xref: /linux/net/ipv6/sit.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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 <linux/slab.h>
32 #include <asm/uaccess.h>
33 #include <linux/init.h>
34 #include <linux/netfilter_ipv4.h>
35 #include <linux/if_ether.h>
36 
37 #include <net/sock.h>
38 #include <net/snmp.h>
39 
40 #include <net/ipv6.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_fib.h>
44 #include <net/ip6_route.h>
45 #include <net/ndisc.h>
46 #include <net/addrconf.h>
47 #include <net/ip.h>
48 #include <net/udp.h>
49 #include <net/icmp.h>
50 #include <net/ipip.h>
51 #include <net/inet_ecn.h>
52 #include <net/xfrm.h>
53 #include <net/dsfield.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
56 
57 /*
58    This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c
59 
60    For comments look at net/ipv4/ip_gre.c --ANK
61  */
62 
63 #define HASH_SIZE  16
64 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
65 
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 __read_mostly;
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 /*
81  * Locking : hash tables are protected by RCU and a spinlock
82  */
83 static DEFINE_SPINLOCK(ipip6_lock);
84 
85 #define for_each_ip_tunnel_rcu(start) \
86 	for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
87 
88 /*
89  * Must be invoked with rcu_read_lock
90  */
91 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
92 		struct net_device *dev, __be32 remote, __be32 local)
93 {
94 	unsigned h0 = HASH(remote);
95 	unsigned h1 = HASH(local);
96 	struct ip_tunnel *t;
97 	struct sit_net *sitn = net_generic(net, sit_net_id);
98 
99 	for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
100 		if (local == t->parms.iph.saddr &&
101 		    remote == t->parms.iph.daddr &&
102 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
103 		    (t->dev->flags & IFF_UP))
104 			return t;
105 	}
106 	for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
107 		if (remote == t->parms.iph.daddr &&
108 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
109 		    (t->dev->flags & IFF_UP))
110 			return t;
111 	}
112 	for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
113 		if (local == t->parms.iph.saddr &&
114 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
115 		    (t->dev->flags & IFF_UP))
116 			return t;
117 	}
118 	t = rcu_dereference(sitn->tunnels_wc[0]);
119 	if ((t != NULL) && (t->dev->flags & IFF_UP))
120 		return t;
121 	return NULL;
122 }
123 
124 static struct ip_tunnel **__ipip6_bucket(struct sit_net *sitn,
125 		struct ip_tunnel_parm *parms)
126 {
127 	__be32 remote = parms->iph.daddr;
128 	__be32 local = parms->iph.saddr;
129 	unsigned h = 0;
130 	int prio = 0;
131 
132 	if (remote) {
133 		prio |= 2;
134 		h ^= HASH(remote);
135 	}
136 	if (local) {
137 		prio |= 1;
138 		h ^= HASH(local);
139 	}
140 	return &sitn->tunnels[prio][h];
141 }
142 
143 static inline struct ip_tunnel **ipip6_bucket(struct sit_net *sitn,
144 		struct ip_tunnel *t)
145 {
146 	return __ipip6_bucket(sitn, &t->parms);
147 }
148 
149 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
150 {
151 	struct ip_tunnel **tp;
152 
153 	for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
154 		if (t == *tp) {
155 			spin_lock_bh(&ipip6_lock);
156 			*tp = t->next;
157 			spin_unlock_bh(&ipip6_lock);
158 			break;
159 		}
160 	}
161 }
162 
163 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
164 {
165 	struct ip_tunnel **tp = ipip6_bucket(sitn, t);
166 
167 	spin_lock_bh(&ipip6_lock);
