xref: /linux/net/mpls/af_mpls.c (revision b68fc09be48edbc47de1a0f3d42ef8adf6c0ac55)
1 #include <linux/types.h>
2 #include <linux/skbuff.h>
3 #include <linux/socket.h>
4 #include <linux/sysctl.h>
5 #include <linux/net.h>
6 #include <linux/module.h>
7 #include <linux/if_arp.h>
8 #include <linux/ipv6.h>
9 #include <linux/mpls.h>
10 #include <linux/netconf.h>
11 #include <linux/nospec.h>
12 #include <linux/vmalloc.h>
13 #include <linux/percpu.h>
14 #include <net/ip.h>
15 #include <net/dst.h>
16 #include <net/sock.h>
17 #include <net/arp.h>
18 #include <net/ip_fib.h>
19 #include <net/netevent.h>
20 #include <net/ip_tunnels.h>
21 #include <net/netns/generic.h>
22 #if IS_ENABLED(CONFIG_IPV6)
23 #include <net/ipv6.h>
24 #endif
25 #include <net/addrconf.h>
26 #include <net/nexthop.h>
27 #include "internal.h"
28 
29 /* max memory we will use for mpls_route */
30 #define MAX_MPLS_ROUTE_MEM	4096
31 
32 /* Maximum number of labels to look ahead at when selecting a path of
33  * a multipath route
34  */
35 #define MAX_MP_SELECT_LABELS 4
36 
37 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
38 
39 static int zero = 0;
40 static int one = 1;
41 static int label_limit = (1 << 20) - 1;
42 static int ttl_max = 255;
43 
44 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
45 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
46 {
47 	return sizeof(struct mpls_shim_hdr);
48 }
49 
50 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
51 	.encap_hlen	= ipgre_mpls_encap_hlen,
52 };
53 
54 static int ipgre_tunnel_encap_add_mpls_ops(void)
55 {
56 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
57 }
58 
59 static void ipgre_tunnel_encap_del_mpls_ops(void)
60 {
61 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
62 }
63 #else
64 static int ipgre_tunnel_encap_add_mpls_ops(void)
65 {
66 	return 0;
67 }
68 
69 static void ipgre_tunnel_encap_del_mpls_ops(void)
70 {
71 }
72 #endif
73 
74 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
75 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
76 		       unsigned int nlm_flags);
77 
78 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
79 {
80 	struct mpls_route *rt = NULL;
81 
82 	if (index < net->mpls.platform_labels) {
83 		struct mpls_route __rcu **platform_label =
84 			rcu_dereference(net->mpls.platform_label);
85 		rt = rcu_dereference(platform_label[index]);
86 	}
87 	return rt;
88 }
89 
90 bool mpls_output_possible(const struct net_device *dev)
91 {
92 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
93 }
94 EXPORT_SYMBOL_GPL(mpls_output_possible);
95 
96 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
97 {
98 	return (u8 *)nh + rt->rt_via_offset;
99 }
100 
101 static const u8 *mpls_nh_via(const struct mpls_route *rt,
102 			     const struct mpls_nh *nh)
103 {
104 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
105 }
106 
107 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
108 {
109 	/* The size of the layer 2.5 labels to be added for this route */
110 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
111 }
112 
113 unsigned int mpls_dev_mtu(const struct net_device *dev)
114 {
115 	/* The amount of data the layer 2 frame can hold */
116 	return dev->mtu;
117 }
118 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
119 
120 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
121 {
122 	if (skb->len <= mtu)
123 		return false;
124 
125 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
126 		return false;
127 
128 	return true;
129 }
130 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
131 
132 void mpls_stats_inc_outucastpkts(struct net_device *dev,
133 				 const struct sk_buff *skb)
134 {
135 	struct mpls_dev *mdev;
136 
137 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
138 		mdev = mpls_dev_get(dev);
139 		if (mdev)
140 			MPLS_INC_STATS_LEN(mdev, skb->len,
141 					   tx_packets,
142 					   tx_bytes);
143 	} else if (skb->protocol == htons(ETH_P_IP)) {
144 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
145 #if IS_ENABLED(CONFIG_IPV6)
146 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
147 		struct inet6_dev *in6dev = __in6_dev_get(dev);
148 
149 		if (in6dev)
150 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
151 					 IPSTATS_MIB_OUT, skb->len);
152 #endif
153 	}
154 }
155 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
156 
157 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
158 {
159 	struct mpls_entry_decoded dec;
160 	unsigned int mpls_hdr_len = 0;
161 	struct mpls_shim_hdr *hdr;
162 	bool eli_seen = false;
163 	int label_index;
164 	u32 hash = 0;
165 
166 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
167 	     label_index++) {
168 		mpls_hdr_len += sizeof(*hdr);
169 		if (!pskb_may_pull(skb, mpls_hdr_len))
170 			break;
171 
172 		/* Read and decode the current label */
173 		hdr = mpls_hdr(skb) + label_index;
174 		dec = mpls_entry_decode(hdr);
175 
176 		/* RFC6790 - reserved labels MUST NOT be used as keys
177 		 * for the load-balancing function
178 		 */
179 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
180 			hash = jhash_1word(dec.label, hash);
181 
182 			/* The entropy label follows the entropy label
183 			 * indicator, so this means that the entropy
184 			 * label was just added to the hash - no need to
185 			 * go any deeper either in the label stack or in the
186 			 * payload
187 			 */
188 			if (eli_seen)
189 				break;
190 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
191 			eli_seen = true;
192 		}
193 
194 		if (!dec.bos)
195 			continue;
196 
197 		/* found bottom label; does skb have room for a header? */
198 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
199 			const struct iphdr *v4hdr;
200 
201 			v4hdr = (const struct iphdr *)(hdr + 1);
202 			if (v4hdr->version == 4) {
203 				hash = jhash_3words(ntohl(v4hdr->saddr),
204 						    ntohl(v4hdr->daddr),
205 						    v4hdr->protocol, hash);
206 			} else if (v4hdr->version == 6 &&
207 				   pskb_may_pull(skb, mpls_hdr_len +
208 						 sizeof(struct ipv6hdr))) {
209 				const struct ipv6hdr *v6hdr;
210 
211 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
212 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
213 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
214 				hash = jhash_1word(v6hdr->nexthdr, hash);
215 			}
216 		}
217 
218 		break;
219 	}
220 
221 	return hash;
222 }
223 
224 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
225 {
226 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
227 }
228 
229 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
230  * a next hop (nh->nh_flags) are modified by netdev event handlers.
231  * Since those fields can change at any moment, use READ_ONCE to
232  * access both.
233  */
234 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
235 					     struct sk_buff *skb)
236 {
237 	u32 hash = 0;
238 	int nh_index = 0;
239 	int n = 0;
240 	u8 alive;
241 
242 	/* No need to look further into packet if there's only
243 	 * one path
244 	 */
245 	if (rt->rt_nhn == 1)
246 		return rt->rt_nh;
247 
248 	alive = READ_ONCE(rt->rt_nhn_alive);
249 	if (alive == 0)
250 		return NULL;
251 
252 	hash = mpls_multipath_hash(rt, skb);
253 	nh_index = hash % alive;
254 	if (alive == rt->rt_nhn)
255 		goto out;
256 	for_nexthops(rt) {
257 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
258 
259 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
260 			continue;
261 		if (n == nh_index)
262 			return nh;
263 		n++;
264 	} endfor_nexthops(rt);
265 
266 out:
267 	return mpls_get_nexthop(rt, nh_index);
268 }
269 
270 static bool mpls_egress(struct net *net, struct mpls_route *rt,
271 			struct sk_buff *skb, struct mpls_entry_decoded dec)
272 {
273 	enum mpls_payload_type payload_type;
274 	bool success = false;
275 
276 	/* The IPv4 code below accesses through the IPv4 header
277 	 * checksum, which is 12 bytes into the packet.
278 	 * The IPv6 code below accesses through the IPv6 hop limit
279 	 * which is 8 bytes into the packet.
280 	 *
281 	 * For all supported cases there should always be at least 12
282 	 * bytes of packet data present.  The IPv4 header is 20 bytes
283 	 * without options and the IPv6 header is always 40 bytes
284 	 * long.
285 	 */
286 	if (!pskb_may_pull(skb, 12))
287 		return false;
288 
289 	payload_type = rt->rt_payload_type;
290 	if (payload_type == MPT_UNSPEC)
291 		payload_type = ip_hdr(skb)->version;
292 
293 	switch (payload_type) {
294 	case MPT_IPV4: {
295 		struct iphdr *hdr4 = ip_hdr(skb);
296 		u8 new_ttl;
297 		skb->protocol = htons(ETH_P_IP);
298 
299 		/* If propagating TTL, take the decremented TTL from
300 		 * the incoming MPLS header, otherwise decrement the
301 		 * TTL, but only if not 0 to avoid underflow.
302 		 */
303 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
304 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
305 		     net->mpls.ip_ttl_propagate))
306 			new_ttl = dec.ttl;
307 		else
308 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
309 
310 		csum_replace2(&hdr4->check,
311 			      htons(hdr4->ttl << 8),
312 			      htons(new_ttl << 8));
313 		hdr4->ttl = new_ttl;
314 		success = true;
315 		break;
316 	}
317 	case MPT_IPV6: {
318 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
319 		skb->protocol = htons(ETH_P_IPV6);
320 
321 		/* If propagating TTL, take the decremented TTL from
322 		 * the incoming MPLS header, otherwise decrement the
323 		 * hop limit, but only if not 0 to avoid underflow.
