xref: /openbmc/linux/net/mpls/af_mpls.c (revision 4f3db074)
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/vmalloc.h>
11 #include <net/ip.h>
12 #include <net/dst.h>
13 #include <net/sock.h>
14 #include <net/arp.h>
15 #include <net/ip_fib.h>
16 #include <net/netevent.h>
17 #include <net/netns/generic.h>
18 #include "internal.h"
19 
20 #define LABEL_NOT_SPECIFIED (1<<20)
21 #define MAX_NEW_LABELS 2
22 
23 /* This maximum ha length copied from the definition of struct neighbour */
24 #define MAX_VIA_ALEN (ALIGN(MAX_ADDR_LEN, sizeof(unsigned long)))
25 
26 struct mpls_route { /* next hop label forwarding entry */
27 	struct net_device __rcu *rt_dev;
28 	struct rcu_head		rt_rcu;
29 	u32			rt_label[MAX_NEW_LABELS];
30 	u8			rt_protocol; /* routing protocol that set this entry */
31 	u8			rt_labels;
32 	u8			rt_via_alen;
33 	u8			rt_via_table;
34 	u8			rt_via[0];
35 };
36 
37 static int zero = 0;
38 static int label_limit = (1 << 20) - 1;
39 
40 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
41 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
42 		       unsigned int nlm_flags);
43 
44 static struct mpls_route *mpls_route_input_rcu(struct net *net, unsigned index)
45 {
46 	struct mpls_route *rt = NULL;
47 
48 	if (index < net->mpls.platform_labels) {
49 		struct mpls_route __rcu **platform_label =
50 			rcu_dereference(net->mpls.platform_label);
51 		rt = rcu_dereference(platform_label[index]);
52 	}
53 	return rt;
54 }
55 
56 static inline struct mpls_dev *mpls_dev_get(const struct net_device *dev)
57 {
58 	return rcu_dereference_rtnl(dev->mpls_ptr);
59 }
60 
61 static bool mpls_output_possible(const struct net_device *dev)
62 {
63 	return dev && (dev->flags & IFF_UP) && netif_carrier_ok(dev);
64 }
65 
66 static unsigned int mpls_rt_header_size(const struct mpls_route *rt)
67 {
68 	/* The size of the layer 2.5 labels to be added for this route */
69 	return rt->rt_labels * sizeof(struct mpls_shim_hdr);
70 }
71 
72 static unsigned int mpls_dev_mtu(const struct net_device *dev)
73 {
74 	/* The amount of data the layer 2 frame can hold */
75 	return dev->mtu;
76 }
77 
78 static bool mpls_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
79 {
80 	if (skb->len <= mtu)
81 		return false;
82 
83 	if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
84 		return false;
85 
86 	return true;
87 }
88 
89 static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
90 			struct mpls_entry_decoded dec)
91 {
92 	/* RFC4385 and RFC5586 encode other packets in mpls such that
93 	 * they don't conflict with the ip version number, making
94 	 * decoding by examining the ip version correct in everything
95 	 * except for the strangest cases.
96 	 *
97 	 * The strange cases if we choose to support them will require
98 	 * manual configuration.
99 	 */
100 	struct iphdr *hdr4;
101 	bool success = true;
102 
103 	/* The IPv4 code below accesses through the IPv4 header
104 	 * checksum, which is 12 bytes into the packet.
105 	 * The IPv6 code below accesses through the IPv6 hop limit
106 	 * which is 8 bytes into the packet.
107 	 *
108 	 * For all supported cases there should always be at least 12
109 	 * bytes of packet data present.  The IPv4 header is 20 bytes
110 	 * without options and the IPv6 header is always 40 bytes
111 	 * long.
