xref: /openbmc/linux/net/mpls/af_mpls.c (revision 4f139972b489f8bc2c821aa25ac65018d92af3f7)
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/netns/generic.h>
20 #if IS_ENABLED(CONFIG_IPV6)
21 #include <net/ipv6.h>
22 #endif
23 #include <net/addrconf.h>
24 #include <net/nexthop.h>
25 #include "internal.h"
26 
27 /* max memory we will use for mpls_route */
28 #define MAX_MPLS_ROUTE_MEM	4096
29 
30 /* Maximum number of labels to look ahead at when selecting a path of
31  * a multipath route
32  */
33 #define MAX_MP_SELECT_LABELS 4
34 
35 #define MPLS_NEIGH_TABLE_UNSPEC (NEIGH_LINK_TABLE + 1)
36 
37 static int zero = 0;
38 static int one = 1;
39 static int label_limit = (1 << 20) - 1;
40 static int ttl_max = 255;
41 
42 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
43 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
44 		       unsigned int nlm_flags);
45 
46 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
47 {
48 	struct mpls_route *rt = NULL;
49 
50 	if (index < net->mpls.platform_labels) {
51 		struct mpls_route __rcu **platform_label =
52 			rcu_dereference(net->mpls.platform_label);
53 		rt = rcu_dereference(platform_label[index]);
54 	}
55 	return rt;
56 }
57 
58 bool mpls_output_possible(const struct net_device *dev)
59 {
60 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
61 }
62 EXPORT_SYMBOL_GPL(mpls_output_possible);
63 
64 static u8 *__mpls_nh_via(struct mpls_route *rt, struct mpls_nh *nh)
65 {
66 	return (u8 *)nh + rt->rt_via_offset;
67 }
68 
69 static const u8 *mpls_nh_via(const struct mpls_route *rt,
70 			     const struct mpls_nh *nh)
71 {
72 	return __mpls_nh_via((struct mpls_route *)rt, (struct mpls_nh *)nh);
73 }
74 
75 static unsigned int mpls_nh_header_size(const struct mpls_nh *nh)
76 {
77 	/* The size of the layer 2.5 labels to be added for this route */
78 	return nh->nh_labels * sizeof(struct mpls_shim_hdr);
79 }
80 
81 unsigned int mpls_dev_mtu(const struct net_device *dev)
82 {
83 	/* The amount of data the layer 2 frame can hold */
84 	return dev->mtu;
85 }
86 EXPORT_SYMBOL_GPL(mpls_dev_mtu);
87 
88 bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
89 {
90 	if (skb->len <= mtu)
91 		return false;
92 
93 	if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
94 		return false;
95 
96 	return true;
97 }
98 EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
99 
100 void mpls_stats_inc_outucastpkts(struct net_device *dev,
101 				 const struct sk_buff *skb)
102 {
103 	struct mpls_dev *mdev;
104 
105 	if (skb->protocol == htons(ETH_P_MPLS_UC)) {
106 		mdev = mpls_dev_get(dev);
107 		if (mdev)
108 			MPLS_INC_STATS_LEN(mdev, skb->len,
109 					   tx_packets,
110 					   tx_bytes);
111 	} else if (skb->protocol == htons(ETH_P_IP)) {
112 		IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
113 #if IS_ENABLED(CONFIG_IPV6)
114 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
115 		struct inet6_dev *in6dev = __in6_dev_get(dev);
116 
117 		if (in6dev)
118 			IP6_UPD_PO_STATS(dev_net(dev), in6dev,
119 					 IPSTATS_MIB_OUT, skb->len);
120 #endif
121 	}
122 }
123 EXPORT_SYMBOL_GPL(mpls_stats_inc_outucastpkts);
124 
125 static u32 mpls_multipath_hash(struct mpls_route *rt, struct sk_buff *skb)
126 {
127 	struct mpls_entry_decoded dec;
128 	unsigned int mpls_hdr_len = 0;
129 	struct mpls_shim_hdr *hdr;
130 	bool eli_seen = false;
131 	int label_index;
132 	u32 hash = 0;
133 
134 	for (label_index = 0; label_index < MAX_MP_SELECT_LABELS;
135 	     label_index++) {
136 		mpls_hdr_len += sizeof(*hdr);
137 		if (!pskb_may_pull(skb, mpls_hdr_len))
138 			break;
139 
140 		/* Read and decode the current label */
141 		hdr = mpls_hdr(skb) + label_index;
142 		dec = mpls_entry_decode(hdr);
143 
144 		/* RFC6790 - reserved labels MUST NOT be used as keys
145 		 * for the load-balancing function
146 		 */
147 		if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
148 			hash = jhash_1word(dec.label, hash);
149 
150 			/* The entropy label follows the entropy label
151 			 * indicator, so this means that the entropy
152 			 * label was just added to the hash - no need to
153 			 * go any deeper either in the label stack or in the
154 			 * payload
155 			 */
156 			if (eli_seen)
157 				break;
158 		} else if (dec.label == MPLS_LABEL_ENTROPY) {
159 			eli_seen = true;
160 		}
161 
162 		if (!dec.bos)
163 			continue;
164 
165 		/* found bottom label; does skb have room for a header? */
166 		if (pskb_may_pull(skb, mpls_hdr_len + sizeof(struct iphdr))) {
167 			const struct iphdr *v4hdr;
168 
169 			v4hdr = (const struct iphdr *)(hdr + 1);
170 			if (v4hdr->version == 4) {
171 				hash = jhash_3words(ntohl(v4hdr->saddr),
172 						    ntohl(v4hdr->daddr),
173 						    v4hdr->protocol, hash);
174 			} else if (v4hdr->version == 6 &&
175 				   pskb_may_pull(skb, mpls_hdr_len +
176 						 sizeof(struct ipv6hdr))) {
177 				const struct ipv6hdr *v6hdr;
178 
179 				v6hdr = (const struct ipv6hdr *)(hdr + 1);
180 				hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
181 				hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
182 				hash = jhash_1word(v6hdr->nexthdr, hash);
183 			}
184 		}
185 
186 		break;
187 	}
188 
189 	return hash;
190 }
191 
192 static struct mpls_nh *mpls_get_nexthop(struct mpls_route *rt, u8 index)
193 {
194 	return (struct mpls_nh *)((u8 *)rt->rt_nh + index * rt->rt_nh_size);
195 }
196 
197 /* number of alive nexthops (rt->rt_nhn_alive) and the flags for
198  * a next hop (nh->nh_flags) are modified by netdev event handlers.
199  * Since those fields can change at any moment, use READ_ONCE to
200  * access both.
201  */
202 static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
203 					     struct sk_buff *skb)
204 {
205 	u32 hash = 0;
206 	int nh_index = 0;
207 	int n = 0;
208 	u8 alive;
209 
210 	/* No need to look further into packet if there's only
211 	 * one path
212 	 */
213 	if (rt->rt_nhn == 1)
214 		return rt->rt_nh;
215 
216 	alive = READ_ONCE(rt->rt_nhn_alive);
217 	if (alive == 0)
218 		return NULL;
219 
220 	hash = mpls_multipath_hash(rt, skb);
221 	nh_index = hash % alive;
222 	if (alive == rt->rt_nhn)
223 		goto out;
224 	for_nexthops(rt) {
225 		unsigned int nh_flags = READ_ONCE(nh->nh_flags);
226 
227 		if (nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
228 			continue;
229 		if (n == nh_index)
230 			return nh;
231 		n++;
232 	} endfor_nexthops(rt);
233 
234 out:
235 	return mpls_get_nexthop(rt, nh_index);
236 }
237 
238 static bool mpls_egress(struct net *net, struct mpls_route *rt,
239 			struct sk_buff *skb, struct mpls_entry_decoded dec)
240 {
241 	enum mpls_payload_type payload_type;
242 	bool success = false;
243 
244 	/* The IPv4 code below accesses through the IPv4 header
245 	 * checksum, which is 12 bytes into the packet.
246 	 * The IPv6 code below accesses through the IPv6 hop limit
247 	 * which is 8 bytes into the packet.
248 	 *
249 	 * For all supported cases there should always be at least 12
250 	 * bytes of packet data present.  The IPv4 header is 20 bytes
251 	 * without options and the IPv6 header is always 40 bytes
252 	 * long.
253 	 */
254 	if (!pskb_may_pull(skb, 12))
255 		return false;
256 
257 	payload_type = rt->rt_payload_type;
258 	if (payload_type == MPT_UNSPEC)
259 		payload_type = ip_hdr(skb)->version;
260 
261 	switch (payload_type) {
262 	case MPT_IPV4: {
263 		struct iphdr *hdr4 = ip_hdr(skb);
264 		u8 new_ttl;
265 		skb->protocol = htons(ETH_P_IP);
266 
267 		/* If propagating TTL, take the decremented TTL from
268 		 * the incoming MPLS header, otherwise decrement the
269 		 * TTL, but only if not 0 to avoid underflow.
270 		 */
271 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
272 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
273 		     net->mpls.ip_ttl_propagate))
274 			new_ttl = dec.ttl;
275 		else
276 			new_ttl = hdr4->ttl ? hdr4->ttl - 1 : 0;
277 
278 		csum_replace2(&hdr4->check,
279 			      htons(hdr4->ttl << 8),
280 			      htons(new_ttl << 8));
281 		hdr4->ttl = new_ttl;
282 		success = true;
283 		break;
284 	}
285 	case MPT_IPV6: {
286 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
287 		skb->protocol = htons(ETH_P_IPV6);
288 
289 		/* If propagating TTL, take the decremented TTL from
290 		 * the incoming MPLS header, otherwise decrement the
291 		 * hop limit, but only if not 0 to avoid underflow.