324 		 */
325 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
326 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
327 		     net->mpls.ip_ttl_propagate))
328 			hdr6->hop_limit = dec.ttl;
329 		else if (hdr6->hop_limit)
330 			hdr6->hop_limit = hdr6->hop_limit - 1;
331 		success = true;
332 		break;
333 	}
334 	case MPT_UNSPEC:
335 		/* Should have decided which protocol it is by now */
336 		break;
337 	}
338 
339 	return success;
340 }
341 
342 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
343 			struct packet_type *pt, struct net_device *orig_dev)
344 {
345 	struct net *net = dev_net(dev);
346 	struct mpls_shim_hdr *hdr;
347 	struct mpls_route *rt;
348 	struct mpls_nh *nh;
349 	struct mpls_entry_decoded dec;
350 	struct net_device *out_dev;
351 	struct mpls_dev *out_mdev;
352 	struct mpls_dev *mdev;
353 	unsigned int hh_len;
354 	unsigned int new_header_size;
355 	unsigned int mtu;
356 	int err;
357 
358 	/* Careful this entire function runs inside of an rcu critical section */
359 
360 	mdev = mpls_dev_get(dev);
361 	if (!mdev)
362 		goto drop;
363 
364 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
365 			   rx_bytes);
366 
367 	if (!mdev->input_enabled) {
368 		MPLS_INC_STATS(mdev, rx_dropped);
369 		goto drop;
370 	}
371 
372 	if (skb->pkt_type != PACKET_HOST)
373 		goto err;
374 
375 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
376 		goto err;
377 
378 	if (!pskb_may_pull(skb, sizeof(*hdr)))
379 		goto err;
380 
381 	/* Read and decode the label */
382 	hdr = mpls_hdr(skb);
383 	dec = mpls_entry_decode(hdr);
384 
385 	rt = mpls_route_input_rcu(net, dec.label);
386 	if (!rt) {
387 		MPLS_INC_STATS(mdev, rx_noroute);
388 		goto drop;
389 	}
390 
391 	nh = mpls_select_multipath(rt, skb);
392 	if (!nh)
393 		goto err;
394 
395 	/* Pop the label */
396 	skb_pull(skb, sizeof(*hdr));
397 	skb_reset_network_header(skb);
398 
399 	skb_orphan(skb);
400 
401 	if (skb_warn_if_lro(skb))
402 		goto err;
403 
404 	skb_forward_csum(skb);
405 
406 	/* Verify ttl is valid */
407 	if (dec.ttl <= 1)
408 		goto err;
409 	dec.ttl -= 1;
410 
411 	/* Find the output device */
412 	out_dev = rcu_dereference(nh->nh_dev);
413 	if (!mpls_output_possible(out_dev))
414 		goto tx_err;
415 
416 	/* Verify the destination can hold the packet */
417 	new_header_size = mpls_nh_header_size(nh);
418 	mtu = mpls_dev_mtu(out_dev);
419 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
420 		goto tx_err;
421 
422 	hh_len = LL_RESERVED_SPACE(out_dev);
423 	if (!out_dev->header_ops)
424 		hh_len = 0;
425 
426 	/* Ensure there is enough space for the headers in the skb */
427 	if (skb_cow(skb, hh_len + new_header_size))
428 		goto tx_err;
429 
430 	skb->dev = out_dev;
431 	skb->protocol = htons(ETH_P_MPLS_UC);
432 
433 	if (unlikely(!new_header_size && dec.bos)) {
434 		/* Penultimate hop popping */
435 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
436 			goto err;
437 	} else {
438 		bool bos;
439 		int i;
440 		skb_push(skb, new_header_size);
441 		skb_reset_network_header(skb);
442 		/* Push the new labels */
443 		hdr = mpls_hdr(skb);
444 		bos = dec.bos;
445 		for (i = nh->nh_labels - 1; i >= 0; i--) {
446 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
447 						   dec.ttl, 0, bos);
448 			bos = false;
449 		}
450 	}
451 
452 	mpls_stats_inc_outucastpkts(out_dev, skb);
453 
454 	/* If via wasn't specified then send out using device address */
455 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
456 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
457 				 out_dev->dev_addr, skb);
458 	else
459 		err = neigh_xmit(nh->nh_via_table, out_dev,
460 				 mpls_nh_via(rt, nh), skb);
461 	if (err)
462 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
463 				    __func__, err);
464 	return 0;
465 
466 tx_err:
467 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
468 	if (out_mdev)
469 		MPLS_INC_STATS(out_mdev, tx_errors);
470 	goto drop;
471 err:
472 	MPLS_INC_STATS(mdev, rx_errors);
473 drop:
474 	kfree_skb(skb);
475 	return NET_RX_DROP;
476 }
477 
478 static struct packet_type mpls_packet_type __read_mostly = {
479 	.type = cpu_to_be16(ETH_P_MPLS_UC),
480 	.func = mpls_forward,
481 };
482 
483 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
484 	[RTA_DST]		= { .type = NLA_U32 },
485 	[RTA_OIF]		= { .type = NLA_U32 },
486 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
487 };
488 
489 struct mpls_route_config {
490 	u32			rc_protocol;
491 	u32			rc_ifindex;
492 	u8			rc_via_table;
493 	u8			rc_via_alen;
494 	u8			rc_via[MAX_VIA_ALEN];
495 	u32			rc_label;
496 	u8			rc_ttl_propagate;
497 	u8			rc_output_labels;
498 	u32			rc_output_label[MAX_NEW_LABELS];
499 	u32			rc_nlflags;
500 	enum mpls_payload_type	rc_payload_type;
501 	struct nl_info		rc_nlinfo;
502 	struct rtnexthop	*rc_mp;
503 	int			rc_mp_len;
504 };
505 
506 /* all nexthops within a route have the same size based on max
507  * number of labels and max via length for a hop
508  */
509 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
510 {
511 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
512 	struct mpls_route *rt;
513 	size_t size;
514 
515 	size = sizeof(*rt) + num_nh * nh_size;
516 	if (size > MAX_MPLS_ROUTE_MEM)
517 		return ERR_PTR(-EINVAL);
518 
519 	rt = kzalloc(size, GFP_KERNEL);
520 	if (!rt)
521 		return ERR_PTR(-ENOMEM);
522 
523 	rt->rt_nhn = num_nh;
524 	rt->rt_nhn_alive = num_nh;
525 	rt->rt_nh_size = nh_size;
526 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
527 
528 	return rt;
529 }
530 
531 static void mpls_rt_free(struct mpls_route *rt)
532 {
533 	if (rt)
534 		kfree_rcu(rt, rt_rcu);
535 }
536 
537 static void mpls_notify_route(struct net *net, unsigned index,
538 			      struct mpls_route *old, struct mpls_route *new,
539 			      const struct nl_info *info)
540 {
541 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
542 	unsigned portid = info ? info->portid : 0;
543 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
544 	struct mpls_route *rt = new ? new : old;
545 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
546 	/* Ignore reserved labels for now */
547 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
548 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
549 }
550 
551 static void mpls_route_update(struct net *net, unsigned index,
552 			      struct mpls_route *new,
553 			      const struct nl_info *info)
554 {
555 	struct mpls_route __rcu **platform_label;
556 	struct mpls_route *rt;
557 
558 	ASSERT_RTNL();
559 
560 	platform_label = rtnl_dereference(net->mpls.platform_label);
561 	rt = rtnl_dereference(platform_label[index]);
562 	rcu_assign_pointer(platform_label[index], new);
563 
564 	mpls_notify_route(net, index, rt, new, info);
565 
566 	/* If we removed a route free it now */
567 	mpls_rt_free(rt);
568 }
569 
570 static unsigned find_free_label(struct net *net)
571 {
572 	struct mpls_route __rcu **platform_label;
573 	size_t platform_labels;
574 	unsigned index;
575 
576 	platform_label = rtnl_dereference(net->mpls.platform_label);
577 	platform_labels = net->mpls.platform_labels;
578 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
579 	     index++) {
580 		if (!rtnl_dereference(platform_label[index]))
581 			return index;
582 	}
583 	return LABEL_NOT_SPECIFIED;
584 }
585 
586 #if IS_ENABLED(CONFIG_INET)
587 static struct net_device *inet_fib_lookup_dev(struct net *net,
588 					      const void *addr)
589 {
590 	struct net_device *dev;
591 	struct rtable *rt;
592 	struct in_addr daddr;
593 
594 	memcpy(&daddr, addr, sizeof(struct in_addr));
595 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
596 	if (IS_ERR(rt))
597 		return ERR_CAST(rt);
598 
599 	dev = rt->dst.dev;
600 	dev_hold(dev);
601 
602 	ip_rt_put(rt);
603 
604 	return dev;
605 }
606 #else
607 static struct net_device *inet_fib_lookup_dev(struct net *net,
608 					      const void *addr)
609 {
610 	return ERR_PTR(-EAFNOSUPPORT);
611 }
612 #endif
613 
614 #if IS_ENABLED(CONFIG_IPV6)
615 static struct net_device *inet6_fib_lookup_dev(struct net *net,
616 					       const void *addr)
617 {
618 	struct net_device *dev;
619 	struct dst_entry *dst;
620 	struct flowi6 fl6;
621 	int err;
622 
623 	if (!ipv6_stub)
624 		return ERR_PTR(-EAFNOSUPPORT);
625 
626 	memset(&fl6, 0, sizeof(fl6));
627 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
628 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
629 	if (err)
630 		return ERR_PTR(err);
631 
632 	dev = dst->dev;
633 	dev_hold(dev);
634 	dst_release(dst);
635 
636 	return dev;
637 }
638 #else
639 static struct net_device *inet6_fib_lookup_dev(struct net *net,
640 					       const void *addr)
641 {
642 	return ERR_PTR(-EAFNOSUPPORT);
643 }
644 #endif
645 
646 static struct net_device *find_outdev(struct net *net,
647 				      struct mpls_route *rt,
648 				      struct mpls_nh *nh, int oif)
649 {
650 	struct net_device *dev = NULL;
651 
652 	if (!oif) {
653 		switch (nh->nh_via_table) {
654 		case NEIGH_ARP_TABLE:
655 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
656 			break;
657 		case NEIGH_ND_TABLE:
