xref: /openbmc/linux/net/mpls/af_mpls.c (revision c67ce71d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/types.h>
3 #include <linux/skbuff.h>
4 #include <linux/socket.h>
5 #include <linux/sysctl.h>
6 #include <linux/net.h>
7 #include <linux/module.h>
8 #include <linux/if_arp.h>
9 #include <linux/ipv6.h>
10 #include <linux/mpls.h>
11 #include <linux/netconf.h>
12 #include <linux/nospec.h>
13 #include <linux/vmalloc.h>
14 #include <linux/percpu.h>
15 #include <net/gso.h>
16 #include <net/ip.h>
17 #include <net/dst.h>
18 #include <net/sock.h>
19 #include <net/arp.h>
20 #include <net/ip_fib.h>
21 #include <net/netevent.h>
22 #include <net/ip_tunnels.h>
23 #include <net/netns/generic.h>
24 #if IS_ENABLED(CONFIG_IPV6)
25 #include <net/ipv6.h>
26 #endif
27 #include <net/ipv6_stubs.h>
28 #include <net/rtnh.h>
29 #include "internal.h"
30 
31 /* max memory we will use for mpls_route */
32 #define MAX_MPLS_ROUTE_MEM	4096
33 
34 /* Maximum number of labels to look ahead at when selecting a path of
35  * a multipath route
36  */
37 #define MAX_MP_SELECT_LABELS 4
38 
39 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
40 
41 static int label_limit = (1 << 20) - 1;
42 static int ttl_max = 255;
43 
44 #if IS_ENABLED(CONFIG_NET_IP_TUNNEL)
45 static size_t ipgre_mpls_encap_hlen(struct ip_tunnel_encap *e)
46 {
47 	return sizeof(struct mpls_shim_hdr);
48 }
49 
50 static const struct ip_tunnel_encap_ops mpls_iptun_ops = {
51 	.encap_hlen	= ipgre_mpls_encap_hlen,
52 };
53 
54 static int ipgre_tunnel_encap_add_mpls_ops(void)
55 {
56 	return ip_tunnel_encap_add_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
57 }
58 
59 static void ipgre_tunnel_encap_del_mpls_ops(void)
60 {
61 	ip_tunnel_encap_del_ops(&mpls_iptun_ops, TUNNEL_ENCAP_MPLS);
62 }
63 #else
64 static int ipgre_tunnel_encap_add_mpls_ops(void)
65 {
66 	return 0;
67 }
68 
69 static void ipgre_tunnel_encap_del_mpls_ops(void)
70 {
71 }
72 #endif
73 
74 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
75 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
76 		       unsigned int nlm_flags);
77 
78 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
79 {
80 	struct mpls_route *rt = NULL;
81 
82 	if (index < net->mpls.platform_labels) {
83 		struct mpls_route __rcu **platform_label =
84 			rcu_dereference(net->mpls.platform_label);
85 		rt = rcu_dereference(platform_label[index]);
86 	}
87 	return rt;
88 }
89 
90 bool mpls_output_possible(const struct net_device *dev)
91 {
92 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
93 }
94 EXPORT_SYMBOL_GPL(mpls_output_possible);
95 
96 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
97 {
98 	return (u8 *)nh + rt->rt_via_offset;
99 }
100 
101 static const u8 *mpls_nh_via(const struct mpls_route *rt,
102 			     const struct mpls_nh *nh)
103 {
104 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
105 }
106 
107 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
108 {
109 	/* The size of the layer 2.5 labels to be added for this route */
110 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
111 }
112 
113 unsigned int mpls_dev_mtu(const struct net_device *dev)
114 {
115 	/* The amount of data the layer 2 frame can hold */
116 	return dev->mtu;
117 }
118 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
119 
120 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
121 {
122 	if (skb->len <= mtu)
123 		return false;
124 
125 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
126 		return false;
127 
128 	return true;
129 }
130 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
131 
132 void mpls_stats_inc_outucastpkts(struct net_device *dev,
133 				 const struct sk_buff *skb)
134 {
135 	struct mpls_dev *mdev;
136 
137 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
138 		mdev = mpls_dev_get(dev);
139 		if (mdev)
140 			MPLS_INC_STATS_LEN(mdev, skb->len,
141 					   tx_packets,
142 					   tx_bytes);
143 	} else if (skb->protocol == htons(ETH_P_IP)) {
144 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
145 #if IS_ENABLED(CONFIG_IPV6)
146 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
147 		struct inet6_dev *in6dev = __in6_dev_get(dev);
148 
149 		if (in6dev)
150 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
151 					 IPSTATS_MIB_OUT, skb->len);
152 #endif
153 	}
154 }
155 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
156 
157 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
158 {
159 	struct mpls_entry_decoded dec;
160 	unsigned int mpls_hdr_len = 0;
161 	struct mpls_shim_hdr *hdr;
162 	bool eli_seen = false;
163 	int label_index;
164 	u32 hash = 0;
165 
166 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
167 	     label_index++) {
168 		mpls_hdr_len += sizeof(*hdr);
169 		if (!pskb_may_pull(skb, mpls_hdr_len))
170 			break;
171 
172 		/* Read and decode the current label */
173 		hdr = mpls_hdr(skb) + label_index;
174 		dec = mpls_entry_decode(hdr);
175 
176 		/* RFC6790 - reserved labels MUST NOT be used as keys
177 		 * for the load-balancing function
178 		 */
179 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
180 			hash = jhash_1word(dec.label, hash);
181 
182 			/* The entropy label follows the entropy label
183 			 * indicator, so this means that the entropy
184 			 * label was just added to the hash - no need to
185 			 * go any deeper either in the label stack or in the
186 			 * payload
187 			 */
188 			if (eli_seen)
189 				break;
190 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
191 			eli_seen = true;
192 		}
193 
194 		if (!dec.bos)
195 			continue;
196 
197 		/* found bottom label; does skb have room for a header? */
198 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
199 			const struct iphdr *v4hdr;
200 
201 			v4hdr = (const struct iphdr *)(hdr + 1);
202 			if (v4hdr->version == 4) {
203 				hash = jhash_3words(ntohl(v4hdr->saddr),
204 						    ntohl(v4hdr->daddr),
205 						    v4hdr->protocol, hash);
206 			} else if (v4hdr->version == 6 &&
207 				   pskb_may_pull(skb, mpls_hdr_len +
208 						 sizeof(struct ipv6hdr))) {
209 				const struct ipv6hdr *v6hdr;
210 
211 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
212 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
213 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
214 				hash = jhash_1word(v6hdr->nexthdr, hash);
215 			}
216 		}
217 
218 		break;
219 	}
220 
221 	return hash;
222 }
223 
224 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
225 {
226 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
227 }
228 
229 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
230  * a next hop (nh->nh_flags) are modified by netdev event handlers.
231  * Since those fields can change at any moment, use READ_ONCE to
232  * access both.
233  */
234 static const struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
235 						   struct sk_buff *skb)
236 {
237 	u32 hash = 0;
238 	int nh_index = 0;
239 	int n = 0;
240 	u8 alive;
241 
242 	/* No need to look further into packet if there's only
243 	 * one path
244 	 */
245 	if (rt->rt_nhn == 1)
246 		return rt->rt_nh;
247 
248 	alive = READ_ONCE(rt->rt_nhn_alive);
249 	if (alive == 0)
250 		return NULL;
251 
252 	hash = mpls_multipath_hash(rt, skb);
253 	nh_index = hash % alive;
254 	if (alive == rt->rt_nhn)
255 		goto out;
256 	for_nexthops(rt) {
257 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
258 
259 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
260 			continue;
261 		if (n == nh_index)
262 			return nh;
263 		n++;
264 	} endfor_nexthops(rt);
265 
266 out:
267 	return mpls_get_nexthop(rt, nh_index);
268 }
269 
270 static bool mpls_egress(struct net *net, struct mpls_route *rt,
271 			struct sk_buff *skb, struct mpls_entry_decoded dec)
272 {
273 	enum mpls_payload_type payload_type;
274 	bool success = false;
275 
276 	/* The IPv4 code below accesses through the IPv4 header
277 	 * checksum, which is 12 bytes into the packet.
278 	 * The IPv6 code below accesses through the IPv6 hop limit
279 	 * which is 8 bytes into the packet.
280 	 *
281 	 * For all supported cases there should always be at least 12
282 	 * bytes of packet data present.  The IPv4 header is 20 bytes
283 	 * without options and the IPv6 header is always 40 bytes
284 	 * long.
285 	 */
286 	if (!pskb_may_pull(skb, 12))
287 		return false;
288 
289 	payload_type = rt->rt_payload_type;
290 	if (payload_type == MPT_UNSPEC)
291 		payload_type = ip_hdr(skb)->version;
292 
293 	switch (payload_type) {
294 	case MPT_IPV4: {
295 		struct iphdr *hdr4 = ip_hdr(skb);
296 		u8 new_ttl;
297 		skb->protocol = htons(ETH_P_IP);
298 
299 		/* If propagating TTL, take the decremented TTL from
300 		 * the incoming MPLS header, otherwise decrement the
301 		 * TTL, but only if not 0 to avoid underflow.
