xref: /openbmc/linux/net/mpls/af_mpls.c (revision ae213c44)
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/ip.h>
16 #include <net/dst.h>
17 #include <net/sock.h>
18 #include <net/arp.h>
19 #include <net/ip_fib.h>
20 #include <net/netevent.h>
21 #include <net/ip_tunnels.h>
22 #include <net/netns/generic.h>
23 #if IS_ENABLED(CONFIG_IPV6)
24 #include <net/ipv6.h>
25 #endif
26 #include <net/ipv6_stubs.h>
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 zero = 0;
41 static int one = 1;
42 static int label_limit = (1 << 20) - 1;
43 static int ttl_max = 255;
44 
45 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
46 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
47 {
48 	return sizeof(struct mpls_shim_hdr);
49 }
50 
51 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
52 	.encap_hlen	= ipgre_mpls_encap_hlen,
53 };
54 
55 static int ipgre_tunnel_encap_add_mpls_ops(void)
56 {
57 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
58 }
59 
60 static void ipgre_tunnel_encap_del_mpls_ops(void)
61 {
62 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
63 }
64 #else
65 static int ipgre_tunnel_encap_add_mpls_ops(void)
66 {
67 	return 0;
68 }
69 
70 static void ipgre_tunnel_encap_del_mpls_ops(void)
71 {
72 }
73 #endif
74 
75 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
76 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
77 		       unsigned int nlm_flags);
78 
79 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
80 {
81 	struct mpls_route *rt = NULL;
82 
83 	if (index < net->mpls.platform_labels) {
84 		struct mpls_route __rcu **platform_label =
85 			rcu_dereference(net->mpls.platform_label);
86 		rt = rcu_dereference(platform_label[index]);
87 	}
88 	return rt;
89 }
90 
91 bool mpls_output_possible(const struct net_device *dev)
92 {
93 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
94 }
95 EXPORT_SYMBOL_GPL(mpls_output_possible);
96 
97 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
98 {
99 	return (u8 *)nh + rt->rt_via_offset;
100 }
101 
102 static const u8 *mpls_nh_via(const struct mpls_route *rt,
103 			     const struct mpls_nh *nh)
104 {
105 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
106 }
107 
108 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
109 {
110 	/* The size of the layer 2.5 labels to be added for this route */
111 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
112 }
113 
114 unsigned int mpls_dev_mtu(const struct net_device *dev)
115 {
116 	/* The amount of data the layer 2 frame can hold */
117 	return dev->mtu;
118 }
119 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
120 
121 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
122 {
123 	if (skb->len <= mtu)
124 		return false;
125 
126 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
127 		return false;
128 
129 	return true;
130 }
131 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
132 
133 void mpls_stats_inc_outucastpkts(struct net_device *dev,
134 				 const struct sk_buff *skb)
135 {
136 	struct mpls_dev *mdev;
137 
138 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
139 		mdev = mpls_dev_get(dev);
140 		if (mdev)
141 			MPLS_INC_STATS_LEN(mdev, skb->len,
142 					   tx_packets,
143 					   tx_bytes);
144 	} else if (skb->protocol == htons(ETH_P_IP)) {
145 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
146 #if IS_ENABLED(CONFIG_IPV6)
147 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
148 		struct inet6_dev *in6dev = __in6_dev_get(dev);
149 
150 		if (in6dev)
151 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
152 					 IPSTATS_MIB_OUT, skb->len);
153 #endif
154 	}
155 }
156 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
157 
158 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
159 {
160 	struct mpls_entry_decoded dec;
161 	unsigned int mpls_hdr_len = 0;
162 	struct mpls_shim_hdr *hdr;
163 	bool eli_seen = false;
164 	int label_index;
165 	u32 hash = 0;
166 
167 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
168 	     label_index++) {
169 		mpls_hdr_len += sizeof(*hdr);
170 		if (!pskb_may_pull(skb, mpls_hdr_len))
171 			break;
172 
173 		/* Read and decode the current label */
174 		hdr = mpls_hdr(skb) + label_index;
175 		dec = mpls_entry_decode(hdr);
176 
177 		/* RFC6790 - reserved labels MUST NOT be used as keys
178 		 * for the load-balancing function
179 		 */
180 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
181 			hash = jhash_1word(dec.label, hash);
182 
183 			/* The entropy label follows the entropy label
184 			 * indicator, so this means that the entropy
185 			 * label was just added to the hash - no need to
186 			 * go any deeper either in the label stack or in the
187 			 * payload
188 			 */
189 			if (eli_seen)
190 				break;
191 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
192 			eli_seen = true;
193 		}
194 
195 		if (!dec.bos)
196 			continue;
197 
198 		/* found bottom label; does skb have room for a header? */
199 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
200 			const struct iphdr *v4hdr;
201 
202 			v4hdr = (const struct iphdr *)(hdr + 1);
203 			if (v4hdr->version == 4) {
204 				hash = jhash_3words(ntohl(v4hdr->saddr),
205 						    ntohl(v4hdr->daddr),
206 						    v4hdr->protocol, hash);
207 			} else if (v4hdr->version == 6 &&
208 				   pskb_may_pull(skb, mpls_hdr_len +
209 						 sizeof(struct ipv6hdr))) {
210 				const struct ipv6hdr *v6hdr;
211 
212 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
213 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
214 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
215 				hash = jhash_1word(v6hdr->nexthdr, hash);
216 			}
217 		}
218 
219 		break;
220 	}
221 
222 	return hash;
223 }
224 
225 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
226 {
227 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
228 }
229 
230 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
231  * a next hop (nh->nh_flags) are modified by netdev event handlers.
232  * Since those fields can change at any moment, use READ_ONCE to
233  * access both.
234  */
235 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
236 					     struct sk_buff *skb)
237 {
238 	u32 hash = 0;
239 	int nh_index = 0;
240 	int n = 0;
241 	u8 alive;
242 
243 	/* No need to look further into packet if there's only
244 	 * one path
245 	 */
246 	if (rt->rt_nhn == 1)
247 		return rt->rt_nh;
248 
249 	alive = READ_ONCE(rt->rt_nhn_alive);
250 	if (alive == 0)
251 		return NULL;
252 
253 	hash = mpls_multipath_hash(rt, skb);
254 	nh_index = hash % alive;
255 	if (alive == rt->rt_nhn)
256 		goto out;
257 	for_nexthops(rt) {
258 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
259 
260 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
261 			continue;
262 		if (n == nh_index)
263 			return nh;
264 		n++;
265 	} endfor_nexthops(rt);
266 
267 out:
268 	return mpls_get_nexthop(rt, nh_index);
269 }
270 
271 static bool mpls_egress(struct net *net, struct mpls_route *rt,
272 			struct sk_buff *skb, struct mpls_entry_decoded dec)
273 {
274 	enum mpls_payload_type payload_type;
275 	bool success = false;
276 
277 	/* The IPv4 code below accesses through the IPv4 header
278 	 * checksum, which is 12 bytes into the packet.
279 	 * The IPv6 code below accesses through the IPv6 hop limit
280 	 * which is 8 bytes into the packet.
281 	 *
282 	 * For all supported cases there should always be at least 12
283 	 * bytes of packet data present.  The IPv4 header is 20 bytes
284 	 * without options and the IPv6 header is always 40 bytes
285 	 * long.
286 	 */
287 	if (!pskb_may_pull(skb, 12))
288 		return false;
289 
290 	payload_type = rt->rt_payload_type;
291 	if (payload_type == MPT_UNSPEC)
292 		payload_type = ip_hdr(skb)->version;
293 
294 	switch (payload_type) {
295 	case MPT_IPV4: {
296 		struct iphdr *hdr4 = ip_hdr(skb);
297 		u8 new_ttl;
298 		skb->protocol = htons(ETH_P_IP);
299 
300 		/* If propagating TTL, take the decremented TTL from
301 		 * the incoming MPLS header, otherwise decrement the
302 		 * TTL, but only if not 0 to avoid underflow.