292 		 */
293 		if (rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED ||
294 		    (rt->rt_ttl_propagate == MPLS_TTL_PROP_DEFAULT &&
295 		     net->mpls.ip_ttl_propagate))
296 			hdr6->hop_limit = dec.ttl;
297 		else if (hdr6->hop_limit)
298 			hdr6->hop_limit = hdr6->hop_limit - 1;
299 		success = true;
300 		break;
301 	}
302 	case MPT_UNSPEC:
303 		/* Should have decided which protocol it is by now */
304 		break;
305 	}
306 
307 	return success;
308 }
309 
310 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
311 			struct packet_type *pt, struct net_device *orig_dev)
312 {
313 	struct net *net = dev_net(dev);
314 	struct mpls_shim_hdr *hdr;
315 	struct mpls_route *rt;
316 	struct mpls_nh *nh;
317 	struct mpls_entry_decoded dec;
318 	struct net_device *out_dev;
319 	struct mpls_dev *out_mdev;
320 	struct mpls_dev *mdev;
321 	unsigned int hh_len;
322 	unsigned int new_header_size;
323 	unsigned int mtu;
324 	int err;
325 
326 	/* Careful this entire function runs inside of an rcu critical section */
327 
328 	mdev = mpls_dev_get(dev);
329 	if (!mdev)
330 		goto drop;
331 
332 	MPLS_INC_STATS_LEN(mdev, skb->len, rx_packets,
333 			   rx_bytes);
334 
335 	if (!mdev->input_enabled) {
336 		MPLS_INC_STATS(mdev, rx_dropped);
337 		goto drop;
338 	}
339 
340 	if (skb->pkt_type != PACKET_HOST)
341 		goto err;
342 
343 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
344 		goto err;
345 
346 	if (!pskb_may_pull(skb, sizeof(*hdr)))
347 		goto err;
348 
349 	/* Read and decode the label */
350 	hdr = mpls_hdr(skb);
351 	dec = mpls_entry_decode(hdr);
352 
353 	rt = mpls_route_input_rcu(net, dec.label);
354 	if (!rt) {
355 		MPLS_INC_STATS(mdev, rx_noroute);
356 		goto drop;
357 	}
358 
359 	nh = mpls_select_multipath(rt, skb);
360 	if (!nh)
361 		goto err;
362 
363 	/* Pop the label */
364 	skb_pull(skb, sizeof(*hdr));
365 	skb_reset_network_header(skb);
366 
367 	skb_orphan(skb);
368 
369 	if (skb_warn_if_lro(skb))
370 		goto err;
371 
372 	skb_forward_csum(skb);
373 
374 	/* Verify ttl is valid */
375 	if (dec.ttl <= 1)
376 		goto err;
377 	dec.ttl -= 1;
378 
379 	/* Find the output device */
380 	out_dev = rcu_dereference(nh->nh_dev);
381 	if (!mpls_output_possible(out_dev))
382 		goto tx_err;
383 
384 	/* Verify the destination can hold the packet */
385 	new_header_size = mpls_nh_header_size(nh);
386 	mtu = mpls_dev_mtu(out_dev);
387 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
388 		goto tx_err;
389 
390 	hh_len = LL_RESERVED_SPACE(out_dev);
391 	if (!out_dev->header_ops)
392 		hh_len = 0;
393 
394 	/* Ensure there is enough space for the headers in the skb */
395 	if (skb_cow(skb, hh_len + new_header_size))
396 		goto tx_err;
397 
398 	skb->dev = out_dev;
399 	skb->protocol = htons(ETH_P_MPLS_UC);
400 
401 	if (unlikely(!new_header_size && dec.bos)) {
402 		/* Penultimate hop popping */
403 		if (!mpls_egress(dev_net(out_dev), rt, skb, dec))
404 			goto err;
405 	} else {
406 		bool bos;
407 		int i;
408 		skb_push(skb, new_header_size);
409 		skb_reset_network_header(skb);
410 		/* Push the new labels */
411 		hdr = mpls_hdr(skb);
412 		bos = dec.bos;
413 		for (i = nh->nh_labels - 1; i >= 0; i--) {
414 			hdr[i] = mpls_entry_encode(nh->nh_label[i],
415 						   dec.ttl, 0, bos);
416 			bos = false;
417 		}
418 	}
419 
420 	mpls_stats_inc_outucastpkts(out_dev, skb);
421 
422 	/* If via wasn't specified then send out using device address */
423 	if (nh->nh_via_table == MPLS_NEIGH_TABLE_UNSPEC)
424 		err = neigh_xmit(NEIGH_LINK_TABLE, out_dev,
425 				 out_dev->dev_addr, skb);
426 	else
427 		err = neigh_xmit(nh->nh_via_table, out_dev,
428 				 mpls_nh_via(rt, nh), skb);
429 	if (err)
430 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
431 				    __func__, err);
432 	return 0;
433 
434 tx_err:
435 	out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL;
436 	if (out_mdev)
437 		MPLS_INC_STATS(out_mdev, tx_errors);
438 	goto drop;
439 err:
440 	MPLS_INC_STATS(mdev, rx_errors);
441 drop:
442 	kfree_skb(skb);
443 	return NET_RX_DROP;
444 }
445 
446 static struct packet_type mpls_packet_type __read_mostly = {
447 	.type = cpu_to_be16(ETH_P_MPLS_UC),
448 	.func = mpls_forward,
449 };
450 
451 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
452 	[RTA_DST]		= { .type = NLA_U32 },
453 	[RTA_OIF]		= { .type = NLA_U32 },
454 	[RTA_TTL_PROPAGATE]	= { .type = NLA_U8 },
455 };
456 
457 struct mpls_route_config {
458 	u32			rc_protocol;
459 	u32			rc_ifindex;
460 	u8			rc_via_table;
461 	u8			rc_via_alen;
462 	u8			rc_via[MAX_VIA_ALEN];
463 	u32			rc_label;
464 	u8			rc_ttl_propagate;
465 	u8			rc_output_labels;
466 	u32			rc_output_label[MAX_NEW_LABELS];
467 	u32			rc_nlflags;
468 	enum mpls_payload_type	rc_payload_type;
469 	struct nl_info		rc_nlinfo;
470 	struct rtnexthop	*rc_mp;
471 	int			rc_mp_len;
472 };
473 
474 /* all nexthops within a route have the same size based on max
475  * number of labels and max via length for a hop
476  */
477 static struct mpls_route *mpls_rt_alloc(u8 num_nh, u8 max_alen, u8 max_labels)
478 {
479 	u8 nh_size = MPLS_NH_SIZE(max_labels, max_alen);
480 	struct mpls_route *rt;
481 	size_t size;
482 
483 	size = sizeof(*rt) + num_nh * nh_size;
484 	if (size > MAX_MPLS_ROUTE_MEM)
485 		return ERR_PTR(-EINVAL);
486 
487 	rt = kzalloc(size, GFP_KERNEL);
488 	if (!rt)
489 		return ERR_PTR(-ENOMEM);
490 
491 	rt->rt_nhn = num_nh;
492 	rt->rt_nhn_alive = num_nh;
493 	rt->rt_nh_size = nh_size;
494 	rt->rt_via_offset = MPLS_NH_VIA_OFF(max_labels);
495 
496 	return rt;
497 }
498 
499 static void mpls_rt_free(struct mpls_route *rt)
500 {
501 	if (rt)
502 		kfree_rcu(rt, rt_rcu);
503 }
504 
505 static void mpls_notify_route(struct net *net, unsigned index,
506 			      struct mpls_route *old, struct mpls_route *new,
507 			      const struct nl_info *info)
508 {
509 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
510 	unsigned portid = info ? info->portid : 0;
511 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
512 	struct mpls_route *rt = new ? new : old;
513 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
514 	/* Ignore reserved labels for now */
515 	if (rt && (index >= MPLS_LABEL_FIRST_UNRESERVED))
516 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
517 }
518 
519 static void mpls_route_update(struct net *net, unsigned index,
520 			      struct mpls_route *new,
521 			      const struct nl_info *info)
522 {
523 	struct mpls_route __rcu **platform_label;
524 	struct mpls_route *rt;
525 
526 	ASSERT_RTNL();
527 
528 	platform_label = rtnl_dereference(net->mpls.platform_label);