302 		 */
303 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
304 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
305 		     net->mpls.ip_ttl_propagate))
306 			new_ttl = dec.ttl;
307 		else
308 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
309 
310 		csum_replace2(&hdr4->check,
311 			      htons(hdr4->ttl << 8),
312 			      htons(new_ttl << 8));
313 		hdr4->ttl = new_ttl;
314 		success = true;
315 		break;
316 	}
317 	case MPT_IPV6: {
318 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
319 		skb->protocol = htons(ETH_P_IPV6);
320 
321 		/* If propagating TTL, take the decremented TTL from
322 		 * the incoming MPLS header, otherwise decrement the
323 		 * hop limit, but only if not 0 to avoid underflow.
324 		 */
325 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
326 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
327 		     net->mpls.ip_ttl_propagate))
328 			hdr6->hop_limit = dec.ttl;
329 		else if (hdr6->hop_limit)
330 			hdr6->hop_limit = hdr6->hop_limit - 1;
331 		success = true;
332 		break;
333 	}
334 	case MPT_UNSPEC:
335 		/* Should have decided which protocol it is by now */
336 		break;
337 	}
338 
339 	return success;
340 }
341 
342 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
343 			struct packet_type *pt, struct net_device *orig_dev)
344 {
345 	struct net *net = dev_net(dev);
346 	struct mpls_shim_hdr *hdr;
347 	const struct mpls_nh *nh;
348 	struct mpls_route *rt;
349 	struct mpls_entry_decoded dec;
350 	struct net_device *out_dev;
351 	struct mpls_dev *out_mdev;
352 	struct mpls_dev *mdev;
353 	unsigned int hh_len;
354 	unsigned int new_header_size;
355 	unsigned int mtu;
356 	int err;
357 
358 	/* Careful this entire function runs inside of an rcu critical section */
359 
360 	mdev = mpls_dev_get(dev);
361 	if (!mdev)
362 		goto drop;
363 
364 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
365 			   rx_bytes);
366 
367 	if (!mdev->input_enabled) {
368 		MPLS_INC_STATS(mdev, rx_dropped);
369 		goto drop;
370 	}
371 
372 	if (skb->pkt_type != PACKET_HOST)
373 		goto err;
374 
375 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
376 		goto err;
377 
378 	if (!pskb_may_pull(skb, sizeof(*hdr)))
379 		goto err;
380 
381 	skb_dst_drop(skb);
382 
383 	/* Read and decode the label */
384 	hdr = mpls_hdr(skb);
385 	dec = mpls_entry_decode(hdr);
386 
387 	rt = mpls_route_input_rcu(net, dec.label);
388 	if (!rt) {
389 		MPLS_INC_STATS(mdev, rx_noroute);
390 		goto drop;
391 	}
392 
393 	nh = mpls_select_multipath(rt, skb);
394 	if (!nh)
395 		goto err;
396 
397 	/* Pop the label */
398 	skb_pull(skb, sizeof(*hdr));
399 	skb_reset_network_header(skb);
400 
401 	skb_orphan(skb);
402 
403 	if (skb_warn_if_lro(skb))
404 		goto err;
405 
406 	skb_forward_csum(skb);
407 
408 	/* Verify ttl is valid */
409 	if (dec.ttl <= 1)
410 		goto err;
411 
412 	/* Find the output device */
413 	out_dev = 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 	dec.ttl -= 1;
435 	if (unlikely(!new_header_size && dec.bos)) {
436 		/* Penultimate hop popping */
437 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
438 			goto err;
439 	} else {
440 		bool bos;
441 		int i;
442 		skb_push(skb, new_header_size);
443 		skb_reset_network_header(skb);
444 		/* Push the new labels */
445 		hdr = mpls_hdr(skb);
446 		bos = dec.bos;
447 		for (i = nh->nh_labels - 1; i >= 0; i--) {
448 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
449 						   dec.ttl, 0, bos);
450 			bos = false;
451 		}
452 	}
453 
454 	mpls_stats_inc_outucastpkts(out_dev, skb);
455 
456 	/* If via wasn't specified then send out using device address */
457 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
458 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
459 				 out_dev->dev_addr, skb);
460 	else
461 		err = neigh_xmit(nh->nh_via_table, out_dev,
462 				 mpls_nh_via(rt, nh), skb);
463 	if (err)
464 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
465 				    __func__, err);
466 	return 0;
467 
468 tx_err:
469 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
470 	if (out_mdev)
471 		MPLS_INC_STATS(out_mdev, tx_errors);
472 	goto drop;
473 err:
474 	MPLS_INC_STATS(mdev, rx_errors);
475 drop:
476 	kfree_skb(skb);
477 	return NET_RX_DROP;
478 }
479 
480 static struct packet_type mpls_packet_type __read_mostly = {
481 	.type = cpu_to_be16(ETH_P_MPLS_UC),
482 	.func = mpls_forward,
483 };
484 
485 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
486 	[RTA_DST]		= { .type = NLA_U32 },
487 	[RTA_OIF]		= { .type = NLA_U32 },
488 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
489 };
490 
491 struct mpls_route_config {
492 	u32			rc_protocol;
493 	u32			rc_ifindex;
494 	u8			rc_via_table;
495 	u8			rc_via_alen;
496 	u8			rc_via[MAX_VIA_ALEN];
497 	u32			rc_label;
498 	u8			rc_ttl_propagate;
499 	u8			rc_output_labels;
500 	u32			rc_output_label[MAX_NEW_LABELS];
501 	u32			rc_nlflags;
502 	enum mpls_payload_type	rc_payload_type;
503 	struct nl_info		rc_nlinfo;
504 	struct rtnexthop	*rc_mp;
505 	int			rc_mp_len;
506 };
507 
508 /* all nexthops within a route have the same size based on max
509  * number of labels and max via length for a hop
510  */
511 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
512 {
513 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
514 	struct mpls_route *rt;
515 	size_t size;
516 
517 	size = sizeof(*rt) + num_nh * nh_size;
518 	if (size > MAX_MPLS_ROUTE_MEM)
519 		return ERR_PTR(-EINVAL);
520 
521 	rt = kzalloc(size, GFP_KERNEL);
522 	if (!rt)
523 		return ERR_PTR(-ENOMEM);
524 
525 	rt->rt_nhn = num_nh;
526 	rt->rt_nhn_alive = num_nh;
527 	rt->rt_nh_size = nh_size;
528 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
529 
530 	return rt;
531 }
532 
533 static void mpls_rt_free(struct mpls_route *rt)
534 {
535 	if (rt)
536 		kfree_rcu(rt, rt_rcu);
537 }
538 
539 static void mpls_notify_route(struct net *net, unsigned index,
540 			      struct mpls_route *old, struct mpls_route *new,
541 			      const struct nl_info *info)
542 {
543 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
544 	unsigned portid = info ? info->portid : 0;
545 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
546 	struct mpls_route *rt = new ? new : old;
547 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
548 	/* Ignore reserved labels for now */
549 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
550 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
551 }
552 
553 static void mpls_route_update(struct net *net, unsigned index,
554 			      struct mpls_route *new,
555 			      const struct nl_info *info)
556 {
557 	struct mpls_route __rcu **platform_label;
558 	struct mpls_route *rt;
559 
560 	ASSERT_RTNL();
561 
562 	platform_label = rtnl_dereference(net->mpls.platform_label);
563 	rt = rtnl_dereference(platform_label[index]);
564 	rcu_assign_pointer(platform_label[index], new);
565 
566 	mpls_notify_route(net, index, rt, new, info);
567 
568 	/* If we removed a route free it now */
569 	mpls_rt_free(rt);
570 }
571 
572 static unsigned find_free_label(struct net *net)
573 {
574 	struct mpls_route __rcu **platform_label;
575 	size_t platform_labels;
576 	unsigned index;
577 
578 	platform_label = rtnl_dereference(net->mpls.platform_label);
579 	platform_labels = net->mpls.platform_labels;
580 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
581 	     index++) {
582 		if (!rtnl_dereference(platform_label[index]))
583 			return index;
584 	}
585 	return LABEL_NOT_SPECIFIED;
586 }
587 
588 #if IS_ENABLED(CONFIG_INET)
589 static struct net_device *inet_fib_lookup_dev(struct net *net,
590 					      const void *addr)
591 {
592 	struct net_device *dev;
593 	struct rtable *rt;
594 	struct in_addr daddr;
595 
596 	memcpy(&daddr, addr, sizeof(struct in_addr));
597 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
598 	if (IS_ERR(rt))
599 		return ERR_CAST(rt);
600 
601 	dev = rt->dst.dev;
602 	dev_hold(dev);
603 
604 	ip_rt_put(rt);
605 
606 	return dev;
607 }
608 #else
609 static struct net_device *inet_fib_lookup_dev(struct net *net,
610 					      const void *addr)
611 {
612 	return ERR_PTR(-EAFNOSUPPORT);
613 }
614 #endif
615 
616 #if IS_ENABLED(CONFIG_IPV6)
617 static struct net_device *inet6_fib_lookup_dev(struct net *net,
618 					       const void *addr)
619 {
620 	struct net_device *dev;
621 	struct dst_entry *dst;
622 	struct flowi6 fl6;
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 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
630 	if (IS_ERR(dst))
631 		return ERR_CAST(dst);
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 	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 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
1024 	if (IS_ERR(rt)) {
1025 		err = PTR_ERR(rt);
1026 		goto errout;
1027 	}
1028 
1029 	rt->rt_protocol = cfg->rc_protocol;
1030 	rt->rt_payload_type = cfg->rc_payload_type;
1031 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
1032 
1033 	if (cfg->rc_mp)
1034 		err = mpls_nh_build_multi(cfg, rt, max_labels, extack);
1035 	else
1036 		err = mpls_nh_build_from_cfg(cfg, rt);
1037 	if (err)
1038 		goto freert;
1039 
1040 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