658 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
659 			break;
660 		case NEIGH_LINK_TABLE:
661 			break;
662 		}
663 	} else {
664 		dev = dev_get_by_index(net, oif);
665 	}
666 
667 	if (!dev)
668 		return ERR_PTR(-ENODEV);
669 
670 	if (IS_ERR(dev))
671 		return dev;
672 
673 	/* The caller is holding rtnl anyways, so release the dev reference */
674 	dev_put(dev);
675 
676 	return dev;
677 }
678 
679 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
680 			      struct mpls_nh *nh, int oif)
681 {
682 	struct net_device *dev = NULL;
683 	int err = -ENODEV;
684 
685 	dev = find_outdev(net, rt, nh, oif);
686 	if (IS_ERR(dev)) {
687 		err = PTR_ERR(dev);
688 		dev = NULL;
689 		goto errout;
690 	}
691 
692 	/* Ensure this is a supported device */
693 	err = -EINVAL;
694 	if (!mpls_dev_get(dev))
695 		goto errout;
696 
697 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
698 	    (dev->addr_len != nh->nh_via_alen))
699 		goto errout;
700 
701 	RCU_INIT_POINTER(nh->nh_dev, dev);
702 
703 	if (!(dev->flags & IFF_UP)) {
704 		nh->nh_flags |= RTNH_F_DEAD;
705 	} else {
706 		unsigned int flags;
707 
708 		flags = dev_get_flags(dev);
709 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
710 			nh->nh_flags |= RTNH_F_LINKDOWN;
711 	}
712 
713 	return 0;
714 
715 errout:
716 	return err;
717 }
718 
719 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
720 		       u8 via_addr[], struct netlink_ext_ack *extack)
721 {
722 	struct rtvia *via = nla_data(nla);
723 	int err = -EINVAL;
724 	int alen;
725 
726 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
727 		NL_SET_ERR_MSG_ATTR(extack, nla,
728 				    "Invalid attribute length for RTA_VIA");
729 		goto errout;
730 	}
731 	alen = nla_len(nla) -
732 			offsetof(struct rtvia, rtvia_addr);
733 	if (alen > MAX_VIA_ALEN) {
734 		NL_SET_ERR_MSG_ATTR(extack, nla,
735 				    "Invalid address length for RTA_VIA");
736 		goto errout;
737 	}
738 
739 	/* Validate the address family */
740 	switch (via->rtvia_family) {
741 	case AF_PACKET:
742 		*via_table = NEIGH_LINK_TABLE;
743 		break;
744 	case AF_INET:
745 		*via_table = NEIGH_ARP_TABLE;
746 		if (alen != 4)
747 			goto errout;
748 		break;
749 	case AF_INET6:
750 		*via_table = NEIGH_ND_TABLE;
751 		if (alen != 16)
752 			goto errout;
753 		break;
754 	default:
755 		/* Unsupported address family */
756 		goto errout;
757 	}
758 
759 	memcpy(via_addr, via->rtvia_addr, alen);
760 	*via_alen = alen;
761 	err = 0;
762 
763 errout:
764 	return err;
765 }
766 
767 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
768 				  struct mpls_route *rt)
769 {
770 	struct net *net = cfg->rc_nlinfo.nl_net;
771 	struct mpls_nh *nh = rt->rt_nh;
772 	int err;
773 	int i;
774 
775 	if (!nh)
776 		return -ENOMEM;
777 
778 	nh->nh_labels = cfg->rc_output_labels;
779 	for (i = 0; i < nh->nh_labels; i++)
780 		nh->nh_label[i] = cfg->rc_output_label[i];
781 
782 	nh->nh_via_table = cfg->rc_via_table;
783 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
784 	nh->nh_via_alen = cfg->rc_via_alen;
785 
786 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
787 	if (err)
788 		goto errout;
789 
790 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
791 		rt->rt_nhn_alive--;
792 
793 	return 0;
794 
795 errout:
796 	return err;
797 }
798 
799 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
800 			 struct mpls_nh *nh, int oif, struct nlattr *via,
801 			 struct nlattr *newdst, u8 max_labels,
802 			 struct netlink_ext_ack *extack)
803 {
804 	int err = -ENOMEM;
805 
806 	if (!nh)
807 		goto errout;
808 
809 	if (newdst) {
810 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
811 				     nh->nh_label, extack);
812 		if (err)
813 			goto errout;
814 	}
815 
816 	if (via) {
817 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
818 				  __mpls_nh_via(rt, nh), extack);
819 		if (err)
820 			goto errout;
821 	} else {
822 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
823 	}
824 
825 	err = mpls_nh_assign_dev(net, rt, nh, oif);
826 	if (err)
827 		goto errout;
828 
829 	return 0;
830 
831 errout:
832 	return err;
833 }
834 
835 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
836 			      u8 cfg_via_alen, u8 *max_via_alen,
837 			      u8 *max_labels)
838 {
839 	int remaining = len;
840 	u8 nhs = 0;
841 
842 	*max_via_alen = 0;
843 	*max_labels = 0;
844 
845 	while (rtnh_ok(rtnh, remaining)) {
846 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
847 		int attrlen;
848 		u8 n_labels = 0;
849 
850 		attrlen = rtnh_attrlen(rtnh);
851 		nla = nla_find(attrs, attrlen, RTA_VIA);
852 		if (nla && nla_len(nla) >=
853 		    offsetof(struct rtvia, rtvia_addr)) {
854 			int via_alen = nla_len(nla) -
855 				offsetof(struct rtvia, rtvia_addr);
856 
857 			if (via_alen <= MAX_VIA_ALEN)
858 				*max_via_alen = max_t(u16, *max_via_alen,
859 						      via_alen);
860 		}
861 
862 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
863 		if (nla &&
864 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
865 				   NULL, NULL) != 0)
866 			return 0;
867 
868 		*max_labels = max_t(u8, *max_labels, n_labels);
869 
870 		/* number of nexthops is tracked by a u8.
871 		 * Check for overflow.
872 		 */
873 		if (nhs == 255)
874 			return 0;
875 		nhs++;
876 
877 		rtnh = rtnh_next(rtnh, &remaining);
878 	}
879 
880 	/* leftover implies invalid nexthop configuration, discard it */
881 	return remaining > 0 ? 0 : nhs;
882 }
883 
884 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
885 			       struct mpls_route *rt, u8 max_labels,
886 			       struct netlink_ext_ack *extack)
887 {
888 	struct rtnexthop *rtnh = cfg->rc_mp;
889 	struct nlattr *nla_via, *nla_newdst;
890 	int remaining = cfg->rc_mp_len;
891 	int err = 0;
892 	u8 nhs = 0;
893 
894 	change_nexthops(rt) {
895 		int attrlen;
896 
897 		nla_via = NULL;
898 		nla_newdst = NULL;
899 
900 		err = -EINVAL;
901 		if (!rtnh_ok(rtnh, remaining))
902 			goto errout;
903 
904 		/* neither weighted multipath nor any flags
905 		 * are supported
906 		 */
907 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
908 			goto errout;
909 
910 		attrlen = rtnh_attrlen(rtnh);
911 		if (attrlen > 0) {
912 			struct nlattr *attrs = rtnh_attrs(rtnh);
913 
914 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
915 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
916 		}
917 
918 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
919 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
920 				    max_labels, extack);
921 		if (err)
922 			goto errout;
923 
924 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
925 			rt->rt_nhn_alive--;
926 
927 		rtnh = rtnh_next(rtnh, &remaining);
928 		nhs++;
929 	} endfor_nexthops(rt);
930 
931 	rt->rt_nhn = nhs;
932 
933 	return 0;
934 
935 errout:
936 	return err;
937 }
938 
939 static bool mpls_label_ok(struct net *net, unsigned int *index,
940 			  struct netlink_ext_ack *extack)
941 {
942 	bool is_ok = true;
943 
944 	/* Reserved labels may not be set */
945 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
946 		NL_SET_ERR_MSG(extack,
947 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
948 		is_ok = false;
949 	}
950 
951 	/* The full 20 bit range may not be supported. */
952 	if (is_ok && *index >= net->mpls.platform_labels) {
953 		NL_SET_ERR_MSG(extack,
954 			       "Label >= configured maximum in platform_labels");
955 		is_ok = false;
956 	}
957 
958 	*index = array_index_nospec(*index, net->mpls.platform_labels);
959 	return is_ok;
960 }
961 
962 static int mpls_route_add(struct mpls_route_config *cfg,
963 			  struct netlink_ext_ack *extack)
964 {
965 	struct mpls_route __rcu **platform_label;
966 	struct net *net = cfg->rc_nlinfo.nl_net;
967 	struct mpls_route *rt, *old;
968 	int err = -EINVAL;
969 	u8 max_via_alen;
970 	unsigned index;
971 	u8 max_labels;
972 	u8 nhs;
973 
974 	index = cfg->rc_label;
975 
976 	/* If a label was not specified during insert pick one */
977 	if ((index == LABEL_NOT_SPECIFIED) &&
978 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
979 		index = find_free_label(net);
980 	}
981 
982 	if (!mpls_label_ok(net, &index, extack))
983 		goto errout;
984 
985 	/* Append makes no sense with mpls */
986 	err = -EOPNOTSUPP;
987 	if (cfg->rc_nlflags & NLM_F_APPEND) {
988 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
989 		goto errout;
990 	}
991 
992 	err = -EEXIST;
993 	platform_label = rtnl_dereference(net->mpls.platform_label);
994 	old = rtnl_dereference(platform_label[index]);
995 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
996 		goto errout;
997 
998 	err = -EEXIST;
999 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1000 		goto errout;
1001 
1002 	err = -ENOENT;
1003 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1004 		goto errout;
1005 
1006 	err = -EINVAL;
1007 	if (cfg->rc_mp) {
1008 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1009 					  cfg->rc_via_alen, &max_via_alen,
1010 					  &max_labels);
1011 	} else {
1012 		max_via_alen = cfg->rc_via_alen;
1013 		max_labels = cfg->rc_output_labels;
1014 		nhs = 1;
1015 	}
1016 
1017 	if (nhs == 0) {
