xref: /linux/net/mpls/af_mpls.c (revision 330296ea9e158758aa65631f5ec64aa74806b7e2)
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(struct net *net, unsigned int index)
79 {
80 	struct mpls_route __rcu **platform_label;
81 
82 	platform_label = mpls_dereference(net, net->mpls.platform_label);
83 	return mpls_dereference(net, platform_label[index]);
84 }
85 
86 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned int index)
87 {
88 	struct mpls_route __rcu **platform_label;
89 
90 	if (index >= net->mpls.platform_labels)
91 		return NULL;
92 
93 	platform_label = rcu_dereference(net->mpls.platform_label);
94 	return rcu_dereference(platform_label[index]);
95 }
96 
97 bool mpls_output_possible(const struct net_device *dev)
98 {
99 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
100 }
101 EXPORT_SYMBOL_GPL(mpls_output_possible);
102 
103 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
104 {
105 	return (u8 *)nh + rt->rt_via_offset;
106 }
107 
108 static const u8 *mpls_nh_via(const struct mpls_route *rt,
109 			     const struct mpls_nh *nh)
110 {
111 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
112 }
113 
114 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
115 {
116 	/* The size of the layer 2.5 labels to be added for this route */
117 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
118 }
119 
120 unsigned int mpls_dev_mtu(const struct net_device *dev)
121 {
122 	/* The amount of data the layer 2 frame can hold */
123 	return dev->mtu;
124 }
125 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
126 
127 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
128 {
129 	if (skb->len <= mtu)
130 		return false;
131 
132 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
133 		return false;
134 
135 	return true;
136 }
137 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
138 
139 void mpls_stats_inc_outucastpkts(struct net *net,
140 				 struct net_device *dev,
141 				 const struct sk_buff *skb)
142 {
143 	struct mpls_dev *mdev;
144 
145 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
146 		mdev = mpls_dev_rcu(dev);
147 		if (mdev)
148 			MPLS_INC_STATS_LEN(mdev, skb->len,
149 					   tx_packets,
150 					   tx_bytes);
151 	} else if (skb->protocol == htons(ETH_P_IP)) {
152 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
153 #if IS_ENABLED(CONFIG_IPV6)
154 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
155 		struct inet6_dev *in6dev = in6_dev_rcu(dev);
156 
157 		if (in6dev)
158 			IP6_UPD_PO_STATS(net, in6dev,
159 					 IPSTATS_MIB_OUT, skb->len);
160 #endif
161 	}
162 }
163 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
164 
165 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
166 {
167 	struct mpls_entry_decoded dec;
168 	unsigned int mpls_hdr_len = 0;
169 	struct mpls_shim_hdr *hdr;
170 	bool eli_seen = false;
171 	int label_index;
172 	u32 hash = 0;
173 
174 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
175 	     label_index++) {
176 		mpls_hdr_len += sizeof(*hdr);
177 		if (!pskb_may_pull(skb, mpls_hdr_len))
178 			break;
179 
180 		/* Read and decode the current label */
181 		hdr = mpls_hdr(skb) + label_index;
182 		dec = mpls_entry_decode(hdr);
183 
184 		/* RFC6790 - reserved labels MUST NOT be used as keys
185 		 * for the load-balancing function
186 		 */
187 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
188 			hash = jhash_1word(dec.label, hash);
189 
190 			/* The entropy label follows the entropy label
191 			 * indicator, so this means that the entropy
192 			 * label was just added to the hash - no need to
193 			 * go any deeper either in the label stack or in the
194 			 * payload
195 			 */
196 			if (eli_seen)
197 				break;
198 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
199 			eli_seen = true;
200 		}
201 
202 		if (!dec.bos)
203 			continue;
204 
205 		/* found bottom label; does skb have room for a header? */
206 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
207 			const struct iphdr *v4hdr;
208 
209 			v4hdr = (const struct iphdr *)(hdr + 1);
210 			if (v4hdr->version == 4) {
211 				hash = jhash_3words(ntohl(v4hdr->saddr),
212 						    ntohl(v4hdr->daddr),
213 						    v4hdr->protocol, hash);
214 			} else if (v4hdr->version == 6 &&
215 				   pskb_may_pull(skb, mpls_hdr_len +
216 						 sizeof(struct ipv6hdr))) {
217 				const struct ipv6hdr *v6hdr;
218 
219 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
220 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
221 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
222 				hash = jhash_1word(v6hdr->nexthdr, hash);
223 			}
224 		}
225 
226 		break;
227 	}
228 
229 	return hash;
230 }
231 
232 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
233 {
234 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
235 }
236 
237 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
238  * a next hop (nh->nh_flags) are modified by netdev event handlers.
239  * Since those fields can change at any moment, use READ_ONCE to
240  * access both.
241  */
242 static const struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
243 						   struct sk_buff *skb)
244 {
245 	u32 hash = 0;
246 	int nh_index = 0;
247 	int n = 0;
248 	u8 alive;
249 
250 	/* No need to look further into packet if there's only
251 	 * one path
252 	 */
253 	if (rt->rt_nhn == 1)
254 		return rt->rt_nh;
255 
256 	alive = READ_ONCE(rt->rt_nhn_alive);
257 	if (alive == 0)
258 		return NULL;
259 
260 	hash = mpls_multipath_hash(rt, skb);
261 	nh_index = hash % alive;
262 	if (alive == rt->rt_nhn)
263 		goto out;
264 	for_nexthops(rt) {
265 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
266 
267 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
268 			continue;
269 		if (n == nh_index)
270 			return nh;
271 		n++;
272 	} endfor_nexthops(rt);
273 
274 out:
275 	return mpls_get_nexthop(rt, nh_index);
276 }
277 
278 static bool mpls_egress(struct net *net, struct mpls_route *rt,
279 			struct sk_buff *skb, struct mpls_entry_decoded dec)
280 {
281 	enum mpls_payload_type payload_type;
282 	bool success = false;
283 
284 	/* The IPv4 code below accesses through the IPv4 header
285 	 * checksum, which is 12 bytes into the packet.
286 	 * The IPv6 code below accesses through the IPv6 hop limit
287 	 * which is 8 bytes into the packet.
288 	 *
289 	 * For all supported cases there should always be at least 12
290 	 * bytes of packet data present.  The IPv4 header is 20 bytes
291 	 * without options and the IPv6 header is always 40 bytes
292 	 * long.
293 	 */
294 	if (!pskb_may_pull(skb, 12))
295 		return false;
296 
297 	payload_type = rt->rt_payload_type;
298 	if (payload_type == MPT_UNSPEC)
299 		payload_type = ip_hdr(skb)->version;
300 
301 	switch (payload_type) {
302 	case MPT_IPV4: {
303 		struct iphdr *hdr4 = ip_hdr(skb);
304 		u8 new_ttl;
305 		skb->protocol = htons(ETH_P_IP);
306 
307 		/* If propagating TTL, take the decremented TTL from
308 		 * the incoming MPLS header, otherwise decrement the
309 		 * TTL, but only if not 0 to avoid underflow.
310 		 */
311 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
312 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
313 		     net->mpls.ip_ttl_propagate))
314 			new_ttl = dec.ttl;
315 		else
316 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
317 
318 		csum_replace2(&hdr4->check,
319 			      htons(hdr4->ttl << 8),
320 			      htons(new_ttl << 8));
321 		hdr4->ttl = new_ttl;
322 		success = true;
323 		break;
324 	}
325 	case MPT_IPV6: {
326 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
327 		skb->protocol = htons(ETH_P_IPV6);
328 
329 		/* If propagating TTL, take the decremented TTL from
330 		 * the incoming MPLS header, otherwise decrement the
331 		 * hop limit, but only if not 0 to avoid underflow.