303 		 */
304 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
305 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
306 		     net->mpls.ip_ttl_propagate))
307 			new_ttl = dec.ttl;
308 		else
309 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
310 
311 		csum_replace2(&hdr4->check,
312 			      htons(hdr4->ttl << 8),
313 			      htons(new_ttl << 8));
314 		hdr4->ttl = new_ttl;
315 		success = true;
316 		break;
317 	}
318 	case MPT_IPV6: {
319 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
320 		skb->protocol = htons(ETH_P_IPV6);
321 
322 		/* If propagating TTL, take the decremented TTL from
323 		 * the incoming MPLS header, otherwise decrement the
324 		 * hop limit, 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 			hdr6->hop_limit = dec.ttl;
330 		else if (hdr6->hop_limit)
331 			hdr6->hop_limit = hdr6->hop_limit - 1;
332 		success = true;
333 		break;
334 	}
335 	case MPT_UNSPEC:
336 		/* Should have decided which protocol it is by now */
337 		break;
338 	}
339 
340 	return success;
341 }
342 
343 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
344 			struct packet_type *pt, struct net_device *orig_dev)
345 {
346 	struct net *net = dev_net(dev);
347 	struct mpls_shim_hdr *hdr;
348 	struct mpls_route *rt;
349 	struct mpls_nh *nh;
350 	struct mpls_entry_decoded dec;
351 	struct net_device *out_dev;
352 	struct mpls_dev *out_mdev;
353 	struct mpls_dev *mdev;
354 	unsigned int hh_len;
355 	unsigned int new_header_size;
356 	unsigned int mtu;
357 	int err;
358 
359 	/* Careful this entire function runs inside of an rcu critical section */
360 
361 	mdev = mpls_dev_get(dev);
362 	if (!mdev)
363 		goto drop;
364 
365 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
366 			   rx_bytes);
367 
368 	if (!mdev->input_enabled) {
369 		MPLS_INC_STATS(mdev, rx_dropped);
370 		goto drop;
371 	}
372 
373 	if (skb->pkt_type != PACKET_HOST)
374 		goto err;
375 
376 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
377 		goto err;
378 
379 	if (!pskb_may_pull(skb, sizeof(*hdr)))
380 		goto err;
381 
382 	/* Read and decode the label */
383 	hdr = mpls_hdr(skb);
384 	dec = mpls_entry_decode(hdr);
385 
386 	rt = mpls_route_input_rcu(net, dec.label);
387 	if (!rt) {
388 		MPLS_INC_STATS(mdev, rx_noroute);
389 		goto drop;
390 	}
391 
392 	nh = mpls_select_multipath(rt, skb);
393 	if (!nh)
394 		goto err;
395 
396 	/* Pop the label */
397 	skb_pull(skb, sizeof(*hdr));
398 	skb_reset_network_header(skb);
399 
400 	skb_orphan(skb);
401 
402 	if (skb_warn_if_lro(skb))
403 		goto err;
404 
405 	skb_forward_csum(skb);
406 
407 	/* Verify ttl is valid */
408 	if (dec.ttl <= 1)
409 		goto err;
410 	dec.ttl -= 1;
411 
412 	/* Find the output device */
413 	out_dev = rcu_dereference(nh->nh_dev);
414 	if (!mpls_output_possible(out_dev))
415 		goto tx_err;
416 
417 	/* Verify the destination can hold the packet */
418 	new_header_size = mpls_nh_header_size(nh);
419 	mtu = mpls_dev_mtu(out_dev);
420 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
421 		goto tx_err;
422 
423 	hh_len = LL_RESERVED_SPACE(out_dev);
424 	if (!out_dev->header_ops)
425 		hh_len = 0;
426 
427 	/* Ensure there is enough space for the headers in the skb */
428 	if (skb_cow(skb, hh_len + new_header_size))
429 		goto tx_err;
430 
431 	skb->dev = out_dev;
432 	skb->protocol = htons(ETH_P_MPLS_UC);
433 
434 	if (unlikely(!new_header_size && dec.bos)) {
435 		/* Penultimate hop popping */
436 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
437 			goto err;
438 	} else {
439 		bool bos;
440 		int i;
441 		skb_push(skb, new_header_size);
442 		skb_reset_network_header(skb);
443 		/* Push the new labels */
444 		hdr = mpls_hdr(skb);
445 		bos = dec.bos;
446 		for (i = nh->nh_labels - 1; i >= 0; i--) {
447 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
448 						   dec.ttl, 0, bos);
449 			bos = false;
450 		}
451 	}
452 
453 	mpls_stats_inc_outucastpkts(out_dev, skb);
454 
455 	/* If via wasn't specified then send out using device address */
456 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
457 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
458 				 out_dev->dev_addr, skb);
459 	else
460 		err = neigh_xmit(nh->nh_via_table, out_dev,
461 				 mpls_nh_via(rt, nh), skb);
462 	if (err)
463 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
464 				    __func__, err);
465 	return 0;
466 
467 tx_err:
468 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
469 	if (out_mdev)
470 		MPLS_INC_STATS(out_mdev, tx_errors);
471 	goto drop;
472 err:
473 	MPLS_INC_STATS(mdev, rx_errors);
474 drop:
475 	kfree_skb(skb);
476 	return NET_RX_DROP;
477 }
478 
479 static struct packet_type mpls_packet_type __read_mostly = {
480 	.type = cpu_to_be16(ETH_P_MPLS_UC),
481 	.func = mpls_forward,
482 };
483 
484 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
485 	[RTA_DST]		= { .type = NLA_U32 },
486 	[RTA_OIF]		= { .type = NLA_U32 },
487 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
488 };
489 
490 struct mpls_route_config {
491 	u32			rc_protocol;
492 	u32			rc_ifindex;
493 	u8			rc_via_table;
494 	u8			rc_via_alen;
495 	u8			rc_via[MAX_VIA_ALEN];
496 	u32			rc_label;
497 	u8			rc_ttl_propagate;
498 	u8			rc_output_labels;
499 	u32			rc_output_label[MAX_NEW_LABELS];
500 	u32			rc_nlflags;
501 	enum mpls_payload_type	rc_payload_type;
502 	struct nl_info		rc_nlinfo;
503 	struct rtnexthop	*rc_mp;
504 	int			rc_mp_len;
505 };
506 
507 /* all nexthops within a route have the same size based on max
508  * number of labels and max via length for a hop
509  */
510 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
511 {
512 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
513 	struct mpls_route *rt;
514 	size_t size;
515 
516 	size = sizeof(*rt) + num_nh * nh_size;
517 	if (size > MAX_MPLS_ROUTE_MEM)
518 		return ERR_PTR(-EINVAL);
519 
520 	rt = kzalloc(size, GFP_KERNEL);
521 	if (!rt)
522 		return ERR_PTR(-ENOMEM);
523 
524 	rt->rt_nhn = num_nh;
525 	rt->rt_nhn_alive = num_nh;
526 	rt->rt_nh_size = nh_size;
527 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
528 
529 	return rt;
530 }
531 
532 static void mpls_rt_free(struct mpls_route *rt)
533 {
534 	if (rt)
535 		kfree_rcu(rt, rt_rcu);
536 }
537 
538 static void mpls_notify_route(struct net *net, unsigned index,
539 			      struct mpls_route *old, struct mpls_route *new,
540 			      const struct nl_info *info)
541 {
542 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
543 	unsigned portid = info ? info->portid : 0;
544 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
545 	struct mpls_route *rt = new ? new : old;
546 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
547 	/* Ignore reserved labels for now */
548 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
549 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
550 }
551 
552 static void mpls_route_update(struct net *net, unsigned index,
553 			      struct mpls_route *new,
554 			      const struct nl_info *info)
555 {
556 	struct mpls_route __rcu **platform_label;
557 	struct mpls_route *rt;
558 
559 	ASSERT_RTNL();
560 
561 	platform_label = rtnl_dereference(net->mpls.platform_label);
562 	rt = rtnl_dereference(platform_label[index]);
563 	rcu_assign_pointer(platform_label[index], new);
564 
565 	mpls_notify_route(net, index, rt, new, info);
566 
567 	/* If we removed a route free it now */
568 	mpls_rt_free(rt);
569 }
570 
571 static unsigned find_free_label(struct net *net)
572 {
573 	struct mpls_route __rcu **platform_label;
574 	size_t platform_labels;
575 	unsigned index;
576 
577 	platform_label = rtnl_dereference(net->mpls.platform_label);
578 	platform_labels = net->mpls.platform_labels;
579 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
580 	     index++) {
581 		if (!rtnl_dereference(platform_label[index]))
582 			return index;
583 	}
584 	return LABEL_NOT_SPECIFIED;
585 }
586 
587 #if IS_ENABLED(CONFIG_INET)
588 static struct net_device *inet_fib_lookup_dev(struct net *net,
589 					      const void *addr)
590 {
591 	struct net_device *dev;
592 	struct rtable *rt;
593 	struct in_addr daddr;
594 
595 	memcpy(&daddr, addr, sizeof(struct in_addr));
596 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
597 	if (IS_ERR(rt))
598 		return ERR_CAST(rt);
599 
600 	dev = rt->dst.dev;
601 	dev_hold(dev);
602 
603 	ip_rt_put(rt);
604 
605 	return dev;
606 }
607 #else
608 static struct net_device *inet_fib_lookup_dev(struct net *net,
609 					      const void *addr)
610 {
611 	return ERR_PTR(-EAFNOSUPPORT);
612 }
613 #endif
614 
615 #if IS_ENABLED(CONFIG_IPV6)
616 static struct net_device *inet6_fib_lookup_dev(struct net *net,
617 					       const void *addr)
618 {
619 	struct net_device *dev;
620 	struct dst_entry *dst;
621 	struct flowi6 fl6;
622 	int err;
623 
624 	if (!ipv6_stub)
625 		return ERR_PTR(-EAFNOSUPPORT);
626 
627 	memset(&fl6, 0, sizeof(fl6));
628 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
629 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
630 	if (err)
631 		return ERR_PTR(err);
632 
633 	dev = dst->dev;
634 	dev_hold(dev);
635 	dst_release(dst);
636 
637 	return dev;
638 }
639 #else
640 static struct net_device *inet6_fib_lookup_dev(struct net *net,
641 					       const void *addr)
642 {
643 	return ERR_PTR(-EAFNOSUPPORT);
644 }
645 #endif
646 
647 static struct net_device *find_outdev(struct net *net,
648 				      struct mpls_route *rt,
649 				      struct mpls_nh *nh, int oif)
650 {
651 	struct net_device *dev = NULL;
652 
653 	if (!oif) {
654 		switch (nh->nh_via_table) {
655 		case NEIGH_ARP_TABLE:
656 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
657 			break;
658 		case NEIGH_ND_TABLE:
659 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
660 			break;
661 		case NEIGH_LINK_TABLE:
662 			break;
663 		}