112 	 */
113 	if (!pskb_may_pull(skb, 12))
114 		return false;
115 
116 	/* Use ip_hdr to find the ip protocol version */
117 	hdr4 = ip_hdr(skb);
118 	if (hdr4->version == 4) {
119 		skb->protocol = htons(ETH_P_IP);
120 		csum_replace2(&hdr4->check,
121 			      htons(hdr4->ttl << 8),
122 			      htons(dec.ttl << 8));
123 		hdr4->ttl = dec.ttl;
124 	}
125 	else if (hdr4->version == 6) {
126 		struct ipv6hdr *hdr6 = ipv6_hdr(skb);
127 		skb->protocol = htons(ETH_P_IPV6);
128 		hdr6->hop_limit = dec.ttl;
129 	}
130 	else
131 		/* version 0 and version 1 are used by pseudo wires */
132 		success = false;
133 	return success;
134 }
135 
136 static int mpls_forward(struct sk_buff *skb, struct net_device *dev,
137 			struct packet_type *pt, struct net_device *orig_dev)
138 {
139 	struct net *net = dev_net(dev);
140 	struct mpls_shim_hdr *hdr;
141 	struct mpls_route *rt;
142 	struct mpls_entry_decoded dec;
143 	struct net_device *out_dev;
144 	struct mpls_dev *mdev;
145 	unsigned int hh_len;
146 	unsigned int new_header_size;
147 	unsigned int mtu;
148 	int err;
149 
150 	/* Careful this entire function runs inside of an rcu critical section */
151 
152 	mdev = mpls_dev_get(dev);
153 	if (!mdev || !mdev->input_enabled)
154 		goto drop;
155 
156 	if (skb->pkt_type != PACKET_HOST)
157 		goto drop;
158 
159 	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
160 		goto drop;
161 
162 	if (!pskb_may_pull(skb, sizeof(*hdr)))
163 		goto drop;
164 
165 	/* Read and decode the label */
166 	hdr = mpls_hdr(skb);
167 	dec = mpls_entry_decode(hdr);
168 
169 	/* Pop the label */
170 	skb_pull(skb, sizeof(*hdr));
171 	skb_reset_network_header(skb);
172 
173 	skb_orphan(skb);
174 
175 	rt = mpls_route_input_rcu(net, dec.label);
176 	if (!rt)
177 		goto drop;
178 
179 	/* Find the output device */
180 	out_dev = rcu_dereference(rt->rt_dev);
181 	if (!mpls_output_possible(out_dev))
182 		goto drop;
183 
184 	if (skb_warn_if_lro(skb))
185 		goto drop;
186 
187 	skb_forward_csum(skb);
188 
189 	/* Verify ttl is valid */
190 	if (dec.ttl <= 1)
191 		goto drop;
192 	dec.ttl -= 1;
193 
194 	/* Verify the destination can hold the packet */
195 	new_header_size = mpls_rt_header_size(rt);
196 	mtu = mpls_dev_mtu(out_dev);
197 	if (mpls_pkt_too_big(skb, mtu - new_header_size))
198 		goto drop;
199 
200 	hh_len = LL_RESERVED_SPACE(out_dev);
201 	if (!out_dev->header_ops)
202 		hh_len = 0;
203 
204 	/* Ensure there is enough space for the headers in the skb */
205 	if (skb_cow(skb, hh_len + new_header_size))
206 		goto drop;
207 
208 	skb->dev = out_dev;
209 	skb->protocol = htons(ETH_P_MPLS_UC);
210 
211 	if (unlikely(!new_header_size && dec.bos)) {
212 		/* Penultimate hop popping */
213 		if (!mpls_egress(rt, skb, dec))
214 			goto drop;
215 	} else {
216 		bool bos;
217 		int i;
218 		skb_push(skb, new_header_size);
219 		skb_reset_network_header(skb);
220 		/* Push the new labels */
221 		hdr = mpls_hdr(skb);
222 		bos = dec.bos;
223 		for (i = rt->rt_labels - 1; i >= 0; i--) {
224 			hdr[i] = mpls_entry_encode(rt->rt_label[i], dec.ttl, 0, bos);
225 			bos = false;
226 		}
227 	}
228 
229 	err = neigh_xmit(rt->rt_via_table, out_dev, rt->rt_via, skb);
230 	if (err)
231 		net_dbg_ratelimited("%s: packet transmission failed: %d\n",
