (rt->rt_nhn == 1) {
2146 		struct mpls_nh *nh = rt->rt_nh;
2147 
2148 		if (nh->nh_dev == dev)
2149 			return true;
2150 	} else {
2151 		for_nexthops(rt) {
2152 			if (nh->nh_dev == dev)
2153 				return true;
2154 		} endfor_nexthops(rt);
2155 	}
2156 
2157 	return false;
2158 }
2159 
2160 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2161 {
2162 	const struct nlmsghdr *nlh = cb->nlh;
2163 	struct net *net = sock_net(skb->sk);
2164 	struct mpls_route __rcu **platform_label;
2165 	struct fib_dump_filter filter = {
2166 		.rtnl_held = true,
2167 	};
2168 	unsigned int flags = NLM_F_MULTI;
2169 	size_t platform_labels;
2170 	unsigned int index;
2171 
2172 	ASSERT_RTNL();
2173 
2174 	if (cb->strict_check) {
2175 		int err;
2176 
2177 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2178 		if (err < 0)
2179 			return err;
2180 
2181 		/* for MPLS, there is only 1 table with fixed type and flags.
2182 		 * If either are set in the filter then return nothing.
2183 		 */
2184 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2185 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2186 		     filter.flags)
2187 			return skb->len;
2188 	}
2189 
2190 	index = cb->args[0];
2191 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2192 		index = MPLS_LABEL_FIRST_UNRESERVED;
2193 
2194 	platform_label = rtnl_dereference(net->mpls.platform_label);
2195 	platform_labels = net->mpls.platform_labels;
2196 
2197 	if (filter.filter_set)
2198 		flags |= NLM_F_DUMP_FILTERED;
2199 
2200 	for (; index < platform_labels; index++) {
2201 		struct mpls_route *rt;
2202 
2203 		rt = rtnl_dereference(platform_label[index]);
2204 		if (!rt)
2205 			continue;
2206 
2207 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2208 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2209 			continue;
2210 
2211 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2212 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2213 				    index, rt, flags) < 0)
2214 			break;
2215 	}
2216 	cb->args[0] = index;
2217 
2218 	return skb->len;
2219 }
2220 
2221 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2222 {
2223 	size_t payload =
2224 		NLMSG_ALIGN(sizeof(struct rtmsg))
2225 		+ nla_total_size(4)			/* RTA_DST */
2226 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2227 
2228 	if (rt->rt_nhn == 1) {
2229 		struct mpls_nh *nh = rt->rt_nh;
2230 
2231 		if (nh->nh_dev)
2232 			payload += nla_total_size(4); /* RTA_OIF */
2233 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2234 			payload += nla_total_size(2 + nh->nh_via_alen);
2235 		if (nh->nh_labels) /* RTA_NEWDST */
2236 			payload += nla_total_size(nh->nh_labels * 4);
2237 	} else {
2238 		/* each nexthop is packed in an attribute */
2239 		size_t nhsize = 0;
2240 
2241 		for_nexthops(rt) {
2242 			if (!nh->nh_dev)
2243 				continue;
2244 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2245 			/* RTA_VIA */
2246 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2247 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2248 			if (nh->nh_labels)
2249 				nhsize += nla_total_size(nh->nh_labels * 4);
2250 		} endfor_nexthops(rt);
2251 		/* nested attribute */
2252 		payload += nla_total_size(nhsize);
2253 	}
2254 
2255 	return payload;
2256 }
2257 
2258 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2259 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2260 		       unsigned int nlm_flags)
2261 {
2262 	struct sk_buff *skb;
2263 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2264 	int err = -ENOBUFS;
2265 
2266 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2267 	if (skb == NULL)
2268 		goto errout;
2269 
2270 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2271 	if (err < 0) {