1018 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1019 		goto errout;
1020 	}
1021 
1022 	err = -ENOMEM;
1023 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1024 	if (IS_ERR(rt)) {
1025 		err = PTR_ERR(rt);
1026 		goto errout;
1027 	}
1028 
1029 	rt->rt_protocol = cfg->rc_protocol;
1030 	rt->rt_payload_type = cfg->rc_payload_type;
1031 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1032 
1033 	if (cfg->rc_mp)
1034 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1035 	else
1036 		err = mpls_nh_build_from_cfg(cfg, rt);
1037 	if (err)
1038 		goto freert;
1039 
1040 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1041 
1042 	return 0;
1043 
1044 freert:
1045 	mpls_rt_free(rt);
1046 errout:
1047 	return err;
1048 }
1049 
1050 static int mpls_route_del(struct mpls_route_config *cfg,
1051 			  struct netlink_ext_ack *extack)
1052 {
1053 	struct net *net = cfg->rc_nlinfo.nl_net;
1054 	unsigned index;
1055 	int err = -EINVAL;
1056 
1057 	index = cfg->rc_label;
1058 
1059 	if (!mpls_label_ok(net, &index, extack))
1060 		goto errout;
1061 
1062 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1063 
1064 	err = 0;
1065 errout:
1066 	return err;
1067 }
1068 
1069 static void mpls_get_stats(struct mpls_dev *mdev,
1070 			   struct mpls_link_stats *stats)
1071 {
1072 	struct mpls_pcpu_stats *p;
1073 	int i;
1074 
1075 	memset(stats, 0, sizeof(*stats));
1076 
1077 	for_each_possible_cpu(i) {
1078 		struct mpls_link_stats local;
1079 		unsigned int start;
1080 
1081 		p = per_cpu_ptr(mdev->stats, i);
1082 		do {
1083 			start = u64_stats_fetch_begin(&p->syncp);
1084 			local = p->stats;
1085 		} while (u64_stats_fetch_retry(&p->syncp, start));
1086 
1087 		stats->rx_packets	+= local.rx_packets;
1088 		stats->rx_bytes		+= local.rx_bytes;
1089 		stats->tx_packets	+= local.tx_packets;
1090 		stats->tx_bytes		+= local.tx_bytes;
1091 		stats->rx_errors	+= local.rx_errors;
1092 		stats->tx_errors	+= local.tx_errors;
1093 		stats->rx_dropped	+= local.rx_dropped;
1094 		stats->tx_dropped	+= local.tx_dropped;
1095 		stats->rx_noroute	+= local.rx_noroute;
1096 	}
1097 }
1098 
1099 static int mpls_fill_stats_af(struct sk_buff *skb,
1100 			      const struct net_device *dev)
1101 {
1102 	struct mpls_link_stats *stats;
1103 	struct mpls_dev *mdev;
1104 	struct nlattr *nla;
1105 
1106 	mdev = mpls_dev_get(dev);
1107 	if (!mdev)
1108 		return -ENODATA;
1109 
1110 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1111 				sizeof(struct mpls_link_stats),
1112 				MPLS_STATS_UNSPEC);
1113 	if (!nla)
1114 		return -EMSGSIZE;
1115 
1116 	stats = nla_data(nla);
1117 	mpls_get_stats(mdev, stats);
1118 
1119 	return 0;
1120 }
1121 
1122 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1123 {
1124 	struct mpls_dev *mdev;
1125 
1126 	mdev = mpls_dev_get(dev);
1127 	if (!mdev)
1128 		return 0;
1129 
1130 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1131 }
1132 
1133 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1134 				     u32 portid, u32 seq, int event,
1135 				     unsigned int flags, int type)
1136 {
1137 	struct nlmsghdr  *nlh;
1138 	struct netconfmsg *ncm;
1139 	bool all = false;
1140 
1141 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1142 			flags);
1143 	if (!nlh)
1144 		return -EMSGSIZE;
1145 
1146 	if (type == NETCONFA_ALL)
1147 		all = true;
1148 
1149 	ncm = nlmsg_data(nlh);
1150 	ncm->ncm_family = AF_MPLS;
1151 
1152 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1153 		goto nla_put_failure;
1154 
1155 	if ((all || type == NETCONFA_INPUT) &&
1156 	    nla_put_s32(skb, NETCONFA_INPUT,
1157 			mdev->input_enabled) < 0)
1158 		goto nla_put_failure;
1159 
1160 	nlmsg_end(skb, nlh);
1161 	return 0;
1162 
1163 nla_put_failure:
1164 	nlmsg_cancel(skb, nlh);
1165 	return -EMSGSIZE;
1166 }
1167 
1168 static int mpls_netconf_msgsize_devconf(int type)
1169 {
1170 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1171 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1172 	bool all = false;
1173 
1174 	if (type == NETCONFA_ALL)
1175 		all = true;
1176 
1177 	if (all || type == NETCONFA_INPUT)
1178 		size += nla_total_size(4);
1179 
1180 	return size;
1181 }
1182 
1183 static void mpls_netconf_notify_devconf(struct net *net, int event,
1184 					int type, struct mpls_dev *mdev)
1185 {
1186 	struct sk_buff *skb;
1187 	int err = -ENOBUFS;
1188 
1189 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1190 	if (!skb)
1191 		goto errout;
1192 
1193 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1194 	if (err < 0) {
1195 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1196 		WARN_ON(err == -EMSGSIZE);
1197 		kfree_skb(skb);
1198 		goto errout;
1199 	}
1200 
1201 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1202 	return;
1203 errout:
1204 	if (err < 0)
1205 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1206 }
1207 
1208 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1209 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1210 };
1211 
1212 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1213 				    struct nlmsghdr *nlh,
1214 				    struct netlink_ext_ack *extack)
1215 {
1216 	struct net *net = sock_net(in_skb->sk);
1217 	struct nlattr *tb[NETCONFA_MAX + 1];
1218 	struct netconfmsg *ncm;
1219 	struct net_device *dev;
1220 	struct mpls_dev *mdev;
1221 	struct sk_buff *skb;
1222 	int ifindex;
1223 	int err;
1224 
1225 	err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1226 			  devconf_mpls_policy, NULL);
1227 	if (err < 0)
1228 		goto errout;
1229 
1230 	err = -EINVAL;
1231 	if (!tb[NETCONFA_IFINDEX])
1232 		goto errout;
1233 
1234 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1235 	dev = __dev_get_by_index(net, ifindex);
1236 	if (!dev)
1237 		goto errout;
1238 
1239 	mdev = mpls_dev_get(dev);
1240 	if (!mdev)
1241 		goto errout;
1242 
1243 	err = -ENOBUFS;
1244 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1245 	if (!skb)
1246 		goto errout;
1247 
1248 	err = mpls_netconf_fill_devconf(skb, mdev,
1249 					NETLINK_CB(in_skb).portid,
1250 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1251 					NETCONFA_ALL);
1252 	if (err < 0) {
1253 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1254 		WARN_ON(err == -EMSGSIZE);
1255 		kfree_skb(skb);
1256 		goto errout;
1257 	}
1258 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1259 errout:
1260 	return err;
1261 }
1262 
1263 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1264 				     struct netlink_callback *cb)
1265 {
1266 	struct net *net = sock_net(skb->sk);
1267 	struct hlist_head *head;
1268 	struct net_device *dev;
1269 	struct mpls_dev *mdev;
1270 	int idx, s_idx;
1271 	int h, s_h;
1272 
1273 	s_h = cb->args[0];
1274 	s_idx = idx = cb->args[1];
1275 
1276 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1277 		idx = 0;
1278 		head = &net->dev_index_head[h];
1279 		rcu_read_lock();
1280 		cb->seq = net->dev_base_seq;
1281 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1282 			if (idx < s_idx)
1283 				goto cont;
1284 			mdev = mpls_dev_get(dev);
1285 			if (!mdev)
1286 				goto cont;
1287 			if (mpls_netconf_fill_devconf(skb, mdev,
1288 						      NETLINK_CB(cb->skb).portid,
1289 						      cb->nlh->nlmsg_seq,
1290 						      RTM_NEWNETCONF,
1291 						      NLM_F_MULTI,
1292 						      NETCONFA_ALL) < 0) {
1293 				rcu_read_unlock();
1294 				goto done;
1295 			}
1296 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1297 cont:
1298 			idx++;
1299 		}
1300 		rcu_read_unlock();
1301 	}
1302 done:
1303 	cb->args[0] = h;
1304 	cb->args[1] = idx;
1305 
1306 	return skb->len;
1307 }
1308 
1309 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1310 	(&((struct mpls_dev *)0)->field)
1311 
1312 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1313 			  void __user *buffer,
1314 			  size_t *lenp, loff_t *ppos)
1315 {
1316 	int oval = *(int *)ctl->data;
1317 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1318 
1319 	if (write) {
1320 		struct mpls_dev *mdev = ctl->extra1;
1321 		int i = (int *)ctl->data - (int *)mdev;
1322 		struct net *net = ctl->extra2;
1323 		int val = *(int *)ctl->data;
1324 
1325 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1326 		    val != oval) {
1327 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1328 						    NETCONFA_INPUT, mdev);
1329 		}
1330 	}
1331 
1332 	return ret;
1333 }
1334 
1335 static const struct ctl_table mpls_dev_table[] = {
1336 	{
1337 		.procname	= "input",
1338 		.maxlen		= sizeof(int),
1339 		.mode		= 0644,
1340 		.proc_handler	= mpls_conf_proc,
1341 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1342 	},
1343 	{ }
1344 };
1345 
1346 static int mpls_dev_sysctl_register(struct net_device *dev,
1347 				    struct mpls_dev *mdev)
1348 {
1349 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1350 	struct net *net = dev_net(dev);
1351 	struct ctl_table *table;
1352 	int i;
1353 
1354 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1355 	if (!table)
1356 		goto out;
1357 
1358 	/* Table data contains only offsets relative to the base of
1359 	 * the mdev at this point, so make them absolute.