232 				    __func__, err);
233 	return 0;
234 
235 drop:
236 	kfree_skb(skb);
237 	return NET_RX_DROP;
238 }
239 
240 static struct packet_type mpls_packet_type __read_mostly = {
241 	.type = cpu_to_be16(ETH_P_MPLS_UC),
242 	.func = mpls_forward,
243 };
244 
245 static const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
246 	[RTA_DST]		= { .type = NLA_U32 },
247 	[RTA_OIF]		= { .type = NLA_U32 },
248 };
249 
250 struct mpls_route_config {
251 	u32		rc_protocol;
252 	u32		rc_ifindex;
253 	u16		rc_via_table;
254 	u16		rc_via_alen;
255 	u8		rc_via[MAX_VIA_ALEN];
256 	u32		rc_label;
257 	u32		rc_output_labels;
258 	u32		rc_output_label[MAX_NEW_LABELS];
259 	u32		rc_nlflags;
260 	struct nl_info	rc_nlinfo;
261 };
262 
263 static struct mpls_route *mpls_rt_alloc(size_t alen)
264 {
265 	struct mpls_route *rt;
266 
267 	rt = kzalloc(sizeof(*rt) + alen, GFP_KERNEL);
268 	if (rt)
269 		rt->rt_via_alen = alen;
270 	return rt;
271 }
272 
273 static void mpls_rt_free(struct mpls_route *rt)
274 {
275 	if (rt)
276 		kfree_rcu(rt, rt_rcu);
277 }
278 
279 static void mpls_notify_route(struct net *net, unsigned index,
280 			      struct mpls_route *old, struct mpls_route *new,
281 			      const struct nl_info *info)
282 {
283 	struct nlmsghdr *nlh = info ? info->nlh : NULL;
284 	unsigned portid = info ? info->portid : 0;
285 	int event = new ? RTM_NEWROUTE : RTM_DELROUTE;
286 	struct mpls_route *rt = new ? new : old;
287 	unsigned nlm_flags = (old && new) ? NLM_F_REPLACE : 0;
288 	/* Ignore reserved labels for now */
289 	if (rt && (index >= 16))
290 		rtmsg_lfib(event, index, rt, nlh, net, portid, nlm_flags);
291 }
292 
293 static void mpls_route_update(struct net *net, unsigned index,
294 			      struct net_device *dev, struct mpls_route *new,
295 			      const struct nl_info *info)
296 {
297 	struct mpls_route __rcu **platform_label;
298 	struct mpls_route *rt, *old = NULL;
299 
300 	ASSERT_RTNL();
301 
302 	platform_label = rtnl_dereference(net->mpls.platform_label);
303 	rt = rtnl_dereference(platform_label[index]);
304 	if (!dev || (rt && (rtnl_dereference(rt->rt_dev) == dev))) {
305 		rcu_assign_pointer(platform_label[index], new);
306 		old = rt;
307 	}
308 
309 	mpls_notify_route(net, index, old, new, info);
310 
311 	/* If we removed a route free it now */
312 	mpls_rt_free(old);
313 }
314 
315 static unsigned find_free_label(struct net *net)
316 {
317 	struct mpls_route __rcu **platform_label;
318 	size_t platform_labels;
319 	unsigned index;
320 
321 	platform_label = rtnl_dereference(net->mpls.platform_label);
322 	platform_labels = net->mpls.platform_labels;
323 	for (index = 16; index < platform_labels; index++) {
324 		if (!rtnl_dereference(platform_label[index]))
325 			return index;
326 	}
327 	return LABEL_NOT_SPECIFIED;
328 }
329 
330 static int mpls_route_add(struct mpls_route_config *cfg)
331 {
332 	struct mpls_route __rcu **platform_label;
333 	struct net *net = cfg->rc_nlinfo.nl_net;
334 	struct net_device *dev = NULL;
335 	struct mpls_route *rt, *old;
336 	unsigned index;
337 	int i;
338 	int err = -EINVAL;
339 
340 	index = cfg->rc_label;
341 
342 	/* If a label was not specified during insert pick one */
343 	if ((index == LABEL_NOT_SPECIFIED) &&
344 	    (cfg->rc_nlflags & NLM_F_CREATE)) {
