xref: /linux/net/ipv6/sit.c (revision 913df4453f85f1fe79b35ecf3c9a0c0b707d22a2)
1 /*
2  *	IPv6 over IPv4 tunnel device - Simple Internet Transition (SIT)
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *	Alexey Kuznetsov	<kuznet@ms2.inr.ac.ru>
8  *
9  *	This program is free software; you can redistribute it and/or
10  *      modify it under the terms of the GNU General Public License
11  *      as published by the Free Software Foundation; either version
12  *      2 of the License, or (at your option) any later version.
13  *
14  *	Changes:
15  * Roger Venning <r.venning@telstra.com>:	6to4 support
16  * Nate Thompson <nate@thebog.net>:		6to4 support
17  * Fred Templin <fred.l.templin@boeing.com>:	isatap support
18  */
19 
20 #include <linux/module.h>
21 #include <linux/capability.h>
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/net.h>
27 #include <linux/in6.h>
28 #include <linux/netdevice.h>
29 #include <linux/if_arp.h>
30 #include <linux/icmp.h>
31 #include <asm/uaccess.h>
32 #include <linux/init.h>
33 #include <linux/netfilter_ipv4.h>
34 #include <linux/if_ether.h>
35 
36 #include <net/sock.h>
37 #include <net/snmp.h>
38 
39 #include <net/ipv6.h>
40 #include <net/protocol.h>
41 #include <net/transp_v6.h>
42 #include <net/ip6_fib.h>
43 #include <net/ip6_route.h>
44 #include <net/ndisc.h>
45 #include <net/addrconf.h>
46 #include <net/ip.h>
47 #include <net/udp.h>
48 #include <net/icmp.h>
49 #include <net/ipip.h>
50 #include <net/inet_ecn.h>
51 #include <net/xfrm.h>
52 #include <net/dsfield.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
55 
56 /*
57    This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c
58 
59    For comments look at net/ipv4/ip_gre.c --ANK
60  */
61 
62 #define HASH_SIZE  16
63 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
64 
65 static void ipip6_fb_tunnel_init(struct net_device *dev);
66 static void ipip6_tunnel_init(struct net_device *dev);
67 static void ipip6_tunnel_setup(struct net_device *dev);
68 
69 static int sit_net_id;
70 struct sit_net {
71 	struct ip_tunnel *tunnels_r_l[HASH_SIZE];
72 	struct ip_tunnel *tunnels_r[HASH_SIZE];
73 	struct ip_tunnel *tunnels_l[HASH_SIZE];
74 	struct ip_tunnel *tunnels_wc[1];
75 	struct ip_tunnel **tunnels[4];
76 
77 	struct net_device *fb_tunnel_dev;
78 };
79 
80 static DEFINE_RWLOCK(ipip6_lock);
81 
82 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
83 		struct net_device *dev, __be32 remote, __be32 local)
84 {
85 	unsigned h0 = HASH(remote);
86 	unsigned h1 = HASH(local);
87 	struct ip_tunnel *t;
88 	struct sit_net *sitn = net_generic(net, sit_net_id);
89 
90 	for (t = sitn->tunnels_r_l[h0^h1]; t; t = t->next) {
91 		if (local == t->parms.iph.saddr &&
92 		    remote == t->parms.iph.daddr &&
93 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
94 		    (t->dev->flags & IFF_UP))
95 			return t;
96 	}
97 	for (t = sitn->tunnels_r[h0]; t; t = t->next) {
98 		if (remote == t->parms.iph.daddr &&
99 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
100 		    (t->dev->flags & IFF_UP))
101 			return t;
102 	}
103 	for (t = sitn->tunnels_l[h1]; t; t = t->next) {
104 		if (local == t->parms.iph.saddr &&
105 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
106 		    (t->dev->flags & IFF_UP))
107 			return t;
108 	}
109 	t = sitn->tunnels_wc[0];
110 	if ((t != NULL) && (t->dev->flags & IFF_UP))
111 		return t;
112 	return NULL;
113 }
114 
115 static struct ip_tunnel **__ipip6_bucket(struct sit_net *sitn,
116 		struct ip_tunnel_parm *parms)
117 {
118 	__be32 remote = parms->iph.daddr;
119 	__be32 local = parms->iph.saddr;
