xref: /linux/net/ipv6/ip6_tunnel.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
1 /*
2  *	IPv6 tunneling device
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Ville Nuorvala		<vnuorval@tcs.hut.fi>
7  *	Yasuyuki Kozakai	<kozakai@linux-ipv6.org>
8  *
9  *      Based on:
10  *      linux/net/ipv6/sit.c and linux/net/ipv4/ipip.c
11  *
12  *      RFC 2473
13  *
14  *	This program is free software; you can redistribute it and/or
15  *      modify it under the terms of the GNU General Public License
16  *      as published by the Free Software Foundation; either version
17  *      2 of the License, or (at your option) any later version.
18  *
19  */
20 
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 
23 #include <linux/module.h>
24 #include <linux/capability.h>
25 #include <linux/errno.h>
26 #include <linux/types.h>
27 #include <linux/sockios.h>
28 #include <linux/icmp.h>
29 #include <linux/if.h>
30 #include <linux/in.h>
31 #include <linux/ip.h>
32 #include <linux/if_tunnel.h>
33 #include <linux/net.h>
34 #include <linux/in6.h>
35 #include <linux/netdevice.h>
36 #include <linux/if_arp.h>
37 #include <linux/icmpv6.h>
38 #include <linux/init.h>
39 #include <linux/route.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/netfilter_ipv6.h>
42 #include <linux/slab.h>
43 
44 #include <asm/uaccess.h>
45 #include <linux/atomic.h>
46 
47 #include <net/icmp.h>
48 #include <net/ip.h>
49 #include <net/ipv6.h>
50 #include <net/ip6_route.h>
51 #include <net/addrconf.h>
52 #include <net/ip6_tunnel.h>
53 #include <net/xfrm.h>
54 #include <net/dsfield.h>
55 #include <net/inet_ecn.h>
56 #include <net/net_namespace.h>
57 #include <net/netns/generic.h>
58 
59 MODULE_AUTHOR("Ville Nuorvala");
60 MODULE_DESCRIPTION("IPv6 tunneling device");
61 MODULE_LICENSE("GPL");
62 MODULE_ALIAS_NETDEV("ip6tnl0");
63 
64 #ifdef IP6_TNL_DEBUG
65 #define IP6_TNL_TRACE(x...) pr_debug("%s:" x "\n", __func__)
66 #else
67 #define IP6_TNL_TRACE(x...) do {;} while(0)
68 #endif
69 
70 #define IPV6_TCLASS_MASK (IPV6_FLOWINFO_MASK & ~IPV6_FLOWLABEL_MASK)
71 #define IPV6_TCLASS_SHIFT 20
72 
73 #define HASH_SIZE  32
74 
75 #define HASH(addr) ((__force u32)((addr)->s6_addr32[0] ^ (addr)->s6_addr32[1] ^ \
76 		     (addr)->s6_addr32[2] ^ (addr)->s6_addr32[3]) & \
77 		    (HASH_SIZE - 1))
78 
79 static int ip6_tnl_dev_init(struct net_device *dev);
80 static void ip6_tnl_dev_setup(struct net_device *dev);
81 
82 static int ip6_tnl_net_id __read_mostly;
83 struct ip6_tnl_net {
84 	/* the IPv6 tunnel fallback device */
85 	struct net_device *fb_tnl_dev;
86 	/* lists for storing tunnels in use */
87 	struct ip6_tnl __rcu *tnls_r_l[HASH_SIZE];
88 	struct ip6_tnl __rcu *tnls_wc[1];
89 	struct ip6_tnl __rcu **tnls[2];
90 };
91 
92 /* often modified stats are per cpu, other are shared (netdev->stats) */
93 struct pcpu_tstats {
94 	unsigned long	rx_packets;
95 	unsigned long	rx_bytes;
96 	unsigned long	tx_packets;
97 	unsigned long	tx_bytes;
98 } __attribute__((aligned(4*sizeof(unsigned long))));
99 
100 static struct net_device_stats *ip6_get_stats(struct net_device *dev)
101 {
102 	struct pcpu_tstats sum = { 0 };
103 	int i;
104 
105 	for_each_possible_cpu(i) {
106 		const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
107 
108 		sum.rx_packets += tstats->rx_packets;
109 		sum.rx_bytes   += tstats->rx_bytes;
110 		sum.tx_packets += tstats->tx_packets;
111 		sum.tx_bytes   += tstats->tx_bytes;
112 	}
113 	dev->stats.rx_packets = sum.rx_packets;
114 	dev->stats.rx_bytes   = sum.rx_bytes;
115 	dev->stats.tx_packets = sum.tx_packets;
116 	dev->stats.tx_bytes   = sum.tx_bytes;
117 	return &dev->stats;
118 }
119 
120 /*
121  * Locking : hash tables are protected by RCU and RTNL
122  */
123 
124 static inline struct dst_entry *ip6_tnl_dst_check(struct ip6_tnl *t)
125 {
126 	struct dst_entry *dst = t->dst_cache;
127 
128 	if (dst && dst->obsolete &&
129 	    dst->ops->check(dst, t->dst_cookie) == NULL) {
130 		t->dst_cache = NULL;
131 		dst_release(dst);
132 		return NULL;
133 	}
134 
135 	return dst;
136 }
137 
138 static inline void ip6_tnl_dst_reset(struct ip6_tnl *t)
139 {
140 	dst_release(t->dst_cache);
141 	t->dst_cache = NULL;
142 }
143 
144 static inline void ip6_tnl_dst_store(struct ip6_tnl *t, struct dst_entry *dst)
145 {
146 	struct rt6_info *rt = (struct rt6_info *) dst;
147 	t->dst_cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
148 	dst_release(t->dst_cache);
149 	t->dst_cache = dst;
150 }
151 
152 /**
153  * ip6_tnl_lookup - fetch tunnel matching the end-point addresses
154  *   @remote: the address of the tunnel exit-point
155  *   @local: the address of the tunnel entry-point
156  *
157  * Return:
158  *   tunnel matching given end-points if found,
159  *   else fallback tunnel if its device is up,
160  *   else %NULL
161  **/
162 
163 #define for_each_ip6_tunnel_rcu(start) \
164 	for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
165 
166 static struct ip6_tnl *
167 ip6_tnl_lookup(struct net *net, const struct in6_addr *remote, const struct in6_addr *local)
168 {
169 	unsigned int h0 = HASH(remote);
170 	unsigned int h1 = HASH(local);
171 	struct ip6_tnl *t;
172 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
173 
174 	for_each_ip6_tunnel_rcu(ip6n->tnls_r_l[h0 ^ h1]) {
175 		if (ipv6_addr_equal(local, &t->parms.laddr) &&
176 		    ipv6_addr_equal(remote, &t->parms.raddr) &&
177 		    (t->dev->flags & IFF_UP))
178 			return t;
179 	}
180 	t = rcu_dereference(ip6n->tnls_wc[0]);
181 	if (t && (t->dev->flags & IFF_UP))
182 		return t;
183 
184 	return NULL;
185 }
186 
187 /**
188  * ip6_tnl_bucket - get head of list matching given tunnel parameters
189  *   @p: parameters containing tunnel end-points
190  *
191  * Description:
192  *   ip6_tnl_bucket() returns the head of the list matching the
193  *   &struct in6_addr entries laddr and raddr in @p.