2272 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2273 		WARN_ON(err == -EMSGSIZE);
2274 		kfree_skb(skb);
2275 		goto errout;
2276 	}
2277 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2278 
2279 	return;
2280 errout:
2281 	if (err < 0)
2282 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2283 }
2284 
2285 static int mpls_valid_getroute_req(struct sk_buff *skb,
2286 				   const struct nlmsghdr *nlh,
2287 				   struct nlattr **tb,
2288 				   struct netlink_ext_ack *extack)
2289 {
2290 	struct rtmsg *rtm;
2291 	int i, err;
2292 
2293 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2294 		NL_SET_ERR_MSG_MOD(extack,
2295 				   "Invalid header for get route request");
2296 		return -EINVAL;
2297 	}
2298 
2299 	if (!netlink_strict_get_check(skb))
2300 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2301 					      rtm_mpls_policy, extack);
2302 
2303 	rtm = nlmsg_data(nlh);
2304 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2305 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2306 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2307 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2308 		return -EINVAL;
2309 	}
2310 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2311 		NL_SET_ERR_MSG_MOD(extack,
2312 				   "Invalid flags for get route request");
2313 		return -EINVAL;
2314 	}
2315 
2316 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2317 					    rtm_mpls_policy, extack);
2318 	if (err)
2319 		return err;
2320 
2321 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2322 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2323 		return -EINVAL;
2324 	}
2325 
2326 	for (i = 0; i <= RTA_MAX; i++) {
2327 		if (!tb[i])
2328 			continue;
2329 
2330 		switch (i) {
2331 		case RTA_DST:
2332 		case RTA_NEWDST:
2333 			break;
2334 		default:
2335 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2336 			return -EINVAL;
2337 		}
2338 	}
2339 
2340 	return 0;
2341 }
2342 
2343 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2344 			 struct netlink_ext_ack *extack)
2345 {
2346 	struct net *net = sock_net(in_skb->sk);
2347 	u32 portid = NETLINK_CB(in_skb).portid;
2348 	u32 in_label = LABEL_NOT_SPECIFIED;
2349 	struct nlattr *tb[RTA_MAX + 1];
2350 	u32 labels[MAX_NEW_LABELS];
2351 	struct mpls_shim_hdr *hdr;
2352 	unsigned int hdr_size = 0;
2353 	const struct mpls_nh *nh;
2354 	struct net_device *dev;
2355 	struct mpls_route *rt;
2356 	struct rtmsg *rtm, *r;
2357 	struct nlmsghdr *nlh;
2358 	struct sk_buff *skb;
2359 	u8 n_labels;
2360 	int err;
2361 
2362 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2363 	if (err < 0)
2364 		goto errout;
2365 
2366 	rtm = nlmsg_data(in_nlh);
2367 
2368 	if (tb[RTA_DST]) {
2369 		u8 label_count;
2370 
2371 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2372 				   &in_label, extack)) {
2373 			err = -EINVAL;
2374 			goto errout;
2375 		}
2376 
2377 		if (!mpls_label_ok(net, &in_label, extack)) {
2378 			err = -EINVAL;
2379 			goto errout;
2380 		}
2381 	}
2382 
2383 	rt = mpls_route_input_rcu(net, in_label);
2384 	if (!rt) {
2385 		err = -ENETUNREACH;
2386 		goto errout;
2387 	}
2388 
2389 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2390 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2391 		if (!skb) {
2392 			err = -ENOBUFS;
2393 			goto errout;
2394 		}
2395 
2396 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2397 				      RTM_NEWROUTE, in_label, rt, 0);
2398 		if (err < 0) {
2399 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2400 			WARN_ON(err == -EMSGSIZE);
2401 			goto errout_free;
2402 		}
2403 
2404 		return rtnl_unicast(skb, net, portid);
2405 	}
2406 
2407 	if (tb[RTA_NEWDST]) {
2408 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2409 				   labels, extack) != 0) {