345 		index = find_free_label(net);
346 	}
347 
348 	/* The first 16 labels are reserved, and may not be set */
349 	if (index < 16)
350 		goto errout;
351 
352 	/* The full 20 bit range may not be supported. */
353 	if (index >= net->mpls.platform_labels)
354 		goto errout;
355 
356 	/* Ensure only a supported number of labels are present */
357 	if (cfg->rc_output_labels > MAX_NEW_LABELS)
358 		goto errout;
359 
360 	err = -ENODEV;
361 	dev = dev_get_by_index(net, cfg->rc_ifindex);
362 	if (!dev)
363 		goto errout;
364 
365 	/* Ensure this is a supported device */
366 	err = -EINVAL;
367 	if (!mpls_dev_get(dev))
368 		goto errout;
369 
370 	err = -EINVAL;
371 	if ((cfg->rc_via_table == NEIGH_LINK_TABLE) &&
372 	    (dev->addr_len != cfg->rc_via_alen))
373 		goto errout;
374 
375 	/* Append makes no sense with mpls */
376 	err = -EOPNOTSUPP;
377 	if (cfg->rc_nlflags & NLM_F_APPEND)
378 		goto errout;
379 
380 	err = -EEXIST;
381 	platform_label = rtnl_dereference(net->mpls.platform_label);
382 	old = rtnl_dereference(platform_label[index]);
383 	if ((cfg->rc_nlflags & NLM_F_EXCL) && old)
384 		goto errout;
385 
386 	err = -EEXIST;
387 	if (!(cfg->rc_nlflags & NLM_F_REPLACE) && old)
388 		goto errout;
389 
390 	err = -ENOENT;
391 	if (!(cfg->rc_nlflags & NLM_F_CREATE) && !old)
392 		goto errout;
393 
394 	err = -ENOMEM;
395 	rt = mpls_rt_alloc(cfg->rc_via_alen);
396 	if (!rt)
397 		goto errout;
398 
399 	rt->rt_labels = cfg->rc_output_labels;
400 	for (i = 0; i < rt->rt_labels; i++)
401 		rt->rt_label[i] = cfg->rc_output_label[i];
402 	rt->rt_protocol = cfg->rc_protocol;
403 	RCU_INIT_POINTER(rt->rt_dev, dev);
404 	rt->rt_via_table = cfg->rc_via_table;
405 	memcpy(rt->rt_via, cfg->rc_via, cfg->rc_via_alen);
406 
407 	mpls_route_update(net, index, NULL, rt, &cfg->rc_nlinfo);
408 
409 	dev_put(dev);
410 	return 0;
411 
412 errout:
413 	if (dev)
414 		dev_put(dev);
415 	return err;
416 }
417 
418 static int mpls_route_del(struct mpls_route_config *cfg)
419 {
420 	struct net *net = cfg->rc_nlinfo.nl_net;
421 	unsigned index;
422 	int err = -EINVAL;
423 
424 	index = cfg->rc_label;
425 
426 	/* The first 16 labels are reserved, and may not be removed */
427 	if (index < 16)
428 		goto errout;
429 
430 	/* The full 20 bit range may not be supported */
431 	if (index >= net->mpls.platform_labels)
432 		goto errout;
433 
434 	mpls_route_update(net, index, NULL, NULL, &cfg->rc_nlinfo);
435 
436 	err = 0;
437 errout:
438 	return err;
439 }
440 
441 #define MPLS_PERDEV_SYSCTL_OFFSET(field)	\
442 	(&((struct mpls_dev *)0)->field)
443 
444 static const struct ctl_table mpls_dev_table[] = {
445 	{
446 		.procname	= "input",
447 		.maxlen		= sizeof(int),
448 		.mode		= 0644,
449 		.proc_handler	= proc_dointvec,
450 		.data		= MPLS_PERDEV_SYSCTL_OFFSET(input_enabled),
451 	},
452 	{ }
453 };
454 
455 static int mpls_dev_sysctl_register(struct net_device *dev,
456 				    struct mpls_dev *mdev)
457 {
458 	char path[sizeof("net/mpls/conf/") + IFNAMSIZ];
459 	struct ctl_table *table;
460 	int i;
461 
462 	table = kmemdup(&mpls_dev_table, sizeof(mpls_dev_table), GFP_KERNEL);
463 	if (!table)
464 		goto out;
465 
466 	/* Table data contains only offsets relative to the base of
467 	 * the mdev at this point, so make them absolute.