168 	t->next = *tp;
169 	rcu_assign_pointer(*tp, t);
170 	spin_unlock_bh(&ipip6_lock);
171 }
172 
173 static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
174 {
175 #ifdef CONFIG_IPV6_SIT_6RD
176 	struct ip_tunnel *t = netdev_priv(dev);
177 
178 	if (t->dev == sitn->fb_tunnel_dev) {
179 		ipv6_addr_set(&t->ip6rd.prefix, htonl(0x20020000), 0, 0, 0);
180 		t->ip6rd.relay_prefix = 0;
181 		t->ip6rd.prefixlen = 16;
182 		t->ip6rd.relay_prefixlen = 0;
183 	} else {
184 		struct ip_tunnel *t0 = netdev_priv(sitn->fb_tunnel_dev);
185 		memcpy(&t->ip6rd, &t0->ip6rd, sizeof(t->ip6rd));
186 	}
187 #endif
188 }
189 
190 static struct ip_tunnel * ipip6_tunnel_locate(struct net *net,
191 		struct ip_tunnel_parm *parms, int create)
192 {
193 	__be32 remote = parms->iph.daddr;
194 	__be32 local = parms->iph.saddr;
195 	struct ip_tunnel *t, **tp, *nt;
196 	struct net_device *dev;
197 	char name[IFNAMSIZ];
198 	struct sit_net *sitn = net_generic(net, sit_net_id);
199 
200 	for (tp = __ipip6_bucket(sitn, parms); (t = *tp) != NULL; tp = &t->next) {
201 		if (local == t->parms.iph.saddr &&
202 		    remote == t->parms.iph.daddr &&
203 		    parms->link == t->parms.link) {
204 			if (create)
205 				return NULL;
206 			else
207 				return t;
208 		}
209 	}
210 	if (!create)
211 		goto failed;
212 
213 	if (parms->name[0])
214 		strlcpy(name, parms->name, IFNAMSIZ);
215 	else
216 		sprintf(name, "sit%%d");
217 
218 	dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
219 	if (dev == NULL)
220 		return NULL;
221 
222 	dev_net_set(dev, net);
223 
224 	if (strchr(name, '%')) {
225 		if (dev_alloc_name(dev, name) < 0)
226 			goto failed_free;
227 	}
228 
229 	nt = netdev_priv(dev);
230 
231 	nt->parms = *parms;
232 	ipip6_tunnel_init(dev);
233 	ipip6_tunnel_clone_6rd(dev, sitn);
234 
235 	if (parms->i_flags & SIT_ISATAP)
236 		dev->priv_flags |= IFF_ISATAP;
237 
238 	if (register_netdevice(dev) < 0)
239 		goto failed_free;
240 
241 	dev_hold(dev);
242 
243 	ipip6_tunnel_link(sitn, nt);
244 	return nt;
245 
246 failed_free:
247 	free_netdev(dev);
248 failed:
249 	return NULL;
250 }
251 
252 static DEFINE_SPINLOCK(ipip6_prl_lock);
253 
254 #define for_each_prl_rcu(start)			\
255 	for (prl = rcu_dereference(start);	\
256 	     prl;				\
257 	     prl = rcu_dereference(prl->next))
258 
259 static struct ip_tunnel_prl_entry *
260 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
261 {
262 	struct ip_tunnel_prl_entry *prl;
263 
264 	for_each_prl_rcu(t->prl)
265 		if (prl->addr == addr)
266 			break;
267 	return prl;
268 
269 }
270 
271 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
272 				struct ip_tunnel_prl __user *a)
273 {
274 	struct ip_tunnel_prl kprl, *kp;
275 	struct ip_tunnel_prl_entry *prl;
276 	unsigned int cmax, c = 0, ca, len;
277 	int ret = 0;
278 
279 	if (copy_from_user(&kprl, a, sizeof(kprl)))
280 		return -EFAULT;
281 	cmax = kprl.datalen / sizeof(kprl);
282 	if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
283 		cmax = 1;
284 
285 	/* For simple GET or for root users,
286 	 * we try harder to allocate.
287 	 */
288 	kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
289 		kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
290 		NULL;
291 
292 	rcu_read_lock();
293 
294 	ca = t->prl_count < cmax ? t->prl_count : cmax;
295 
296 	if (!kp) {
297 		/* We don't try hard to allocate much memory for
298 		 * non-root users.
299 		 * For root users, retry allocating enough memory for
300 		 * the answer.