1041 
1042 	return 0;
1043 
1044 freert:
1045 	mpls_rt_free(rt);
1046 errout:
1047 	return err;
1048 }
1049 
1050 static int mpls_route_del(struct mpls_route_config *cfg,
1051 			  struct netlink_ext_ack *extack)
1052 {
1053 	struct net *net = cfg->rc_nlinfo.nl_net;
1054 	unsigned index;
1055 	int err = -EINVAL;
1056 
1057 	index = cfg->rc_label;
1058 
1059 	if (!mpls_label_ok(net, &index, extack))
1060 		goto errout;
1061 
1062 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
1063 
1064 	err = 0;
1065 errout:
1066 	return err;
1067 }
1068 
1069 static void mpls_get_stats(struct mpls_dev *mdev,
1070 			   struct mpls_link_stats *stats)
1071 {
1072 	struct mpls_pcpu_stats *p;
1073 	int i;
1074 
1075 	memset(stats, 0, sizeof(*stats));
1076 
1077 	for_each_possible_cpu(i) {
1078 		struct mpls_link_stats local;
1079 		unsigned int start;
1080 
1081 		p = per_cpu_ptr(mdev->stats, i);
1082 		do {
1083 			start = u64_stats_fetch_begin(&p->syncp);
1084 			local = p->stats;
1085 		} while (u64_stats_fetch_retry(&p->syncp, start));
1086 
1087 		stats->rx_packets	+= local.rx_packets;
1088 		stats->rx_bytes		+= local.rx_bytes;
1089 		stats->tx_packets	+= local.tx_packets;
1090 		stats->tx_bytes		+= local.tx_bytes;
1091 		stats->rx_errors	+= local.rx_errors;
1092 		stats->tx_errors	+= local.tx_errors;
1093 		stats->rx_dropped	+= local.rx_dropped;
1094 		stats->tx_dropped	+= local.tx_dropped;
1095 		stats->rx_noroute	+= local.rx_noroute;
1096 	}
1097 }
1098 
1099 static int mpls_fill_stats_af(struct sk_buff *skb,
1100 			      const struct net_device *dev)
1101 {
1102 	struct mpls_link_stats *stats;
1103 	struct mpls_dev *mdev;
1104 	struct nlattr *nla;
1105 
1106 	mdev = mpls_dev_get(dev);
1107 	if (!mdev)
1108 		return -ENODATA;
1109 
1110 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1111 				sizeof(struct mpls_link_stats),
1112 				MPLS_STATS_UNSPEC);
1113 	if (!nla)
1114 		return -EMSGSIZE;
1115 
1116 	stats = nla_data(nla);
1117 	mpls_get_stats(mdev, stats);
1118 
1119 	return 0;
1120 }
1121 
1122 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1123 {
1124 	struct mpls_dev *mdev;
1125 
1126 	mdev = mpls_dev_get(dev);
1127 	if (!mdev)
1128 		return 0;
1129 
1130 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1131 }
1132 
1133 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1134 				     u32 portid, u32 seq, int event,
1135 				     unsigned int flags, int type)
1136 {
1137 	struct nlmsghdr  *nlh;
1138 	struct netconfmsg *ncm;
1139 	bool all = false;
1140 
1141 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1142 			flags);
1143 	if (!nlh)
1144 		return -EMSGSIZE;
1145 
1146 	if (type == NETCONFA_ALL)
1147 		all = true;
1148 
1149 	ncm = nlmsg_data(nlh);
1150 	ncm->ncm_family = AF_MPLS;
1151 
1152 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1153 		goto nla_put_failure;
1154 
1155 	if ((all || type == NETCONFA_INPUT) &&
1156 	    nla_put_s32(skb, NETCONFA_INPUT,
1157 			READ_ONCE(mdev->input_enabled)) < 0)
1158 		goto nla_put_failure;
1159 
1160 	nlmsg_end(skb, nlh);
1161 	return 0;
1162 
1163 nla_put_failure:
1164 	nlmsg_cancel(skb, nlh);
1165 	return -EMSGSIZE;
1166 }
1167 
1168 static int mpls_netconf_msgsize_devconf(int type)
1169 {
1170 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1171 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1172 	bool all = false;
1173 
1174 	if (type == NETCONFA_ALL)
1175 		all = true;
1176 
1177 	if (all || type == NETCONFA_INPUT)
1178 		size += nla_total_size(4);
1179 
1180 	return size;
1181 }
1182 
1183 static void mpls_netconf_notify_devconf(struct net *net, int event,
1184 					int type, struct mpls_dev *mdev)
1185 {
1186 	struct sk_buff *skb;
1187 	int err = -ENOBUFS;
1188 
1189 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1190 	if (!skb)
1191 		goto errout;
1192 
1193 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1194 	if (err < 0) {
1195 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1196 		WARN_ON(err == -EMSGSIZE);
1197 		kfree_skb(skb);
1198 		goto errout;
1199 	}
1200 
1201 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1202 	return;
1203 errout:
1204 	if (err < 0)
1205 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1206 }
1207 
1208 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1209 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1210 };
1211 
1212 static int mpls_netconf_valid_get_req(struct sk_buff *skb,
1213 				      const struct nlmsghdr *nlh,
1214 				      struct nlattr **tb,
1215 				      struct netlink_ext_ack *extack)
1216 {
1217 	int i, err;
1218 
1219 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(struct netconfmsg))) {
1220 		NL_SET_ERR_MSG_MOD(extack,
1221 				   "Invalid header for netconf get request");
1222 		return -EINVAL;
1223 	}
1224 
1225 	if (!netlink_strict_get_check(skb))
1226 		return nlmsg_parse_deprecated(nlh, sizeof(struct netconfmsg),
1227 					      tb, NETCONFA_MAX,
1228 					      devconf_mpls_policy, extack);
1229 
1230 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct netconfmsg),
1231 					    tb, NETCONFA_MAX,
1232 					    devconf_mpls_policy, extack);
1233 	if (err)
1234 		return err;
1235 
1236 	for (i = 0; i <= NETCONFA_MAX; i++) {
1237 		if (!tb[i])
1238 			continue;
1239 
1240 		switch (i) {
1241 		case NETCONFA_IFINDEX:
1242 			break;
1243 		default:
1244 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in netconf get request");
1245 			return -EINVAL;
1246 		}
1247 	}
1248 
1249 	return 0;
1250 }
1251 
1252 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1253 				    struct nlmsghdr *nlh,
1254 				    struct netlink_ext_ack *extack)
1255 {
1256 	struct net *net = sock_net(in_skb->sk);
1257 	struct nlattr *tb[NETCONFA_MAX + 1];
1258 	struct net_device *dev;
1259 	struct mpls_dev *mdev;
1260 	struct sk_buff *skb;
1261 	int ifindex;
1262 	int err;
1263 
1264 	err = mpls_netconf_valid_get_req(in_skb, nlh, tb, extack);
1265 	if (err < 0)
1266 		goto errout;
1267 
1268 	err = -EINVAL;
1269 	if (!tb[NETCONFA_IFINDEX])
1270 		goto errout;
1271 
1272 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1273 	dev = __dev_get_by_index(net, ifindex);
1274 	if (!dev)
1275 		goto errout;
1276 
1277 	mdev = mpls_dev_get(dev);
1278 	if (!mdev)
1279 		goto errout;
1280 
1281 	err = -ENOBUFS;
1282 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1283 	if (!skb)
1284 		goto errout;
1285 
1286 	err = mpls_netconf_fill_devconf(skb, mdev,
1287 					NETLINK_CB(in_skb).portid,
1288 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1289 					NETCONFA_ALL);
1290 	if (err < 0) {
1291 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1292 		WARN_ON(err == -EMSGSIZE);
1293 		kfree_skb(skb);
1294 		goto errout;
1295 	}
1296 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1297 errout:
1298 	return err;
1299 }
1300 
1301 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1302 				     struct netlink_callback *cb)
1303 {
1304 	const struct nlmsghdr *nlh = cb->nlh;
1305 	struct net *net = sock_net(skb->sk);
1306 	struct {
1307 		unsigned long ifindex;
1308 	} *ctx = (void *)cb->ctx;
1309 	struct net_device *dev;
1310 	struct mpls_dev *mdev;
1311 	int err = 0;
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 	rcu_read_lock();
1329 	for_each_netdev_dump(net, dev, ctx->ifindex) {
1330 		mdev = mpls_dev_get(dev);
1331 		if (!mdev)
1332 			continue;
1333 		err = mpls_netconf_fill_devconf(skb, mdev,
1334 						NETLINK_CB(cb->skb).portid,
1335 						nlh->nlmsg_seq,
1336 						RTM_NEWNETCONF,
1337 						NLM_F_MULTI,
1338 						NETCONFA_ALL);
1339 		if (err < 0)
1340 			break;
1341 	}
1342 	rcu_read_unlock();
1343 
1344 	return err;
1345 }
1346 
1347 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1348 	(&((struct mpls_dev *)0)->field)
1349 
1350 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1351 			  void *buffer, size_t *lenp, loff_t *ppos)
1352 {
1353 	int oval = *(int *)ctl->data;
1354 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1355 
1356 	if (write) {
1357 		struct mpls_dev *mdev = ctl->extra1;
1358 		int i = (int *)ctl->data - (int *)mdev;
1359 		struct net *net = ctl->extra2;
1360 		int val = *(int *)ctl->data;
1361 
1362 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1363 		    val != oval) {
1364 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1365 						    NETCONFA_INPUT, mdev);
1366 		}
1367 	}
1368 
1369 	return ret;
1370 }
1371 
1372 static const struct ctl_table mpls_dev_table[] = {
1373 	{
1374 		.procname	= "input",
1375 		.maxlen		= sizeof(int),
1376 		.mode		= 0644,
1377 		.proc_handler	= mpls_conf_proc,
1378 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1379 	},
1380 	{ }
1381 };
1382 
1383 static int mpls_dev_sysctl_register(struct net_device *dev,
1384 				    struct mpls_dev *mdev)
1385 {
1386 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1387 	struct net *net = dev_net(dev);
1388 	struct ctl_table *table;
1389 	int i;
1390 
1391 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1392 	if (!table)
1393 		goto out;
1394 
1395 	/* Table data contains only offsets relative to the base of
1396 	 * the mdev at this point, so make them absolute.
