xref: /openbmc/linux/net/mpls/af_mpls.c (revision a72594ca)
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/vmalloc.h>
12 #include <linux/percpu.h>
13 #include <net/ip.h>
14 #include <net/dst.h>
15 #include <net/sock.h>
16 #include <net/arp.h>
17 #include <net/ip_fib.h>
18 #include <net/netevent.h>
19 #include <net/ip_tunnels.h>
20 #include <net/netns/generic.h>
21 #if IS_ENABLED(CONFIG_IPV6)
22 #include <net/ipv6.h>
23 #endif
24 #include <net/addrconf.h>
25 #include <net/nexthop.h>
26 #include "internal.h"
27 
28 /* max memory we will use for mpls_route */
29 #define MAX_MPLS_ROUTE_MEM	4096
30 
31 /* Maximum number of labels to look ahead at when selecting a path of
32  * a multipath route
33  */
34 #define MAX_MP_SELECT_LABELS 4
35 
36 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
37 
38 static int zero = 0;
39 static int one = 1;
40 static int label_limit = (1 << 20) - 1;
41 static int ttl_max = 255;
42 
43 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
44 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
45 {
46 	return sizeof(struct mpls_shim_hdr);
47 }
48 
49 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
50 	.encap_hlen	= ipgre_mpls_encap_hlen,
51 };
52 
53 static int ipgre_tunnel_encap_add_mpls_ops(void)
54 {
55 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
56 }
57 
58 static void ipgre_tunnel_encap_del_mpls_ops(void)
59 {
60 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
61 }
62 #else
63 static int ipgre_tunnel_encap_add_mpls_ops(void)
64 {
65 	return 0;
66 }
67 
68 static void ipgre_tunnel_encap_del_mpls_ops(void)
69 {
70 }
71 #endif
72 
73 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
74 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
75 		       unsigned int nlm_flags);
76 
77 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
78 {
79 	struct mpls_route *rt = NULL;
80 
81 	if (index < net->mpls.platform_labels) {
82 		struct mpls_route __rcu **platform_label =
83 			rcu_dereference(net->mpls.platform_label);
84 		rt = rcu_dereference(platform_label[index]);
85 	}
86 	return rt;
87 }
88 
89 bool mpls_output_possible(const struct net_device *dev)
90 {
91 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
92 }
93 EXPORT_SYMBOL_GPL(mpls_output_possible);
94 
95 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
96 {
97 	return (u8 *)nh + rt->rt_via_offset;
98 }
99 
100 static const u8 *mpls_nh_via(const struct mpls_route *rt,
101 			     const struct mpls_nh *nh)
102 {
103 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
104 }
105 
106 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
107 {
108 	/* The size of the layer 2.5 labels to be added for this route */
109 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
110 }
111 
112 unsigned int mpls_dev_mtu(const struct net_device *dev)
113 {
114 	/* The amount of data the layer 2 frame can hold */
115 	return dev->mtu;
116 }
117 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
118 
119 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
120 {
121 	if (skb->len <= mtu)
122 		return false;
123 
124 	if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
125 		return false;
126 
127 	return true;
128 }
129 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
130 
131 void mpls_stats_inc_outucastpkts(struct net_device *dev,
132 				 const struct sk_buff *skb)
133 {
134 	struct mpls_dev *mdev;
135 
136 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
137 		mdev = mpls_dev_get(dev);
138 		if (mdev)
139 			MPLS_INC_STATS_LEN(mdev, skb->len,
140 					   tx_packets,
141 					   tx_bytes);
142 	} else if (skb->protocol == htons(ETH_P_IP)) {
143 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
144 #if IS_ENABLED(CONFIG_IPV6)
145 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
146 		struct inet6_dev *in6dev = __in6_dev_get(dev);
147 
148 		if (in6dev)
149 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
150 					 IPSTATS_MIB_OUT, skb->len);
151 #endif
152 	}
153 }
154 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
155 
156 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
157 {
158 	struct mpls_entry_decoded dec;
159 	unsigned int mpls_hdr_len = 0;
160 	struct mpls_shim_hdr *hdr;
161 	bool eli_seen = false;
162 	int label_index;
163 	u32 hash = 0;
164 
165 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
166 	     label_index++) {
167 		mpls_hdr_len += sizeof(*hdr);
168 		if (!pskb_may_pull(skb, mpls_hdr_len))
169 			break;
170 
171 		/* Read and decode the current label */
172 		hdr = mpls_hdr(skb) + label_index;
173 		dec = mpls_entry_decode(hdr);
174 
175 		/* RFC6790 - reserved labels MUST NOT be used as keys
176 		 * for the load-balancing function
177 		 */
178 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
179 			hash = jhash_1word(dec.label, hash);
180 
181 			/* The entropy label follows the entropy label
182 			 * indicator, so this means that the entropy
183 			 * label was just added to the hash - no need to
184 			 * go any deeper either in the label stack or in the
185 			 * payload
186 			 */
187 			if (eli_seen)
188 				break;
189 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
190 			eli_seen = true;
191 		}
192 
193 		if (!dec.bos)
194 			continue;
195 
196 		/* found bottom label; does skb have room for a header? */
197 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
198 			const struct iphdr *v4hdr;
199 
200 			v4hdr = (const struct iphdr *)(hdr + 1);
201 			if (v4hdr->version == 4) {
202 				hash = jhash_3words(ntohl(v4hdr->saddr),
203 						    ntohl(v4hdr->daddr),
204 						    v4hdr->protocol, hash);
205 			} else if (v4hdr->version == 6 &&
206 				   pskb_may_pull(skb, mpls_hdr_len +
207 						 sizeof(struct ipv6hdr))) {
208 				const struct ipv6hdr *v6hdr;
209 
210 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
211 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
212 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
213 				hash = jhash_1word(v6hdr->nexthdr, hash);
214 			}
215 		}
216 
217 		break;
218 	}
219 
220 	return hash;
221 }
222 
223 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
224 {
225 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
226 }
227 
228 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
229  * a next hop (nh->nh_flags) are modified by netdev event handlers.
230  * Since those fields can change at any moment, use READ_ONCE to
231  * access both.
232  */
233 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
234 					     struct sk_buff *skb)
235 {
236 	u32 hash = 0;
237 	int nh_index = 0;
238 	int n = 0;
239 	u8 alive;
240 
241 	/* No need to look further into packet if there's only
242 	 * one path
243 	 */
244 	if (rt->rt_nhn == 1)
245 		return rt->rt_nh;
246 
247 	alive = READ_ONCE(rt->rt_nhn_alive);
248 	if (alive == 0)
249 		return NULL;
250 
251 	hash = mpls_multipath_hash(rt, skb);
252 	nh_index = hash % alive;
253 	if (alive == rt->rt_nhn)
254 		goto out;
255 	for_nexthops(rt) {
256 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
257 
258 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
259 			continue;
260 		if (n == nh_index)
261 			return nh;
262 		n++;
263 	} endfor_nexthops(rt);
264 
265 out:
266 	return mpls_get_nexthop(rt, nh_index);
267 }
268 
269 static bool mpls_egress(struct net *net, struct mpls_route *rt,
270 			struct sk_buff *skb, struct mpls_entry_decoded dec)
271 {
272 	enum mpls_payload_type payload_type;
273 	bool success = false;
274 
275 	/* The IPv4 code below accesses through the IPv4 header
276 	 * checksum, which is 12 bytes into the packet.
277 	 * The IPv6 code below accesses through the IPv6 hop limit
278 	 * which is 8 bytes into the packet.
279 	 *
280 	 * For all supported cases there should always be at least 12
281 	 * bytes of packet data present.  The IPv4 header is 20 bytes
282 	 * without options and the IPv6 header is always 40 bytes
283 	 * long.
284 	 */
285 	if (!pskb_may_pull(skb, 12))
286 		return false;
287 
288 	payload_type = rt->rt_payload_type;
289 	if (payload_type == MPT_UNSPEC)
290 		payload_type = ip_hdr(skb)->version;
291 
292 	switch (payload_type) {
293 	case MPT_IPV4: {
294 		struct iphdr *hdr4 = ip_hdr(skb);
295 		u8 new_ttl;
296 		skb->protocol = htons(ETH_P_IP);
297 
298 		/* If propagating TTL, take the decremented TTL from
299 		 * the incoming MPLS header, otherwise decrement the
300 		 * TTL, but only if not 0 to avoid underflow.
301 		 */
302 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
303 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
304 		     net->mpls.ip_ttl_propagate))
305 			new_ttl = dec.ttl;
306 		else
307 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
308 
309 		csum_replace2(&hdr4->check,
310 			      htons(hdr4->ttl << 8),
311 			      htons(new_ttl << 8));
312 		hdr4->ttl = new_ttl;
313 		success = true;
314 		break;
315 	}
316 	case MPT_IPV6: {
317 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
318 		skb->protocol = htons(ETH_P_IPV6);
319 
320 		/* If propagating TTL, take the decremented TTL from
321 		 * the incoming MPLS header, otherwise decrement the
322 		 * hop limit, but only if not 0 to avoid underflow.