664 	} else {
665 		dev = dev_get_by_index(net, oif);
666 	}
667 
668 	if (!dev)
669 		return ERR_PTR(-ENODEV);
670 
671 	if (IS_ERR(dev))
672 		return dev;
673 
674 	/* The caller is holding rtnl anyways, so release the dev reference */
675 	dev_put(dev);
676 
677 	return dev;
678 }
679 
680 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
681 			      struct mpls_nh *nh, int oif)
682 {
683 	struct net_device *dev = NULL;
684 	int err = -ENODEV;
685 
686 	dev = find_outdev(net, rt, nh, oif);
687 	if (IS_ERR(dev)) {
688 		err = PTR_ERR(dev);
689 		dev = NULL;
690 		goto errout;
691 	}
692 
693 	/* Ensure this is a supported device */
694 	err = -EINVAL;
695 	if (!mpls_dev_get(dev))
696 		goto errout;
697 
698 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
699 	    (dev->addr_len != nh->nh_via_alen))
700 		goto errout;
701 
702 	RCU_INIT_POINTER(nh->nh_dev, dev);
703 
704 	if (!(dev->flags & IFF_UP)) {
705 		nh->nh_flags |= RTNH_F_DEAD;
706 	} else {
707 		unsigned int flags;
708 
709 		flags = dev_get_flags(dev);
710 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
711 			nh->nh_flags |= RTNH_F_LINKDOWN;
712 	}
713 
714 	return 0;
715 
716 errout:
717 	return err;
718 }
719 
720 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
721 		       u8 via_addr[], struct netlink_ext_ack *extack)
722 {
723 	struct rtvia *via = nla_data(nla);
724 	int err = -EINVAL;
725 	int alen;
726 
727 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
728 		NL_SET_ERR_MSG_ATTR(extack, nla,
729 				    "Invalid attribute length for RTA_VIA");
730 		goto errout;
731 	}
732 	alen = nla_len(nla) -
733 			offsetof(struct rtvia, rtvia_addr);
734 	if (alen > MAX_VIA_ALEN) {
735 		NL_SET_ERR_MSG_ATTR(extack, nla,
736 				    "Invalid address length for RTA_VIA");
737 		goto errout;
738 	}
739 
740 	/* Validate the address family */
741 	switch (via->rtvia_family) {
742 	case AF_PACKET:
743 		*via_table = NEIGH_LINK_TABLE;
744 		break;
745 	case AF_INET:
746 		*via_table = NEIGH_ARP_TABLE;
747 		if (alen != 4)
748 			goto errout;
749 		break;
750 	case AF_INET6:
751 		*via_table = NEIGH_ND_TABLE;
752 		if (alen != 16)
753 			goto errout;
754 		break;
755 	default:
756 		/* Unsupported address family */
757 		goto errout;
758 	}
759 
760 	memcpy(via_addr, via->rtvia_addr, alen);
761 	*via_alen = alen;
762 	err = 0;
763 
764 errout:
765 	return err;
766 }
767 
768 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
769 				  struct mpls_route *rt)
770 {
771 	struct net *net = cfg->rc_nlinfo.nl_net;
772 	struct mpls_nh *nh = rt->rt_nh;
773 	int err;
774 	int i;
775 
776 	if (!nh)
777 		return -ENOMEM;
778 
779 	nh->nh_labels = cfg->rc_output_labels;
780 	for (i = 0; i < nh->nh_labels; i++)
781 		nh->nh_label[i] = cfg->rc_output_label[i];
782 
783 	nh->nh_via_table = cfg->rc_via_table;
784 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
785 	nh->nh_via_alen = cfg->rc_via_alen;
786 
787 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
788 	if (err)
789 		goto errout;
790 
791 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
792 		rt->rt_nhn_alive--;
793 
794 	return 0;
795 
796 errout:
797 	return err;
798 }
799 
800 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
801 			 struct mpls_nh *nh, int oif, struct nlattr *via,
802 			 struct nlattr *newdst, u8 max_labels,
803 			 struct netlink_ext_ack *extack)
804 {
805 	int err = -ENOMEM;
806 
807 	if (!nh)
808 		goto errout;
809 
810 	if (newdst) {
811 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
812 				     nh->nh_label, extack);
813 		if (err)
814 			goto errout;
815 	}
816 
817 	if (via) {
818 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
819 				  __mpls_nh_via(rt, nh), extack);
820 		if (err)
821 			goto errout;
822 	} else {
823 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
824 	}
825 
826 	err = mpls_nh_assign_dev(net, rt, nh, oif);
827 	if (err)
828 		goto errout;
829 
830 	return 0;
831 
832 errout:
833 	return err;
834 }
835 
836 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
837 			      u8 cfg_via_alen, u8 *max_via_alen,
838 			      u8 *max_labels)
839 {
840 	int remaining = len;
841 	u8 nhs = 0;
842 
843 	*max_via_alen = 0;
844 	*max_labels = 0;
845 
846 	while (rtnh_ok(rtnh, remaining)) {
847 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
848 		int attrlen;
849 		u8 n_labels = 0;
850 
851 		attrlen = rtnh_attrlen(rtnh);
852 		nla = nla_find(attrs, attrlen, RTA_VIA);
853 		if (nla && nla_len(nla) >=
854 		    offsetof(struct rtvia, rtvia_addr)) {
855 			int via_alen = nla_len(nla) -
856 				offsetof(struct rtvia, rtvia_addr);
857 
858 			if (via_alen <= MAX_VIA_ALEN)
859 				*max_via_alen = max_t(u16, *max_via_alen,
860 						      via_alen);
861 		}
862 
863 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
864 		if (nla &&
865 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
866 				   NULL, NULL) != 0)
867 			return 0;
868 
869 		*max_labels = max_t(u8, *max_labels, n_labels);
870 
871 		/* number of nexthops is tracked by a u8.
872 		 * Check for overflow.
873 		 */
874 		if (nhs == 255)
875 			return 0;
876 		nhs++;
877 
878 		rtnh = rtnh_next(rtnh, &remaining);
879 	}
880 
881 	/* leftover implies invalid nexthop configuration, discard it */
882 	return remaining > 0 ? 0 : nhs;
883 }
884 
885 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
886 			       struct mpls_route *rt, u8 max_labels,
887 			       struct netlink_ext_ack *extack)
888 {
889 	struct rtnexthop *rtnh = cfg->rc_mp;
890 	struct nlattr *nla_via, *nla_newdst;
891 	int remaining = cfg->rc_mp_len;
892 	int err = 0;
893 	u8 nhs = 0;
894 
895 	change_nexthops(rt) {
896 		int attrlen;
897 
898 		nla_via = NULL;
899 		nla_newdst = NULL;
900 
901 		err = -EINVAL;
902 		if (!rtnh_ok(rtnh, remaining))
903 			goto errout;
904 
905 		/* neither weighted multipath nor any flags
906 		 * are supported
907 		 */
908 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
909 			goto errout;
910 
911 		attrlen = rtnh_attrlen(rtnh);
912 		if (attrlen > 0) {
913 			struct nlattr *attrs = rtnh_attrs(rtnh);
914 
915 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
916 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
917 		}
918 
919 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
920 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
921 				    max_labels, extack);
922 		if (err)
923 			goto errout;
924 
925 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
926 			rt->rt_nhn_alive--;
927 
928 		rtnh = rtnh_next(rtnh, &remaining);
929 		nhs++;
930 	} endfor_nexthops(rt);
931 
932 	rt->rt_nhn = nhs;
933 
934 	return 0;
935 
936 errout:
937 	return err;
938 }
939 
940 static bool mpls_label_ok(struct net *net, unsigned int *index,
941 			  struct netlink_ext_ack *extack)
942 {
943 	bool is_ok = true;
944 
945 	/* Reserved labels may not be set */
946 	if (*index < MPLS_LABEL_FIRST_UNRESERVED) {
947 		NL_SET_ERR_MSG(extack,
948 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
949 		is_ok = false;
950 	}
951 
952 	/* The full 20 bit range may not be supported. */
953 	if (is_ok && *index >= net->mpls.platform_labels) {
954 		NL_SET_ERR_MSG(extack,
955 			       "Label >= configured maximum in platform_labels");
956 		is_ok = false;
957 	}
958 
959 	*index = array_index_nospec(*index, net->mpls.platform_labels);
960 	return is_ok;
961 }
962 
963 static int mpls_route_add(struct mpls_route_config *cfg,
964 			  struct netlink_ext_ack *extack)
965 {
966 	struct mpls_route __rcu **platform_label;
967 	struct net *net = cfg->rc_nlinfo.nl_net;
968 	struct mpls_route *rt, *old;
969 	int err = -EINVAL;
970 	u8 max_via_alen;
971 	unsigned index;
972 	u8 max_labels;
973 	u8 nhs;
974 
975 	index = cfg->rc_label;
976 
977 	/* If a label was not specified during insert pick one */
978 	if ((index == LABEL_NOT_SPECIFIED) &&
979 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
980 		index = find_free_label(net);
981 	}
982 
983 	if (!mpls_label_ok(net, &index, extack))
984 		goto errout;
985 
986 	/* Append makes no sense with mpls */
987 	err = -EOPNOTSUPP;
988 	if (cfg->rc_nlflags & NLM_F_APPEND) {
989 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
990 		goto errout;
991 	}
992 
993 	err = -EEXIST;
994 	platform_label = rtnl_dereference(net->mpls.platform_label);
995 	old = rtnl_dereference(platform_label[index]);
996 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
997 		goto errout;
998 
999 	err = -EEXIST;
1000 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
1001 		goto errout;
1002 
1003 	err = -ENOENT;
1004 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1005 		goto errout;
1006 
1007 	err = -EINVAL;
1008 	if (cfg->rc_mp) {
1009 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1010 					  cfg->rc_via_alen, &max_via_alen,
1011 					  &max_labels);
1012 	} else {
1013 		max_via_alen = cfg->rc_via_alen;
1014 		max_labels = cfg->rc_output_labels;
1015 		nhs = 1;
1016 	}
1017 
1018 	if (nhs == 0) {
1019 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1020 		goto errout;
1021 	}
1022 
1023 	err = -ENOMEM;
1024 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1025 	if (IS_ERR(rt)) {
1026 		err = PTR_ERR(rt);
1027 		goto errout;
1028 	}
1029 
1030 	rt->rt_protocol = cfg->rc_protocol;
1031 	rt->rt_payload_type = cfg->rc_payload_type;
1032 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1033 
1034 	if (cfg->rc_mp)
1035 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1036 	else
1037 		err = mpls_nh_build_from_cfg(cfg, rt);
1038 	if (err)
1039 		goto freert;
1040 
1041 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1042 
1043 	return 0;
1044 
1045 freert:
1046 	mpls_rt_free(rt);
1047 errout:
1048 	return err;
1049 }
1050 
1051 static int mpls_route_del(struct mpls_route_config *cfg,