1397 	 */
1398 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1399 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1400 		table[i].extra1 = mdev;
1401 		table[i].extra2 = net;
1402 	}
1403 
1404 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1405 
1406 	mdev->sysctl = register_net_sysctl_sz(net, path, table,
1407 					      ARRAY_SIZE(mpls_dev_table));
1408 	if (!mdev->sysctl)
1409 		goto free;
1410 
1411 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1412 	return 0;
1413 
1414 free:
1415 	kfree(table);
1416 out:
1417 	mdev->sysctl = NULL;
1418 	return -ENOBUFS;
1419 }
1420 
1421 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1422 				       struct mpls_dev *mdev)
1423 {
1424 	struct net *net = dev_net(dev);
1425 	struct ctl_table *table;
1426 
1427 	if (!mdev->sysctl)
1428 		return;
1429 
1430 	table = mdev->sysctl->ctl_table_arg;
1431 	unregister_net_sysctl_table(mdev->sysctl);
1432 	kfree(table);
1433 
1434 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1435 }
1436 
1437 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1438 {
1439 	struct mpls_dev *mdev;
1440 	int err = -ENOMEM;
1441 	int i;
1442 
1443 	ASSERT_RTNL();
1444 
1445 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1446 	if (!mdev)
1447 		return ERR_PTR(err);
1448 
1449 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1450 	if (!mdev->stats)
1451 		goto free;
1452 
1453 	for_each_possible_cpu(i) {
1454 		struct mpls_pcpu_stats *mpls_stats;
1455 
1456 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1457 		u64_stats_init(&mpls_stats->syncp);
1458 	}
1459 
1460 	mdev->dev = dev;
1461 
1462 	err = mpls_dev_sysctl_register(dev, mdev);
1463 	if (err)
1464 		goto free;
1465 
1466 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1467 
1468 	return mdev;
1469 
1470 free:
1471 	free_percpu(mdev->stats);
1472 	kfree(mdev);
1473 	return ERR_PTR(err);
1474 }
1475 
1476 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1477 {
1478 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1479 
1480 	free_percpu(mdev->stats);
1481 	kfree(mdev);
1482 }
1483 
1484 static int mpls_ifdown(struct net_device *dev, int event)
1485 {
1486 	struct mpls_route __rcu **platform_label;
1487 	struct net *net = dev_net(dev);
1488 	unsigned index;
1489 
1490 	platform_label = rtnl_dereference(net->mpls.platform_label);
1491 	for (index = 0; index < net->mpls.platform_labels; index++) {
1492 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1493 		bool nh_del = false;
1494 		u8 alive = 0;
1495 
1496 		if (!rt)
1497 			continue;
1498 
1499 		if (event == NETDEV_UNREGISTER) {
1500 			u8 deleted = 0;
1501 
1502 			for_nexthops(rt) {
1503 				if (!nh->nh_dev || nh->nh_dev == dev)
1504 					deleted++;
1505 				if (nh->nh_dev == dev)
1506 					nh_del = true;
1507 			} endfor_nexthops(rt);
1508 
1509 			/* if there are no more nexthops, delete the route */
1510 			if (deleted == rt->rt_nhn) {
1511 				mpls_route_update(net, index, NULL, NULL);
1512 				continue;
1513 			}
1514 
1515 			if (nh_del) {
1516 				size_t size = sizeof(*rt) + rt->rt_nhn *
1517 					rt->rt_nh_size;
1518 				struct mpls_route *orig = rt;
1519 
1520 				rt = kmemdup(orig, size, GFP_KERNEL);
1521 				if (!rt)
1522 					return -ENOMEM;
1523 			}
1524 		}
1525 
1526 		change_nexthops(rt) {
1527 			unsigned int nh_flags = nh->nh_flags;
1528 
1529 			if (nh->nh_dev != dev)
1530 				goto next;
1531 
1532 			switch (event) {
1533 			case NETDEV_DOWN:
1534 			case NETDEV_UNREGISTER:
1535 				nh_flags |= RTNH_F_DEAD;
1536 				fallthrough;
1537 			case NETDEV_CHANGE:
1538 				nh_flags |= RTNH_F_LINKDOWN;
1539 				break;
1540 			}
1541 			if (event == NETDEV_UNREGISTER)
1542 				nh->nh_dev = NULL;
1543 
1544 			if (nh->nh_flags != nh_flags)
1545 				WRITE_ONCE(nh->nh_flags, nh_flags);
1546 next:
1547 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1548 				alive++;
1549 		} endfor_nexthops(rt);
1550 
1551 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1552 
1553 		if (nh_del)
1554 			mpls_route_update(net, index, rt, NULL);
1555 	}
1556 
1557 	return 0;
1558 }
1559 
1560 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1561 {
1562 	struct mpls_route __rcu **platform_label;
1563 	struct net *net = dev_net(dev);
1564 	unsigned index;
1565 	u8 alive;
1566 
1567 	platform_label = rtnl_dereference(net->mpls.platform_label);
1568 	for (index = 0; index < net->mpls.platform_labels; index++) {
1569 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1570 
1571 		if (!rt)
1572 			continue;
1573 
1574 		alive = 0;
1575 		change_nexthops(rt) {
1576 			unsigned int nh_flags = nh->nh_flags;
1577 
1578 			if (!(nh_flags & flags)) {
1579 				alive++;
1580 				continue;
1581 			}
1582 			if (nh->nh_dev != dev)
1583 				continue;
1584 			alive++;
1585 			nh_flags &= ~flags;
1586 			WRITE_ONCE(nh->nh_flags, nh_flags);
1587 		} endfor_nexthops(rt);
1588 
1589 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1590 	}
1591 }
1592 
1593 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1594 			   void *ptr)
1595 {
1596 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1597 	struct mpls_dev *mdev;
1598 	unsigned int flags;
1599 	int err;
1600 
1601 	if (event == NETDEV_REGISTER) {
1602 		mdev = mpls_add_dev(dev);
1603 		if (IS_ERR(mdev))
1604 			return notifier_from_errno(PTR_ERR(mdev));
1605 
1606 		return NOTIFY_OK;
1607 	}
1608 
1609 	mdev = mpls_dev_get(dev);
1610 	if (!mdev)
1611 		return NOTIFY_OK;
1612 
1613 	switch (event) {
1614 
1615 	case NETDEV_DOWN:
1616 		err = mpls_ifdown(dev, event);
1617 		if (err)
1618 			return notifier_from_errno(err);
1619 		break;
1620 	case NETDEV_UP:
1621 		flags = dev_get_flags(dev);
1622 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1623 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1624 		else
1625 			mpls_ifup(dev, RTNH_F_DEAD);
1626 		break;
1627 	case NETDEV_CHANGE:
1628 		flags = dev_get_flags(dev);
1629 		if (flags & (IFF_RUNNING | IFF_LOWER_UP)) {
1630 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1631 		} else {
1632 			err = mpls_ifdown(dev, event);
1633 			if (err)
1634 				return notifier_from_errno(err);
1635 		}
1636 		break;
1637 	case NETDEV_UNREGISTER:
1638 		err = mpls_ifdown(dev, event);
1639 		if (err)
1640 			return notifier_from_errno(err);
1641 		mdev = mpls_dev_get(dev);
1642 		if (mdev) {
1643 			mpls_dev_sysctl_unregister(dev, mdev);
1644 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1645 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1646 		}
1647 		break;
1648 	case NETDEV_CHANGENAME:
1649 		mdev = mpls_dev_get(dev);
1650 		if (mdev) {
1651 			mpls_dev_sysctl_unregister(dev, mdev);
1652 			err = mpls_dev_sysctl_register(dev, mdev);
1653 			if (err)
1654 				return notifier_from_errno(err);
1655 		}
1656 		break;
1657 	}
1658 	return NOTIFY_OK;
1659 }
1660 
1661 static struct notifier_block mpls_dev_notifier = {
1662 	.notifier_call = mpls_dev_notify,
1663 };
1664 
1665 static int nla_put_via(struct sk_buff *skb,
1666 		       u8 table, const void *addr, int alen)
1667 {
1668 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1669 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1670 	};
1671 	struct nlattr *nla;
1672 	struct rtvia *via;
1673 	int family = AF_UNSPEC;
1674 
1675 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1676 	if (!nla)
1677 		return -EMSGSIZE;
1678 
1679 	if (table <= NEIGH_NR_TABLES)
1680 		family = table_to_family[table];
1681 
1682 	via = nla_data(nla);
1683 	via->rtvia_family = family;
1684 	memcpy(via->rtvia_addr, addr, alen);
1685 	return 0;
1686 }
1687 
1688 int nla_put_labels(struct sk_buff *skb, int attrtype,
1689 		   u8 labels, const u32 label[])
1690 {
1691 	struct nlattr *nla;
1692 	struct mpls_shim_hdr *nla_label;
1693 	bool bos;
1694 	int i;
1695 	nla = nla_reserve(skb, attrtype, labels*4);
1696 	if (!nla)
1697 		return -EMSGSIZE;
1698 
1699 	nla_label = nla_data(nla);
1700 	bos = true;
1701 	for (i = labels - 1; i >= 0; i--) {
1702 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1703 		bos = false;
1704 	}
1705 
1706 	return 0;
1707 }
1708 EXPORT_SYMBOL_GPL(nla_put_labels);
1709 
1710 int nla_get_labels(const struct nlattr *nla, u8 max_labels, u8 *labels,
1711 		   u32 label[], struct netlink_ext_ack *extack)
1712 {
1713 	unsigned len = nla_len(nla);
1714 	struct mpls_shim_hdr *nla_label;
1715 	u8 nla_labels;
1716 	bool bos;
1717 	int i;
1718 
1719 	/* len needs to be an even multiple of 4 (the label size). Number
1720 	 * of labels is a u8 so check for overflow.