529 	rt = rtnl_dereference(platform_label[index]);
530 	rcu_assign_pointer(platform_label[index], new);
531 
532 	mpls_notify_route(net, index, rt, new, info);
533 
534 	/* If we removed a route free it now */
535 	mpls_rt_free(rt);
536 }
537 
538 static unsigned find_free_label(struct net *net)
539 {
540 	struct mpls_route __rcu **platform_label;
541 	size_t platform_labels;
542 	unsigned index;
543 
544 	platform_label = rtnl_dereference(net->mpls.platform_label);
545 	platform_labels = net->mpls.platform_labels;
546 	for (index = MPLS_LABEL_FIRST_UNRESERVED; index < platform_labels;
547 	     index++) {
548 		if (!rtnl_dereference(platform_label[index]))
549 			return index;
550 	}
551 	return LABEL_NOT_SPECIFIED;
552 }
553 
554 #if IS_ENABLED(CONFIG_INET)
555 static struct net_device *inet_fib_lookup_dev(struct net *net,
556 					      const void *addr)
557 {
558 	struct net_device *dev;
559 	struct rtable *rt;
560 	struct in_addr daddr;
561 
562 	memcpy(&daddr, addr, sizeof(struct in_addr));
563 	rt = ip_route_output(net, daddr.s_addr, 0, 0, 0);
564 	if (IS_ERR(rt))
565 		return ERR_CAST(rt);
566 
567 	dev = rt->dst.dev;
568 	dev_hold(dev);
569 
570 	ip_rt_put(rt);
571 
572 	return dev;
573 }
574 #else
575 static struct net_device *inet_fib_lookup_dev(struct net *net,
576 					      const void *addr)
577 {
578 	return ERR_PTR(-EAFNOSUPPORT);
579 }
580 #endif
581 
582 #if IS_ENABLED(CONFIG_IPV6)
583 static struct net_device *inet6_fib_lookup_dev(struct net *net,
584 					       const void *addr)
585 {
586 	struct net_device *dev;
587 	struct dst_entry *dst;
588 	struct flowi6 fl6;
589 	int err;
590 
591 	if (!ipv6_stub)
592 		return ERR_PTR(-EAFNOSUPPORT);
593 
594 	memset(&fl6, 0, sizeof(fl6));
595 	memcpy(&fl6.daddr, addr, sizeof(struct in6_addr));
596 	err = ipv6_stub->ipv6_dst_lookup(net, NULL, &dst, &fl6);
597 	if (err)
598 		return ERR_PTR(err);
599 
600 	dev = dst->dev;
601 	dev_hold(dev);
602 	dst_release(dst);
603 
604 	return dev;
605 }
606 #else
607 static struct net_device *inet6_fib_lookup_dev(struct net *net,
608 					       const void *addr)
609 {
610 	return ERR_PTR(-EAFNOSUPPORT);
611 }
612 #endif
613 
614 static struct net_device *find_outdev(struct net *net,
615 				      struct mpls_route *rt,
616 				      struct mpls_nh *nh, int oif)
617 {
618 	struct net_device *dev = NULL;
619 
620 	if (!oif) {
621 		switch (nh->nh_via_table) {
622 		case NEIGH_ARP_TABLE:
623 			dev = inet_fib_lookup_dev(net, mpls_nh_via(rt, nh));
624 			break;
625 		case NEIGH_ND_TABLE:
626 			dev = inet6_fib_lookup_dev(net, mpls_nh_via(rt, nh));
627 			break;
628 		case NEIGH_LINK_TABLE:
629 			break;
630 		}
631 	} else {
632 		dev = dev_get_by_index(net, oif);
633 	}
634 
635 	if (!dev)
636 		return ERR_PTR(-ENODEV);
637 
638 	if (IS_ERR(dev))
639 		return dev;
640 
641 	/* The caller is holding rtnl anyways, so release the dev reference */
642 	dev_put(dev);
643 
644 	return dev;
645 }
646 
647 static int mpls_nh_assign_dev(struct net *net, struct mpls_route *rt,
648 			      struct mpls_nh *nh, int oif)
649 {
650 	struct net_device *dev = NULL;
651 	int err = -ENODEV;
652 
653 	dev = find_outdev(net, rt, nh, oif);
654 	if (IS_ERR(dev)) {
655 		err = PTR_ERR(dev);
656 		dev = NULL;
657 		goto errout;
658 	}
659 
660 	/* Ensure this is a supported device */
661 	err = -EINVAL;
662 	if (!mpls_dev_get(dev))
663 		goto errout;
664 
665 	if ((nh->nh_via_table == NEIGH_LINK_TABLE) &&
666 	    (dev->addr_len != nh->nh_via_alen))
667 		goto errout;
668 
669 	RCU_INIT_POINTER(nh->nh_dev, dev);
670 
671 	if (!(dev->flags & IFF_UP)) {
672 		nh->nh_flags |= RTNH_F_DEAD;
673 	} else {
674 		unsigned int flags;
675 
676 		flags = dev_get_flags(dev);
677 		if (!(flags & (IFF_RUNNING | IFF_LOWER_UP)))
678 			nh->nh_flags |= RTNH_F_LINKDOWN;
679 	}
680 
681 	return 0;
682 
683 errout:
684 	return err;
685 }
686 
687 static int mpls_nh_build_from_cfg(struct mpls_route_config *cfg,
688 				  struct mpls_route *rt)
689 {
690 	struct net *net = cfg->rc_nlinfo.nl_net;
691 	struct mpls_nh *nh = rt->rt_nh;
692 	int err;
693 	int i;
694 
695 	if (!nh)
696 		return -ENOMEM;
697 
698 	err = -EINVAL;
699 
700 	nh->nh_labels = cfg->rc_output_labels;
701 	for (i = 0; i < nh->nh_labels; i++)
702 		nh->nh_label[i] = cfg->rc_output_label[i];
703 
704 	nh->nh_via_table = cfg->rc_via_table;
705 	memcpy(__mpls_nh_via(rt, nh), cfg->rc_via, cfg->rc_via_alen);
706 	nh->nh_via_alen = cfg->rc_via_alen;
707 
708 	err = mpls_nh_assign_dev(net, rt, nh, cfg->rc_ifindex);
709 	if (err)
710 		goto errout;
711 
712 	if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
713 		rt->rt_nhn_alive--;
714 
715 	return 0;
716 
717 errout:
718 	return err;
719 }
720 
721 static int mpls_nh_build(struct net *net, struct mpls_route *rt,
722 			 struct mpls_nh *nh, int oif, struct nlattr *via,
723 			 struct nlattr *newdst, u8 max_labels)
724 {
725 	int err = -ENOMEM;
726 
727 	if (!nh)
728 		goto errout;
729 
730 	if (newdst) {
731 		err = nla_get_labels(newdst, max_labels,
732 				     &nh->nh_labels, nh->nh_label);
733 		if (err)
734 			goto errout;
735 	}
736 
737 	if (via) {
738 		err = nla_get_via(via, &nh->nh_via_alen, &nh->nh_via_table,
739 				  __mpls_nh_via(rt, nh));
740 		if (err)
741 			goto errout;
742 	} else {
743 		nh->nh_via_table = MPLS_NEIGH_TABLE_UNSPEC;
744 	}
745 
746 	err = mpls_nh_assign_dev(net, rt, nh, oif);
747 	if (err)
748 		goto errout;
749 
750 	return 0;
751 
752 errout:
753 	return err;
754 }
755 
756 static u8 mpls_count_nexthops(struct rtnexthop *rtnh, int len,
757 			      u8 cfg_via_alen, u8 *max_via_alen,
758 			      u8 *max_labels)
759 {
760 	int remaining = len;
761 	u8 nhs = 0;
762 
763 	*max_via_alen = 0;
764 	*max_labels = 0;
765 
766 	while (rtnh_ok(rtnh, remaining)) {
767 		struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
768 		int attrlen;
769 		u8 n_labels = 0;
770 
771 		attrlen = rtnh_attrlen(rtnh);
772 		nla = nla_find(attrs, attrlen, RTA_VIA);
773 		if (nla && nla_len(nla) >=
774 		    offsetof(struct rtvia, rtvia_addr)) {
775 			int via_alen = nla_len(nla) -
776 				offsetof(struct rtvia, rtvia_addr);
777 
778 			if (via_alen <= MAX_VIA_ALEN)
779 				*max_via_alen = max_t(u16, *max_via_alen,
780 						      via_alen);
781 		}
782 
783 		nla = nla_find(attrs, attrlen, RTA_NEWDST);
784 		if (nla &&
785 		    nla_get_labels(nla, MAX_NEW_LABELS, &n_labels, NULL) != 0)
786 			return 0;
787 
788 		*max_labels = max_t(u8, *max_labels, n_labels);
789 
790 		/* number of nexthops is tracked by a u8.
791 		 * Check for overflow.