1052 			  struct netlink_ext_ack *extack)
1053 {
1054 	struct net *net = cfg->rc_nlinfo.nl_net;
1055 	unsigned index;
1056 	int err = -EINVAL;
1057 
1058 	index = cfg->rc_label;
1059 
1060 	if (!mpls_label_ok(net, &index, extack))
1061 		goto errout;
1062 
1063 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1064 
1065 	err = 0;
1066 errout:
1067 	return err;
1068 }
1069 
1070 static void mpls_get_stats(struct mpls_dev *mdev,
1071 			   struct mpls_link_stats *stats)
1072 {
1073 	struct mpls_pcpu_stats *p;
1074 	int i;
1075 
1076 	memset(stats, 0, sizeof(*stats));
1077 
1078 	for_each_possible_cpu(i) {
1079 		struct mpls_link_stats local;
1080 		unsigned int start;
1081 
1082 		p = per_cpu_ptr(mdev->stats, i);
1083 		do {
1084 			start = u64_stats_fetch_begin(&p->syncp);
1085 			local = p->stats;
1086 		} while (u64_stats_fetch_retry(&p->syncp, start));
1087 
1088 		stats->rx_packets	+= local.rx_packets;
1089 		stats->rx_bytes		+= local.rx_bytes;
1090 		stats->tx_packets	+= local.tx_packets;
1091 		stats->tx_bytes		+= local.tx_bytes;
1092 		stats->rx_errors	+= local.rx_errors;
1093 		stats->tx_errors	+= local.tx_errors;
1094 		stats->rx_dropped	+= local.rx_dropped;
1095 		stats->tx_dropped	+= local.tx_dropped;
1096 		stats->rx_noroute	+= local.rx_noroute;
1097 	}
1098 }
1099 
1100 static int mpls_fill_stats_af(struct sk_buff *skb,
1101 			      const struct net_device *dev)
1102 {
1103 	struct mpls_link_stats *stats;
1104 	struct mpls_dev *mdev;
1105 	struct nlattr *nla;
1106 
1107 	mdev = mpls_dev_get(dev);
1108 	if (!mdev)
1109 		return -ENODATA;
1110 
1111 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1112 				sizeof(struct mpls_link_stats),
1113 				MPLS_STATS_UNSPEC);
1114 	if (!nla)
1115 		return -EMSGSIZE;
1116 
1117 	stats = nla_data(nla);
1118 	mpls_get_stats(mdev, stats);
1119 
1120 	return 0;
1121 }
1122 
1123 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1124 {
1125 	struct mpls_dev *mdev;
1126 
1127 	mdev = mpls_dev_get(dev);
1128 	if (!mdev)
1129 		return 0;
1130 
1131 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1132 }
1133 
1134 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1135 				     u32 portid, u32 seq, int event,
1136 				     unsigned int flags, int type)
1137 {
1138 	struct nlmsghdr  *nlh;
1139 	struct netconfmsg *ncm;
1140 	bool all = false;
1141 
1142 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1143 			flags);
1144 	if (!nlh)
1145 		return -EMSGSIZE;
1146 
1147 	if (type == NETCONFA_ALL)
1148 		all = true;
1149 
1150 	ncm = nlmsg_data(nlh);
1151 	ncm->ncm_family = AF_MPLS;
1152 
1153 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1154 		goto nla_put_failure;
1155 
1156 	if ((all || type == NETCONFA_INPUT) &&
1157 	    nla_put_s32(skb, NETCONFA_INPUT,
1158 			mdev->input_enabled) < 0)
1159 		goto nla_put_failure;
1160 
1161 	nlmsg_end(skb, nlh);
1162 	return 0;
1163 
1164 nla_put_failure:
1165 	nlmsg_cancel(skb, nlh);
1166 	return -EMSGSIZE;
1167 }
1168 
1169 static int mpls_netconf_msgsize_devconf(int type)
1170 {
1171 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1172 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1173 	bool all = false;
1174 
1175 	if (type == NETCONFA_ALL)
1176 		all = true;
1177 
1178 	if (all || type == NETCONFA_INPUT)
1179 		size += nla_total_size(4);
1180 
1181 	return size;
1182 }
1183 
1184 static void mpls_netconf_notify_devconf(struct net *net, int event,
1185 					int type, struct mpls_dev *mdev)
1186 {
1187 	struct sk_buff *skb;
1188 	int err = -ENOBUFS;
1189 
1190 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1191 	if (!skb)
1192 		goto errout;
1193 
1194 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1195 	if (err < 0) {
1196 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1197 		WARN_ON(err == -EMSGSIZE);
1198 		kfree_skb(skb);
1199 		goto errout;
1200 	}
1201 
1202 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1203 	return;
1204 errout:
1205 	if (err < 0)
1206 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1207 }
1208 
1209 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1210 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1211 };
1212 
1213 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1214 				      const struct nlmsghdr *nlh,
1215 				      struct nlattr **tb,
1216 				      struct netlink_ext_ack *extack)
1217 {
1218 	int i, err;
1219 
1220 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1221 		NL_SET_ERR_MSG_MOD(extack,
1222 				   "Invalid header for netconf get request");
1223 		return -EINVAL;
1224 	}
1225 
1226 	if (!netlink_strict_get_check(skb))
1227 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1228 					      tb, NETCONFA_MAX,
1229 					      devconf_mpls_policy, extack);
1230 
1231 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1232 					    tb, NETCONFA_MAX,
1233 					    devconf_mpls_policy, extack);
1234 	if (err)
1235 		return err;
1236 
1237 	for (i = 0; i <= NETCONFA_MAX; i++) {
1238 		if (!tb[i])
1239 			continue;
1240 
1241 		switch (i) {
1242 		case NETCONFA_IFINDEX:
1243 			break;
1244 		default:
1245 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1246 			return -EINVAL;
1247 		}
1248 	}
1249 
1250 	return 0;
1251 }
1252 
1253 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1254 				    struct nlmsghdr *nlh,
1255 				    struct netlink_ext_ack *extack)
1256 {
1257 	struct net *net = sock_net(in_skb->sk);
1258 	struct nlattr *tb[NETCONFA_MAX + 1];
1259 	struct net_device *dev;
1260 	struct mpls_dev *mdev;
1261 	struct sk_buff *skb;
1262 	int ifindex;
1263 	int err;
1264 
1265 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1266 	if (err < 0)
1267 		goto errout;
1268 
1269 	err = -EINVAL;
1270 	if (!tb[NETCONFA_IFINDEX])
1271 		goto errout;
1272 
1273 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1274 	dev = __dev_get_by_index(net, ifindex);
1275 	if (!dev)
1276 		goto errout;
1277 
1278 	mdev = mpls_dev_get(dev);
1279 	if (!mdev)
1280 		goto errout;
1281 
1282 	err = -ENOBUFS;
1283 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1284 	if (!skb)
1285 		goto errout;
1286 
1287 	err = mpls_netconf_fill_devconf(skb, mdev,
1288 					NETLINK_CB(in_skb).portid,
1289 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1290 					NETCONFA_ALL);
1291 	if (err < 0) {
1292 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1293 		WARN_ON(err == -EMSGSIZE);
1294 		kfree_skb(skb);
1295 		goto errout;
1296 	}
1297 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1298 errout:
1299 	return err;
1300 }
1301 
1302 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1303 				     struct netlink_callback *cb)
1304 {
1305 	const struct nlmsghdr *nlh = cb->nlh;
1306 	struct net *net = sock_net(skb->sk);
1307 	struct hlist_head *head;
1308 	struct net_device *dev;
1309 	struct mpls_dev *mdev;
1310 	int idx, s_idx;
1311 	int h, s_h;
1312 
1313 	if (cb->strict_check) {
1314 		struct netlink_ext_ack *extack = cb->extack;
1315 		struct netconfmsg *ncm;
1316 
1317 		if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ncm))) {
1318 			NL_SET_ERR_MSG_MOD(extack, "Invalid header for netconf dump request");
1319 			return -EINVAL;
1320 		}
1321 
1322 		if (nlmsg_attrlen(nlh, sizeof(*ncm))) {
1323 			NL_SET_ERR_MSG_MOD(extack, "Invalid data after header in netconf dump request");
1324 			return -EINVAL;
1325 		}
1326 	}
1327 
1328 	s_h = cb->args[0];
1329 	s_idx = idx = cb->args[1];
1330 
1331 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1332 		idx = 0;
1333 		head = &net->dev_index_head[h];
1334 		rcu_read_lock();
1335 		cb->seq = net->dev_base_seq;
1336 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1337 			if (idx < s_idx)
1338 				goto cont;
1339 			mdev = mpls_dev_get(dev);
1340 			if (!mdev)
1341 				goto cont;
1342 			if (mpls_netconf_fill_devconf(skb, mdev,
1343 						      NETLINK_CB(cb->skb).portid,
1344 						      nlh->nlmsg_seq,
1345 						      RTM_NEWNETCONF,
1346 						      NLM_F_MULTI,
1347 						      NETCONFA_ALL) < 0) {
1348 				rcu_read_unlock();
1349 				goto done;
1350 			}
1351 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1352 cont:
1353 			idx++;
1354 		}
1355 		rcu_read_unlock();
1356 	}
1357 done:
1358 	cb->args[0] = h;
1359 	cb->args[1] = idx;
1360 
1361 	return skb->len;
1362 }
1363 
1364 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1365 	(&((struct mpls_dev *)0)->field)
1366 
1367 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1368 			  void __user *buffer,
1369 			  size_t *lenp, loff_t *ppos)
1370 {
1371 	int oval = *(int *)ctl->data;
1372 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1373 
1374 	if (write) {
1375 		struct mpls_dev *mdev = ctl->extra1;
1376 		int i = (int *)ctl->data - (int *)mdev;
1377 		struct net *net = ctl->extra2;
1378 		int val = *(int *)ctl->data;
1379 
1380 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1381 		    val != oval) {
1382 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1383 						    NETCONFA_INPUT, mdev);
1384 		}
1385 	}
1386 
1387 	return ret;
1388 }
1389 
1390 static const struct ctl_table mpls_dev_table[] = {
1391 	{
1392 		.procname	= "input",
1393 		.maxlen		= sizeof(int),
1394 		.mode		= 0644,
1395 		.proc_handler	= mpls_conf_proc,
1396 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1397 	},
1398 	{ }
1399 };
1400 
1401 static int mpls_dev_sysctl_register(struct net_device *dev,
1402 				    struct mpls_dev *mdev)
1403 {
1404 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1405 	struct net *net = dev_net(dev);
1406 	struct ctl_table *table;
1407 	int i;
1408 
1409 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1410 	if (!table)
1411 		goto out;
1412 
1413 	/* Table data contains only offsets relative to the base of
1414 	 * the mdev at this point, so make them absolute.