323 		 */
324 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
325 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
326 		     net->mpls.ip_ttl_propagate))
327 			hdr6->hop_limit = dec.ttl;
328 		else if (hdr6->hop_limit)
329 			hdr6->hop_limit = hdr6->hop_limit - 1;
330 		success = true;
331 		break;
332 	}
333 	case MPT_UNSPEC:
334 		/* Should have decided which protocol it is by now */
335 		break;
336 	}
337 
338 	return success;
339 }
340 
341 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
342 			struct packet_type *pt, struct net_device *orig_dev)
343 {
344 	struct net *net = dev_net(dev);
345 	struct mpls_shim_hdr *hdr;
346 	struct mpls_route *rt;
347 	struct mpls_nh *nh;
348 	struct mpls_entry_decoded dec;
349 	struct net_device *out_dev;
350 	struct mpls_dev *out_mdev;
351 	struct mpls_dev *mdev;
352 	unsigned int hh_len;
353 	unsigned int new_header_size;
354 	unsigned int mtu;
355 	int err;
356 
357 	/* Careful this entire function runs inside of an rcu critical section */
358 
359 	mdev = mpls_dev_get(dev);
360 	if (!mdev)
361 		goto drop;
362 
363 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
364 			   rx_bytes);
365 
366 	if (!mdev->input_enabled) {
367 		MPLS_INC_STATS(mdev, rx_dropped);
368 		goto drop;
369 	}
370 
371 	if (skb->pkt_type != PACKET_HOST)
372 		goto err;
373 
374 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
375 		goto err;
376 
377 	if (!pskb_may_pull(skb, sizeof(*hdr)))
378 		goto err;
379 
380 	/* Read and decode the label */
381 	hdr = mpls_hdr(skb);
382 	dec = mpls_entry_decode(hdr);
383 
384 	rt = mpls_route_input_rcu(net, dec.label);
385 	if (!rt) {
386 		MPLS_INC_STATS(mdev, rx_noroute);
387 		goto drop;
388 	}
389 
390 	nh = mpls_select_multipath(rt, skb);
391 	if (!nh)
392 		goto err;
393 
394 	/* Pop the label */
395 	skb_pull(skb, sizeof(*hdr));
396 	skb_reset_network_header(skb);
397 
398 	skb_orphan(skb);
399 
400 	if (skb_warn_if_lro(skb))
401 		goto err;
402 
403 	skb_forward_csum(skb);
404 
405 	/* Verify ttl is valid */
406 	if (dec.ttl <= 1)
407 		goto err;
408 	dec.ttl -= 1;
409 
410 	/* Find the output device */
411 	out_dev = rcu_dereference(nh->nh_dev);
412 	if (!mpls_output_possible(out_dev))
413 		goto tx_err;
414 
415 	/* Verify the destination can hold the packet */
416 	new_header_size = mpls_nh_header_size(nh);
417 	mtu = mpls_dev_mtu(out_dev);
418 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
419 		goto tx_err;
420 
421 	hh_len = LL_RESERVED_SPACE(out_dev);
422 	if (!out_dev->header_ops)
423 		hh_len = 0;
424 
425 	/* Ensure there is enough space for the headers in the skb */
426 	if (skb_cow(skb, hh_len + new_header_size))
427 		goto tx_err;
428 
429 	skb->dev = out_dev;
430 	skb->protocol = htons(ETH_P_MPLS_UC);
431 
432 	if (unlikely(!new_header_size && dec.bos)) {
433 		/* Penultimate hop popping */
434 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
435 			goto err;
436 	} else {
437 		bool bos;
438 		int i;
439 		skb_push(skb, new_header_size);
440 		skb_reset_network_header(skb);
441 		/* Push the new labels */
442 		hdr = mpls_hdr(skb);
443 		bos = dec.bos;
444 		for (i = nh->nh_labels - 1; i >= 0; i--) {
445 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
446 						   dec.ttl, 0, bos);
447 			bos = false;
448 		}
449 	}
450 
451 	mpls_stats_inc_outucastpkts(out_dev, skb);
452 
453 	/* If via wasn't specified then send out using device address */
454 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
455 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
456 				 out_dev->dev_addr, skb);
457 	else
458 		err = neigh_xmit(nh->nh_via_table, out_dev,
459 				 mpls_nh_via(rt, nh), skb);
460 	if (err)
461 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
462 				    __func__, err);
463 	return 0;
464 
465 tx_err:
466 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
467 	if (out_mdev)
468 		MPLS_INC_STATS(out_mdev, tx_errors);
469 	goto drop;
470 err:
471 	MPLS_INC_STATS(mdev, rx_errors);
472 drop:
473 	kfree_skb(skb);
474 	return NET_RX_DROP;
475 }
476 
477 static struct packet_type mpls_packet_type __read_mostly = {
478 	.type = cpu_to_be16(ETH_P_MPLS_UC),
479 	.func = mpls_forward,
480 };
481 
482 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
483 	[RTA_DST]		= { .type = NLA_U32 },
484 	[RTA_OIF]		= { .type = NLA_U32 },
485 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
486 };
487 
488 struct mpls_route_config {
489 	u32			rc_protocol;
490 	u32			rc_ifindex;
491 	u8			rc_via_table;
492 	u8			rc_via_alen;
493 	u8			rc_via[MAX_VIA_ALEN];
494 	u32			rc_label;
495 	u8			rc_ttl_propagate;
496 	u8			rc_output_labels;
497 	u32			rc_output_label[MAX_NEW_LABELS];
498 	u32			rc_nlflags;
499 	enum mpls_payload_type	rc_payload_type;
500 	struct nl_info		rc_nlinfo;
501 	struct rtnexthop	*rc_mp;
502 	int			rc_mp_len;
503 };
504 
505 /* all nexthops within a route have the same size based on max
506  * number of labels and max via length for a hop
507  */
508 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
509 {
510 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
511 	struct mpls_route *rt;
512 	size_t size;
513 
514 	size = sizeof(*rt) + num_nh * nh_size;
515 	if (size > MAX_MPLS_ROUTE_MEM)
516 		return ERR_PTR(-EINVAL);
517 
518 	rt = kzalloc(size, GFP_KERNEL);
519 	if (!rt)
520 		return ERR_PTR(-ENOMEM);
521 
522 	rt->rt_nhn = num_nh;
523 	rt->rt_nhn_alive = num_nh;
524 	rt->rt_nh_size = nh_size;
525 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
526 
527 	return rt;
528 }
529 
530 static void mpls_rt_free(struct mpls_route *rt)
531 {
532 	if (rt)
533 		kfree_rcu(rt, rt_rcu);
534 }
535 
536 static void mpls_notify_route(struct net *net, unsigned index,
537 			      struct mpls_route *old, struct mpls_route *new,
538 			      const struct nl_info *info)
539 {
540 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
541 	unsigned portid = info ? info->portid : 0;
542 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
543 	struct mpls_route *rt = new ? new : old;
544 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
545 	/* Ignore reserved labels for now */
546 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
547 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
548 }
549 
550 static void mpls_route_update(struct net *net, unsigned index,
551 			      struct mpls_route *new,
552 			      const struct nl_info *info)
553 {
554 	struct mpls_route __rcu **platform_label;
555 	struct mpls_route *rt;
556 
557 	ASSERT_RTNL();
558 
559 	platform_label = rtnl_dereference(net->mpls.platform_label);
560 	rt = rtnl_dereference(platform_label[index]);
561 	rcu_assign_pointer(platform_label[index], new);
562 
563 	mpls_notify_route(net, index, rt, new, info);
564 
565 	/* If we removed a route free it now */
566 	mpls_rt_free(rt);
567 }
568 
569 static unsigned find_free_label(struct net *net)
570 {
571 	struct mpls_route __rcu **platform_label;
572 	size_t platform_labels;
573 	unsigned index;
574 
575 	platform_label = rtnl_dereference(net->mpls.platform_label);
576 	platform_labels = net->mpls.platform_labels;
577 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
578 	     index++) {
579 		if (!rtnl_dereference(platform_label[index]))
580 			return index;
581 	}
582 	return LABEL_NOT_SPECIFIED;
583 }
584 
585 #if IS_ENABLED(CONFIG_INET)
586 static struct net_device *inet_fib_lookup_dev(struct net *net,
587 					      const void *addr)
588 {
589 	struct net_device *dev;
590 	struct rtable *rt;
591 	struct in_addr daddr;
592 
593 	memcpy(&daddr, addr, sizeof(struct in_addr));
594 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
595 	if (IS_ERR(rt))
596 		return ERR_CAST(rt);
597 
598 	dev = rt->dst.dev;
599 	dev_hold(dev);
600 
601 	ip_rt_put(rt);
602 
603 	return dev;
604 }
605 #else
606 static struct net_device *inet_fib_lookup_dev(struct net *net,
607 					      const void *addr)
608 {
609 	return ERR_PTR(-EAFNOSUPPORT);
610 }
611 #endif
612 
613 #if IS_ENABLED(CONFIG_IPV6)
614 static struct net_device *inet6_fib_lookup_dev(struct net *net,
615 					       const void *addr)
616 {
617 	struct net_device *dev;
618 	struct dst_entry *dst;
619 	struct flowi6 fl6;
620 	int err;
621 
622 	if (!ipv6_stub)
623 		return ERR_PTR(-EAFNOSUPPORT);
624 
625 	memset(&fl6, 0, sizeof(fl6));
626 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
627 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
628 	if (err)
629 		return ERR_PTR(err);
630 
631 	dev = dst->dev;
632 	dev_hold(dev);
633 	dst_release(dst);
634 
635 	return dev;
636 }
637 #else
638 static struct net_device *inet6_fib_lookup_dev(struct net *net,
639 					       const void *addr)
640 {
641 	return ERR_PTR(-EAFNOSUPPORT);
642 }
643 #endif
644 
645 static struct net_device *find_outdev(struct net *net,
646 				      struct mpls_route *rt,
647 				      struct mpls_nh *nh, int oif)
648 {
649 	struct net_device *dev = NULL;
650 
651 	if (!oif) {
652 		switch (nh->nh_via_table) {
653 		case NEIGH_ARP_TABLE:
654 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
655 			break;
656 		case NEIGH_ND_TABLE:
