xref: /linux/net/mpls/af_mpls.c (revision 8ffb33d7709b59ff60560f48960a73bd8a55be95)
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/rtnh.h>
28 #include "internal.h"
29 
30 /* max memory we will use for mpls_route */
31 #define MAX_MPLS_ROUTE_MEM	4096
32 
33 /* Maximum number of labels to look ahead at when selecting a path of
34  * a multipath route
35  */
36 #define MAX_MP_SELECT_LABELS 4
37 
38 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
39 
40 static int label_limit = (1 << 20) - 1;
41 static int ttl_max = 255;
42 
43 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
44 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
45 {
46 	return sizeof(struct mpls_shim_hdr);
47 }
48 
49 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
50 	.encap_hlen	= ipgre_mpls_encap_hlen,
51 };
52 
53 static int ipgre_tunnel_encap_add_mpls_ops(void)
54 {
55 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
56 }
57 
58 static void ipgre_tunnel_encap_del_mpls_ops(void)
59 {
60 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
61 }
62 #else
63 static int ipgre_tunnel_encap_add_mpls_ops(void)
64 {
65 	return 0;
66 }
67 
68 static void ipgre_tunnel_encap_del_mpls_ops(void)
69 {
70 }
71 #endif
72 
73 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
74 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
75 		       unsigned int nlm_flags);
76 
77 static struct mpls_route *mpls_route_input(struct net *net, unsigned int index)
78 {
79 	struct mpls_route __rcu **platform_label;
80 
81 	platform_label = mpls_dereference(net, net->mpls.platform_label);
82 	return mpls_dereference(net, platform_label[index]);
83 }
84 
85 static struct mpls_route __rcu **mpls_platform_label_rcu(struct net *net, size_t *platform_labels)
86 {
87 	struct mpls_route __rcu **platform_label;
88 	unsigned int sequence;
89 
90 	do {
91 		sequence = read_seqcount_begin(&net->mpls.platform_label_seq);
92 		platform_label = rcu_dereference(net->mpls.platform_label);
93 		*platform_labels = net->mpls.platform_labels;
94 	} while (read_seqcount_retry(&net->mpls.platform_label_seq, sequence));
95 
96 	return platform_label;
97 }
98 
99 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned int index)
100 {
101 	struct mpls_route __rcu **platform_label;
102 	size_t platform_labels;
103 
104 	platform_label = mpls_platform_label_rcu(net, &platform_labels);
105 
106 	if (index >= platform_labels)
107 		return NULL;
108 
109 	return rcu_dereference(platform_label[index]);
110 }
111 
112 bool mpls_output_possible(const struct net_device *dev)
113 {
114 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
115 }
116 EXPORT_SYMBOL_GPL(mpls_output_possible);
117 
118 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
119 {
120 	return (u8 *)nh + rt->rt_via_offset;
121 }
122 
123 static const u8 *mpls_nh_via(const struct mpls_route *rt,
124 			     const struct mpls_nh *nh)
125 {
126 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
127 }
128 
129 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
130 {
131 	/* The size of the layer 2.5 labels to be added for this route */
132 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
133 }
134 
135 unsigned int mpls_dev_mtu(const struct net_device *dev)
136 {
137 	/* The amount of data the layer 2 frame can hold */
138 	return dev->mtu;
139 }
140 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
141 
142 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
143 {
144 	if (skb->len <= mtu)
145 		return false;
146 
147 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
148 		return false;
149 
150 	return true;
151 }
152 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
153 
154 void mpls_stats_inc_outucastpkts(struct net *net,
155 				 struct net_device *dev,
156 				 const struct sk_buff *skb)
157 {
158 	struct mpls_dev *mdev;
159 
160 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
161 		mdev = mpls_dev_rcu(dev);
162 		if (mdev)
163 			MPLS_INC_STATS_LEN(mdev, skb->len,
164 					   tx_packets,
165 					   tx_bytes);
166 	} else if (skb->protocol == htons(ETH_P_IP)) {
167 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
168 #if IS_ENABLED(CONFIG_IPV6)
169 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
170 		struct inet6_dev *in6dev = in6_dev_rcu(dev);
171 
172 		if (in6dev)
173 			IP6_UPD_PO_STATS(net, in6dev,
174 					 IPSTATS_MIB_OUT, skb->len);
175 #endif
176 	}
177 }
178 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
179 
180 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
181 {
182 	struct mpls_entry_decoded dec;
183 	unsigned int mpls_hdr_len = 0;
184 	struct mpls_shim_hdr *hdr;
185 	bool eli_seen = false;
186 	int label_index;
187 	u32 hash = 0;
188 
189 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
190 	     label_index++) {
191 		mpls_hdr_len += sizeof(*hdr);
192 		if (!pskb_may_pull(skb, mpls_hdr_len))
193 			break;
194 
195 		/* Read and decode the current label */
196 		hdr = mpls_hdr(skb) + label_index;
197 		dec = mpls_entry_decode(hdr);
198 
199 		/* RFC6790 - reserved labels MUST NOT be used as keys
200 		 * for the load-balancing function
201 		 */
202 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
203 			hash = jhash_1word(dec.label, hash);
204 
205 			/* The entropy label follows the entropy label
206 			 * indicator, so this means that the entropy
207 			 * label was just added to the hash - no need to
208 			 * go any deeper either in the label stack or in the
209 			 * payload
210 			 */
211 			if (eli_seen)
212 				break;
213 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
214 			eli_seen = true;
215 		}
216 
217 		if (!dec.bos)
218 			continue;
219 
220 		/* found bottom label; does skb have room for a header? */
221 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
222 			const struct iphdr *v4hdr;
223 
224 			v4hdr = (const struct iphdr *)(hdr + 1);
225 			if (v4hdr->version == 4) {
226 				hash = jhash_3words(ntohl(v4hdr->saddr),
227 						    ntohl(v4hdr->daddr),
228 						    v4hdr->protocol, hash);
229 			} else if (v4hdr->version == 6 &&
230 				   pskb_may_pull(skb, mpls_hdr_len +
231 						 sizeof(struct ipv6hdr))) {
232 				const struct ipv6hdr *v6hdr;
233 
234 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
235 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
236 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
237 				hash = jhash_1word(v6hdr->nexthdr, hash);
238 			}
239 		}
240 
241 		break;
242 	}
243 
244 	return hash;
245 }
246 
247 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
248 {
249 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
250 }
251 
252 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
253  * a next hop (nh->nh_flags) are modified by netdev event handlers.
254  * Since those fields can change at any moment, use READ_ONCE to
255  * access both.
256  */
257 static const struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
258 						   struct sk_buff *skb)
259 {
260 	u32 hash = 0;
261 	int nh_index = 0;
262 	int n = 0;
263 	u8 alive;
264 
265 	/* No need to look further into packet if there's only
266 	 * one path
267 	 */
268 	if (rt->rt_nhn == 1)
269 		return rt->rt_nh;
270 
271 	alive = READ_ONCE(rt->rt_nhn_alive);
272 	if (alive == 0)
273 		return NULL;
274 
275 	hash = mpls_multipath_hash(rt, skb);
276 	nh_index = hash % alive;
277 	if (alive == rt->rt_nhn)
278 		goto out;
279 	for_nexthops(rt) {
280 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
281 
282 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
283 			continue;
284 		if (n == nh_index)
285 			return nh;
286 		n++;
287 	} endfor_nexthops(rt);
288 
289 out:
290 	return mpls_get_nexthop(rt, nh_index);
291 }
292 
293 static bool mpls_egress(struct net *net, struct mpls_route *rt,
294 			struct sk_buff *skb, struct mpls_entry_decoded dec)
295 {
296 	enum mpls_payload_type payload_type;
297 	bool success = false;
298 
299 	/* The IPv4 code below accesses through the IPv4 header
300 	 * checksum, which is 12 bytes into the packet.
301 	 * The IPv6 code below accesses through the IPv6 hop limit
302 	 * which is 8 bytes into the packet.
303 	 *
304 	 * For all supported cases there should always be at least 12
305 	 * bytes of packet data present.  The IPv4 header is 20 bytes
306 	 * without options and the IPv6 header is always 40 bytes
307 	 * long.
308 	 */
309 	if (!pskb_may_pull(skb, 12))
310 		return false;
311 
312 	payload_type = rt->rt_payload_type;
313 	if (payload_type == MPT_UNSPEC)
314 		payload_type = ip_hdr(skb)->version;
315 
316 	switch (payload_type) {
317 	case MPT_IPV4: {
318 		struct iphdr *hdr4 = ip_hdr(skb);
319 		u8 new_ttl;
320 		skb->protocol = htons(ETH_P_IP);
321 
322 		/* If propagating TTL, take the decremented TTL from
323 		 * the incoming MPLS header, otherwise decrement the
324 		 * TTL, but only if not 0 to avoid underflow.
325 		 */
326 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
327 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
328 		     net->mpls.ip_ttl_propagate))
329 			new_ttl = dec.ttl;
330 		else
331 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
332 
333 		csum_replace2(&hdr4->check,
334 			      htons(hdr4->ttl << 8),
335 			      htons(new_ttl << 8));
336 		hdr4->ttl = new_ttl;
337 		success = true;
338 		break;
339 	}
340 	case MPT_IPV6: {
341 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
342 		skb->protocol = htons(ETH_P_IPV6);
343 
344 		/* If propagating TTL, take the decremented TTL from
345 		 * the incoming MPLS header, otherwise decrement the
346 		 * hop limit, but only if not 0 to avoid underflow.