301 		 */
302 		kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
303 		if (!kp) {
304 			ret = -ENOMEM;
305 			goto out;
306 		}
307 	}
308 
309 	c = 0;
310 	for_each_prl_rcu(t->prl) {
311 		if (c >= cmax)
312 			break;
313 		if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
314 			continue;
315 		kp[c].addr = prl->addr;
316 		kp[c].flags = prl->flags;
317 		c++;
318 		if (kprl.addr != htonl(INADDR_ANY))
319 			break;
320 	}
321 out:
322 	rcu_read_unlock();
323 
324 	len = sizeof(*kp) * c;
325 	ret = 0;
326 	if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
327 		ret = -EFAULT;
328 
329 	kfree(kp);
330 
331 	return ret;
332 }
333 
334 static int
335 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
336 {
337 	struct ip_tunnel_prl_entry *p;
338 	int err = 0;
339 
340 	if (a->addr == htonl(INADDR_ANY))
341 		return -EINVAL;
342 
343 	spin_lock(&ipip6_prl_lock);
344 
345 	for (p = t->prl; p; p = p->next) {
346 		if (p->addr == a->addr) {
347 			if (chg) {
348 				p->flags = a->flags;
349 				goto out;
350 			}
351 			err = -EEXIST;
352 			goto out;
353 		}
354 	}
355 
356 	if (chg) {
357 		err = -ENXIO;
358 		goto out;
359 	}
360 
361 	p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
362 	if (!p) {
363 		err = -ENOBUFS;
364 		goto out;
365 	}
366 
367 	p->next = t->prl;
368 	p->addr = a->addr;
369 	p->flags = a->flags;
370 	t->prl_count++;
371 	rcu_assign_pointer(t->prl, p);
372 out:
373 	spin_unlock(&ipip6_prl_lock);
374 	return err;
375 }
376 
377 static void prl_entry_destroy_rcu(struct rcu_head *head)
378 {
379 	kfree(container_of(head, struct ip_tunnel_prl_entry, rcu_head));
380 }
381 
382 static void prl_list_destroy_rcu(struct rcu_head *head)
383 {
384 	struct ip_tunnel_prl_entry *p, *n;
385 
386 	p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
387 	do {
388 		n = p->next;
389 		kfree(p);
390 		p = n;
391 	} while (p);
392 }
393 
394 static int
395 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
396 {
397 	struct ip_tunnel_prl_entry *x, **p;
398 	int err = 0;
399 
400 	spin_lock(&ipip6_prl_lock);
401 
402 	if (a && a->addr != htonl(INADDR_ANY)) {
403 		for (p = &t->prl; *p; p = &(*p)->next) {
404 			if ((*p)->addr == a->addr) {
405 				x = *p;
406 				*p = x->next;
407 				call_rcu(&x->rcu_head, prl_entry_destroy_rcu);
408 				t->prl_count--;
409 				goto out;
410 			}
411 		}
412 		err = -ENXIO;
413 	} else {
414 		if (t->prl) {
415 			t->prl_count = 0;
416 			x = t->prl;
417 			call_rcu(&x->rcu_head, prl_list_destroy_rcu);
418 			t->prl = NULL;
419 		}
420 	}
421 out:
422 	spin_unlock(&ipip6_prl_lock);
423 	return err;
424 }
425 
426 static int
427 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
428 {
429 	struct ip_tunnel_prl_entry *p;
430 	int ok = 1;
431 
432 	rcu_read_lock();
433 	p = __ipip6_tunnel_locate_prl(t, iph->saddr);
434 	if (p) {
435 		if (p->flags & PRL_DEFAULT)
436 			skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
437 		else
438 			skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
439 	} else {
440 		struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
441 		if (ipv6_addr_is_isatap(addr6) &&
442 		    (addr6->s6_addr32[3] == iph->saddr) &&
443 		    ipv6_chk_prefix(addr6, t->dev))
444 			skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
445 		else
446 			ok = 0;
447 	}
448 	rcu_read_unlock();
449 	return ok;
450 }
451 
452 static void ipip6_tunnel_uninit(struct net_device *dev)
453 {
454 	struct net *net = dev_net(dev);
455 	struct sit_net *sitn = net_generic(net, sit_net_id);
456 
457 	if (dev == sitn->fb_tunnel_dev) {
458 		spin_lock_bh(&ipip6_lock);
459 		sitn->tunnels_wc[0] = NULL;
460 		spin_unlock_bh(&ipip6_lock);
461 		dev_put(dev);
462 	} else {
463 		ipip6_tunnel_unlink(sitn, netdev_priv(dev));
464 		ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
465 		dev_put(dev);
466 	}
467 }
468 
469 
470 static int ipip6_err(struct sk_buff *skb, u32 info)