120 	unsigned h = 0;
121 	int prio = 0;
122 
123 	if (remote) {
124 		prio |= 2;
125 		h ^= HASH(remote);
126 	}
127 	if (local) {
128 		prio |= 1;
129 		h ^= HASH(local);
130 	}
131 	return &sitn->tunnels[prio][h];
132 }
133 
134 static inline struct ip_tunnel **ipip6_bucket(struct sit_net *sitn,
135 		struct ip_tunnel *t)
136 {
137 	return __ipip6_bucket(sitn, &t->parms);
138 }
139 
140 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
141 {
142 	struct ip_tunnel **tp;
143 
144 	for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
145 		if (t == *tp) {
146 			write_lock_bh(&ipip6_lock);
147 			*tp = t->next;
148 			write_unlock_bh(&ipip6_lock);
149 			break;
150 		}
151 	}
152 }
153 
154 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
155 {
156 	struct ip_tunnel **tp = ipip6_bucket(sitn, t);
157 
158 	t->next = *tp;
159 	write_lock_bh(&ipip6_lock);
160 	*tp = t;
161 	write_unlock_bh(&ipip6_lock);
162 }
163 
164 static struct ip_tunnel * ipip6_tunnel_locate(struct net *net,
165 		struct ip_tunnel_parm *parms, int create)
166 {
167 	__be32 remote = parms->iph.daddr;
168 	__be32 local = parms->iph.saddr;
169 	struct ip_tunnel *t, **tp, *nt;
170 	struct net_device *dev;
171 	char name[IFNAMSIZ];
172 	struct sit_net *sitn = net_generic(net, sit_net_id);
173 
174 	for (tp = __ipip6_bucket(sitn, parms); (t = *tp) != NULL; tp = &t->next) {
175 		if (local == t->parms.iph.saddr &&
176 		    remote == t->parms.iph.daddr &&
177 		    parms->link == t->parms.link) {
178 			if (create)
179 				return NULL;
180 			else
181 				return t;
182 		}
183 	}
184 	if (!create)
185 		goto failed;
186 
187 	if (parms->name[0])
188 		strlcpy(name, parms->name, IFNAMSIZ);
189 	else
190 		sprintf(name, "sit%%d");
191 
192 	dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
193 	if (dev == NULL)
194 		return NULL;
195 
196 	dev_net_set(dev, net);
197 
198 	if (strchr(name, '%')) {
199 		if (dev_alloc_name(dev, name) < 0)
200 			goto failed_free;
201 	}
202 
203 	nt = netdev_priv(dev);
204 
205 	nt->parms = *parms;
206 	ipip6_tunnel_init(dev);
207 
208 	if (parms->i_flags & SIT_ISATAP)
209 		dev->priv_flags |= IFF_ISATAP;
210 
211 	if (register_netdevice(dev) < 0)
212 		goto failed_free;
213 
214 	dev_hold(dev);
215 
216 	ipip6_tunnel_link(sitn, nt);
217 	return nt;
218 
219 failed_free:
220 	free_netdev(dev);
221 failed:
222 	return NULL;
223 }
224 
225 static struct ip_tunnel_prl_entry *
226 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
227 {
228 	struct ip_tunnel_prl_entry *p = (struct ip_tunnel_prl_entry *)NULL;
229 
230 	for (p = t->prl; p; p = p->next)
231 		if (p->addr == addr)
232 			break;
233 	return p;
234 
235 }
236 
237 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
238 				struct ip_tunnel_prl __user *a)
239 {
240 	struct ip_tunnel_prl kprl, *kp;
241 	struct ip_tunnel_prl_entry *prl;
242 	unsigned int cmax, c = 0, ca, len;
243 	int ret = 0;
244 
245 	if (copy_from_user(&kprl, a, sizeof(kprl)))
246 		return -EFAULT;
247 	cmax = kprl.datalen / sizeof(kprl);
248 	if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
249 		cmax = 1;
250 
251 	/* For simple GET or for root users,
252 	 * we try harder to allocate.
253 	 */
254 	kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
255 		kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
256 		NULL;
257 
258 	read_lock(&ipip6_lock);
259 
260 	ca = t->prl_count < cmax ? t->prl_count : cmax;
261 
262 	if (!kp) {
263 		/* We don't try hard to allocate much memory for
264 		 * non-root users.
265 		 * For root users, retry allocating enough memory for
266 		 * the answer.