332 		 */
333 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
334 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
335 		     net->mpls.ip_ttl_propagate))
336 			hdr6->hop_limit = dec.ttl;
337 		else if (hdr6->hop_limit)
338 			hdr6->hop_limit = hdr6->hop_limit - 1;
339 		success = true;
340 		break;
341 	}
342 	case MPT_UNSPEC:
343 		/* Should have decided which protocol it is by now */
344 		break;
345 	}
346 
347 	return success;
348 }
349 
350 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
351 			struct packet_type *pt, struct net_device *orig_dev)
352 {
353 	struct net *net = dev_net_rcu(dev);
354 	struct mpls_shim_hdr *hdr;
355 	const struct mpls_nh *nh;
356 	struct mpls_route *rt;
357 	struct mpls_entry_decoded dec;
358 	struct net_device *out_dev;
359 	struct mpls_dev *out_mdev;
360 	struct mpls_dev *mdev;
361 	unsigned int hh_len;
362 	unsigned int new_header_size;
363 	unsigned int mtu;
364 	int err;
365 
366 	/* Careful this entire function runs inside of an rcu critical section */
367 
368 	mdev = mpls_dev_rcu(dev);
369 	if (!mdev)
370 		goto drop;
371 
372 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
373 			   rx_bytes);
374 
375 	if (!mdev->input_enabled) {
376 		MPLS_INC_STATS(mdev, rx_dropped);
377 		goto drop;
378 	}
379 
380 	if (skb->pkt_type != PACKET_HOST)
381 		goto err;
382 
383 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
384 		goto err;
385 
386 	if (!pskb_may_pull(skb, sizeof(*hdr)))
387 		goto err;
388 
389 	skb_dst_drop(skb);
390 
391 	/* Read and decode the label */
392 	hdr = mpls_hdr(skb);
393 	dec = mpls_entry_decode(hdr);
394 
395 	rt = mpls_route_input_rcu(net, dec.label);
396 	if (!rt) {
397 		MPLS_INC_STATS(mdev, rx_noroute);
398 		goto drop;
399 	}
400 
401 	nh = mpls_select_multipath(rt, skb);
402 	if (!nh)
403 		goto err;
404 
405 	/* Pop the label */
406 	skb_pull(skb, sizeof(*hdr));
407 	skb_reset_network_header(skb);
408 
409 	skb_orphan(skb);
410 
411 	if (skb_warn_if_lro(skb))
412 		goto err;
413 
414 	skb_forward_csum(skb);
415 
416 	/* Verify ttl is valid */
417 	if (dec.ttl <= 1)
418 		goto err;
419 
420 	/* Find the output device */
421 	out_dev = nh->nh_dev;
422 	if (!mpls_output_possible(out_dev))
423 		goto tx_err;
424 
425 	/* Verify the destination can hold the packet */
426 	new_header_size = mpls_nh_header_size(nh);
427 	mtu = mpls_dev_mtu(out_dev);
428 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
429 		goto tx_err;
430 
431 	hh_len = LL_RESERVED_SPACE(out_dev);
432 	if (!out_dev->header_ops)
433 		hh_len = 0;
434 
435 	/* Ensure there is enough space for the headers in the skb */
436 	if (skb_cow(skb, hh_len + new_header_size))
437 		goto tx_err;
438 
439 	skb->dev = out_dev;
440 	skb->protocol = htons(ETH_P_MPLS_UC);
441 
442 	dec.ttl -= 1;
443 	if (unlikely(!new_header_size && dec.bos)) {
444 		/* Penultimate hop popping */
445 		if (!mpls_egress(net, rt, skb, dec))
446 			goto err;
447 	} else {
448 		bool bos;
449 		int i;
450 		skb_push(skb, new_header_size);
451 		skb_reset_network_header(skb);
452 		/* Push the new labels */
453 		hdr = mpls_hdr(skb);
454 		bos = dec.bos;
455 		for (i = nh->nh_labels - 1; i >= 0; i--) {
456 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
457 						   dec.ttl, 0, bos);
458 			bos = false;
459 		}
460 	}
461 
462 	mpls_stats_inc_outucastpkts(net, out_dev, skb);
463 
464 	/* If via wasn't specified then send out using device address */
465 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
466 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
467 				 out_dev->dev_addr, skb);
468 	else
469 		err = neigh_xmit(nh->nh_via_table, out_dev,
470 				 mpls_nh_via(rt, nh), skb);
471 	if (err)
472 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
473 				    __func__, err);
474 	return 0;
475 
476 tx_err:
477 	out_mdev = out_dev ? mpls_dev_rcu(out_dev) : NULL;
478 	if (out_mdev)
479 		MPLS_INC_STATS(out_mdev, tx_errors);
480 	goto drop;
481 err:
482 	MPLS_INC_STATS(mdev, rx_errors);
483 drop:
484 	kfree_skb(skb);
485 	return NET_RX_DROP;
486 }
487 
488 static struct packet_type mpls_packet_type __read_mostly = {
489 	.type = cpu_to_be16(ETH_P_MPLS_UC),
490 	.func = mpls_forward,
491 };
492 
493 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
494 	[RTA_DST]		= { .type = NLA_U32 },
495 	[RTA_OIF]		= { .type = NLA_U32 },
496 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
497 };
498 
499 struct mpls_route_config {
500 	u32			rc_protocol;
501 	u32			rc_ifindex;
502 	u8			rc_via_table;
503 	u8			rc_via_alen;
504 	u8			rc_via[MAX_VIA_ALEN];
505 	u32			rc_label;
506 	u8			rc_ttl_propagate;
507 	u8			rc_output_labels;
508 	u32			rc_output_label[MAX_NEW_LABELS];
509 	u32			rc_nlflags;
510 	enum mpls_payload_type	rc_payload_type;
511 	struct nl_info		rc_nlinfo;
512 	struct rtnexthop	*rc_mp;
513 	int			rc_mp_len;
514 };
515 
516 /* all nexthops within a route have the same size based on max
517  * number of labels and max via length for a hop
518  */
519 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
520 {
521 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
522 	struct mpls_route *rt;
523 	size_t size;
524 
525 	size = sizeof(*rt) + num_nh * nh_size;
526 	if (size > MAX_MPLS_ROUTE_MEM)
527 		return ERR_PTR(-EINVAL);
528 
529 	rt = kzalloc(size, GFP_KERNEL);
530 	if (!rt)
531 		return ERR_PTR(-ENOMEM);
532 
533 	rt->rt_nhn = num_nh;
534 	rt->rt_nhn_alive = num_nh;
535 	rt->rt_nh_size = nh_size;
536 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
537 
538 	return rt;
539 }
540 
541 static void mpls_rt_free_rcu(struct rcu_head *head)
542 {
543 	struct mpls_route *rt;
544 
545 	rt = container_of(head, struct mpls_route, rt_rcu);
546 
547 	change_nexthops(rt) {
548 		netdev_put(nh->nh_dev, &nh->nh_dev_tracker);
549 	} endfor_nexthops(rt);
550 
551 	kfree(rt);
552 }
553 
554 static void mpls_rt_free(struct mpls_route *rt)
555 {
556 	if (rt)
557 		call_rcu(&rt->rt_rcu, mpls_rt_free_rcu);
558 }
559 
560 static void mpls_notify_route(struct net *net, unsigned index,
561 			      struct mpls_route *old, struct mpls_route *new,
562 			      const struct nl_info *info)
563 {
564 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
565 	unsigned portid = info ? info->portid : 0;
566 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
567 	struct mpls_route *rt = new ? new : old;
568 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
569 	/* Ignore reserved labels for now */
570 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
571 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
572 }
573 
574 static void mpls_route_update(struct net *net, unsigned index,
575 			      struct mpls_route *new,
576 			      const struct nl_info *info)
577 {
578 	struct mpls_route __rcu **platform_label;
579 	struct mpls_route *rt;
580 
581 	platform_label = mpls_dereference(net, net->mpls.platform_label);
582 	rt = mpls_dereference(net, platform_label[index]);
583 	rcu_assign_pointer(platform_label[index], new);
584 
585 	mpls_notify_route(net, index, rt, new, info);
586 
587 	/* If we removed a route free it now */
588 	mpls_rt_free(rt);
589 }
590 
591 static unsigned int find_free_label(struct net *net)
592 {
593 	unsigned int index;
594 
595 	for (index = MPLS_LABEL_FIRST_UNRESERVED;
596 	     index < net->mpls.platform_labels;
597 	     index++) {
598 		if (!mpls_route_input(net, index))
599 			return index;
600 	}
601 
602 	return LABEL_NOT_SPECIFIED;
603 }
604 
605 #if IS_ENABLED(CONFIG_INET)
606 static struct net_device *inet_fib_lookup_dev(struct net *net,
607 					      struct mpls_nh *nh,
608 					      const void *addr)
609 {
610 	struct net_device *dev;
611 	struct rtable *rt;
612 	struct in_addr daddr;
613 
614 	memcpy(&daddr, addr, sizeof(struct in_addr));
615 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0, RT_SCOPE_UNIVERSE);
616 	if (IS_ERR(rt))
617 		return ERR_CAST(rt);
618 
619 	dev = rt->dst.dev;
620 	netdev_hold(dev, &nh->nh_dev_tracker, GFP_KERNEL);
621 	ip_rt_put(rt);
622 
623 	return dev;
624 }
625 #else
626 static struct net_device *inet_fib_lookup_dev(struct net *net,
627 					      struct mpls_nh *nh,
628 					      const void *addr)
629 {
630 	return ERR_PTR(-EAFNOSUPPORT);
631 }
632 #endif
633 
634 #if IS_ENABLED(CONFIG_IPV6)
635 static struct net_device *inet6_fib_lookup_dev(struct net *net,
636 					       struct mpls_nh *nh,
637 					       const void *addr)
638 {
639 	struct net_device *dev;
640 	struct dst_entry *dst;
641 	struct flowi6 fl6;
642 
643 	memset(&fl6, 0, sizeof(fl6));
644 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
645 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
646 	if (IS_ERR(dst))
647 		return ERR_CAST(dst);
648 
649 	dev = dst->dev;
650 	netdev_hold(dev, &nh->nh_dev_tracker, GFP_KERNEL);
651 	dst_release(dst);
652 
653 	return dev;
654 }
655 #else
656 static struct net_device *inet6_fib_lookup_dev(struct net *net,
657 					       struct mpls_nh *nh,
658 					       const void *addr)
659 {
660 	return ERR_PTR(-EAFNOSUPPORT);
661 }
662 #endif
663 
664 static struct net_device *find_outdev(struct net *net,
665 				      struct mpls_route *rt,
666 				      struct mpls_nh *nh, int oif)
667 {
668 	struct net_device *dev = NULL;
669 
670 	if (!oif) {
671 		switch (nh->nh_via_table) {
672 		case NEIGH_ARP_TABLE:
673 			dev = inet_fib_lookup_dev(net, nh, mpls_nh_via(rt, nh));
674 			break;
675 		case NEIGH_ND_TABLE:
676 			dev = inet6_fib_lookup_dev(net, nh, mpls_nh_via(rt, nh));
677 			break;
678 		case NEIGH_LINK_TABLE:
679 			break;
680 		}
681 	} else {
682 		dev = netdev_get_by_index(net, oif,