347 		 */
348 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
349 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
350 		     net->mpls.ip_ttl_propagate))
351 			hdr6->hop_limit = dec.ttl;
352 		else if (hdr6->hop_limit)
353 			hdr6->hop_limit = hdr6->hop_limit - 1;
354 		success = true;
355 		break;
356 	}
357 	case MPT_UNSPEC:
358 		/* Should have decided which protocol it is by now */
359 		break;
360 	}
361 
362 	return success;
363 }
364 
365 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
366 			struct packet_type *pt, struct net_device *orig_dev)
367 {
368 	struct net *net = dev_net_rcu(dev);
369 	struct mpls_shim_hdr *hdr;
370 	const struct mpls_nh *nh;
371 	struct mpls_route *rt;
372 	struct mpls_entry_decoded dec;
373 	struct net_device *out_dev;
374 	struct mpls_dev *out_mdev;
375 	struct mpls_dev *mdev;
376 	unsigned int hh_len;
377 	unsigned int new_header_size;
378 	unsigned int mtu;
379 	int err;
380 
381 	/* Careful this entire function runs inside of an rcu critical section */
382 
383 	mdev = mpls_dev_rcu(dev);
384 	if (!mdev)
385 		goto drop;
386 
387 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
388 			   rx_bytes);
389 
390 	if (!mdev->input_enabled) {
391 		MPLS_INC_STATS(mdev, rx_dropped);
392 		goto drop;
393 	}
394 
395 	if (skb->pkt_type != PACKET_HOST)
396 		goto err;
397 
398 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
399 		goto err;
400 
401 	if (!pskb_may_pull(skb, sizeof(*hdr)))
402 		goto err;
403 
404 	skb_dst_drop(skb);
405 
406 	/* Read and decode the label */
407 	hdr = mpls_hdr(skb);
408 	dec = mpls_entry_decode(hdr);
409 
410 	rt = mpls_route_input_rcu(net, dec.label);
411 	if (!rt) {
412 		MPLS_INC_STATS(mdev, rx_noroute);
413 		goto drop;
414 	}
415 
416 	nh = mpls_select_multipath(rt, skb);
417 	if (!nh)
418 		goto err;
419 
420 	/* Pop the label */
421 	skb_pull(skb, sizeof(*hdr));
422 	skb_reset_network_header(skb);
423 
424 	skb_orphan(skb);
425 
426 	if (skb_warn_if_lro(skb))
427 		goto err;
428 
429 	skb_forward_csum(skb);
430 
431 	/* Verify ttl is valid */
432 	if (dec.ttl <= 1)
433 		goto err;
434 
435 	/* Find the output device */
436 	out_dev = nh->nh_dev;
437 	if (!mpls_output_possible(out_dev))
438 		goto tx_err;
439 
440 	/* Verify the destination can hold the packet */
441 	new_header_size = mpls_nh_header_size(nh);
442 	mtu = mpls_dev_mtu(out_dev);
443 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
444 		goto tx_err;
445 
446 	hh_len = LL_RESERVED_SPACE(out_dev);
447 	if (!out_dev->header_ops)
448 		hh_len = 0;
449 
450 	/* Ensure there is enough space for the headers in the skb */
451 	if (skb_cow(skb, hh_len + new_header_size))
452 		goto tx_err;
453 
454 	skb->dev = out_dev;
455 	skb->protocol = htons(ETH_P_MPLS_UC);
456 
457 	dec.ttl -= 1;
458 	if (unlikely(!new_header_size && dec.bos)) {
459 		/* Penultimate hop popping */
460 		if (!mpls_egress(net, rt, skb, dec))
461 			goto err;
462 	} else {
463 		bool bos;
464 		int i;
465 		skb_push(skb, new_header_size);
466 		skb_reset_network_header(skb);
467 		/* Push the new labels */
468 		hdr = mpls_hdr(skb);
469 		bos = dec.bos;
470 		for (i = nh->nh_labels - 1; i >= 0; i--) {
471 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
472 						   dec.ttl, 0, bos);
473 			bos = false;
474 		}
475 	}
476 
477 	mpls_stats_inc_outucastpkts(net, out_dev, skb);
478 
479 	/* If via wasn't specified then send out using device address */
480 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
481 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
482 				 out_dev->dev_addr, skb);
483 	else
484 		err = neigh_xmit(nh->nh_via_table, out_dev,
485 				 mpls_nh_via(rt, nh), skb);
486 	if (err)
487 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
488 				    __func__, err);
489 	return 0;
490 
491 tx_err:
492 	out_mdev = out_dev ? mpls_dev_rcu(out_dev) : NULL;
493 	if (out_mdev)
494 		MPLS_INC_STATS(out_mdev, tx_errors);
495 	goto drop;
496 err:
497 	MPLS_INC_STATS(mdev, rx_errors);
498 drop:
499 	kfree_skb(skb);
500 	return NET_RX_DROP;
501 }
502 
503 static struct packet_type mpls_packet_type __read_mostly = {
504 	.type = cpu_to_be16(ETH_P_MPLS_UC),
505 	.func = mpls_forward,
506 };
507 
508 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
509 	[RTA_DST]		= { .type = NLA_U32 },
510 	[RTA_OIF]		= { .type = NLA_U32 },
511 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
512 };
513 
514 struct mpls_route_config {
515 	u32			rc_protocol;
516 	u32			rc_ifindex;
517 	u8			rc_via_table;
518 	u8			rc_via_alen;
519 	u8			rc_via[MAX_VIA_ALEN];
520 	u32			rc_label;
521 	u8			rc_ttl_propagate;
522 	u8			rc_output_labels;
523 	u32			rc_output_label[MAX_NEW_LABELS];
524 	u32			rc_nlflags;
525 	enum mpls_payload_type	rc_payload_type;
526 	struct nl_info		rc_nlinfo;
527 	struct rtnexthop	*rc_mp;
528 	int			rc_mp_len;
529 };
530 
531 /* all nexthops within a route have the same size based on max
532  * number of labels and max via length for a hop
533  */
534 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
535 {
536 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
537 	struct mpls_route *rt;
538 	size_t size;
539 
540 	size = sizeof(*rt) + num_nh * nh_size;
541 	if (size > MAX_MPLS_ROUTE_MEM)
542 		return ERR_PTR(-EINVAL);
543 
544 	rt = kzalloc(size, GFP_KERNEL);
545 	if (!rt)
546 		return ERR_PTR(-ENOMEM);
547 
548 	rt->rt_nhn = num_nh;
549 	rt->rt_nhn_alive = num_nh;
550 	rt->rt_nh_size = nh_size;
551 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
552 
553 	return rt;
554 }
555 
556 static void mpls_rt_free_rcu(struct rcu_head *head)
557 {
558 	struct mpls_route *rt;
559 
560 	rt = container_of(head, struct mpls_route, rt_rcu);
561 
562 	change_nexthops(rt) {
563 		netdev_put(nh->nh_dev, &nh->nh_dev_tracker);
564 	} endfor_nexthops(rt);
565 
566 	kfree(rt);
567 }
568 
569 static void mpls_rt_free(struct mpls_route *rt)
570 {
571 	if (rt)
572 		call_rcu(&rt->rt_rcu, mpls_rt_free_rcu);
573 }
574 
575 static void mpls_notify_route(struct net *net, unsigned index,
576 			      struct mpls_route *old, struct mpls_route *new,
577 			      const struct nl_info *info)
578 {
579 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
580 	unsigned portid = info ? info->portid : 0;
581 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
582 	struct mpls_route *rt = new ? new : old;
583 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
584 	/* Ignore reserved labels for now */
585 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
586 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
587 }
588 
589 static void mpls_route_update(struct net *net, unsigned index,
590 			      struct mpls_route *new,
591 			      const struct nl_info *info)
592 {
593 	struct mpls_route __rcu **platform_label;
594 	struct mpls_route *rt;
595 
596 	platform_label = mpls_dereference(net, net->mpls.platform_label);
597 	rt = mpls_dereference(net, platform_label[index]);
598 	rcu_assign_pointer(platform_label[index], new);
599 
600 	mpls_notify_route(net, index, rt, new, info);
601 
602 	/* If we removed a route free it now */
603 	mpls_rt_free(rt);
604 }
605 
606 static unsigned int find_free_label(struct net *net)
607 {
608 	unsigned int index;
609 
610 	for (index = MPLS_LABEL_FIRST_UNRESERVED;
611 	     index < net->mpls.platform_labels;
612 	     index++) {
613 		if (!mpls_route_input(net, index))
614 			return index;
615 	}
616 
617 	return LABEL_NOT_SPECIFIED;
618 }
619 
620 #if IS_ENABLED(CONFIG_INET)
621 static struct net_device *inet_fib_lookup_dev(struct net *net,
622 					      struct mpls_nh *nh,
623 					      const void *addr)
624 {
625 	struct net_device *dev;
626 	struct rtable *rt;
627 	struct in_addr daddr;
628 
629 	memcpy(&daddr, addr, sizeof(struct in_addr));
630 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0, RT_SCOPE_UNIVERSE);
631 	if (IS_ERR(rt))
632 		return ERR_CAST(rt);
633 
634 	dev = rt->dst.dev;
635 	netdev_hold(dev, &nh->nh_dev_tracker, GFP_KERNEL);
636 	ip_rt_put(rt);
637 
638 	return dev;
639 }
640 #else
641 static struct net_device *inet_fib_lookup_dev(struct net *net,
642 					      struct mpls_nh *nh,
643 					      const void *addr)
644 {
645 	return ERR_PTR(-EAFNOSUPPORT);
646 }
647 #endif
648 
649 #if IS_ENABLED(CONFIG_IPV6)
650 static struct net_device *inet6_fib_lookup_dev(struct net *net,
651 					       struct mpls_nh *nh,
652 					       const void *addr)
653 {
654 	struct net_device *dev;
655 	struct dst_entry *dst;
656 	struct flowi6 fl6;
657 
658 	memset(&fl6, 0, sizeof(fl6));
659 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
660 	dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL);
661 	if (IS_ERR(dst))
662 		return ERR_CAST(dst);
663 
664 	dev = dst->dev;
665 	netdev_hold(dev, &nh->nh_dev_tracker, GFP_KERNEL);
666 	dst_release(dst);
667 
668 	return dev;
669 }
670 #else
671 static struct net_device *inet6_fib_lookup_dev(struct net *net,
672 					       struct mpls_nh *nh,
673 					       const void *addr)
674 {
675 	return ERR_PTR(-EAFNOSUPPORT);
676 }
677 #endif
678 
679 static struct net_device *find_outdev(struct net *net,
680 				      struct mpls_route *rt,
681 				      struct mpls_nh *nh, int oif)
682 {
683 	struct net_device *dev = NULL;
684 
685 	if (!oif) {
686 		switch (nh->nh_via_table) {
687 		case NEIGH_ARP_TABLE:
688 			dev = inet_fib_lookup_dev(net, nh, mpls_nh_via(rt, nh));
689 			break;
690 		case NEIGH_ND_TABLE:
691 			dev = inet6_fib_lookup_dev(net, nh, mpls_nh_via(rt, nh));
692 			break;
693 		case NEIGH_LINK_TABLE:
694 			break;
695 		}
696 	} else {
697 		dev = netdev_get_by_index(net, oif,