267 		 */
268 		kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
269 		if (!kp) {
270 			ret = -ENOMEM;
271 			goto out;
272 		}
273 	}
274 
275 	c = 0;
276 	for (prl = t->prl; prl; prl = prl->next) {
277 		if (c >= cmax)
278 			break;
279 		if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
280 			continue;
281 		kp[c].addr = prl->addr;
282 		kp[c].flags = prl->flags;
283 		c++;
284 		if (kprl.addr != htonl(INADDR_ANY))
285 			break;
286 	}
287 out:
288 	read_unlock(&ipip6_lock);
289 
290 	len = sizeof(*kp) * c;
291 	ret = 0;
292 	if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
293 		ret = -EFAULT;
294 
295 	kfree(kp);
296 
297 	return ret;
298 }
299 
300 static int
301 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
302 {
303 	struct ip_tunnel_prl_entry *p;
304 	int err = 0;
305 
306 	if (a->addr == htonl(INADDR_ANY))
307 		return -EINVAL;
308 
309 	write_lock(&ipip6_lock);
310 
311 	for (p = t->prl; p; p = p->next) {
312 		if (p->addr == a->addr) {
313 			if (chg)
314 				goto update;
315 			err = -EEXIST;
316 			goto out;
317 		}
318 	}
319 
320 	if (chg) {
321 		err = -ENXIO;
322 		goto out;
323 	}
324 
325 	p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
326 	if (!p) {
327 		err = -ENOBUFS;
328 		goto out;
329 	}
330 
331 	p->next = t->prl;
332 	t->prl = p;
333 	t->prl_count++;
334 update:
335 	p->addr = a->addr;
336 	p->flags = a->flags;
337 out:
338 	write_unlock(&ipip6_lock);
339 	return err;
340 }
341 
342 static int
343 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
344 {
345 	struct ip_tunnel_prl_entry *x, **p;
346 	int err = 0;
347 
348 	write_lock(&ipip6_lock);
349 
350 	if (a && a->addr != htonl(INADDR_ANY)) {
351 		for (p = &t->prl; *p; p = &(*p)->next) {
352 			if ((*p)->addr == a->addr) {
353 				x = *p;
354 				*p = x->next;
355 				kfree(x);
356 				t->prl_count--;
357 				goto out;
358 			}
359 		}
360 		err = -ENXIO;
361 	} else {
362 		while (t->prl) {
363 			x = t->prl;
364 			t->prl = t->prl->next;
365 			kfree(x);
366 			t->prl_count--;
367 		}
368 	}
369 out:
370 	write_unlock(&ipip6_lock);
371 	return err;
372 }
373 
374 static int
375 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
376 {
377 	struct ip_tunnel_prl_entry *p;
378 	int ok = 1;
379 
380 	read_lock(&ipip6_lock);
381 	p = __ipip6_tunnel_locate_prl(t, iph->saddr);
382 	if (p) {
383 		if (p->flags & PRL_DEFAULT)
384 			skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
385 		else
386 			skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
387 	} else {
388 		struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
389 		if (ipv6_addr_is_isatap(addr6) &&
390 		    (addr6->s6_addr32[3] == iph->saddr) &&
391 		    ipv6_chk_prefix(addr6, t->dev))
392 			skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
393 		else
394 			ok = 0;
395 	}
396 	read_unlock(&ipip6_lock);
397 	return ok;
398 }
399 
400 static void ipip6_tunnel_uninit(struct net_device *dev)
401 {
402 	struct net *net = dev_net(dev);
403 	struct sit_net *sitn = net_generic(net, sit_net_id);
404 
405 	if (dev == sitn->fb_tunnel_dev) {
406 		write_lock_bh(&ipip6_lock);
407 		sitn->tunnels_wc[0] = NULL;
408 		write_unlock_bh(&ipip6_lock);
409 		dev_put(dev);
410 	} else {
411 		ipip6_tunnel_unlink(sitn, netdev_priv(dev));
412 		ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
413 		dev_put(dev);
414 	}
415 }
416 
417 
418 static int ipip6_err(struct sk_buff *skb, u32 info)
419 {
420 
421 /* All the routers (except for Linux) return only
422    8 bytes of packet payload. It means, that precise relaying of
423    ICMP in the real Internet is absolutely infeasible.
424  */
425 	struct iphdr *iph = (struct iphdr*)skb->data;
426 	const int type = icmp_hdr(skb)->type;
427 	const int code = icmp_hdr(skb)->code;
428 	struct ip_tunnel *t;
429 	int err;
430 
431 	switch (type) {
432 	default:
433 	case ICMP_PARAMETERPROB:
434 		return 0;
435 
436 	case ICMP_DEST_UNREACH:
437 		switch (code) {
438 		case ICMP_SR_FAILED:
439 		case ICMP_PORT_UNREACH:
440 			/* Impossible event. */
441 			return 0;
442 		case ICMP_FRAG_NEEDED:
443 			/* Soft state for pmtu is maintained by IP core. */
444 			return 0;
445 		default:
446 			/* All others are translated to HOST_UNREACH.