683 					  &nh->nh_dev_tracker, GFP_KERNEL);
684 	}
685 
686 	if (!dev)
687 		return ERR_PTR(-ENODEV);
688 
689 	if (IS_ERR(dev))
690 		return dev;
691 
692 	nh->nh_dev = dev;
693 
694 	return dev;
695 }
696 
697 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
698 			      struct mpls_nh *nh, int oif)
699 {
700 	struct net_device *dev = NULL;
701 	int err = -ENODEV;
702 
703 	dev = find_outdev(net, rt, nh, oif);
704 	if (IS_ERR(dev)) {
705 		err = PTR_ERR(dev);
706 		goto errout;
707 	}
708 
709 	/* Ensure this is a supported device */
710 	err = -EINVAL;
711 	if (!mpls_dev_get(net, dev))
712 		goto errout_put;
713 
714 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
715 	    (dev->addr_len != nh->nh_via_alen))
716 		goto errout_put;
717 
718 	if (!(dev->flags & IFF_UP)) {
719 		nh->nh_flags |= RTNH_F_DEAD;
720 	} else {
721 		unsigned int flags;
722 
723 		flags = netif_get_flags(dev);
724 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
725 			nh->nh_flags |= RTNH_F_LINKDOWN;
726 	}
727 
728 	return 0;
729 
730 errout_put:
731 	netdev_put(nh->nh_dev, &nh->nh_dev_tracker);
732 	nh->nh_dev = NULL;
733 errout:
734 	return err;
735 }
736 
737 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
738 		       u8 via_addr[], struct netlink_ext_ack *extack)
739 {
740 	struct rtvia *via = nla_data(nla);
741 	int err = -EINVAL;
742 	int alen;
743 
744 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
745 		NL_SET_ERR_MSG_ATTR(extack, nla,
746 				    "Invalid attribute length for RTA_VIA");
747 		goto errout;
748 	}
749 	alen = nla_len(nla) -
750 			offsetof(struct rtvia, rtvia_addr);
751 	if (alen > MAX_VIA_ALEN) {
752 		NL_SET_ERR_MSG_ATTR(extack, nla,
753 				    "Invalid address length for RTA_VIA");
754 		goto errout;
755 	}
756 
757 	/* Validate the address family */
758 	switch (via->rtvia_family) {
759 	case AF_PACKET:
760 		*via_table = NEIGH_LINK_TABLE;
761 		break;
762 	case AF_INET:
763 		*via_table = NEIGH_ARP_TABLE;
764 		if (alen != 4)
765 			goto errout;
766 		break;
767 	case AF_INET6:
768 		*via_table = NEIGH_ND_TABLE;
769 		if (alen != 16)
770 			goto errout;
771 		break;
772 	default:
773 		/* Unsupported address family */
774 		goto errout;
775 	}
776 
777 	memcpy(via_addr, via->rtvia_addr, alen);
778 	*via_alen = alen;
779 	err = 0;
780 
781 errout:
782 	return err;
783 }
784 
785 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
786 				  struct mpls_route *rt)
787 {
788 	struct net *net = cfg->rc_nlinfo.nl_net;
789 	struct mpls_nh *nh = rt->rt_nh;
790 	int err;
791 	int i;
792 
793 	if (!nh)
794 		return -ENOMEM;
795 
796 	nh->nh_labels = cfg->rc_output_labels;
797 	for (i = 0; i < nh->nh_labels; i++)
798 		nh->nh_label[i] = cfg->rc_output_label[i];
799 
800 	nh->nh_via_table = cfg->rc_via_table;
801 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
802 	nh->nh_via_alen = cfg->rc_via_alen;
803 
804 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
805 	if (err)
806 		goto errout;
807 
808 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
809 		rt->rt_nhn_alive--;
810 
811 	return 0;
812 
813 errout:
814 	return err;
815 }
816 
817 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
818 			 struct mpls_nh *nh, int oif, struct nlattr *via,
819 			 struct nlattr *newdst, u8 max_labels,
820 			 struct netlink_ext_ack *extack)
821 {
822 	int err = -ENOMEM;
823 
824 	if (!nh)
825 		goto errout;
826 
827 	if (newdst) {
828 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
829 				     nh->nh_label, extack);
830 		if (err)
831 			goto errout;
832 	}
833 
834 	if (via) {
835 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
836 				  __mpls_nh_via(rt, nh), extack);
837 		if (err)
838 			goto errout;
839 	} else {
840 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
841 	}
842 
843 	err = mpls_nh_assign_dev(net, rt, nh, oif);
844 	if (err)
845 		goto errout;
846 
847 	return 0;
848 
849 errout:
850 	return err;
851 }
852 
853 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
854 			      u8 cfg_via_alen, u8 *max_via_alen,
855 			      u8 *max_labels)
856 {
857 	int remaining = len;
858 	u8 nhs = 0;
859 
860 	*max_via_alen = 0;
861 	*max_labels = 0;
862 
863 	while (rtnh_ok(rtnh, remaining)) {
864 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
865 		int attrlen;
866 		u8 n_labels = 0;
867 
868 		attrlen = rtnh_attrlen(rtnh);
869 		nla = nla_find(attrs, attrlen, RTA_VIA);
870 		if (nla && nla_len(nla) >=
871 		    offsetof(struct rtvia, rtvia_addr)) {
872 			int via_alen = nla_len(nla) -
873 				offsetof(struct rtvia, rtvia_addr);
874 
875 			if (via_alen <= MAX_VIA_ALEN)
876 				*max_via_alen = max_t(u16, *max_via_alen,
877 						      via_alen);
878 		}
879 
880 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
881 		if (nla &&
882 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
883 				   NULL, NULL) != 0)
884 			return 0;
885 
886 		*max_labels = max_t(u8, *max_labels, n_labels);
887 
888 		/* number of nexthops is tracked by a u8.
889 		 * Check for overflow.
890 		 */
891 		if (nhs == 255)
892 			return 0;
893 		nhs++;
894 
895 		rtnh = rtnh_next(rtnh, &remaining);
896 	}
897 
898 	/* leftover implies invalid nexthop configuration, discard it */
899 	return remaining > 0 ? 0 : nhs;
900 }
901 
902 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
903 			       struct mpls_route *rt, u8 max_labels,
904 			       struct netlink_ext_ack *extack)
905 {
906 	struct rtnexthop *rtnh = cfg->rc_mp;
907 	struct nlattr *nla_via, *nla_newdst;
908 	int remaining = cfg->rc_mp_len;
909 	int err = 0;
910 
911 	rt->rt_nhn = 0;
912 
913 	change_nexthops(rt) {
914 		int attrlen;
915 
916 		nla_via = NULL;
917 		nla_newdst = NULL;
918 
919 		err = -EINVAL;
920 		if (!rtnh_ok(rtnh, remaining))
921 			goto errout;
922 
923 		/* neither weighted multipath nor any flags
924 		 * are supported
925 		 */
926 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
927 			goto errout;
928 
929 		attrlen = rtnh_attrlen(rtnh);
930 		if (attrlen > 0) {
931 			struct nlattr *attrs = rtnh_attrs(rtnh);
932 
933 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
934 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
935 		}
936 
937 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
938 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
939 				    max_labels, extack);
940 		if (err)
941 			goto errout;
942 
943 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
944 			rt->rt_nhn_alive--;
945 
946 		rtnh = rtnh_next(rtnh, &remaining);
947 		rt->rt_nhn++;
948 	} endfor_nexthops(rt);
949 
950 	return 0;
951 
952 errout:
953 	return err;
954 }
955 
956 static bool mpls_label_ok(struct net *net, unsigned int *index,
957 			  struct netlink_ext_ack *extack)
958 {
959 	/* Reserved labels may not be set */
960 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
961 		NL_SET_ERR_MSG(extack,
962 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
963 		return false;
964 	}
965 
966 	/* The full 20 bit range may not be supported. */
967 	if (*index >= net->mpls.platform_labels) {
968 		NL_SET_ERR_MSG(extack,
969 			       "Label >= configured maximum in platform_labels");
970 		return false;
971 	}
972 
973 	*index = array_index_nospec(*index, net->mpls.platform_labels);
974 
975 	return true;
976 }
977 
978 static int mpls_route_add(struct mpls_route_config *cfg,
979 			  struct netlink_ext_ack *extack)
980 {
981 	struct net *net = cfg->rc_nlinfo.nl_net;
982 	struct mpls_route *rt, *old;
983 	int err = -EINVAL;
984 	u8 max_via_alen;
985 	unsigned index;
986 	u8 max_labels;
987 	u8 nhs;
988 
989 	index = cfg->rc_label;
990 
991 	/* If a label was not specified during insert pick one */
992 	if ((index == LABEL_NOT_SPECIFIED) &&
993 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
994 		index = find_free_label(net);
995 	}
996 
997 	if (!mpls_label_ok(net, &index, extack))
998 		goto errout;
999 
1000 	/* Append makes no sense with mpls */
1001 	err = -EOPNOTSUPP;
1002 	if (cfg->rc_nlflags & NLM_F_APPEND) {
1003 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
1004 		goto errout;
1005 	}
1006 
1007 	err = -EEXIST;
1008 	old = mpls_route_input(net, index);
1009 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
1010 		goto errout;
1011 
1012 	err = -EEXIST;
1013 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1014 		goto errout;
1015 
1016 	err = -ENOENT;
1017 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1018 		goto errout;
1019 
1020 	err = -EINVAL;
1021 	if (cfg->rc_mp) {
1022 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1023 					  cfg->rc_via_alen, &max_via_alen,
1024 					  &max_labels);
1025 	} else {
1026 		max_via_alen = cfg->rc_via_alen;
1027 		max_labels = cfg->rc_output_labels;
1028 		nhs = 1;
1029 	}
1030 
1031 	if (nhs == 0) {
1032 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1033 		goto errout;
1034 	}
1035 
1036 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1037 	if (IS_ERR(rt)) {
1038 		err = PTR_ERR(rt);
1039 		goto errout;
1040 	}
1041 
1042 	rt->rt_protocol = cfg->rc_protocol;
1043 	rt->rt_payload_type = cfg->rc_payload_type;
1044 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1045 
1046 	if (cfg->rc_mp)
1047 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1048 	else
1049 		err = mpls_nh_build_from_cfg(cfg, rt);
1050 	if (err)
1051 		goto freert;
1052 
1053 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1054 
1055 	return 0;
1056 
1057 freert:
1058 	mpls_rt_free(rt);
1059 errout:
1060 	return err;
1061 }
1062 
1063 static int mpls_route_del(struct mpls_route_config *cfg,