657 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
658 			break;
659 		case NEIGH_LINK_TABLE:
660 			break;
661 		}
662 	} else {
663 		dev = dev_get_by_index(net, oif);
664 	}
665 
666 	if (!dev)
667 		return ERR_PTR(-ENODEV);
668 
669 	if (IS_ERR(dev))
670 		return dev;
671 
672 	/* The caller is holding rtnl anyways, so release the dev reference */
673 	dev_put(dev);
674 
675 	return dev;
676 }
677 
678 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
679 			      struct mpls_nh *nh, int oif)
680 {
681 	struct net_device *dev = NULL;
682 	int err = -ENODEV;
683 
684 	dev = find_outdev(net, rt, nh, oif);
685 	if (IS_ERR(dev)) {
686 		err = PTR_ERR(dev);
687 		dev = NULL;
688 		goto errout;
689 	}
690 
691 	/* Ensure this is a supported device */
692 	err = -EINVAL;
693 	if (!mpls_dev_get(dev))
694 		goto errout;
695 
696 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
697 	    (dev->addr_len != nh->nh_via_alen))
698 		goto errout;
699 
700 	RCU_INIT_POINTER(nh->nh_dev, dev);
701 
702 	if (!(dev->flags & IFF_UP)) {
703 		nh->nh_flags |= RTNH_F_DEAD;
704 	} else {
705 		unsigned int flags;
706 
707 		flags = dev_get_flags(dev);
708 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
709 			nh->nh_flags |= RTNH_F_LINKDOWN;
710 	}
711 
712 	return 0;
713 
714 errout:
715 	return err;
716 }
717 
718 static int nla_get_via(const struct nlattr *nla, u8 *via_alen, u8 *via_table,
719 		       u8 via_addr[], struct netlink_ext_ack *extack)
720 {
721 	struct rtvia *via = nla_data(nla);
722 	int err = -EINVAL;
723 	int alen;
724 
725 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
726 		NL_SET_ERR_MSG_ATTR(extack, nla,
727 				    "Invalid attribute length for RTA_VIA");
728 		goto errout;
729 	}
730 	alen = nla_len(nla) -
731 			offsetof(struct rtvia, rtvia_addr);
732 	if (alen > MAX_VIA_ALEN) {
733 		NL_SET_ERR_MSG_ATTR(extack, nla,
734 				    "Invalid address length for RTA_VIA");
735 		goto errout;
736 	}
737 
738 	/* Validate the address family */
739 	switch (via->rtvia_family) {
740 	case AF_PACKET:
741 		*via_table = NEIGH_LINK_TABLE;
742 		break;
743 	case AF_INET:
744 		*via_table = NEIGH_ARP_TABLE;
745 		if (alen != 4)
746 			goto errout;
747 		break;
748 	case AF_INET6:
749 		*via_table = NEIGH_ND_TABLE;
750 		if (alen != 16)
751 			goto errout;
752 		break;
753 	default:
754 		/* Unsupported address family */
755 		goto errout;
756 	}
757 
758 	memcpy(via_addr, via->rtvia_addr, alen);
759 	*via_alen = alen;
760 	err = 0;
761 
762 errout:
763 	return err;
764 }
765 
766 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
767 				  struct mpls_route *rt)
768 {
769 	struct net *net = cfg->rc_nlinfo.nl_net;
770 	struct mpls_nh *nh = rt->rt_nh;
771 	int err;
772 	int i;
773 
774 	if (!nh)
775 		return -ENOMEM;
776 
777 	nh->nh_labels = cfg->rc_output_labels;
778 	for (i = 0; i < nh->nh_labels; i++)
779 		nh->nh_label[i] = cfg->rc_output_label[i];
780 
781 	nh->nh_via_table = cfg->rc_via_table;
782 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
783 	nh->nh_via_alen = cfg->rc_via_alen;
784 
785 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
786 	if (err)
787 		goto errout;
788 
789 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
790 		rt->rt_nhn_alive--;
791 
792 	return 0;
793 
794 errout:
795 	return err;
796 }
797 
798 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
799 			 struct mpls_nh *nh, int oif, struct nlattr *via,
800 			 struct nlattr *newdst, u8 max_labels,
801 			 struct netlink_ext_ack *extack)
802 {
803 	int err = -ENOMEM;
804 
805 	if (!nh)
806 		goto errout;
807 
808 	if (newdst) {
809 		err = nla_get_labels(newdst, max_labels, &nh->nh_labels,
810 				     nh->nh_label, extack);
811 		if (err)
812 			goto errout;
813 	}
814 
815 	if (via) {
816 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
817 				  __mpls_nh_via(rt, nh), extack);
818 		if (err)
819 			goto errout;
820 	} else {
821 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
822 	}
823 
824 	err = mpls_nh_assign_dev(net, rt, nh, oif);
825 	if (err)
826 		goto errout;
827 
828 	return 0;
829 
830 errout:
831 	return err;
832 }
833 
834 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
835 			      u8 cfg_via_alen, u8 *max_via_alen,
836 			      u8 *max_labels)
837 {
838 	int remaining = len;
839 	u8 nhs = 0;
840 
841 	*max_via_alen = 0;
842 	*max_labels = 0;
843 
844 	while (rtnh_ok(rtnh, remaining)) {
845 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
846 		int attrlen;
847 		u8 n_labels = 0;
848 
849 		attrlen = rtnh_attrlen(rtnh);
850 		nla = nla_find(attrs, attrlen, RTA_VIA);
851 		if (nla && nla_len(nla) >=
852 		    offsetof(struct rtvia, rtvia_addr)) {
853 			int via_alen = nla_len(nla) -
854 				offsetof(struct rtvia, rtvia_addr);
855 
856 			if (via_alen <= MAX_VIA_ALEN)
857 				*max_via_alen = max_t(u16, *max_via_alen,
858 						      via_alen);
859 		}
860 
861 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
862 		if (nla &&
863 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels,
864 				   NULL, NULL) != 0)
865 			return 0;
866 
867 		*max_labels = max_t(u8, *max_labels, n_labels);
868 
869 		/* number of nexthops is tracked by a u8.
870 		 * Check for overflow.
871 		 */
872 		if (nhs == 255)
873 			return 0;
874 		nhs++;
875 
876 		rtnh = rtnh_next(rtnh, &remaining);
877 	}
878 
879 	/* leftover implies invalid nexthop configuration, discard it */
880 	return remaining > 0 ? 0 : nhs;
881 }
882 
883 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
884 			       struct mpls_route *rt, u8 max_labels,
885 			       struct netlink_ext_ack *extack)
886 {
887 	struct rtnexthop *rtnh = cfg->rc_mp;
888 	struct nlattr *nla_via, *nla_newdst;
889 	int remaining = cfg->rc_mp_len;
890 	int err = 0;
891 	u8 nhs = 0;
892 
893 	change_nexthops(rt) {
894 		int attrlen;
895 
896 		nla_via = NULL;
897 		nla_newdst = NULL;
898 
899 		err = -EINVAL;
900 		if (!rtnh_ok(rtnh, remaining))
901 			goto errout;
902 
903 		/* neither weighted multipath nor any flags
904 		 * are supported
905 		 */
906 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
907 			goto errout;
908 
909 		attrlen = rtnh_attrlen(rtnh);
910 		if (attrlen > 0) {
911 			struct nlattr *attrs = rtnh_attrs(rtnh);
912 
913 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
914 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
915 		}
916 
917 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
918 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
919 				    max_labels, extack);
920 		if (err)
921 			goto errout;
922 
923 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
924 			rt->rt_nhn_alive--;
925 
926 		rtnh = rtnh_next(rtnh, &remaining);
927 		nhs++;
928 	} endfor_nexthops(rt);
929 
930 	rt->rt_nhn = nhs;
931 
932 	return 0;
933 
934 errout:
935 	return err;
936 }
937 
938 static bool mpls_label_ok(struct net *net, unsigned int index,
939 			  struct netlink_ext_ack *extack)
940 {
941 	/* Reserved labels may not be set */
942 	if (index < MPLS_LABEL_FIRST_UNRESERVED) {
943 		NL_SET_ERR_MSG(extack,
944 			       "Invalid label - must be MPLS_LABEL_FIRST_UNRESERVED or higher");
945 		return false;
946 	}
947 
948 	/* The full 20 bit range may not be supported. */
949 	if (index >= net->mpls.platform_labels) {
950 		NL_SET_ERR_MSG(extack,
951 			       "Label >= configured maximum in platform_labels");
952 		return false;
953 	}
954 
955 	return true;
956 }
957 
958 static int mpls_route_add(struct mpls_route_config *cfg,
959 			  struct netlink_ext_ack *extack)
960 {
961 	struct mpls_route __rcu **platform_label;
962 	struct net *net = cfg->rc_nlinfo.nl_net;
963 	struct mpls_route *rt, *old;
964 	int err = -EINVAL;
965 	u8 max_via_alen;
966 	unsigned index;
967 	u8 max_labels;
968 	u8 nhs;
969 
970 	index = cfg->rc_label;
971 
972 	/* If a label was not specified during insert pick one */
973 	if ((index == LABEL_NOT_SPECIFIED) &&
974 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
975 		index = find_free_label(net);
976 	}
977 
978 	if (!mpls_label_ok(net, index, extack))
979 		goto errout;
980 
981 	/* Append makes no sense with mpls */
982 	err = -EOPNOTSUPP;
983 	if (cfg->rc_nlflags & NLM_F_APPEND) {
984 		NL_SET_ERR_MSG(extack, "MPLS does not support route append");
985 		goto errout;
986 	}
987 
988 	err = -EEXIST;
989 	platform_label = rtnl_dereference(net->mpls.platform_label);
990 	old = rtnl_dereference(platform_label[index]);
991 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
992 		goto errout;
993 
994 	err = -EEXIST;
995 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
996 		goto errout;
997 
998 	err = -ENOENT;
999 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
1000 		goto errout;
1001 
1002 	err = -EINVAL;
1003 	if (cfg->rc_mp) {
1004 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
1005 					  cfg->rc_via_alen, &max_via_alen,
1006 					  &max_labels);
1007 	} else {
1008 		max_via_alen = cfg->rc_via_alen;
1009 		max_labels = cfg->rc_output_labels;
1010 		nhs = 1;
1011 	}
1012 
1013 	if (nhs == 0) {