447 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
448 			   I believe they are just ether pollution. --ANK
449 			 */
450 			break;
451 		}
452 		break;
453 	case ICMP_TIME_EXCEEDED:
454 		if (code != ICMP_EXC_TTL)
455 			return 0;
456 		break;
457 	}
458 
459 	err = -ENOENT;
460 
461 	read_lock(&ipip6_lock);
462 	t = ipip6_tunnel_lookup(dev_net(skb->dev),
463 				skb->dev,
464 				iph->daddr,
465 				iph->saddr);
466 	if (t == NULL || t->parms.iph.daddr == 0)
467 		goto out;
468 
469 	err = 0;
470 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
471 		goto out;
472 
473 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
474 		t->err_count++;
475 	else
476 		t->err_count = 1;
477 	t->err_time = jiffies;
478 out:
479 	read_unlock(&ipip6_lock);
480 	return err;
481 }
482 
483 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
484 {
485 	if (INET_ECN_is_ce(iph->tos))
486 		IP6_ECN_set_ce(ipv6_hdr(skb));
487 }
488 
489 static int ipip6_rcv(struct sk_buff *skb)
490 {
491 	struct iphdr *iph;
492 	struct ip_tunnel *tunnel;
493 
494 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
495 		goto out;
496 
497 	iph = ip_hdr(skb);
498 
499 	read_lock(&ipip6_lock);
500 	tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
501 				     iph->saddr, iph->daddr);
502 	if (tunnel != NULL) {
503 		secpath_reset(skb);
504 		skb->mac_header = skb->network_header;
505 		skb_reset_network_header(skb);
506 		IPCB(skb)->flags = 0;
507 		skb->protocol = htons(ETH_P_IPV6);
508 		skb->pkt_type = PACKET_HOST;
509 
510 		if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
511 		    !isatap_chksrc(skb, iph, tunnel)) {
512 			tunnel->dev->stats.rx_errors++;
513 			read_unlock(&ipip6_lock);
514 			kfree_skb(skb);
515 			return 0;
516 		}
517 		tunnel->dev->stats.rx_packets++;
518 		tunnel->dev->stats.rx_bytes += skb->len;
519 		skb->dev = tunnel->dev;
520 		skb_dst_drop(skb);
521 		nf_reset(skb);
522 		ipip6_ecn_decapsulate(iph, skb);
523 		netif_rx(skb);
524 		read_unlock(&ipip6_lock);
525 		return 0;
526 	}
527 
528 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
529 	read_unlock(&ipip6_lock);
530 out:
531 	kfree_skb(skb);
532 	return 0;
533 }
534 
535 /* Returns the embedded IPv4 address if the IPv6 address
536    comes from 6to4 (RFC 3056) addr space */
537 
538 static inline __be32 try_6to4(struct in6_addr *v6dst)
539 {
540 	__be32 dst = 0;
541 
542 	if (v6dst->s6_addr16[0] == htons(0x2002)) {
543 		/* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
544 		memcpy(&dst, &v6dst->s6_addr16[1], 4);
545 	}
546 	return dst;
547 }
548 
549 /*
550  *	This function assumes it is being called from dev_queue_xmit()
551  *	and that skb is filled properly by that function.