1415 	 */
1416 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1417 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1418 		table[i].extra1 = mdev;
1419 		table[i].extra2 = net;
1420 	}
1421 
1422 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1423 
1424 	mdev->sysctl = register_net_sysctl(net, path, table);
1425 	if (!mdev->sysctl)
1426 		goto free;
1427 
1428 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1429 	return 0;
1430 
1431 free:
1432 	kfree(table);
1433 out:
1434 	return -ENOBUFS;
1435 }
1436 
1437 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1438 				       struct mpls_dev *mdev)
1439 {
1440 	struct net *net = dev_net(dev);
1441 	struct ctl_table *table;
1442 
1443 	table = mdev->sysctl->ctl_table_arg;
1444 	unregister_net_sysctl_table(mdev->sysctl);
1445 	kfree(table);
1446 
1447 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1448 }
1449 
1450 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1451 {
1452 	struct mpls_dev *mdev;
1453 	int err = -ENOMEM;
1454 	int i;
1455 
1456 	ASSERT_RTNL();
1457 
1458 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1459 	if (!mdev)
1460 		return ERR_PTR(err);
1461 
1462 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1463 	if (!mdev->stats)
1464 		goto free;
1465 
1466 	for_each_possible_cpu(i) {
1467 		struct mpls_pcpu_stats *mpls_stats;
1468 
1469 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1470 		u64_stats_init(&mpls_stats->syncp);
1471 	}
1472 
1473 	mdev->dev = dev;
1474 
1475 	err = mpls_dev_sysctl_register(dev, mdev);
1476 	if (err)
1477 		goto free;
1478 
1479 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1480 
1481 	return mdev;
1482 
1483 free:
1484 	free_percpu(mdev->stats);
1485 	kfree(mdev);
1486 	return ERR_PTR(err);
1487 }
1488 
1489 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1490 {
1491 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1492 
1493 	free_percpu(mdev->stats);
1494 	kfree(mdev);
1495 }
1496 
1497 static void mpls_ifdown(struct net_device *dev, int event)
1498 {
1499 	struct mpls_route __rcu **platform_label;
1500 	struct net *net = dev_net(dev);
1501 	u8 alive, deleted;
1502 	unsigned index;
1503 
1504 	platform_label = rtnl_dereference(net->mpls.platform_label);
1505 	for (index = 0; index < net->mpls.platform_labels; index++) {
1506 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1507 
1508 		if (!rt)
1509 			continue;
1510 
1511 		alive = 0;
1512 		deleted = 0;
1513 		change_nexthops(rt) {
1514 			unsigned int nh_flags = nh->nh_flags;
1515 
1516 			if (rtnl_dereference(nh->nh_dev) != dev)
1517 				goto next;
1518 
1519 			switch (event) {
1520 			case NETDEV_DOWN:
1521 			case NETDEV_UNREGISTER:
1522 				nh_flags |= RTNH_F_DEAD;
1523 				/* fall through */
1524 			case NETDEV_CHANGE:
1525 				nh_flags |= RTNH_F_LINKDOWN;
1526 				break;
1527 			}
1528 			if (event == NETDEV_UNREGISTER)
1529 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1530 
1531 			if (nh->nh_flags != nh_flags)
1532 				WRITE_ONCE(nh->nh_flags, nh_flags);
1533 next:
1534 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1535 				alive++;
1536 			if (!rtnl_dereference(nh->nh_dev))
1537 				deleted++;
1538 		} endfor_nexthops(rt);
1539 
1540 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1541 
1542 		/* if there are no more nexthops, delete the route */
1543 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1544 			mpls_route_update(net, index, NULL, NULL);
1545 	}
1546 }
1547 
1548 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1549 {
1550 	struct mpls_route __rcu **platform_label;
1551 	struct net *net = dev_net(dev);
1552 	unsigned index;
1553 	u8 alive;
1554 
1555 	platform_label = rtnl_dereference(net->mpls.platform_label);
1556 	for (index = 0; index < net->mpls.platform_labels; index++) {
1557 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1558 
1559 		if (!rt)
1560 			continue;
1561 
1562 		alive = 0;
1563 		change_nexthops(rt) {
1564 			unsigned int nh_flags = nh->nh_flags;
1565 			struct net_device *nh_dev =
1566 				rtnl_dereference(nh->nh_dev);
1567 
1568 			if (!(nh_flags & flags)) {
1569 				alive++;
1570 				continue;
1571 			}
1572 			if (nh_dev != dev)
1573 				continue;
1574 			alive++;
1575 			nh_flags &= ~flags;
1576 			WRITE_ONCE(nh->nh_flags, nh_flags);
1577 		} endfor_nexthops(rt);
1578 
1579 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1580 	}
1581 }
1582 
1583 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1584 			   void *ptr)
1585 {
1586 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1587 	struct mpls_dev *mdev;
1588 	unsigned int flags;
1589 
1590 	if (event == NETDEV_REGISTER) {
1591 
1592 		/* For now just support Ethernet, IPGRE, IP6GRE, SIT and
1593 		 * IPIP devices
1594 		 */
1595 		if (dev->type == ARPHRD_ETHER ||
1596 		    dev->type == ARPHRD_LOOPBACK ||
1597 		    dev->type == ARPHRD_IPGRE ||
1598 		    dev->type == ARPHRD_IP6GRE ||
1599 		    dev->type == ARPHRD_SIT ||
1600 		    dev->type == ARPHRD_TUNNEL) {
1601 			mdev = mpls_add_dev(dev);
1602 			if (IS_ERR(mdev))
1603 				return notifier_from_errno(PTR_ERR(mdev));
1604 		}
1605 		return NOTIFY_OK;
1606 	}
1607 
1608 	mdev = mpls_dev_get(dev);
1609 	if (!mdev)
1610 		return NOTIFY_OK;
1611 
1612 	switch (event) {
1613 	case NETDEV_DOWN:
1614 		mpls_ifdown(dev, event);
1615 		break;
1616 	case NETDEV_UP:
1617 		flags = dev_get_flags(dev);
1618 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1619 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1620 		else
1621 			mpls_ifup(dev, RTNH_F_DEAD);
1622 		break;
1623 	case NETDEV_CHANGE:
1624 		flags = dev_get_flags(dev);
1625 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1626 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1627 		else
1628 			mpls_ifdown(dev, event);
1629 		break;
1630 	case NETDEV_UNREGISTER:
1631 		mpls_ifdown(dev, event);
1632 		mdev = mpls_dev_get(dev);
1633 		if (mdev) {
1634 			mpls_dev_sysctl_unregister(dev, mdev);
1635 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1636 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1637 		}
1638 		break;
1639 	case NETDEV_CHANGENAME:
1640 		mdev = mpls_dev_get(dev);
1641 		if (mdev) {
1642 			int err;
1643 
1644 			mpls_dev_sysctl_unregister(dev, mdev);
1645 			err = mpls_dev_sysctl_register(dev, mdev);
1646 			if (err)
1647 				return notifier_from_errno(err);
1648 		}
1649 		break;
1650 	}
1651 	return NOTIFY_OK;
1652 }
1653 
1654 static struct notifier_block mpls_dev_notifier = {
1655 	.notifier_call = mpls_dev_notify,
1656 };
1657 
1658 static int nla_put_via(struct sk_buff *skb,
1659 		       u8 table, const void *addr, int alen)
1660 {
1661 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1662 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1663 	};
1664 	struct nlattr *nla;
1665 	struct rtvia *via;
1666 	int family = AF_UNSPEC;
1667 
1668 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1669 	if (!nla)
1670 		return -EMSGSIZE;
1671 
1672 	if (table <= NEIGH_NR_TABLES)
1673 		family = table_to_family[table];
1674 
1675 	via = nla_data(nla);
1676 	via->rtvia_family = family;
1677 	memcpy(via->rtvia_addr, addr, alen);
1678 	return 0;
1679 }
1680 
1681 int nla_put_labels(struct sk_buff *skb, int attrtype,
1682 		   u8 labels, const u32 label[])
1683 {
1684 	struct nlattr *nla;
1685 	struct mpls_shim_hdr *nla_label;
1686 	bool bos;
1687 	int i;
1688 	nla = nla_reserve(skb, attrtype, labels*4);
1689 	if (!nla)
1690 		return -EMSGSIZE;
1691 
1692 	nla_label = nla_data(nla);
1693 	bos = true;
1694 	for (i = labels - 1; i >= 0; i--) {
1695 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1696 		bos = false;
1697 	}
1698 
1699 	return 0;
1700 }
1701 EXPORT_SYMBOL_GPL(nla_put_labels);
1702 
1703 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1704 		   u32 label[], struct netlink_ext_ack *extack)
1705 {
1706 	unsigned len = nla_len(nla);
1707 	struct mpls_shim_hdr *nla_label;
1708 	u8 nla_labels;
1709 	bool bos;
1710 	int i;
1711 
1712 	/* len needs to be an even multiple of 4 (the label size). Number
1713 	 * of labels is a u8 so check for overflow.
1714 	 */
1715 	if (len & 3 || len / 4 > 255) {
1716 		NL_SET_ERR_MSG_ATTR(extack, nla,
1717 				    "Invalid length for labels attribute");
1718 		return -EINVAL;
1719 	}
1720 
1721 	/* Limit the number of new labels allowed */
1722 	nla_labels = len/4;
1723 	if (nla_labels > max_labels) {
1724 		NL_SET_ERR_MSG(extack, "Too many labels");
1725 		return -EINVAL;
1726 	}
1727 
1728 	/* when label == NULL, caller wants number of labels */
1729 	if (!label)
1730 		goto out;
1731 
1732 	nla_label = nla_data(nla);
1733 	bos = true;
1734 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1735 		struct mpls_entry_decoded dec;
1736 		dec = mpls_entry_decode(nla_label + i);
1737 
1738 		/* Ensure the bottom of stack flag is properly set
1739 		 * and ttl and tc are both clear.
1740 		 */
1741 		if (dec.ttl) {
1742 			NL_SET_ERR_MSG_ATTR(extack, nla,
1743 					    "TTL in label must be 0");
1744 			return -EINVAL;
1745 		}
1746 
1747 		if (dec.tc) {
1748 			NL_SET_ERR_MSG_ATTR(extack, nla,
1749 					    "Traffic class in label must be 0");
1750 			return -EINVAL;
1751 		}
1752 
1753 		if (dec.bos != bos) {
1754 			NL_SET_BAD_ATTR(extack, nla);
1755 			if (bos) {
1756 				NL_SET_ERR_MSG(extack,
1757 					       "BOS bit must be set in first label");
1758 			} else {
1759 				NL_SET_ERR_MSG(extack,
1760 					       "BOS bit can only be set in first label");
1761 			}
1762 			return -EINVAL;
1763 		}
1764 
1765 		switch (dec.label) {
1766 		case MPLS_LABEL_IMPLNULL:
1767 			/* RFC3032: This is a label that an LSR may
1768 			 * assign and distribute, but which never
1769 			 * actually appears in the encapsulation.