792 		 */
793 		if (nhs == 255)
794 			return 0;
795 		nhs++;
796 
797 		rtnh = rtnh_next(rtnh, &remaining);
798 	}
799 
800 	/* leftover implies invalid nexthop configuration, discard it */
801 	return remaining > 0 ? 0 : nhs;
802 }
803 
804 static int mpls_nh_build_multi(struct mpls_route_config *cfg,
805 			       struct mpls_route *rt, u8 max_labels)
806 {
807 	struct rtnexthop *rtnh = cfg->rc_mp;
808 	struct nlattr *nla_via, *nla_newdst;
809 	int remaining = cfg->rc_mp_len;
810 	int err = 0;
811 	u8 nhs = 0;
812 
813 	change_nexthops(rt) {
814 		int attrlen;
815 
816 		nla_via = NULL;
817 		nla_newdst = NULL;
818 
819 		err = -EINVAL;
820 		if (!rtnh_ok(rtnh, remaining))
821 			goto errout;
822 
823 		/* neither weighted multipath nor any flags
824 		 * are supported
825 		 */
826 		if (rtnh->rtnh_hops || rtnh->rtnh_flags)
827 			goto errout;
828 
829 		attrlen = rtnh_attrlen(rtnh);
830 		if (attrlen > 0) {
831 			struct nlattr *attrs = rtnh_attrs(rtnh);
832 
833 			nla_via = nla_find(attrs, attrlen, RTA_VIA);
834 			nla_newdst = nla_find(attrs, attrlen, RTA_NEWDST);
835 		}
836 
837 		err = mpls_nh_build(cfg->rc_nlinfo.nl_net, rt, nh,
838 				    rtnh->rtnh_ifindex, nla_via, nla_newdst,
839 				    max_labels);
840 		if (err)
841 			goto errout;
842 
843 		if (nh->nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
844 			rt->rt_nhn_alive--;
845 
846 		rtnh = rtnh_next(rtnh, &remaining);
847 		nhs++;
848 	} endfor_nexthops(rt);
849 
850 	rt->rt_nhn = nhs;
851 
852 	return 0;
853 
854 errout:
855 	return err;
856 }
857 
858 static int mpls_route_add(struct mpls_route_config *cfg)
859 {
860 	struct mpls_route __rcu **platform_label;
861 	struct net *net = cfg->rc_nlinfo.nl_net;
862 	struct mpls_route *rt, *old;
863 	int err = -EINVAL;
864 	u8 max_via_alen;
865 	unsigned index;
866 	u8 max_labels;
867 	u8 nhs;
868 
869 	index = cfg->rc_label;
870 
871 	/* If a label was not specified during insert pick one */
872 	if ((index == LABEL_NOT_SPECIFIED) &&
873 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
874 		index = find_free_label(net);
875 	}
876 
877 	/* Reserved labels may not be set */
878 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
879 		goto errout;
880 
881 	/* The full 20 bit range may not be supported. */
882 	if (index >= net->mpls.platform_labels)
883 		goto errout;
884 
885 	/* Append makes no sense with mpls */
886 	err = -EOPNOTSUPP;
887 	if (cfg->rc_nlflags & NLM_F_APPEND)
888 		goto errout;
889 
890 	err = -EEXIST;
891 	platform_label = rtnl_dereference(net->mpls.platform_label);
892 	old = rtnl_dereference(platform_label[index]);
893 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
894 		goto errout;
895 
896 	err = -EEXIST;
897 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
898 		goto errout;
899 
900 	err = -ENOENT;
901 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
902 		goto errout;
903 
904 	err = -EINVAL;
905 	if (cfg->rc_mp) {
906 		nhs = mpls_count_nexthops(cfg->rc_mp, cfg->rc_mp_len,
907 					  cfg->rc_via_alen, &max_via_alen,
908 					  &max_labels);
909 	} else {
910 		max_via_alen = cfg->rc_via_alen;
911 		max_labels = cfg->rc_output_labels;
912 		nhs = 1;
913 	}
914 
915 	if (nhs == 0)
916 		goto errout;
917 
918 	err = -ENOMEM;
919 	rt = mpls_rt_alloc(nhs, max_via_alen, max_labels);
920 	if (IS_ERR(rt)) {
921 		err = PTR_ERR(rt);
922 		goto errout;
923 	}
924 
925 	rt->rt_protocol = cfg->rc_protocol;
926 	rt->rt_payload_type = cfg->rc_payload_type;
927 	rt->rt_ttl_propagate = cfg->rc_ttl_propagate;
928 
929 	if (cfg->rc_mp)
930 		err = mpls_nh_build_multi(cfg, rt, max_labels);
931 	else
932 		err = mpls_nh_build_from_cfg(cfg, rt);
933 	if (err)
934 		goto freert;
935 
936 	mpls_route_update(net, index, rt, &cfg->rc_nlinfo);
937 
938 	return 0;
939 
940 freert:
941 	mpls_rt_free(rt);
942 errout:
943 	return err;
944 }
945 
946 static int mpls_route_del(struct mpls_route_config *cfg)
947 {
948 	struct net *net = cfg->rc_nlinfo.nl_net;
949 	unsigned index;
950 	int err = -EINVAL;
951 
952 	index = cfg->rc_label;
953 
954 	/* Reserved labels may not be removed */
955 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
956 		goto errout;
957 
958 	/* The full 20 bit range may not be supported */
959 	if (index >= net->mpls.platform_labels)
960 		goto errout;
961 
962 	mpls_route_update(net, index, NULL, &cfg->rc_nlinfo);
963 
964 	err = 0;
965 errout:
966 	return err;
967 }
968 
969 static void mpls_get_stats(struct mpls_dev *mdev,
970 			   struct mpls_link_stats *stats)
971 {
972 	struct mpls_pcpu_stats *p;
973 	int i;
974 
975 	memset(stats, 0, sizeof(*stats));
976 
977 	for_each_possible_cpu(i) {
978 		struct mpls_link_stats local;
979 		unsigned int start;
980 
981 		p = per_cpu_ptr(mdev->stats, i);
982 		do {
983 			start = u64_stats_fetch_begin(&p->syncp);
984 			local = p->stats;
985 		} while (u64_stats_fetch_retry(&p->syncp, start));
986 
987 		stats->rx_packets	+= local.rx_packets;
988 		stats->rx_bytes		+= local.rx_bytes;
989 		stats->tx_packets	+= local.tx_packets;
990 		stats->tx_bytes		+= local.tx_bytes;
991 		stats->rx_errors	+= local.rx_errors;
992 		stats->tx_errors	+= local.tx_errors;
993 		stats->rx_dropped	+= local.rx_dropped;
994 		stats->tx_dropped	+= local.tx_dropped;
995 		stats->rx_noroute	+= local.rx_noroute;
996 	}
997 }
998 
999 static int mpls_fill_stats_af(struct sk_buff *skb,
1000 			      const struct net_device *dev)
1001 {
1002 	struct mpls_link_stats *stats;
1003 	struct mpls_dev *mdev;
1004 	struct nlattr *nla;
1005 
1006 	mdev = mpls_dev_get(dev);
1007 	if (!mdev)
1008 		return -ENODATA;
1009 
1010 	nla = nla_reserve_64bit(skb, MPLS_STATS_LINK,
1011 				sizeof(struct mpls_link_stats),
1012 				MPLS_STATS_UNSPEC);
1013 	if (!nla)
1014 		return -EMSGSIZE;
1015 
1016 	stats = nla_data(nla);
1017 	mpls_get_stats(mdev, stats);
1018 
1019 	return 0;
1020 }
1021 
1022 static size_t mpls_get_stats_af_size(const struct net_device *dev)
1023 {
1024 	struct mpls_dev *mdev;
1025 
1026 	mdev = mpls_dev_get(dev);
1027 	if (!mdev)
1028 		return 0;
1029 
1030 	return nla_total_size_64bit(sizeof(struct mpls_link_stats));
1031 }
1032 
1033 static int mpls_netconf_fill_devconf(struct sk_buff *skb, struct mpls_dev *mdev,
1034 				     u32 portid, u32 seq, int event,
1035 				     unsigned int flags, int type)
1036 {
1037 	struct nlmsghdr  *nlh;
1038 	struct netconfmsg *ncm;
1039 	bool all = false;
1040 
1041 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(struct netconfmsg),
1042 			flags);
1043 	if (!nlh)
1044 		return -EMSGSIZE;
1045 
1046 	if (type == NETCONFA_ALL)
1047 		all = true;
1048 
1049 	ncm = nlmsg_data(nlh);
1050 	ncm->ncm_family = AF_MPLS;
1051 
1052 	if (nla_put_s32(skb, NETCONFA_IFINDEX, mdev->dev->ifindex) < 0)
1053 		goto nla_put_failure;
1054 
1055 	if ((all || type == NETCONFA_INPUT) &&
1056 	    nla_put_s32(skb, NETCONFA_INPUT,
1057 			mdev->input_enabled) < 0)
1058 		goto nla_put_failure;
1059 
1060 	nlmsg_end(skb, nlh);
1061 	return 0;
1062 
1063 nla_put_failure:
1064 	nlmsg_cancel(skb, nlh);
1065 	return -EMSGSIZE;
1066 }
1067 
1068 static int mpls_netconf_msgsize_devconf(int type)
1069 {
1070 	int size = NLMSG_ALIGN(sizeof(struct netconfmsg))
1071 			+ nla_total_size(4); /* NETCONFA_IFINDEX */
1072 	bool all = false;
1073 
1074 	if (type == NETCONFA_ALL)
1075 		all = true;
1076 
1077 	if (all || type == NETCONFA_INPUT)
1078 		size += nla_total_size(4);
1079 
1080 	return size;
1081 }
1082 
1083 static void mpls_netconf_notify_devconf(struct net *net, int event,
1084 					int type, struct mpls_dev *mdev)
1085 {
1086 	struct sk_buff *skb;
1087 	int err = -ENOBUFS;
1088 
1089 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(type), GFP_KERNEL);
1090 	if (!skb)
1091 		goto errout;
1092 
1093 	err = mpls_netconf_fill_devconf(skb, mdev, 0, 0, event, 0, type);
1094 	if (err < 0) {
1095 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1096 		WARN_ON(err == -EMSGSIZE);
1097 		kfree_skb(skb);
1098 		goto errout;
1099 	}
1100 
1101 	rtnl_notify(skb, net, 0, RTNLGRP_MPLS_NETCONF, NULL, GFP_KERNEL);
1102 	return;
1103 errout:
1104 	if (err < 0)
1105 		rtnl_set_sk_err(net, RTNLGRP_MPLS_NETCONF, err);
1106 }
1107 
1108 static const struct nla_policy devconf_mpls_policy[NETCONFA_MAX + 1] = {
1109 	[NETCONFA_IFINDEX]	= { .len = sizeof(int) },
1110 };
1111 
1112 static int mpls_netconf_get_devconf(struct sk_buff *in_skb,
1113 				    struct nlmsghdr *nlh)
1114 {
1115 	struct net *net = sock_net(in_skb->sk);