552  */
553 
554 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
555 				     struct net_device *dev)
556 {
557 	struct ip_tunnel *tunnel = netdev_priv(dev);
558 	struct net_device_stats *stats = &tunnel->dev->stats;
559 	struct iphdr  *tiph = &tunnel->parms.iph;
560 	struct ipv6hdr *iph6 = ipv6_hdr(skb);
561 	u8     tos = tunnel->parms.iph.tos;
562 	struct rtable *rt;     			/* Route to the other host */
563 	struct net_device *tdev;			/* Device to other host */
564 	struct iphdr  *iph;			/* Our new IP header */
565 	unsigned int max_headroom;		/* The extra header space needed */
566 	__be32 dst = tiph->daddr;
567 	int    mtu;
568 	struct in6_addr *addr6;
569 	int addr_type;
570 
571 	if (skb->protocol != htons(ETH_P_IPV6))
572 		goto tx_error;
573 
574 	/* ISATAP (RFC4214) - must come before 6to4 */
575 	if (dev->priv_flags & IFF_ISATAP) {
576 		struct neighbour *neigh = NULL;
577 
578 		if (skb_dst(skb))
579 			neigh = skb_dst(skb)->neighbour;
580 
581 		if (neigh == NULL) {
582 			if (net_ratelimit())
583 				printk(KERN_DEBUG "sit: nexthop == NULL\n");
584 			goto tx_error;
585 		}
586 
587 		addr6 = (struct in6_addr*)&neigh->primary_key;
588 		addr_type = ipv6_addr_type(addr6);
589 
590 		if ((addr_type & IPV6_ADDR_UNICAST) &&
591 		     ipv6_addr_is_isatap(addr6))
592 			dst = addr6->s6_addr32[3];
593 		else
594 			goto tx_error;
595 	}
596 
597 	if (!dst)
598 		dst = try_6to4(&iph6->daddr);
599 
600 	if (!dst) {
601 		struct neighbour *neigh = NULL;
602 
603 		if (skb_dst(skb))
604 			neigh = skb_dst(skb)->neighbour;
605 
606 		if (neigh == NULL) {
607 			if (net_ratelimit())
608 				printk(KERN_DEBUG "sit: nexthop == NULL\n");
609 			goto tx_error;
610 		}
611 
612 		addr6 = (struct in6_addr*)&neigh->primary_key;
613 		addr_type = ipv6_addr_type(addr6);
614 
615 		if (addr_type == IPV6_ADDR_ANY) {
616 			addr6 = &ipv6_hdr(skb)->daddr;
617 			addr_type = ipv6_addr_type(addr6);
618 		}
619 
620 		if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
621 			goto tx_error_icmp;
622 
623 		dst = addr6->s6_addr32[3];
624 	}
625 
626 	{
627 		struct flowi fl = { .nl_u = { .ip4_u =
628 					      { .daddr = dst,
629 						.saddr = tiph->saddr,
630 						.tos = RT_TOS(tos) } },
631 				    .oif = tunnel->parms.link,
632 				    .proto = IPPROTO_IPV6 };
633 		if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
634 			stats->tx_carrier_errors++;
635 			goto tx_error_icmp;
636 		}
637 	}
638 	if (rt->rt_type != RTN_UNICAST) {
639 		ip_rt_put(rt);
640 		stats->tx_carrier_errors++;
641 		goto tx_error_icmp;
642 	}
643 	tdev = rt->u.dst.dev;
644 
645 	if (tdev == dev) {
646 		ip_rt_put(rt);
647 		stats->collisions++;
648 		goto tx_error;
649 	}
650 
651 	if (tiph->frag_off)
652 		mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
653 	else
654 		mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
655 
656 	if (mtu < 68) {
657 		stats->collisions++;
658 		ip_rt_put(rt);
659 		goto tx_error;
660 	}
661 	if (mtu < IPV6_MIN_MTU)
662 		mtu = IPV6_MIN_MTU;
663 	if (tunnel->parms.iph.daddr && skb_dst(skb))
664 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
665 
666 	if (skb->len > mtu) {
667 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
668 		ip_rt_put(rt);
669 		goto tx_error;
670 	}
671 
672 	if (tunnel->err_count > 0) {
673 		if (time_before(jiffies,
674 				tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
675 			tunnel->err_count--;
676 			dst_link_failure(skb);
677 		} else
678 			tunnel->err_count = 0;
679 	}
680 
681 	/*
682 	 * Okay, now see if we can stuff it in the buffer as-is.
683 	 */
684 	max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
685 
686 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
687 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
688 		struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
689 		if (!new_skb) {
690 			ip_rt_put(rt);
691 			stats->tx_dropped++;
692 			dev_kfree_skb(skb);
693 			return NETDEV_TX_OK;
694 		}
695 		if (skb->sk)
696 			skb_set_owner_w(new_skb, skb->sk);
697 		dev_kfree_skb(skb);
698 		skb = new_skb;
699 		iph6 = ipv6_hdr(skb);
700 	}
701 
702 	skb->transport_header = skb->network_header;
703 	skb_push(skb, sizeof(struct iphdr));
704 	skb_reset_network_header(skb);
705 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
706 	IPCB(skb)->flags = 0;
707 	skb_dst_drop(skb);
708 	skb_dst_set(skb, &rt->u.dst);
709 
710 	/*
711 	 *	Push down and install the IPIP header.