1064 			  struct netlink_ext_ack *extack)
1065 {
1066 	struct net *net = cfg->rc_nlinfo.nl_net;
1067 	unsigned index;
1068 	int err = -EINVAL;
1069 
1070 	index = cfg->rc_label;
1071 
1072 	if (!mpls_label_ok(net, &index, extack))
1073 		goto errout;
1074 
1075 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1076 
1077 	err = 0;
1078 errout:
1079 	return err;
1080 }
1081 
1082 static void mpls_get_stats(struct mpls_dev *mdev,
1083 			   struct mpls_link_stats *stats)
1084 {
1085 	struct mpls_pcpu_stats *p;
1086 	int i;
1087 
1088 	memset(stats, 0, sizeof(*stats));
1089 
1090 	for_each_possible_cpu(i) {
1091 		struct mpls_link_stats local;
1092 		unsigned int start;
1093 
1094 		p = per_cpu_ptr(mdev->stats, i);
1095 		do {
1096 			start = u64_stats_fetch_begin(&p->syncp);
1097 			local = p->stats;
1098 		} while (u64_stats_fetch_retry(&p->syncp, start));
1099 
1100 		stats->rx_packets	+= local.rx_packets;
1101 		stats->rx_bytes		+= local.rx_bytes;
1102 		stats->tx_packets	+= local.tx_packets;
1103 		stats->tx_bytes		+= local.tx_bytes;
1104 		stats->rx_errors	+= local.rx_errors;
1105 		stats->tx_errors	+= local.tx_errors;
1106 		stats->rx_dropped	+= local.rx_dropped;
1107 		stats->tx_dropped	+= local.tx_dropped;
1108 		stats->rx_noroute	+= local.rx_noroute;
1109 	}
1110 }
1111 
1112 static int mpls_fill_stats_af(struct sk_buff *skb,
1113 			      const struct net_device *dev)
1114 {
1115 	struct mpls_link_stats *stats;
1116 	struct mpls_dev *mdev;
1117 	struct nlattr *nla;
1118 
1119 	mdev = mpls_dev_rcu(dev);
1120 	if (!mdev)
1121 		return -ENODATA;
1122 
1123 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1124 				sizeof(struct mpls_link_stats),
1125 				MPLS_STATS_UNSPEC);
1126 	if (!nla)
1127 		return -EMSGSIZE;
1128 
1129 	stats = nla_data(nla);
1130 	mpls_get_stats(mdev, stats);
1131 
1132 	return 0;
1133 }
1134 
1135 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1136 {
1137 	struct mpls_dev *mdev;
1138 
1139 	mdev = mpls_dev_rcu(dev);
1140 	if (!mdev)
1141 		return 0;
1142 
1143 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1144 }
1145 
1146 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1147 				     u32 portid, u32 seq, int event,
1148 				     unsigned int flags, int type)
1149 {
1150 	struct nlmsghdr  *nlh;
1151 	struct netconfmsg *ncm;
1152 	bool all = false;
1153 
1154 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1155 			flags);
1156 	if (!nlh)
1157 		return -EMSGSIZE;
1158 
1159 	if (type == NETCONFA_ALL)
1160 		all = true;
1161 
1162 	ncm = nlmsg_data(nlh);
1163 	ncm->ncm_family = AF_MPLS;
1164 
1165 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1166 		goto nla_put_failure;
1167 
1168 	if ((all || type == NETCONFA_INPUT) &&
1169 	    nla_put_s32(skb, NETCONFA_INPUT,
1170 			READ_ONCE(mdev->input_enabled)) < 0)
1171 		goto nla_put_failure;
1172 
1173 	nlmsg_end(skb, nlh);
1174 	return 0;
1175 
1176 nla_put_failure:
1177 	nlmsg_cancel(skb, nlh);
1178 	return -EMSGSIZE;
1179 }
1180 
1181 static int mpls_netconf_msgsize_devconf(int type)
1182 {
1183 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1184 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1185 	bool all = false;
1186 
1187 	if (type == NETCONFA_ALL)
1188 		all = true;
1189 
1190 	if (all || type == NETCONFA_INPUT)
1191 		size += nla_total_size(4);
1192 
1193 	return size;
1194 }
1195 
1196 static void mpls_netconf_notify_devconf(struct net *net, int event,
1197 					int type, struct mpls_dev *mdev)
1198 {
1199 	struct sk_buff *skb;
1200 	int err = -ENOBUFS;
1201 
1202 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1203 	if (!skb)
1204 		goto errout;
1205 
1206 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1207 	if (err < 0) {
1208 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1209 		WARN_ON(err == -EMSGSIZE);
1210 		kfree_skb(skb);
1211 		goto errout;
1212 	}
1213 
1214 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1215 	return;
1216 errout:
1217 	rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1218 }
1219 
1220 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1221 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1222 };
1223 
1224 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1225 				      const struct nlmsghdr *nlh,
1226 				      struct nlattr **tb,
1227 				      struct netlink_ext_ack *extack)
1228 {
1229 	int i, err;
1230 
1231 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1232 		NL_SET_ERR_MSG_MOD(extack,
1233 				   "Invalid header for netconf get request");
1234 		return -EINVAL;
1235 	}
1236 
1237 	if (!netlink_strict_get_check(skb))
1238 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1239 					      tb, NETCONFA_MAX,
1240 					      devconf_mpls_policy, extack);
1241 
1242 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1243 					    tb, NETCONFA_MAX,
1244 					    devconf_mpls_policy, extack);
1245 	if (err)
1246 		return err;
1247 
1248 	for (i = 0; i <= NETCONFA_MAX; i++) {
1249 		if (!tb[i])
1250 			continue;
1251 
1252 		switch (i) {
1253 		case NETCONFA_IFINDEX:
1254 			break;
1255 		default:
1256 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1257 			return -EINVAL;
1258 		}
1259 	}
1260 
1261 	return 0;
1262 }
1263 
1264 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1265 				    struct nlmsghdr *nlh,
1266 				    struct netlink_ext_ack *extack)
1267 {
1268 	struct net *net = sock_net(in_skb->sk);
1269 	struct nlattr *tb[NETCONFA_MAX + 1];
1270 	struct net_device *dev;
1271 	struct mpls_dev *mdev;
1272 	struct sk_buff *skb;
1273 	int ifindex;
1274 	int err;
1275 
1276 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1277 	if (err < 0)
1278 		goto errout;
1279 
1280 	if (!tb[NETCONFA_IFINDEX]) {
1281 		err = -EINVAL;
1282 		goto errout;
1283 	}
1284 
1285 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1286 
1287 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1288 	if (!skb) {
1289 		err = -ENOBUFS;
1290 		goto errout;
1291 	}
1292 
1293 	rcu_read_lock();
1294 
1295 	dev = dev_get_by_index_rcu(net, ifindex);
1296 	if (!dev) {
1297 		err = -EINVAL;
1298 		goto errout_unlock;
1299 	}
1300 
1301 	mdev = mpls_dev_rcu(dev);
1302 	if (!mdev) {
1303 		err = -EINVAL;
1304 		goto errout_unlock;
1305 	}
1306 
1307 	err = mpls_netconf_fill_devconf(skb, mdev,
1308 					NETLINK_CB(in_skb).portid,
1309 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1310 					NETCONFA_ALL);
1311 	if (err < 0) {
1312 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1313 		WARN_ON(err == -EMSGSIZE);
1314 		goto errout_unlock;
1315 	}
1316 
1317 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1318 
1319 	rcu_read_unlock();
1320 errout:
1321 	return err;
1322 
1323 errout_unlock:
1324 	rcu_read_unlock();
1325 	kfree_skb(skb);
1326 	goto errout;
1327 }
1328 
1329 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1330 				     struct netlink_callback *cb)
1331 {
1332 	const struct nlmsghdr *nlh = cb->nlh;
1333 	struct net *net = sock_net(skb->sk);
1334 	struct {
1335 		unsigned long ifindex;
1336 	} *ctx = (void *)cb->ctx;
1337 	struct net_device *dev;
1338 	struct mpls_dev *mdev;
1339 	int err = 0;
1340 
1341 	if (cb->strict_check) {
1342 		struct netlink_ext_ack *extack = cb->extack;
1343 		struct netconfmsg *ncm;
1344 
1345 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1346 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1347 			return -EINVAL;
1348 		}
1349 
1350 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1351 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1352 			return -EINVAL;
1353 		}
1354 	}
1355 
1356 	rcu_read_lock();
1357 	for_each_netdev_dump(net, dev, ctx->ifindex) {
1358 		mdev = mpls_dev_rcu(dev);
1359 		if (!mdev)
1360 			continue;
1361 		err = mpls_netconf_fill_devconf(skb, mdev,
1362 						NETLINK_CB(cb->skb).portid,
1363 						nlh->nlmsg_seq,
1364 						RTM_NEWNETCONF,
1365 						NLM_F_MULTI,
1366 						NETCONFA_ALL);
1367 		if (err < 0)
1368 			break;
1369 	}
1370 	rcu_read_unlock();
1371 
1372 	return err;
1373 }
1374 
1375 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1376 	(&((struct mpls_dev *)0)->field)
1377 
1378 static int mpls_conf_proc(const struct ctl_table *ctl, int write,
1379 			  void *buffer, size_t *lenp, loff_t *ppos)
1380 {
1381 	int oval = *(int *)ctl->data;
1382 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1383 
1384 	if (write) {
1385 		struct mpls_dev *mdev = ctl->extra1;
1386 		int i = (int *)ctl->data - (int *)mdev;
1387 		struct net *net = ctl->extra2;
1388 		int val = *(int *)ctl->data;
1389 
1390 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1391 		    val != oval) {
1392 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1393 						    NETCONFA_INPUT, mdev);
1394 		}
1395 	}
1396 
1397 	return ret;
1398 }
1399 
1400 static const struct ctl_table mpls_dev_table[] = {
1401 	{
1402 		.procname	= "input",
1403 		.maxlen		= sizeof(int),
1404 		.mode		= 0644,
1405 		.proc_handler	= mpls_conf_proc,
1406 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1407 	},
1408 };
1409 
1410 static int mpls_dev_sysctl_register(struct net_device *dev,
1411 				    struct mpls_dev *mdev)
1412 {
1413 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1414 	size_t table_size = ARRAY_SIZE(mpls_dev_table);
1415 	struct net *net = dev_net(dev);
1416 	struct ctl_table *table;
1417 	int i;
1418 
1419 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1420 	if (!table)
1421 		goto out;
1422 
1423 	/* Table data contains only offsets relative to the base of
1424 	 * the mdev at this point, so make them absolute.