1014 		NL_SET_ERR_MSG(extack, "Route does not contain a nexthop");
1015 		goto errout;
1016 	}
1017 
1018 	err = -ENOMEM;
1019 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1020 	if (IS_ERR(rt)) {
1021 		err = PTR_ERR(rt);
1022 		goto errout;
1023 	}
1024 
1025 	rt->rt_protocol = cfg->rc_protocol;
1026 	rt->rt_payload_type = cfg->rc_payload_type;
1027 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1028 
1029 	if (cfg->rc_mp)
1030 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1031 	else
1032 		err = mpls_nh_build_from_cfg(cfg, rt);
1033 	if (err)
1034 		goto freert;
1035 
1036 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1037 
1038 	return 0;
1039 
1040 freert:
1041 	mpls_rt_free(rt);
1042 errout:
1043 	return err;
1044 }
1045 
1046 static int mpls_route_del(struct mpls_route_config *cfg,
1047 			  struct netlink_ext_ack *extack)
1048 {
1049 	struct net *net = cfg->rc_nlinfo.nl_net;
1050 	unsigned index;
1051 	int err = -EINVAL;
1052 
1053 	index = cfg->rc_label;
1054 
1055 	if (!mpls_label_ok(net, index, extack))
1056 		goto errout;
1057 
1058 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1059 
1060 	err = 0;
1061 errout:
1062 	return err;
1063 }
1064 
1065 static void mpls_get_stats(struct mpls_dev *mdev,
1066 			   struct mpls_link_stats *stats)
1067 {
1068 	struct mpls_pcpu_stats *p;
1069 	int i;
1070 
1071 	memset(stats, 0, sizeof(*stats));
1072 
1073 	for_each_possible_cpu(i) {
1074 		struct mpls_link_stats local;
1075 		unsigned int start;
1076 
1077 		p = per_cpu_ptr(mdev->stats, i);
1078 		do {
1079 			start = u64_stats_fetch_begin(&p->syncp);
1080 			local = p->stats;
1081 		} while (u64_stats_fetch_retry(&p->syncp, start));
1082 
1083 		stats->rx_packets	+= local.rx_packets;
1084 		stats->rx_bytes		+= local.rx_bytes;
1085 		stats->tx_packets	+= local.tx_packets;
1086 		stats->tx_bytes		+= local.tx_bytes;
1087 		stats->rx_errors	+= local.rx_errors;
1088 		stats->tx_errors	+= local.tx_errors;
1089 		stats->rx_dropped	+= local.rx_dropped;
1090 		stats->tx_dropped	+= local.tx_dropped;
1091 		stats->rx_noroute	+= local.rx_noroute;
1092 	}
1093 }
1094 
1095 static int mpls_fill_stats_af(struct sk_buff *skb,
1096 			      const struct net_device *dev)
1097 {
1098 	struct mpls_link_stats *stats;
1099 	struct mpls_dev *mdev;
1100 	struct nlattr *nla;
1101 
1102 	mdev = mpls_dev_get(dev);
1103 	if (!mdev)
1104 		return -ENODATA;
1105 
1106 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1107 				sizeof(struct mpls_link_stats),
1108 				MPLS_STATS_UNSPEC);
1109 	if (!nla)
1110 		return -EMSGSIZE;
1111 
1112 	stats = nla_data(nla);
1113 	mpls_get_stats(mdev, stats);
1114 
1115 	return 0;
1116 }
1117 
1118 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1119 {
1120 	struct mpls_dev *mdev;
1121 
1122 	mdev = mpls_dev_get(dev);
1123 	if (!mdev)
1124 		return 0;
1125 
1126 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1127 }
1128 
1129 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1130 				     u32 portid, u32 seq, int event,
1131 				     unsigned int flags, int type)
1132 {
1133 	struct nlmsghdr  *nlh;
1134 	struct netconfmsg *ncm;
1135 	bool all = false;
1136 
1137 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1138 			flags);
1139 	if (!nlh)
1140 		return -EMSGSIZE;
1141 
1142 	if (type == NETCONFA_ALL)
1143 		all = true;
1144 
1145 	ncm = nlmsg_data(nlh);
1146 	ncm->ncm_family = AF_MPLS;
1147 
1148 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1149 		goto nla_put_failure;
1150 
1151 	if ((all || type == NETCONFA_INPUT) &&
1152 	    nla_put_s32(skb, NETCONFA_INPUT,
1153 			mdev->input_enabled) < 0)
1154 		goto nla_put_failure;
1155 
1156 	nlmsg_end(skb, nlh);
1157 	return 0;
1158 
1159 nla_put_failure:
1160 	nlmsg_cancel(skb, nlh);
1161 	return -EMSGSIZE;
1162 }
1163 
1164 static int mpls_netconf_msgsize_devconf(int type)
1165 {
1166 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1167 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1168 	bool all = false;
1169 
1170 	if (type == NETCONFA_ALL)
1171 		all = true;
1172 
1173 	if (all || type == NETCONFA_INPUT)
1174 		size += nla_total_size(4);
1175 
1176 	return size;
1177 }
1178 
1179 static void mpls_netconf_notify_devconf(struct net *net, int event,
1180 					int type, struct mpls_dev *mdev)
1181 {
1182 	struct sk_buff *skb;
1183 	int err = -ENOBUFS;
1184 
1185 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1186 	if (!skb)
1187 		goto errout;
1188 
1189 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1190 	if (err < 0) {
1191 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1192 		WARN_ON(err == -EMSGSIZE);
1193 		kfree_skb(skb);
1194 		goto errout;
1195 	}
1196 
1197 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1198 	return;
1199 errout:
1200 	if (err < 0)
1201 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1202 }
1203 
1204 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1205 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1206 };
1207 
1208 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1209 				    struct nlmsghdr *nlh,
1210 				    struct netlink_ext_ack *extack)
1211 {
1212 	struct net *net = sock_net(in_skb->sk);
1213 	struct nlattr *tb[NETCONFA_MAX + 1];
1214 	struct netconfmsg *ncm;
1215 	struct net_device *dev;
1216 	struct mpls_dev *mdev;
1217 	struct sk_buff *skb;
1218 	int ifindex;
1219 	int err;
1220 
1221 	err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1222 			  devconf_mpls_policy, NULL);
1223 	if (err < 0)
1224 		goto errout;
1225 
1226 	err = -EINVAL;
1227 	if (!tb[NETCONFA_IFINDEX])
1228 		goto errout;
1229 
1230 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1231 	dev = __dev_get_by_index(net, ifindex);
1232 	if (!dev)
1233 		goto errout;
1234 
1235 	mdev = mpls_dev_get(dev);
1236 	if (!mdev)
1237 		goto errout;
1238 
1239 	err = -ENOBUFS;
1240 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1241 	if (!skb)
1242 		goto errout;
1243 
1244 	err = mpls_netconf_fill_devconf(skb, mdev,
1245 					NETLINK_CB(in_skb).portid,
1246 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1247 					NETCONFA_ALL);
1248 	if (err < 0) {
1249 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1250 		WARN_ON(err == -EMSGSIZE);
1251 		kfree_skb(skb);
1252 		goto errout;
1253 	}
1254 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1255 errout:
1256 	return err;
1257 }
1258 
1259 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1260 				     struct netlink_callback *cb)
1261 {
1262 	struct net *net = sock_net(skb->sk);
1263 	struct hlist_head *head;
1264 	struct net_device *dev;
1265 	struct mpls_dev *mdev;
1266 	int idx, s_idx;
1267 	int h, s_h;
1268 
1269 	s_h = cb->args[0];
1270 	s_idx = idx = cb->args[1];
1271 
1272 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1273 		idx = 0;
1274 		head = &net->dev_index_head[h];
1275 		rcu_read_lock();
1276 		cb->seq = net->dev_base_seq;
1277 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1278 			if (idx < s_idx)
1279 				goto cont;
1280 			mdev = mpls_dev_get(dev);
1281 			if (!mdev)
1282 				goto cont;
1283 			if (mpls_netconf_fill_devconf(skb, mdev,
1284 						      NETLINK_CB(cb->skb).portid,
1285 						      cb->nlh->nlmsg_seq,
1286 						      RTM_NEWNETCONF,
1287 						      NLM_F_MULTI,
1288 						      NETCONFA_ALL) < 0) {
1289 				rcu_read_unlock();
1290 				goto done;
1291 			}
1292 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1293 cont:
1294 			idx++;
1295 		}
1296 		rcu_read_unlock();
1297 	}
1298 done:
1299 	cb->args[0] = h;
1300 	cb->args[1] = idx;
1301 
1302 	return skb->len;
1303 }
1304 
1305 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1306 	(&((struct mpls_dev *)0)->field)
1307 
1308 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1309 			  void __user *buffer,
1310 			  size_t *lenp, loff_t *ppos)
1311 {
1312 	int oval = *(int *)ctl->data;
1313 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1314 
1315 	if (write) {
1316 		struct mpls_dev *mdev = ctl->extra1;
1317 		int i = (int *)ctl->data - (int *)mdev;
1318 		struct net *net = ctl->extra2;
1319 		int val = *(int *)ctl->data;
1320 
1321 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1322 		    val != oval) {
1323 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1324 						    NETCONFA_INPUT, mdev);
1325 		}
1326 	}
1327 
1328 	return ret;
1329 }
1330 
1331 static const struct ctl_table mpls_dev_table[] = {
1332 	{
1333 		.procname	= "input",
1334 		.maxlen		= sizeof(int),
1335 		.mode		= 0644,
1336 		.proc_handler	= mpls_conf_proc,
1337 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1338 	},
1339 	{ }
1340 };
1341 
1342 static int mpls_dev_sysctl_register(struct net_device *dev,
1343 				    struct mpls_dev *mdev)
1344 {
1345 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1346 	struct net *net = dev_net(dev);
1347 	struct ctl_table *table;
1348 	int i;
1349 
1350 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1351 	if (!table)
1352 		goto out;
1353 
1354 	/* Table data contains only offsets relative to the base of
1355 	 * the mdev at this point, so make them absolute.