1770 			 */
1771 			NL_SET_ERR_MSG_ATTR(extack, nla,
1772 					    "Implicit NULL Label (3) can not be used in encapsulation");
1773 			return -EINVAL;
1774 		}
1775 
1776 		label[i] = dec.label;
1777 	}
1778 out:
1779 	*labels = nla_labels;
1780 	return 0;
1781 }
1782 EXPORT_SYMBOL_GPL(nla_get_labels);
1783 
1784 static int rtm_to_route_config(struct sk_buff *skb,
1785 			       struct nlmsghdr *nlh,
1786 			       struct mpls_route_config *cfg,
1787 			       struct netlink_ext_ack *extack)
1788 {
1789 	struct rtmsg *rtm;
1790 	struct nlattr *tb[RTA_MAX+1];
1791 	int index;
1792 	int err;
1793 
1794 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1795 				     rtm_mpls_policy, extack);
1796 	if (err < 0)
1797 		goto errout;
1798 
1799 	err = -EINVAL;
1800 	rtm = nlmsg_data(nlh);
1801 
1802 	if (rtm->rtm_family != AF_MPLS) {
1803 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1804 		goto errout;
1805 	}
1806 	if (rtm->rtm_dst_len != 20) {
1807 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1808 		goto errout;
1809 	}
1810 	if (rtm->rtm_src_len != 0) {
1811 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1812 		goto errout;
1813 	}
1814 	if (rtm->rtm_tos != 0) {
1815 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1816 		goto errout;
1817 	}
1818 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1819 		NL_SET_ERR_MSG(extack,
1820 			       "MPLS only supports the main route table");
1821 		goto errout;
1822 	}
1823 	/* Any value is acceptable for rtm_protocol */
1824 
1825 	/* As mpls uses destination specific addresses
1826 	 * (or source specific address in the case of multicast)
1827 	 * all addresses have universal scope.
1828 	 */
1829 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1830 		NL_SET_ERR_MSG(extack,
1831 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1832 		goto errout;
1833 	}
1834 	if (rtm->rtm_type != RTN_UNICAST) {
1835 		NL_SET_ERR_MSG(extack,
1836 			       "Invalid route type - MPLS only supports UNICAST");
1837 		goto errout;
1838 	}
1839 	if (rtm->rtm_flags != 0) {
1840 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1841 		goto errout;
1842 	}
1843 
1844 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1845 	cfg->rc_protocol	= rtm->rtm_protocol;
1846 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1847 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1848 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1849 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1850 	cfg->rc_nlinfo.nlh	= nlh;
1851 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1852 
1853 	for (index = 0; index <= RTA_MAX; index++) {
1854 		struct nlattr *nla = tb[index];
1855 		if (!nla)
1856 			continue;
1857 
1858 		switch (index) {
1859 		case RTA_OIF:
1860 			cfg->rc_ifindex = nla_get_u32(nla);
1861 			break;
1862 		case RTA_NEWDST:
1863 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1864 					   &cfg->rc_output_labels,
1865 					   cfg->rc_output_label, extack))
1866 				goto errout;
1867 			break;
1868 		case RTA_DST:
1869 		{
1870 			u8 label_count;
1871 			if (nla_get_labels(nla, 1, &label_count,
1872 					   &cfg->rc_label, extack))
1873 				goto errout;
1874 
1875 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1876 					   &cfg->rc_label, extack))
1877 				goto errout;
1878 			break;
1879 		}
1880 		case RTA_GATEWAY:
1881 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1882 			goto errout;
1883 		case RTA_VIA:
1884 		{
1885 			if (nla_get_via(nla, &cfg->rc_via_alen,
1886 					&cfg->rc_via_table, cfg->rc_via,
1887 					extack))
1888 				goto errout;
1889 			break;
1890 		}
1891 		case RTA_MULTIPATH:
1892 		{
1893 			cfg->rc_mp = nla_data(nla);
1894 			cfg->rc_mp_len = nla_len(nla);
1895 			break;
1896 		}
1897 		case RTA_TTL_PROPAGATE:
1898 		{
1899 			u8 ttl_propagate = nla_get_u8(nla);
1900 
1901 			if (ttl_propagate > 1) {
1902 				NL_SET_ERR_MSG_ATTR(extack, nla,
1903 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1904 				goto errout;
1905 			}
1906 			cfg->rc_ttl_propagate = ttl_propagate ?
1907 				MPLS_TTL_PROP_ENABLED :
1908 				MPLS_TTL_PROP_DISABLED;
1909 			break;
1910 		}
1911 		default:
1912 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1913 			/* Unsupported attribute */
1914 			goto errout;
1915 		}
1916 	}
1917 
1918 	err = 0;
1919 errout:
1920 	return err;
1921 }
1922 
1923 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1924 			     struct netlink_ext_ack *extack)
1925 {
1926 	struct mpls_route_config *cfg;
1927 	int err;
1928 
1929 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1930 	if (!cfg)
1931 		return -ENOMEM;
1932 
1933 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1934 	if (err < 0)
1935 		goto out;
1936 
1937 	err = mpls_route_del(cfg, extack);
1938 out:
1939 	kfree(cfg);
1940 
1941 	return err;
1942 }
1943 
1944 
1945 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1946 			     struct netlink_ext_ack *extack)
1947 {
1948 	struct mpls_route_config *cfg;
1949 	int err;
1950 
1951 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1952 	if (!cfg)
1953 		return -ENOMEM;
1954 
1955 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1956 	if (err < 0)
1957 		goto out;
1958 
1959 	err = mpls_route_add(cfg, extack);
1960 out:
1961 	kfree(cfg);
1962 
1963 	return err;
1964 }
1965 
1966 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1967 			   u32 label, struct mpls_route *rt, int flags)
1968 {
1969 	struct net_device *dev;
1970 	struct nlmsghdr *nlh;
1971 	struct rtmsg *rtm;
1972 
1973 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1974 	if (nlh == NULL)
1975 		return -EMSGSIZE;
1976 
1977 	rtm = nlmsg_data(nlh);
1978 	rtm->rtm_family = AF_MPLS;
1979 	rtm->rtm_dst_len = 20;
1980 	rtm->rtm_src_len = 0;
1981 	rtm->rtm_tos = 0;
1982 	rtm->rtm_table = RT_TABLE_MAIN;
1983 	rtm->rtm_protocol = rt->rt_protocol;
1984 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1985 	rtm->rtm_type = RTN_UNICAST;
1986 	rtm->rtm_flags = 0;
1987 
1988 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1989 		goto nla_put_failure;
1990 
1991 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1992 		bool ttl_propagate =
1993 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1994 
1995 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1996 			       ttl_propagate))
1997 			goto nla_put_failure;
1998 	}
1999 	if (rt->rt_nhn == 1) {
2000 		const struct mpls_nh *nh = rt->rt_nh;
2001 
2002 		if (nh->nh_labels &&
2003 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2004 				   nh->nh_label))
2005 			goto nla_put_failure;
2006 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2007 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2008 				nh->nh_via_alen))
2009 			goto nla_put_failure;
2010 		dev = rtnl_dereference(nh->nh_dev);
2011 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2012 			goto nla_put_failure;
2013 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2014 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2015 		if (nh->nh_flags & RTNH_F_DEAD)
2016 			rtm->rtm_flags |= RTNH_F_DEAD;
2017 	} else {
2018 		struct rtnexthop *rtnh;
2019 		struct nlattr *mp;
2020 		u8 linkdown = 0;
2021 		u8 dead = 0;
2022 
2023 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2024 		if (!mp)
2025 			goto nla_put_failure;
2026 
2027 		for_nexthops(rt) {
2028 			dev = rtnl_dereference(nh->nh_dev);
2029 			if (!dev)
2030 				continue;
2031 
2032 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2033 			if (!rtnh)
2034 				goto nla_put_failure;
2035 
2036 			rtnh->rtnh_ifindex = dev->ifindex;
2037 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2038 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2039 				linkdown++;
2040 			}
2041 			if (nh->nh_flags & RTNH_F_DEAD) {
2042 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2043 				dead++;
2044 			}
2045 
2046 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2047 							    nh->nh_labels,
2048 							    nh->nh_label))
2049 				goto nla_put_failure;
2050 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2051 			    nla_put_via(skb, nh->nh_via_table,
2052 					mpls_nh_via(rt, nh),
2053 					nh->nh_via_alen))
2054 				goto nla_put_failure;
2055 
2056 			/* length of rtnetlink header + attributes */
2057 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2058 		} endfor_nexthops(rt);
2059 
2060 		if (linkdown == rt->rt_nhn)
2061 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2062 		if (dead == rt->rt_nhn)
2063 			rtm->rtm_flags |= RTNH_F_DEAD;
2064 
2065 		nla_nest_end(skb, mp);
2066 	}
2067 
2068 	nlmsg_end(skb, nlh);
2069 	return 0;
2070 
2071 nla_put_failure:
2072 	nlmsg_cancel(skb, nlh);
2073 	return -EMSGSIZE;
2074 }
2075 
2076 #if IS_ENABLED(CONFIG_INET)
2077 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2078 				   struct fib_dump_filter *filter,
2079 				   struct netlink_callback *cb)
2080 {
2081 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2082 }
2083 #else
2084 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2085 				   struct fib_dump_filter *filter,
2086 				   struct netlink_callback *cb)
2087 {
2088 	struct netlink_ext_ack *extack = cb->extack;
2089 	struct nlattr *tb[RTA_MAX + 1];
2090 	struct rtmsg *rtm;
2091 	int err, i;
2092 
2093 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2094 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2095 		return -EINVAL;
2096 	}
2097 
2098 	rtm = nlmsg_data(nlh);
2099 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2100 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2101 	    rtm->rtm_flags) {
2102 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2103 		return -EINVAL;
2104 	}
2105 
2106 	if (rtm->rtm_protocol) {
2107 		filter->protocol = rtm->rtm_protocol;
2108 		filter->filter_set = 1;
2109 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2110 	}
2111 
2112 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2113 					    rtm_mpls_policy, extack);
2114 	if (err < 0)
2115 		return err;
2116 
2117 	for (i = 0; i <= RTA_MAX; ++i) {
2118 		int ifindex;
2119 
2120 		if (i == RTA_OIF) {
2121 			ifindex = nla_get_u32(tb[i]);
2122 			filter->dev = __dev_get_by_index(net, ifindex);
2123 			if (!filter->dev)
2124 				return -ENODEV;
2125 			filter->filter_set = 1;
2126 		} else if (tb[i]) {
2127 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2128 			return -EINVAL;
2129 		}
2130 	}
2131 
2132 	return 0;
2133 }
2134 #endif
2135 
2136 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2137 			     const struct net_device *dev)
2138 {
2139 	struct net_device *nh_dev;
2140 
2141 	if (rt->rt_nhn == 1) {
2142 		struct mpls_nh *nh = rt->rt_nh;
2143 
2144 		nh_dev = rtnl_dereference(nh->nh_dev);
2145 		if (dev == nh_dev)
2146 			return true;
2147 	} else {
2148 		for_nexthops(rt) {
2149 			nh_dev = rtnl_dereference(nh->nh_dev);
2150 			if (nh_dev == dev)
2151 				return true;
2152 		} endfor_nexthops(rt);
2153 	}
2154 
2155 	return false;
2156 }
2157 
2158 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2159 {
2160 	const struct nlmsghdr *nlh = cb->nlh;
2161 	struct net *net = sock_net(skb->sk);
2162 	struct mpls_route __rcu **platform_label;
2163 	struct fib_dump_filter filter = {};
2164 	unsigned int flags = NLM_F_MULTI;
2165 	size_t platform_labels;
2166 	unsigned int index;
2167 
2168 	ASSERT_RTNL();
2169 
2170 	if (cb->strict_check) {
2171 		int err;
2172 
2173 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2174 		if (err < 0)
2175 			return err;
2176 
2177 		/* for MPLS, there is only 1 table with fixed type and flags.