194  *
195  * Return: head of IPv6 tunnel list
196  **/
197 
198 static struct ip6_tnl __rcu **
199 ip6_tnl_bucket(struct ip6_tnl_net *ip6n, const struct ip6_tnl_parm *p)
200 {
201 	const struct in6_addr *remote = &p->raddr;
202 	const struct in6_addr *local = &p->laddr;
203 	unsigned int h = 0;
204 	int prio = 0;
205 
206 	if (!ipv6_addr_any(remote) || !ipv6_addr_any(local)) {
207 		prio = 1;
208 		h = HASH(remote) ^ HASH(local);
209 	}
210 	return &ip6n->tnls[prio][h];
211 }
212 
213 /**
214  * ip6_tnl_link - add tunnel to hash table
215  *   @t: tunnel to be added
216  **/
217 
218 static void
219 ip6_tnl_link(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
220 {
221 	struct ip6_tnl __rcu **tp = ip6_tnl_bucket(ip6n, &t->parms);
222 
223 	rcu_assign_pointer(t->next , rtnl_dereference(*tp));
224 	rcu_assign_pointer(*tp, t);
225 }
226 
227 /**
228  * ip6_tnl_unlink - remove tunnel from hash table
229  *   @t: tunnel to be removed
230  **/
231 
232 static void
233 ip6_tnl_unlink(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
234 {
235 	struct ip6_tnl __rcu **tp;
236 	struct ip6_tnl *iter;
237 
238 	for (tp = ip6_tnl_bucket(ip6n, &t->parms);
239 	     (iter = rtnl_dereference(*tp)) != NULL;
240 	     tp = &iter->next) {
241 		if (t == iter) {
242 			rcu_assign_pointer(*tp, t->next);
243 			break;
244 		}
245 	}
246 }
247 
248 static void ip6_dev_free(struct net_device *dev)
249 {
250 	free_percpu(dev->tstats);
251 	free_netdev(dev);
252 }
253 
254 /**
255  * ip6_tnl_create() - create a new tunnel
256  *   @p: tunnel parameters
257  *   @pt: pointer to new tunnel
258  *
259  * Description:
260  *   Create tunnel matching given parameters.
261  *
262  * Return:
263  *   created tunnel or NULL
264  **/
265 
266 static struct ip6_tnl *ip6_tnl_create(struct net *net, struct ip6_tnl_parm *p)
267 {
268 	struct net_device *dev;
269 	struct ip6_tnl *t;
270 	char name[IFNAMSIZ];
271 	int err;
272 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
273 
274 	if (p->name[0])
275 		strlcpy(name, p->name, IFNAMSIZ);
276 	else
277 		sprintf(name, "ip6tnl%%d");
278 
279 	dev = alloc_netdev(sizeof (*t), name, ip6_tnl_dev_setup);
280 	if (dev == NULL)
281 		goto failed;
282 
283 	dev_net_set(dev, net);
284 
285 	t = netdev_priv(dev);
286 	t->parms = *p;
287 	err = ip6_tnl_dev_init(dev);
288 	if (err < 0)
289 		goto failed_free;
290 
291 	if ((err = register_netdevice(dev)) < 0)
292 		goto failed_free;
293 
294 	strcpy(t->parms.name, dev->name);
295 
296 	dev_hold(dev);
297 	ip6_tnl_link(ip6n, t);
298 	return t;
299 
300 failed_free:
301 	ip6_dev_free(dev);
302 failed:
303 	return NULL;
304 }
305 
306 /**
307  * ip6_tnl_locate - find or create tunnel matching given parameters
308  *   @p: tunnel parameters
309  *   @create: != 0 if allowed to create new tunnel if no match found
310  *
311  * Description:
312  *   ip6_tnl_locate() first tries to locate an existing tunnel
313  *   based on @parms. If this is unsuccessful, but @create is set a new
314  *   tunnel device is created and registered for use.
315  *
316  * Return:
317  *   matching tunnel or NULL
318  **/
319 
320 static struct ip6_tnl *ip6_tnl_locate(struct net *net,
321 		struct ip6_tnl_parm *p, int create)
322 {
323 	const struct in6_addr *remote = &p->raddr;
324 	const struct in6_addr *local = &p->laddr;
325 	struct ip6_tnl __rcu **tp;
326 	struct ip6_tnl *t;
327 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
328 
329 	for (tp = ip6_tnl_bucket(ip6n, p);
330 	     (t = rtnl_dereference(*tp)) != NULL;
331 	     tp = &t->next) {
332 		if (ipv6_addr_equal(local, &t->parms.laddr) &&
333 		    ipv6_addr_equal(remote, &t->parms.raddr))
334 			return t;
335 	}
336 	if (!create)
337 		return NULL;
338 	return ip6_tnl_create(net, p);
339 }
340 
341 /**
342  * ip6_tnl_dev_uninit - tunnel device uninitializer
343  *   @dev: the device to be destroyed
344  *
345  * Description:
346  *   ip6_tnl_dev_uninit() removes tunnel from its list
347  **/
348 
349 static void
350 ip6_tnl_dev_uninit(struct net_device *dev)
351 {
352 	struct ip6_tnl *t = netdev_priv(dev);
353 	struct net *net = dev_net(dev);
354 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
355 
356 	if (dev == ip6n->fb_tnl_dev)
357 		RCU_INIT_POINTER(ip6n->tnls_wc[0], NULL);
358 	else
359 		ip6_tnl_unlink(ip6n, t);
360 	ip6_tnl_dst_reset(t);
361 	dev_put(dev);
362 }
363 
364 /**
365  * parse_tvl_tnl_enc_lim - handle encapsulation limit option
366  *   @skb: received socket buffer
367  *
368  * Return:
369  *   0 if none was found,
370  *   else index to encapsulation limit
371  **/
372 
373 static __u16
374 parse_tlv_tnl_enc_lim(struct sk_buff *skb, __u8 * raw)
375 {
376 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *) raw;
377 	__u8 nexthdr = ipv6h->nexthdr;
378 	__u16 off = sizeof (*ipv6h);
379 
380 	while (ipv6_ext_hdr(nexthdr) && nexthdr != NEXTHDR_NONE) {
381 		__u16 optlen = 0;
382 		struct ipv6_opt_hdr *hdr;
383 		if (raw + off + sizeof (*hdr) > skb->data &&
384 		    !pskb_may_pull(skb, raw - skb->data + off + sizeof (*hdr)))
385 			break;
386 
387 		hdr = (struct ipv6_opt_hdr *) (raw + off);
388 		if (nexthdr == NEXTHDR_FRAGMENT) {
389 			struct frag_hdr *frag_hdr = (struct frag_hdr *) hdr;
390 			if (frag_hdr->frag_off)
391 				break;
392 			optlen = 8;
393 		} else if (nexthdr == NEXTHDR_AUTH) {
394 			optlen = (hdr->hdrlen + 2) << 2;
395 		} else {
396 			optlen = ipv6_optlen(hdr);
397 		}
398 		if (nexthdr == NEXTHDR_DEST) {
399 			__u16 i = off + 2;
400 			while (1) {
401 				struct ipv6_tlv_tnl_enc_lim *tel;
402 
403 				/* No more room for encapsulation limit */
404 				if (i + sizeof (*tel) > off + optlen)
405 					break;
406 
407 				tel = (struct ipv6_tlv_tnl_enc_lim *) &raw[i];
408 				/* return index of option if found and valid */
409 				if (tel->type == IPV6_TLV_TNL_ENCAP_LIMIT &&
410 				    tel->length == 1)
411 					return i;
412 				/* else jump to next option */
413 				if (tel->type)
414 					i += tel->length + 2;
415 				else
416 					i++;
417 			}
418 		}
419 		nexthdr = hdr->nexthdr;
420 		off += optlen;
421 	}
422 	return 0;
423 }
424 
425 /**
426  * ip6_tnl_err - tunnel error handler
427  *
428  * Description:
429  *   ip6_tnl_err() should handle errors in the tunnel according
430  *   to the specifications in RFC 2473.