1425 	 */
1426 	for (i = 0; i < table_size; i++) {
1427 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1428 		table[i].extra1 = mdev;
1429 		table[i].extra2 = net;
1430 	}
1431 
1432 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1433 
1434 	mdev->sysctl = register_net_sysctl_sz(net, path, table, table_size);
1435 	if (!mdev->sysctl)
1436 		goto free;
1437 
1438 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1439 	return 0;
1440 
1441 free:
1442 	kfree(table);
1443 out:
1444 	mdev->sysctl = NULL;
1445 	return -ENOBUFS;
1446 }
1447 
1448 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1449 				       struct mpls_dev *mdev)
1450 {
1451 	struct net *net = dev_net(dev);
1452 	const struct ctl_table *table;
1453 
1454 	if (!mdev->sysctl)
1455 		return;
1456 
1457 	table = mdev->sysctl->ctl_table_arg;
1458 	unregister_net_sysctl_table(mdev->sysctl);
1459 	kfree(table);
1460 
1461 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1462 }
1463 
1464 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1465 {
1466 	struct mpls_dev *mdev;
1467 	int err = -ENOMEM;
1468 	int i;
1469 
1470 	mdev = kzalloc_obj(*mdev);
1471 	if (!mdev)
1472 		return ERR_PTR(err);
1473 
1474 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1475 	if (!mdev->stats)
1476 		goto free;
1477 
1478 	for_each_possible_cpu(i) {
1479 		struct mpls_pcpu_stats *mpls_stats;
1480 
1481 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1482 		u64_stats_init(&mpls_stats->syncp);
1483 	}
1484 
1485 	mdev->dev = dev;
1486 
1487 	err = mpls_dev_sysctl_register(dev, mdev);
1488 	if (err)
1489 		goto free;
1490 
1491 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1492 
1493 	return mdev;
1494 
1495 free:
1496 	free_percpu(mdev->stats);
1497 	kfree(mdev);
1498 	return ERR_PTR(err);
1499 }
1500 
1501 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1502 {
1503 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1504 
1505 	free_percpu(mdev->stats);
1506 	kfree(mdev);
1507 }
1508 
1509 static int mpls_ifdown(struct net_device *dev, int event)
1510 {
1511 	struct net *net = dev_net(dev);
1512 	unsigned int index;
1513 
1514 	for (index = 0; index < net->mpls.platform_labels; index++) {
1515 		struct mpls_route *rt;
1516 		bool nh_del = false;
1517 		u8 alive = 0;
1518 
1519 		rt = mpls_route_input(net, index);
1520 		if (!rt)
1521 			continue;
1522 
1523 		if (event == NETDEV_UNREGISTER) {
1524 			u8 deleted = 0;
1525 
1526 			for_nexthops(rt) {
1527 				if (!nh->nh_dev || nh->nh_dev == dev)
1528 					deleted++;
1529 				if (nh->nh_dev == dev)
1530 					nh_del = true;
1531 			} endfor_nexthops(rt);
1532 
1533 			/* if there are no more nexthops, delete the route */
1534 			if (deleted == rt->rt_nhn) {
1535 				mpls_route_update(net, index, NULL, NULL);
1536 				continue;
1537 			}
1538 
1539 			if (nh_del) {
1540 				size_t size = sizeof(*rt) + rt->rt_nhn *
1541 					rt->rt_nh_size;
1542 				struct mpls_route *orig = rt;
1543 
1544 				rt = kmemdup(orig, size, GFP_KERNEL);
1545 				if (!rt)
1546 					return -ENOMEM;
1547 			}
1548 		}
1549 
1550 		change_nexthops(rt) {
1551 			unsigned int nh_flags = nh->nh_flags;
1552 
1553 			if (nh->nh_dev != dev) {
1554 				if (nh_del)
1555 					netdev_hold(nh->nh_dev, &nh->nh_dev_tracker,
1556 						    GFP_KERNEL);
1557 				goto next;
1558 			}
1559 
1560 			switch (event) {
1561 			case NETDEV_DOWN:
1562 			case NETDEV_UNREGISTER:
1563 				nh_flags |= RTNH_F_DEAD;
1564 				fallthrough;
1565 			case NETDEV_CHANGE:
1566 				nh_flags |= RTNH_F_LINKDOWN;
1567 				break;
1568 			}
1569 			if (event == NETDEV_UNREGISTER)
1570 				nh->nh_dev = NULL;
1571 
1572 			if (nh->nh_flags != nh_flags)
1573 				WRITE_ONCE(nh->nh_flags, nh_flags);
1574 next:
1575 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1576 				alive++;
1577 		} endfor_nexthops(rt);
1578 
1579 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1580 
1581 		if (nh_del)
1582 			mpls_route_update(net, index, rt, NULL);
1583 	}
1584 
1585 	return 0;
1586 }
1587 
1588 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1589 {
1590 	struct net *net = dev_net(dev);
1591 	unsigned int index;
1592 	u8 alive;
1593 
1594 	for (index = 0; index < net->mpls.platform_labels; index++) {
1595 		struct mpls_route *rt;
1596 
1597 		rt = mpls_route_input(net, index);
1598 		if (!rt)
1599 			continue;
1600 
1601 		alive = 0;
1602 		change_nexthops(rt) {
1603 			unsigned int nh_flags = nh->nh_flags;
1604 
1605 			if (!(nh_flags & flags)) {
1606 				alive++;
1607 				continue;
1608 			}
1609 			if (nh->nh_dev != dev)
1610 				continue;
1611 			alive++;
1612 			nh_flags &= ~flags;
1613 			WRITE_ONCE(nh->nh_flags, nh_flags);
1614 		} endfor_nexthops(rt);
1615 
1616 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1617 	}
1618 }
1619 
1620 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1621 			   void *ptr)
1622 {
1623 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1624 	struct net *net = dev_net(dev);
1625 	struct mpls_dev *mdev;
1626 	unsigned int flags;
1627 	int err;
1628 
1629 	mutex_lock(&net->mpls.platform_mutex);
1630 
1631 	if (event == NETDEV_REGISTER) {
1632 		mdev = mpls_add_dev(dev);
1633 		if (IS_ERR(mdev)) {
1634 			err = PTR_ERR(mdev);
1635 			goto err;
1636 		}
1637 
1638 		goto out;
1639 	}
1640 
1641 	mdev = mpls_dev_get(net, dev);
1642 	if (!mdev)
1643 		goto out;
1644 
1645 	switch (event) {
1646 
1647 	case NETDEV_DOWN:
1648 		err = mpls_ifdown(dev, event);
1649 		if (err)
1650 			goto err;
1651 		break;
1652 	case NETDEV_UP:
1653 		flags = netif_get_flags(dev);
1654 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1655 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1656 		else
1657 			mpls_ifup(dev, RTNH_F_DEAD);
1658 		break;
1659 	case NETDEV_CHANGE:
1660 		flags = netif_get_flags(dev);
1661 		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1662 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1663 		} else {
1664 			err = mpls_ifdown(dev, event);
1665 			if (err)
1666 				goto err;
1667 		}
1668 		break;
1669 	case NETDEV_UNREGISTER:
1670 		err = mpls_ifdown(dev, event);
1671 		if (err)
1672 			goto err;
1673 
1674 		mdev = mpls_dev_get(net, dev);
1675 		if (mdev) {
1676 			mpls_dev_sysctl_unregister(dev, mdev);
1677 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1678 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1679 		}
1680 		break;
1681 	case NETDEV_CHANGENAME:
1682 		mdev = mpls_dev_get(net, dev);
1683 		if (mdev) {
1684 			mpls_dev_sysctl_unregister(dev, mdev);
1685 			err = mpls_dev_sysctl_register(dev, mdev);
1686 			if (err)
1687 				goto err;
1688 		}
1689 		break;
1690 	}
1691 
1692 out:
1693 	mutex_unlock(&net->mpls.platform_mutex);
1694 	return NOTIFY_OK;
1695 
1696 err:
1697 	mutex_unlock(&net->mpls.platform_mutex);
1698 	return notifier_from_errno(err);
1699 }
1700 
1701 static struct notifier_block mpls_dev_notifier = {
1702 	.notifier_call = mpls_dev_notify,
1703 };
1704 
1705 static int nla_put_via(struct sk_buff *skb,
1706 		       u8 table, const void *addr, int alen)
1707 {
1708 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1709 		AF_INET, AF_INET6, AF_PACKET,
1710 	};
1711 	struct nlattr *nla;
1712 	struct rtvia *via;
1713 	int family = AF_UNSPEC;
1714 
1715 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1716 	if (!nla)
1717 		return -EMSGSIZE;
1718 
1719 	if (table <= NEIGH_NR_TABLES)
1720 		family = table_to_family[table];
1721 
1722 	via = nla_data(nla);
1723 	via->rtvia_family = family;
1724 	memcpy(via->rtvia_addr, addr, alen);
1725 	return 0;
1726 }
1727 
1728 int nla_put_labels(struct sk_buff *skb, int attrtype,
1729 		   u8 labels, const u32 label[])
1730 {
1731 	struct nlattr *nla;
1732 	struct mpls_shim_hdr *nla_label;
1733 	bool bos;
1734 	int i;
1735 	nla = nla_reserve(skb, attrtype, labels*4);
1736 	if (!nla)
1737 		return -EMSGSIZE;
1738 
1739 	nla_label = nla_data(nla);
1740 	bos = true;
1741 	for (i = labels - 1; i >= 0; i--) {
1742 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1743 		bos = false;
1744 	}
1745 
1746 	return 0;
1747 }
1748 EXPORT_SYMBOL_GPL(nla_put_labels);
1749 
1750 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1751 		   u32 label[], struct netlink_ext_ack *extack)
1752 {
1753 	unsigned len = nla_len(nla);
1754 	struct mpls_shim_hdr *nla_label;
1755 	u8 nla_labels;
1756 	bool bos;
1757 	int i;
1758 
1759 	/* len needs to be an even multiple of 4 (the label size). Number
1760 	 * of labels is a u8 so check for overflow.
1761 	 */
1762 	if (len & 3 || len / 4 > 255) {
1763 		NL_SET_ERR_MSG_ATTR(extack, nla,
1764 				    "Invalid length for labels attribute");
1765 		return -EINVAL;
1766 	}
1767 
1768 	/* Limit the number of new labels allowed */
1769 	nla_labels = len/4;
1770 	if (nla_labels > max_labels) {
1771 		NL_SET_ERR_MSG(extack, "Too many labels");
1772 		return -EINVAL;
1773 	}
1774 
1775 	/* when label == NULL, caller wants number of labels */
1776 	if (!label)
1777 		goto out;
1778 
1779 	nla_label = nla_data(nla);
1780 	bos = true;
1781 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1782 		struct mpls_entry_decoded dec;
1783 		dec = mpls_entry_decode(nla_label + i);
1784 
1785 		/* Ensure the bottom of stack flag is properly set
1786 		 * and ttl and tc are both clear.
1787 		 */
1788 		if (dec.ttl) {
1789 			NL_SET_ERR_MSG_ATTR(extack, nla,
1790 					    "TTL in label must be 0");
1791 			return -EINVAL;
1792 		}
1793 
1794 		if (dec.tc) {
1795 			NL_SET_ERR_MSG_ATTR(extack, nla,
1796 					    "Traffic class in label must be 0");
1797 			return -EINVAL;
1798 		}
1799 
1800 		if (dec.bos != bos) {
1801 			NL_SET_BAD_ATTR(extack, nla);
1802 			if (bos) {
1803 				NL_SET_ERR_MSG(extack,
1804 					       "BOS bit must be set in first label");
1805 			} else {
1806 				NL_SET_ERR_MSG(extack,
1807 					       "BOS bit can only be set in first label");
1808 			}
1809 			return -EINVAL;
1810 		}
1811 
1812 		switch (dec.label) {
1813 		case MPLS_LABEL_IMPLNULL:
1814 			/* RFC3032: This is a label that an LSR may
1815 			 * assign and distribute, but which never
1816 			 * actually appears in the encapsulation.
1817 			 */
1818 			NL_SET_ERR_MSG_ATTR(extack, nla,
1819 					    "Implicit NULL Label (3) can not be used in encapsulation");
1820 			return -EINVAL;
1821 		}
1822 
1823 		label[i] = dec.label;
1824 	}
1825 out:
1826 	*labels = nla_labels;
1827 	return 0;
1828 }
1829 EXPORT_SYMBOL_GPL(nla_get_labels);
1830 
1831 static int rtm_to_route_config(struct sk_buff *skb,
1832 			       struct nlmsghdr *nlh,
1833 			       struct mpls_route_config *cfg,
1834 			       struct netlink_ext_ack *extack)
1835 {
1836 	struct rtmsg *rtm;
1837 	struct nlattr *tb[RTA_MAX+1];
1838 	int index;
1839 	int err;
1840 
1841 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1842 				     rtm_mpls_policy, extack);
1843 	if (err < 0)
1844 		goto errout;
1845 
1846 	err = -EINVAL;
1847 	rtm = nlmsg_data(nlh);
1848 
1849 	if (rtm->rtm_family != AF_MPLS) {
1850 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1851 		goto errout;
1852 	}
1853 	if (rtm->rtm_dst_len != 20) {
1854 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1855 		goto errout;
1856 	}
1857 	if (rtm->rtm_src_len != 0) {
1858 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1859 		goto errout;
1860 	}
1861 	if (rtm->rtm_tos != 0) {
1862 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1863 		goto errout;
1864 	}
1865 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1866 		NL_SET_ERR_MSG(extack,
1867 			       "MPLS only supports the main route table");
1868 		goto errout;
1869 	}
1870 	/* Any value is acceptable for rtm_protocol */
1871 
1872 	/* As mpls uses destination specific addresses
1873 	 * (or source specific address in the case of multicast)
1874 	 * all addresses have universal scope.