2410 			err = -EINVAL;
2411 			goto errout;
2412 		}
2413 
2414 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2415 	}
2416 
2417 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2418 	if (!skb) {
2419 		err = -ENOBUFS;
2420 		goto errout;
2421 	}
2422 
2423 	skb->protocol = htons(ETH_P_MPLS_UC);
2424 
2425 	if (hdr_size) {
2426 		bool bos;
2427 		int i;
2428 
2429 		if (skb_cow(skb, hdr_size)) {
2430 			err = -ENOBUFS;
2431 			goto errout_free;
2432 		}
2433 
2434 		skb_reserve(skb, hdr_size);
2435 		skb_push(skb, hdr_size);
2436 		skb_reset_network_header(skb);
2437 
2438 		/* Push new labels */
2439 		hdr = mpls_hdr(skb);
2440 		bos = true;
2441 		for (i = n_labels - 1; i >= 0; i--) {
2442 			hdr[i] = mpls_entry_encode(labels[i],
2443 						   1, 0, bos);
2444 			bos = false;
2445 		}
2446 	}
2447 
2448 	nh = mpls_select_multipath(rt, skb);
2449 	if (!nh) {
2450 		err = -ENETUNREACH;
2451 		goto errout_free;
2452 	}
2453 
2454 	if (hdr_size) {
2455 		skb_pull(skb, hdr_size);
2456 		skb_reset_network_header(skb);
2457 	}
2458 
2459 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2460 			RTM_NEWROUTE, sizeof(*r), 0);
2461 	if (!nlh) {
2462 		err = -EMSGSIZE;
2463 		goto errout_free;
2464 	}
2465 
2466 	r = nlmsg_data(nlh);
2467 	r->rtm_family	 = AF_MPLS;
2468 	r->rtm_dst_len	= 20;
2469 	r->rtm_src_len	= 0;
2470 	r->rtm_table	= RT_TABLE_MAIN;
2471 	r->rtm_type	= RTN_UNICAST;
2472 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2473 	r->rtm_protocol = rt->rt_protocol;
2474 	r->rtm_flags	= 0;
2475 
2476 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2477 		goto nla_put_failure;
2478 
2479 	if (nh->nh_labels &&
2480 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2481 			   nh->nh_label))
2482 		goto nla_put_failure;
2483 
2484 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2485 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2486 			nh->nh_via_alen))
2487 		goto nla_put_failure;
2488 	dev = nh->nh_dev;
2489 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2490 		goto nla_put_failure;
2491 
2492 	nlmsg_end(skb, nlh);
2493 
2494 	err = rtnl_unicast(skb, net, portid);
2495 errout:
2496 	return err;
2497 
2498 nla_put_failure:
2499 	nlmsg_cancel(skb, nlh);
2500 	err = -EMSGSIZE;
2501 errout_free:
2502 	kfree_skb(skb);
2503 	return err;
2504 }
2505 
2506 static int resize_platform_label_table(struct net *net, size_t limit)
2507 {
2508 	size_t size = sizeof(struct mpls_route *) * limit;
2509 	size_t old_limit;
2510 	size_t cp_size;
2511 	struct mpls_route __rcu **labels = NULL, **old;
2512 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2513 	unsigned index;
2514 
2515 	if (size) {
2516 		labels = kvzalloc(size, GFP_KERNEL);
2517 		if (!labels)
2518 			goto nolabels;
2519 	}
2520 
2521 	/* In case the predefined labels need to be populated */
2522 	if (limit > MPLS_LABEL_IPV4NULL) {
2523 		struct net_device *lo = net->loopback_dev;
2524 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2525 		if (IS_ERR(rt0))
2526 			goto nort0;
2527 		rt0->rt_nh->nh_dev = lo;
2528 		rt0->rt_protocol = RTPROT_KERNEL;
2529 		rt0->rt_payload_type = MPT_IPV4;
2530 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2531 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2532 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2533 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2534 		       lo->addr_len);
2535 	}
2536 	if (limit > MPLS_LABEL_IPV6NULL) {
2537 		struct net_device *lo = net->loopback_dev;
2538 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2539 		if (IS_ERR(rt2))
2540 			goto nort2;
2541 		rt2->rt_nh->nh_dev = lo;
2542 		rt2->rt_protocol = RTPROT_KERNEL;
2543 		rt2->rt_payload_type = MPT_IPV6;