431  **/
432 
433 static int
434 ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
435 	    u8 *type, u8 *code, int *msg, __u32 *info, int offset)
436 {
437 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *) skb->data;
438 	struct ip6_tnl *t;
439 	int rel_msg = 0;
440 	u8 rel_type = ICMPV6_DEST_UNREACH;
441 	u8 rel_code = ICMPV6_ADDR_UNREACH;
442 	__u32 rel_info = 0;
443 	__u16 len;
444 	int err = -ENOENT;
445 
446 	/* If the packet doesn't contain the original IPv6 header we are
447 	   in trouble since we might need the source address for further
448 	   processing of the error. */
449 
450 	rcu_read_lock();
451 	if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->daddr,
452 					&ipv6h->saddr)) == NULL)
453 		goto out;
454 
455 	if (t->parms.proto != ipproto && t->parms.proto != 0)
456 		goto out;
457 
458 	err = 0;
459 
460 	switch (*type) {
461 		__u32 teli;
462 		struct ipv6_tlv_tnl_enc_lim *tel;
463 		__u32 mtu;
464 	case ICMPV6_DEST_UNREACH:
465 		net_warn_ratelimited("%s: Path to destination invalid or inactive!\n",
466 				     t->parms.name);
467 		rel_msg = 1;
468 		break;
469 	case ICMPV6_TIME_EXCEED:
470 		if ((*code) == ICMPV6_EXC_HOPLIMIT) {
471 			net_warn_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
472 					     t->parms.name);
473 			rel_msg = 1;
474 		}
475 		break;
476 	case ICMPV6_PARAMPROB:
477 		teli = 0;
478 		if ((*code) == ICMPV6_HDR_FIELD)
479 			teli = parse_tlv_tnl_enc_lim(skb, skb->data);
480 
481 		if (teli && teli == *info - 2) {
482 			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
483 			if (tel->encap_limit == 0) {
484 				net_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
485 						     t->parms.name);
486 				rel_msg = 1;
487 			}
488 		} else {
489 			net_warn_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
490 					     t->parms.name);
491 		}
492 		break;
493 	case ICMPV6_PKT_TOOBIG:
494 		mtu = *info - offset;
495 		if (mtu < IPV6_MIN_MTU)
496 			mtu = IPV6_MIN_MTU;
497 		t->dev->mtu = mtu;
498 
499 		if ((len = sizeof (*ipv6h) + ntohs(ipv6h->payload_len)) > mtu) {
500 			rel_type = ICMPV6_PKT_TOOBIG;
501 			rel_code = 0;
502 			rel_info = mtu;
503 			rel_msg = 1;
504 		}
505 		break;
506 	}
507 
508 	*type = rel_type;
509 	*code = rel_code;
510 	*info = rel_info;
511 	*msg = rel_msg;
512 
513 out:
514 	rcu_read_unlock();
515 	return err;
516 }
517 
518 static int
519 ip4ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
520 	   u8 type, u8 code, int offset, __be32 info)
521 {
522 	int rel_msg = 0;
523 	u8 rel_type = type;
524 	u8 rel_code = code;
525 	__u32 rel_info = ntohl(info);
526 	int err;
527 	struct sk_buff *skb2;
528 	const struct iphdr *eiph;
529 	struct rtable *rt;
530 	struct flowi4 fl4;
531 
532 	err = ip6_tnl_err(skb, IPPROTO_IPIP, opt, &rel_type, &rel_code,
533 			  &rel_msg, &rel_info, offset);
534 	if (err < 0)
535 		return err;
536 
537 	if (rel_msg == 0)
538 		return 0;
539 
540 	switch (rel_type) {
541 	case ICMPV6_DEST_UNREACH:
542 		if (rel_code != ICMPV6_ADDR_UNREACH)
543 			return 0;
544 		rel_type = ICMP_DEST_UNREACH;
545 		rel_code = ICMP_HOST_UNREACH;
546 		break;
547 	case ICMPV6_PKT_TOOBIG:
548 		if (rel_code != 0)
549 			return 0;
550 		rel_type = ICMP_DEST_UNREACH;
551 		rel_code = ICMP_FRAG_NEEDED;
552 		break;
553 	default:
554 		return 0;
555 	}
556 
557 	if (!pskb_may_pull(skb, offset + sizeof(struct iphdr)))
558 		return 0;
559 
560 	skb2 = skb_clone(skb, GFP_ATOMIC);
561 	if (!skb2)
562 		return 0;
563 
564 	skb_dst_drop(skb2);
565 
566 	skb_pull(skb2, offset);
567 	skb_reset_network_header(skb2);
568 	eiph = ip_hdr(skb2);
569 
570 	/* Try to guess incoming interface */
571 	rt = ip_route_output_ports(dev_net(skb->dev), &fl4, NULL,
572 				   eiph->saddr, 0,
573 				   0, 0,
574 				   IPPROTO_IPIP, RT_TOS(eiph->tos), 0);
575 	if (IS_ERR(rt))
576 		goto out;
577 
578 	skb2->dev = rt->dst.dev;
579 
580 	/* route "incoming" packet */
581 	if (rt->rt_flags & RTCF_LOCAL) {
582 		ip_rt_put(rt);
583 		rt = NULL;
584 		rt = ip_route_output_ports(dev_net(skb->dev), &fl4, NULL,
585 					   eiph->daddr, eiph->saddr,
586 					   0, 0,
587 					   IPPROTO_IPIP,
588 					   RT_TOS(eiph->tos), 0);
589 		if (IS_ERR(rt) ||
590 		    rt->dst.dev->type != ARPHRD_TUNNEL) {
591 			if (!IS_ERR(rt))
592 				ip_rt_put(rt);
593 			goto out;
594 		}
595 		skb_dst_set(skb2, &rt->dst);
596 	} else {
597 		ip_rt_put(rt);
598 		if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos,
599 				   skb2->dev) ||
600 		    skb_dst(skb2)->dev->type != ARPHRD_TUNNEL)
601 			goto out;
602 	}
603 
604 	/* change mtu on this route */
605 	if (rel_type == ICMP_DEST_UNREACH && rel_code == ICMP_FRAG_NEEDED) {
606 		if (rel_info > dst_mtu(skb_dst(skb2)))
607 			goto out;
608 
609 		skb_dst(skb2)->ops->update_pmtu(skb_dst(skb2), rel_info);
610 	}
611 
612 	icmp_send(skb2, rel_type, rel_code, htonl(rel_info));
613 
614 out:
615 	kfree_skb(skb2);
616 	return 0;
617 }
618 
619 static int
620 ip6ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
621 	   u8 type, u8 code, int offset, __be32 info)
622 {
623 	int rel_msg = 0;
624 	u8 rel_type = type;
625 	u8 rel_code = code;