698 					  &nh->nh_dev_tracker, GFP_KERNEL);
699 	}
700 
701 	if (!dev)
702 		return ERR_PTR(-ENODEV);
703 
704 	if (IS_ERR(dev))
705 		return dev;
706 
707 	nh->nh_dev = dev;
708 
709 	return dev;
710 }
711 
712 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
713 			      struct mpls_nh *nh, int oif)
714 {
715 	struct net_device *dev = NULL;
716 	int err = -ENODEV;
717 
718 	dev = find_outdev(net, rt, nh, oif);
719 	if (IS_ERR(dev)) {
720 		err = PTR_ERR(dev);
721 		goto errout;
722 	}
723 
724 	/* Ensure this is a supported device */
725 	err = -EINVAL;
726 	if (!mpls_dev_get(net, dev))
727 		goto errout_put;
728 
729 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
730 	    (dev->addr_len != nh->nh_via_alen))
731 		goto errout_put;
732 
733 	if (!(dev->flags & IFF_UP)) {
734 		nh->nh_flags |= RTNH_F_DEAD;
735 	} else {
736 		unsigned int flags;
737 
738 		flags = netif_get_flags(dev);
739 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
740 			nh->nh_flags |= RTNH_F_LINKDOWN;
741 	}
742 
743 	return 0;
744 
745 errout_put:
746 	netdev_put(nh->nh_dev, &nh->nh_dev_tracker);
747 	nh->nh_dev = NULL;
748 errout:
749 	return err;
750 }
751 
752 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
753 		       u8 via_addr[], struct netlink_ext_ack *extack)
754 {
755 	struct rtvia *via = nla_data(nla);
756 	int err = -EINVAL;
757 	int alen;
758 
759 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
760 		NL_SET_ERR_MSG_ATTR(extack, nla,
761 				    "Invalid attribute length for RTA_VIA");
762 		goto errout;
763 	}
764 	alen = nla_len(nla) -
765 			offsetof(struct rtvia, rtvia_addr);
766 	if (alen > MAX_VIA_ALEN) {
767 		NL_SET_ERR_MSG_ATTR(extack, nla,
768 				    "Invalid address length for RTA_VIA");
769 		goto errout;
770 	}
771 
772 	/* Validate the address family */
773 	switch (via->rtvia_family) {
774 	case AF_PACKET:
775 		*via_table = NEIGH_LINK_TABLE;
776 		break;
777 	case AF_INET:
778 		*via_table = NEIGH_ARP_TABLE;
779 		if (alen != 4)
780 			goto errout;
781 		break;
782 	case AF_INET6:
783 		*via_table = NEIGH_ND_TABLE;
784 		if (alen != 16)
785 			goto errout;
786 		break;
787 	default:
788 		/* Unsupported address family */
789 		goto errout;
790 	}
791 
792 	memcpy(via_addr, via->rtvia_addr, alen);
793 	*via_alen = alen;
794 	err = 0;
795 
796 errout:
797 	return err;
798 }
799 
800 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
801 				  struct mpls_route *rt)
802 {
803 	struct net *net = cfg->rc_nlinfo.nl_net;
804 	struct mpls_nh *nh = rt->rt_nh;
805 	int err;
806 	int i;
807 
808 	if (!nh)
809 		return -ENOMEM;
810 
811 	nh->nh_labels = cfg->rc_output_labels;
812 	for (i = 0; i < nh->nh_labels; i++)
813 		nh->nh_label[i] = cfg->rc_output_label[i];
814 
815 	nh->nh_via_table = cfg->rc_via_table;
816 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
817 	nh->nh_via_alen = cfg->rc_via_alen;
818 
819 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
820 	if (err)
821 		goto errout;
822 
823 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
824 		rt->rt_nhn_alive--;
825 
826 	return 0;
827 
828 errout:
829 	return err;
830 }
831 
832 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
833 			 struct mpls_nh *nh, int oif, struct nlattr *via,
834 			 struct nlattr *newdst, u8 max_labels,
835 			 struct netlink_ext_ack *extack)
836 {
837 	int err = -ENOMEM;
838 
839 	if (!nh)
840 		goto errout;
841 
842 	if (newdst) {
843 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
844 				     nh->nh_label, extack);
845 		if (err)
846 			goto errout;
847 	}
848 
849 	if (via) {
850 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
851 				  __mpls_nh_via(rt, nh), extack);
852 		if (err)
853 			goto errout;
854 	} else {
855 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
856 	}
857 
858 	err = mpls_nh_assign_dev(net, rt, nh, oif);
859 	if (err)
860 		goto errout;
861 
862 	return 0;
863 
864 errout:
865 	return err;
866 }
867 
868 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
869 			      u8 cfg_via_alen, u8 *max_via_alen,
870 			      u8 *max_labels)
871 {
872 	int remaining = len;
873 	u8 nhs = 0;
874 
875 	*max_via_alen = 0;
876 	*max_labels = 0;
877 
878 	while (rtnh_ok(rtnh, remaining)) {
879 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
880 		int attrlen;
881 		u8 n_labels = 0;
882 
883 		attrlen = rtnh_attrlen(rtnh);
884 		nla = nla_find(attrs, attrlen, RTA_VIA);
885 		if (nla && nla_len(nla) >=
886 		    offsetof(struct rtvia, rtvia_addr)) {
887 			int via_alen = nla_len(nla) -
888 				offsetof(struct rtvia, rtvia_addr);
889 
890 			if (via_alen <= MAX_VIA_ALEN)
891 				*max_via_alen = max_t(u16, *max_via_alen,
892 						      via_alen);
893 		}
894 
895 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
896 		if (nla &&
897 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
898 				   NULL, NULL) != 0)
899 			return 0;
900 
901 		*max_labels = max_t(u8, *max_labels, n_labels);
902 
903 		/* number of nexthops is tracked by a u8.
904 		 * Check for overflow.
905 		 */
906 		if (nhs == 255)
907 			return 0;
908 		nhs++;
909 
910 		rtnh = rtnh_next(rtnh, &remaining);
911 	}
912 
913 	/* leftover implies invalid nexthop configuration, discard it */
914 	return remaining > 0 ? 0 : nhs;
915 }
916 
917 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
918 			       struct mpls_route *rt, u8 max_labels,
919 			       struct netlink_ext_ack *extack)
920 {
921 	struct rtnexthop *rtnh = cfg->rc_mp;
922 	struct nlattr *nla_via, *nla_newdst;
923 	int remaining = cfg->rc_mp_len;
924 	int err = 0;
925 
926 	rt->rt_nhn = 0;
927 
928 	change_nexthops(rt) {
929 		int attrlen;
930 
931 		nla_via = NULL;
932 		nla_newdst = NULL;
933 
934 		err = -EINVAL;
935 		if (!rtnh_ok(rtnh, remaining))
936 			goto errout;
937 
938 		/* neither weighted multipath nor any flags
939 		 * are supported
940 		 */
941 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
942 			goto errout;
943 
944 		attrlen = rtnh_attrlen(rtnh);
945 		if (attrlen > 0) {
946 			struct nlattr *attrs = rtnh_attrs(rtnh);
947 
948 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
949 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
950 		}
951 
952 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
953 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
954 				    max_labels, extack);
955 		if (err)
956 			goto errout;
957 
958 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
959 			rt->rt_nhn_alive--;
960 
961 		rtnh = rtnh_next(rtnh, &remaining);
962 		rt->rt_nhn++;
963 	} endfor_nexthops(rt);
964 
965 	return 0;
966 
967 errout:
968 	return err;
969 }
970 
971 static bool mpls_label_ok(struct net *net, unsigned int *index,
972 			  struct netlink_ext_ack *extack)
973 {
974 	/* Reserved labels may not be set */
975 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
976 		NL_SET_ERR_MSG(extack,
977 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
978 		return false;
979 	}
980 
981 	/* The full 20 bit range may not be supported. */
982 	if (*index >= net->mpls.platform_labels) {
983 		NL_SET_ERR_MSG(extack,
984 			       "Label >= configured maximum in platform_labels");
985 		return false;
986 	}
987 
988 	*index = array_index_nospec(*index, net->mpls.platform_labels);
989 
990 	return true;
991 }
992 
993 static int mpls_route_add(struct mpls_route_config *cfg,
994 			  struct netlink_ext_ack *extack)
995 {
996 	struct net *net = cfg->rc_nlinfo.nl_net;
997 	struct mpls_route *rt, *old;
998 	int err = -EINVAL;
999 	u8 max_via_alen;
1000 	unsigned index;
1001 	u8 max_labels;
1002 	u8 nhs;
1003 
1004 	index = cfg->rc_label;
1005 
1006 	/* If a label was not specified during insert pick one */
1007 	if ((index == LABEL_NOT_SPECIFIED) &&
1008 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
1009 		index = find_free_label(net);
1010 	}
1011 
1012 	if (!mpls_label_ok(net, &index, extack))
1013 		goto errout;
1014 
1015 	/* Append makes no sense with mpls */
1016 	err = -EOPNOTSUPP;
1017 	if (cfg->rc_nlflags & NLM_F_APPEND) {
1018 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
1019 		goto errout;
1020 	}
1021 
1022 	err = -EEXIST;
1023 	old = mpls_route_input(net, index);
1024 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
1025 		goto errout;
1026 
1027 	err = -EEXIST;
1028 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1029 		goto errout;
1030 
1031 	err = -ENOENT;
1032 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1033 		goto errout;
1034 
1035 	err = -EINVAL;
1036 	if (cfg->rc_mp) {
1037 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1038 					  cfg->rc_via_alen, &max_via_alen,
1039 					  &max_labels);
1040 	} else {
1041 		max_via_alen = cfg->rc_via_alen;
1042 		max_labels = cfg->rc_output_labels;
1043 		nhs = 1;
1044 	}
1045 
1046 	if (nhs == 0) {
1047 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1048 		goto errout;
1049 	}
1050 
1051 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1052 	if (IS_ERR(rt)) {
1053 		err = PTR_ERR(rt);
1054 		goto errout;
1055 	}
1056 
1057 	rt->rt_protocol = cfg->rc_protocol;
1058 	rt->rt_payload_type = cfg->rc_payload_type;
1059 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1060 
1061 	if (cfg->rc_mp)
1062 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1063 	else
1064 		err = mpls_nh_build_from_cfg(cfg, rt);
1065 	if (err)
1066 		goto freert;
1067 
1068 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1069 
1070 	return 0;
1071 
1072 freert:
1073 	mpls_rt_free(rt);
1074 errout:
1075 	return err;
1076 }
1077 
1078 static int mpls_route_del(struct mpls_route_config *cfg,