712 	 */
713 
714 	iph 			=	ip_hdr(skb);
715 	iph->version		=	4;
716 	iph->ihl		=	sizeof(struct iphdr)>>2;
717 	if (mtu > IPV6_MIN_MTU)
718 		iph->frag_off	=	tiph->frag_off;
719 	else
720 		iph->frag_off	=	0;
721 
722 	iph->protocol		=	IPPROTO_IPV6;
723 	iph->tos		=	INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
724 	iph->daddr		=	rt->rt_dst;
725 	iph->saddr		=	rt->rt_src;
726 
727 	if ((iph->ttl = tiph->ttl) == 0)
728 		iph->ttl	=	iph6->hop_limit;
729 
730 	nf_reset(skb);
731 
732 	IPTUNNEL_XMIT();
733 	return NETDEV_TX_OK;
734 
735 tx_error_icmp:
736 	dst_link_failure(skb);
737 tx_error:
738 	stats->tx_errors++;
739 	dev_kfree_skb(skb);
740 	return NETDEV_TX_OK;
741 }
742 
743 static void ipip6_tunnel_bind_dev(struct net_device *dev)
744 {
745 	struct net_device *tdev = NULL;
746 	struct ip_tunnel *tunnel;
747 	struct iphdr *iph;
748 
749 	tunnel = netdev_priv(dev);
750 	iph = &tunnel->parms.iph;
751 
752 	if (iph->daddr) {
753 		struct flowi fl = { .nl_u = { .ip4_u =
754 					      { .daddr = iph->daddr,
755 						.saddr = iph->saddr,
756 						.tos = RT_TOS(iph->tos) } },
757 				    .oif = tunnel->parms.link,
758 				    .proto = IPPROTO_IPV6 };
759 		struct rtable *rt;
760 		if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
761 			tdev = rt->u.dst.dev;
762 			ip_rt_put(rt);
763 		}
764 		dev->flags |= IFF_POINTOPOINT;
765 	}
766 
767 	if (!tdev && tunnel->parms.link)
768 		tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
769 
770 	if (tdev) {
771 		dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
772 		dev->mtu = tdev->mtu - sizeof(struct iphdr);
773 		if (dev->mtu < IPV6_MIN_MTU)
774 			dev->mtu = IPV6_MIN_MTU;
775 	}
776 	dev->iflink = tunnel->parms.link;
777 }
778 
779 static int
780 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
781 {
782 	int err = 0;
783 	struct ip_tunnel_parm p;
784 	struct ip_tunnel_prl prl;
785 	struct ip_tunnel *t;
786 	struct net *net = dev_net(dev);
787 	struct sit_net *sitn = net_generic(net, sit_net_id);
788 
789 	switch (cmd) {
790 	case SIOCGETTUNNEL:
791 		t = NULL;
792 		if (dev == sitn->fb_tunnel_dev) {
793 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
794 				err = -EFAULT;
795 				break;
796 			}
797 			t = ipip6_tunnel_locate(net, &p, 0);
798 		}
799 		if (t == NULL)
800 			t = netdev_priv(dev);
801 		memcpy(&p, &t->parms, sizeof(p));
802 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
803 			err = -EFAULT;
804 		break;
805 
806 	case SIOCADDTUNNEL:
807 	case SIOCCHGTUNNEL:
808 		err = -EPERM;
809 		if (!capable(CAP_NET_ADMIN))
810 			goto done;
811 
812 		err = -EFAULT;
813 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
814 			goto done;
815 
816 		err = -EINVAL;
817 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
818 		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
819 			goto done;
820 		if (p.iph.ttl)
821 			p.iph.frag_off |= htons(IP_DF);
822 
823 		t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
824 
825 		if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
826 			if (t != NULL) {
827 				if (t->dev != dev) {
828 					err = -EEXIST;
829 					break;
830 				}
831 			} else {
832 				if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
833 				    (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
834 					err = -EINVAL;
835 					break;
836 				}
837 				t = netdev_priv(dev);
838 				ipip6_tunnel_unlink(sitn, t);
839 				t->parms.iph.saddr = p.iph.saddr;