1356 	 */
1357 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1358 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1359 		table[i].extra1 = mdev;
1360 		table[i].extra2 = net;
1361 	}
1362 
1363 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1364 
1365 	mdev->sysctl = register_net_sysctl(net, path, table);
1366 	if (!mdev->sysctl)
1367 		goto free;
1368 
1369 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1370 	return 0;
1371 
1372 free:
1373 	kfree(table);
1374 out:
1375 	return -ENOBUFS;
1376 }
1377 
1378 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1379 				       struct mpls_dev *mdev)
1380 {
1381 	struct net *net = dev_net(dev);
1382 	struct ctl_table *table;
1383 
1384 	table = mdev->sysctl->ctl_table_arg;
1385 	unregister_net_sysctl_table(mdev->sysctl);
1386 	kfree(table);
1387 
1388 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1389 }
1390 
1391 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1392 {
1393 	struct mpls_dev *mdev;
1394 	int err = -ENOMEM;
1395 	int i;
1396 
1397 	ASSERT_RTNL();
1398 
1399 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1400 	if (!mdev)
1401 		return ERR_PTR(err);
1402 
1403 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1404 	if (!mdev->stats)
1405 		goto free;
1406 
1407 	for_each_possible_cpu(i) {
1408 		struct mpls_pcpu_stats *mpls_stats;
1409 
1410 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1411 		u64_stats_init(&mpls_stats->syncp);
1412 	}
1413 
1414 	mdev->dev = dev;
1415 
1416 	err = mpls_dev_sysctl_register(dev, mdev);
1417 	if (err)
1418 		goto free;
1419 
1420 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1421 
1422 	return mdev;
1423 
1424 free:
1425 	free_percpu(mdev->stats);
1426 	kfree(mdev);
1427 	return ERR_PTR(err);
1428 }
1429 
1430 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1431 {
1432 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1433 
1434 	free_percpu(mdev->stats);
1435 	kfree(mdev);
1436 }
1437 
1438 static void mpls_ifdown(struct net_device *dev, int event)
1439 {
1440 	struct mpls_route __rcu **platform_label;
1441 	struct net *net = dev_net(dev);
1442 	u8 alive, deleted;
1443 	unsigned index;
1444 
1445 	platform_label = rtnl_dereference(net->mpls.platform_label);
1446 	for (index = 0; index < net->mpls.platform_labels; index++) {
1447 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1448 
1449 		if (!rt)
1450 			continue;
1451 
1452 		alive = 0;
1453 		deleted = 0;
1454 		change_nexthops(rt) {
1455 			unsigned int nh_flags = nh->nh_flags;
1456 
1457 			if (rtnl_dereference(nh->nh_dev) != dev)
1458 				goto next;
1459 
1460 			switch (event) {
1461 			case NETDEV_DOWN:
1462 			case NETDEV_UNREGISTER:
1463 				nh_flags |= RTNH_F_DEAD;
1464 				/* fall through */
1465 			case NETDEV_CHANGE:
1466 				nh_flags |= RTNH_F_LINKDOWN;
1467 				break;
1468 			}
1469 			if (event == NETDEV_UNREGISTER)
1470 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1471 
1472 			if (nh->nh_flags != nh_flags)
1473 				WRITE_ONCE(nh->nh_flags, nh_flags);
1474 next:
1475 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1476 				alive++;
1477 			if (!rtnl_dereference(nh->nh_dev))
1478 				deleted++;
1479 		} endfor_nexthops(rt);
1480 
1481 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1482 
1483 		/* if there are no more nexthops, delete the route */
1484 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1485 			mpls_route_update(net, index, NULL, NULL);
1486 	}
1487 }
1488 
1489 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1490 {
1491 	struct mpls_route __rcu **platform_label;
1492 	struct net *net = dev_net(dev);
1493 	unsigned index;
1494 	u8 alive;
1495 
1496 	platform_label = rtnl_dereference(net->mpls.platform_label);
1497 	for (index = 0; index < net->mpls.platform_labels; index++) {
1498 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1499 
1500 		if (!rt)
1501 			continue;
1502 
1503 		alive = 0;
1504 		change_nexthops(rt) {
1505 			unsigned int nh_flags = nh->nh_flags;
1506 			struct net_device *nh_dev =
1507 				rtnl_dereference(nh->nh_dev);
1508 
1509 			if (!(nh_flags & flags)) {
1510 				alive++;
1511 				continue;
1512 			}
1513 			if (nh_dev != dev)
1514 				continue;
1515 			alive++;
1516 			nh_flags &= ~flags;
1517 			WRITE_ONCE(nh->nh_flags, nh_flags);
1518 		} endfor_nexthops(rt);
1519 
1520 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1521 	}
1522 }
1523 
1524 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1525 			   void *ptr)
1526 {
1527 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1528 	struct mpls_dev *mdev;
1529 	unsigned int flags;
1530 
1531 	if (event == NETDEV_REGISTER) {
1532 		/* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1533 		if (dev->type == ARPHRD_ETHER ||
1534 		    dev->type == ARPHRD_LOOPBACK ||
1535 		    dev->type == ARPHRD_IPGRE ||
1536 		    dev->type == ARPHRD_SIT ||
1537 		    dev->type == ARPHRD_TUNNEL) {
1538 			mdev = mpls_add_dev(dev);
1539 			if (IS_ERR(mdev))
1540 				return notifier_from_errno(PTR_ERR(mdev));
1541 		}
1542 		return NOTIFY_OK;
1543 	}
1544 
1545 	mdev = mpls_dev_get(dev);
1546 	if (!mdev)
1547 		return NOTIFY_OK;
1548 
1549 	switch (event) {
1550 	case NETDEV_DOWN:
1551 		mpls_ifdown(dev, event);
1552 		break;
1553 	case NETDEV_UP:
1554 		flags = dev_get_flags(dev);
1555 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1556 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1557 		else
1558 			mpls_ifup(dev, RTNH_F_DEAD);
1559 		break;
1560 	case NETDEV_CHANGE:
1561 		flags = dev_get_flags(dev);
1562 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1563 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1564 		else
1565 			mpls_ifdown(dev, event);
1566 		break;
1567 	case NETDEV_UNREGISTER:
1568 		mpls_ifdown(dev, event);
1569 		mdev = mpls_dev_get(dev);
1570 		if (mdev) {
1571 			mpls_dev_sysctl_unregister(dev, mdev);
1572 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1573 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1574 		}
1575 		break;
1576 	case NETDEV_CHANGENAME:
1577 		mdev = mpls_dev_get(dev);
1578 		if (mdev) {
1579 			int err;
1580 
1581 			mpls_dev_sysctl_unregister(dev, mdev);
1582 			err = mpls_dev_sysctl_register(dev, mdev);
1583 			if (err)
1584 				return notifier_from_errno(err);
1585 		}
1586 		break;
1587 	}
1588 	return NOTIFY_OK;
1589 }
1590 
1591 static struct notifier_block mpls_dev_notifier = {
1592 	.notifier_call = mpls_dev_notify,
1593 };
1594 
1595 static int nla_put_via(struct sk_buff *skb,
1596 		       u8 table, const void *addr, int alen)
1597 {
1598 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1599 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1600 	};
1601 	struct nlattr *nla;
1602 	struct rtvia *via;
1603 	int family = AF_UNSPEC;
1604 
1605 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1606 	if (!nla)
1607 		return -EMSGSIZE;
1608 
1609 	if (table <= NEIGH_NR_TABLES)
1610 		family = table_to_family[table];
1611 
1612 	via = nla_data(nla);
1613 	via->rtvia_family = family;
1614 	memcpy(via->rtvia_addr, addr, alen);
1615 	return 0;
1616 }
1617 
1618 int nla_put_labels(struct sk_buff *skb, int attrtype,
1619 		   u8 labels, const u32 label[])
1620 {
1621 	struct nlattr *nla;
1622 	struct mpls_shim_hdr *nla_label;
1623 	bool bos;
1624 	int i;
1625 	nla = nla_reserve(skb, attrtype, labels*4);
1626 	if (!nla)
1627 		return -EMSGSIZE;
1628 
1629 	nla_label = nla_data(nla);
1630 	bos = true;
1631 	for (i = labels - 1; i >= 0; i--) {
1632 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1633 		bos = false;
1634 	}
1635 
1636 	return 0;
1637 }
1638 EXPORT_SYMBOL_GPL(nla_put_labels);
1639 
1640 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1641 		   u32 label[], struct netlink_ext_ack *extack)
1642 {
1643 	unsigned len = nla_len(nla);
1644 	struct mpls_shim_hdr *nla_label;
1645 	u8 nla_labels;
1646 	bool bos;
1647 	int i;
1648 
1649 	/* len needs to be an even multiple of 4 (the label size). Number
1650 	 * of labels is a u8 so check for overflow.