1079 			  struct netlink_ext_ack *extack)
1080 {
1081 	struct net *net = cfg->rc_nlinfo.nl_net;
1082 	unsigned index;
1083 	int err = -EINVAL;
1084 
1085 	index = cfg->rc_label;
1086 
1087 	if (!mpls_label_ok(net, &index, extack))
1088 		goto errout;
1089 
1090 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1091 
1092 	err = 0;
1093 errout:
1094 	return err;
1095 }
1096 
1097 static void mpls_get_stats(struct mpls_dev *mdev,
1098 			   struct mpls_link_stats *stats)
1099 {
1100 	struct mpls_pcpu_stats *p;
1101 	int i;
1102 
1103 	memset(stats, 0, sizeof(*stats));
1104 
1105 	for_each_possible_cpu(i) {
1106 		struct mpls_link_stats local;
1107 		unsigned int start;
1108 
1109 		p = per_cpu_ptr(mdev->stats, i);
1110 		do {
1111 			start = u64_stats_fetch_begin(&p->syncp);
1112 			local = p->stats;
1113 		} while (u64_stats_fetch_retry(&p->syncp, start));
1114 
1115 		stats->rx_packets	+= local.rx_packets;
1116 		stats->rx_bytes		+= local.rx_bytes;
1117 		stats->tx_packets	+= local.tx_packets;
1118 		stats->tx_bytes		+= local.tx_bytes;
1119 		stats->rx_errors	+= local.rx_errors;
1120 		stats->tx_errors	+= local.tx_errors;
1121 		stats->rx_dropped	+= local.rx_dropped;
1122 		stats->tx_dropped	+= local.tx_dropped;
1123 		stats->rx_noroute	+= local.rx_noroute;
1124 	}
1125 }
1126 
1127 static int mpls_fill_stats_af(struct sk_buff *skb,
1128 			      const struct net_device *dev)
1129 {
1130 	struct mpls_link_stats *stats;
1131 	struct mpls_dev *mdev;
1132 	struct nlattr *nla;
1133 
1134 	mdev = mpls_dev_rcu(dev);
1135 	if (!mdev)
1136 		return -ENODATA;
1137 
1138 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1139 				sizeof(struct mpls_link_stats),
1140 				MPLS_STATS_UNSPEC);
1141 	if (!nla)
1142 		return -EMSGSIZE;
1143 
1144 	stats = nla_data(nla);
1145 	mpls_get_stats(mdev, stats);
1146 
1147 	return 0;
1148 }
1149 
1150 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1151 {
1152 	struct mpls_dev *mdev;
1153 
1154 	mdev = mpls_dev_rcu(dev);
1155 	if (!mdev)
1156 		return 0;
1157 
1158 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1159 }
1160 
1161 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1162 				     u32 portid, u32 seq, int event,
1163 				     unsigned int flags, int type)
1164 {
1165 	struct nlmsghdr  *nlh;
1166 	struct netconfmsg *ncm;
1167 	bool all = false;
1168 
1169 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1170 			flags);
1171 	if (!nlh)
1172 		return -EMSGSIZE;
1173 
1174 	if (type == NETCONFA_ALL)
1175 		all = true;
1176 
1177 	ncm = nlmsg_data(nlh);
1178 	ncm->ncm_family = AF_MPLS;
1179 
1180 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1181 		goto nla_put_failure;
1182 
1183 	if ((all || type == NETCONFA_INPUT) &&
1184 	    nla_put_s32(skb, NETCONFA_INPUT,
1185 			READ_ONCE(mdev->input_enabled)) < 0)
1186 		goto nla_put_failure;
1187 
1188 	nlmsg_end(skb, nlh);
1189 	return 0;
1190 
1191 nla_put_failure:
1192 	nlmsg_cancel(skb, nlh);
1193 	return -EMSGSIZE;
1194 }
1195 
1196 static int mpls_netconf_msgsize_devconf(int type)
1197 {
1198 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1199 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1200 	bool all = false;
1201 
1202 	if (type == NETCONFA_ALL)
1203 		all = true;
1204 
1205 	if (all || type == NETCONFA_INPUT)
1206 		size += nla_total_size(4);
1207 
1208 	return size;
1209 }
1210 
1211 static void mpls_netconf_notify_devconf(struct net *net, int event,
1212 					int type, struct mpls_dev *mdev)
1213 {
1214 	struct sk_buff *skb;
1215 	int err = -ENOBUFS;
1216 
1217 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1218 	if (!skb)
1219 		goto errout;
1220 
1221 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1222 	if (err < 0) {
1223 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1224 		WARN_ON(err == -EMSGSIZE);
1225 		kfree_skb(skb);
1226 		goto errout;
1227 	}
1228 
1229 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1230 	return;
1231 errout:
1232 	rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1233 }
1234 
1235 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1236 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1237 };
1238 
1239 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1240 				      const struct nlmsghdr *nlh,
1241 				      struct nlattr **tb,
1242 				      struct netlink_ext_ack *extack)
1243 {
1244 	int i, err;
1245 
1246 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1247 		NL_SET_ERR_MSG_MOD(extack,
1248 				   "Invalid header for netconf get request");
1249 		return -EINVAL;
1250 	}
1251 
1252 	if (!netlink_strict_get_check(skb))
1253 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1254 					      tb, NETCONFA_MAX,
1255 					      devconf_mpls_policy, extack);
1256 
1257 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1258 					    tb, NETCONFA_MAX,
1259 					    devconf_mpls_policy, extack);
1260 	if (err)
1261 		return err;
1262 
1263 	for (i = 0; i <= NETCONFA_MAX; i++) {
1264 		if (!tb[i])
1265 			continue;
1266 
1267 		switch (i) {
1268 		case NETCONFA_IFINDEX:
1269 			break;
1270 		default:
1271 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1272 			return -EINVAL;
1273 		}
1274 	}
1275 
1276 	return 0;
1277 }
1278 
1279 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1280 				    struct nlmsghdr *nlh,
1281 				    struct netlink_ext_ack *extack)
1282 {
1283 	struct net *net = sock_net(in_skb->sk);
1284 	struct nlattr *tb[NETCONFA_MAX + 1];
1285 	struct net_device *dev;
1286 	struct mpls_dev *mdev;
1287 	struct sk_buff *skb;
1288 	int ifindex;
1289 	int err;
1290 
1291 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1292 	if (err < 0)
1293 		goto errout;
1294 
1295 	if (!tb[NETCONFA_IFINDEX]) {
1296 		err = -EINVAL;
1297 		goto errout;
1298 	}
1299 
1300 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1301 
1302 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1303 	if (!skb) {
1304 		err = -ENOBUFS;
1305 		goto errout;
1306 	}
1307 
1308 	rcu_read_lock();
1309 
1310 	dev = dev_get_by_index_rcu(net, ifindex);
1311 	if (!dev) {
1312 		err = -EINVAL;
1313 		goto errout_unlock;
1314 	}
1315 
1316 	mdev = mpls_dev_rcu(dev);
1317 	if (!mdev) {
1318 		err = -EINVAL;
1319 		goto errout_unlock;
1320 	}
1321 
1322 	err = mpls_netconf_fill_devconf(skb, mdev,
1323 					NETLINK_CB(in_skb).portid,
1324 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1325 					NETCONFA_ALL);
1326 	if (err < 0) {
1327 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1328 		WARN_ON(err == -EMSGSIZE);
1329 		goto errout_unlock;
1330 	}
1331 
1332 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1333 
1334 	rcu_read_unlock();
1335 errout:
1336 	return err;
1337 
1338 errout_unlock:
1339 	rcu_read_unlock();
1340 	kfree_skb(skb);
1341 	goto errout;
1342 }
1343 
1344 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1345 				     struct netlink_callback *cb)
1346 {
1347 	const struct nlmsghdr *nlh = cb->nlh;
1348 	struct net *net = sock_net(skb->sk);
1349 	struct {
1350 		unsigned long ifindex;
1351 	} *ctx = (void *)cb->ctx;
1352 	struct net_device *dev;
1353 	struct mpls_dev *mdev;
1354 	int err = 0;
1355 
1356 	if (cb->strict_check) {
1357 		struct netlink_ext_ack *extack = cb->extack;
1358 		struct netconfmsg *ncm;
1359 
1360 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1361 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1362 			return -EINVAL;
1363 		}
1364 
1365 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1366 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1367 			return -EINVAL;
1368 		}
1369 	}
1370 
1371 	rcu_read_lock();
1372 	for_each_netdev_dump(net, dev, ctx->ifindex) {
1373 		mdev = mpls_dev_rcu(dev);
1374 		if (!mdev)
1375 			continue;
1376 		err = mpls_netconf_fill_devconf(skb, mdev,
1377 						NETLINK_CB(cb->skb).portid,
1378 						nlh->nlmsg_seq,
1379 						RTM_NEWNETCONF,
1380 						NLM_F_MULTI,
1381 						NETCONFA_ALL);
1382 		if (err < 0)
1383 			break;
1384 	}
1385 	rcu_read_unlock();
1386 
1387 	return err;
1388 }
1389 
1390 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1391 	(&((struct mpls_dev *)0)->field)
1392 
1393 static int mpls_conf_proc(const struct ctl_table *ctl, int write,
1394 			  void *buffer, size_t *lenp, loff_t *ppos)
1395 {
1396 	int oval = *(int *)ctl->data;
1397 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1398 
1399 	if (write) {
1400 		struct mpls_dev *mdev = ctl->extra1;
1401 		int i = (int *)ctl->data - (int *)mdev;
1402 		struct net *net = ctl->extra2;
1403 		int val = *(int *)ctl->data;
1404 
1405 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1406 		    val != oval) {
1407 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1408 						    NETCONFA_INPUT, mdev);
1409 		}
1410 	}
1411 
1412 	return ret;
1413 }
1414 
1415 static const struct ctl_table mpls_dev_table[] = {
1416 	{
1417 		.procname	= "input",
1418 		.maxlen		= sizeof(int),
1419 		.mode		= 0644,
1420 		.proc_handler	= mpls_conf_proc,
1421 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1422 	},
1423 };
1424 
1425 static int mpls_dev_sysctl_register(struct net_device *dev,
1426 				    struct mpls_dev *mdev)
1427 {
1428 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1429 	size_t table_size = ARRAY_SIZE(mpls_dev_table);
1430 	struct net *net = dev_net(dev);
1431 	struct ctl_table *table;
1432 	int i;
1433 
1434 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1435 	if (!table)
1436 		goto out;
1437 
1438 	/* Table data contains only offsets relative to the base of
1439 	 * the mdev at this point, so make them absolute.