1360 	 */
1361 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1362 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1363 		table[i].extra1 = mdev;
1364 		table[i].extra2 = net;
1365 	}
1366 
1367 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1368 
1369 	mdev->sysctl = register_net_sysctl(net, path, table);
1370 	if (!mdev->sysctl)
1371 		goto free;
1372 
1373 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1374 	return 0;
1375 
1376 free:
1377 	kfree(table);
1378 out:
1379 	return -ENOBUFS;
1380 }
1381 
1382 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1383 				       struct mpls_dev *mdev)
1384 {
1385 	struct net *net = dev_net(dev);
1386 	struct ctl_table *table;
1387 
1388 	table = mdev->sysctl->ctl_table_arg;
1389 	unregister_net_sysctl_table(mdev->sysctl);
1390 	kfree(table);
1391 
1392 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1393 }
1394 
1395 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1396 {
1397 	struct mpls_dev *mdev;
1398 	int err = -ENOMEM;
1399 	int i;
1400 
1401 	ASSERT_RTNL();
1402 
1403 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1404 	if (!mdev)
1405 		return ERR_PTR(err);
1406 
1407 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1408 	if (!mdev->stats)
1409 		goto free;
1410 
1411 	for_each_possible_cpu(i) {
1412 		struct mpls_pcpu_stats *mpls_stats;
1413 
1414 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1415 		u64_stats_init(&mpls_stats->syncp);
1416 	}
1417 
1418 	mdev->dev = dev;
1419 
1420 	err = mpls_dev_sysctl_register(dev, mdev);
1421 	if (err)
1422 		goto free;
1423 
1424 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1425 
1426 	return mdev;
1427 
1428 free:
1429 	free_percpu(mdev->stats);
1430 	kfree(mdev);
1431 	return ERR_PTR(err);
1432 }
1433 
1434 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1435 {
1436 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1437 
1438 	free_percpu(mdev->stats);
1439 	kfree(mdev);
1440 }
1441 
1442 static void mpls_ifdown(struct net_device *dev, int event)
1443 {
1444 	struct mpls_route __rcu **platform_label;
1445 	struct net *net = dev_net(dev);
1446 	u8 alive, deleted;
1447 	unsigned index;
1448 
1449 	platform_label = rtnl_dereference(net->mpls.platform_label);
1450 	for (index = 0; index < net->mpls.platform_labels; index++) {
1451 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1452 
1453 		if (!rt)
1454 			continue;
1455 
1456 		alive = 0;
1457 		deleted = 0;
1458 		change_nexthops(rt) {
1459 			unsigned int nh_flags = nh->nh_flags;
1460 
1461 			if (rtnl_dereference(nh->nh_dev) != dev)
1462 				goto next;
1463 
1464 			switch (event) {
1465 			case NETDEV_DOWN:
1466 			case NETDEV_UNREGISTER:
1467 				nh_flags |= RTNH_F_DEAD;
1468 				/* fall through */
1469 			case NETDEV_CHANGE:
1470 				nh_flags |= RTNH_F_LINKDOWN;
1471 				break;
1472 			}
1473 			if (event == NETDEV_UNREGISTER)
1474 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1475 
1476 			if (nh->nh_flags != nh_flags)
1477 				WRITE_ONCE(nh->nh_flags, nh_flags);
1478 next:
1479 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1480 				alive++;
1481 			if (!rtnl_dereference(nh->nh_dev))
1482 				deleted++;
1483 		} endfor_nexthops(rt);
1484 
1485 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1486 
1487 		/* if there are no more nexthops, delete the route */
1488 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1489 			mpls_route_update(net, index, NULL, NULL);
1490 	}
1491 }
1492 
1493 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1494 {
1495 	struct mpls_route __rcu **platform_label;
1496 	struct net *net = dev_net(dev);
1497 	unsigned index;
1498 	u8 alive;
1499 
1500 	platform_label = rtnl_dereference(net->mpls.platform_label);
1501 	for (index = 0; index < net->mpls.platform_labels; index++) {
1502 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1503 
1504 		if (!rt)
1505 			continue;
1506 
1507 		alive = 0;
1508 		change_nexthops(rt) {
1509 			unsigned int nh_flags = nh->nh_flags;
1510 			struct net_device *nh_dev =
1511 				rtnl_dereference(nh->nh_dev);
1512 
1513 			if (!(nh_flags & flags)) {
1514 				alive++;
1515 				continue;
1516 			}
1517 			if (nh_dev != dev)
1518 				continue;
1519 			alive++;
1520 			nh_flags &= ~flags;
1521 			WRITE_ONCE(nh->nh_flags, nh_flags);
1522 		} endfor_nexthops(rt);
1523 
1524 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1525 	}
1526 }
1527 
1528 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1529 			   void *ptr)
1530 {
1531 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1532 	struct mpls_dev *mdev;
1533 	unsigned int flags;
1534 
1535 	if (event == NETDEV_REGISTER) {
1536 
1537 		/* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1538 		 * IPIP devices
1539 		 */
1540 		if (dev->type == ARPHRD_ETHER ||
1541 		    dev->type == ARPHRD_LOOPBACK ||
1542 		    dev->type == ARPHRD_IPGRE ||
1543 		    dev->type == ARPHRD_IP6GRE ||
1544 		    dev->type == ARPHRD_SIT ||
1545 		    dev->type == ARPHRD_TUNNEL) {
1546 			mdev = mpls_add_dev(dev);
1547 			if (IS_ERR(mdev))
1548 				return notifier_from_errno(PTR_ERR(mdev));
1549 		}
1550 		return NOTIFY_OK;
1551 	}
1552 
1553 	mdev = mpls_dev_get(dev);
1554 	if (!mdev)
1555 		return NOTIFY_OK;
1556 
1557 	switch (event) {
1558 	case NETDEV_DOWN:
1559 		mpls_ifdown(dev, event);
1560 		break;
1561 	case NETDEV_UP:
1562 		flags = dev_get_flags(dev);
1563 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1564 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1565 		else
1566 			mpls_ifup(dev, RTNH_F_DEAD);
1567 		break;
1568 	case NETDEV_CHANGE:
1569 		flags = dev_get_flags(dev);
1570 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1571 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1572 		else
1573 			mpls_ifdown(dev, event);
1574 		break;
1575 	case NETDEV_UNREGISTER:
1576 		mpls_ifdown(dev, event);
1577 		mdev = mpls_dev_get(dev);
1578 		if (mdev) {
1579 			mpls_dev_sysctl_unregister(dev, mdev);
1580 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1581 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1582 		}
1583 		break;
1584 	case NETDEV_CHANGENAME:
1585 		mdev = mpls_dev_get(dev);
1586 		if (mdev) {
1587 			int err;
1588 
1589 			mpls_dev_sysctl_unregister(dev, mdev);
1590 			err = mpls_dev_sysctl_register(dev, mdev);
1591 			if (err)
1592 				return notifier_from_errno(err);
1593 		}
1594 		break;
1595 	}
1596 	return NOTIFY_OK;
1597 }
1598 
1599 static struct notifier_block mpls_dev_notifier = {
1600 	.notifier_call = mpls_dev_notify,
1601 };
1602 
1603 static int nla_put_via(struct sk_buff *skb,
1604 		       u8 table, const void *addr, int alen)
1605 {
1606 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1607 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1608 	};
1609 	struct nlattr *nla;
1610 	struct rtvia *via;
1611 	int family = AF_UNSPEC;
1612 
1613 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1614 	if (!nla)
1615 		return -EMSGSIZE;
1616 
1617 	if (table <= NEIGH_NR_TABLES)
1618 		family = table_to_family[table];
1619 
1620 	via = nla_data(nla);
1621 	via->rtvia_family = family;
1622 	memcpy(via->rtvia_addr, addr, alen);
1623 	return 0;
1624 }
1625 
1626 int nla_put_labels(struct sk_buff *skb, int attrtype,
1627 		   u8 labels, const u32 label[])
1628 {
1629 	struct nlattr *nla;
1630 	struct mpls_shim_hdr *nla_label;
1631 	bool bos;
1632 	int i;
1633 	nla = nla_reserve(skb, attrtype, labels*4);
1634 	if (!nla)
1635 		return -EMSGSIZE;
1636 
1637 	nla_label = nla_data(nla);
1638 	bos = true;
1639 	for (i = labels - 1; i >= 0; i--) {
1640 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1641 		bos = false;
1642 	}
1643 
1644 	return 0;
1645 }
1646 EXPORT_SYMBOL_GPL(nla_put_labels);
1647 
1648 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1649 		   u32 label[], struct netlink_ext_ack *extack)
1650 {
1651 	unsigned len = nla_len(nla);
1652 	struct mpls_shim_hdr *nla_label;
1653 	u8 nla_labels;
1654 	bool bos;
1655 	int i;
1656 
1657 	/* len needs to be an even multiple of 4 (the label size). Number
1658 	 * of labels is a u8 so check for overflow.