840 				t->parms.iph.daddr = p.iph.daddr;
841 				memcpy(dev->dev_addr, &p.iph.saddr, 4);
842 				memcpy(dev->broadcast, &p.iph.daddr, 4);
843 				ipip6_tunnel_link(sitn, t);
844 				netdev_state_change(dev);
845 			}
846 		}
847 
848 		if (t) {
849 			err = 0;
850 			if (cmd == SIOCCHGTUNNEL) {
851 				t->parms.iph.ttl = p.iph.ttl;
852 				t->parms.iph.tos = p.iph.tos;
853 				if (t->parms.link != p.link) {
854 					t->parms.link = p.link;
855 					ipip6_tunnel_bind_dev(dev);
856 					netdev_state_change(dev);
857 				}
858 			}
859 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
860 				err = -EFAULT;
861 		} else
862 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
863 		break;
864 
865 	case SIOCDELTUNNEL:
866 		err = -EPERM;
867 		if (!capable(CAP_NET_ADMIN))
868 			goto done;
869 
870 		if (dev == sitn->fb_tunnel_dev) {
871 			err = -EFAULT;
872 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
873 				goto done;
874 			err = -ENOENT;
875 			if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
876 				goto done;
877 			err = -EPERM;
878 			if (t == netdev_priv(sitn->fb_tunnel_dev))
879 				goto done;
880 			dev = t->dev;
881 		}
882 		unregister_netdevice(dev);
883 		err = 0;
884 		break;
885 
886 	case SIOCGETPRL:
887 		err = -EINVAL;
888 		if (dev == sitn->fb_tunnel_dev)
889 			goto done;
890 		err = -ENOENT;
891 		if (!(t = netdev_priv(dev)))
892 			goto done;
893 		err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
894 		break;
895 
896 	case SIOCADDPRL:
897 	case SIOCDELPRL:
898 	case SIOCCHGPRL:
899 		err = -EPERM;
900 		if (!capable(CAP_NET_ADMIN))
901 			goto done;
902 		err = -EINVAL;
903 		if (dev == sitn->fb_tunnel_dev)
904 			goto done;
905 		err = -EFAULT;
906 		if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
907 			goto done;
908 		err = -ENOENT;
909 		if (!(t = netdev_priv(dev)))
910 			goto done;
911 
912 		switch (cmd) {
913 		case SIOCDELPRL:
914 			err = ipip6_tunnel_del_prl(t, &prl);
915 			break;
916 		case SIOCADDPRL:
917 		case SIOCCHGPRL:
918 			err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
919 			break;
920 		}
921 		netdev_state_change(dev);
922 		break;
923 
924 	default:
925 		err = -EINVAL;
926 	}
927 
928 done:
929 	return err;
930 }
931 
932 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
933 {
934 	if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
935 		return -EINVAL;
936 	dev->mtu = new_mtu;
937 	return 0;
938 }
939 
940 static const struct net_device_ops ipip6_netdev_ops = {
941 	.ndo_uninit	= ipip6_tunnel_uninit,
942 	.ndo_start_xmit	= ipip6_tunnel_xmit,
943 	.ndo_do_ioctl	= ipip6_tunnel_ioctl,
944 	.ndo_change_mtu	= ipip6_tunnel_change_mtu,
945 };
946 
947 static void ipip6_tunnel_setup(struct net_device *dev)
948 {
949 	dev->netdev_ops		= &ipip6_netdev_ops;
950 	dev->destructor 	= free_netdev;
951 
952 	dev->type		= ARPHRD_SIT;
953 	dev->hard_header_len 	= LL_MAX_HEADER + sizeof(struct iphdr);
954 	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr);
955 	dev->flags		= IFF_NOARP;
956 	dev->priv_flags	       &= ~IFF_XMIT_DST_RELEASE;
957 	dev->iflink		= 0;
958 	dev->addr_len		= 4;
959 	dev->features		|= NETIF_F_NETNS_LOCAL;
960 }
961 
962 static void ipip6_tunnel_init(struct net_device *dev)
963 {
964 	struct ip_tunnel *tunnel = netdev_priv(dev);
965 
966 	tunnel->dev = dev;
967 	strcpy(tunnel->parms.name, dev->name);
968 
969 	memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