626 	__u32 rel_info = ntohl(info);
627 	int err;
628 
629 	err = ip6_tnl_err(skb, IPPROTO_IPV6, opt, &rel_type, &rel_code,
630 			  &rel_msg, &rel_info, offset);
631 	if (err < 0)
632 		return err;
633 
634 	if (rel_msg && pskb_may_pull(skb, offset + sizeof(struct ipv6hdr))) {
635 		struct rt6_info *rt;
636 		struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
637 
638 		if (!skb2)
639 			return 0;
640 
641 		skb_dst_drop(skb2);
642 		skb_pull(skb2, offset);
643 		skb_reset_network_header(skb2);
644 
645 		/* Try to guess incoming interface */
646 		rt = rt6_lookup(dev_net(skb->dev), &ipv6_hdr(skb2)->saddr,
647 				NULL, 0, 0);
648 
649 		if (rt && rt->dst.dev)
650 			skb2->dev = rt->dst.dev;
651 
652 		icmpv6_send(skb2, rel_type, rel_code, rel_info);
653 
654 		if (rt)
655 			dst_release(&rt->dst);
656 
657 		kfree_skb(skb2);
658 	}
659 
660 	return 0;
661 }
662 
663 static void ip4ip6_dscp_ecn_decapsulate(const struct ip6_tnl *t,
664 					const struct ipv6hdr *ipv6h,
665 					struct sk_buff *skb)
666 {
667 	__u8 dsfield = ipv6_get_dsfield(ipv6h) & ~INET_ECN_MASK;
668 
669 	if (t->parms.flags & IP6_TNL_F_RCV_DSCP_COPY)
670 		ipv4_change_dsfield(ip_hdr(skb), INET_ECN_MASK, dsfield);
671 
672 	if (INET_ECN_is_ce(dsfield))
673 		IP_ECN_set_ce(ip_hdr(skb));
674 }
675 
676 static void ip6ip6_dscp_ecn_decapsulate(const struct ip6_tnl *t,
677 					const struct ipv6hdr *ipv6h,
678 					struct sk_buff *skb)
679 {
680 	if (t->parms.flags & IP6_TNL_F_RCV_DSCP_COPY)
681 		ipv6_copy_dscp(ipv6_get_dsfield(ipv6h), ipv6_hdr(skb));
682 
683 	if (INET_ECN_is_ce(ipv6_get_dsfield(ipv6h)))
684 		IP6_ECN_set_ce(ipv6_hdr(skb));
685 }
686 
687 /* called with rcu_read_lock() */
688 static inline int ip6_tnl_rcv_ctl(struct ip6_tnl *t)
689 {
690 	struct ip6_tnl_parm *p = &t->parms;
691 	int ret = 0;
692 	struct net *net = dev_net(t->dev);
693 
694 	if (p->flags & IP6_TNL_F_CAP_RCV) {
695 		struct net_device *ldev = NULL;
696 
697 		if (p->link)
698 			ldev = dev_get_by_index_rcu(net, p->link);
699 
700 		if ((ipv6_addr_is_multicast(&p->laddr) ||
701 		     likely(ipv6_chk_addr(net, &p->laddr, ldev, 0))) &&
702 		    likely(!ipv6_chk_addr(net, &p->raddr, NULL, 0)))
703 			ret = 1;
704 
705 	}
706 	return ret;
707 }
708 
709 /**
710  * ip6_tnl_rcv - decapsulate IPv6 packet and retransmit it locally
711  *   @skb: received socket buffer
712  *   @protocol: ethernet protocol ID
713  *   @dscp_ecn_decapsulate: the function to decapsulate DSCP code and ECN
714  *
715  * Return: 0
716  **/
717 
718 static int ip6_tnl_rcv(struct sk_buff *skb, __u16 protocol,
719 		       __u8 ipproto,
720 		       void (*dscp_ecn_decapsulate)(const struct ip6_tnl *t,
721 						    const struct ipv6hdr *ipv6h,
722 						    struct sk_buff *skb))
723 {
724 	struct ip6_tnl *t;
725 	const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
726 
727 	rcu_read_lock();
728 
729 	if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->saddr,
730 					&ipv6h->daddr)) != NULL) {
731 		struct pcpu_tstats *tstats;
732 
733 		if (t->parms.proto != ipproto && t->parms.proto != 0) {
734 			rcu_read_unlock();
735 			goto discard;
736 		}
737 
738 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
739 			rcu_read_unlock();
740 			goto discard;
741 		}
742 
743 		if (!ip6_tnl_rcv_ctl(t)) {
744 			t->dev->stats.rx_dropped++;
745 			rcu_read_unlock();
746 			goto discard;
747 		}
748 		secpath_reset(skb);
749 		skb->mac_header = skb->network_header;
750 		skb_reset_network_header(skb);
751 		skb->protocol = htons(protocol);
752 		skb->pkt_type = PACKET_HOST;
753 		memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
754 
755 		tstats = this_cpu_ptr(t->dev->tstats);
756 		tstats->rx_packets++;
757 		tstats->rx_bytes += skb->len;
758 
759 		__skb_tunnel_rx(skb, t->dev);
760 
761 		dscp_ecn_decapsulate(t, ipv6h, skb);
762 
763 		netif_rx(skb);
764 
765 		rcu_read_unlock();
766 		return 0;
767 	}
768 	rcu_read_unlock();
769 	return 1;
770 
771 discard:
772 	kfree_skb(skb);
773 	return 0;
774 }
775 
776 static int ip4ip6_rcv(struct sk_buff *skb)
777 {
778 	return ip6_tnl_rcv(skb, ETH_P_IP, IPPROTO_IPIP,
779 			   ip4ip6_dscp_ecn_decapsulate);
780 }
781 
782 static int ip6ip6_rcv(struct sk_buff *skb)
783 {
784 	return ip6_tnl_rcv(skb, ETH_P_IPV6, IPPROTO_IPV6,
785 			   ip6ip6_dscp_ecn_decapsulate);
786 }
787 
788 struct ipv6_tel_txoption {
789 	struct ipv6_txoptions ops;
790 	__u8 dst_opt[8];
791 };
792 
793 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
794 {
795 	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
796 
797 	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
798 	opt->dst_opt[3] = 1;
799 	opt->dst_opt[4] = encap_limit;
800 	opt->dst_opt[5] = IPV6_TLV_PADN;
801 	opt->dst_opt[6] = 1;
802 
803 	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
804 	opt->ops.opt_nflen = 8;
805 }
806 
807 /**
808  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
809  *   @t: the outgoing tunnel device
810  *   @hdr: IPv6 header from the incoming packet
811  *
812  * Description:
813  *   Avoid trivial tunneling loop by checking that tunnel exit-point
814  *   doesn't match source of incoming packet.