471 {
472 
473 /* All the routers (except for Linux) return only
474    8 bytes of packet payload. It means, that precise relaying of
475    ICMP in the real Internet is absolutely infeasible.
476  */
477 	struct iphdr *iph = (struct iphdr*)skb->data;
478 	const int type = icmp_hdr(skb)->type;
479 	const int code = icmp_hdr(skb)->code;
480 	struct ip_tunnel *t;
481 	int err;
482 
483 	switch (type) {
484 	default:
485 	case ICMP_PARAMETERPROB:
486 		return 0;
487 
488 	case ICMP_DEST_UNREACH:
489 		switch (code) {
490 		case ICMP_SR_FAILED:
491 		case ICMP_PORT_UNREACH:
492 			/* Impossible event. */
493 			return 0;
494 		case ICMP_FRAG_NEEDED:
495 			/* Soft state for pmtu is maintained by IP core. */
496 			return 0;
497 		default:
498 			/* All others are translated to HOST_UNREACH.
499 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
500 			   I believe they are just ether pollution. --ANK
501 			 */
502 			break;
503 		}
504 		break;
505 	case ICMP_TIME_EXCEEDED:
506 		if (code != ICMP_EXC_TTL)
507 			return 0;
508 		break;
509 	}
510 
511 	err = -ENOENT;
512 
513 	rcu_read_lock();
514 	t = ipip6_tunnel_lookup(dev_net(skb->dev),
515 				skb->dev,
516 				iph->daddr,
517 				iph->saddr);
518 	if (t == NULL || t->parms.iph.daddr == 0)
519 		goto out;
520 
521 	err = 0;
522 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
523 		goto out;
524 
525 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
526 		t->err_count++;
527 	else
528 		t->err_count = 1;
529 	t->err_time = jiffies;
530 out:
531 	rcu_read_unlock();
532 	return err;
533 }
534 
535 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
536 {
537 	if (INET_ECN_is_ce(iph->tos))
538 		IP6_ECN_set_ce(ipv6_hdr(skb));
539 }
540 
541 static int ipip6_rcv(struct sk_buff *skb)
542 {
543 	struct iphdr *iph;
544 	struct ip_tunnel *tunnel;
545 
546 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
547 		goto out;
548 
549 	iph = ip_hdr(skb);
550 
551 	rcu_read_lock();
552 	tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
553 				     iph->saddr, iph->daddr);
554 	if (tunnel != NULL) {
555 		secpath_reset(skb);
556 		skb->mac_header = skb->network_header;
557 		skb_reset_network_header(skb);
558 		IPCB(skb)->flags = 0;
559 		skb->protocol = htons(ETH_P_IPV6);
560 		skb->pkt_type = PACKET_HOST;
561 
562 		if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
563 		    !isatap_chksrc(skb, iph, tunnel)) {
564 			tunnel->dev->stats.rx_errors++;
565 			rcu_read_unlock();
566 			kfree_skb(skb);
567 			return 0;
568 		}
569 		tunnel->dev->stats.rx_packets++;
570 		tunnel->dev->stats.rx_bytes += skb->len;
571 		skb->dev = tunnel->dev;
572 		skb_dst_drop(skb);
573 		nf_reset(skb);
574 		ipip6_ecn_decapsulate(iph, skb);
575 		netif_rx(skb);
576 		rcu_read_unlock();
577 		return 0;
578 	}
579 
580 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
581 	rcu_read_unlock();
582 out:
583 	kfree_skb(skb);
584 	return 0;
585 }
586 
587 /*
588  * Returns the embedded IPv4 address if the IPv6 address
589  * comes from 6rd / 6to4 (RFC 3056) addr space.