1440 	 */
1441 	for (i = 0; i < table_size; i++) {
1442 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1443 		table[i].extra1 = mdev;
1444 		table[i].extra2 = net;
1445 	}
1446 
1447 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1448 
1449 	mdev->sysctl = register_net_sysctl_sz(net, path, table, table_size);
1450 	if (!mdev->sysctl)
1451 		goto free;
1452 
1453 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1454 	return 0;
1455 
1456 free:
1457 	kfree(table);
1458 out:
1459 	mdev->sysctl = NULL;
1460 	return -ENOBUFS;
1461 }
1462 
1463 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1464 				       struct mpls_dev *mdev)
1465 {
1466 	struct net *net = dev_net(dev);
1467 	const struct ctl_table *table;
1468 
1469 	if (!mdev->sysctl)
1470 		return;
1471 
1472 	table = mdev->sysctl->ctl_table_arg;
1473 	unregister_net_sysctl_table(mdev->sysctl);
1474 	kfree(table);
1475 
1476 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1477 }
1478 
1479 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1480 {
1481 	struct mpls_dev *mdev;
1482 	int err = -ENOMEM;
1483 	int i;
1484 
1485 	mdev = kzalloc_obj(*mdev);
1486 	if (!mdev)
1487 		return ERR_PTR(err);
1488 
1489 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1490 	if (!mdev->stats)
1491 		goto free;
1492 
1493 	for_each_possible_cpu(i) {
1494 		struct mpls_pcpu_stats *mpls_stats;
1495 
1496 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1497 		u64_stats_init(&mpls_stats->syncp);
1498 	}
1499 
1500 	mdev->dev = dev;
1501 
1502 	err = mpls_dev_sysctl_register(dev, mdev);
1503 	if (err)
1504 		goto free;
1505 
1506 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1507 
1508 	return mdev;
1509 
1510 free:
1511 	free_percpu(mdev->stats);
1512 	kfree(mdev);
1513 	return ERR_PTR(err);
1514 }
1515 
1516 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1517 {
1518 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1519 
1520 	free_percpu(mdev->stats);
1521 	kfree(mdev);
1522 }
1523 
1524 static int mpls_ifdown(struct net_device *dev, int event)
1525 {
1526 	struct net *net = dev_net(dev);
1527 	unsigned int index;
1528 
1529 	for (index = 0; index < net->mpls.platform_labels; index++) {
1530 		struct mpls_route *rt;
1531 		bool nh_del = false;
1532 		u8 alive = 0;
1533 
1534 		rt = mpls_route_input(net, index);
1535 		if (!rt)
1536 			continue;
1537 
1538 		if (event == NETDEV_UNREGISTER) {
1539 			u8 deleted = 0;
1540 
1541 			for_nexthops(rt) {
1542 				if (!nh->nh_dev || nh->nh_dev == dev)
1543 					deleted++;
1544 				if (nh->nh_dev == dev)
1545 					nh_del = true;
1546 			} endfor_nexthops(rt);
1547 
1548 			/* if there are no more nexthops, delete the route */
1549 			if (deleted == rt->rt_nhn) {
1550 				mpls_route_update(net, index, NULL, NULL);
1551 				continue;
1552 			}
1553 
1554 			if (nh_del) {
1555 				size_t size = sizeof(*rt) + rt->rt_nhn *
1556 					rt->rt_nh_size;
1557 				struct mpls_route *orig = rt;
1558 
1559 				rt = kmemdup(orig, size, GFP_KERNEL);
1560 				if (!rt)
1561 					return -ENOMEM;
1562 			}
1563 		}
1564 
1565 		change_nexthops(rt) {
1566 			unsigned int nh_flags = nh->nh_flags;
1567 
1568 			if (nh->nh_dev != dev) {
1569 				if (nh_del)
1570 					netdev_hold(nh->nh_dev, &nh->nh_dev_tracker,
1571 						    GFP_KERNEL);
1572 				goto next;
1573 			}
1574 
1575 			switch (event) {
1576 			case NETDEV_DOWN:
1577 			case NETDEV_UNREGISTER:
1578 				nh_flags |= RTNH_F_DEAD;
1579 				fallthrough;
1580 			case NETDEV_CHANGE:
1581 				nh_flags |= RTNH_F_LINKDOWN;
1582 				break;
1583 			}
1584 			if (event == NETDEV_UNREGISTER)
1585 				nh->nh_dev = NULL;
1586 
1587 			if (nh->nh_flags != nh_flags)
1588 				WRITE_ONCE(nh->nh_flags, nh_flags);
1589 next:
1590 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1591 				alive++;
1592 		} endfor_nexthops(rt);
1593 
1594 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1595 
1596 		if (nh_del)
1597 			mpls_route_update(net, index, rt, NULL);
1598 	}
1599 
1600 	return 0;
1601 }
1602 
1603 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1604 {
1605 	struct net *net = dev_net(dev);
1606 	unsigned int index;
1607 	u8 alive;
1608 
1609 	for (index = 0; index < net->mpls.platform_labels; index++) {
1610 		struct mpls_route *rt;
1611 
1612 		rt = mpls_route_input(net, index);
1613 		if (!rt)
1614 			continue;
1615 
1616 		alive = 0;
1617 		change_nexthops(rt) {
1618 			unsigned int nh_flags = nh->nh_flags;
1619 
1620 			if (!(nh_flags & flags)) {
1621 				alive++;
1622 				continue;
1623 			}
1624 			if (nh->nh_dev != dev)
1625 				continue;
1626 			alive++;
1627 			nh_flags &= ~flags;
1628 			WRITE_ONCE(nh->nh_flags, nh_flags);
1629 		} endfor_nexthops(rt);
1630 
1631 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1632 	}
1633 }
1634 
1635 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1636 			   void *ptr)
1637 {
1638 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1639 	struct net *net = dev_net(dev);
1640 	struct mpls_dev *mdev;
1641 	unsigned int flags;
1642 	int err;
1643 
1644 	mutex_lock(&net->mpls.platform_mutex);
1645 
1646 	if (event == NETDEV_REGISTER) {
1647 		mdev = mpls_add_dev(dev);
1648 		if (IS_ERR(mdev)) {
1649 			err = PTR_ERR(mdev);
1650 			goto err;
1651 		}
1652 
1653 		goto out;
1654 	}
1655 
1656 	mdev = mpls_dev_get(net, dev);
1657 	if (!mdev)
1658 		goto out;
1659 
1660 	switch (event) {
1661 
1662 	case NETDEV_DOWN:
1663 		err = mpls_ifdown(dev, event);
1664 		if (err)
1665 			goto err;
1666 		break;
1667 	case NETDEV_UP:
1668 		flags = netif_get_flags(dev);
1669 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1670 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1671 		else
1672 			mpls_ifup(dev, RTNH_F_DEAD);
1673 		break;
1674 	case NETDEV_CHANGE:
1675 		flags = netif_get_flags(dev);
1676 		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1677 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1678 		} else {
1679 			err = mpls_ifdown(dev, event);
1680 			if (err)
1681 				goto err;
1682 		}
1683 		break;
1684 	case NETDEV_UNREGISTER:
1685 		err = mpls_ifdown(dev, event);
1686 		if (err)
1687 			goto err;
1688 
1689 		mdev = mpls_dev_get(net, dev);
1690 		if (mdev) {
1691 			mpls_dev_sysctl_unregister(dev, mdev);
1692 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1693 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1694 		}
1695 		break;
1696 	case NETDEV_CHANGENAME:
1697 		mdev = mpls_dev_get(net, dev);
1698 		if (mdev) {
1699 			mpls_dev_sysctl_unregister(dev, mdev);
1700 			err = mpls_dev_sysctl_register(dev, mdev);
1701 			if (err)
1702 				goto err;
1703 		}
1704 		break;
1705 	}
1706 
1707 out:
1708 	mutex_unlock(&net->mpls.platform_mutex);
1709 	return NOTIFY_OK;
1710 
1711 err:
1712 	mutex_unlock(&net->mpls.platform_mutex);
1713 	return notifier_from_errno(err);
1714 }
1715 
1716 static struct notifier_block mpls_dev_notifier = {
1717 	.notifier_call = mpls_dev_notify,
1718 };
1719 
1720 static int nla_put_via(struct sk_buff *skb,
1721 		       u8 table, const void *addr, int alen)
1722 {
1723 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1724 		AF_INET, AF_INET6, AF_PACKET,
1725 	};
1726 	struct nlattr *nla;
1727 	struct rtvia *via;
1728 	int family = AF_UNSPEC;
1729 
1730 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1731 	if (!nla)
1732 		return -EMSGSIZE;
1733 
1734 	if (table <= NEIGH_NR_TABLES)
1735 		family = table_to_family[table];
1736 
1737 	via = nla_data(nla);
1738 	via->rtvia_family = family;
1739 	memcpy(via->rtvia_addr, addr, alen);
1740 	return 0;
1741 }
1742 
1743 int nla_put_labels(struct sk_buff *skb, int attrtype,
1744 		   u8 labels, const u32 label[])
1745 {
1746 	struct nlattr *nla;
1747 	struct mpls_shim_hdr *nla_label;
1748 	bool bos;
1749 	int i;
1750 	nla = nla_reserve(skb, attrtype, labels*4);
1751 	if (!nla)
1752 		return -EMSGSIZE;
1753 
1754 	nla_label = nla_data(nla);
1755 	bos = true;
1756 	for (i = labels - 1; i >= 0; i--) {
1757 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1758 		bos = false;
1759 	}
1760 
1761 	return 0;
1762 }
1763 EXPORT_SYMBOL_GPL(nla_put_labels);
1764 
1765 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1766 		   u32 label[], struct netlink_ext_ack *extack)
1767 {
1768 	unsigned len = nla_len(nla);
1769 	struct mpls_shim_hdr *nla_label;
1770 	u8 nla_labels;
1771 	bool bos;
1772 	int i;
1773 
1774 	/* len needs to be an even multiple of 4 (the label size). Number
1775 	 * of labels is a u8 so check for overflow.
1776 	 */
1777 	if (len & 3 || len / 4 > 255) {
1778 		NL_SET_ERR_MSG_ATTR(extack, nla,
1779 				    "Invalid length for labels attribute");
1780 		return -EINVAL;
1781 	}
1782 
1783 	/* Limit the number of new labels allowed */
1784 	nla_labels = len/4;
1785 	if (nla_labels > max_labels) {
1786 		NL_SET_ERR_MSG(extack, "Too many labels");
1787 		return -EINVAL;
1788 	}
1789 
1790 	/* when label == NULL, caller wants number of labels */
1791 	if (!label)
1792 		goto out;
1793 
1794 	nla_label = nla_data(nla);
1795 	bos = true;
1796 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1797 		struct mpls_entry_decoded dec;
1798 		dec = mpls_entry_decode(nla_label + i);
1799 
1800 		/* Ensure the bottom of stack flag is properly set
1801 		 * and ttl and tc are both clear.
1802 		 */
1803 		if (dec.ttl) {
1804 			NL_SET_ERR_MSG_ATTR(extack, nla,
1805 					    "TTL in label must be 0");
1806 			return -EINVAL;
1807 		}
1808 
1809 		if (dec.tc) {
1810 			NL_SET_ERR_MSG_ATTR(extack, nla,
1811 					    "Traffic class in label must be 0");
1812 			return -EINVAL;
1813 		}
1814 
1815 		if (dec.bos != bos) {
1816 			NL_SET_BAD_ATTR(extack, nla);
1817 			if (bos) {
1818 				NL_SET_ERR_MSG(extack,
1819 					       "BOS bit must be set in first label");
1820 			} else {
1821 				NL_SET_ERR_MSG(extack,
1822 					       "BOS bit can only be set in first label");
1823 			}
1824 			return -EINVAL;
1825 		}
1826 
1827 		switch (dec.label) {
1828 		case MPLS_LABEL_IMPLNULL:
1829 			/* RFC3032: This is a label that an LSR may
1830 			 * assign and distribute, but which never
1831 			 * actually appears in the encapsulation.