815  *
816  * Return:
817  *   1 if conflict,
818  *   0 else
819  **/
820 
821 static inline bool
822 ip6_tnl_addr_conflict(const struct ip6_tnl *t, const struct ipv6hdr *hdr)
823 {
824 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
825 }
826 
827 static inline int ip6_tnl_xmit_ctl(struct ip6_tnl *t)
828 {
829 	struct ip6_tnl_parm *p = &t->parms;
830 	int ret = 0;
831 	struct net *net = dev_net(t->dev);
832 
833 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
834 		struct net_device *ldev = NULL;
835 
836 		rcu_read_lock();
837 		if (p->link)
838 			ldev = dev_get_by_index_rcu(net, p->link);
839 
840 		if (unlikely(!ipv6_chk_addr(net, &p->laddr, ldev, 0)))
841 			pr_warn("%s xmit: Local address not yet configured!\n",
842 				p->name);
843 		else if (!ipv6_addr_is_multicast(&p->raddr) &&
844 			 unlikely(ipv6_chk_addr(net, &p->raddr, NULL, 0)))
845 			pr_warn("%s xmit: Routing loop! Remote address found on this node!\n",
846 				p->name);
847 		else
848 			ret = 1;
849 		rcu_read_unlock();
850 	}
851 	return ret;
852 }
853 /**
854  * ip6_tnl_xmit2 - encapsulate packet and send
855  *   @skb: the outgoing socket buffer
856  *   @dev: the outgoing tunnel device
857  *   @dsfield: dscp code for outer header
858  *   @fl: flow of tunneled packet
859  *   @encap_limit: encapsulation limit
860  *   @pmtu: Path MTU is stored if packet is too big
861  *
862  * Description:
863  *   Build new header and do some sanity checks on the packet before sending
864  *   it.
865  *
866  * Return:
867  *   0 on success
868  *   -1 fail
869  *   %-EMSGSIZE message too big. return mtu in this case.
870  **/
871 
872 static int ip6_tnl_xmit2(struct sk_buff *skb,
873 			 struct net_device *dev,
874 			 __u8 dsfield,
875 			 struct flowi6 *fl6,
876 			 int encap_limit,
877 			 __u32 *pmtu)
878 {
879 	struct net *net = dev_net(dev);
880 	struct ip6_tnl *t = netdev_priv(dev);
881 	struct net_device_stats *stats = &t->dev->stats;
882 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
883 	struct ipv6_tel_txoption opt;
884 	struct dst_entry *dst = NULL, *ndst = NULL;
885 	struct net_device *tdev;
886 	int mtu;
887 	unsigned int max_headroom = sizeof(struct ipv6hdr);
888 	u8 proto;
889 	int err = -1;
890 	int pkt_len;
891 
892 	if (!fl6->flowi6_mark)
893 		dst = ip6_tnl_dst_check(t);
894 	if (!dst) {
895 		ndst = ip6_route_output(net, NULL, fl6);
896 
897 		if (ndst->error)
898 			goto tx_err_link_failure;
899 		ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
900 		if (IS_ERR(ndst)) {
901 			err = PTR_ERR(ndst);
902 			ndst = NULL;
903 			goto tx_err_link_failure;
904 		}
905 		dst = ndst;
906 	}
907 
908 	tdev = dst->dev;
909 
910 	if (tdev == dev) {
911 		stats->collisions++;
912 		net_warn_ratelimited("%s: Local routing loop detected!\n",
913 				     t->parms.name);
914 		goto tx_err_dst_release;
915 	}
916 	mtu = dst_mtu(dst) - sizeof (*ipv6h);
917 	if (encap_limit >= 0) {
918 		max_headroom += 8;
919 		mtu -= 8;
920 	}
921 	if (mtu < IPV6_MIN_MTU)
922 		mtu = IPV6_MIN_MTU;
923 	if (skb_dst(skb))
924 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
925 	if (skb->len > mtu) {
926 		*pmtu = mtu;
927 		err = -EMSGSIZE;
928 		goto tx_err_dst_release;
929 	}
930 
931 	/*
932 	 * Okay, now see if we can stuff it in the buffer as-is.
933 	 */
934 	max_headroom += LL_RESERVED_SPACE(tdev);
935 
936 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
937 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
938 		struct sk_buff *new_skb;
939 
940 		if (!(new_skb = skb_realloc_headroom(skb, max_headroom)))
941 			goto tx_err_dst_release;
942 
943 		if (skb->sk)
944 			skb_set_owner_w(new_skb, skb->sk);
945 		consume_skb(skb);
946 		skb = new_skb;
947 	}
948 	skb_dst_drop(skb);
949 	if (fl6->flowi6_mark) {
950 		skb_dst_set(skb, dst);
951 		ndst = NULL;
952 	} else {
953 		skb_dst_set_noref(skb, dst);
954 	}
955 	skb->transport_header = skb->network_header;
956 
957 	proto = fl6->flowi6_proto;
958 	if (encap_limit >= 0) {
959 		init_tel_txopt(&opt, encap_limit);
960 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
961 	}
962 	skb_push(skb, sizeof(struct ipv6hdr));
963 	skb_reset_network_header(skb);
964 	ipv6h = ipv6_hdr(skb);
965 	*(__be32*)ipv6h = fl6->flowlabel | htonl(0x60000000);
966 	dsfield = INET_ECN_encapsulate(0, dsfield);
967 	ipv6_change_dsfield(ipv6h, ~INET_ECN_MASK, dsfield);
968 	ipv6h->hop_limit = t->parms.hop_limit;
969 	ipv6h->nexthdr = proto;
970 	ipv6h->saddr = fl6->saddr;
971 	ipv6h->daddr = fl6->daddr;
972 	nf_reset(skb);
973 	pkt_len = skb->len;
974 	err = ip6_local_out(skb);
975 
976 	if (net_xmit_eval(err) == 0) {
977 		struct pcpu_tstats *tstats = this_cpu_ptr(t->dev->tstats);
978 
979 		tstats->tx_bytes += pkt_len;
980 		tstats->tx_packets++;
981 	} else {
982 		stats->tx_errors++;
983 		stats->tx_aborted_errors++;
984 	}
985 	if (ndst)
986 		ip6_tnl_dst_store(t, ndst);
987 	return 0;
988 tx_err_link_failure:
989 	stats->tx_carrier_errors++;
990 	dst_link_failure(skb);
991 tx_err_dst_release:
992 	dst_release(ndst);
993 	return err;
994 }
995 
996 static inline int
997 ip4ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
998 {
999 	struct ip6_tnl *t = netdev_priv(dev);
1000 	const struct iphdr  *iph = ip_hdr(skb);
1001 	int encap_limit = -1;
1002 	struct flowi6 fl6;
1003 	__u8 dsfield;
1004 	__u32 mtu;
1005 	int err;
1006 
1007 	if ((t->parms.proto != IPPROTO_IPIP && t->parms.proto != 0) ||
1008 	    !ip6_tnl_xmit_ctl(t))
1009 		return -1;
1010 
1011 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1012 		encap_limit = t->parms.encap_limit;
1013 
1014 	memcpy(&fl6, &t->fl.u.ip6, sizeof (fl6));
1015 	fl6.flowi6_proto = IPPROTO_IPIP;
1016 
1017 	dsfield = ipv4_get_dsfield(iph);
1018 
1019 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
1020 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
1021 					  & IPV6_TCLASS_MASK;
1022 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
1023 		fl6.flowi6_mark = skb->mark;
1024 
1025 	err = ip6_tnl_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