1721 	 */
1722 	if (len & 3 || len / 4 > 255) {
1723 		NL_SET_ERR_MSG_ATTR(extack, nla,
1724 				    "Invalid length for labels attribute");
1725 		return -EINVAL;
1726 	}
1727 
1728 	/* Limit the number of new labels allowed */
1729 	nla_labels = len/4;
1730 	if (nla_labels > max_labels) {
1731 		NL_SET_ERR_MSG(extack, "Too many labels");
1732 		return -EINVAL;
1733 	}
1734 
1735 	/* when label == NULL, caller wants number of labels */
1736 	if (!label)
1737 		goto out;
1738 
1739 	nla_label = nla_data(nla);
1740 	bos = true;
1741 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1742 		struct mpls_entry_decoded dec;
1743 		dec = mpls_entry_decode(nla_label + i);
1744 
1745 		/* Ensure the bottom of stack flag is properly set
1746 		 * and ttl and tc are both clear.
1747 		 */
1748 		if (dec.ttl) {
1749 			NL_SET_ERR_MSG_ATTR(extack, nla,
1750 					    "TTL in label must be 0");
1751 			return -EINVAL;
1752 		}
1753 
1754 		if (dec.tc) {
1755 			NL_SET_ERR_MSG_ATTR(extack, nla,
1756 					    "Traffic class in label must be 0");
1757 			return -EINVAL;
1758 		}
1759 
1760 		if (dec.bos != bos) {
1761 			NL_SET_BAD_ATTR(extack, nla);
1762 			if (bos) {
1763 				NL_SET_ERR_MSG(extack,
1764 					       "BOS bit must be set in first label");
1765 			} else {
1766 				NL_SET_ERR_MSG(extack,
1767 					       "BOS bit can only be set in first label");
1768 			}
1769 			return -EINVAL;
1770 		}
1771 
1772 		switch (dec.label) {
1773 		case MPLS_LABEL_IMPLNULL:
1774 			/* RFC3032: This is a label that an LSR may
1775 			 * assign and distribute, but which never
1776 			 * actually appears in the encapsulation.
1777 			 */
1778 			NL_SET_ERR_MSG_ATTR(extack, nla,
1779 					    "Implicit NULL Label (3) can not be used in encapsulation");
1780 			return -EINVAL;
1781 		}
1782 
1783 		label[i] = dec.label;
1784 	}
1785 out:
1786 	*labels = nla_labels;
1787 	return 0;
1788 }
1789 EXPORT_SYMBOL_GPL(nla_get_labels);
1790 
1791 static int rtm_to_route_config(struct sk_buff *skb,
1792 			       struct nlmsghdr *nlh,
1793 			       struct mpls_route_config *cfg,
1794 			       struct netlink_ext_ack *extack)
1795 {
1796 	struct rtmsg *rtm;
1797 	struct nlattr *tb[RTA_MAX+1];
1798 	int index;
1799 	int err;
1800 
1801 	err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1802 				     rtm_mpls_policy, extack);
1803 	if (err < 0)
1804 		goto errout;
1805 
1806 	err = -EINVAL;
1807 	rtm = nlmsg_data(nlh);
1808 
1809 	if (rtm->rtm_family != AF_MPLS) {
1810 		NL_SET_ERR_MSG(extack, "Invalid address family in rtmsg");
1811 		goto errout;
1812 	}
1813 	if (rtm->rtm_dst_len != 20) {
1814 		NL_SET_ERR_MSG(extack, "rtm_dst_len must be 20 for MPLS");
1815 		goto errout;
1816 	}
1817 	if (rtm->rtm_src_len != 0) {
1818 		NL_SET_ERR_MSG(extack, "rtm_src_len must be 0 for MPLS");
1819 		goto errout;
1820 	}
1821 	if (rtm->rtm_tos != 0) {
1822 		NL_SET_ERR_MSG(extack, "rtm_tos must be 0 for MPLS");
1823 		goto errout;
1824 	}
1825 	if (rtm->rtm_table != RT_TABLE_MAIN) {
1826 		NL_SET_ERR_MSG(extack,
1827 			       "MPLS only supports the main route table");
1828 		goto errout;
1829 	}
1830 	/* Any value is acceptable for rtm_protocol */
1831 
1832 	/* As mpls uses destination specific addresses
1833 	 * (or source specific address in the case of multicast)
1834 	 * all addresses have universal scope.
1835 	 */
1836 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE) {
1837 		NL_SET_ERR_MSG(extack,
1838 			       "Invalid route scope  - MPLS only supports UNIVERSE");
1839 		goto errout;
1840 	}
1841 	if (rtm->rtm_type != RTN_UNICAST) {
1842 		NL_SET_ERR_MSG(extack,
1843 			       "Invalid route type - MPLS only supports UNICAST");
1844 		goto errout;
1845 	}
1846 	if (rtm->rtm_flags != 0) {
1847 		NL_SET_ERR_MSG(extack, "rtm_flags must be 0 for MPLS");
1848 		goto errout;
1849 	}
1850 
1851 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1852 	cfg->rc_protocol	= rtm->rtm_protocol;
1853 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1854 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1855 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1856 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1857 	cfg->rc_nlinfo.nlh	= nlh;
1858 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1859 
1860 	for (index = 0; index <= RTA_MAX; index++) {
1861 		struct nlattr *nla = tb[index];
1862 		if (!nla)
1863 			continue;
1864 
1865 		switch (index) {
1866 		case RTA_OIF:
1867 			cfg->rc_ifindex = nla_get_u32(nla);
1868 			break;
1869 		case RTA_NEWDST:
1870 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1871 					   &cfg->rc_output_labels,
1872 					   cfg->rc_output_label, extack))
1873 				goto errout;
1874 			break;
1875 		case RTA_DST:
1876 		{
1877 			u8 label_count;
1878 			if (nla_get_labels(nla, 1, &label_count,
1879 					   &cfg->rc_label, extack))
1880 				goto errout;
1881 
1882 			if (!mpls_label_ok(cfg->rc_nlinfo.nl_net,
1883 					   &cfg->rc_label, extack))
1884 				goto errout;
1885 			break;
1886 		}
1887 		case RTA_GATEWAY:
1888 			NL_SET_ERR_MSG(extack, "MPLS does not support RTA_GATEWAY attribute");
1889 			goto errout;
1890 		case RTA_VIA:
1891 		{
1892 			if (nla_get_via(nla, &cfg->rc_via_alen,
1893 					&cfg->rc_via_table, cfg->rc_via,
1894 					extack))
1895 				goto errout;
1896 			break;
1897 		}
1898 		case RTA_MULTIPATH:
1899 		{
1900 			cfg->rc_mp = nla_data(nla);
1901 			cfg->rc_mp_len = nla_len(nla);
1902 			break;
1903 		}
1904 		case RTA_TTL_PROPAGATE:
1905 		{
1906 			u8 ttl_propagate = nla_get_u8(nla);
1907 
1908 			if (ttl_propagate > 1) {
1909 				NL_SET_ERR_MSG_ATTR(extack, nla,
1910 						    "RTA_TTL_PROPAGATE can only be 0 or 1");
1911 				goto errout;
1912 			}
1913 			cfg->rc_ttl_propagate = ttl_propagate ?