2178 		 * If either are set in the filter then return nothing.
2179 		 */
2180 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2181 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2182 		     filter.flags)
2183 			return skb->len;
2184 	}
2185 
2186 	index = cb->args[0];
2187 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2188 		index = MPLS_LABEL_FIRST_UNRESERVED;
2189 
2190 	platform_label = rtnl_dereference(net->mpls.platform_label);
2191 	platform_labels = net->mpls.platform_labels;
2192 
2193 	if (filter.filter_set)
2194 		flags |= NLM_F_DUMP_FILTERED;
2195 
2196 	for (; index < platform_labels; index++) {
2197 		struct mpls_route *rt;
2198 
2199 		rt = rtnl_dereference(platform_label[index]);
2200 		if (!rt)
2201 			continue;
2202 
2203 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2204 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2205 			continue;
2206 
2207 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2208 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2209 				    index, rt, flags) < 0)
2210 			break;
2211 	}
2212 	cb->args[0] = index;
2213 
2214 	return skb->len;
2215 }
2216 
2217 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2218 {
2219 	size_t payload =
2220 		NLMSG_ALIGN(sizeof(struct rtmsg))
2221 		+ nla_total_size(4)			/* RTA_DST */
2222 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2223 
2224 	if (rt->rt_nhn == 1) {
2225 		struct mpls_nh *nh = rt->rt_nh;
2226 
2227 		if (nh->nh_dev)
2228 			payload += nla_total_size(4); /* RTA_OIF */
2229 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2230 			payload += nla_total_size(2 + nh->nh_via_alen);
2231 		if (nh->nh_labels) /* RTA_NEWDST */
2232 			payload += nla_total_size(nh->nh_labels * 4);
2233 	} else {
2234 		/* each nexthop is packed in an attribute */
2235 		size_t nhsize = 0;
2236 
2237 		for_nexthops(rt) {
2238 			if (!rtnl_dereference(nh->nh_dev))
2239 				continue;
2240 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2241 			/* RTA_VIA */
2242 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2243 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2244 			if (nh->nh_labels)
2245 				nhsize += nla_total_size(nh->nh_labels * 4);
2246 		} endfor_nexthops(rt);
2247 		/* nested attribute */
2248 		payload += nla_total_size(nhsize);
2249 	}
2250 
2251 	return payload;
2252 }
2253 
2254 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2255 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2256 		       unsigned int nlm_flags)
2257 {
2258 	struct sk_buff *skb;
2259 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2260 	int err = -ENOBUFS;
2261 
2262 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2263 	if (skb == NULL)
2264 		goto errout;
2265 
2266 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2267 	if (err < 0) {
2268 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2269 		WARN_ON(err == -EMSGSIZE);
2270 		kfree_skb(skb);
2271 		goto errout;
2272 	}
2273 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2274 
2275 	return;
2276 errout:
2277 	if (err < 0)
2278 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2279 }
2280 
2281 static int mpls_valid_getroute_req(struct sk_buff *skb,
2282 				   const struct nlmsghdr *nlh,
2283 				   struct nlattr **tb,
2284 				   struct netlink_ext_ack *extack)
2285 {
2286 	struct rtmsg *rtm;
2287 	int i, err;
2288 
2289 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2290 		NL_SET_ERR_MSG_MOD(extack,
2291 				   "Invalid header for get route request");
2292 		return -EINVAL;
2293 	}
2294 
2295 	if (!netlink_strict_get_check(skb))
2296 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2297 					      rtm_mpls_policy, extack);
2298 
2299 	rtm = nlmsg_data(nlh);
2300 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2301 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2302 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2303 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2304 		return -EINVAL;
2305 	}
2306 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2307 		NL_SET_ERR_MSG_MOD(extack,
2308 				   "Invalid flags for get route request");
2309 		return -EINVAL;
2310 	}
2311 
2312 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2313 					    rtm_mpls_policy, extack);
2314 	if (err)
2315 		return err;
2316 
2317 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2318 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2319 		return -EINVAL;
2320 	}
2321 
2322 	for (i = 0; i <= RTA_MAX; i++) {
2323 		if (!tb[i])
2324 			continue;
2325 
2326 		switch (i) {
2327 		case RTA_DST:
2328 		case RTA_NEWDST:
2329 			break;
2330 		default:
2331 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2332 			return -EINVAL;
2333 		}
2334 	}
2335 
2336 	return 0;
2337 }
2338 
2339 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2340 			 struct netlink_ext_ack *extack)
2341 {
2342 	struct net *net = sock_net(in_skb->sk);
2343 	u32 portid = NETLINK_CB(in_skb).portid;
2344 	u32 in_label = LABEL_NOT_SPECIFIED;
2345 	struct nlattr *tb[RTA_MAX + 1];
2346 	u32 labels[MAX_NEW_LABELS];
2347 	struct mpls_shim_hdr *hdr;
2348 	unsigned int hdr_size = 0;
2349 	struct net_device *dev;
2350 	struct mpls_route *rt;
2351 	struct rtmsg *rtm, *r;
2352 	struct nlmsghdr *nlh;
2353 	struct sk_buff *skb;
2354 	struct mpls_nh *nh;
2355 	u8 n_labels;
2356 	int err;
2357 
2358 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2359 	if (err < 0)
2360 		goto errout;
2361 
2362 	rtm = nlmsg_data(in_nlh);
2363 
2364 	if (tb[RTA_DST]) {
2365 		u8 label_count;
2366 
2367 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2368 				   &in_label, extack)) {
2369 			err = -EINVAL;
2370 			goto errout;
2371 		}
2372 
2373 		if (!mpls_label_ok(net, &in_label, extack)) {
2374 			err = -EINVAL;
2375 			goto errout;
2376 		}
2377 	}
2378 
2379 	rt = mpls_route_input_rcu(net, in_label);
2380 	if (!rt) {
2381 		err = -ENETUNREACH;
2382 		goto errout;
2383 	}
2384 
2385 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2386 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2387 		if (!skb) {
2388 			err = -ENOBUFS;
2389 			goto errout;
2390 		}
2391 
2392 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2393 				      RTM_NEWROUTE, in_label, rt, 0);
2394 		if (err < 0) {
2395 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2396 			WARN_ON(err == -EMSGSIZE);
2397 			goto errout_free;
2398 		}
2399 
2400 		return rtnl_unicast(skb, net, portid);
2401 	}
2402 
2403 	if (tb[RTA_NEWDST]) {
2404 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2405 				   labels, extack) != 0) {
2406 			err = -EINVAL;
2407 			goto errout;
2408 		}
2409 
2410 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2411 	}
2412 
2413 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2414 	if (!skb) {
2415 		err = -ENOBUFS;
2416 		goto errout;
2417 	}
2418 
2419 	skb->protocol = htons(ETH_P_MPLS_UC);
2420 
2421 	if (hdr_size) {
2422 		bool bos;
2423 		int i;
2424 
2425 		if (skb_cow(skb, hdr_size)) {
2426 			err = -ENOBUFS;
2427 			goto errout_free;
2428 		}
2429 
2430 		skb_reserve(skb, hdr_size);
2431 		skb_push(skb, hdr_size);
2432 		skb_reset_network_header(skb);
2433 
2434 		/* Push new labels */
2435 		hdr = mpls_hdr(skb);
2436 		bos = true;
2437 		for (i = n_labels - 1; i >= 0; i--) {
2438 			hdr[i] = mpls_entry_encode(labels[i],
2439 						   1, 0, bos);
2440 			bos = false;
2441 		}
2442 	}
2443 
2444 	nh = mpls_select_multipath(rt, skb);
2445 	if (!nh) {
2446 		err = -ENETUNREACH;
2447 		goto errout_free;
2448 	}
2449 
2450 	if (hdr_size) {
2451 		skb_pull(skb, hdr_size);
2452 		skb_reset_network_header(skb);
2453 	}
2454 
2455 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2456 			RTM_NEWROUTE, sizeof(*r), 0);
2457 	if (!nlh) {
2458 		err = -EMSGSIZE;
2459 		goto errout_free;
2460 	}
2461 
2462 	r = nlmsg_data(nlh);
2463 	r->rtm_family	 = AF_MPLS;
2464 	r->rtm_dst_len	= 20;
2465 	r->rtm_src_len	= 0;
2466 	r->rtm_table	= RT_TABLE_MAIN;
2467 	r->rtm_type	= RTN_UNICAST;
2468 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2469 	r->rtm_protocol = rt->rt_protocol;
2470 	r->rtm_flags	= 0;
2471 
2472 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2473 		goto nla_put_failure;
2474 
2475 	if (nh->nh_labels &&
2476 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2477 			   nh->nh_label))
2478 		goto nla_put_failure;
2479 
2480 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2481 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2482 			nh->nh_via_alen))
2483 		goto nla_put_failure;
2484 	dev = rtnl_dereference(nh->nh_dev);
2485 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2486 		goto nla_put_failure;
2487 
2488 	nlmsg_end(skb, nlh);
2489 
2490 	err = rtnl_unicast(skb, net, portid);
2491 errout:
2492 	return err;
2493 
2494 nla_put_failure:
2495 	nlmsg_cancel(skb, nlh);