468 	 */
469 	for (i = 0; i < ARRAY_SIZE(mpls_dev_table); i++)
470 		table[i].data = (char *)mdev + (uintptr_t)table[i].data;
471 
472 	snprintf(path, sizeof(path), "net/mpls/conf/%s", dev->name);
473 
474 	mdev->sysctl = register_net_sysctl(dev_net(dev), path, table);
475 	if (!mdev->sysctl)
476 		goto free;
477 
478 	return 0;
479 
480 free:
481 	kfree(table);
482 out:
483 	return -ENOBUFS;
484 }
485 
486 static void mpls_dev_sysctl_unregister(struct mpls_dev *mdev)
487 {
488 	struct ctl_table *table;
489 
490 	table = mdev->sysctl->ctl_table_arg;
491 	unregister_net_sysctl_table(mdev->sysctl);
492 	kfree(table);
493 }
494 
495 static struct mpls_dev *mpls_add_dev(struct net_device *dev)
496 {
497 	struct mpls_dev *mdev;
498 	int err = -ENOMEM;
499 
500 	ASSERT_RTNL();
501 
502 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
503 	if (!mdev)
504 		return ERR_PTR(err);
505 
506 	err = mpls_dev_sysctl_register(dev, mdev);
507 	if (err)
508 		goto free;
509 
510 	rcu_assign_pointer(dev->mpls_ptr, mdev);
511 
512 	return mdev;
513 
514 free:
515 	kfree(mdev);
516 	return ERR_PTR(err);
517 }
518 
519 static void mpls_ifdown(struct net_device *dev)
520 {
521 	struct mpls_route __rcu **platform_label;
522 	struct net *net = dev_net(dev);
523 	struct mpls_dev *mdev;
524 	unsigned index;
525 
526 	platform_label = rtnl_dereference(net->mpls.platform_label);
527 	for (index = 0; index < net->mpls.platform_labels; index++) {
528 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
529 		if (!rt)
530 			continue;
531 		if (rtnl_dereference(rt->rt_dev) != dev)
532 			continue;
533 		rt->rt_dev = NULL;
534 	}
535 
536 	mdev = mpls_dev_get(dev);
537 	if (!mdev)
538 		return;
539 
540 	mpls_dev_sysctl_unregister(mdev);
541 
542 	RCU_INIT_POINTER(dev->mpls_ptr, NULL);
543 
544 	kfree(mdev);
545 }
546 
547 static int mpls_dev_notify(struct notifier_block *this, unsigned long event,
548 			   void *ptr)
549 {
550 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
551 	struct mpls_dev *mdev;
552 
553 	switch(event) {
554 	case NETDEV_REGISTER:
555 		/* For now just support ethernet devices */
556 		if ((dev->type == ARPHRD_ETHER) ||
557 		    (dev->type == ARPHRD_LOOPBACK)) {
558 			mdev = mpls_add_dev(dev);
559 			if (IS_ERR(mdev))
560 				return notifier_from_errno(PTR_ERR(mdev));
561 		}
562 		break;
563 
564 	case NETDEV_UNREGISTER:
565 		mpls_ifdown(dev);
566 		break;
567 	}
568 	return NOTIFY_OK;
569 }
570 
571 static struct notifier_block mpls_dev_notifier = {
572 	.notifier_call = mpls_dev_notify,
573 };
574 
575 static int nla_put_via(struct sk_buff *skb,
576 		       u8 table, const void *addr, int alen)
577 {
578 	static const int table_to_family[NEIGH_NR_TABLES + 1] = {
579 		AF_INET, AF_INET6, AF_DECnet, AF_PACKET,
580 	};
581 	struct nlattr *nla;
582 	struct rtvia *via;
583 	int family = AF_UNSPEC;
584 
585 	nla = nla_reserve(skb, RTA_VIA, alen + 2);
586 	if (!nla)
587 		return -EMSGSIZE;
588 
589 	if (table <= NEIGH_NR_TABLES)
590 		family = table_to_family[table];
591 
592 	via = nla_data(nla);
593 	via->rtvia_family = family;
594 	memcpy(via->rtvia_addr, addr, alen);
595 	return 0;
596 }
597 
598 int nla_put_labels(struct sk_buff *skb, int attrtype,
599 		   u8 labels, const u32 label[])
600 {
601 	struct nlattr *nla;
602 	struct mpls_shim_hdr *nla_label;
603 	bool bos;
604 	int i;
605 	nla = nla_reserve(skb, attrtype, labels*4);
606 	if (!nla)
607 		return -EMSGSIZE;
608 
609 	nla_label = nla_data(nla);
610 	bos = true;
611 	for (i = labels - 1; i >= 0; i--) {
612 		nla_label[i] = mpls_entry_encode(label[i], 0, 0, bos);
613 		bos = false;
614 	}
615 
616 	return 0;
617 }
618 
619 int nla_get_labels(const struct nlattr *nla,
620 		   u32 max_labels, u32 *labels, u32 label[])
621 {
622 	unsigned len = nla_len(nla);
623 	unsigned nla_labels;
624 	struct mpls_shim_hdr *nla_label;
625 	bool bos;
626 	int i;
627 
628 	/* len needs to be an even multiple of 4 (the label size) */
629 	if (len & 3)
630 		return -EINVAL;
631 
632 	/* Limit the number of new labels allowed */
633 	nla_labels = len/4;
634 	if (nla_labels > max_labels)
635 		return -EINVAL;
636 
637 	nla_label = nla_data(nla);
638 	bos = true;
639 	for (i = nla_labels - 1; i >= 0; i--, bos = false) {
640 		struct mpls_entry_decoded dec;
641 		dec = mpls_entry_decode(nla_label + i);
642 
643 		/* Ensure the bottom of stack flag is properly set
644 		 * and ttl and tc are both clear.
645 		 */
646 		if ((dec.bos != bos) || dec.ttl || dec.tc)
647 			return -EINVAL;
648 
649 		switch (dec.label) {
650 		case LABEL_IMPLICIT_NULL:
651 			/* RFC3032: This is a label that an LSR may
652 			 * assign and distribute, but which never
653 			 * actually appears in the encapsulation.
654 			 */
655 			return -EINVAL;
656 		}
657 
658 		label[i] = dec.label;
659 	}
660 	*labels = nla_labels;
661 	return 0;
662 }
663 
664 static int rtm_to_route_config(struct sk_buff *skb,  struct nlmsghdr *nlh,
665 			       struct mpls_route_config *cfg)
666 {
667 	struct rtmsg *rtm;
668 	struct nlattr *tb[RTA_MAX+1];
669 	int index;
670 	int err;
671 
672 	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
673 	if (err < 0)
674 		goto errout;
675 
676 	err = -EINVAL;
677 	rtm = nlmsg_data(nlh);
678 	memset(cfg, 0, sizeof(*cfg));
679 
680 	if (rtm->rtm_family != AF_MPLS)
681 		goto errout;
682 	if (rtm->rtm_dst_len != 20)
683 		goto errout;
684 	if (rtm->rtm_src_len != 0)
685 		goto errout;
686 	if (rtm->rtm_tos != 0)
687 		goto errout;
688 	if (rtm->rtm_table != RT_TABLE_MAIN)
689 		goto errout;
690 	/* Any value is acceptable for rtm_protocol */
691 
692 	/* As mpls uses destination specific addresses
693 	 * (or source specific address in the case of multicast)
694 	 * all addresses have universal scope.