1914 				MPLS_TTL_PROP_ENABLED :
1915 				MPLS_TTL_PROP_DISABLED;
1916 			break;
1917 		}
1918 		default:
1919 			NL_SET_ERR_MSG_ATTR(extack, nla, "Unknown attribute");
1920 			/* Unsupported attribute */
1921 			goto errout;
1922 		}
1923 	}
1924 
1925 	err = 0;
1926 errout:
1927 	return err;
1928 }
1929 
1930 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1931 			     struct netlink_ext_ack *extack)
1932 {
1933 	struct mpls_route_config *cfg;
1934 	int err;
1935 
1936 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1937 	if (!cfg)
1938 		return -ENOMEM;
1939 
1940 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1941 	if (err < 0)
1942 		goto out;
1943 
1944 	err = mpls_route_del(cfg, extack);
1945 out:
1946 	kfree(cfg);
1947 
1948 	return err;
1949 }
1950 
1951 
1952 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1953 			     struct netlink_ext_ack *extack)
1954 {
1955 	struct mpls_route_config *cfg;
1956 	int err;
1957 
1958 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1959 	if (!cfg)
1960 		return -ENOMEM;
1961 
1962 	err = rtm_to_route_config(skb, nlh, cfg, extack);
1963 	if (err < 0)
1964 		goto out;
1965 
1966 	err = mpls_route_add(cfg, extack);
1967 out:
1968 	kfree(cfg);
1969 
1970 	return err;
1971 }
1972 
1973 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1974 			   u32 label, struct mpls_route *rt, int flags)
1975 {
1976 	struct net_device *dev;
1977 	struct nlmsghdr *nlh;
1978 	struct rtmsg *rtm;
1979 
1980 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1981 	if (nlh == NULL)
1982 		return -EMSGSIZE;
1983 
1984 	rtm = nlmsg_data(nlh);
1985 	rtm->rtm_family = AF_MPLS;
1986 	rtm->rtm_dst_len = 20;
1987 	rtm->rtm_src_len = 0;
1988 	rtm->rtm_tos = 0;
1989 	rtm->rtm_table = RT_TABLE_MAIN;
1990 	rtm->rtm_protocol = rt->rt_protocol;
1991 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1992 	rtm->rtm_type = RTN_UNICAST;
1993 	rtm->rtm_flags = 0;
1994 
1995 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1996 		goto nla_put_failure;
1997 
1998 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1999 		bool ttl_propagate =
2000 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
2001 
2002 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
2003 			       ttl_propagate))
2004 			goto nla_put_failure;
2005 	}
2006 	if (rt->rt_nhn == 1) {
2007 		const struct mpls_nh *nh = rt->rt_nh;
2008 
2009 		if (nh->nh_labels &&
2010 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2011 				   nh->nh_label))
2012 			goto nla_put_failure;
2013 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2014 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2015 				nh->nh_via_alen))
2016 			goto nla_put_failure;
2017 		dev = nh->nh_dev;
2018 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2019 			goto nla_put_failure;
2020 		if (nh->nh_flags & RTNH_F_LINKDOWN)
2021 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2022 		if (nh->nh_flags & RTNH_F_DEAD)
2023 			rtm->rtm_flags |= RTNH_F_DEAD;
2024 	} else {
2025 		struct rtnexthop *rtnh;
2026 		struct nlattr *mp;
2027 		u8 linkdown = 0;
2028 		u8 dead = 0;
2029 
2030 		mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
2031 		if (!mp)
2032 			goto nla_put_failure;
2033 
2034 		for_nexthops(rt) {
2035 			dev = nh->nh_dev;
2036 			if (!dev)
2037 				continue;
2038 
2039 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
2040 			if (!rtnh)
2041 				goto nla_put_failure;
2042 
2043 			rtnh->rtnh_ifindex = dev->ifindex;
2044 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
2045 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
2046 				linkdown++;
2047 			}
2048 			if (nh->nh_flags & RTNH_F_DEAD) {
2049 				rtnh->rtnh_flags |= RTNH_F_DEAD;
2050 				dead++;
2051 			}
2052 
2053 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
2054 							    nh->nh_labels,
2055 							    nh->nh_label))
2056 				goto nla_put_failure;
2057 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2058 			    nla_put_via(skb, nh->nh_via_table,
2059 					mpls_nh_via(rt, nh),
2060 					nh->nh_via_alen))
2061 				goto nla_put_failure;
2062 
2063 			/* length of rtnetlink header + attributes */
2064 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
2065 		} endfor_nexthops(rt);
2066 
2067 		if (linkdown == rt->rt_nhn)
2068 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
2069 		if (dead == rt->rt_nhn)
2070 			rtm->rtm_flags |= RTNH_F_DEAD;
2071 
2072 		nla_nest_end(skb, mp);
2073 	}
2074 
2075 	nlmsg_end(skb, nlh);
2076 	return 0;
2077 
2078 nla_put_failure:
2079 	nlmsg_cancel(skb, nlh);
2080 	return -EMSGSIZE;
2081 }
2082 
2083 #if IS_ENABLED(CONFIG_INET)
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 	return ip_valid_fib_dump_req(net, nlh, filter, cb);
2089 }
2090 #else
2091 static int mpls_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
2092 				   struct fib_dump_filter *filter,
2093 				   struct netlink_callback *cb)
2094 {
2095 	struct netlink_ext_ack *extack = cb->extack;
2096 	struct nlattr *tb[RTA_MAX + 1];
2097 	struct rtmsg *rtm;
2098 	int err, i;
2099 
2100 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2101 		NL_SET_ERR_MSG_MOD(extack, "Invalid header for FIB dump request");
2102 		return -EINVAL;
2103 	}
2104 
2105 	rtm = nlmsg_data(nlh);
2106 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
2107 	    rtm->rtm_table   || rtm->rtm_scope    || rtm->rtm_type  ||
2108 	    rtm->rtm_flags) {
2109 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for FIB dump request");
2110 		return -EINVAL;
2111 	}
2112 
2113 	if (rtm->rtm_protocol) {
2114 		filter->protocol = rtm->rtm_protocol;
2115 		filter->filter_set = 1;
2116 		cb->answer_flags = NLM_F_DUMP_FILTERED;
2117 	}
2118 
2119 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2120 					    rtm_mpls_policy, extack);
2121 	if (err < 0)
2122 		return err;
2123 
2124 	for (i = 0; i <= RTA_MAX; ++i) {
2125 		int ifindex;
2126 
2127 		if (i == RTA_OIF) {
2128 			ifindex = nla_get_u32(tb[i]);
2129 			filter->dev = __dev_get_by_index(net, ifindex);
2130 			if (!filter->dev)
2131 				return -ENODEV;
2132 			filter->filter_set = 1;
2133 		} else if (tb[i]) {
2134 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in dump request");
2135 			return -EINVAL;
2136 		}
2137 	}
2138 
2139 	return 0;
2140 }
2141 #endif
2142 
2143 static bool mpls_rt_uses_dev(struct mpls_route *rt,
2144 			     const struct net_device *dev)
2145 {
2146 	if (rt->rt_nhn == 1) {
2147 		struct mpls_nh *nh = rt->rt_nh;
2148 
2149 		if (nh->nh_dev == dev)
2150 			return true;
2151 	} else {
2152 		for_nexthops(rt) {
2153 			if (nh->nh_dev == dev)
2154 				return true;
2155 		} endfor_nexthops(rt);
2156 	}
2157 
2158 	return false;
2159 }
2160 
2161 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
2162 {
2163 	const struct nlmsghdr *nlh = cb->nlh;
2164 	struct net *net = sock_net(skb->sk);
2165 	struct mpls_route __rcu **platform_label;
2166 	struct fib_dump_filter filter = {};
2167 	unsigned int flags = NLM_F_MULTI;
2168 	size_t platform_labels;
2169 	unsigned int index;
2170 
2171 	ASSERT_RTNL();
2172 
2173 	if (cb->strict_check) {
2174 		int err;
2175 
2176 		err = mpls_valid_fib_dump_req(net, nlh, &filter, cb);
2177 		if (err < 0)
2178 			return err;
2179 
2180 		/* for MPLS, there is only 1 table with fixed type and flags.
2181 		 * If either are set in the filter then return nothing.