1026 	if (err != 0) {
1027 		/* XXX: send ICMP error even if DF is not set. */
1028 		if (err == -EMSGSIZE)
1029 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
1030 				  htonl(mtu));
1031 		return -1;
1032 	}
1033 
1034 	return 0;
1035 }
1036 
1037 static inline int
1038 ip6ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
1039 {
1040 	struct ip6_tnl *t = netdev_priv(dev);
1041 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1042 	int encap_limit = -1;
1043 	__u16 offset;
1044 	struct flowi6 fl6;
1045 	__u8 dsfield;
1046 	__u32 mtu;
1047 	int err;
1048 
1049 	if ((t->parms.proto != IPPROTO_IPV6 && t->parms.proto != 0) ||
1050 	    !ip6_tnl_xmit_ctl(t) || ip6_tnl_addr_conflict(t, ipv6h))
1051 		return -1;
1052 
1053 	offset = parse_tlv_tnl_enc_lim(skb, skb_network_header(skb));
1054 	if (offset > 0) {
1055 		struct ipv6_tlv_tnl_enc_lim *tel;
1056 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
1057 		if (tel->encap_limit == 0) {
1058 			icmpv6_send(skb, ICMPV6_PARAMPROB,
1059 				    ICMPV6_HDR_FIELD, offset + 2);
1060 			return -1;
1061 		}
1062 		encap_limit = tel->encap_limit - 1;
1063 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1064 		encap_limit = t->parms.encap_limit;
1065 
1066 	memcpy(&fl6, &t->fl.u.ip6, sizeof (fl6));
1067 	fl6.flowi6_proto = IPPROTO_IPV6;
1068 
1069 	dsfield = ipv6_get_dsfield(ipv6h);
1070 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
1071 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
1072 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
1073 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_FLOWLABEL_MASK);
1074 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
1075 		fl6.flowi6_mark = skb->mark;
1076 
1077 	err = ip6_tnl_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
1078 	if (err != 0) {
1079 		if (err == -EMSGSIZE)
1080 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1081 		return -1;
1082 	}
1083 
1084 	return 0;
1085 }
1086 
1087 static netdev_tx_t
1088 ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
1089 {
1090 	struct ip6_tnl *t = netdev_priv(dev);
1091 	struct net_device_stats *stats = &t->dev->stats;
1092 	int ret;
1093 
1094 	switch (skb->protocol) {
1095 	case htons(ETH_P_IP):
1096 		ret = ip4ip6_tnl_xmit(skb, dev);
1097 		break;
1098 	case htons(ETH_P_IPV6):
1099 		ret = ip6ip6_tnl_xmit(skb, dev);
1100 		break;
1101 	default:
1102 		goto tx_err;
1103 	}
1104 
1105 	if (ret < 0)
1106 		goto tx_err;
1107 
1108 	return NETDEV_TX_OK;
1109 
1110 tx_err:
1111 	stats->tx_errors++;
1112 	stats->tx_dropped++;
1113 	kfree_skb(skb);
1114 	return NETDEV_TX_OK;
1115 }
1116 
1117 static void ip6_tnl_set_cap(struct ip6_tnl *t)
1118 {
1119 	struct ip6_tnl_parm *p = &t->parms;
1120 	int ltype = ipv6_addr_type(&p->laddr);
1121 	int rtype = ipv6_addr_type(&p->raddr);
1122 
1123 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV);
1124 
1125 	if (ltype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
1126 	    rtype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
1127 	    !((ltype|rtype) & IPV6_ADDR_LOOPBACK) &&
1128 	    (!((ltype|rtype) & IPV6_ADDR_LINKLOCAL) || p->link)) {
1129 		if (ltype&IPV6_ADDR_UNICAST)
1130 			p->flags |= IP6_TNL_F_CAP_XMIT;
1131 		if (rtype&IPV6_ADDR_UNICAST)
1132 			p->flags |= IP6_TNL_F_CAP_RCV;
1133 	}
1134 }
1135 
1136 static void ip6_tnl_link_config(struct ip6_tnl *t)
1137 {
1138 	struct net_device *dev = t->dev;
1139 	struct ip6_tnl_parm *p = &t->parms;
1140 	struct flowi6 *fl6 = &t->fl.u.ip6;
1141 
1142 	memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
1143 	memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
1144 
1145 	/* Set up flowi template */
1146 	fl6->saddr = p->laddr;
1147 	fl6->daddr = p->raddr;
1148 	fl6->flowi6_oif = p->link;
1149 	fl6->flowlabel = 0;
1150 
1151 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
1152 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
1153 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
1154 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
1155 
1156 	ip6_tnl_set_cap(t);
1157 
1158 	if (p->flags&IP6_TNL_F_CAP_XMIT && p->flags&IP6_TNL_F_CAP_RCV)
1159 		dev->flags |= IFF_POINTOPOINT;
1160 	else
1161 		dev->flags &= ~IFF_POINTOPOINT;
1162 
1163 	dev->iflink = p->link;
1164 
1165 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
1166 		int strict = (ipv6_addr_type(&p->raddr) &
1167 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
1168 
1169 		struct rt6_info *rt = rt6_lookup(dev_net(dev),
1170 						 &p->raddr, &p->laddr,
1171 						 p->link, strict);
1172 
1173 		if (rt == NULL)
1174 			return;
1175 
1176 		if (rt->dst.dev) {
1177 			dev->hard_header_len = rt->dst.dev->hard_header_len +
1178 				sizeof (struct ipv6hdr);
1179 
1180 			dev->mtu = rt->dst.dev->mtu - sizeof (struct ipv6hdr);
1181 			if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1182 				dev->mtu-=8;
1183 
1184 			if (dev->mtu < IPV6_MIN_MTU)
1185 				dev->mtu = IPV6_MIN_MTU;
1186 		}
1187 		dst_release(&rt->dst);
1188 	}
1189 }
1190 
1191 /**
1192  * ip6_tnl_change - update the tunnel parameters
1193  *   @t: tunnel to be changed
1194  *   @p: tunnel configuration parameters
1195  *
1196  * Description:
1197  *   ip6_tnl_change() updates the tunnel parameters
1198  **/
1199 
1200 static int
1201 ip6_tnl_change(struct ip6_tnl *t, struct ip6_tnl_parm *p)
1202 {
1203 	t->parms.laddr = p->laddr;
1204 	t->parms.raddr = p->raddr;
1205 	t->parms.flags = p->flags;
1206 	t->parms.hop_limit = p->hop_limit;
1207 	t->parms.encap_limit = p->encap_limit;
1208 	t->parms.flowinfo = p->flowinfo;
1209 	t->parms.link = p->link;
1210 	t->parms.proto = p->proto;
1211 	ip6_tnl_dst_reset(t);
1212 	ip6_tnl_link_config(t);
1213 	return 0;
1214 }
1215 
1216 /**
1217  * ip6_tnl_ioctl - configure ipv6 tunnels from userspace
1218  *   @dev: virtual device associated with tunnel
1219  *   @ifr: parameters passed from userspace
1220  *   @cmd: command to be performed
1221  *
1222  * Description:
1223  *   ip6_tnl_ioctl() is used for managing IPv6 tunnels
1224  *   from userspace.