2544 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2545 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2546 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2547 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2548 		       lo->addr_len);
2549 	}
2550 
2551 	rtnl_lock();
2552 	/* Remember the original table */
2553 	old = rtnl_dereference(net->mpls.platform_label);
2554 	old_limit = net->mpls.platform_labels;
2555 
2556 	/* Free any labels beyond the new table */
2557 	for (index = limit; index < old_limit; index++)
2558 		mpls_route_update(net, index, NULL, NULL);
2559 
2560 	/* Copy over the old labels */
2561 	cp_size = size;
2562 	if (old_limit < limit)
2563 		cp_size = old_limit * sizeof(struct mpls_route *);
2564 
2565 	memcpy(labels, old, cp_size);
2566 
2567 	/* If needed set the predefined labels */
2568 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2569 	    (limit > MPLS_LABEL_IPV6NULL)) {
2570 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2571 		rt2 = NULL;
2572 	}
2573 
2574 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2575 	    (limit > MPLS_LABEL_IPV4NULL)) {
2576 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2577 		rt0 = NULL;
2578 	}
2579 
2580 	/* Update the global pointers */
2581 	net->mpls.platform_labels = limit;
2582 	rcu_assign_pointer(net->mpls.platform_label, labels);
2583 
2584 	rtnl_unlock();
2585 
2586 	mpls_rt_free(rt2);
2587 	mpls_rt_free(rt0);
2588 
2589 	if (old) {
2590 		synchronize_rcu();
2591 		kvfree(old);
2592 	}
2593 	return 0;
2594 
2595 nort2:
2596 	mpls_rt_free(rt0);
2597 nort0:
2598 	kvfree(labels);
2599 nolabels:
2600 	return -ENOMEM;
2601 }
2602 
2603 static int mpls_platform_labels(struct ctl_table *table, int write,
2604 				void *buffer, size_t *lenp, loff_t *ppos)
2605 {
2606 	struct net *net = table->data;
2607 	int platform_labels = net->mpls.platform_labels;
2608 	int ret;
2609 	struct ctl_table tmp = {
2610 		.procname	= table->procname,
2611 		.data		= &platform_labels,
2612 		.maxlen		= sizeof(int),
2613 		.mode		= table->mode,
2614 		.extra1		= SYSCTL_ZERO,
2615 		.extra2		= &label_limit,
2616 	};
2617 
2618 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2619 
2620 	if (write && ret == 0)
2621 		ret = resize_platform_label_table(net, platform_labels);
2622 
2623 	return ret;
2624 }
2625 
2626 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2627 	(&((struct net *)0)->field)
2628 
2629 static const struct ctl_table mpls_table[] = {
2630 	{
2631 		.procname	= "platform_labels",
2632 		.data		= NULL,
2633 		.maxlen		= sizeof(int),
2634 		.mode		= 0644,
2635 		.proc_handler	= mpls_platform_labels,
2636 	},
2637 	{
2638 		.procname	= "ip_ttl_propagate",
2639 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2640 		.maxlen		= sizeof(int),
2641 		.mode		= 0644,
2642 		.proc_handler	= proc_dointvec_minmax,
2643 		.extra1		= SYSCTL_ZERO,
2644 		.extra2		= SYSCTL_ONE,
2645 	},
2646 	{
2647 		.procname	= "default_ttl",
2648 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2649 		.maxlen		= sizeof(int),
2650 		.mode		= 0644,
2651 		.proc_handler	= proc_dointvec_minmax,
2652 		.extra1		= SYSCTL_ONE,
2653 		.extra2		= &ttl_max,
2654 	},
2655 };
2656 
2657 static int mpls_net_init(struct net *net)
2658 {
2659 	size_t table_size = ARRAY_SIZE(mpls_table);
2660 	struct ctl_table *table;
2661 	int i;
2662 
2663 	net->mpls.platform_labels = 0;
2664 	net->mpls.platform_label = NULL;
2665 	net->mpls.ip_ttl_propagate = 1;
2666 	net->mpls.default_ttl = 255;
2667 
2668 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2669 	if (table == NULL)
2670 		return -ENOMEM;
2671 
2672 	/* Table data contains only offsets relative to the base of
2673 	 * the mdev at this point, so make them absolute.