1875 	 */
1876 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1877 		NL_SET_ERR_MSG(extack,
1878 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1879 		goto errout;
1880 	}
1881 	if (rtm->rtm_type != RTN_UNICAST) {
1882 		NL_SET_ERR_MSG(extack,
1883 			       "Invalid route type - MPLS only supports UNICAST");
1884 		goto errout;
1885 	}
1886 	if (rtm->rtm_flags != 0) {
1887 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1888 		goto errout;
1889 	}
1890 
1891 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1892 	cfg->rc_protocol	= rtm->rtm_protocol;
1893 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1894 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1895 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1896 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1897 	cfg->rc_nlinfo.nlh	= nlh;
1898 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1899 
1900 	for (index = 0; index <= RTA_MAX; index++) {
1901 		struct nlattr *nla = tb[index];
1902 		if (!nla)
1903 			continue;
1904 
1905 		switch (index) {
1906 		case RTA_OIF:
1907 			cfg->rc_ifindex = nla_get_u32(nla);
1908 			break;
1909 		case RTA_NEWDST:
1910 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1911 					   &cfg->rc_output_labels,
1912 					   cfg->rc_output_label, extack))
1913 				goto errout;
1914 			break;
1915 		case RTA_DST:
1916 		{
1917 			u8 label_count;
1918 			if (nla_get_labels(nla, 1, &label_count,
1919 					   &cfg->rc_label, extack))
1920 				goto errout;
1921 
1922 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1923 					   &cfg->rc_label, extack))
1924 				goto errout;
1925 			break;
1926 		}
1927 		case RTA_GATEWAY:
1928 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1929 			goto errout;
1930 		case RTA_VIA:
1931 		{
1932 			if (nla_get_via(nla, &cfg->rc_via_alen,
1933 					&cfg->rc_via_table, cfg->rc_via,
1934 					extack))
1935 				goto errout;
1936 			break;
1937 		}
1938 		case RTA_MULTIPATH:
1939 		{
1940 			cfg->rc_mp = nla_data(nla);
1941 			cfg->rc_mp_len = nla_len(nla);
1942 			break;
1943 		}
1944 		case RTA_TTL_PROPAGATE:
1945 		{
1946 			u8 ttl_propagate = nla_get_u8(nla);
1947 
1948 			if (ttl_propagate > 1) {
1949 				NL_SET_ERR_MSG_ATTR(extack, nla,
1950 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1951 				goto errout;
1952 			}
1953 			cfg->rc_ttl_propagate = ttl_propagate ?
1954 				MPLS_TTL_PROP_ENABLED :
1955 				MPLS_TTL_PROP_DISABLED;
1956 			break;
1957 		}
1958 		default:
1959 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1960 			/* Unsupported attribute */
1961 			goto errout;
1962 		}
1963 	}
1964 
1965 	err = 0;
1966 errout:
1967 	return err;
1968 }
1969 
1970 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1971 			     struct netlink_ext_ack *extack)
1972 {
1973 	struct net *net = sock_net(skb->sk);
1974 	struct mpls_route_config *cfg;
1975 	int err;
1976 
1977 	cfg = kzalloc_obj(*cfg);
1978 	if (!cfg)
1979 		return -ENOMEM;
1980 
1981 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1982 	if (err < 0)
1983 		goto out;
1984 
1985 	mutex_lock(&net->mpls.platform_mutex);
1986 	err = mpls_route_del(cfg, extack);
1987 	mutex_unlock(&net->mpls.platform_mutex);
1988 out:
1989 	kfree(cfg);
1990 
1991 	return err;
1992 }
1993 
1994 
1995 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1996 			     struct netlink_ext_ack *extack)
1997 {
1998 	struct net *net = sock_net(skb->sk);
1999 	struct mpls_route_config *cfg;
2000 	int err;
2001 
2002 	cfg = kzalloc_obj(*cfg);
2003 	if (!cfg)
2004 		return -ENOMEM;
2005 
2006 	err = rtm_to_route_config(skb, nlh, cfg, extack);
2007 	if (err < 0)
2008 		goto out;
2009 
2010 	mutex_lock(&net->mpls.platform_mutex);
2011 	err = mpls_route_add(cfg, extack);
2012 	mutex_unlock(&net->mpls.platform_mutex);
2013 out:
2014 	kfree(cfg);
2015 
2016 	return err;
2017 }
2018 
2019 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
2020 			   u32 label, struct mpls_route *rt, int flags)
2021 {
2022 	struct net_device *dev;
2023 	struct nlmsghdr *nlh;
2024 	struct rtmsg *rtm;
2025 
2026 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
2027 	if (nlh == NULL)
2028 		return -EMSGSIZE;
2029 
2030 	rtm = nlmsg_data(nlh);
2031 	rtm->rtm_family = AF_MPLS;
2032 	rtm->rtm_dst_len = 20;
2033 	rtm->rtm_src_len = 0;
2034 	rtm->rtm_tos = 0;
2035 	rtm->rtm_table = RT_TABLE_MAIN;
2036 	rtm->rtm_protocol = rt->rt_protocol;
2037 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
2038 	rtm->rtm_type = RTN_UNICAST;
2039 	rtm->rtm_flags = 0;
2040 
2041 	if (nla_put_labels(skb, RTA_DST, 1, &label))
2042 		goto nla_put_failure;
2043 
2044 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
2045 		bool ttl_propagate =
2046 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2047 
2048 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2049 			       ttl_propagate))
2050 			goto nla_put_failure;
2051 	}
2052 	if (rt->rt_nhn == 1) {
2053 		const struct mpls_nh *nh = rt->rt_nh;
2054 
2055 		if (nh->nh_labels &&
2056 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2057 				   nh->nh_label))
2058 			goto nla_put_failure;
2059 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2060 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2061 				nh->nh_via_alen))
2062 			goto nla_put_failure;
2063 		dev = nh->nh_dev;
2064 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2065 			goto nla_put_failure;
2066 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2067 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2068 		if (nh->nh_flags & RTNH_F_DEAD)
2069 			rtm->rtm_flags |= RTNH_F_DEAD;
2070 	} else {
2071 		struct rtnexthop *rtnh;
2072 		struct nlattr *mp;
2073 		u8 linkdown = 0;
2074 		u8 dead = 0;
2075 
2076 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2077 		if (!mp)
2078 			goto nla_put_failure;
2079 
2080 		for_nexthops(rt) {
2081 			dev = nh->nh_dev;
2082 			if (!dev)
2083 				continue;
2084 
2085 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2086 			if (!rtnh)
2087 				goto nla_put_failure;
2088 
2089 			rtnh->rtnh_ifindex = dev->ifindex;
2090 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2091 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2092 				linkdown++;
2093 			}
2094 			if (nh->nh_flags & RTNH_F_DEAD) {
2095 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2096 				dead++;
2097 			}
2098 
2099 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2100 							    nh->nh_labels,
2101 							    nh->nh_label))
2102 				goto nla_put_failure;
2103 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2104 			    nla_put_via(skb, nh->nh_via_table,
2105 					mpls_nh_via(rt, nh),
2106 					nh->nh_via_alen))
2107 				goto nla_put_failure;
2108 
2109 			/* length of rtnetlink header + attributes */
2110 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2111 		} endfor_nexthops(rt);
2112 
2113 		if (linkdown == rt->rt_nhn)
2114 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2115 		if (dead == rt->rt_nhn)
2116 			rtm->rtm_flags |= RTNH_F_DEAD;
2117 
2118 		nla_nest_end(skb, mp);
2119 	}
2120 
2121 	nlmsg_end(skb, nlh);
2122 	return 0;
2123 
2124 nla_put_failure:
2125 	nlmsg_cancel(skb, nlh);
2126 	return -EMSGSIZE;
2127 }
2128 
2129 #if IS_ENABLED(CONFIG_INET)
2130 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2131 				   struct fib_dump_filter *filter,
2132 				   struct netlink_callback *cb)
2133 {
2134 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2135 }
2136 #else
2137 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2138 				   struct fib_dump_filter *filter,
2139 				   struct netlink_callback *cb)
2140 {
2141 	struct netlink_ext_ack *extack = cb->extack;
2142 	struct nlattr *tb[RTA_MAX + 1];
2143 	struct rtmsg *rtm;
2144 	int err, i;
2145 
2146 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
2147 	if (!rtm) {
2148 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2149 		return -EINVAL;
2150 	}
2151 
2152 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2153 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2154 	    rtm->rtm_flags) {
2155 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2156 		return -EINVAL;
2157 	}
2158 
2159 	if (rtm->rtm_protocol) {
2160 		filter->protocol = rtm->rtm_protocol;
2161 		filter->filter_set = 1;
2162 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2163 	}
2164 
2165 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2166 					    rtm_mpls_policy, extack);
2167 	if (err < 0)
2168 		return err;
2169 
2170 	for (i = 0; i <= RTA_MAX; ++i) {
2171 		int ifindex;
2172 
2173 		if (i == RTA_OIF) {
2174 			ifindex = nla_get_u32(tb[i]);
2175 			filter->dev = dev_get_by_index_rcu(net, ifindex);
2176 			if (!filter->dev)
2177 				return -ENODEV;
2178 			filter->filter_set = 1;
2179 		} else if (tb[i]) {
2180 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2181 			return -EINVAL;
2182 		}
2183 	}
2184 
2185 	return 0;
2186 }
2187 #endif
2188 
2189 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2190 			     const struct net_device *dev)
2191 {
2192 	if (rt->rt_nhn == 1) {
2193 		struct mpls_nh *nh = rt->rt_nh;
2194 
2195 		if (nh->nh_dev == dev)
2196 			return true;
2197 	} else {
2198 		for_nexthops(rt) {
2199 			if (nh->nh_dev == dev)
2200 				return true;
2201 		} endfor_nexthops(rt);
2202 	}
2203 
2204 	return false;
2205 }
2206 
2207 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2208 {
2209 	const struct nlmsghdr *nlh = cb->nlh;
2210 	struct net *net = sock_net(skb->sk);
2211 	struct mpls_route __rcu **platform_label;
2212 	struct fib_dump_filter filter = {
2213 		.rtnl_held = false,
2214 	};
2215 	unsigned int flags = NLM_F_MULTI;
2216 	size_t platform_labels;
2217 	unsigned int index;
2218 	int err;
2219 
2220 	rcu_read_lock();
2221 
2222 	if (cb->strict_check) {
2223 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2224 		if (err < 0)
2225 			goto err;
2226 
2227 		/* for MPLS, there is only 1 table with fixed type and flags.