1225  *
1226  *   The possible commands are the following:
1227  *     %SIOCGETTUNNEL: get tunnel parameters for device
1228  *     %SIOCADDTUNNEL: add tunnel matching given tunnel parameters
1229  *     %SIOCCHGTUNNEL: change tunnel parameters to those given
1230  *     %SIOCDELTUNNEL: delete tunnel
1231  *
1232  *   The fallback device "ip6tnl0", created during module
1233  *   initialization, can be used for creating other tunnel devices.
1234  *
1235  * Return:
1236  *   0 on success,
1237  *   %-EFAULT if unable to copy data to or from userspace,
1238  *   %-EPERM if current process hasn't %CAP_NET_ADMIN set
1239  *   %-EINVAL if passed tunnel parameters are invalid,
1240  *   %-EEXIST if changing a tunnel's parameters would cause a conflict
1241  *   %-ENODEV if attempting to change or delete a nonexisting device
1242  **/
1243 
1244 static int
1245 ip6_tnl_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1246 {
1247 	int err = 0;
1248 	struct ip6_tnl_parm p;
1249 	struct ip6_tnl *t = NULL;
1250 	struct net *net = dev_net(dev);
1251 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
1252 
1253 	switch (cmd) {
1254 	case SIOCGETTUNNEL:
1255 		if (dev == ip6n->fb_tnl_dev) {
1256 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p))) {
1257 				err = -EFAULT;
1258 				break;
1259 			}
1260 			t = ip6_tnl_locate(net, &p, 0);
1261 		}
1262 		if (t == NULL)
1263 			t = netdev_priv(dev);
1264 		memcpy(&p, &t->parms, sizeof (p));
1265 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof (p))) {
1266 			err = -EFAULT;
1267 		}
1268 		break;
1269 	case SIOCADDTUNNEL:
1270 	case SIOCCHGTUNNEL:
1271 		err = -EPERM;
1272 		if (!capable(CAP_NET_ADMIN))
1273 			break;
1274 		err = -EFAULT;
1275 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
1276 			break;
1277 		err = -EINVAL;
1278 		if (p.proto != IPPROTO_IPV6 && p.proto != IPPROTO_IPIP &&
1279 		    p.proto != 0)
1280 			break;
1281 		t = ip6_tnl_locate(net, &p, cmd == SIOCADDTUNNEL);
1282 		if (dev != ip6n->fb_tnl_dev && cmd == SIOCCHGTUNNEL) {
1283 			if (t != NULL) {
1284 				if (t->dev != dev) {
1285 					err = -EEXIST;
1286 					break;
1287 				}
1288 			} else
1289 				t = netdev_priv(dev);
1290 
1291 			ip6_tnl_unlink(ip6n, t);
1292 			synchronize_net();
1293 			err = ip6_tnl_change(t, &p);
1294 			ip6_tnl_link(ip6n, t);
1295 			netdev_state_change(dev);
1296 		}
1297 		if (t) {
1298 			err = 0;
1299 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof (p)))
1300 				err = -EFAULT;
1301 
1302 		} else
1303 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1304 		break;
1305 	case SIOCDELTUNNEL:
1306 		err = -EPERM;
1307 		if (!capable(CAP_NET_ADMIN))
1308 			break;
1309 
1310 		if (dev == ip6n->fb_tnl_dev) {
1311 			err = -EFAULT;
1312 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
1313 				break;
1314 			err = -ENOENT;
1315 			if ((t = ip6_tnl_locate(net, &p, 0)) == NULL)
1316 				break;
1317 			err = -EPERM;
1318 			if (t->dev == ip6n->fb_tnl_dev)
1319 				break;
1320 			dev = t->dev;
1321 		}
1322 		err = 0;
1323 		unregister_netdevice(dev);
1324 		break;
1325 	default:
1326 		err = -EINVAL;
1327 	}
1328 	return err;
1329 }
1330 
1331 /**
1332  * ip6_tnl_change_mtu - change mtu manually for tunnel device
1333  *   @dev: virtual device associated with tunnel
1334  *   @new_mtu: the new mtu
1335  *
1336  * Return:
1337  *   0 on success,
1338  *   %-EINVAL if mtu too small
1339  **/
1340 
1341 static int
1342 ip6_tnl_change_mtu(struct net_device *dev, int new_mtu)
1343 {
1344 	if (new_mtu < IPV6_MIN_MTU) {
1345 		return -EINVAL;
1346 	}
1347 	dev->mtu = new_mtu;
1348 	return 0;
1349 }
1350 
1351 
1352 static const struct net_device_ops ip6_tnl_netdev_ops = {
1353 	.ndo_uninit	= ip6_tnl_dev_uninit,
1354 	.ndo_start_xmit = ip6_tnl_xmit,
1355 	.ndo_do_ioctl	= ip6_tnl_ioctl,
1356 	.ndo_change_mtu = ip6_tnl_change_mtu,
1357 	.ndo_get_stats	= ip6_get_stats,
1358 };
1359 
1360 
1361 /**
1362  * ip6_tnl_dev_setup - setup virtual tunnel device
1363  *   @dev: virtual device associated with tunnel
1364  *
1365  * Description:
1366  *   Initialize function pointers and device parameters
1367  **/
1368 
1369 static void ip6_tnl_dev_setup(struct net_device *dev)
1370 {
1371 	struct ip6_tnl *t;
1372 
1373 	dev->netdev_ops = &ip6_tnl_netdev_ops;
1374 	dev->destructor = ip6_dev_free;
1375 
1376 	dev->type = ARPHRD_TUNNEL6;
1377 	dev->hard_header_len = LL_MAX_HEADER + sizeof (struct ipv6hdr);
1378 	dev->mtu = ETH_DATA_LEN - sizeof (struct ipv6hdr);
1379 	t = netdev_priv(dev);
1380 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1381 		dev->mtu-=8;
1382 	dev->flags |= IFF_NOARP;
1383 	dev->addr_len = sizeof(struct in6_addr);
1384 	dev->features |= NETIF_F_NETNS_LOCAL;
1385 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1386 }
1387 
1388 
1389 /**
1390  * ip6_tnl_dev_init_gen - general initializer for all tunnel devices
1391  *   @dev: virtual device associated with tunnel
1392  **/