590  */
591 static inline
592 __be32 try_6rd(struct in6_addr *v6dst, struct ip_tunnel *tunnel)
593 {
594 	__be32 dst = 0;
595 
596 #ifdef CONFIG_IPV6_SIT_6RD
597 	if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
598 			      tunnel->ip6rd.prefixlen)) {
599 		unsigned pbw0, pbi0;
600 		int pbi1;
601 		u32 d;
602 
603 		pbw0 = tunnel->ip6rd.prefixlen >> 5;
604 		pbi0 = tunnel->ip6rd.prefixlen & 0x1f;
605 
606 		d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
607 		    tunnel->ip6rd.relay_prefixlen;
608 
609 		pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
610 		if (pbi1 > 0)
611 			d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
612 			     (32 - pbi1);
613 
614 		dst = tunnel->ip6rd.relay_prefix | htonl(d);
615 	}
616 #else
617 	if (v6dst->s6_addr16[0] == htons(0x2002)) {
618 		/* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
619 		memcpy(&dst, &v6dst->s6_addr16[1], 4);
620 	}
621 #endif
622 	return dst;
623 }
624 
625 /*
626  *	This function assumes it is being called from dev_queue_xmit()
627  *	and that skb is filled properly by that function.
628  */
629 
630 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
631 				     struct net_device *dev)
632 {
633 	struct ip_tunnel *tunnel = netdev_priv(dev);
634 	struct net_device_stats *stats = &dev->stats;
635 	struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
636 	struct iphdr  *tiph = &tunnel->parms.iph;
637 	struct ipv6hdr *iph6 = ipv6_hdr(skb);
638 	u8     tos = tunnel->parms.iph.tos;
639 	__be16 df = tiph->frag_off;
640 	struct rtable *rt;     			/* Route to the other host */
641 	struct net_device *tdev;			/* Device to other host */
642 	struct iphdr  *iph;			/* Our new IP header */
643 	unsigned int max_headroom;		/* The extra header space needed */
644 	__be32 dst = tiph->daddr;
645 	int    mtu;
646 	struct in6_addr *addr6;
647 	int addr_type;
648 
649 	if (skb->protocol != htons(ETH_P_IPV6))
650 		goto tx_error;
651 
652 	/* ISATAP (RFC4214) - must come before 6to4 */
653 	if (dev->priv_flags & IFF_ISATAP) {
654 		struct neighbour *neigh = NULL;
655 
656 		if (skb_dst(skb))
657 			neigh = skb_dst(skb)->neighbour;
658 
659 		if (neigh == NULL) {
660 			if (net_ratelimit())
661 				printk(KERN_DEBUG "sit: nexthop == NULL\n");
662 			goto tx_error;
663 		}
664 
665 		addr6 = (struct in6_addr*)&neigh->primary_key;
666 		addr_type = ipv6_addr_type(addr6);
667 
668 		if ((addr_type & IPV6_ADDR_UNICAST) &&
669 		     ipv6_addr_is_isatap(addr6))
670 			dst = addr6->s6_addr32[3];
671 		else
672 			goto tx_error;
673 	}
674 
675 	if (!dst)
676 		dst = try_6rd(&iph6->daddr, tunnel);
677 
678 	if (!dst) {
679 		struct neighbour *neigh = NULL;
680 
681 		if (skb_dst(skb))
682 			neigh = skb_dst(skb)->neighbour;
683 
684 		if (neigh == NULL) {
685 			if (net_ratelimit())