1832 			 */
1833 			NL_SET_ERR_MSG_ATTR(extack, nla,
1834 					    "Implicit NULL Label (3) can not be used in encapsulation");
1835 			return -EINVAL;
1836 		}
1837 
1838 		label[i] = dec.label;
1839 	}
1840 out:
1841 	*labels = nla_labels;
1842 	return 0;
1843 }
1844 EXPORT_SYMBOL_GPL(nla_get_labels);
1845 
1846 static int rtm_to_route_config(struct sk_buff *skb,
1847 			       struct nlmsghdr *nlh,
1848 			       struct mpls_route_config *cfg,
1849 			       struct netlink_ext_ack *extack)
1850 {
1851 	struct rtmsg *rtm;
1852 	struct nlattr *tb[RTA_MAX+1];
1853 	int index;
1854 	int err;
1855 
1856 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1857 				     rtm_mpls_policy, extack);
1858 	if (err < 0)
1859 		goto errout;
1860 
1861 	err = -EINVAL;
1862 	rtm = nlmsg_data(nlh);
1863 
1864 	if (rtm->rtm_family != AF_MPLS) {
1865 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1866 		goto errout;
1867 	}
1868 	if (rtm->rtm_dst_len != 20) {
1869 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1870 		goto errout;
1871 	}
1872 	if (rtm->rtm_src_len != 0) {
1873 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1874 		goto errout;
1875 	}
1876 	if (rtm->rtm_tos != 0) {
1877 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1878 		goto errout;
1879 	}
1880 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1881 		NL_SET_ERR_MSG(extack,
1882 			       "MPLS only supports the main route table");
1883 		goto errout;
1884 	}
1885 	/* Any value is acceptable for rtm_protocol */
1886 
1887 	/* As mpls uses destination specific addresses
1888 	 * (or source specific address in the case of multicast)
1889 	 * all addresses have universal scope.
1890 	 */
1891 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1892 		NL_SET_ERR_MSG(extack,
1893 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1894 		goto errout;
1895 	}
1896 	if (rtm->rtm_type != RTN_UNICAST) {
1897 		NL_SET_ERR_MSG(extack,
1898 			       "Invalid route type - MPLS only supports UNICAST");
1899 		goto errout;
1900 	}
1901 	if (rtm->rtm_flags != 0) {
1902 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1903 		goto errout;
1904 	}
1905 
1906 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1907 	cfg->rc_protocol	= rtm->rtm_protocol;
1908 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1909 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1910 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1911 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1912 	cfg->rc_nlinfo.nlh	= nlh;
1913 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1914 
1915 	for (index = 0; index <= RTA_MAX; index++) {
1916 		struct nlattr *nla = tb[index];
1917 		if (!nla)
1918 			continue;
1919 
1920 		switch (index) {
1921 		case RTA_OIF:
1922 			cfg->rc_ifindex = nla_get_u32(nla);
1923 			break;
1924 		case RTA_NEWDST:
1925 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1926 					   &cfg->rc_output_labels,
1927 					   cfg->rc_output_label, extack))
1928 				goto errout;
1929 			break;
1930 		case RTA_DST:
1931 		{
1932 			u8 label_count;
1933 			if (nla_get_labels(nla, 1, &label_count,
1934 					   &cfg->rc_label, extack))
1935 				goto errout;
1936 
1937 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1938 					   &cfg->rc_label, extack))
1939 				goto errout;
1940 			break;
1941 		}
1942 		case RTA_GATEWAY:
1943 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1944 			goto errout;
1945 		case RTA_VIA:
1946 		{
1947 			if (nla_get_via(nla, &cfg->rc_via_alen,
1948 					&cfg->rc_via_table, cfg->rc_via,
1949 					extack))
1950 				goto errout;
1951 			break;
1952 		}
1953 		case RTA_MULTIPATH:
1954 		{
1955 			cfg->rc_mp = nla_data(nla);
1956 			cfg->rc_mp_len = nla_len(nla);
1957 			break;
1958 		}
1959 		case RTA_TTL_PROPAGATE:
1960 		{
1961 			u8 ttl_propagate = nla_get_u8(nla);
1962 
1963 			if (ttl_propagate > 1) {
1964 				NL_SET_ERR_MSG_ATTR(extack, nla,
1965 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1966 				goto errout;
1967 			}
1968 			cfg->rc_ttl_propagate = ttl_propagate ?
1969 				MPLS_TTL_PROP_ENABLED :
1970 				MPLS_TTL_PROP_DISABLED;
1971 			break;
1972 		}
1973 		default:
1974 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1975 			/* Unsupported attribute */
1976 			goto errout;
1977 		}
1978 	}
1979 
1980 	err = 0;
1981 errout:
1982 	return err;
1983 }
1984 
1985 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1986 			     struct netlink_ext_ack *extack)
1987 {
1988 	struct net *net = sock_net(skb->sk);
1989 	struct mpls_route_config *cfg;
1990 	int err;
1991 
1992 	cfg = kzalloc_obj(*cfg);
1993 	if (!cfg)
1994 		return -ENOMEM;
1995 
1996 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1997 	if (err < 0)
1998 		goto out;
1999 
2000 	mutex_lock(&net->mpls.platform_mutex);
2001 	err = mpls_route_del(cfg, extack);
2002 	mutex_unlock(&net->mpls.platform_mutex);
2003 out:
2004 	kfree(cfg);
2005 
2006 	return err;
2007 }
2008 
2009 
2010 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
2011 			     struct netlink_ext_ack *extack)
2012 {
2013 	struct net *net = sock_net(skb->sk);
2014 	struct mpls_route_config *cfg;
2015 	int err;
2016 
2017 	cfg = kzalloc_obj(*cfg);
2018 	if (!cfg)
2019 		return -ENOMEM;
2020 
2021 	err = rtm_to_route_config(skb, nlh, cfg, extack);
2022 	if (err < 0)
2023 		goto out;
2024 
2025 	mutex_lock(&net->mpls.platform_mutex);
2026 	err = mpls_route_add(cfg, extack);
2027 	mutex_unlock(&net->mpls.platform_mutex);
2028 out:
2029 	kfree(cfg);
2030 
2031 	return err;
2032 }
2033 
2034 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
2035 			   u32 label, struct mpls_route *rt, int flags)
2036 {
2037 	struct net_device *dev;
2038 	struct nlmsghdr *nlh;
2039 	struct rtmsg *rtm;
2040 
2041 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
2042 	if (nlh == NULL)
2043 		return -EMSGSIZE;
2044 
2045 	rtm = nlmsg_data(nlh);
2046 	rtm->rtm_family = AF_MPLS;
2047 	rtm->rtm_dst_len = 20;
2048 	rtm->rtm_src_len = 0;
2049 	rtm->rtm_tos = 0;
2050 	rtm->rtm_table = RT_TABLE_MAIN;
2051 	rtm->rtm_protocol = rt->rt_protocol;
2052 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
2053 	rtm->rtm_type = RTN_UNICAST;
2054 	rtm->rtm_flags = 0;
2055 
2056 	if (nla_put_labels(skb, RTA_DST, 1, &label))
2057 		goto nla_put_failure;
2058 
2059 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
2060 		bool ttl_propagate =
2061 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2062 
2063 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2064 			       ttl_propagate))
2065 			goto nla_put_failure;
2066 	}
2067 	if (rt->rt_nhn == 1) {
2068 		const struct mpls_nh *nh = rt->rt_nh;
2069 
2070 		if (nh->nh_labels &&
2071 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2072 				   nh->nh_label))
2073 			goto nla_put_failure;
2074 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2075 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2076 				nh->nh_via_alen))
2077 			goto nla_put_failure;
2078 		dev = nh->nh_dev;
2079 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2080 			goto nla_put_failure;
2081 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2082 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2083 		if (nh->nh_flags & RTNH_F_DEAD)
2084 			rtm->rtm_flags |= RTNH_F_DEAD;
2085 	} else {
2086 		struct rtnexthop *rtnh;
2087 		struct nlattr *mp;
2088 		u8 linkdown = 0;
2089 		u8 dead = 0;
2090 
2091 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2092 		if (!mp)
2093 			goto nla_put_failure;
2094 
2095 		for_nexthops(rt) {
2096 			dev = nh->nh_dev;
2097 			if (!dev)
2098 				continue;
2099 
2100 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2101 			if (!rtnh)
2102 				goto nla_put_failure;
2103 
2104 			rtnh->rtnh_ifindex = dev->ifindex;
2105 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2106 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2107 				linkdown++;
2108 			}
2109 			if (nh->nh_flags & RTNH_F_DEAD) {
2110 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2111 				dead++;
2112 			}
2113 
2114 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2115 							    nh->nh_labels,
2116 							    nh->nh_label))
2117 				goto nla_put_failure;
2118 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2119 			    nla_put_via(skb, nh->nh_via_table,
2120 					mpls_nh_via(rt, nh),
2121 					nh->nh_via_alen))
2122 				goto nla_put_failure;
2123 
2124 			/* length of rtnetlink header + attributes */
2125 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2126 		} endfor_nexthops(rt);
2127 
2128 		if (linkdown == rt->rt_nhn)
2129 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2130 		if (dead == rt->rt_nhn)
2131 			rtm->rtm_flags |= RTNH_F_DEAD;
2132 
2133 		nla_nest_end(skb, mp);
2134 	}
2135 
2136 	nlmsg_end(skb, nlh);
2137 	return 0;
2138 
2139 nla_put_failure:
2140 	nlmsg_cancel(skb, nlh);
2141 	return -EMSGSIZE;
2142 }
2143 
2144 #if IS_ENABLED(CONFIG_INET)
2145 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2146 				   struct fib_dump_filter *filter,
2147 				   struct netlink_callback *cb)
2148 {
2149 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2150 }
2151 #else
2152 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2153 				   struct fib_dump_filter *filter,
2154 				   struct netlink_callback *cb)
2155 {
2156 	struct netlink_ext_ack *extack = cb->extack;
2157 	struct nlattr *tb[RTA_MAX + 1];
2158 	struct rtmsg *rtm;
2159 	int err, i;
2160 
2161 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
2162 	if (!rtm) {
2163 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2164 		return -EINVAL;
2165 	}
2166 
2167 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2168 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2169 	    rtm->rtm_flags) {
2170 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2171 		return -EINVAL;
2172 	}
2173 
2174 	if (rtm->rtm_protocol) {
2175 		filter->protocol = rtm->rtm_protocol;
2176 		filter->filter_set = 1;
2177 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2178 	}
2179 
2180 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2181 					    rtm_mpls_policy, extack);
2182 	if (err < 0)
2183 		return err;
2184 
2185 	for (i = 0; i <= RTA_MAX; ++i) {
2186 		int ifindex;
2187 
2188 		if (i == RTA_OIF) {
2189 			ifindex = nla_get_u32(tb[i]);
2190 			filter->dev = dev_get_by_index_rcu(net, ifindex);
2191 			if (!filter->dev)
2192 				return -ENODEV;
2193 			filter->filter_set = 1;
2194 		} else if (tb[i]) {
2195 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2196 			return -EINVAL;
2197 		}
2198 	}
2199 
2200 	return 0;
2201 }
2202 #endif
2203 
2204 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2205 			     const struct net_device *dev)
2206 {
2207 	if (rt->rt_nhn == 1) {
2208 		struct mpls_nh *nh = rt->rt_nh;
2209 
2210 		if (nh->nh_dev == dev)
2211 			return true;
2212 	} else {
2213 		for_nexthops(rt) {
2214 			if (nh->nh_dev == dev)
2215 				return true;
2216 		} endfor_nexthops(rt);
2217 	}
2218 
2219 	return false;
2220 }
2221 
2222 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2223 {
2224 	const struct nlmsghdr *nlh = cb->nlh;
2225 	struct net *net = sock_net(skb->sk);
2226 	struct mpls_route __rcu **platform_label;
2227 	struct fib_dump_filter filter = {
2228 		.rtnl_held = false,
2229 	};
2230 	unsigned int flags = NLM_F_MULTI;
2231 	size_t platform_labels;
2232 	unsigned int index;
2233 	int err;
2234 
2235 	rcu_read_lock();
2236 
2237 	if (cb->strict_check) {
2238 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2239 		if (err < 0)
2240 			goto err;
2241 
2242 		/* for MPLS, there is only 1 table with fixed type and flags.