1659 	 */
1660 	if (len & 3 || len / 4 > 255) {
1661 		NL_SET_ERR_MSG_ATTR(extack, nla,
1662 				    "Invalid length for labels attribute");
1663 		return -EINVAL;
1664 	}
1665 
1666 	/* Limit the number of new labels allowed */
1667 	nla_labels = len/4;
1668 	if (nla_labels > max_labels) {
1669 		NL_SET_ERR_MSG(extack, "Too many labels");
1670 		return -EINVAL;
1671 	}
1672 
1673 	/* when label == NULL, caller wants number of labels */
1674 	if (!label)
1675 		goto out;
1676 
1677 	nla_label = nla_data(nla);
1678 	bos = true;
1679 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1680 		struct mpls_entry_decoded dec;
1681 		dec = mpls_entry_decode(nla_label + i);
1682 
1683 		/* Ensure the bottom of stack flag is properly set
1684 		 * and ttl and tc are both clear.
1685 		 */
1686 		if (dec.ttl) {
1687 			NL_SET_ERR_MSG_ATTR(extack, nla,
1688 					    "TTL in label must be 0");
1689 			return -EINVAL;
1690 		}
1691 
1692 		if (dec.tc) {
1693 			NL_SET_ERR_MSG_ATTR(extack, nla,
1694 					    "Traffic class in label must be 0");
1695 			return -EINVAL;
1696 		}
1697 
1698 		if (dec.bos != bos) {
1699 			NL_SET_BAD_ATTR(extack, nla);
1700 			if (bos) {
1701 				NL_SET_ERR_MSG(extack,
1702 					       "BOS bit must be set in first label");
1703 			} else {
1704 				NL_SET_ERR_MSG(extack,
1705 					       "BOS bit can only be set in first label");
1706 			}
1707 			return -EINVAL;
1708 		}
1709 
1710 		switch (dec.label) {
1711 		case MPLS_LABEL_IMPLNULL:
1712 			/* RFC3032: This is a label that an LSR may
1713 			 * assign and distribute, but which never
1714 			 * actually appears in the encapsulation.
1715 			 */
1716 			NL_SET_ERR_MSG_ATTR(extack, nla,
1717 					    "Implicit NULL Label (3) can not be used in encapsulation");
1718 			return -EINVAL;
1719 		}
1720 
1721 		label[i] = dec.label;
1722 	}
1723 out:
1724 	*labels = nla_labels;
1725 	return 0;
1726 }
1727 EXPORT_SYMBOL_GPL(nla_get_labels);
1728 
1729 static int rtm_to_route_config(struct sk_buff *skb,
1730 			       struct nlmsghdr *nlh,
1731 			       struct mpls_route_config *cfg,
1732 			       struct netlink_ext_ack *extack)
1733 {
1734 	struct rtmsg *rtm;
1735 	struct nlattr *tb[RTA_MAX+1];
1736 	int index;
1737 	int err;
1738 
1739 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy,
1740 			  extack);
1741 	if (err < 0)
1742 		goto errout;
1743 
1744 	err = -EINVAL;
1745 	rtm = nlmsg_data(nlh);
1746 
1747 	if (rtm->rtm_family != AF_MPLS) {
1748 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1749 		goto errout;
1750 	}
1751 	if (rtm->rtm_dst_len != 20) {
1752 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1753 		goto errout;
1754 	}
1755 	if (rtm->rtm_src_len != 0) {
1756 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1757 		goto errout;
1758 	}
1759 	if (rtm->rtm_tos != 0) {
1760 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1761 		goto errout;
1762 	}
1763 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1764 		NL_SET_ERR_MSG(extack,
1765 			       "MPLS only supports the main route table");
1766 		goto errout;
1767 	}
1768 	/* Any value is acceptable for rtm_protocol */
1769 
1770 	/* As mpls uses destination specific addresses
1771 	 * (or source specific address in the case of multicast)
1772 	 * all addresses have universal scope.
1773 	 */
1774 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1775 		NL_SET_ERR_MSG(extack,
1776 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1777 		goto errout;
1778 	}
1779 	if (rtm->rtm_type != RTN_UNICAST) {
1780 		NL_SET_ERR_MSG(extack,
1781 			       "Invalid route type - MPLS only supports UNICAST");
1782 		goto errout;
1783 	}
1784 	if (rtm->rtm_flags != 0) {
1785 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1786 		goto errout;
1787 	}
1788 
1789 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1790 	cfg->rc_protocol	= rtm->rtm_protocol;
1791 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1792 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1793 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1794 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1795 	cfg->rc_nlinfo.nlh	= nlh;
1796 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1797 
1798 	for (index = 0; index <= RTA_MAX; index++) {
1799 		struct nlattr *nla = tb[index];
1800 		if (!nla)
1801 			continue;
1802 
1803 		switch (index) {
1804 		case RTA_OIF:
1805 			cfg->rc_ifindex = nla_get_u32(nla);
1806 			break;
1807 		case RTA_NEWDST:
1808 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1809 					   &cfg->rc_output_labels,
1810 					   cfg->rc_output_label, extack))
1811 				goto errout;
1812 			break;
1813 		case RTA_DST:
1814 		{
1815 			u8 label_count;
1816 			if (nla_get_labels(nla, 1, &label_count,
1817 					   &cfg->rc_label, extack))
1818 				goto errout;
1819 
1820 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1821 					   &cfg->rc_label, extack))
1822 				goto errout;
1823 			break;
1824 		}
1825 		case RTA_VIA:
1826 		{
1827 			if (nla_get_via(nla, &cfg->rc_via_alen,
1828 					&cfg->rc_via_table, cfg->rc_via,
1829 					extack))
1830 				goto errout;
1831 			break;
1832 		}
1833 		case RTA_MULTIPATH:
1834 		{
1835 			cfg->rc_mp = nla_data(nla);
1836 			cfg->rc_mp_len = nla_len(nla);
1837 			break;
1838 		}
1839 		case RTA_TTL_PROPAGATE:
1840 		{
1841 			u8 ttl_propagate = nla_get_u8(nla);
1842 
1843 			if (ttl_propagate > 1) {
1844 				NL_SET_ERR_MSG_ATTR(extack, nla,
1845 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1846 				goto errout;
1847 			}
1848 			cfg->rc_ttl_propagate = ttl_propagate ?