1393 
1394 static inline int
1395 ip6_tnl_dev_init_gen(struct net_device *dev)
1396 {
1397 	struct ip6_tnl *t = netdev_priv(dev);
1398 
1399 	t->dev = dev;
1400 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1401 	if (!dev->tstats)
1402 		return -ENOMEM;
1403 	return 0;
1404 }
1405 
1406 /**
1407  * ip6_tnl_dev_init - initializer for all non fallback tunnel devices
1408  *   @dev: virtual device associated with tunnel
1409  **/
1410 
1411 static int ip6_tnl_dev_init(struct net_device *dev)
1412 {
1413 	struct ip6_tnl *t = netdev_priv(dev);
1414 	int err = ip6_tnl_dev_init_gen(dev);
1415 
1416 	if (err)
1417 		return err;
1418 	ip6_tnl_link_config(t);
1419 	return 0;
1420 }
1421 
1422 /**
1423  * ip6_fb_tnl_dev_init - initializer for fallback tunnel device
1424  *   @dev: fallback device
1425  *
1426  * Return: 0
1427  **/
1428 
1429 static int __net_init ip6_fb_tnl_dev_init(struct net_device *dev)
1430 {
1431 	struct ip6_tnl *t = netdev_priv(dev);
1432 	struct net *net = dev_net(dev);
1433 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
1434 	int err = ip6_tnl_dev_init_gen(dev);
1435 
1436 	if (err)
1437 		return err;
1438 
1439 	t->parms.proto = IPPROTO_IPV6;
1440 	dev_hold(dev);
1441 	rcu_assign_pointer(ip6n->tnls_wc[0], t);
1442 	return 0;
1443 }
1444 
1445 static struct xfrm6_tunnel ip4ip6_handler __read_mostly = {
1446 	.handler	= ip4ip6_rcv,
1447 	.err_handler	= ip4ip6_err,
1448 	.priority	=	1,
1449 };
1450 
1451 static struct xfrm6_tunnel ip6ip6_handler __read_mostly = {
1452 	.handler	= ip6ip6_rcv,
1453 	.err_handler	= ip6ip6_err,
1454 	.priority	=	1,
1455 };
1456 
1457 static void __net_exit ip6_tnl_destroy_tunnels(struct ip6_tnl_net *ip6n)
1458 {
1459 	int h;
1460 	struct ip6_tnl *t;
1461 	LIST_HEAD(list);
1462 
1463 	for (h = 0; h < HASH_SIZE; h++) {
1464 		t = rtnl_dereference(ip6n->tnls_r_l[h]);
1465 		while (t != NULL) {
1466 			unregister_netdevice_queue(t->dev, &list);
1467 			t = rtnl_dereference(t->next);
1468 		}
1469 	}
1470 
1471 	t = rtnl_dereference(ip6n->tnls_wc[0]);
1472 	unregister_netdevice_queue(t->dev, &list);
1473 	unregister_netdevice_many(&list);
1474 }
1475 
1476 static int __net_init ip6_tnl_init_net(struct net *net)
1477 {
1478 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
1479 	struct ip6_tnl *t = NULL;
1480 	int err;
1481 
1482 	ip6n->tnls[0] = ip6n->tnls_wc;
1483 	ip6n->tnls[1] = ip6n->tnls_r_l;
1484 
1485 	err = -ENOMEM;
1486 	ip6n->fb_tnl_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6tnl0",
1487 				      ip6_tnl_dev_setup);
1488 
1489 	if (!ip6n->fb_tnl_dev)
1490 		goto err_alloc_dev;
1491 	dev_net_set(ip6n->fb_tnl_dev, net);
1492 
1493 	err = ip6_fb_tnl_dev_init(ip6n->fb_tnl_dev);
1494 	if (err < 0)
1495 		goto err_register;
1496 
1497 	err = register_netdev(ip6n->fb_tnl_dev);
1498 	if (err < 0)
1499 		goto err_register;
1500 
1501 	t = netdev_priv(ip6n->fb_tnl_dev);
1502 
1503 	strcpy(t->parms.name, ip6n->fb_tnl_dev->name);
1504 	return 0;
1505 
1506 err_register:
1507 	ip6_dev_free(ip6n->fb_tnl_dev);
1508 err_alloc_dev:
1509 	return err;
1510 }
1511 
1512 static void __net_exit ip6_tnl_exit_net(struct net *net)
1513 {
1514 	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
1515 
1516 	rtnl_lock();
1517 	ip6_tnl_destroy_tunnels(ip6n);
1518 	rtnl_unlock();
1519 }
1520 
1521 static struct pernet_operations ip6_tnl_net_ops = {
1522 	.init = ip6_tnl_init_net,
1523 	.exit = ip6_tnl_exit_net,
1524 	.id   = &ip6_tnl_net_id,
1525 	.size = sizeof(struct ip6_tnl_net),
1526 };
1527 
1528 /**
1529  * ip6_tunnel_init - register protocol and reserve needed resources
1530  *
1531  * Return: 0 on success
1532  **/
1533 
1534 static int __init ip6_tunnel_init(void)
1535 {
1536 	int  err;
1537 
1538 	err = register_pernet_device(&ip6_tnl_net_ops);
1539 	if (err < 0)
1540 		goto out_pernet;
1541 
1542 	err = xfrm6_tunnel_register(&ip4ip6_handler, AF_INET);
1543 	if (err < 0) {
1544 		pr_err("%s: can't register ip4ip6\n", __func__);
1545 		goto out_ip4ip6;
1546 	}
1547 
1548 	err = xfrm6_tunnel_register(&ip6ip6_handler, AF_INET6);
1549 	if (err < 0) {
1550 		pr_err("%s: can't register ip6ip6\n", __func__);
1551 		goto out_ip6ip6;
1552 	}
1553 
1554 	return 0;
1555 
1556 out_ip6ip6:
1557 	xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET);
1558 out_ip4ip6:
1559 	unregister_pernet_device(&ip6_tnl_net_ops);
1560 out_pernet:
1561 	return err;
1562 }
1563 
1564 /**
1565  * ip6_tunnel_cleanup - free resources and unregister protocol
1566  **/
1567 
1568 static void __exit ip6_tunnel_cleanup(void)
1569 {
1570 	if (xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET))
1571 		pr_info("%s: can't deregister ip4ip6\n", __func__);
1572 
1573 	if (xfrm6_tunnel_deregister(&ip6ip6_handler, AF_INET6))
1574 		pr_info("%s: can't deregister ip6ip6\n", __func__);
1575 
1576 	unregister_pernet_device(&ip6_tnl_net_ops);
1577 }
1578 
1579 module_init(ip6_tunnel_init);
1580 module_exit(ip6_tunnel_cleanup);
1581