686 				printk(KERN_DEBUG "sit: nexthop == NULL\n");
687 			goto tx_error;
688 		}
689 
690 		addr6 = (struct in6_addr*)&neigh->primary_key;
691 		addr_type = ipv6_addr_type(addr6);
692 
693 		if (addr_type == IPV6_ADDR_ANY) {
694 			addr6 = &ipv6_hdr(skb)->daddr;
695 			addr_type = ipv6_addr_type(addr6);
696 		}
697 
698 		if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
699 			goto tx_error_icmp;
700 
701 		dst = addr6->s6_addr32[3];
702 	}
703 
704 	{
705 		struct flowi fl = { .nl_u = { .ip4_u =
706 					      { .daddr = dst,
707 						.saddr = tiph->saddr,
708 						.tos = RT_TOS(tos) } },
709 				    .oif = tunnel->parms.link,
710 				    .proto = IPPROTO_IPV6 };
711 		if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
712 			stats->tx_carrier_errors++;
713 			goto tx_error_icmp;
714 		}
715 	}
716 	if (rt->rt_type != RTN_UNICAST) {
717 		ip_rt_put(rt);
718 		stats->tx_carrier_errors++;
719 		goto tx_error_icmp;
720 	}
721 	tdev = rt->u.dst.dev;
722 
723 	if (tdev == dev) {
724 		ip_rt_put(rt);
725 		stats->collisions++;
726 		goto tx_error;
727 	}
728 
729 	if (df) {
730 		mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
731 
732 		if (mtu < 68) {
733 			stats->collisions++;
734 			ip_rt_put(rt);
735 			goto tx_error;
736 		}
737 
738 		if (mtu < IPV6_MIN_MTU) {
739 			mtu = IPV6_MIN_MTU;
740 			df = 0;
741 		}
742 
743 		if (tunnel->parms.iph.daddr && skb_dst(skb))
744 			skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
745 
746 		if (skb->len > mtu) {
747 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
748 			ip_rt_put(rt);
749 			goto tx_error;
750 		}
751 	}
752 
753 	if (tunnel->err_count > 0) {
754 		if (time_before(jiffies,
755 				tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
756 			tunnel->err_count--;
757 			dst_link_failure(skb);
758 		} else
759 			tunnel->err_count = 0;
760 	}
761 
762 	/*
763 	 * Okay, now see if we can stuff it in the buffer as-is.
764 	 */
765 	max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
766 
767 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
768 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
769 		struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
770 		if (!new_skb) {
771 			ip_rt_put(rt);
772 			txq->tx_dropped++;
773 			dev_kfree_skb(skb);
774 			return NETDEV_TX_OK;
775 		}
776 		if (skb->sk)
777 			skb_set_owner_w(new_skb, skb->sk);
778 		dev_kfree_skb(skb);
779 		skb = new_skb;
780 		iph6 = ipv6_hdr(skb);
781 	}
782 
783 	skb->transport_header = skb->network_header;
784 	skb_push(skb, sizeof(struct iphdr));
785 	skb_reset_network_header(skb);
786 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
787 	IPCB(skb)->flags = 0;
788 	skb_dst_drop(skb);
789 	skb_dst_set(skb, &rt->u.dst);
790 
791 	/*
792 	 *	Push down and install the IPIP header.