1849 				MPLS_TTL_PROP_ENABLED :
1850 				MPLS_TTL_PROP_DISABLED;
1851 			break;
1852 		}
1853 		default:
1854 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1855 			/* Unsupported attribute */
1856 			goto errout;
1857 		}
1858 	}
1859 
1860 	err = 0;
1861 errout:
1862 	return err;
1863 }
1864 
1865 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1866 			     struct netlink_ext_ack *extack)
1867 {
1868 	struct mpls_route_config *cfg;
1869 	int err;
1870 
1871 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1872 	if (!cfg)
1873 		return -ENOMEM;
1874 
1875 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1876 	if (err < 0)
1877 		goto out;
1878 
1879 	err = mpls_route_del(cfg, extack);
1880 out:
1881 	kfree(cfg);
1882 
1883 	return err;
1884 }
1885 
1886 
1887 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1888 			     struct netlink_ext_ack *extack)
1889 {
1890 	struct mpls_route_config *cfg;
1891 	int err;
1892 
1893 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1894 	if (!cfg)
1895 		return -ENOMEM;
1896 
1897 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1898 	if (err < 0)
1899 		goto out;
1900 
1901 	err = mpls_route_add(cfg, extack);
1902 out:
1903 	kfree(cfg);
1904 
1905 	return err;
1906 }
1907 
1908 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1909 			   u32 label, struct mpls_route *rt, int flags)
1910 {
1911 	struct net_device *dev;
1912 	struct nlmsghdr *nlh;
1913 	struct rtmsg *rtm;
1914 
1915 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1916 	if (nlh == NULL)
1917 		return -EMSGSIZE;
1918 
1919 	rtm = nlmsg_data(nlh);
1920 	rtm->rtm_family = AF_MPLS;
1921 	rtm->rtm_dst_len = 20;
1922 	rtm->rtm_src_len = 0;
1923 	rtm->rtm_tos = 0;
1924 	rtm->rtm_table = RT_TABLE_MAIN;
1925 	rtm->rtm_protocol = rt->rt_protocol;
1926 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1927 	rtm->rtm_type = RTN_UNICAST;
1928 	rtm->rtm_flags = 0;
1929 
1930 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1931 		goto nla_put_failure;
1932 
1933 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1934 		bool ttl_propagate =
1935 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1936 
1937 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1938 			       ttl_propagate))
1939 			goto nla_put_failure;
1940 	}
1941 	if (rt->rt_nhn == 1) {
1942 		const struct mpls_nh *nh = rt->rt_nh;
1943 
1944 		if (nh->nh_labels &&
1945 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1946 				   nh->nh_label))
1947 			goto nla_put_failure;
1948 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1949 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1950 				nh->nh_via_alen))
1951 			goto nla_put_failure;
1952 		dev = rtnl_dereference(nh->nh_dev);
1953 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1954 			goto nla_put_failure;
1955 		if (nh->nh_flags & RTNH_F_LINKDOWN)
1956 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1957 		if (nh->nh_flags & RTNH_F_DEAD)
1958 			rtm->rtm_flags |= RTNH_F_DEAD;
1959 	} else {
1960 		struct rtnexthop *rtnh;
1961 		struct nlattr *mp;
1962 		u8 linkdown = 0;
1963 		u8 dead = 0;
1964 
1965 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1966 		if (!mp)
1967 			goto nla_put_failure;
1968 
1969 		for_nexthops(rt) {
1970 			dev = rtnl_dereference(nh->nh_dev);
1971 			if (!dev)
1972 				continue;
1973 
1974 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1975 			if (!rtnh)
1976 				goto nla_put_failure;
1977 
1978 			rtnh->rtnh_ifindex = dev->ifindex;
1979 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1980 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1981 				linkdown++;
1982 			}
1983 			if (nh->nh_flags & RTNH_F_DEAD) {
1984 				rtnh->rtnh_flags |= RTNH_F_DEAD;
1985 				dead++;
1986 			}
1987 
1988 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1989 							    nh->nh_labels,
1990 							    nh->nh_label))
1991 				goto nla_put_failure;
1992 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1993 			    nla_put_via(skb, nh->nh_via_table,
1994 					mpls_nh_via(rt, nh),
1995 					nh->nh_via_alen))
1996 				goto nla_put_failure;
1997 
1998 			/* length of rtnetlink header + attributes */
1999 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2000 		} endfor_nexthops(rt);
2001 
2002 		if (linkdown == rt->rt_nhn)
2003 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2004 		if (dead == rt->rt_nhn)
2005 			rtm->rtm_flags |= RTNH_F_DEAD;
2006 
2007 		nla_nest_end(skb, mp);
2008 	}
2009 
2010 	nlmsg_end(skb, nlh);
2011 	return 0;
2012 
2013 nla_put_failure:
2014 	nlmsg_cancel(skb, nlh);
2015 	return -EMSGSIZE;
2016 }
2017 
2018 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2019 {
2020 	struct net *net = sock_net(skb->sk);
2021 	struct mpls_route __rcu **platform_label;
2022 	size_t platform_labels;
2023 	unsigned int index;
2024 
2025 	ASSERT_RTNL();
2026 
2027 	index = cb->args[0];
2028 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2029 		index = MPLS_LABEL_FIRST_UNRESERVED;
2030 
2031 	platform_label = rtnl_dereference(net->mpls.platform_label);
2032 	platform_labels = net->mpls.platform_labels;
2033 	for (; index < platform_labels; index++) {
2034 		struct mpls_route *rt;
2035 		rt = rtnl_dereference(platform_label[index]);
2036 		if (!rt)
2037 			continue;
2038 
2039 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2040 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2041 				    index, rt, NLM_F_MULTI) < 0)
2042 			break;
2043 	}
2044 	cb->args[0] = index;
2045 
2046 	return skb->len;
2047 }
2048 
2049 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2050 {
2051 	size_t payload =
2052 		NLMSG_ALIGN(sizeof(struct rtmsg))
2053 		+ nla_total_size(4)			/* RTA_DST */
2054 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2055 
2056 	if (rt->rt_nhn == 1) {
2057 		struct mpls_nh *nh = rt->rt_nh;
2058 
2059 		if (nh->nh_dev)
2060 			payload += nla_total_size(4); /* RTA_OIF */
2061 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2062 			payload += nla_total_size(2 + nh->nh_via_alen);
2063 		if (nh->nh_labels) /* RTA_NEWDST */
2064 			payload += nla_total_size(nh->nh_labels * 4);
2065 	} else {
2066 		/* each nexthop is packed in an attribute */
2067 		size_t nhsize = 0;
2068 
2069 		for_nexthops(rt) {
2070 			if (!rtnl_dereference(nh->nh_dev))
2071 				continue;
2072 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2073 			/* RTA_VIA */
2074 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2075 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2076 			if (nh->nh_labels)
2077 				nhsize += nla_total_size(nh->nh_labels * 4);
2078 		} endfor_nexthops(rt);
2079 		/* nested attribute */
2080 		payload += nla_total_size(nhsize);
2081 	}
2082 
2083 	return payload;
2084 }
2085 
2086 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2087 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2088 		       unsigned int nlm_flags)
2089 {
2090 	struct sk_buff *skb;
2091 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2092 	int err = -ENOBUFS;
2093 
2094 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2095 	if (skb == NULL)
2096 		goto errout;
2097 
2098 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2099 	if (err < 0) {
2100 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2101 		WARN_ON(err == -EMSGSIZE);
2102 		kfree_skb(skb);
2103 		goto errout;
2104 	}
2105 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2106 
2107 	return;
2108 errout:
2109 	if (err < 0)
2110 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2111 }
2112 
2113 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2114 			 struct netlink_ext_ack *extack)
2115 {
2116 	struct net *net = sock_net(in_skb->sk);
2117 	u32 portid = NETLINK_CB(in_skb).portid;
2118 	u32 in_label = LABEL_NOT_SPECIFIED;
2119 	struct nlattr *tb[RTA_MAX + 1];
2120 	u32 labels[MAX_NEW_LABELS];
2121 	struct mpls_shim_hdr *hdr;
2122 	unsigned int hdr_size = 0;
2123 	struct net_device *dev;
2124 	struct mpls_route *rt;
2125 	struct rtmsg *rtm, *r;
2126 	struct nlmsghdr *nlh;
2127 	struct sk_buff *skb;
2128 	struct mpls_nh *nh;
2129 	u8 n_labels;
2130 	int err;
2131 
2132 	err = nlmsg_parse(in_nlh, sizeof(*rtm), tb, RTA_MAX,
2133 			  rtm_mpls_policy, extack);
2134 	if (err < 0)
2135 		goto errout;
2136 
2137 	rtm = nlmsg_data(in_nlh);
2138 
2139 	if (tb[RTA_DST]) {
2140 		u8 label_count;
2141 
2142 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2143 				   &in_label, extack)) {
2144 			err = -EINVAL;
2145 			goto errout;
2146 		}
2147 
2148 		if (!mpls_label_ok(net, &in_label, extack)) {
2149 			err = -EINVAL;
2150 			goto errout;
2151 		}
2152 	}
2153 
2154 	rt = mpls_route_input_rcu(net, in_label);
2155 	if (!rt) {
2156 		err = -ENETUNREACH;
2157 		goto errout;
2158 	}
2159 
2160 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2161 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2162 		if (!skb) {
2163 			err = -ENOBUFS;
2164 			goto errout;
2165 		}
2166 
2167 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2168 				      RTM_NEWROUTE, in_label, rt, 0);
2169 		if (err < 0) {
2170 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2171 			WARN_ON(err == -EMSGSIZE);
2172 			goto errout_free;
2173 		}
2174 
2175 		return rtnl_unicast(skb, net, portid);
2176 	}
2177 
2178 	if (tb[RTA_NEWDST]) {
2179 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2180 				   labels, extack) != 0) {
2181 			err = -EINVAL;
2182 			goto errout;
2183 		}
2184 
2185 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2186 	}
2187 
2188 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2189 	if (!skb) {
2190 		err = -ENOBUFS;
2191 		goto errout;
2192 	}
2193 
2194 	skb->protocol = htons(ETH_P_MPLS_UC);
2195 
2196 	if (hdr_size) {
2197 		bool bos;
2198 		int i;
2199 
2200 		if (skb_cow(skb, hdr_size)) {
2201 			err = -ENOBUFS;
2202 			goto errout_free;
2203 		}
2204 
2205 		skb_reserve(skb, hdr_size);
2206 		skb_push(skb, hdr_size);
2207 		skb_reset_network_header(skb);
2208 
2209 		/* Push new labels */
2210 		hdr = mpls_hdr(skb);
2211 		bos = true;
2212 		for (i = n_labels - 1; i >= 0; i--) {
2213 			hdr[i] = mpls_entry_encode(labels[i],
2214 						   1, 0, bos);
2215 			bos = false;
2216 		}
2217 	}