2228 		 * If either are set in the filter then return nothing.
2229 		 */
2230 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2231 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2232 		     filter.flags)
2233 			goto unlock;
2234 	}
2235 
2236 	index = cb->args[0];
2237 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2238 		index = MPLS_LABEL_FIRST_UNRESERVED;
2239 
2240 	platform_label = rcu_dereference(net->mpls.platform_label);
2241 	platform_labels = net->mpls.platform_labels;
2242 
2243 	if (filter.filter_set)
2244 		flags |= NLM_F_DUMP_FILTERED;
2245 
2246 	for (; index < platform_labels; index++) {
2247 		struct mpls_route *rt;
2248 
2249 		rt = rcu_dereference(platform_label[index]);
2250 		if (!rt)
2251 			continue;
2252 
2253 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2254 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2255 			continue;
2256 
2257 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2258 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2259 				    index, rt, flags) < 0)
2260 			break;
2261 	}
2262 	cb->args[0] = index;
2263 
2264 unlock:
2265 	rcu_read_unlock();
2266 	return skb->len;
2267 
2268 err:
2269 	rcu_read_unlock();
2270 	return err;
2271 }
2272 
2273 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2274 {
2275 	size_t payload =
2276 		NLMSG_ALIGN(sizeof(struct rtmsg))
2277 		+ nla_total_size(4)			/* RTA_DST */
2278 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2279 
2280 	if (rt->rt_nhn == 1) {
2281 		struct mpls_nh *nh = rt->rt_nh;
2282 
2283 		if (nh->nh_dev)
2284 			payload += nla_total_size(4); /* RTA_OIF */
2285 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2286 			payload += nla_total_size(2 + nh->nh_via_alen);
2287 		if (nh->nh_labels) /* RTA_NEWDST */
2288 			payload += nla_total_size(nh->nh_labels * 4);
2289 	} else {
2290 		/* each nexthop is packed in an attribute */
2291 		size_t nhsize = 0;
2292 
2293 		for_nexthops(rt) {
2294 			if (!nh->nh_dev)
2295 				continue;
2296 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2297 			/* RTA_VIA */
2298 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2299 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2300 			if (nh->nh_labels)
2301 				nhsize += nla_total_size(nh->nh_labels * 4);
2302 		} endfor_nexthops(rt);
2303 		/* nested attribute */
2304 		payload += nla_total_size(nhsize);
2305 	}
2306 
2307 	return payload;
2308 }
2309 
2310 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2311 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2312 		       unsigned int nlm_flags)
2313 {
2314 	struct sk_buff *skb;
2315 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2316 	int err = -ENOBUFS;
2317 
2318 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2319 	if (skb == NULL)
2320 		goto errout;
2321 
2322 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2323 	if (err < 0) {
2324 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2325 		WARN_ON(err == -EMSGSIZE);
2326 		kfree_skb(skb);
2327 		goto errout;
2328 	}
2329 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2330 
2331 	return;
2332 errout:
2333 	rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2334 }
2335 
2336 static int mpls_valid_getroute_req(struct sk_buff *skb,
2337 				   const struct nlmsghdr *nlh,
2338 				   struct nlattr **tb,
2339 				   struct netlink_ext_ack *extack)
2340 {
2341 	struct rtmsg *rtm;
2342 	int i, err;
2343 
2344 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
2345 	if (!rtm) {
2346 		NL_SET_ERR_MSG_MOD(extack,
2347 				   "Invalid header for get route request");
2348 		return -EINVAL;
2349 	}
2350 
2351 	if (!netlink_strict_get_check(skb))
2352 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2353 					      rtm_mpls_policy, extack);
2354 
2355 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2356 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2357 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2358 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2359 		return -EINVAL;
2360 	}
2361 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2362 		NL_SET_ERR_MSG_MOD(extack,
2363 				   "Invalid flags for get route request");
2364 		return -EINVAL;
2365 	}
2366 
2367 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2368 					    rtm_mpls_policy, extack);
2369 	if (err)
2370 		return err;
2371 
2372 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2373 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2374 		return -EINVAL;
2375 	}
2376 
2377 	for (i = 0; i <= RTA_MAX; i++) {
2378 		if (!tb[i])
2379 			continue;
2380 
2381 		switch (i) {
2382 		case RTA_DST:
2383 		case RTA_NEWDST:
2384 			break;
2385 		default:
2386 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2387 			return -EINVAL;
2388 		}
2389 	}
2390 
2391 	return 0;
2392 }
2393 
2394 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2395 			 struct netlink_ext_ack *extack)
2396 {
2397 	struct net *net = sock_net(in_skb->sk);
2398 	u32 portid = NETLINK_CB(in_skb).portid;
2399 	u32 in_label = LABEL_NOT_SPECIFIED;
2400 	struct nlattr *tb[RTA_MAX + 1];
2401 	struct mpls_route *rt = NULL;
2402 	u32 labels[MAX_NEW_LABELS];
2403 	struct mpls_shim_hdr *hdr;
2404 	unsigned int hdr_size = 0;
2405 	const struct mpls_nh *nh;
2406 	struct net_device *dev;
2407 	struct rtmsg *rtm, *r;
2408 	struct nlmsghdr *nlh;
2409 	struct sk_buff *skb;
2410 	u8 n_labels;
2411 	int err;
2412 
2413 	mutex_lock(&net->mpls.platform_mutex);
2414 
2415 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2416 	if (err < 0)
2417 		goto errout;
2418 
2419 	rtm = nlmsg_data(in_nlh);
2420 
2421 	if (tb[RTA_DST]) {
2422 		u8 label_count;
2423 
2424 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2425 				   &in_label, extack)) {
2426 			err = -EINVAL;
2427 			goto errout;
2428 		}
2429 
2430 		if (!mpls_label_ok(net, &in_label, extack)) {
2431 			err = -EINVAL;
2432 			goto errout;
2433 		}
2434 	}
2435 
2436 	if (in_label < net->mpls.platform_labels)
2437 		rt = mpls_route_input(net, in_label);
2438 	if (!rt) {
2439 		err = -ENETUNREACH;
2440 		goto errout;
2441 	}
2442 
2443 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2444 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2445 		if (!skb) {
2446 			err = -ENOBUFS;
2447 			goto errout;
2448 		}
2449 
2450 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2451 				      RTM_NEWROUTE, in_label, rt, 0);
2452 		if (err < 0) {
2453 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2454 			WARN_ON(err == -EMSGSIZE);
2455 			goto errout_free;
2456 		}
2457 
2458 		err = rtnl_unicast(skb, net, portid);
2459 		goto errout;
2460 	}
2461 
2462 	if (tb[RTA_NEWDST]) {
2463 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2464 				   labels, extack) != 0) {
2465 			err = -EINVAL;
2466 			goto errout;
2467 		}
2468 
2469 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2470 	}
2471 
2472 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2473 	if (!skb) {
2474 		err = -ENOBUFS;
2475 		goto errout;
2476 	}
2477 
2478 	skb->protocol = htons(ETH_P_MPLS_UC);
2479 
2480 	if (hdr_size) {
2481 		bool bos;
2482 		int i;
2483 
2484 		if (skb_cow(skb, hdr_size)) {
2485 			err = -ENOBUFS;
2486 			goto errout_free;
2487 		}
2488 
2489 		skb_reserve(skb, hdr_size);
2490 		skb_push(skb, hdr_size);
2491 		skb_reset_network_header(skb);
2492 
2493 		/* Push new labels */
2494 		hdr = mpls_hdr(skb);
2495 		bos = true;
2496 		for (i = n_labels - 1; i >= 0; i--) {
2497 			hdr[i] = mpls_entry_encode(labels[i],
2498 						   1, 0, bos);
2499 			bos = false;
2500 		}
2501 	}
2502 
2503 	nh = mpls_select_multipath(rt, skb);
2504 	if (!nh) {
2505 		err = -ENETUNREACH;
2506 		goto errout_free;
2507 	}
2508 
2509 	if (hdr_size) {
2510 		skb_pull(skb, hdr_size);
2511 		skb_reset_network_header(skb);
2512 	}
2513 
2514 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2515 			RTM_NEWROUTE, sizeof(*r), 0);
2516 	if (!nlh) {
2517 		err = -EMSGSIZE;
2518 		goto errout_free;
2519 	}
2520 
2521 	r = nlmsg_data(nlh);
2522 	r->rtm_family	 = AF_MPLS;
2523 	r->rtm_dst_len	= 20;
2524 	r->rtm_src_len	= 0;
2525 	r->rtm_table	= RT_TABLE_MAIN;
2526 	r->rtm_type	= RTN_UNICAST;
2527 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2528 	r->rtm_protocol = rt->rt_protocol;
2529 	r->rtm_flags	= 0;
2530 
2531 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2532 		goto nla_put_failure;
2533 
2534 	if (nh->nh_labels &&
2535 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2536 			   nh->nh_label))
2537 		goto nla_put_failure;
2538 
2539 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2540 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2541 			nh->nh_via_alen))
2542 		goto nla_put_failure;
2543 	dev = nh->nh_dev;
2544 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2545 		goto nla_put_failure;
2546 
2547 	nlmsg_end(skb, nlh);
2548 
2549 	err = rtnl_unicast(skb, net, portid);
2550 errout:
2551 	mutex_unlock(&net->mpls.platform_mutex);
2552 	return err;
2553 
2554 nla_put_failure:
2555 	nlmsg_cancel(skb, nlh);
2556 	err = -EMSGSIZE;
2557 errout_free:
2558 	mutex_unlock(&net->mpls.platform_mutex);
2559 	kfree_skb(skb);
2560 	return err;
2561 }
2562 
2563 static int resize_platform_label_table(struct net *net, size_t limit)
2564 {
2565 	size_t size = sizeof(struct mpls_route *) * limit;
2566 	size_t old_limit;
2567 	size_t cp_size;
2568 	struct mpls_route __rcu **labels = NULL, **old;