1651 	 */
1652 	if (len & 3 || len / 4 > 255) {
1653 		NL_SET_ERR_MSG_ATTR(extack, nla,
1654 				    "Invalid length for labels attribute");
1655 		return -EINVAL;
1656 	}
1657 
1658 	/* Limit the number of new labels allowed */
1659 	nla_labels = len/4;
1660 	if (nla_labels > max_labels) {
1661 		NL_SET_ERR_MSG(extack, "Too many labels");
1662 		return -EINVAL;
1663 	}
1664 
1665 	/* when label == NULL, caller wants number of labels */
1666 	if (!label)
1667 		goto out;
1668 
1669 	nla_label = nla_data(nla);
1670 	bos = true;
1671 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1672 		struct mpls_entry_decoded dec;
1673 		dec = mpls_entry_decode(nla_label + i);
1674 
1675 		/* Ensure the bottom of stack flag is properly set
1676 		 * and ttl and tc are both clear.
1677 		 */
1678 		if (dec.ttl) {
1679 			NL_SET_ERR_MSG_ATTR(extack, nla,
1680 					    "TTL in label must be 0");
1681 			return -EINVAL;
1682 		}
1683 
1684 		if (dec.tc) {
1685 			NL_SET_ERR_MSG_ATTR(extack, nla,
1686 					    "Traffic class in label must be 0");
1687 			return -EINVAL;
1688 		}
1689 
1690 		if (dec.bos != bos) {
1691 			NL_SET_BAD_ATTR(extack, nla);
1692 			if (bos) {
1693 				NL_SET_ERR_MSG(extack,
1694 					       "BOS bit must be set in first label");
1695 			} else {
1696 				NL_SET_ERR_MSG(extack,
1697 					       "BOS bit can only be set in first label");
1698 			}
1699 			return -EINVAL;
1700 		}
1701 
1702 		switch (dec.label) {
1703 		case MPLS_LABEL_IMPLNULL:
1704 			/* RFC3032: This is a label that an LSR may
1705 			 * assign and distribute, but which never
1706 			 * actually appears in the encapsulation.
1707 			 */
1708 			NL_SET_ERR_MSG_ATTR(extack, nla,
1709 					    "Implicit NULL Label (3) can not be used in encapsulation");
1710 			return -EINVAL;
1711 		}
1712 
1713 		label[i] = dec.label;
1714 	}
1715 out:
1716 	*labels = nla_labels;
1717 	return 0;
1718 }
1719 EXPORT_SYMBOL_GPL(nla_get_labels);
1720 
1721 static int rtm_to_route_config(struct sk_buff *skb,
1722 			       struct nlmsghdr *nlh,
1723 			       struct mpls_route_config *cfg,
1724 			       struct netlink_ext_ack *extack)
1725 {
1726 	struct rtmsg *rtm;
1727 	struct nlattr *tb[RTA_MAX+1];
1728 	int index;
1729 	int err;
1730 
1731 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy,
1732 			  extack);
1733 	if (err < 0)
1734 		goto errout;
1735 
1736 	err = -EINVAL;
1737 	rtm = nlmsg_data(nlh);
1738 
1739 	if (rtm->rtm_family != AF_MPLS) {
1740 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1741 		goto errout;
1742 	}
1743 	if (rtm->rtm_dst_len != 20) {
1744 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1745 		goto errout;
1746 	}
1747 	if (rtm->rtm_src_len != 0) {
1748 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1749 		goto errout;
1750 	}
1751 	if (rtm->rtm_tos != 0) {
1752 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1753 		goto errout;
1754 	}
1755 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1756 		NL_SET_ERR_MSG(extack,
1757 			       "MPLS only supports the main route table");
1758 		goto errout;
1759 	}
1760 	/* Any value is acceptable for rtm_protocol */
1761 
1762 	/* As mpls uses destination specific addresses
1763 	 * (or source specific address in the case of multicast)
1764 	 * all addresses have universal scope.
1765 	 */
1766 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1767 		NL_SET_ERR_MSG(extack,
1768 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1769 		goto errout;
1770 	}
1771 	if (rtm->rtm_type != RTN_UNICAST) {
1772 		NL_SET_ERR_MSG(extack,
1773 			       "Invalid route type - MPLS only supports UNICAST");
1774 		goto errout;
1775 	}
1776 	if (rtm->rtm_flags != 0) {
1777 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1778 		goto errout;
1779 	}
1780 
1781 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1782 	cfg->rc_protocol	= rtm->rtm_protocol;
1783 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1784 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1785 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1786 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1787 	cfg->rc_nlinfo.nlh	= nlh;
1788 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1789 
1790 	for (index = 0; index <= RTA_MAX; index++) {
1791 		struct nlattr *nla = tb[index];
1792 		if (!nla)
1793 			continue;
1794 
1795 		switch (index) {
1796 		case RTA_OIF:
1797 			cfg->rc_ifindex = nla_get_u32(nla);
1798 			break;
1799 		case RTA_NEWDST:
1800 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1801 					   &cfg->rc_output_labels,
1802 					   cfg->rc_output_label, extack))
1803 				goto errout;
1804 			break;
1805 		case RTA_DST:
1806 		{
1807 			u8 label_count;
1808 			if (nla_get_labels(nla, 1, &label_count,
1809 					   &cfg->rc_label, extack))
1810 				goto errout;
1811 
1812 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1813 					   cfg->rc_label, extack))
1814 				goto errout;
1815 			break;
1816 		}
1817 		case RTA_VIA:
1818 		{
1819 			if (nla_get_via(nla, &cfg->rc_via_alen,
1820 					&cfg->rc_via_table, cfg->rc_via,
1821 					extack))
1822 				goto errout;
1823 			break;
1824 		}
1825 		case RTA_MULTIPATH:
1826 		{
1827 			cfg->rc_mp = nla_data(nla);
1828 			cfg->rc_mp_len = nla_len(nla);
1829 			break;
1830 		}
1831 		case RTA_TTL_PROPAGATE:
1832 		{
1833 			u8 ttl_propagate = nla_get_u8(nla);
1834 
1835 			if (ttl_propagate > 1) {
1836 				NL_SET_ERR_MSG_ATTR(extack, nla,
1837 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1838 				goto errout;
1839 			}
1840 			cfg->rc_ttl_propagate = ttl_propagate ?