2496 	err = -EMSGSIZE;
2497 errout_free:
2498 	kfree_skb(skb);
2499 	return err;
2500 }
2501 
2502 static int resize_platform_label_table(struct net *net, size_t limit)
2503 {
2504 	size_t size = sizeof(struct mpls_route *) * limit;
2505 	size_t old_limit;
2506 	size_t cp_size;
2507 	struct mpls_route __rcu **labels = NULL, **old;
2508 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2509 	unsigned index;
2510 
2511 	if (size) {
2512 		labels = kvzalloc(size, GFP_KERNEL);
2513 		if (!labels)
2514 			goto nolabels;
2515 	}
2516 
2517 	/* In case the predefined labels need to be populated */
2518 	if (limit > MPLS_LABEL_IPV4NULL) {
2519 		struct net_device *lo = net->loopback_dev;
2520 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2521 		if (IS_ERR(rt0))
2522 			goto nort0;
2523 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2524 		rt0->rt_protocol = RTPROT_KERNEL;
2525 		rt0->rt_payload_type = MPT_IPV4;
2526 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2527 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2528 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2529 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2530 		       lo->addr_len);
2531 	}
2532 	if (limit > MPLS_LABEL_IPV6NULL) {
2533 		struct net_device *lo = net->loopback_dev;
2534 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2535 		if (IS_ERR(rt2))
2536 			goto nort2;
2537 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2538 		rt2->rt_protocol = RTPROT_KERNEL;
2539 		rt2->rt_payload_type = MPT_IPV6;
2540 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2541 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2542 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2543 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2544 		       lo->addr_len);
2545 	}
2546 
2547 	rtnl_lock();
2548 	/* Remember the original table */
2549 	old = rtnl_dereference(net->mpls.platform_label);
2550 	old_limit = net->mpls.platform_labels;
2551 
2552 	/* Free any labels beyond the new table */
2553 	for (index = limit; index < old_limit; index++)
2554 		mpls_route_update(net, index, NULL, NULL);
2555 
2556 	/* Copy over the old labels */
2557 	cp_size = size;
2558 	if (old_limit < limit)
2559 		cp_size = old_limit * sizeof(struct mpls_route *);
2560 
2561 	memcpy(labels, old, cp_size);
2562 
2563 	/* If needed set the predefined labels */
2564 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2565 	    (limit > MPLS_LABEL_IPV6NULL)) {
2566 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2567 		rt2 = NULL;
2568 	}
2569 
2570 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2571 	    (limit > MPLS_LABEL_IPV4NULL)) {
2572 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2573 		rt0 = NULL;
2574 	}
2575 
2576 	/* Update the global pointers */
2577 	net->mpls.platform_labels = limit;
2578 	rcu_assign_pointer(net->mpls.platform_label, labels);
2579 
2580 	rtnl_unlock();
2581 
2582 	mpls_rt_free(rt2);
2583 	mpls_rt_free(rt0);
2584 
2585 	if (old) {
2586 		synchronize_rcu();
2587 		kvfree(old);
2588 	}
2589 	return 0;
2590 
2591 nort2:
2592 	mpls_rt_free(rt0);
2593 nort0:
2594 	kvfree(labels);
2595 nolabels:
2596 	return -ENOMEM;
2597 }
2598 
2599 static int mpls_platform_labels(struct ctl_table *table, int write,
2600 				void __user *buffer, size_t *lenp, loff_t *ppos)
2601 {
2602 	struct net *net = table->data;
2603 	int platform_labels = net->mpls.platform_labels;
2604 	int ret;
2605 	struct ctl_table tmp = {
2606 		.procname	= table->procname,
2607 		.data		= &platform_labels,
2608 		.maxlen		= sizeof(int),
2609 		.mode		= table->mode,
2610 		.extra1		= &zero,
2611 		.extra2		= &label_limit,
2612 	};
2613 
2614 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2615 
2616 	if (write && ret == 0)
2617 		ret = resize_platform_label_table(net, platform_labels);
2618 
2619 	return ret;
2620 }
2621 
2622 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2623 	(&((struct net *)0)->field)
2624 
2625 static const struct ctl_table mpls_table[] = {
2626 	{
2627 		.procname	= "platform_labels",
2628 		.data		= NULL,
2629 		.maxlen		= sizeof(int),
2630 		.mode		= 0644,
2631 		.proc_handler	= mpls_platform_labels,
2632 	},
2633 	{
2634 		.procname	= "ip_ttl_propagate",
2635 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2636 		.maxlen		= sizeof(int),
2637 		.mode		= 0644,
2638 		.proc_handler	= proc_dointvec_minmax,
2639 		.extra1		= &zero,
2640 		.extra2		= &one,
2641 	},
2642 	{
2643 		.procname	= "default_ttl",
2644 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2645 		.maxlen		= sizeof(int),
2646 		.mode		= 0644,
2647 		.proc_handler	= proc_dointvec_minmax,
2648 		.extra1		= &one,
2649 		.extra2		= &ttl_max,
2650 	},
2651 	{ }
2652 };
2653 
2654 static int mpls_net_init(struct net *net)
2655 {
2656 	struct ctl_table *table;
2657 	int i;
2658 
2659 	net->mpls.platform_labels = 0;
2660 	net->mpls.platform_label = NULL;
2661 	net->mpls.ip_ttl_propagate = 1;
2662 	net->mpls.default_ttl = 255;
2663 
2664 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2665 	if (table == NULL)
2666 		return -ENOMEM;
2667 
2668 	/* Table data contains only offsets relative to the base of
2669 	 * the mdev at this point, so make them absolute.
2670 	 */
2671 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2672 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2673 
2674 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2675 	if (net->mpls.ctl == NULL) {
2676 		kfree(table);
2677 		return -ENOMEM;
2678 	}
2679 
2680 	return 0;
2681 }
2682 
2683 static void mpls_net_exit(struct net *net)
2684 {
2685 	struct mpls_route __rcu **platform_label;
2686 	size_t platform_labels;
2687 	struct ctl_table *table;
2688 	unsigned int index;
2689 
2690 	table = net->mpls.ctl->ctl_table_arg;
2691 	unregister_net_sysctl_table(net->mpls.ctl);
2692 	kfree(table);
2693 
2694 	/* An rcu grace period has passed since there was a device in
2695 	 * the network namespace (and thus the last in flight packet)
2696 	 * left this network namespace.  This is because
2697 	 * unregister_netdevice_many and netdev_run_todo has completed
2698 	 * for each network device that was in this network namespace.
2699 	 *
2700 	 * As such no additional rcu synchronization is necessary when
2701 	 * freeing the platform_label table.
2702 	 */
2703 	rtnl_lock();
2704 	platform_label = rtnl_dereference(net->mpls.platform_label);
2705 	platform_labels = net->mpls.platform_labels;
2706 	for (index = 0; index < platform_labels; index++) {
2707 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2708 		RCU_INIT_POINTER(platform_label[index], NULL);
2709 		mpls_notify_route(net, index, rt, NULL, NULL);
2710 		mpls_rt_free(rt);
2711 	}
2712 	rtnl_unlock();
2713 
2714 	kvfree(platform_label);
2715 }
2716 
2717 static struct pernet_operations mpls_net_ops = {
2718 	.init = mpls_net_init,
2719 	.exit = mpls_net_exit,
2720 };
2721 
2722 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2723 	.family		   = AF_MPLS,
2724 	.fill_stats_af	   = mpls_fill_stats_af,
2725 	.get_stats_af_size = mpls_get_stats_af_size,
2726 };
2727 
2728 static int __init mpls_init(void)
2729 {
2730 	int err;
2731 
2732 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2733 
2734 	err = register_pernet_subsys(&mpls_net_ops);
2735 	if (err)
2736 		goto out;
2737 
2738 	err = register_netdevice_notifier(&mpls_dev_notifier);
2739 	if (err)
2740 		goto out_unregister_pernet;
2741 
2742 	dev_add_pack(&mpls_packet_type);
2743 
2744 	rtnl_af_register(&mpls_af_ops);
2745 
2746 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_NEWROUTE,
2747 			     mpls_rtm_newroute, NULL, 0);
2748 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_DELROUTE,
2749 			     mpls_rtm_delroute, NULL, 0);
2750 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETROUTE,
2751 			     mpls_getroute, mpls_dump_routes, 0);
2752 	rtnl_register_module(THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2753 			     mpls_netconf_get_devconf,
2754 			     mpls_netconf_dump_devconf, 0);
2755 	err = ipgre_tunnel_encap_add_mpls_ops();
2756 	if (err)
2757 		pr_err("Can't add mpls over gre tunnel ops\n");
2758 
2759 	err = 0;
2760 out:
2761 	return err;
2762 
2763 out_unregister_pernet:
2764 	unregister_pernet_subsys(&mpls_net_ops);
2765 	goto out;
2766 }
2767 module_init(mpls_init);
2768 
2769 static void __exit mpls_exit(void)
2770 {
2771 	rtnl_unregister_all(PF_MPLS);
2772 	rtnl_af_unregister(&mpls_af_ops);
2773 	dev_remove_pack(&mpls_packet_type);
2774 	unregister_netdevice_notifier(&mpls_dev_notifier);
2775 	unregister_pernet_subsys(&mpls_net_ops);
2776 	ipgre_tunnel_encap_del_mpls_ops();
2777 }
2778 module_exit(mpls_exit);
2779 
2780 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2781 MODULE_LICENSE("GPL v2");
2782 MODULE_ALIAS_NETPROTO(PF_MPLS);
2783