695 	 */
696 	if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
697 		goto errout;
698 	if (rtm->rtm_type != RTN_UNICAST)
699 		goto errout;
700 	if (rtm->rtm_flags != 0)
701 		goto errout;
702 
703 	cfg->rc_label		= LABEL_NOT_SPECIFIED;
704 	cfg->rc_protocol	= rtm->rtm_protocol;
705 	cfg->rc_nlflags		= nlh->nlmsg_flags;
706 	cfg->rc_nlinfo.portid	= NETLINK_CB(skb).portid;
707 	cfg->rc_nlinfo.nlh	= nlh;
708 	cfg->rc_nlinfo.nl_net	= sock_net(skb->sk);
709 
710 	for (index = 0; index <= RTA_MAX; index++) {
711 		struct nlattr *nla = tb[index];
712 		if (!nla)
713 			continue;
714 
715 		switch(index) {
716 		case RTA_OIF:
717 			cfg->rc_ifindex = nla_get_u32(nla);
718 			break;
719 		case RTA_NEWDST:
720 			if (nla_get_labels(nla, MAX_NEW_LABELS,
721 					   &cfg->rc_output_labels,
722 					   cfg->rc_output_label))
723 				goto errout;
724 			break;
725 		case RTA_DST:
726 		{
727 			u32 label_count;
728 			if (nla_get_labels(nla, 1, &label_count,
729 					   &cfg->rc_label))
730 				goto errout;
731 
732 			/* The first 16 labels are reserved, and may not be set */
733 			if (cfg->rc_label < 16)
734 				goto errout;
735 
736 			break;
737 		}
738 		case RTA_VIA:
739 		{
740 			struct rtvia *via = nla_data(nla);
741 			if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr))
742 				goto errout;
743 			cfg->rc_via_alen   = nla_len(nla) -
744 				offsetof(struct rtvia, rtvia_addr);
745 			if (cfg->rc_via_alen > MAX_VIA_ALEN)
746 				goto errout;
747 
748 			/* Validate the address family */
749 			switch(via->rtvia_family) {
750 			case AF_PACKET:
751 				cfg->rc_via_table = NEIGH_LINK_TABLE;
752 				break;
753 			case AF_INET:
754 				cfg->rc_via_table = NEIGH_ARP_TABLE;
755 				if (cfg->rc_via_alen != 4)
756 					goto errout;
757 				break;
758 			case AF_INET6:
759 				cfg->rc_via_table = NEIGH_ND_TABLE;
760 				if (cfg->rc_via_alen != 16)
761 					goto errout;
762 				break;
763 			default:
764 				/* Unsupported address family */
765 				goto errout;
766 			}
767 
768 			memcpy(cfg->rc_via, via->rtvia_addr, cfg->rc_via_alen);
769 			break;
770 		}
771 		default:
772 			/* Unsupported attribute */
773 			goto errout;
774 		}
775 	}
776 
777 	err = 0;
778 errout:
779 	return err;
780 }
781 
782 static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
783 {
784 	struct mpls_route_config cfg;
785 	int err;
786 
787 	err = rtm_to_route_config(skb, nlh, &cfg);
788 	if (err < 0)
789 		return err;
790 
791 	return mpls_route_del(&cfg);
792 }
793 
794 
795 static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
796 {
797 	struct mpls_route_config cfg;
798 	int err;
799 
800 	err = rtm_to_route_config(skb, nlh, &cfg);
801 	if (err < 0)
802 		return err;
803 
804 	return mpls_route_add(&cfg);
805 }
806 
807 static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
808 			   u32 label, struct mpls_route *rt, int flags)
809 {
810 	struct net_device *dev;
811 	struct nlmsghdr *nlh;
812 	struct rtmsg *rtm;
813 
814 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
815 	if (nlh == NULL)
816 		return -EMSGSIZE;
817 
818 	rtm = nlmsg_data(nlh);
819 	rtm->rtm_family = AF_MPLS;
820 	rtm->rtm_dst_len = 20;
821 	rtm->rtm_src_len = 0;
822 	rtm->rtm_tos = 0;
823 	rtm->rtm_table = RT_TABLE_MAIN;
824 	rtm->rtm_protocol = rt->rt_protocol;
825 	rtm->rtm_scope = RT_SCOPE_UNIVERSE;
826 	rtm->rtm_type = RTN_UNICAST;
827 	rtm->rtm_flags = 0;
828 
829 	if (rt->rt_labels &&
830 	    nla_put_labels(skb, RTA_NEWDST, rt->rt_labels, rt->rt_label))
831 		goto nla_put_failure;
832 	if (nla_put_via(skb, rt->rt_via_table, rt->rt_via, rt->rt_via_alen))
833 		goto nla_put_failure;
834 	dev = rtnl_dereference(rt->rt_dev);
835 	if (dev && nla_put_u32(skb, RTA_OIF, dev->ifindex))
836 		goto nla_put_failure;
837 	if (nla_put_labels(skb, RTA_DST, 1, &label))
838 		goto nla_put_failure;
839 
840 	nlmsg_end(skb, nlh);
841 	return 0;
842 
843 nla_put_failure:
844 	nlmsg_cancel(skb, nlh);
845 	return -EMSGSIZE;
846 }
847 
848 static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
849 {
850 	struct net *net = sock_net(skb->sk);
851 	struct mpls_route __rcu **platform_label;
852 	size_t platform_labels;
853 	unsigned int index;
854 