1841 				MPLS_TTL_PROP_ENABLED :
1842 				MPLS_TTL_PROP_DISABLED;
1843 			break;
1844 		}
1845 		default:
1846 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1847 			/* Unsupported attribute */
1848 			goto errout;
1849 		}
1850 	}
1851 
1852 	err = 0;
1853 errout:
1854 	return err;
1855 }
1856 
1857 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1858 			     struct netlink_ext_ack *extack)
1859 {
1860 	struct mpls_route_config *cfg;
1861 	int err;
1862 
1863 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1864 	if (!cfg)
1865 		return -ENOMEM;
1866 
1867 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1868 	if (err < 0)
1869 		goto out;
1870 
1871 	err = mpls_route_del(cfg, extack);
1872 out:
1873 	kfree(cfg);
1874 
1875 	return err;
1876 }
1877 
1878 
1879 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1880 			     struct netlink_ext_ack *extack)
1881 {
1882 	struct mpls_route_config *cfg;
1883 	int err;
1884 
1885 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1886 	if (!cfg)
1887 		return -ENOMEM;
1888 
1889 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1890 	if (err < 0)
1891 		goto out;
1892 
1893 	err = mpls_route_add(cfg, extack);
1894 out:
1895 	kfree(cfg);
1896 
1897 	return err;
1898 }
1899 
1900 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1901 			   u32 label, struct mpls_route *rt, int flags)
1902 {
1903 	struct net_device *dev;
1904 	struct nlmsghdr *nlh;
1905 	struct rtmsg *rtm;
1906 
1907 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1908 	if (nlh == NULL)
1909 		return -EMSGSIZE;
1910 
1911 	rtm = nlmsg_data(nlh);
1912 	rtm->rtm_family = AF_MPLS;
1913 	rtm->rtm_dst_len = 20;
1914 	rtm->rtm_src_len = 0;
1915 	rtm->rtm_tos = 0;
1916 	rtm->rtm_table = RT_TABLE_MAIN;
1917 	rtm->rtm_protocol = rt->rt_protocol;
1918 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1919 	rtm->rtm_type = RTN_UNICAST;
1920 	rtm->rtm_flags = 0;
1921 
1922 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1923 		goto nla_put_failure;
1924 
1925 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1926 		bool ttl_propagate =
1927 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1928 
1929 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1930 			       ttl_propagate))
1931 			goto nla_put_failure;
1932 	}
1933 	if (rt->rt_nhn == 1) {
1934 		const struct mpls_nh *nh = rt->rt_nh;
1935 
1936 		if (nh->nh_labels &&
1937 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1938 				   nh->nh_label))
1939 			goto nla_put_failure;
1940 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1941 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1942 				nh->nh_via_alen))
1943 			goto nla_put_failure;
1944 		dev = rtnl_dereference(nh->nh_dev);
1945 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1946 			goto nla_put_failure;
1947 		if (nh->nh_flags & RTNH_F_LINKDOWN)
1948 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1949 		if (nh->nh_flags & RTNH_F_DEAD)
1950 			rtm->rtm_flags |= RTNH_F_DEAD;
1951 	} else {
1952 		struct rtnexthop *rtnh;
1953 		struct nlattr *mp;
1954 		u8 linkdown = 0;
1955 		u8 dead = 0;
1956 
1957 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1958 		if (!mp)
1959 			goto nla_put_failure;
1960 
1961 		for_nexthops(rt) {
1962 			dev = rtnl_dereference(nh->nh_dev);
1963 			if (!dev)
1964 				continue;
1965 
1966 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1967 			if (!rtnh)
1968 				goto nla_put_failure;
1969 
1970 			rtnh->rtnh_ifindex = dev->ifindex;
1971 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1972 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1973 				linkdown++;
1974 			}
1975 			if (nh->nh_flags & RTNH_F_DEAD) {
1976 				rtnh->rtnh_flags |= RTNH_F_DEAD;
1977 				dead++;
1978 			}
1979 
1980 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1981 							    nh->nh_labels,
1982 							    nh->nh_label))
1983 				goto nla_put_failure;
1984 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1985 			    nla_put_via(skb, nh->nh_via_table,
1986 					mpls_nh_via(rt, nh),
1987 					nh->nh_via_alen))
1988 				goto nla_put_failure;
1989 
1990 			/* length of rtnetlink header + attributes */
1991 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1992 		} endfor_nexthops(rt);
1993 
1994 		if (linkdown == rt->rt_nhn)
1995 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1996 		if (dead == rt->rt_nhn)
1997 			rtm->rtm_flags |= RTNH_F_DEAD;
1998 
1999 		nla_nest_end(skb, mp);
2000 	}
2001 
2002 	nlmsg_end(skb, nlh);
2003 	return 0;
2004 
2005 nla_put_failure:
2006 	nlmsg_cancel(skb, nlh);
2007 	return -EMSGSIZE;
2008 }
2009 
2010 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2011 {
2012 	struct net *net = sock_net(skb->sk);
2013 	struct mpls_route __rcu **platform_label;
2014 	size_t platform_labels;
2015 	unsigned int index;
2016 
2017 	ASSERT_RTNL();
2018 
2019 	index = cb->args[0];
2020 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2021 		index = MPLS_LABEL_FIRST_UNRESERVED;
2022 
2023 	platform_label = rtnl_dereference(net->mpls.platform_label);
2024 	platform_labels = net->mpls.platform_labels;
2025 	for (; index < platform_labels; index++) {
2026 		struct mpls_route *rt;
2027 		rt = rtnl_dereference(platform_label[index]);
2028 		if (!rt)
2029 			continue;
2030 
2031 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2032 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2033 				    index, rt, NLM_F_MULTI) < 0)
2034 			break;
2035 	}
2036 	cb->args[0] = index;
2037 
2038 	return skb->len;
2039 }
2040 
2041 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2042 {
2043 	size_t payload =
2044 		NLMSG_ALIGN(sizeof(struct rtmsg))
2045 		+ nla_total_size(4)			/* RTA_DST */
2046 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2047 
2048 	if (rt->rt_nhn == 1) {
2049 		struct mpls_nh *nh = rt->rt_nh;
2050 
2051 		if (nh->nh_dev)
2052 			payload += nla_total_size(4); /* RTA_OIF */
2053 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2054 			payload += nla_total_size(2 + nh->nh_via_alen);
2055 		if (nh->nh_labels) /* RTA_NEWDST */
2056 			payload += nla_total_size(nh->nh_labels * 4);
2057 	} else {
2058 		/* each nexthop is packed in an attribute */
2059 		size_t nhsize = 0;
2060 
2061 		for_nexthops(rt) {
2062 			if (!rtnl_dereference(nh->nh_dev))
2063 				continue;
2064 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2065 			/* RTA_VIA */
2066 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2067 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2068 			if (nh->nh_labels)
2069 				nhsize += nla_total_size(nh->nh_labels * 4);
2070 		} endfor_nexthops(rt);
2071 		/* nested attribute */
2072 		payload += nla_total_size(nhsize);
2073 	}
2074 
2075 	return payload;
2076 }
2077 
2078 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2079 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2080 		       unsigned int nlm_flags)
2081 {
2082 	struct sk_buff *skb;
2083 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2084 	int err = -ENOBUFS;
2085 
2086 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2087 	if (skb == NULL)
2088 		goto errout;
2089 
2090 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2091 	if (err < 0) {
2092 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2093 		WARN_ON(err == -EMSGSIZE);
2094 		kfree_skb(skb);
2095 		goto errout;
2096 	}
2097 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2098 
2099 	return;
2100 errout:
2101 	if (err < 0)
2102 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2103 }
2104 
2105 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2106 			 struct netlink_ext_ack *extack)
2107 {
2108 	struct net *net = sock_net(in_skb->sk);
2109 	u32 portid = NETLINK_CB(in_skb).portid;
2110 	u32 in_label = LABEL_NOT_SPECIFIED;
2111 	struct nlattr *tb[RTA_MAX + 1];
2112 	u32 labels[MAX_NEW_LABELS];
2113 	struct mpls_shim_hdr *hdr;
2114 	unsigned int hdr_size = 0;
2115 	struct net_device *dev;
2116 	struct mpls_route *rt;
2117 	struct rtmsg *rtm, *r;
2118 	struct nlmsghdr *nlh;
2119 	struct sk_buff *skb;
2120 	struct mpls_nh *nh;
2121 	u8 n_labels;
2122 	int err;
2123 
2124 	err = nlmsg_parse(in_nlh, sizeof(*rtm), tb, RTA_MAX,
2125 			  rtm_mpls_policy, extack);
2126 	if (err < 0)
2127 		goto errout;
2128 
2129 	rtm = nlmsg_data(in_nlh);
2130 
2131 	if (tb[RTA_DST]) {
2132 		u8 label_count;
2133 
2134 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2135 				   &in_label, extack)) {
2136 			err = -EINVAL;
2137 			goto errout;
2138 		}
2139 
2140 		if (!mpls_label_ok(net, in_label, extack)) {
2141 			err = -EINVAL;
2142 			goto errout;
2143 		}
2144 	}
2145 
2146 	rt = mpls_route_input_rcu(net, in_label);
2147 	if (!rt) {
2148 		err = -ENETUNREACH;
2149 		goto errout;
2150 	}
2151 
2152 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2153 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2154 		if (!skb) {
2155 			err = -ENOBUFS;
2156 			goto errout;
2157 		}
2158 
2159 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2160 				      RTM_NEWROUTE, in_label, rt, 0);
2161 		if (err < 0) {
2162 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2163 			WARN_ON(err == -EMSGSIZE);
2164 			goto errout_free;
2165 		}
2166 
2167 		return rtnl_unicast(skb, net, portid);
2168 	}
2169 
2170 	if (tb[RTA_NEWDST]) {
2171 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2172 				   labels, extack) != 0) {
2173 			err = -EINVAL;
2174 			goto errout;
2175 		}
2176 
2177 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2178 	}
2179 
2180 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2181 	if (!skb) {
2182 		err = -ENOBUFS;
2183 		goto errout;
2184 	}
2185 
2186 	skb->protocol = htons(ETH_P_MPLS_UC);
2187 
2188 	if (hdr_size) {
2189 		bool bos;
2190 		int i;
2191 
2192 		if (skb_cow(skb, hdr_size)) {
2193 			err = -ENOBUFS;
2194 			goto errout_free;
2195 		}
2196 
2197 		skb_reserve(skb, hdr_size);
2198 		skb_push(skb, hdr_size);
2199 		skb_reset_network_header(skb);
2200 
2201 		/* Push new labels */
2202 		hdr = mpls_hdr(skb);
2203 		bos = true;
2204 		for (i = n_labels - 1; i >= 0; i--) {
2205 			hdr[i] = mpls_entry_encode(labels[i],