2218 
2219 	nh = mpls_select_multipath(rt, skb);
2220 	if (!nh) {
2221 		err = -ENETUNREACH;
2222 		goto errout_free;
2223 	}
2224 
2225 	if (hdr_size) {
2226 		skb_pull(skb, hdr_size);
2227 		skb_reset_network_header(skb);
2228 	}
2229 
2230 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2231 			RTM_NEWROUTE, sizeof(*r), 0);
2232 	if (!nlh) {
2233 		err = -EMSGSIZE;
2234 		goto errout_free;
2235 	}
2236 
2237 	r = nlmsg_data(nlh);
2238 	r->rtm_family	 = AF_MPLS;
2239 	r->rtm_dst_len	= 20;
2240 	r->rtm_src_len	= 0;
2241 	r->rtm_table	= RT_TABLE_MAIN;
2242 	r->rtm_type	= RTN_UNICAST;
2243 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2244 	r->rtm_protocol = rt->rt_protocol;
2245 	r->rtm_flags	= 0;
2246 
2247 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2248 		goto nla_put_failure;
2249 
2250 	if (nh->nh_labels &&
2251 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2252 			   nh->nh_label))
2253 		goto nla_put_failure;
2254 
2255 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2256 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2257 			nh->nh_via_alen))
2258 		goto nla_put_failure;
2259 	dev = rtnl_dereference(nh->nh_dev);
2260 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2261 		goto nla_put_failure;
2262 
2263 	nlmsg_end(skb, nlh);
2264 
2265 	err = rtnl_unicast(skb, net, portid);
2266 errout:
2267 	return err;
2268 
2269 nla_put_failure:
2270 	nlmsg_cancel(skb, nlh);
2271 	err = -EMSGSIZE;
2272 errout_free:
2273 	kfree_skb(skb);
2274 	return err;
2275 }
2276 
2277 static int resize_platform_label_table(struct net *net, size_t limit)
2278 {
2279 	size_t size = sizeof(struct mpls_route *) * limit;
2280 	size_t old_limit;
2281 	size_t cp_size;
2282 	struct mpls_route __rcu **labels = NULL, **old;
2283 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2284 	unsigned index;
2285 
2286 	if (size) {
2287 		labels = kvzalloc(size, GFP_KERNEL);
2288 		if (!labels)
2289 			goto nolabels;
2290 	}
2291 
2292 	/* In case the predefined labels need to be populated */
2293 	if (limit > MPLS_LABEL_IPV4NULL) {
2294 		struct net_device *lo = net->loopback_dev;
2295 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2296 		if (IS_ERR(rt0))
2297 			goto nort0;
2298 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2299 		rt0->rt_protocol = RTPROT_KERNEL;
2300 		rt0->rt_payload_type = MPT_IPV4;
2301 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2302 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2303 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2304 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2305 		       lo->addr_len);
2306 	}
2307 	if (limit > MPLS_LABEL_IPV6NULL) {
2308 		struct net_device *lo = net->loopback_dev;
2309 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2310 		if (IS_ERR(rt2))
2311 			goto nort2;
2312 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2313 		rt2->rt_protocol = RTPROT_KERNEL;
2314 		rt2->rt_payload_type = MPT_IPV6;
2315 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2316 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2317 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2318 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2319 		       lo->addr_len);
2320 	}
2321 
2322 	rtnl_lock();
2323 	/* Remember the original table */
2324 	old = rtnl_dereference(net->mpls.platform_label);
2325 	old_limit = net->mpls.platform_labels;
2326 
2327 	/* Free any labels beyond the new table */
2328 	for (index = limit; index < old_limit; index++)
2329 		mpls_route_update(net, index, NULL, NULL);
2330 
2331 	/* Copy over the old labels */
2332 	cp_size = size;
2333 	if (old_limit < limit)
2334 		cp_size = old_limit * sizeof(struct mpls_route *);
2335 
2336 	memcpy(labels, old, cp_size);
2337 
2338 	/* If needed set the predefined labels */
2339 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2340 	    (limit > MPLS_LABEL_IPV6NULL)) {
2341 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2342 		rt2 = NULL;
2343 	}
2344 
2345 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2346 	    (limit > MPLS_LABEL_IPV4NULL)) {
2347 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2348 		rt0 = NULL;
2349 	}
2350 
2351 	/* Update the global pointers */
2352 	net->mpls.platform_labels = limit;
2353 	rcu_assign_pointer(net->mpls.platform_label, labels);
2354 
2355 	rtnl_unlock();
2356 
2357 	mpls_rt_free(rt2);
2358 	mpls_rt_free(rt0);
2359 
2360 	if (old) {
2361 		synchronize_rcu();
2362 		kvfree(old);
2363 	}
2364 	return 0;
2365 
2366 nort2:
2367 	mpls_rt_free(rt0);
2368 nort0:
2369 	kvfree(labels);
2370 nolabels:
2371 	return -ENOMEM;
2372 }
2373 
2374 static int mpls_platform_labels(struct ctl_table *table, int write,
2375 				void __user *buffer, size_t *lenp, loff_t *ppos)
2376 {
2377 	struct net *net = table->data;
2378 	int platform_labels = net->mpls.platform_labels;
2379 	int ret;
2380 	struct ctl_table tmp = {
2381 		.procname	= table->procname,
2382 		.data		= &platform_labels,
2383 		.maxlen		= sizeof(int),
2384 		.mode		= table->mode,
2385 		.extra1		= &zero,
2386 		.extra2		= &label_limit,
2387 	};
2388 
2389 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2390 
2391 	if (write && ret == 0)
2392 		ret = resize_platform_label_table(net, platform_labels);
2393 
2394 	return ret;
2395 }
2396 
2397 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2398 	(&((struct net *)0)->field)
2399 
2400 static const struct ctl_table mpls_table[] = {
2401 	{
2402 		.procname	= "platform_labels",
2403 		.data		= NULL,
2404 		.maxlen		= sizeof(int),
2405 		.mode		= 0644,
2406 		.proc_handler	= mpls_platform_labels,
2407 	},
2408 	{
2409 		.procname	= "ip_ttl_propagate",
2410 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2411 		.maxlen		= sizeof(int),
2412 		.mode		= 0644,
2413 		.proc_handler	= proc_dointvec_minmax,
2414 		.extra1		= &zero,
2415 		.extra2		= &one,
2416 	},
2417 	{
2418 		.procname	= "default_ttl",
2419 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2420 		.maxlen		= sizeof(int),
2421 		.mode		= 0644,
2422 		.proc_handler	= proc_dointvec_minmax,
2423 		.extra1		= &one,
2424 		.extra2		= &ttl_max,
2425 	},
2426 	{ }
2427 };
2428 
2429 static int mpls_net_init(struct net *net)
2430 {
2431 	struct ctl_table *table;
2432 	int i;
2433 
2434 	net->mpls.platform_labels = 0;
2435 	net->mpls.platform_label = NULL;
2436 	net->mpls.ip_ttl_propagate = 1;
2437 	net->mpls.default_ttl = 255;
2438 
2439 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2440 	if (table == NULL)
2441 		return -ENOMEM;
2442 
2443 	/* Table data contains only offsets relative to the base of
2444 	 * the mdev at this point, so make them absolute.
2445 	 */
2446 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2447 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2448 
2449 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2450 	if (net->mpls.ctl == NULL) {
2451 		kfree(table);
2452 		return -ENOMEM;
2453 	}
2454 
2455 	return 0;
2456 }
2457 
2458 static void mpls_net_exit(struct net *net)
2459 {
2460 	struct mpls_route __rcu **platform_label;
2461 	size_t platform_labels;
2462 	struct ctl_table *table;
2463 	unsigned int index;
2464 
2465 	table = net->mpls.ctl->ctl_table_arg;
2466 	unregister_net_sysctl_table(net->mpls.ctl);
2467 	kfree(table);
2468 
2469 	/* An rcu grace period has passed since there was a device in
2470 	 * the network namespace (and thus the last in flight packet)
2471 	 * left this network namespace.  This is because
2472 	 * unregister_netdevice_many and netdev_run_todo has completed
2473 	 * for each network device that was in this network namespace.
2474 	 *
2475 	 * As such no additional rcu synchronization is necessary when
2476 	 * freeing the platform_label table.
2477 	 */
2478 	rtnl_lock();
2479 	platform_label = rtnl_dereference(net->mpls.platform_label);
2480 	platform_labels = net->mpls.platform_labels;
2481 	for (index = 0; index < platform_labels; index++) {
2482 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2483 		RCU_INIT_POINTER(platform_label[index], NULL);
2484 		mpls_notify_route(net, index, rt, NULL, NULL);
2485 		mpls_rt_free(rt);
2486 	}
2487 	rtnl_unlock();
2488 
2489 	kvfree(platform_label);
2490 }
2491 
2492 static struct pernet_operations mpls_net_ops = {
2493 	.init = mpls_net_init,
2494 	.exit = mpls_net_exit,
2495 };
2496 
2497 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2498 	.family		   = AF_MPLS,
2499 	.fill_stats_af	   = mpls_fill_stats_af,
2500 	.get_stats_af_size = mpls_get_stats_af_size,
2501 };
2502 
2503 static int __init mpls_init(void)
2504 {
2505 	int err;
2506 
2507 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2508 
2509 	err = register_pernet_subsys(&mpls_net_ops);
2510 	if (err)
2511 		goto out;
2512 
2513 	err = register_netdevice_notifier(&mpls_dev_notifier);
2514 	if (err)
2515 		goto out_unregister_pernet;
2516 
2517 	dev_add_pack(&mpls_packet_type);
2518 
2519 	rtnl_af_register(&mpls_af_ops);
2520 
2521 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2522 			     mpls_rtm_newroute, NULL, 0);
2523 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2524 			     mpls_rtm_delroute, NULL, 0);
2525 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2526 			     mpls_getroute, mpls_dump_routes, 0);
2527 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2528 			     mpls_netconf_get_devconf,
2529 			     mpls_netconf_dump_devconf, 0);
2530 	err = ipgre_tunnel_encap_add_mpls_ops();
2531 	if (err)
2532 		pr_err("Can't add mpls over gre tunnel ops\n");
2533 
2534 	err = 0;
2535 out:
2536 	return err;
2537 
2538 out_unregister_pernet:
2539 	unregister_pernet_subsys(&mpls_net_ops);
2540 	goto out;
2541 }
2542 module_init(mpls_init);
2543 
2544 static void __exit mpls_exit(void)
2545 {
2546 	rtnl_unregister_all(PF_MPLS);
2547 	rtnl_af_unregister(&mpls_af_ops);
2548 	dev_remove_pack(&mpls_packet_type);
2549 	unregister_netdevice_notifier(&mpls_dev_notifier);
2550 	unregister_pernet_subsys(&mpls_net_ops);
2551 	ipgre_tunnel_encap_del_mpls_ops();
2552 }
2553 module_exit(mpls_exit);
2554 
2555 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2556 MODULE_LICENSE("GPL v2");
2557 MODULE_ALIAS_NETPROTO(PF_MPLS);
2558