2243 		 * If either are set in the filter then return nothing.
2244 		 */
2245 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2246 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2247 		     filter.flags)
2248 			goto unlock;
2249 	}
2250 
2251 	index = cb->args[0];
2252 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2253 		index = MPLS_LABEL_FIRST_UNRESERVED;
2254 
2255 	platform_label = mpls_platform_label_rcu(net, &platform_labels);
2256 
2257 	if (filter.filter_set)
2258 		flags |= NLM_F_DUMP_FILTERED;
2259 
2260 	for (; index < platform_labels; index++) {
2261 		struct mpls_route *rt;
2262 
2263 		rt = rcu_dereference(platform_label[index]);
2264 		if (!rt)
2265 			continue;
2266 
2267 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2268 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2269 			continue;
2270 
2271 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2272 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2273 				    index, rt, flags) < 0)
2274 			break;
2275 	}
2276 	cb->args[0] = index;
2277 
2278 unlock:
2279 	rcu_read_unlock();
2280 	return skb->len;
2281 
2282 err:
2283 	rcu_read_unlock();
2284 	return err;
2285 }
2286 
2287 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2288 {
2289 	size_t payload =
2290 		NLMSG_ALIGN(sizeof(struct rtmsg))
2291 		+ nla_total_size(4)			/* RTA_DST */
2292 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2293 
2294 	if (rt->rt_nhn == 1) {
2295 		struct mpls_nh *nh = rt->rt_nh;
2296 
2297 		if (nh->nh_dev)
2298 			payload += nla_total_size(4); /* RTA_OIF */
2299 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2300 			payload += nla_total_size(2 + nh->nh_via_alen);
2301 		if (nh->nh_labels) /* RTA_NEWDST */
2302 			payload += nla_total_size(nh->nh_labels * 4);
2303 	} else {
2304 		/* each nexthop is packed in an attribute */
2305 		size_t nhsize = 0;
2306 
2307 		for_nexthops(rt) {
2308 			if (!nh->nh_dev)
2309 				continue;
2310 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2311 			/* RTA_VIA */
2312 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2313 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2314 			if (nh->nh_labels)
2315 				nhsize += nla_total_size(nh->nh_labels * 4);
2316 		} endfor_nexthops(rt);
2317 		/* nested attribute */
2318 		payload += nla_total_size(nhsize);
2319 	}
2320 
2321 	return payload;
2322 }
2323 
2324 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2325 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2326 		       unsigned int nlm_flags)
2327 {
2328 	struct sk_buff *skb;
2329 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2330 	int err = -ENOBUFS;
2331 
2332 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2333 	if (skb == NULL)
2334 		goto errout;
2335 
2336 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2337 	if (err < 0) {
2338 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2339 		WARN_ON(err == -EMSGSIZE);
2340 		kfree_skb(skb);
2341 		goto errout;
2342 	}
2343 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2344 
2345 	return;
2346 errout:
2347 	rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2348 }
2349 
2350 static int mpls_valid_getroute_req(struct sk_buff *skb,
2351 				   const struct nlmsghdr *nlh,
2352 				   struct nlattr **tb,
2353 				   struct netlink_ext_ack *extack)
2354 {
2355 	struct rtmsg *rtm;
2356 	int i, err;
2357 
2358 	rtm = nlmsg_payload(nlh, sizeof(*rtm));
2359 	if (!rtm) {
2360 		NL_SET_ERR_MSG_MOD(extack,
2361 				   "Invalid header for get route request");
2362 		return -EINVAL;
2363 	}
2364 
2365 	if (!netlink_strict_get_check(skb))
2366 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2367 					      rtm_mpls_policy, extack);
2368 
2369 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2370 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2371 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2372 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2373 		return -EINVAL;
2374 	}
2375 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2376 		NL_SET_ERR_MSG_MOD(extack,
2377 				   "Invalid flags for get route request");
2378 		return -EINVAL;
2379 	}
2380 
2381 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2382 					    rtm_mpls_policy, extack);
2383 	if (err)
2384 		return err;
2385 
2386 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2387 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2388 		return -EINVAL;
2389 	}
2390 
2391 	for (i = 0; i <= RTA_MAX; i++) {
2392 		if (!tb[i])
2393 			continue;
2394 
2395 		switch (i) {
2396 		case RTA_DST:
2397 		case RTA_NEWDST:
2398 			break;
2399 		default:
2400 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2401 			return -EINVAL;
2402 		}
2403 	}
2404 
2405 	return 0;
2406 }
2407 
2408 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2409 			 struct netlink_ext_ack *extack)
2410 {
2411 	struct net *net = sock_net(in_skb->sk);
2412 	u32 portid = NETLINK_CB(in_skb).portid;
2413 	u32 in_label = LABEL_NOT_SPECIFIED;
2414 	struct nlattr *tb[RTA_MAX + 1];
2415 	struct mpls_route *rt = NULL;
2416 	u32 labels[MAX_NEW_LABELS];
2417 	struct mpls_shim_hdr *hdr;
2418 	unsigned int hdr_size = 0;
2419 	const struct mpls_nh *nh;
2420 	struct net_device *dev;
2421 	struct rtmsg *rtm, *r;
2422 	struct nlmsghdr *nlh;
2423 	struct sk_buff *skb;
2424 	u8 n_labels;
2425 	int err;
2426 
2427 	mutex_lock(&net->mpls.platform_mutex);
2428 
2429 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2430 	if (err < 0)
2431 		goto errout;
2432 
2433 	rtm = nlmsg_data(in_nlh);
2434 
2435 	if (tb[RTA_DST]) {
2436 		u8 label_count;
2437 
2438 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2439 				   &in_label, extack)) {
2440 			err = -EINVAL;
2441 			goto errout;
2442 		}
2443 
2444 		if (!mpls_label_ok(net, &in_label, extack)) {
2445 			err = -EINVAL;
2446 			goto errout;
2447 		}
2448 	}
2449 
2450 	if (in_label < net->mpls.platform_labels)
2451 		rt = mpls_route_input(net, in_label);
2452 	if (!rt) {
2453 		err = -ENETUNREACH;
2454 		goto errout;
2455 	}
2456 
2457 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2458 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2459 		if (!skb) {
2460 			err = -ENOBUFS;
2461 			goto errout;
2462 		}
2463 
2464 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2465 				      RTM_NEWROUTE, in_label, rt, 0);
2466 		if (err < 0) {
2467 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2468 			WARN_ON(err == -EMSGSIZE);
2469 			goto errout_free;
2470 		}
2471 
2472 		err = rtnl_unicast(skb, net, portid);
2473 		goto errout;
2474 	}
2475 
2476 	if (tb[RTA_NEWDST]) {
2477 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2478 				   labels, extack) != 0) {
2479 			err = -EINVAL;
2480 			goto errout;
2481 		}
2482 
2483 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2484 	}
2485 
2486 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2487 	if (!skb) {
2488 		err = -ENOBUFS;
2489 		goto errout;
2490 	}
2491 
2492 	skb->protocol = htons(ETH_P_MPLS_UC);
2493 
2494 	if (hdr_size) {
2495 		bool bos;
2496 		int i;
2497 
2498 		if (skb_cow(skb, hdr_size)) {
2499 			err = -ENOBUFS;
2500 			goto errout_free;
2501 		}
2502 
2503 		skb_reserve(skb, hdr_size);
2504 		skb_push(skb, hdr_size);
2505 		skb_reset_network_header(skb);
2506 
2507 		/* Push new labels */
2508 		hdr = mpls_hdr(skb);
2509 		bos = true;
2510 		for (i = n_labels - 1; i >= 0; i--) {
2511 			hdr[i] = mpls_entry_encode(labels[i],
2512 						   1, 0, bos);
2513 			bos = false;
2514 		}
2515 	}
2516 
2517 	nh = mpls_select_multipath(rt, skb);
2518 	if (!nh) {
2519 		err = -ENETUNREACH;
2520 		goto errout_free;
2521 	}
2522 
2523 	if (hdr_size) {
2524 		skb_pull(skb, hdr_size);
2525 		skb_reset_network_header(skb);
2526 	}
2527 
2528 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2529 			RTM_NEWROUTE, sizeof(*r), 0);
2530 	if (!nlh) {
2531 		err = -EMSGSIZE;
2532 		goto errout_free;
2533 	}
2534 
2535 	r = nlmsg_data(nlh);
2536 	r->rtm_family	 = AF_MPLS;
2537 	r->rtm_dst_len	= 20;
2538 	r->rtm_src_len	= 0;
2539 	r->rtm_table	= RT_TABLE_MAIN;
2540 	r->rtm_type	= RTN_UNICAST;
2541 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2542 	r->rtm_protocol = rt->rt_protocol;
2543 	r->rtm_flags	= 0;
2544 
2545 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2546 		goto nla_put_failure;
2547 
2548 	if (nh->nh_labels &&
2549 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2550 			   nh->nh_label))
2551 		goto nla_put_failure;
2552 
2553 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2554 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2555 			nh->nh_via_alen))
2556 		goto nla_put_failure;
2557 	dev = nh->nh_dev;
2558 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2559 		goto nla_put_failure;
2560 
2561 	nlmsg_end(skb, nlh);
2562 
2563 	err = rtnl_unicast(skb, net, portid);
2564 errout:
2565 	mutex_unlock(&net->mpls.platform_mutex);
2566 	return err;
2567 
2568 nla_put_failure:
2569 	nlmsg_cancel(skb, nlh);
2570 	err = -EMSGSIZE;
2571 errout_free:
2572 	mutex_unlock(&net->mpls.platform_mutex);
2573 	kfree_skb(skb);
2574 	return err;
2575 }
2576 
2577 static int resize_platform_label_table(struct net *net, size_t limit)
2578 {
2579 	size_t size = sizeof(struct mpls_route *) * limit;
2580 	size_t old_limit;
2581 	size_t cp_size;
2582 	struct mpls_route __rcu **labels = NULL, **old;
2583 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2584 	unsigned index;
2585 
2586 	if (size) {
2587 		labels = kvzalloc(size, GFP_KERNEL);
2588 		if (!labels)
2589 			goto nolabels;