2182 		 */
2183 		if ((filter.table_id && filter.table_id != RT_TABLE_MAIN) ||
2184 		    (filter.rt_type && filter.rt_type != RTN_UNICAST) ||
2185 		     filter.flags)
2186 			return skb->len;
2187 	}
2188 
2189 	index = cb->args[0];
2190 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
2191 		index = MPLS_LABEL_FIRST_UNRESERVED;
2192 
2193 	platform_label = rtnl_dereference(net->mpls.platform_label);
2194 	platform_labels = net->mpls.platform_labels;
2195 
2196 	if (filter.filter_set)
2197 		flags |= NLM_F_DUMP_FILTERED;
2198 
2199 	for (; index < platform_labels; index++) {
2200 		struct mpls_route *rt;
2201 
2202 		rt = rtnl_dereference(platform_label[index]);
2203 		if (!rt)
2204 			continue;
2205 
2206 		if ((filter.dev && !mpls_rt_uses_dev(rt, filter.dev)) ||
2207 		    (filter.protocol && rt->rt_protocol != filter.protocol))
2208 			continue;
2209 
2210 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
2211 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2212 				    index, rt, flags) < 0)
2213 			break;
2214 	}
2215 	cb->args[0] = index;
2216 
2217 	return skb->len;
2218 }
2219 
2220 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
2221 {
2222 	size_t payload =
2223 		NLMSG_ALIGN(sizeof(struct rtmsg))
2224 		+ nla_total_size(4)			/* RTA_DST */
2225 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
2226 
2227 	if (rt->rt_nhn == 1) {
2228 		struct mpls_nh *nh = rt->rt_nh;
2229 
2230 		if (nh->nh_dev)
2231 			payload += nla_total_size(4); /* RTA_OIF */
2232 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
2233 			payload += nla_total_size(2 + nh->nh_via_alen);
2234 		if (nh->nh_labels) /* RTA_NEWDST */
2235 			payload += nla_total_size(nh->nh_labels * 4);
2236 	} else {
2237 		/* each nexthop is packed in an attribute */
2238 		size_t nhsize = 0;
2239 
2240 		for_nexthops(rt) {
2241 			if (!nh->nh_dev)
2242 				continue;
2243 			nhsize += nla_total_size(sizeof(struct rtnexthop));
2244 			/* RTA_VIA */
2245 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
2246 				nhsize += nla_total_size(2 + nh->nh_via_alen);
2247 			if (nh->nh_labels)
2248 				nhsize += nla_total_size(nh->nh_labels * 4);
2249 		} endfor_nexthops(rt);
2250 		/* nested attribute */
2251 		payload += nla_total_size(nhsize);
2252 	}
2253 
2254 	return payload;
2255 }
2256 
2257 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
2258 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
2259 		       unsigned int nlm_flags)
2260 {
2261 	struct sk_buff *skb;
2262 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
2263 	int err = -ENOBUFS;
2264 
2265 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2266 	if (skb == NULL)
2267 		goto errout;
2268 
2269 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
2270 	if (err < 0) {
2271 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2272 		WARN_ON(err == -EMSGSIZE);
2273 		kfree_skb(skb);
2274 		goto errout;
2275 	}
2276 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
2277 
2278 	return;
2279 errout:
2280 	if (err < 0)
2281 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
2282 }
2283 
2284 static int mpls_valid_getroute_req(struct sk_buff *skb,
2285 				   const struct nlmsghdr *nlh,
2286 				   struct nlattr **tb,
2287 				   struct netlink_ext_ack *extack)
2288 {
2289 	struct rtmsg *rtm;
2290 	int i, err;
2291 
2292 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2293 		NL_SET_ERR_MSG_MOD(extack,
2294 				   "Invalid header for get route request");
2295 		return -EINVAL;
2296 	}
2297 
2298 	if (!netlink_strict_get_check(skb))
2299 		return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2300 					      rtm_mpls_policy, extack);
2301 
2302 	rtm = nlmsg_data(nlh);
2303 	if ((rtm->rtm_dst_len && rtm->rtm_dst_len != 20) ||
2304 	    rtm->rtm_src_len || rtm->rtm_tos || rtm->rtm_table ||
2305 	    rtm->rtm_protocol || rtm->rtm_scope || rtm->rtm_type) {
2306 		NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
2307 		return -EINVAL;
2308 	}
2309 	if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
2310 		NL_SET_ERR_MSG_MOD(extack,
2311 				   "Invalid flags for get route request");
2312 		return -EINVAL;
2313 	}
2314 
2315 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2316 					    rtm_mpls_policy, extack);
2317 	if (err)
2318 		return err;
2319 
2320 	if ((tb[RTA_DST] || tb[RTA_NEWDST]) && !rtm->rtm_dst_len) {
2321 		NL_SET_ERR_MSG_MOD(extack, "rtm_dst_len must be 20 for MPLS");
2322 		return -EINVAL;
2323 	}
2324 
2325 	for (i = 0; i <= RTA_MAX; i++) {
2326 		if (!tb[i])
2327 			continue;
2328 
2329 		switch (i) {
2330 		case RTA_DST:
2331 		case RTA_NEWDST:
2332 			break;
2333 		default:
2334 			NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
2335 			return -EINVAL;
2336 		}
2337 	}
2338 
2339 	return 0;
2340 }
2341 
2342 static int mpls_getroute(struct sk_buff *in_skb, struct nlmsghdr *in_nlh,
2343 			 struct netlink_ext_ack *extack)
2344 {
2345 	struct net *net = sock_net(in_skb->sk);
2346 	u32 portid = NETLINK_CB(in_skb).portid;
2347 	u32 in_label = LABEL_NOT_SPECIFIED;
2348 	struct nlattr *tb[RTA_MAX + 1];
2349 	u32 labels[MAX_NEW_LABELS];
2350 	struct mpls_shim_hdr *hdr;
2351 	unsigned int hdr_size = 0;
2352 	const struct mpls_nh *nh;
2353 	struct net_device *dev;
2354 	struct mpls_route *rt;
2355 	struct rtmsg *rtm, *r;
2356 	struct nlmsghdr *nlh;
2357 	struct sk_buff *skb;
2358 	u8 n_labels;
2359 	int err;
2360 
2361 	err = mpls_valid_getroute_req(in_skb, in_nlh, tb, extack);
2362 	if (err < 0)
2363 		goto errout;
2364 
2365 	rtm = nlmsg_data(in_nlh);
2366 
2367 	if (tb[RTA_DST]) {
2368 		u8 label_count;
2369 
2370 		if (nla_get_labels(tb[RTA_DST], 1, &label_count,
2371 				   &in_label, extack)) {
2372 			err = -EINVAL;
2373 			goto errout;
2374 		}
2375 
2376 		if (!mpls_label_ok(net, &in_label, extack)) {
2377 			err = -EINVAL;
2378 			goto errout;
2379 		}
2380 	}
2381 
2382 	rt = mpls_route_input_rcu(net, in_label);
2383 	if (!rt) {
2384 		err = -ENETUNREACH;
2385 		goto errout;
2386 	}
2387 
2388 	if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2389 		skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
2390 		if (!skb) {
2391 			err = -ENOBUFS;
2392 			goto errout;
2393 		}
2394 
2395 		err = mpls_dump_route(skb, portid, in_nlh->nlmsg_seq,
2396 				      RTM_NEWROUTE, in_label, rt, 0);
2397 		if (err < 0) {
2398 			/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
2399 			WARN_ON(err == -EMSGSIZE);
2400 			goto errout_free;
2401 		}
2402 
2403 		return rtnl_unicast(skb, net, portid);
2404 	}
2405 
2406 	if (tb[RTA_NEWDST]) {
2407 		if (nla_get_labels(tb[RTA_NEWDST], MAX_NEW_LABELS, &n_labels,
2408 				   labels, extack) != 0) {
2409 			err = -EINVAL;
2410 			goto errout;
2411 		}
2412 
2413 		hdr_size = n_labels * sizeof(struct mpls_shim_hdr);
2414 	}
2415 
2416 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2417 	if (!skb) {
2418 		err = -ENOBUFS;
2419 		goto errout;
2420 	}
2421 
2422 	skb->protocol = htons(ETH_P_MPLS_UC);
2423 
2424 	if (hdr_size) {
2425 		bool bos;
2426 		int i;
2427 
2428 		if (skb_cow(skb, hdr_size)) {
2429 			err = -ENOBUFS;
2430 			goto errout_free;
2431 		}
2432 
2433 		skb_reserve(skb, hdr_size);
2434 		skb_push(skb, hdr_size);
2435 		skb_reset_network_header(skb);
2436 
2437 		/* Push new labels */
2438 		hdr = mpls_hdr(skb);
2439 		bos = true;
2440 		for (i = n_labels - 1; i >= 0; i--) {
2441 			hdr[i] = mpls_entry_encode(labels[i],
2442 						   1, 0, bos);
2443 			bos = false;
2444 		}
2445 	}
2446 
2447 	nh = mpls_select_multipath(rt, skb);
2448 	if (!nh) {
2449 		err = -ENETUNREACH;
2450 		goto errout_free;
2451 	}
2452 
2453 	if (hdr_size) {
2454 		skb_pull(skb, hdr_size);
2455 		skb_reset_network_header(skb);
2456 	}
2457 
2458 	nlh = nlmsg_put(skb, portid, in_nlh->nlmsg_seq,
2459 			RTM_NEWROUTE, sizeof(*r), 0);
2460 	if (!nlh) {
2461 		err = -EMSGSIZE;
2462 		goto errout_free;
2463 	}
2464 
2465 	r = nlmsg_data(nlh);
2466 	r->rtm_family	 = AF_MPLS;
2467 	r->rtm_dst_len	= 20;
2468 	r->rtm_src_len	= 0;
2469 	r->rtm_table	= RT_TABLE_MAIN;
2470 	r->rtm_type	= RTN_UNICAST;
2471 	r->rtm_scope	= RT_SCOPE_UNIVERSE;
2472 	r->rtm_protocol = rt->rt_protocol;
2473 	r->rtm_flags	= 0;
2474 
2475 	if (nla_put_labels(skb, RTA_DST, 1, &in_label))
2476 		goto nla_put_failure;
2477 
2478 	if (nh->nh_labels &&
2479 	    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
2480 			   nh->nh_label))
2481 		goto nla_put_failure;
2482 
2483 	if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
2484 	    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
2485 			nh->nh_via_alen))
2486 		goto nla_put_failure;
2487 	dev = nh->nh_dev;
2488 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
2489 		goto nla_put_failure;
2490 
2491 	nlmsg_end(skb, nlh);
2492 
2493 	err = rtnl_unicast(skb, net, portid);
2494 errout:
2495 	return err;
2496 
2497 nla_put_failure:
2498 	nlmsg_cancel(skb, nlh);
2499 	err = -EMSGSIZE;
2500 errout_free:
2501 	kfree_skb(skb);
2502 	return err;
2503 }
2504 
2505 static int resize_platform_label_table(struct net *net, size_t limit)