1116 	struct nlattr *tb[NETCONFA_MAX + 1];
1117 	struct netconfmsg *ncm;
1118 	struct net_device *dev;
1119 	struct mpls_dev *mdev;
1120 	struct sk_buff *skb;
1121 	int ifindex;
1122 	int err;
1123 
1124 	err = nlmsg_parse(nlh, sizeof(*ncm), tb, NETCONFA_MAX,
1125 			  devconf_mpls_policy);
1126 	if (err < 0)
1127 		goto errout;
1128 
1129 	err = -EINVAL;
1130 	if (!tb[NETCONFA_IFINDEX])
1131 		goto errout;
1132 
1133 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
1134 	dev = __dev_get_by_index(net, ifindex);
1135 	if (!dev)
1136 		goto errout;
1137 
1138 	mdev = mpls_dev_get(dev);
1139 	if (!mdev)
1140 		goto errout;
1141 
1142 	err = -ENOBUFS;
1143 	skb = nlmsg_new(mpls_netconf_msgsize_devconf(NETCONFA_ALL), GFP_KERNEL);
1144 	if (!skb)
1145 		goto errout;
1146 
1147 	err = mpls_netconf_fill_devconf(skb, mdev,
1148 					NETLINK_CB(in_skb).portid,
1149 					nlh->nlmsg_seq, RTM_NEWNETCONF, 0,
1150 					NETCONFA_ALL);
1151 	if (err < 0) {
1152 		/* -EMSGSIZE implies BUG in mpls_netconf_msgsize_devconf() */
1153 		WARN_ON(err == -EMSGSIZE);
1154 		kfree_skb(skb);
1155 		goto errout;
1156 	}
1157 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1158 errout:
1159 	return err;
1160 }
1161 
1162 static int mpls_netconf_dump_devconf(struct sk_buff *skb,
1163 				     struct netlink_callback *cb)
1164 {
1165 	struct net *net = sock_net(skb->sk);
1166 	struct hlist_head *head;
1167 	struct net_device *dev;
1168 	struct mpls_dev *mdev;
1169 	int idx, s_idx;
1170 	int h, s_h;
1171 
1172 	s_h = cb->args[0];
1173 	s_idx = idx = cb->args[1];
1174 
1175 	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
1176 		idx = 0;
1177 		head = &net->dev_index_head[h];
1178 		rcu_read_lock();
1179 		cb->seq = net->dev_base_seq;
1180 		hlist_for_each_entry_rcu(dev, head, index_hlist) {
1181 			if (idx < s_idx)
1182 				goto cont;
1183 			mdev = mpls_dev_get(dev);
1184 			if (!mdev)
1185 				goto cont;
1186 			if (mpls_netconf_fill_devconf(skb, mdev,
1187 						      NETLINK_CB(cb->skb).portid,
1188 						      cb->nlh->nlmsg_seq,
1189 						      RTM_NEWNETCONF,
1190 						      NLM_F_MULTI,
1191 						      NETCONFA_ALL) < 0) {
1192 				rcu_read_unlock();
1193 				goto done;
1194 			}
1195 			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1196 cont:
1197 			idx++;
1198 		}
1199 		rcu_read_unlock();
1200 	}
1201 done:
1202 	cb->args[0] = h;
1203 	cb->args[1] = idx;
1204 
1205 	return skb->len;
1206 }
1207 
1208 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
1209 	(&((struct mpls_dev *)0)->field)
1210 
1211 static int mpls_conf_proc(struct ctl_table *ctl, int write,
1212 			  void __user *buffer,
1213 			  size_t *lenp, loff_t *ppos)
1214 {
1215 	int oval = *(int *)ctl->data;
1216 	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1217 
1218 	if (write) {
1219 		struct mpls_dev *mdev = ctl->extra1;
1220 		int i = (int *)ctl->data - (int *)mdev;
1221 		struct net *net = ctl->extra2;
1222 		int val = *(int *)ctl->data;
1223 
1224 		if (i == offsetof(struct mpls_dev, input_enabled) &&
1225 		    val != oval) {
1226 			mpls_netconf_notify_devconf(net, RTM_NEWNETCONF,
1227 						    NETCONFA_INPUT, mdev);
1228 		}
1229 	}
1230 
1231 	return ret;
1232 }
1233 
1234 static const struct ctl_table mpls_dev_table[] = {
1235 	{
1236 		.procname	= "input",
1237 		.maxlen		= sizeof(int),
1238 		.mode		= 0644,
1239 		.proc_handler	= mpls_conf_proc,
1240 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
1241 	},
1242 	{ }
1243 };
1244 
1245 static int mpls_dev_sysctl_register(struct net_device *dev,
1246 				    struct mpls_dev *mdev)
1247 {
1248 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
1249 	struct net *net = dev_net(dev);
1250 	struct ctl_table *table;
1251 	int i;
1252 
1253 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
1254 	if (!table)
1255 		goto out;
1256 
1257 	/* Table data contains only offsets relative to the base of
1258 	 * the mdev at this point, so make them absolute.
1259 	 */
1260 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++) {
1261 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
1262 		table[i].extra1 = mdev;
1263 		table[i].extra2 = net;
1264 	}
1265 
1266 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
1267 
1268 	mdev->sysctl = register_net_sysctl(net, path, table);
1269 	if (!mdev->sysctl)
1270 		goto free;
1271 
1272 	mpls_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_ALL, mdev);
1273 	return 0;
1274 
1275 free:
1276 	kfree(table);
1277 out:
1278 	return -ENOBUFS;
1279 }
1280 
1281 static void mpls_dev_sysctl_unregister(struct net_device *dev,
1282 				       struct mpls_dev *mdev)
1283 {
1284 	struct net *net = dev_net(dev);
1285 	struct ctl_table *table;
1286 
1287 	table = mdev->sysctl->ctl_table_arg;
1288 	unregister_net_sysctl_table(mdev->sysctl);
1289 	kfree(table);
1290 
1291 	mpls_netconf_notify_devconf(net, RTM_DELNETCONF, 0, mdev);
1292 }
1293 
1294 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
1295 {
1296 	struct mpls_dev *mdev;
1297 	int err = -ENOMEM;
1298 	int i;
1299 
1300 	ASSERT_RTNL();
1301 
1302 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
1303 	if (!mdev)
1304 		return ERR_PTR(err);
1305 
1306 	mdev->stats = alloc_percpu(struct mpls_pcpu_stats);
1307 	if (!mdev->stats)
1308 		goto free;
1309 
1310 	for_each_possible_cpu(i) {
1311 		struct mpls_pcpu_stats *mpls_stats;
1312 
1313 		mpls_stats = per_cpu_ptr(mdev->stats, i);
1314 		u64_stats_init(&mpls_stats->syncp);
1315 	}
1316 
1317 	mdev->dev = dev;
1318 
1319 	err = mpls_dev_sysctl_register(dev, mdev);
1320 	if (err)
1321 		goto free;
1322 
1323 	rcu_assign_pointer(dev->mpls_ptr, mdev);
1324 
1325 	return mdev;
1326 
1327 free:
1328 	free_percpu(mdev->stats);
1329 	kfree(mdev);
1330 	return ERR_PTR(err);
1331 }
1332 
1333 static void mpls_dev_destroy_rcu(struct rcu_head *head)
1334 {
1335 	struct mpls_dev *mdev = container_of(head, struct mpls_dev, rcu);
1336 
1337 	free_percpu(mdev->stats);
1338 	kfree(mdev);
1339 }
1340 
1341 static void mpls_ifdown(struct net_device *dev, int event)
1342 {
1343 	struct mpls_route __rcu **platform_label;
1344 	struct net *net = dev_net(dev);
1345 	u8 alive, deleted;
1346 	unsigned index;
1347 
1348 	platform_label = rtnl_dereference(net->mpls.platform_label);
1349 	for (index = 0; index < net->mpls.platform_labels; index++) {
1350 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1351 
1352 		if (!rt)
1353 			continue;
1354 
1355 		alive = 0;
1356 		deleted = 0;
1357 		change_nexthops(rt) {
1358 			unsigned int nh_flags = nh->nh_flags;
1359 
1360 			if (rtnl_dereference(nh->nh_dev) != dev)
1361 				goto next;
1362 
1363 			switch (event) {
1364 			case NETDEV_DOWN:
1365 			case NETDEV_UNREGISTER:
1366 				nh_flags |= RTNH_F_DEAD;
1367 				/* fall through */
1368 			case NETDEV_CHANGE:
1369 				nh_flags |= RTNH_F_LINKDOWN;
1370 				break;
1371 			}
1372 			if (event == NETDEV_UNREGISTER)
1373 				RCU_INIT_POINTER(nh->nh_dev, NULL);
1374 
1375 			if (nh->nh_flags != nh_flags)
1376 				WRITE_ONCE(nh->nh_flags, nh_flags);
1377 next:
1378 			if (!(nh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)))
1379 				alive++;
1380 			if (!rtnl_dereference(nh->nh_dev))
1381 				deleted++;
1382 		} endfor_nexthops(rt);
1383 
1384 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1385 
1386 		/* if there are no more nexthops, delete the route */
1387 		if (event == NETDEV_UNREGISTER && deleted == rt->rt_nhn)
1388 			mpls_route_update(net, index, NULL, NULL);
1389 	}
1390 }
1391 
1392 static void mpls_ifup(struct net_device *dev, unsigned int flags)
1393 {
1394 	struct mpls_route __rcu **platform_label;
1395 	struct net *net = dev_net(dev);
1396 	unsigned index;
1397 	u8 alive;
1398 
1399 	platform_label = rtnl_dereference(net->mpls.platform_label);
1400 	for (index = 0; index < net->mpls.platform_labels; index++) {
1401 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1402 
1403 		if (!rt)
1404 			continue;
1405 
1406 		alive = 0;
1407 		change_nexthops(rt) {
1408 			unsigned int nh_flags = nh->nh_flags;
1409 			struct net_device *nh_dev =
1410 				rtnl_dereference(nh->nh_dev);
1411 
1412 			if (!(nh_flags & flags)) {
1413 				alive++;
1414 				continue;
1415 			}
1416 			if (nh_dev != dev)
1417 				continue;
1418 			alive++;
1419 			nh_flags &= ~flags;
1420 			WRITE_ONCE(nh->nh_flags, flags);
1421 		} endfor_nexthops(rt);
1422 
1423 		WRITE_ONCE(rt->rt_nhn_alive, alive);
1424 	}
1425 }