793 	 */
794 
795 	iph 			=	ip_hdr(skb);
796 	iph->version		=	4;
797 	iph->ihl		=	sizeof(struct iphdr)>>2;
798 	iph->frag_off		=	df;
799 	iph->protocol		=	IPPROTO_IPV6;
800 	iph->tos		=	INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
801 	iph->daddr		=	rt->rt_dst;
802 	iph->saddr		=	rt->rt_src;
803 
804 	if ((iph->ttl = tiph->ttl) == 0)
805 		iph->ttl	=	iph6->hop_limit;
806 
807 	nf_reset(skb);
808 
809 	IPTUNNEL_XMIT();
810 	return NETDEV_TX_OK;
811 
812 tx_error_icmp:
813 	dst_link_failure(skb);
814 tx_error:
815 	stats->tx_errors++;
816 	dev_kfree_skb(skb);
817 	return NETDEV_TX_OK;
818 }
819 
820 static void ipip6_tunnel_bind_dev(struct net_device *dev)
821 {
822 	struct net_device *tdev = NULL;
823 	struct ip_tunnel *tunnel;
824 	struct iphdr *iph;
825 
826 	tunnel = netdev_priv(dev);
827 	iph = &tunnel->parms.iph;
828 
829 	if (iph->daddr) {
830 		struct flowi fl = { .nl_u = { .ip4_u =
831 					      { .daddr = iph->daddr,
832 						.saddr = iph->saddr,
833 						.tos = RT_TOS(iph->tos) } },
834 				    .oif = tunnel->parms.link,
835 				    .proto = IPPROTO_IPV6 };
836 		struct rtable *rt;
837 		if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
838 			tdev = rt->u.dst.dev;
839 			ip_rt_put(rt);
840 		}
841 		dev->flags |= IFF_POINTOPOINT;
842 	}
843 
844 	if (!tdev && tunnel->parms.link)
845 		tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
846 
847 	if (tdev) {
848 		dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
849 		dev->mtu = tdev->mtu - sizeof(struct iphdr);
850 		if (dev->mtu < IPV6_MIN_MTU)
851 			dev->mtu = IPV6_MIN_MTU;
852 	}
853 	dev->iflink = tunnel->parms.link;
854 }
855 
856 static int
857 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
858 {
859 	int err = 0;
860 	struct ip_tunnel_parm p;
861 	struct ip_tunnel_prl prl;
862 	struct ip_tunnel *t;
863 	struct net *net = dev_net(dev);
864 	struct sit_net *sitn = net_generic(net, sit_net_id);
865 #ifdef CONFIG_IPV6_SIT_6RD
866 	struct ip_tunnel_6rd ip6rd;
867 #endif
868 
869 	switch (cmd) {
870 	case SIOCGETTUNNEL:
871 #ifdef CONFIG_IPV6_SIT_6RD
872 	case SIOCGET6RD:
873 #endif
874 		t = NULL;
875 		if (dev == sitn->fb_tunnel_dev) {
876 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
877 				err = -EFAULT;
878 				break;
879 			}
880 			t = ipip6_tunnel_locate(net, &p, 0);
881 		}
882 		if (t == NULL)
883 			t = netdev_priv(dev);
884 
885 		err = -EFAULT;
886 		if (cmd == SIOCGETTUNNEL) {
887 			memcpy(&p, &t->parms, sizeof(p));
888 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
889 					 sizeof(p)))
890 				goto done;
891 #ifdef CONFIG_IPV6_SIT_6RD
892 		} else {
893 			ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
894 			ip6rd.relay_prefix = t->ip6rd.relay_prefix;
895 			ip6rd.prefixlen = t->ip6rd.prefixlen;
896 			ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
897 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
898 					 sizeof(ip6rd)))
899 				goto done;
900 #endif
901 		}
902 		err = 0;
903 		break;
904 
905 	case SIOCADDTUNNEL:
906 	case SIOCCHGTUNNEL:
907 		err = -EPERM;
908 		if (!capable(CAP_NET_ADMIN))
909 			goto done;
910 
911 		err = -EFAULT;
912 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
913 			goto done;
914 
915 		err = -EINVAL;
916 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
917 		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
918 			goto done;
919 		if (p.iph.ttl)
920 			p.iph.frag_off |= htons(IP_DF);
921 
922 		t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
923 
924 		if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
925 			if (t != NULL) {
926 				if (t->dev != dev) {
927 					err = -EEXIST;
928 					break;
929 				}
930 			} else {
931 				if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
932 				    (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