2674 	 */
2675 	for (i = 0; i < table_size; i++)
2676 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2677 
2678 	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2679 					       table_size);
2680 	if (net->mpls.ctl == NULL) {
2681 		kfree(table);
2682 		return -ENOMEM;
2683 	}
2684 
2685 	return 0;
2686 }
2687 
2688 static void mpls_net_exit(struct net *net)
2689 {
2690 	struct mpls_route __rcu **platform_label;
2691 	size_t platform_labels;
2692 	const struct ctl_table *table;
2693 	unsigned int index;
2694 
2695 	table = net->mpls.ctl->ctl_table_arg;
2696 	unregister_net_sysctl_table(net->mpls.ctl);
2697 	kfree(table);
2698 
2699 	/* An rcu grace period has passed since there was a device in
2700 	 * the network namespace (and thus the last in flight packet)
2701 	 * left this network namespace.  This is because
2702 	 * unregister_netdevice_many and netdev_run_todo has completed
2703 	 * for each network device that was in this network namespace.
2704 	 *
2705 	 * As such no additional rcu synchronization is necessary when
2706 	 * freeing the platform_label table.
2707 	 */
2708 	rtnl_lock();
2709 	platform_label = rtnl_dereference(net->mpls.platform_label);
2710 	platform_labels = net->mpls.platform_labels;
2711 	for (index = 0; index < platform_labels; index++) {
2712 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2713 		RCU_INIT_POINTER(platform_label[index], NULL);
2714 		mpls_notify_route(net, index, rt, NULL, NULL);
2715 		mpls_rt_free(rt);
2716 	}
2717 	rtnl_unlock();
2718 
2719 	kvfree(platform_label);
2720 }
2721 
2722 static struct pernet_operations mpls_net_ops = {
2723 	.init = mpls_net_init,
2724 	.exit = mpls_net_exit,
2725 };
2726 
2727 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2728 	.family		   = AF_MPLS,
2729 	.fill_stats_af	   = mpls_fill_stats_af,
2730 	.get_stats_af_size = mpls_get_stats_af_size,
2731 };
2732 
2733 static int __init mpls_init(void)
2734 {
2735 	int err;
2736 
2737 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2738 
2739 	err = register_pernet_subsys(&mpls_net_ops);
2740 	if (err)
2741 		goto out;
2742 
2743 	err = register_netdevice_notifier(&mpls_dev_notifier);
2744 	if (err)
2745 		goto out_unregister_pernet;
2746 
2747 	dev_add_pack(&mpls_packet_type);
2748 
2749 	rtnl_af_register(&mpls_af_ops);
2750 
2751 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2752 			     mpls_rtm_newroute, NULL, 0);
2753 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2754 			     mpls_rtm_delroute, NULL, 0);
2755 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2756 			     mpls_getroute, mpls_dump_routes, 0);
2757 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2758 			     mpls_netconf_get_devconf,
2759 			     mpls_netconf_dump_devconf,
2760 			     RTNL_FLAG_DUMP_UNLOCKED);
2761 	err = ipgre_tunnel_encap_add_mpls_ops();
2762 	if (err)
2763 		pr_err("Can't add mpls over gre tunnel ops\n");
2764 
2765 	err = 0;
2766 out:
2767 	return err;
2768 
2769 out_unregister_pernet:
2770 	unregister_pernet_subsys(&mpls_net_ops);
2771 	goto out;
2772 }
2773 module_init(mpls_init);
2774 
2775 static void __exit mpls_exit(void)
2776 {
2777 	rtnl_unregister_all(PF_MPLS);
2778 	rtnl_af_unregister(&mpls_af_ops);
2779 	dev_remove_pack(&mpls_packet_type);
2780 	unregister_netdevice_notifier(&mpls_dev_notifier);
2781 	unregister_pernet_subsys(&mpls_net_ops);
2782 	ipgre_tunnel_encap_del_mpls_ops();
2783 }
2784 module_exit(mpls_exit);
2785 
2786 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2787 MODULE_LICENSE("GPL v2");
2788 MODULE_ALIAS_NETPROTO(PF_MPLS);
2789