1426 
1427 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
1428 			   void *ptr)
1429 {
1430 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1431 	struct mpls_dev *mdev;
1432 	unsigned int flags;
1433 
1434 	if (event == NETDEV_REGISTER) {
1435 		/* For now just support Ethernet, IPGRE, SIT and IPIP devices */
1436 		if (dev->type == ARPHRD_ETHER ||
1437 		    dev->type == ARPHRD_LOOPBACK ||
1438 		    dev->type == ARPHRD_IPGRE ||
1439 		    dev->type == ARPHRD_SIT ||
1440 		    dev->type == ARPHRD_TUNNEL) {
1441 			mdev = mpls_add_dev(dev);
1442 			if (IS_ERR(mdev))
1443 				return notifier_from_errno(PTR_ERR(mdev));
1444 		}
1445 		return NOTIFY_OK;
1446 	}
1447 
1448 	mdev = mpls_dev_get(dev);
1449 	if (!mdev)
1450 		return NOTIFY_OK;
1451 
1452 	switch (event) {
1453 	case NETDEV_DOWN:
1454 		mpls_ifdown(dev, event);
1455 		break;
1456 	case NETDEV_UP:
1457 		flags = dev_get_flags(dev);
1458 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1459 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1460 		else
1461 			mpls_ifup(dev, RTNH_F_DEAD);
1462 		break;
1463 	case NETDEV_CHANGE:
1464 		flags = dev_get_flags(dev);
1465 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1466 			mpls_ifup(dev, RTNH_F_DEAD | RTNH_F_LINKDOWN);
1467 		else
1468 			mpls_ifdown(dev, event);
1469 		break;
1470 	case NETDEV_UNREGISTER:
1471 		mpls_ifdown(dev, event);
1472 		mdev = mpls_dev_get(dev);
1473 		if (mdev) {
1474 			mpls_dev_sysctl_unregister(dev, mdev);
1475 			RCU_INIT_POINTER(dev->mpls_ptr, NULL);
1476 			call_rcu(&mdev->rcu, mpls_dev_destroy_rcu);
1477 		}
1478 		break;
1479 	case NETDEV_CHANGENAME:
1480 		mdev = mpls_dev_get(dev);
1481 		if (mdev) {
1482 			int err;
1483 
1484 			mpls_dev_sysctl_unregister(dev, mdev);
1485 			err = mpls_dev_sysctl_register(dev, mdev);
1486 			if (err)
1487 				return notifier_from_errno(err);
1488 		}
1489 		break;
1490 	}
1491 	return NOTIFY_OK;
1492 }
1493 
1494 static struct notifier_block mpls_dev_notifier = {
1495 	.notifier_call = mpls_dev_notify,
1496 };
1497 
1498 static int nla_put_via(struct sk_buff *skb,
1499 		       u8 table, const void *addr, int alen)
1500 {
1501 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
1502 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
1503 	};
1504 	struct nlattr *nla;
1505 	struct rtvia *via;
1506 	int family = AF_UNSPEC;
1507 
1508 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
1509 	if (!nla)
1510 		return -EMSGSIZE;
1511 
1512 	if (table <= NEIGH_NR_TABLES)
1513 		family = table_to_family[table];
1514 
1515 	via = nla_data(nla);
1516 	via->rtvia_family = family;
1517 	memcpy(via->rtvia_addr, addr, alen);
1518 	return 0;
1519 }
1520 
1521 int nla_put_labels(struct sk_buff *skb, int attrtype,
1522 		   u8 labels, const u32 label[])
1523 {
1524 	struct nlattr *nla;
1525 	struct mpls_shim_hdr *nla_label;
1526 	bool bos;
1527 	int i;
1528 	nla = nla_reserve(skb, attrtype, labels*4);
1529 	if (!nla)
1530 		return -EMSGSIZE;
1531 
1532 	nla_label = nla_data(nla);
1533 	bos = true;
1534 	for (i = labels - 1; i >= 0; i--) {
1535 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
1536 		bos = false;
1537 	}
1538 
1539 	return 0;
1540 }
1541 EXPORT_SYMBOL_GPL(nla_put_labels);
1542 
1543 int nla_get_labels(const struct nlattr *nla,
1544 		   u8 max_labels, u8 *labels, u32 label[])
1545 {
1546 	unsigned len = nla_len(nla);
1547 	struct mpls_shim_hdr *nla_label;
1548 	u8 nla_labels;
1549 	bool bos;
1550 	int i;
1551 
1552 	/* len needs to be an even multiple of 4 (the label size). Number
1553 	 * of labels is a u8 so check for overflow.
1554 	 */
1555 	if (len & 3 || len / 4 > 255)
1556 		return -EINVAL;
1557 
1558 	/* Limit the number of new labels allowed */
1559 	nla_labels = len/4;
1560 	if (nla_labels > max_labels)
1561 		return -EINVAL;
1562 
1563 	/* when label == NULL, caller wants number of labels */
1564 	if (!label)
1565 		goto out;
1566 
1567 	nla_label = nla_data(nla);
1568 	bos = true;
1569 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
1570 		struct mpls_entry_decoded dec;
1571 		dec = mpls_entry_decode(nla_label + i);
1572 
1573 		/* Ensure the bottom of stack flag is properly set
1574 		 * and ttl and tc are both clear.
1575 		 */
1576 		if ((dec.bos != bos) || dec.ttl || dec.tc)
1577 			return -EINVAL;
1578 
1579 		switch (dec.label) {
1580 		case MPLS_LABEL_IMPLNULL:
1581 			/* RFC3032: This is a label that an LSR may
1582 			 * assign and distribute, but which never
1583 			 * actually appears in the encapsulation.
1584 			 */
1585 			return -EINVAL;
1586 		}
1587 
1588 		label[i] = dec.label;
1589 	}
1590 out:
1591 	*labels = nla_labels;
1592 	return 0;
1593 }
1594 EXPORT_SYMBOL_GPL(nla_get_labels);
1595 
1596 int nla_get_via(const struct nlattr *nla, u8 *via_alen,
1597 		u8 *via_table, u8 via_addr[])
1598 {
1599 	struct rtvia *via = nla_data(nla);
1600 	int err = -EINVAL;
1601 	int alen;
1602 
1603 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
1604 		goto errout;
1605 	alen = nla_len(nla) -
1606 			offsetof(struct rtvia, rtvia_addr);
1607 	if (alen > MAX_VIA_ALEN)
1608 		goto errout;
1609 
1610 	/* Validate the address family */
1611 	switch (via->rtvia_family) {
1612 	case AF_PACKET:
1613 		*via_table = NEIGH_LINK_TABLE;
1614 		break;
1615 	case AF_INET:
1616 		*via_table = NEIGH_ARP_TABLE;
1617 		if (alen != 4)
1618 			goto errout;
1619 		break;
1620 	case AF_INET6:
1621 		*via_table = NEIGH_ND_TABLE;
1622 		if (alen != 16)
1623 			goto errout;
1624 		break;
1625 	default:
1626 		/* Unsupported address family */
1627 		goto errout;
1628 	}
1629 
1630 	memcpy(via_addr, via->rtvia_addr, alen);
1631 	*via_alen = alen;
1632 	err = 0;
1633 
1634 errout:
1635 	return err;
1636 }
1637 
1638 static int rtm_to_route_config(struct sk_buff *skb,  struct nlmsghdr *nlh,
1639 			       struct mpls_route_config *cfg)
1640 {
1641 	struct rtmsg *rtm;
1642 	struct nlattr *tb[RTA_MAX+1];
1643 	int index;
1644 	int err;
1645 
1646 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
1647 	if (err < 0)
1648 		goto errout;
1649 
1650 	err = -EINVAL;
1651 	rtm = nlmsg_data(nlh);
1652 
1653 	if (rtm->rtm_family != AF_MPLS)
1654 		goto errout;
1655 	if (rtm->rtm_dst_len != 20)
1656 		goto errout;
1657 	if (rtm->rtm_src_len != 0)
1658 		goto errout;
1659 	if (rtm->rtm_tos != 0)
1660 		goto errout;
1661 	if (rtm->rtm_table != RT_TABLE_MAIN)
1662 		goto errout;
1663 	/* Any value is acceptable for rtm_protocol */
1664 
1665 	/* As mpls uses destination specific addresses
1666 	 * (or source specific address in the case of multicast)
1667 	 * all addresses have universal scope.
1668 	 */
1669 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
1670 		goto errout;
1671 	if (rtm->rtm_type != RTN_UNICAST)
1672 		goto errout;
1673 	if (rtm->rtm_flags != 0)
1674 		goto errout;
1675 
1676 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
1677 	cfg->rc_protocol	= rtm->rtm_protocol;
1678 	cfg->rc_via_table	= MPLS_NEIGH_TABLE_UNSPEC;
1679 	cfg->rc_ttl_propagate	= MPLS_TTL_PROP_DEFAULT;
1680 	cfg->rc_nlflags		= nlh->nlmsg_flags;
1681 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
1682 	cfg->rc_nlinfo.nlh	= nlh;
1683 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
1684 
1685 	for (index = 0; index <= RTA_MAX; index++) {
1686 		struct nlattr *nla = tb[index];
1687 		if (!nla)
1688 			continue;
1689 
1690 		switch (index) {
1691 		case RTA_OIF:
1692 			cfg->rc_ifindex = nla_get_u32(nla);
1693 			break;
1694 		case RTA_NEWDST:
1695 			if (nla_get_labels(nla, MAX_NEW_LABELS,
1696 					   &cfg->rc_output_labels,
1697 					   cfg->rc_output_label))
1698 				goto errout;
1699 			break;
1700 		case RTA_DST:
1701 		{
1702 			u8 label_count;
1703 			if (nla_get_labels(nla, 1, &label_count,
1704 					   &cfg->rc_label))
1705 				goto errout;
1706 
1707 			/* Reserved labels may not be set */
1708 			if (cfg->rc_label < MPLS_LABEL_FIRST_UNRESERVED)
1709 				goto errout;
1710 
1711 			break;
1712 		}
1713 		case RTA_VIA:
1714 		{
1715 			if (nla_get_via(nla, &cfg->rc_via_alen,
1716 					&cfg->rc_via_table, cfg->rc_via))
1717 				goto errout;
1718 			break;
1719 		}
1720 		case RTA_MULTIPATH:
1721 		{
1722 			cfg->rc_mp = nla_data(nla);
1723 			cfg->rc_mp_len = nla_len(nla);
1724 			break;
1725 		}
1726 		case RTA_TTL_PROPAGATE:
1727 		{
1728 			u8 ttl_propagate = nla_get_u8(nla);
1729 
1730 			if (ttl_propagate > 1)
1731 				goto errout;
1732 			cfg->rc_ttl_propagate = ttl_propagate ?