855 	ASSERT_RTNL();
856 
857 	index = cb->args[0];
858 	if (index < 16)
859 		index = 16;
860 
861 	platform_label = rtnl_dereference(net->mpls.platform_label);
862 	platform_labels = net->mpls.platform_labels;
863 	for (; index < platform_labels; index++) {
864 		struct mpls_route *rt;
865 		rt = rtnl_dereference(platform_label[index]);
866 		if (!rt)
867 			continue;
868 
869 		if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
870 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
871 				    index, rt, NLM_F_MULTI) < 0)
872 			break;
873 	}
874 	cb->args[0] = index;
875 
876 	return skb->len;
877 }
878 
879 static inline size_t lfib_nlmsg_size(struct mpls_route *rt)
880 {
881 	size_t payload =
882 		NLMSG_ALIGN(sizeof(struct rtmsg))
883 		+ nla_total_size(2 + rt->rt_via_alen)	/* RTA_VIA */
884 		+ nla_total_size(4);			/* RTA_DST */
885 	if (rt->rt_labels)				/* RTA_NEWDST */
886 		payload += nla_total_size(rt->rt_labels * 4);
887 	if (rt->rt_dev)					/* RTA_OIF */
888 		payload += nla_total_size(4);
889 	return payload;
890 }
891 
892 static void rtmsg_lfib(int event, u32 label, struct mpls_route *rt,
893 		       struct nlmsghdr *nlh, struct net *net, u32 portid,
894 		       unsigned int nlm_flags)
895 {
896 	struct sk_buff *skb;
897 	u32 seq = nlh ? nlh->nlmsg_seq : 0;
898 	int err = -ENOBUFS;
899 
900 	skb = nlmsg_new(lfib_nlmsg_size(rt), GFP_KERNEL);
901 	if (skb == NULL)
902 		goto errout;
903 
904 	err = mpls_dump_route(skb, portid, seq, event, label, rt, nlm_flags);
905 	if (err < 0) {
906 		/* -EMSGSIZE implies BUG in lfib_nlmsg_size */
907 		WARN_ON(err == -EMSGSIZE);
908 		kfree_skb(skb);
909 		goto errout;
910 	}
911 	rtnl_notify(skb, net, portid, RTNLGRP_MPLS_ROUTE, nlh, GFP_KERNEL);
912 
913 	return;
914 errout:
915 	if (err < 0)
916 		rtnl_set_sk_err(net, RTNLGRP_MPLS_ROUTE, err);
917 }
918 
919 static int resize_platform_label_table(struct net *net, size_t limit)
920 {
921 	size_t size = sizeof(struct mpls_route *) * limit;
922 	size_t old_limit;
923 	size_t cp_size;
924 	struct mpls_route __rcu **labels = NULL, **old;
925 	struct mpls_route *rt0 = NULL, *rt2 = NULL;
926 	unsigned index;
927 
928 	if (size) {
929 		labels = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
930 		if (!labels)
931 			labels = vzalloc(size);
932 
933 		if (!labels)
934 			goto nolabels;
935 	}
936 
937 	/* In case the predefined labels need to be populated */
938 	if (limit > LABEL_IPV4_EXPLICIT_NULL) {
939 		struct net_device *lo = net->loopback_dev;
940 		rt0 = mpls_rt_alloc(lo->addr_len);
941 		if (!rt0)
942 			goto nort0;
943 		RCU_INIT_POINTER(rt0->rt_dev, lo);
944 		rt0->rt_protocol = RTPROT_KERNEL;
945 		rt0->rt_via_table = NEIGH_LINK_TABLE;
946 		memcpy(rt0->rt_via, lo->dev_addr, lo->addr_len);
947 	}
948 	if (limit > LABEL_IPV6_EXPLICIT_NULL) {
949 		struct net_device *lo = net->loopback_dev;
950 		rt2 = mpls_rt_alloc(lo->addr_len);
951 		if (!rt2)
952 			goto nort2;
953 		RCU_INIT_POINTER(rt2->rt_dev, lo);
954 		rt2->rt_protocol = RTPROT_KERNEL;
955 		rt2->rt_via_table = NEIGH_LINK_TABLE;
956 		memcpy(rt2->rt_via, lo->dev_addr, lo->addr_len);
957 	}
958 
959 	rtnl_lock();
960 	/* Remember the original table */
961 	old = rtnl_dereference(net->mpls.platform_label);
962 	old_limit = net->mpls.platform_labels;
963 
964 	/* Free any labels beyond the new table */
965 	for (index = limit; index < old_limit; index++)
966 		mpls_route_update(net, index, NULL, NULL, NULL);
967 
968 	/* Copy over the old labels */
969 	cp_size = size;
970 	if (old_limit < limit)
971 		cp_size = old_limit * sizeof(struct mpls_route *);
972 
973 	memcpy(labels, old, cp_size);
974 
975 	/* If needed set the predefined labels */
976 	if ((old_limit <= LABEL_IPV6_EXPLICIT_NULL) &&
977 	    (limit > LABEL_IPV6_EXPLICIT_NULL)) {
978 		RCU_INIT_POINTER(labels[LABEL_IPV6_EXPLICIT_NULL], rt2);
979 		rt2 = NULL;
980 	}
981 
982 	if ((old_limit <= LABEL_IPV4_EXPLICIT_NULL) &&
983 	    (limit > LABEL_IPV4_EXPLICIT_NULL)) {
984 		RCU_INIT_POINTER(labels[LABEL_IPV4_EXPLICIT_NULL], rt0);
985 		rt0 = NULL;
986 	}
987 
988 	/* Update the global pointers */
989 	net->mpls.platform_labels = limit;
990 	rcu_assign_pointer(net->mpls.platform_label, labels);