2569 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2570 	unsigned index;
2571 
2572 	if (size) {
2573 		labels = kvzalloc(size, GFP_KERNEL);
2574 		if (!labels)
2575 			goto nolabels;
2576 	}
2577 
2578 	/* In case the predefined labels need to be populated */
2579 	if (limit > MPLS_LABEL_IPV4NULL) {
2580 		struct net_device *lo = net->loopback_dev;
2581 
2582 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2583 		if (IS_ERR(rt0))
2584 			goto nort0;
2585 
2586 		rt0->rt_nh->nh_dev = lo;
2587 		netdev_hold(lo, &rt0->rt_nh->nh_dev_tracker, GFP_KERNEL);
2588 		rt0->rt_protocol = RTPROT_KERNEL;
2589 		rt0->rt_payload_type = MPT_IPV4;
2590 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2591 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2592 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2593 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2594 		       lo->addr_len);
2595 	}
2596 	if (limit > MPLS_LABEL_IPV6NULL) {
2597 		struct net_device *lo = net->loopback_dev;
2598 
2599 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2600 		if (IS_ERR(rt2))
2601 			goto nort2;
2602 
2603 		rt2->rt_nh->nh_dev = lo;
2604 		netdev_hold(lo, &rt2->rt_nh->nh_dev_tracker, GFP_KERNEL);
2605 		rt2->rt_protocol = RTPROT_KERNEL;
2606 		rt2->rt_payload_type = MPT_IPV6;
2607 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2608 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2609 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2610 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2611 		       lo->addr_len);
2612 	}
2613 
2614 	mutex_lock(&net->mpls.platform_mutex);
2615 
2616 	/* Remember the original table */
2617 	old = mpls_dereference(net, net->mpls.platform_label);
2618 	old_limit = net->mpls.platform_labels;
2619 
2620 	/* Free any labels beyond the new table */
2621 	for (index = limit; index < old_limit; index++)
2622 		mpls_route_update(net, index, NULL, NULL);
2623 
2624 	/* Copy over the old labels */
2625 	cp_size = size;
2626 	if (old_limit < limit)
2627 		cp_size = old_limit * sizeof(struct mpls_route *);
2628 
2629 	memcpy(labels, old, cp_size);
2630 
2631 	/* If needed set the predefined labels */
2632 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2633 	    (limit > MPLS_LABEL_IPV6NULL)) {
2634 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2635 		rt2 = NULL;
2636 	}
2637 
2638 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2639 	    (limit > MPLS_LABEL_IPV4NULL)) {
2640 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2641 		rt0 = NULL;
2642 	}
2643 
2644 	/* Update the global pointers */
2645 	net->mpls.platform_labels = limit;
2646 	rcu_assign_pointer(net->mpls.platform_label, labels);
2647 
2648 	mutex_unlock(&net->mpls.platform_mutex);
2649 
2650 	mpls_rt_free(rt2);
2651 	mpls_rt_free(rt0);
2652 
2653 	if (old) {
2654 		synchronize_rcu();
2655 		kvfree(old);
2656 	}
2657 	return 0;
2658 
2659 nort2:
2660 	mpls_rt_free(rt0);
2661 nort0:
2662 	kvfree(labels);
2663 nolabels:
2664 	return -ENOMEM;
2665 }
2666 
2667 static int mpls_platform_labels(const struct ctl_table *table, int write,
2668 				void *buffer, size_t *lenp, loff_t *ppos)
2669 {
2670 	struct net *net = table->data;
2671 	int platform_labels = net->mpls.platform_labels;
2672 	int ret;
2673 	struct ctl_table tmp = {
2674 		.procname	= table->procname,
2675 		.data		= &platform_labels,
2676 		.maxlen		= sizeof(int),
2677 		.mode		= table->mode,
2678 		.extra1		= SYSCTL_ZERO,
2679 		.extra2		= &label_limit,
2680 	};
2681 
2682 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2683 
2684 	if (write && ret == 0)
2685 		ret = resize_platform_label_table(net, platform_labels);
2686 
2687 	return ret;
2688 }
2689 
2690 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2691 	(&((struct net *)0)->field)
2692 
2693 static const struct ctl_table mpls_table[] = {
2694 	{
2695 		.procname	= "platform_labels",
2696 		.data		= NULL,
2697 		.maxlen		= sizeof(int),
2698 		.mode		= 0644,
2699 		.proc_handler	= mpls_platform_labels,
2700 	},
2701 	{
2702 		.procname	= "ip_ttl_propagate",
2703 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2704 		.maxlen		= sizeof(int),
2705 		.mode		= 0644,
2706 		.proc_handler	= proc_dointvec_minmax,
2707 		.extra1		= SYSCTL_ZERO,
2708 		.extra2		= SYSCTL_ONE,
2709 	},
2710 	{
2711 		.procname	= "default_ttl",
2712 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2713 		.maxlen		= sizeof(int),
2714 		.mode		= 0644,
2715 		.proc_handler	= proc_dointvec_minmax,
2716 		.extra1		= SYSCTL_ONE,
2717 		.extra2		= &ttl_max,
2718 	},
2719 };
2720 
2721 static __net_init int mpls_net_init(struct net *net)
2722 {
2723 	size_t table_size = ARRAY_SIZE(mpls_table);
2724 	struct ctl_table *table;
2725 	int i;
2726 
2727 	mutex_init(&net->mpls.platform_mutex);
2728 	net->mpls.platform_labels = 0;
2729 	net->mpls.platform_label = NULL;
2730 	net->mpls.ip_ttl_propagate = 1;
2731 	net->mpls.default_ttl = 255;
2732 
2733 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2734 	if (table == NULL)
2735 		return -ENOMEM;
2736 
2737 	/* Table data contains only offsets relative to the base of
2738 	 * the mdev at this point, so make them absolute.
2739 	 */
2740 	for (i = 0; i < table_size; i++)
2741 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2742 
2743 	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2744 					       table_size);
2745 	if (net->mpls.ctl == NULL) {
2746 		kfree(table);
2747 		return -ENOMEM;
2748 	}
2749 
2750 	return 0;
2751 }
2752 
2753 static __net_exit void mpls_net_exit(struct net *net)
2754 {
2755 	struct mpls_route __rcu **platform_label;
2756 	size_t platform_labels;
2757 	const struct ctl_table *table;
2758 	unsigned int index;
2759 
2760 	table = net->mpls.ctl->ctl_table_arg;
2761 	unregister_net_sysctl_table(net->mpls.ctl);
2762 	kfree(table);
2763 
2764 	/* An rcu grace period has passed since there was a device in
2765 	 * the network namespace (and thus the last in flight packet)
2766 	 * left this network namespace.  This is because
2767 	 * unregister_netdevice_many and netdev_run_todo has completed
2768 	 * for each network device that was in this network namespace.
2769 	 *
2770 	 * As such no additional rcu synchronization is necessary when
2771 	 * freeing the platform_label table.
2772 	 */
2773 	mutex_lock(&net->mpls.platform_mutex);
2774 
2775 	platform_label = mpls_dereference(net, net->mpls.platform_label);
2776 	platform_labels = net->mpls.platform_labels;
2777 
2778 	for (index = 0; index < platform_labels; index++) {
2779 		struct mpls_route *rt;
2780 
2781 		rt = mpls_dereference(net, platform_label[index]);
2782 		mpls_notify_route(net, index, rt, NULL, NULL);
2783 		mpls_rt_free(rt);
2784 	}
2785 
2786 	mutex_unlock(&net->mpls.platform_mutex);
2787 
2788 	kvfree(platform_label);
2789 }
2790 
2791 static struct pernet_operations mpls_net_ops = {
2792 	.init = mpls_net_init,
2793 	.exit = mpls_net_exit,
2794 };
2795 
2796 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2797 	.family		   = AF_MPLS,
2798 	.fill_stats_af	   = mpls_fill_stats_af,
2799 	.get_stats_af_size = mpls_get_stats_af_size,
2800 };
2801 
2802 static const struct rtnl_msg_handler mpls_rtnl_msg_handlers[] __initdata_or_module = {
2803 	{THIS_MODULE, PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL,
2804 	 RTNL_FLAG_DOIT_UNLOCKED},
2805 	{THIS_MODULE, PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL,
2806 	 RTNL_FLAG_DOIT_UNLOCKED},
2807 	{THIS_MODULE, PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes,
2808 	 RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
2809 	{THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2810 	 mpls_netconf_get_devconf, mpls_netconf_dump_devconf,
2811 	 RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
2812 };
2813 
2814 static int __init mpls_init(void)
2815 {
2816 	int err;
2817 
2818 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2819 
2820 	err = register_pernet_subsys(&mpls_net_ops);
2821 	if (err)
2822 		goto out;
2823 
2824 	err = register_netdevice_notifier(&mpls_dev_notifier);
2825 	if (err)
2826 		goto out_unregister_pernet;
2827 
2828 	dev_add_pack(&mpls_packet_type);
2829 
2830 	err = rtnl_af_register(&mpls_af_ops);
2831 	if (err)
2832 		goto out_unregister_dev_type;
2833 
2834 	err = rtnl_register_many(mpls_rtnl_msg_handlers);
2835 	if (err)
2836 		goto out_unregister_rtnl_af;
2837 
2838 	err = ipgre_tunnel_encap_add_mpls_ops();
2839 	if (err) {
2840 		pr_err("Can't add mpls over gre tunnel ops\n");
2841 		goto out_unregister_rtnl;
2842 	}
2843 
2844 	err = 0;
2845 out:
2846 	return err;
2847 
2848 out_unregister_rtnl:
2849 	rtnl_unregister_many(mpls_rtnl_msg_handlers);
2850 out_unregister_rtnl_af:
2851 	rtnl_af_unregister(&mpls_af_ops);
2852 out_unregister_dev_type:
2853 	dev_remove_pack(&mpls_packet_type);
2854 	unregister_netdevice_notifier(&mpls_dev_notifier);
2855 out_unregister_pernet:
2856 	unregister_pernet_subsys(&mpls_net_ops);
2857 	goto out;
2858 }
2859 module_init(mpls_init);
2860 
2861 static void __exit mpls_exit(void)
2862 {
2863 	rtnl_unregister_all(PF_MPLS);
2864 	rtnl_af_unregister(&mpls_af_ops);
2865 	dev_remove_pack(&mpls_packet_type);
2866 	unregister_netdevice_notifier(&mpls_dev_notifier);
2867 	unregister_pernet_subsys(&mpls_net_ops);
2868 	ipgre_tunnel_encap_del_mpls_ops();
2869 }
2870 module_exit(mpls_exit);
2871 
2872 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2873 MODULE_LICENSE("GPL v2");
2874 MODULE_ALIAS_NETPROTO(PF_MPLS);
2875