2206 						   1, 0, bos);
2207 			bos = false;
2208 		}
2209 	}
2210 
2211 	nh = mpls_select_multipath(rt, skb);
2212 	if (!nh) {
2213 		err = -ENETUNREACH;
2214 		goto errout_free;
2215 	}
2216 
2217 	if (hdr_size) {
2218 		skb_pull(skb, hdr_size);
2219 		skb_reset_network_header(skb);
2220 	}
2221 
2222 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2223 			RTM_NEWROUTE, sizeof(*r), 0);
2224 	if (!nlh) {
2225 		err = -EMSGSIZE;
2226 		goto errout_free;
2227 	}
2228 
2229 	r = nlmsg_data(nlh);
2230 	r->rtm_family	 = AF_MPLS;
2231 	r->rtm_dst_len	= 20;
2232 	r->rtm_src_len	= 0;
2233 	r->rtm_table	= RT_TABLE_MAIN;
2234 	r->rtm_type	= RTN_UNICAST;
2235 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2236 	r->rtm_protocol = rt->rt_protocol;
2237 	r->rtm_flags	= 0;
2238 
2239 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2240 		goto nla_put_failure;
2241 
2242 	if (nh->nh_labels &&
2243 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2244 			   nh->nh_label))
2245 		goto nla_put_failure;
2246 
2247 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2248 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2249 			nh->nh_via_alen))
2250 		goto nla_put_failure;
2251 	dev = rtnl_dereference(nh->nh_dev);
2252 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2253 		goto nla_put_failure;
2254 
2255 	nlmsg_end(skb, nlh);
2256 
2257 	err = rtnl_unicast(skb, net, portid);
2258 errout:
2259 	return err;
2260 
2261 nla_put_failure:
2262 	nlmsg_cancel(skb, nlh);
2263 	err = -EMSGSIZE;
2264 errout_free:
2265 	kfree_skb(skb);
2266 	return err;
2267 }
2268 
2269 static int resize_platform_label_table(struct net *net, size_t limit)
2270 {
2271 	size_t size = sizeof(struct mpls_route *) * limit;
2272 	size_t old_limit;
2273 	size_t cp_size;
2274 	struct mpls_route __rcu **labels = NULL, **old;
2275 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2276 	unsigned index;
2277 
2278 	if (size) {
2279 		labels = kvzalloc(size, GFP_KERNEL);
2280 		if (!labels)
2281 			goto nolabels;
2282 	}
2283 
2284 	/* In case the predefined labels need to be populated */
2285 	if (limit > MPLS_LABEL_IPV4NULL) {
2286 		struct net_device *lo = net->loopback_dev;
2287 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2288 		if (IS_ERR(rt0))
2289 			goto nort0;
2290 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2291 		rt0->rt_protocol = RTPROT_KERNEL;
2292 		rt0->rt_payload_type = MPT_IPV4;
2293 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2294 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2295 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2296 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2297 		       lo->addr_len);
2298 	}
2299 	if (limit > MPLS_LABEL_IPV6NULL) {
2300 		struct net_device *lo = net->loopback_dev;
2301 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2302 		if (IS_ERR(rt2))
2303 			goto nort2;
2304 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2305 		rt2->rt_protocol = RTPROT_KERNEL;
2306 		rt2->rt_payload_type = MPT_IPV6;
2307 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2308 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2309 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2310 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2311 		       lo->addr_len);
2312 	}
2313 
2314 	rtnl_lock();
2315 	/* Remember the original table */
2316 	old = rtnl_dereference(net->mpls.platform_label);
2317 	old_limit = net->mpls.platform_labels;
2318 
2319 	/* Free any labels beyond the new table */
2320 	for (index = limit; index < old_limit; index++)
2321 		mpls_route_update(net, index, NULL, NULL);
2322 
2323 	/* Copy over the old labels */
2324 	cp_size = size;
2325 	if (old_limit < limit)
2326 		cp_size = old_limit * sizeof(struct mpls_route *);
2327 
2328 	memcpy(labels, old, cp_size);
2329 
2330 	/* If needed set the predefined labels */
2331 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2332 	    (limit > MPLS_LABEL_IPV6NULL)) {
2333 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2334 		rt2 = NULL;
2335 	}
2336 
2337 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2338 	    (limit > MPLS_LABEL_IPV4NULL)) {
2339 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2340 		rt0 = NULL;
2341 	}
2342 
2343 	/* Update the global pointers */
2344 	net->mpls.platform_labels = limit;
2345 	rcu_assign_pointer(net->mpls.platform_label, labels);
2346 
2347 	rtnl_unlock();
2348 
2349 	mpls_rt_free(rt2);
2350 	mpls_rt_free(rt0);
2351 
2352 	if (old) {
2353 		synchronize_rcu();
2354 		kvfree(old);
2355 	}
2356 	return 0;
2357 
2358 nort2:
2359 	mpls_rt_free(rt0);
2360 nort0:
2361 	kvfree(labels);
2362 nolabels:
2363 	return -ENOMEM;
2364 }
2365 
2366 static int mpls_platform_labels(struct ctl_table *table, int write,
2367 				void __user *buffer, size_t *lenp, loff_t *ppos)
2368 {
2369 	struct net *net = table->data;
2370 	int platform_labels = net->mpls.platform_labels;
2371 	int ret;
2372 	struct ctl_table tmp = {
2373 		.procname	= table->procname,
2374 		.data		= &platform_labels,
2375 		.maxlen		= sizeof(int),
2376 		.mode		= table->mode,
2377 		.extra1		= &zero,
2378 		.extra2		= &label_limit,
2379 	};
2380 
2381 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2382 
2383 	if (write && ret == 0)
2384 		ret = resize_platform_label_table(net, platform_labels);
2385 
2386 	return ret;
2387 }
2388 
2389 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2390 	(&((struct net *)0)->field)
2391 
2392 static const struct ctl_table mpls_table[] = {
2393 	{
2394 		.procname	= "platform_labels",
2395 		.data		= NULL,
2396 		.maxlen		= sizeof(int),
2397 		.mode		= 0644,
2398 		.proc_handler	= mpls_platform_labels,
2399 	},
2400 	{
2401 		.procname	= "ip_ttl_propagate",
2402 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2403 		.maxlen		= sizeof(int),
2404 		.mode		= 0644,
2405 		.proc_handler	= proc_dointvec_minmax,
2406 		.extra1		= &zero,
2407 		.extra2		= &one,
2408 	},
2409 	{
2410 		.procname	= "default_ttl",
2411 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2412 		.maxlen		= sizeof(int),
2413 		.mode		= 0644,
2414 		.proc_handler	= proc_dointvec_minmax,
2415 		.extra1		= &one,
2416 		.extra2		= &ttl_max,
2417 	},
2418 	{ }
2419 };
2420 
2421 static int mpls_net_init(struct net *net)
2422 {
2423 	struct ctl_table *table;
2424 	int i;
2425 
2426 	net->mpls.platform_labels = 0;
2427 	net->mpls.platform_label = NULL;
2428 	net->mpls.ip_ttl_propagate = 1;
2429 	net->mpls.default_ttl = 255;
2430 
2431 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2432 	if (table == NULL)
2433 		return -ENOMEM;
2434 
2435 	/* Table data contains only offsets relative to the base of
2436 	 * the mdev at this point, so make them absolute.
2437 	 */
2438 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2439 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2440 
2441 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2442 	if (net->mpls.ctl == NULL) {
2443 		kfree(table);
2444 		return -ENOMEM;
2445 	}
2446 
2447 	return 0;
2448 }
2449 
2450 static void mpls_net_exit(struct net *net)
2451 {
2452 	struct mpls_route __rcu **platform_label;
2453 	size_t platform_labels;
2454 	struct ctl_table *table;
2455 	unsigned int index;
2456 
2457 	table = net->mpls.ctl->ctl_table_arg;
2458 	unregister_net_sysctl_table(net->mpls.ctl);
2459 	kfree(table);
2460 
2461 	/* An rcu grace period has passed since there was a device in
2462 	 * the network namespace (and thus the last in flight packet)
2463 	 * left this network namespace.  This is because
2464 	 * unregister_netdevice_many and netdev_run_todo has completed
2465 	 * for each network device that was in this network namespace.
2466 	 *
2467 	 * As such no additional rcu synchronization is necessary when
2468 	 * freeing the platform_label table.
2469 	 */
2470 	rtnl_lock();
2471 	platform_label = rtnl_dereference(net->mpls.platform_label);
2472 	platform_labels = net->mpls.platform_labels;
2473 	for (index = 0; index < platform_labels; index++) {
2474 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2475 		RCU_INIT_POINTER(platform_label[index], NULL);
2476 		mpls_notify_route(net, index, rt, NULL, NULL);
2477 		mpls_rt_free(rt);
2478 	}
2479 	rtnl_unlock();
2480 
2481 	kvfree(platform_label);
2482 }
2483 
2484 static struct pernet_operations mpls_net_ops = {
2485 	.init = mpls_net_init,
2486 	.exit = mpls_net_exit,
2487 };
2488 
2489 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2490 	.family		   = AF_MPLS,
2491 	.fill_stats_af	   = mpls_fill_stats_af,
2492 	.get_stats_af_size = mpls_get_stats_af_size,
2493 };
2494 
2495 static int __init mpls_init(void)
2496 {
2497 	int err;
2498 
2499 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2500 
2501 	err = register_pernet_subsys(&mpls_net_ops);
2502 	if (err)
2503 		goto out;
2504 
2505 	err = register_netdevice_notifier(&mpls_dev_notifier);
2506 	if (err)
2507 		goto out_unregister_pernet;
2508 
2509 	dev_add_pack(&mpls_packet_type);
2510 
2511 	rtnl_af_register(&mpls_af_ops);
2512 
2513 	rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, 0);
2514 	rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, 0);
2515 	rtnl_register(PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes,
2516 		      0);
2517 	rtnl_register(PF_MPLS, RTM_GETNETCONF, mpls_netconf_get_devconf,
2518 		      mpls_netconf_dump_devconf, 0);
2519 	err = ipgre_tunnel_encap_add_mpls_ops();
2520 	if (err)
2521 		pr_err("Can't add mpls over gre tunnel ops\n");
2522 
2523 	err = 0;
2524 out:
2525 	return err;
2526 
2527 out_unregister_pernet:
2528 	unregister_pernet_subsys(&mpls_net_ops);
2529 	goto out;
2530 }
2531 module_init(mpls_init);
2532 
2533 static void __exit mpls_exit(void)
2534 {
2535 	rtnl_unregister_all(PF_MPLS);
2536 	rtnl_af_unregister(&mpls_af_ops);
2537 	dev_remove_pack(&mpls_packet_type);
2538 	unregister_netdevice_notifier(&mpls_dev_notifier);
2539 	unregister_pernet_subsys(&mpls_net_ops);
2540 	ipgre_tunnel_encap_del_mpls_ops();
2541 }
2542 module_exit(mpls_exit);
2543 
2544 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2545 MODULE_LICENSE("GPL v2");
2546 MODULE_ALIAS_NETPROTO(PF_MPLS);
2547