970 	memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
971 
972 	ipip6_tunnel_bind_dev(dev);
973 }
974 
975 static void ipip6_fb_tunnel_init(struct net_device *dev)
976 {
977 	struct ip_tunnel *tunnel = netdev_priv(dev);
978 	struct iphdr *iph = &tunnel->parms.iph;
979 	struct net *net = dev_net(dev);
980 	struct sit_net *sitn = net_generic(net, sit_net_id);
981 
982 	tunnel->dev = dev;
983 	strcpy(tunnel->parms.name, dev->name);
984 
985 	iph->version		= 4;
986 	iph->protocol		= IPPROTO_IPV6;
987 	iph->ihl		= 5;
988 	iph->ttl		= 64;
989 
990 	dev_hold(dev);
991 	sitn->tunnels_wc[0]	= tunnel;
992 }
993 
994 static struct xfrm_tunnel sit_handler = {
995 	.handler	=	ipip6_rcv,
996 	.err_handler	=	ipip6_err,
997 	.priority	=	1,
998 };
999 
1000 static void sit_destroy_tunnels(struct sit_net *sitn)
1001 {
1002 	int prio;
1003 
1004 	for (prio = 1; prio < 4; prio++) {
1005 		int h;
1006 		for (h = 0; h < HASH_SIZE; h++) {
1007 			struct ip_tunnel *t;
1008 			while ((t = sitn->tunnels[prio][h]) != NULL)
1009 				unregister_netdevice(t->dev);
1010 		}
1011 	}
1012 }
1013 
1014 static int sit_init_net(struct net *net)
1015 {
1016 	int err;
1017 	struct sit_net *sitn;
1018 
1019 	err = -ENOMEM;
1020 	sitn = kzalloc(sizeof(struct sit_net), GFP_KERNEL);
1021 	if (sitn == NULL)
1022 		goto err_alloc;
1023 
1024 	err = net_assign_generic(net, sit_net_id, sitn);
1025 	if (err < 0)
1026 		goto err_assign;
1027 
1028 	sitn->tunnels[0] = sitn->tunnels_wc;
1029 	sitn->tunnels[1] = sitn->tunnels_l;
1030 	sitn->tunnels[2] = sitn->tunnels_r;
1031 	sitn->tunnels[3] = sitn->tunnels_r_l;
1032 
1033 	sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1034 					   ipip6_tunnel_setup);
1035 	if (!sitn->fb_tunnel_dev) {
1036 		err = -ENOMEM;
1037 		goto err_alloc_dev;
1038 	}
1039 	dev_net_set(sitn->fb_tunnel_dev, net);
1040 
1041 	ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1042 
1043 	if ((err = register_netdev(sitn->fb_tunnel_dev)))
1044 		goto err_reg_dev;
1045 
1046 	return 0;
1047 
1048 err_reg_dev:
1049 	dev_put(sitn->fb_tunnel_dev);
1050 	free_netdev(sitn->fb_tunnel_dev);
1051 err_alloc_dev:
1052 	/* nothing */
1053 err_assign:
1054 	kfree(sitn);
1055 err_alloc:
1056 	return err;
1057 }
1058 
1059 static void sit_exit_net(struct net *net)
1060 {
1061 	struct sit_net *sitn;
1062 
1063 	sitn = net_generic(net, sit_net_id);
1064 	rtnl_lock();
1065 	sit_destroy_tunnels(sitn);
1066 	unregister_netdevice(sitn->fb_tunnel_dev);
1067 	rtnl_unlock();
1068 	kfree(sitn);
1069 }
1070 
1071 static struct pernet_operations sit_net_ops = {
1072 	.init = sit_init_net,
1073 	.exit = sit_exit_net,
1074 };
1075 
1076 static void __exit sit_cleanup(void)
1077 {
1078 	xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1079 
1080 	unregister_pernet_gen_device(sit_net_id, &sit_net_ops);
1081 }
1082 
1083 static int __init sit_init(void)
1084 {
1085 	int err;
1086 
1087 	printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1088 
1089 	if (xfrm4_tunnel_register(&sit_handler, AF_INET6) < 0) {
1090 		printk(KERN_INFO "sit init: Can't add protocol\n");
1091 		return -EAGAIN;
1092 	}
1093 
1094 	err = register_pernet_gen_device(&sit_net_id, &sit_net_ops);
1095 	if (err < 0)
1096 		xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1097 
1098 	return err;
1099 }
1100 
1101 module_init(sit_init);
1102 module_exit(sit_cleanup);
1103 MODULE_LICENSE("GPL");
1104 MODULE_ALIAS("sit0");
1105