2506 {
2507 	size_t size = sizeof(struct mpls_route *) * limit;
2508 	size_t old_limit;
2509 	size_t cp_size;
2510 	struct mpls_route __rcu **labels = NULL, **old;
2511 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2512 	unsigned index;
2513 
2514 	if (size) {
2515 		labels = kvzalloc(size, GFP_KERNEL);
2516 		if (!labels)
2517 			goto nolabels;
2518 	}
2519 
2520 	/* In case the predefined labels need to be populated */
2521 	if (limit > MPLS_LABEL_IPV4NULL) {
2522 		struct net_device *lo = net->loopback_dev;
2523 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2524 		if (IS_ERR(rt0))
2525 			goto nort0;
2526 		rt0->rt_nh->nh_dev = lo;
2527 		rt0->rt_protocol = RTPROT_KERNEL;
2528 		rt0->rt_payload_type = MPT_IPV4;
2529 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2530 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2531 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2532 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2533 		       lo->addr_len);
2534 	}
2535 	if (limit > MPLS_LABEL_IPV6NULL) {
2536 		struct net_device *lo = net->loopback_dev;
2537 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2538 		if (IS_ERR(rt2))
2539 			goto nort2;
2540 		rt2->rt_nh->nh_dev = lo;
2541 		rt2->rt_protocol = RTPROT_KERNEL;
2542 		rt2->rt_payload_type = MPT_IPV6;
2543 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2544 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2545 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2546 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2547 		       lo->addr_len);
2548 	}
2549 
2550 	rtnl_lock();
2551 	/* Remember the original table */
2552 	old = rtnl_dereference(net->mpls.platform_label);
2553 	old_limit = net->mpls.platform_labels;
2554 
2555 	/* Free any labels beyond the new table */
2556 	for (index = limit; index < old_limit; index++)
2557 		mpls_route_update(net, index, NULL, NULL);
2558 
2559 	/* Copy over the old labels */
2560 	cp_size = size;
2561 	if (old_limit < limit)
2562 		cp_size = old_limit * sizeof(struct mpls_route *);
2563 
2564 	memcpy(labels, old, cp_size);
2565 
2566 	/* If needed set the predefined labels */
2567 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2568 	    (limit > MPLS_LABEL_IPV6NULL)) {
2569 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2570 		rt2 = NULL;
2571 	}
2572 
2573 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2574 	    (limit > MPLS_LABEL_IPV4NULL)) {
2575 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2576 		rt0 = NULL;
2577 	}
2578 
2579 	/* Update the global pointers */
2580 	net->mpls.platform_labels = limit;
2581 	rcu_assign_pointer(net->mpls.platform_label, labels);
2582 
2583 	rtnl_unlock();
2584 
2585 	mpls_rt_free(rt2);
2586 	mpls_rt_free(rt0);
2587 
2588 	if (old) {
2589 		synchronize_rcu();
2590 		kvfree(old);
2591 	}
2592 	return 0;
2593 
2594 nort2:
2595 	mpls_rt_free(rt0);
2596 nort0:
2597 	kvfree(labels);
2598 nolabels:
2599 	return -ENOMEM;
2600 }
2601 
2602 static int mpls_platform_labels(struct ctl_table *table, int write,
2603 				void *buffer, size_t *lenp, loff_t *ppos)
2604 {
2605 	struct net *net = table->data;
2606 	int platform_labels = net->mpls.platform_labels;
2607 	int ret;
2608 	struct ctl_table tmp = {
2609 		.procname	= table->procname,
2610 		.data		= &platform_labels,
2611 		.maxlen		= sizeof(int),
2612 		.mode		= table->mode,
2613 		.extra1		= SYSCTL_ZERO,
2614 		.extra2		= &label_limit,
2615 	};
2616 
2617 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2618 
2619 	if (write && ret == 0)
2620 		ret = resize_platform_label_table(net, platform_labels);
2621 
2622 	return ret;
2623 }
2624 
2625 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2626 	(&((struct net *)0)->field)
2627 
2628 static const struct ctl_table mpls_table[] = {
2629 	{
2630 		.procname	= "platform_labels",
2631 		.data		= NULL,
2632 		.maxlen		= sizeof(int),
2633 		.mode		= 0644,
2634 		.proc_handler	= mpls_platform_labels,
2635 	},
2636 	{
2637 		.procname	= "ip_ttl_propagate",
2638 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2639 		.maxlen		= sizeof(int),
2640 		.mode		= 0644,
2641 		.proc_handler	= proc_dointvec_minmax,
2642 		.extra1		= SYSCTL_ZERO,
2643 		.extra2		= SYSCTL_ONE,
2644 	},
2645 	{
2646 		.procname	= "default_ttl",
2647 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2648 		.maxlen		= sizeof(int),
2649 		.mode		= 0644,
2650 		.proc_handler	= proc_dointvec_minmax,
2651 		.extra1		= SYSCTL_ONE,
2652 		.extra2		= &ttl_max,
2653 	},
2654 	{ }
2655 };
2656 
2657 static int mpls_net_init(struct net *net)
2658 {
2659 	struct ctl_table *table;
2660 	int i;
2661 
2662 	net->mpls.platform_labels = 0;
2663 	net->mpls.platform_label = NULL;
2664 	net->mpls.ip_ttl_propagate = 1;
2665 	net->mpls.default_ttl = 255;
2666 
2667 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2668 	if (table == NULL)
2669 		return -ENOMEM;
2670 
2671 	/* Table data contains only offsets relative to the base of
2672 	 * the mdev at this point, so make them absolute.
2673 	 */
2674 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2675 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2676 
2677 	net->mpls.ctl = register_net_sysctl_sz(net, "net/mpls", table,
2678 					       ARRAY_SIZE(mpls_table));
2679 	if (net->mpls.ctl == NULL) {
2680 		kfree(table);
2681 		return -ENOMEM;
2682 	}
2683 
2684 	return 0;
2685 }
2686 
2687 static void mpls_net_exit(struct net *net)
2688 {
2689 	struct mpls_route __rcu **platform_label;
2690 	size_t platform_labels;
2691 	struct ctl_table *table;
2692 	unsigned int index;
2693 
2694 	table = net->mpls.ctl->ctl_table_arg;
2695 	unregister_net_sysctl_table(net->mpls.ctl);
2696 	kfree(table);
2697 
2698 	/* An rcu grace period has passed since there was a device in
2699 	 * the network namespace (and thus the last in flight packet)
2700 	 * left this network namespace.  This is because
2701 	 * unregister_netdevice_many and netdev_run_todo has completed
2702 	 * for each network device that was in this network namespace.
2703 	 *
2704 	 * As such no additional rcu synchronization is necessary when
2705 	 * freeing the platform_label table.
2706 	 */
2707 	rtnl_lock();
2708 	platform_label = rtnl_dereference(net->mpls.platform_label);
2709 	platform_labels = net->mpls.platform_labels;
2710 	for (index = 0; index < platform_labels; index++) {
2711 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2712 		RCU_INIT_POINTER(platform_label[index], NULL);
2713 		mpls_notify_route(net, index, rt, NULL, NULL);
2714 		mpls_rt_free(rt);
2715 	}
2716 	rtnl_unlock();
2717 
2718 	kvfree(platform_label);
2719 }
2720 
2721 static struct pernet_operations mpls_net_ops = {
2722 	.init = mpls_net_init,
2723 	.exit = mpls_net_exit,
2724 };
2725 
2726 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2727 	.family		   = AF_MPLS,
2728 	.fill_stats_af	   = mpls_fill_stats_af,
2729 	.get_stats_af_size = mpls_get_stats_af_size,
2730 };
2731 
2732 static const struct rtnl_msg_handler mpls_rtnl_msg_handlers[] __initdata_or_module = {
2733 	{THIS_MODULE, PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, 0},
2734 	{THIS_MODULE, PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, 0},
2735 	{THIS_MODULE, PF_MPLS, RTM_GETROUTE, mpls_getroute, mpls_dump_routes, 0},
2736 	{THIS_MODULE, PF_MPLS, RTM_GETNETCONF,
2737 	 mpls_netconf_get_devconf, mpls_netconf_dump_devconf,
2738 	 RTNL_FLAG_DUMP_UNLOCKED},
2739 };
2740 
2741 static int __init mpls_init(void)
2742 {
2743 	int err;
2744 
2745 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2746 
2747 	err = register_pernet_subsys(&mpls_net_ops);
2748 	if (err)
2749 		goto out;
2750 
2751 	err = register_netdevice_notifier(&mpls_dev_notifier);
2752 	if (err)
2753 		goto out_unregister_pernet;
2754 
2755 	dev_add_pack(&mpls_packet_type);
2756 
2757 	rtnl_af_register(&mpls_af_ops);
2758 
2759 	err = rtnl_register_many(mpls_rtnl_msg_handlers);
2760 	if (err)
2761 		goto out_unregister_rtnl_af;
2762 
2763 	err = ipgre_tunnel_encap_add_mpls_ops();
2764 	if (err) {
2765 		pr_err("Can't add mpls over gre tunnel ops\n");
2766 		goto out_unregister_rtnl;
2767 	}
2768 
2769 	err = 0;
2770 out:
2771 	return err;
2772 
2773 out_unregister_rtnl:
2774 	rtnl_unregister_many(mpls_rtnl_msg_handlers);
2775 out_unregister_rtnl_af:
2776 	rtnl_af_unregister(&mpls_af_ops);
2777 	dev_remove_pack(&mpls_packet_type);
2778 out_unregister_pernet:
2779 	unregister_pernet_subsys(&mpls_net_ops);
2780 	goto out;
2781 }
2782 module_init(mpls_init);
2783 
2784 static void __exit mpls_exit(void)
2785 {
2786 	rtnl_unregister_all(PF_MPLS);
2787 	rtnl_af_unregister(&mpls_af_ops);
2788 	dev_remove_pack(&mpls_packet_type);
2789 	unregister_netdevice_notifier(&mpls_dev_notifier);
2790 	unregister_pernet_subsys(&mpls_net_ops);
2791 	ipgre_tunnel_encap_del_mpls_ops();
2792 }
2793 module_exit(mpls_exit);
2794 
2795 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2796 MODULE_LICENSE("GPL v2");
2797 MODULE_ALIAS_NETPROTO(PF_MPLS);
2798