2590 	}
2591 
2592 	/* In case the predefined labels need to be populated */
2593 	if (limit > MPLS_LABEL_IPV4NULL) {
2594 		struct net_device *lo = net->loopback_dev;
2595 
2596 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2597 		if (IS_ERR(rt0))
2598 			goto nort0;
2599 
2600 		rt0->rt_nh->nh_dev = lo;
2601 		netdev_hold(lo, &rt0->rt_nh->nh_dev_tracker, GFP_KERNEL);
2602 		rt0->rt_protocol = RTPROT_KERNEL;
2603 		rt0->rt_payload_type = MPT_IPV4;
2604 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2605 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2606 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2607 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2608 		       lo->addr_len);
2609 	}
2610 	if (limit > MPLS_LABEL_IPV6NULL) {
2611 		struct net_device *lo = net->loopback_dev;
2612 
2613 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2614 		if (IS_ERR(rt2))
2615 			goto nort2;
2616 
2617 		rt2->rt_nh->nh_dev = lo;
2618 		netdev_hold(lo, &rt2->rt_nh->nh_dev_tracker, GFP_KERNEL);
2619 		rt2->rt_protocol = RTPROT_KERNEL;
2620 		rt2->rt_payload_type = MPT_IPV6;
2621 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2622 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2623 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2624 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2625 		       lo->addr_len);
2626 	}
2627 
2628 	mutex_lock(&net->mpls.platform_mutex);
2629 
2630 	/* Remember the original table */
2631 	old = mpls_dereference(net, net->mpls.platform_label);
2632 	old_limit = net->mpls.platform_labels;
2633 
2634 	/* Free any labels beyond the new table */
2635 	for (index = limit; index < old_limit; index++)
2636 		mpls_route_update(net, index, NULL, NULL);
2637 
2638 	/* Copy over the old labels */
2639 	cp_size = size;
2640 	if (old_limit < limit)
2641 		cp_size = old_limit * sizeof(struct mpls_route *);
2642 
2643 	memcpy(labels, old, cp_size);
2644 
2645 	/* If needed set the predefined labels */
2646 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2647 	    (limit > MPLS_LABEL_IPV6NULL)) {
2648 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2649 		rt2 = NULL;
2650 	}
2651 
2652 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2653 	    (limit > MPLS_LABEL_IPV4NULL)) {
2654 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2655 		rt0 = NULL;
2656 	}
2657 
2658 	/* Update the global pointers */
2659 	local_bh_disable();
2660 	write_seqcount_begin(&net->mpls.platform_label_seq);
2661 	net->mpls.platform_labels = limit;
2662 	rcu_assign_pointer(net->mpls.platform_label, labels);
2663 	write_seqcount_end(&net->mpls.platform_label_seq);
2664 	local_bh_enable();
2665 
2666 	mutex_unlock(&net->mpls.platform_mutex);
2667 
2668 	mpls_rt_free(rt2);
2669 	mpls_rt_free(rt0);
2670 
2671 	if (old) {
2672 		synchronize_rcu();
2673 		kvfree(old);
2674 	}
2675 	return 0;
2676 
2677 nort2:
2678 	mpls_rt_free(rt0);
2679 nort0:
2680 	kvfree(labels);
2681 nolabels:
2682 	return -ENOMEM;
2683 }
2684 
2685 static int mpls_platform_labels(const struct ctl_table *table, int write,
2686 				void *buffer, size_t *lenp, loff_t *ppos)
2687 {
2688 	struct net *net = table->data;
2689 	int platform_labels = net->mpls.platform_labels;
2690 	int ret;
2691 	struct ctl_table tmp = {
2692 		.procname	= table->procname,
2693 		.data		= &platform_labels,
2694 		.maxlen		= sizeof(int),
2695 		.mode		= table->mode,
2696 		.extra1		= SYSCTL_ZERO,
2697 		.extra2		= &label_limit,
2698 	};
2699 
2700 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2701 
2702 	if (write && ret == 0)
2703 		ret = resize_platform_label_table(net, platform_labels);
2704 
2705 	return ret;
2706 }
2707 
2708 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2709 	(&((struct net *)0)->field)
2710 
2711 static const struct ctl_table mpls_table[] = {
2712 	{
2713 		.procname	= "platform_labels",
2714 		.data		= NULL,
2715 		.maxlen		= sizeof(int),
2716 		.mode		= 0644,
2717 		.proc_handler	= mpls_platform_labels,
2718 	},
2719 	{
2720 		.procname	= "ip_ttl_propagate",
2721 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2722 		.maxlen		= sizeof(int),
2723 		.mode		= 0644,
2724 		.proc_handler	= proc_dointvec_minmax,
2725 		.extra1		= SYSCTL_ZERO,
2726 		.extra2		= SYSCTL_ONE,
2727 	},
2728 	{
2729 		.procname	= "default_ttl",
2730 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2731 		.maxlen		= sizeof(int),
2732 		.mode		= 0644,
2733 		.proc_handler	= proc_dointvec_minmax,
2734 		.extra1		= SYSCTL_ONE,
2735 		.extra2		= &ttl_max,
2736 	},
2737 };
2738 
2739 static __net_init int mpls_net_init(struct net *net)
2740 {
2741 	size_t table_size = ARRAY_SIZE(mpls_table);
2742 	struct ctl_table *table;
2743 	int i;
2744 
2745 	mutex_init(&net->mpls.platform_mutex);
2746 	seqcount_mutex_init(&net->mpls.platform_label_seq, &net->mpls.platform_mutex);
2747 
2748 	net->mpls.platform_labels = 0;
2749 	net->mpls.platform_label = NULL;
2750 	net->mpls.ip_ttl_propagate = 1;
2751 	net->mpls.default_ttl = 255;
2752 
2753 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2754 	if (table == NULL)
2755 		return -ENOMEM;
2756 
2757 	/* Table data contains only offsets relative to the base of
2758 	 * the mdev at this point, so make them absolute.
2759 	 */
2760 	for (i = 0; i < table_size; i++)
2761 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2762 
2763 	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2764 					       table_size);
2765 	if (net->mpls.ctl == NULL) {
2766 		kfree(table);
2767 		return -ENOMEM;
2768 	}
2769 
2770 	return 0;
2771 }
2772 
2773 static __net_exit void mpls_net_exit(struct net *net)
2774 {
2775 	struct mpls_route __rcu **platform_label;
2776 	size_t platform_labels;
2777 	const struct ctl_table *table;
2778 	unsigned int index;
2779 
2780 	table = net->mpls.ctl->ctl_table_arg;
2781 	unregister_net_sysctl_table(net->mpls.ctl);
2782 	kfree(table);
2783 
2784 	/* An rcu grace period has passed since there was a device in
2785 	 * the network namespace (and thus the last in flight packet)
2786 	 * left this network namespace.  This is because
2787 	 * unregister_netdevice_many and netdev_run_todo has completed
2788 	 * for each network device that was in this network namespace.
2789 	 *
2790 	 * As such no additional rcu synchronization is necessary when
2791 	 * freeing the platform_label table.
2792 	 */
2793 	mutex_lock(&net->mpls.platform_mutex);
2794 
2795 	platform_label = mpls_dereference(net, net->mpls.platform_label);
2796 	platform_labels = net->mpls.platform_labels;
2797 
2798 	for (index = 0; index < platform_labels; index++) {
2799 		struct mpls_route *rt;
2800 
2801 		rt = mpls_dereference(net, platform_label[index]);
2802 		mpls_notify_route(net, index, rt, NULL, NULL);
2803 		mpls_rt_free(rt);
2804 	}
2805 
2806 	mutex_unlock(&net->mpls.platform_mutex);
2807 
2808 	kvfree(platform_label);
2809 }
2810 
2811 static struct pernet_operations mpls_net_ops = {
2812 	.init = mpls_net_init,
2813 	.exit = mpls_net_exit,
2814 };
2815 
2816 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2817 	.family		   = AF_MPLS,
2818 	.fill_stats_af	   = mpls_fill_stats_af,
2819 	.get_stats_af_size = mpls_get_stats_af_size,
2820 };
2821 
2822 static const struct rtnl_msg_handler mpls_rtnl_msg_handlers[] __initdata_or_module = {
2823 	{THIS_MODULE, PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL,
2824 	 RTNL_FLAG_DOIT_UNLOCKED},
2825 	{THIS_MODULE, PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL,
2826 	 RTNL_FLAG_DOIT_UNLOCKED},
2827 	{THIS_MODULE, PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes,
2828 	 RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
2829 	{THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2830 	 mpls_netconf_get_devconf, mpls_netconf_dump_devconf,
2831 	 RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
2832 };
2833 
2834 static int __init mpls_init(void)
2835 {
2836 	int err;
2837 
2838 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2839 
2840 	err = register_pernet_subsys(&mpls_net_ops);
2841 	if (err)
2842 		goto out;
2843 
2844 	err = register_netdevice_notifier(&mpls_dev_notifier);
2845 	if (err)
2846 		goto out_unregister_pernet;
2847 
2848 	dev_add_pack(&mpls_packet_type);
2849 
2850 	err = rtnl_af_register(&mpls_af_ops);
2851 	if (err)
2852 		goto out_unregister_dev_type;
2853 
2854 	err = rtnl_register_many(mpls_rtnl_msg_handlers);
2855 	if (err)
2856 		goto out_unregister_rtnl_af;
2857 
2858 	err = ipgre_tunnel_encap_add_mpls_ops();
2859 	if (err) {
2860 		pr_err("Can't add mpls over gre tunnel ops\n");
2861 		goto out_unregister_rtnl;
2862 	}
2863 
2864 	err = 0;
2865 out:
2866 	return err;
2867 
2868 out_unregister_rtnl:
2869 	rtnl_unregister_many(mpls_rtnl_msg_handlers);
2870 out_unregister_rtnl_af:
2871 	rtnl_af_unregister(&mpls_af_ops);
2872 out_unregister_dev_type:
2873 	dev_remove_pack(&mpls_packet_type);
2874 	unregister_netdevice_notifier(&mpls_dev_notifier);
2875 out_unregister_pernet:
2876 	unregister_pernet_subsys(&mpls_net_ops);
2877 	goto out;
2878 }
2879 module_init(mpls_init);
2880 
2881 static void __exit mpls_exit(void)
2882 {
2883 	rtnl_unregister_all(PF_MPLS);
2884 	rtnl_af_unregister(&mpls_af_ops);
2885 	dev_remove_pack(&mpls_packet_type);
2886 	unregister_netdevice_notifier(&mpls_dev_notifier);
2887 	unregister_pernet_subsys(&mpls_net_ops);
2888 	ipgre_tunnel_encap_del_mpls_ops();
2889 }
2890 module_exit(mpls_exit);
2891 
2892 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2893 MODULE_LICENSE("GPL v2");
2894 MODULE_ALIAS_NETPROTO(PF_MPLS);
2895