1733 				MPLS_TTL_PROP_ENABLED :
1734 				MPLS_TTL_PROP_DISABLED;
1735 			break;
1736 		}
1737 		default:
1738 			/* Unsupported attribute */
1739 			goto errout;
1740 		}
1741 	}
1742 
1743 	err = 0;
1744 errout:
1745 	return err;
1746 }
1747 
1748 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1749 {
1750 	struct mpls_route_config *cfg;
1751 	int err;
1752 
1753 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1754 	if (!cfg)
1755 		return -ENOMEM;
1756 
1757 	err = rtm_to_route_config(skb, nlh, cfg);
1758 	if (err < 0)
1759 		goto out;
1760 
1761 	err = mpls_route_del(cfg);
1762 out:
1763 	kfree(cfg);
1764 
1765 	return err;
1766 }
1767 
1768 
1769 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
1770 {
1771 	struct mpls_route_config *cfg;
1772 	int err;
1773 
1774 	cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1775 	if (!cfg)
1776 		return -ENOMEM;
1777 
1778 	err = rtm_to_route_config(skb, nlh, cfg);
1779 	if (err < 0)
1780 		goto out;
1781 
1782 	err = mpls_route_add(cfg);
1783 out:
1784 	kfree(cfg);
1785 
1786 	return err;
1787 }
1788 
1789 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
1790 			   u32 label, struct mpls_route *rt, int flags)
1791 {
1792 	struct net_device *dev;
1793 	struct nlmsghdr *nlh;
1794 	struct rtmsg *rtm;
1795 
1796 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1797 	if (nlh == NULL)
1798 		return -EMSGSIZE;
1799 
1800 	rtm = nlmsg_data(nlh);
1801 	rtm->rtm_family = AF_MPLS;
1802 	rtm->rtm_dst_len = 20;
1803 	rtm->rtm_src_len = 0;
1804 	rtm->rtm_tos = 0;
1805 	rtm->rtm_table = RT_TABLE_MAIN;
1806 	rtm->rtm_protocol = rt->rt_protocol;
1807 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1808 	rtm->rtm_type = RTN_UNICAST;
1809 	rtm->rtm_flags = 0;
1810 
1811 	if (nla_put_labels(skb, RTA_DST, 1, &label))
1812 		goto nla_put_failure;
1813 
1814 	if (rt->rt_ttl_propagate != MPLS_TTL_PROP_DEFAULT) {
1815 		bool ttl_propagate =
1816 			rt->rt_ttl_propagate == MPLS_TTL_PROP_ENABLED;
1817 
1818 		if (nla_put_u8(skb, RTA_TTL_PROPAGATE,
1819 			       ttl_propagate))
1820 			goto nla_put_failure;
1821 	}
1822 	if (rt->rt_nhn == 1) {
1823 		const struct mpls_nh *nh = rt->rt_nh;
1824 
1825 		if (nh->nh_labels &&
1826 		    nla_put_labels(skb, RTA_NEWDST, nh->nh_labels,
1827 				   nh->nh_label))
1828 			goto nla_put_failure;
1829 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1830 		    nla_put_via(skb, nh->nh_via_table, mpls_nh_via(rt, nh),
1831 				nh->nh_via_alen))
1832 			goto nla_put_failure;
1833 		dev = rtnl_dereference(nh->nh_dev);
1834 		if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
1835 			goto nla_put_failure;
1836 		if (nh->nh_flags & RTNH_F_LINKDOWN)
1837 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1838 		if (nh->nh_flags & RTNH_F_DEAD)
1839 			rtm->rtm_flags |= RTNH_F_DEAD;
1840 	} else {
1841 		struct rtnexthop *rtnh;
1842 		struct nlattr *mp;
1843 		u8 linkdown = 0;
1844 		u8 dead = 0;
1845 
1846 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1847 		if (!mp)
1848 			goto nla_put_failure;
1849 
1850 		for_nexthops(rt) {
1851 			dev = rtnl_dereference(nh->nh_dev);
1852 			if (!dev)
1853 				continue;
1854 
1855 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1856 			if (!rtnh)
1857 				goto nla_put_failure;
1858 
1859 			rtnh->rtnh_ifindex = dev->ifindex;
1860 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1861 				rtnh->rtnh_flags |= RTNH_F_LINKDOWN;
1862 				linkdown++;
1863 			}
1864 			if (nh->nh_flags & RTNH_F_DEAD) {
1865 				rtnh->rtnh_flags |= RTNH_F_DEAD;
1866 				dead++;
1867 			}
1868 
1869 			if (nh->nh_labels && nla_put_labels(skb, RTA_NEWDST,
1870 							    nh->nh_labels,
1871 							    nh->nh_label))
1872 				goto nla_put_failure;
1873 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC &&
1874 			    nla_put_via(skb, nh->nh_via_table,
1875 					mpls_nh_via(rt, nh),
1876 					nh->nh_via_alen))
1877 				goto nla_put_failure;
1878 
1879 			/* length of rtnetlink header + attributes */
1880 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1881 		} endfor_nexthops(rt);
1882 
1883 		if (linkdown == rt->rt_nhn)
1884 			rtm->rtm_flags |= RTNH_F_LINKDOWN;
1885 		if (dead == rt->rt_nhn)
1886 			rtm->rtm_flags |= RTNH_F_DEAD;
1887 
1888 		nla_nest_end(skb, mp);
1889 	}
1890 
1891 	nlmsg_end(skb, nlh);
1892 	return 0;
1893 
1894 nla_put_failure:
1895 	nlmsg_cancel(skb, nlh);
1896 	return -EMSGSIZE;
1897 }
1898 
1899 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
1900 {
1901 	struct net *net = sock_net(skb->sk);
1902 	struct mpls_route __rcu **platform_label;
1903 	size_t platform_labels;
1904 	unsigned int index;
1905 
1906 	ASSERT_RTNL();
1907 
1908 	index = cb->args[0];
1909 	if (index < MPLS_LABEL_FIRST_UNRESERVED)
1910 		index = MPLS_LABEL_FIRST_UNRESERVED;
1911 
1912 	platform_label = rtnl_dereference(net->mpls.platform_label);
1913 	platform_labels = net->mpls.platform_labels;
1914 	for (; index < platform_labels; index++) {
1915 		struct mpls_route *rt;
1916 		rt = rtnl_dereference(platform_label[index]);
1917 		if (!rt)
1918 			continue;
1919 
1920 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
1921 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1922 				    index, rt, NLM_F_MULTI) < 0)
1923 			break;
1924 	}
1925 	cb->args[0] = index;
1926 
1927 	return skb->len;
1928 }
1929 
1930 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
1931 {
1932 	size_t payload =
1933 		NLMSG_ALIGN(sizeof(struct rtmsg))
1934 		+ nla_total_size(4)			/* RTA_DST */
1935 		+ nla_total_size(1);			/* RTA_TTL_PROPAGATE */
1936 
1937 	if (rt->rt_nhn == 1) {
1938 		struct mpls_nh *nh = rt->rt_nh;
1939 
1940 		if (nh->nh_dev)
1941 			payload += nla_total_size(4); /* RTA_OIF */
1942 		if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC) /* RTA_VIA */
1943 			payload += nla_total_size(2 + nh->nh_via_alen);
1944 		if (nh->nh_labels) /* RTA_NEWDST */
1945 			payload += nla_total_size(nh->nh_labels * 4);
1946 	} else {
1947 		/* each nexthop is packed in an attribute */
1948 		size_t nhsize = 0;
1949 
1950 		for_nexthops(rt) {
1951 			if (!rtnl_dereference(nh->nh_dev))
1952 				continue;
1953 			nhsize += nla_total_size(sizeof(struct rtnexthop));
1954 			/* RTA_VIA */
1955 			if (nh->nh_via_table != MPLS_NEIGH_TABLE_UNSPEC)
1956 				nhsize += nla_total_size(2 + nh->nh_via_alen);
1957 			if (nh->nh_labels)
1958 				nhsize += nla_total_size(nh->nh_labels * 4);
1959 		} endfor_nexthops(rt);
1960 		/* nested attribute */
1961 		payload += nla_total_size(nhsize);
1962 	}
1963 
1964 	return payload;
1965 }
1966 
1967 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
1968 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
1969 		       unsigned int nlm_flags)
1970 {
1971 	struct sk_buff *skb;
1972 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
1973 	int err = -ENOBUFS;
1974 
1975 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
1976 	if (skb == NULL)
1977 		goto errout;
1978 
1979 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
1980 	if (err < 0) {
1981 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
1982 		WARN_ON(err == -EMSGSIZE);
1983 		kfree_skb(skb);
1984 		goto errout;
1985 	}
1986 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
1987 
1988 	return;
1989 errout:
1990 	if (err < 0)
1991 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
1992 }
1993 
1994 static int resize_platform_label_table(struct net *net, size_t limit)
1995 {
1996 	size_t size = sizeof(struct mpls_route *) * limit;
1997 	size_t old_limit;
1998 	size_t cp_size;
1999 	struct mpls_route __rcu **labels = NULL, **old;
2000 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
2001 	unsigned index;
2002 
2003 	if (size) {
2004 		labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
2005 		if (!labels)
2006 			labels = vzalloc(size);
2007 
2008 		if (!labels)
2009 			goto nolabels;
2010 	}
2011 