991 
992 	rtnl_unlock();
993 
994 	mpls_rt_free(rt2);
995 	mpls_rt_free(rt0);
996 
997 	if (old) {
998 		synchronize_rcu();
999 		kvfree(old);
1000 	}
1001 	return 0;
1002 
1003 nort2:
1004 	mpls_rt_free(rt0);
1005 nort0:
1006 	kvfree(labels);
1007 nolabels:
1008 	return -ENOMEM;
1009 }
1010 
1011 static int mpls_platform_labels(struct ctl_table *table, int write,
1012 				void __user *buffer, size_t *lenp, loff_t *ppos)
1013 {
1014 	struct net *net = table->data;
1015 	int platform_labels = net->mpls.platform_labels;
1016 	int ret;
1017 	struct ctl_table tmp = {
1018 		.procname	= table->procname,
1019 		.data		= &platform_labels,
1020 		.maxlen		= sizeof(int),
1021 		.mode		= table->mode,
1022 		.extra1		= &zero,
1023 		.extra2		= &label_limit,
1024 	};
1025 
1026 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
1027 
1028 	if (write && ret == 0)
1029 		ret = resize_platform_label_table(net, platform_labels);
1030 
1031 	return ret;
1032 }
1033 
1034 static const struct ctl_table mpls_table[] = {
1035 	{
1036 		.procname	= "platform_labels",
1037 		.data		= NULL,
1038 		.maxlen		= sizeof(int),
1039 		.mode		= 0644,
1040 		.proc_handler	= mpls_platform_labels,
1041 	},
1042 	{ }
1043 };
1044 
1045 static int mpls_net_init(struct net *net)
1046 {
1047 	struct ctl_table *table;
1048 
1049 	net->mpls.platform_labels = 0;
1050 	net->mpls.platform_label = NULL;
1051 
1052 	table = kmemdup(mpls_table, sizeof(mpls_table), GFP_KERNEL);
1053 	if (table == NULL)
1054 		return -ENOMEM;
1055 
1056 	table[0].data = net;
1057 	net->mpls.ctl = register_net_sysctl(net, "net/mpls", table);
1058 	if (net->mpls.ctl == NULL)
1059 		return -ENOMEM;
1060 
1061 	return 0;
1062 }
1063 
1064 static void mpls_net_exit(struct net *net)
1065 {
1066 	struct mpls_route __rcu **platform_label;
1067 	size_t platform_labels;
1068 	struct ctl_table *table;
1069 	unsigned int index;
1070 
1071 	table = net->mpls.ctl->ctl_table_arg;
1072 	unregister_net_sysctl_table(net->mpls.ctl);
1073 	kfree(table);
1074 
1075 	/* An rcu grace period has passed since there was a device in
1076 	 * the network namespace (and thus the last in flight packet)
1077 	 * left this network namespace.  This is because
1078 	 * unregister_netdevice_many and netdev_run_todo has completed
1079 	 * for each network device that was in this network namespace.
1080 	 *
1081 	 * As such no additional rcu synchronization is necessary when
1082 	 * freeing the platform_label table.
1083 	 */
1084 	rtnl_lock();
1085 	platform_label = rtnl_dereference(net->mpls.platform_label);
1086 	platform_labels = net->mpls.platform_labels;
1087 	for (index = 0; index < platform_labels; index++) {
1088 		struct mpls_route *rt = rtnl_dereference(platform_label[index]);
1089 		RCU_INIT_POINTER(platform_label[index], NULL);
1090 		mpls_rt_free(rt);
1091 	}
1092 	rtnl_unlock();
1093 
1094 	kvfree(platform_label);
1095 }
1096 
1097 static struct pernet_operations mpls_net_ops = {
1098 	.init = mpls_net_init,
1099 	.exit = mpls_net_exit,
1100 };
1101 
1102 static int __init mpls_init(void)
1103 {
1104 	int err;
1105 
1106 	BUILD_BUG_ON(sizeof(struct mpls_shim_hdr) != 4);
1107 
1108 	err = register_pernet_subsys(&mpls_net_ops);
1109 	if (err)
1110 		goto out;
1111 
1112 	err = register_netdevice_notifier(&mpls_dev_notifier);
1113 	if (err)
1114 		goto out_unregister_pernet;
1115 
1116 	dev_add_pack(&mpls_packet_type);
1117 
1118 	rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
1119 	rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
1120 	rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
1121 	err = 0;
1122 out:
1123 	return err;
1124 
1125 out_unregister_pernet:
1126 	unregister_pernet_subsys(&mpls_net_ops);
1127 	goto out;
1128 }
1129 module_init(mpls_init);
1130 
1131 static void __exit mpls_exit(void)
1132 {
1133 	rtnl_unregister_all(PF_MPLS);
1134 	dev_remove_pack(&mpls_packet_type);
1135 	unregister_netdevice_notifier(&mpls_dev_notifier);
1136 	unregister_pernet_subsys(&mpls_net_ops);
1137 }
1138 module_exit(mpls_exit);
1139 
1140 MODULE_DESCRIPTION("MultiProtocol Label Switching");
1141 MODULE_LICENSE("GPL v2");
1142 MODULE_ALIAS_NETPROTO(PF_MPLS);
1143