2012 	/* In case the predefined labels need to be populated */
2013 	if (limit > MPLS_LABEL_IPV4NULL) {
2014 		struct net_device *lo = net->loopback_dev;
2015 		rt0 = mpls_rt_alloc(1, lo->addr_len, 0);
2016 		if (IS_ERR(rt0))
2017 			goto nort0;
2018 		RCU_INIT_POINTER(rt0->rt_nh->nh_dev, lo);
2019 		rt0->rt_protocol = RTPROT_KERNEL;
2020 		rt0->rt_payload_type = MPT_IPV4;
2021 		rt0->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2022 		rt0->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2023 		rt0->rt_nh->nh_via_alen = lo->addr_len;
2024 		memcpy(__mpls_nh_via(rt0, rt0->rt_nh), lo->dev_addr,
2025 		       lo->addr_len);
2026 	}
2027 	if (limit > MPLS_LABEL_IPV6NULL) {
2028 		struct net_device *lo = net->loopback_dev;
2029 		rt2 = mpls_rt_alloc(1, lo->addr_len, 0);
2030 		if (IS_ERR(rt2))
2031 			goto nort2;
2032 		RCU_INIT_POINTER(rt2->rt_nh->nh_dev, lo);
2033 		rt2->rt_protocol = RTPROT_KERNEL;
2034 		rt2->rt_payload_type = MPT_IPV6;
2035 		rt2->rt_ttl_propagate = MPLS_TTL_PROP_DEFAULT;
2036 		rt2->rt_nh->nh_via_table = NEIGH_LINK_TABLE;
2037 		rt2->rt_nh->nh_via_alen = lo->addr_len;
2038 		memcpy(__mpls_nh_via(rt2, rt2->rt_nh), lo->dev_addr,
2039 		       lo->addr_len);
2040 	}
2041 
2042 	rtnl_lock();
2043 	/* Remember the original table */
2044 	old = rtnl_dereference(net->mpls.platform_label);
2045 	old_limit = net->mpls.platform_labels;
2046 
2047 	/* Free any labels beyond the new table */
2048 	for (index = limit; index < old_limit; index++)
2049 		mpls_route_update(net, index, NULL, NULL);
2050 
2051 	/* Copy over the old labels */
2052 	cp_size = size;
2053 	if (old_limit < limit)
2054 		cp_size = old_limit * sizeof(struct mpls_route *);
2055 
2056 	memcpy(labels, old, cp_size);
2057 
2058 	/* If needed set the predefined labels */
2059 	if ((old_limit <= MPLS_LABEL_IPV6NULL) &&
2060 	    (limit > MPLS_LABEL_IPV6NULL)) {
2061 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV6NULL], rt2);
2062 		rt2 = NULL;
2063 	}
2064 
2065 	if ((old_limit <= MPLS_LABEL_IPV4NULL) &&
2066 	    (limit > MPLS_LABEL_IPV4NULL)) {
2067 		RCU_INIT_POINTER(labels[MPLS_LABEL_IPV4NULL], rt0);
2068 		rt0 = NULL;
2069 	}
2070 
2071 	/* Update the global pointers */
2072 	net->mpls.platform_labels = limit;
2073 	rcu_assign_pointer(net->mpls.platform_label, labels);
2074 
2075 	rtnl_unlock();
2076 
2077 	mpls_rt_free(rt2);
2078 	mpls_rt_free(rt0);
2079 
2080 	if (old) {
2081 		synchronize_rcu();
2082 		kvfree(old);
2083 	}
2084 	return 0;
2085 
2086 nort2:
2087 	mpls_rt_free(rt0);
2088 nort0:
2089 	kvfree(labels);
2090 nolabels:
2091 	return -ENOMEM;
2092 }
2093 
2094 static int mpls_platform_labels(struct ctl_table *table, int write,
2095 				void __user *buffer, size_t *lenp, loff_t *ppos)
2096 {
2097 	struct net *net = table->data;
2098 	int platform_labels = net->mpls.platform_labels;
2099 	int ret;
2100 	struct ctl_table tmp = {
2101 		.procname	= table->procname,
2102 		.data		= &platform_labels,
2103 		.maxlen		= sizeof(int),
2104 		.mode		= table->mode,
2105 		.extra1		= &zero,
2106 		.extra2		= &label_limit,
2107 	};
2108 
2109 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2110 
2111 	if (write && ret == 0)
2112 		ret = resize_platform_label_table(net, platform_labels);
2113 
2114 	return ret;
2115 }
2116 
2117 #define MPLS_NS_SYSCTL_OFFSET(field)		\
2118 	(&((struct net *)0)->field)
2119 
2120 static const struct ctl_table mpls_table[] = {
2121 	{
2122 		.procname	= "platform_labels",
2123 		.data		= NULL,
2124 		.maxlen		= sizeof(int),
2125 		.mode		= 0644,
2126 		.proc_handler	= mpls_platform_labels,
2127 	},
2128 	{
2129 		.procname	= "ip_ttl_propagate",
2130 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.ip_ttl_propagate),
2131 		.maxlen		= sizeof(int),
2132 		.mode		= 0644,
2133 		.proc_handler	= proc_dointvec_minmax,
2134 		.extra1		= &zero,
2135 		.extra2		= &one,
2136 	},
2137 	{
2138 		.procname	= "default_ttl",
2139 		.data		= MPLS_NS_SYSCTL_OFFSET(mpls.default_ttl),
2140 		.maxlen		= sizeof(int),
2141 		.mode		= 0644,
2142 		.proc_handler	= proc_dointvec_minmax,
2143 		.extra1		= &one,
2144 		.extra2		= &ttl_max,
2145 	},
2146 	{ }
2147 };
2148 
2149 static int mpls_net_init(struct net *net)
2150 {
2151 	struct ctl_table *table;
2152 	int i;
2153 
2154 	net->mpls.platform_labels = 0;
2155 	net->mpls.platform_label = NULL;
2156 	net->mpls.ip_ttl_propagate = 1;
2157 	net->mpls.default_ttl = 255;
2158 
2159 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
2160 	if (table == NULL)
2161 		return -ENOMEM;
2162 
2163 	/* Table data contains only offsets relative to the base of
2164 	 * the mdev at this point, so make them absolute.
2165 	 */
2166 	for (i = 0; i < ARRAY_SIZE(mpls_table) - 1; i++)
2167 		table[i].data = (char *)net + (uintptr_t)table[i].data;
2168 
2169 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
2170 	if (net->mpls.ctl == NULL) {
2171 		kfree(table);
2172 		return -ENOMEM;
2173 	}
2174 
2175 	return 0;
2176 }
2177 
2178 static void mpls_net_exit(struct net *net)
2179 {
2180 	struct mpls_route __rcu **platform_label;
2181 	size_t platform_labels;
2182 	struct ctl_table *table;
2183 	unsigned int index;
2184 
2185 	table = net->mpls.ctl->ctl_table_arg;
2186 	unregister_net_sysctl_table(net->mpls.ctl);
2187 	kfree(table);
2188 
2189 	/* An rcu grace period has passed since there was a device in
2190 	 * the network namespace (and thus the last in flight packet)
2191 	 * left this network namespace.  This is because
2192 	 * unregister_netdevice_many and netdev_run_todo has completed
2193 	 * for each network device that was in this network namespace.
2194 	 *
2195 	 * As such no additional rcu synchronization is necessary when
2196 	 * freeing the platform_label table.
2197 	 */
2198 	rtnl_lock();
2199 	platform_label = rtnl_dereference(net->mpls.platform_label);
2200 	platform_labels = net->mpls.platform_labels;
2201 	for (index = 0; index < platform_labels; index++) {
2202 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
2203 		RCU_INIT_POINTER(platform_label[index], NULL);
2204 		mpls_notify_route(net, index, rt, NULL, NULL);
2205 		mpls_rt_free(rt);
2206 	}
2207 	rtnl_unlock();
2208 
2209 	kvfree(platform_label);
2210 }
2211 
2212 static struct pernet_operations mpls_net_ops = {
2213 	.init = mpls_net_init,
2214 	.exit = mpls_net_exit,
2215 };
2216 
2217 static struct rtnl_af_ops mpls_af_ops __read_mostly = {
2218 	.family		   = AF_MPLS,
2219 	.fill_stats_af	   = mpls_fill_stats_af,
2220 	.get_stats_af_size = mpls_get_stats_af_size,
2221 };
2222 
2223 static int __init mpls_init(void)
2224 {
2225 	int err;
2226 
2227 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
2228 
2229 	err = register_pernet_subsys(&mpls_net_ops);
2230 	if (err)
2231 		goto out;
2232 
2233 	err = register_netdevice_notifier(&mpls_dev_notifier);
2234 	if (err)
2235 		goto out_unregister_pernet;
2236 
2237 	dev_add_pack(&mpls_packet_type);
2238 
2239 	rtnl_af_register(&mpls_af_ops);
2240 
2241 	rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
2242 	rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
2243 	rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
2244 	rtnl_register(PF_MPLS, RTM_GETNETCONF, mpls_netconf_get_devconf,
2245 		      mpls_netconf_dump_devconf, NULL);
2246 	err = 0;
2247 out:
2248 	return err;
2249 
2250 out_unregister_pernet:
2251 	unregister_pernet_subsys(&mpls_net_ops);
2252 	goto out;
2253 }
2254 module_init(mpls_init);
2255 
2256 static void __exit mpls_exit(void)
2257 {
2258 	rtnl_unregister_all(PF_MPLS);
2259 	rtnl_af_unregister(&mpls_af_ops);
2260 	dev_remove_pack(&mpls_packet_type);
2261 	unregister_netdevice_notifier(&mpls_dev_notifier);
2262 	unregister_pernet_subsys(&mpls_net_ops);
2263 }
2264 module_exit(mpls_exit);
2265 
2266 MODULE_DESCRIPTION("MultiProtocol Label Switching");
2267 MODULE_LICENSE("GPL v2");
2268 MODULE_ALIAS_NETPROTO(PF_MPLS);
2269