933 					err = -EINVAL;
934 					break;
935 				}
936 				t = netdev_priv(dev);
937 				ipip6_tunnel_unlink(sitn, t);
938 				t->parms.iph.saddr = p.iph.saddr;
939 				t->parms.iph.daddr = p.iph.daddr;
940 				memcpy(dev->dev_addr, &p.iph.saddr, 4);
941 				memcpy(dev->broadcast, &p.iph.daddr, 4);
942 				ipip6_tunnel_link(sitn, t);
943 				netdev_state_change(dev);
944 			}
945 		}
946 
947 		if (t) {
948 			err = 0;
949 			if (cmd == SIOCCHGTUNNEL) {
950 				t->parms.iph.ttl = p.iph.ttl;
951 				t->parms.iph.tos = p.iph.tos;
952 				if (t->parms.link != p.link) {
953 					t->parms.link = p.link;
954 					ipip6_tunnel_bind_dev(dev);
955 					netdev_state_change(dev);
956 				}
957 			}
958 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
959 				err = -EFAULT;
960 		} else
961 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
962 		break;
963 
964 	case SIOCDELTUNNEL:
965 		err = -EPERM;
966 		if (!capable(CAP_NET_ADMIN))
967 			goto done;
968 
969 		if (dev == sitn->fb_tunnel_dev) {
970 			err = -EFAULT;
971 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
972 				goto done;
973 			err = -ENOENT;
974 			if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
975 				goto done;
976 			err = -EPERM;
977 			if (t == netdev_priv(sitn->fb_tunnel_dev))
978 				goto done;
979 			dev = t->dev;
980 		}
981 		unregister_netdevice(dev);
982 		err = 0;
983 		break;
984 
985 	case SIOCGETPRL:
986 		err = -EINVAL;
987 		if (dev == sitn->fb_tunnel_dev)
988 			goto done;
989 		err = -ENOENT;
990 		if (!(t = netdev_priv(dev)))
991 			goto done;
992 		err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
993 		break;
994 
995 	case SIOCADDPRL:
996 	case SIOCDELPRL:
997 	case SIOCCHGPRL:
998 		err = -EPERM;
999 		if (!capable(CAP_NET_ADMIN))
1000 			goto done;
1001 		err = -EINVAL;
1002 		if (dev == sitn->fb_tunnel_dev)
1003 			goto done;
1004 		err = -EFAULT;
1005 		if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
1006 			goto done;
1007 		err = -ENOENT;
1008 		if (!(t = netdev_priv(dev)))
1009 			goto done;
1010 
1011 		switch (cmd) {
1012 		case SIOCDELPRL:
1013 			err = ipip6_tunnel_del_prl(t, &prl);
1014 			break;
1015 		case SIOCADDPRL:
1016 		case SIOCCHGPRL:
1017 			err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1018 			break;
1019 		}
1020 		netdev_state_change(dev);
1021 		break;
1022 
1023 #ifdef CONFIG_IPV6_SIT_6RD
1024 	case SIOCADD6RD:
1025 	case SIOCCHG6RD:
1026 	case SIOCDEL6RD:
1027 		err = -EPERM;
1028 		if (!capable(CAP_NET_ADMIN))
1029 			goto done;
1030 
1031 		err = -EFAULT;
1032 		if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
1033 				   sizeof(ip6rd)))
1034 			goto done;
1035 
1036 		t = netdev_priv(dev);
1037 
1038 		if (cmd != SIOCDEL6RD) {
1039 			struct in6_addr prefix;
1040 			__be32 relay_prefix;
1041 
1042 			err = -EINVAL;
1043 			if (ip6rd.relay_prefixlen > 32 ||
1044 			    ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
1045 				goto done;
1046 
1047 			ipv6_addr_prefix(&prefix, &ip6rd.prefix,
1048 					 ip6rd.prefixlen);
1049 			if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
1050 				goto done;
1051 			if (ip6rd.relay_prefixlen)
1052 				relay_prefix = ip6rd.relay_prefix &
1053 					       htonl(0xffffffffUL <<
1054 						     (32 - ip6rd.relay_prefixlen));
1055 			else
1056 				relay_prefix = 0;
1057 			if (relay_prefix != ip6rd.relay_prefix)
1058 				goto done;
1059 
1060 			ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
1061 			t->ip6rd.relay_prefix = relay_prefix;
1062 			t->ip6rd.prefixlen = ip6rd.prefixlen;
1063 			t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
1064 		} else
1065 			ipip6_tunnel_clone_6rd(dev, sitn);
1066 
1067 		err = 0;
1068 		break;
1069 #endif
1070 
1071 	default:
1072 		err = -EINVAL;
1073 	}
1074 
1075 done:
1076 	return err;
1077 }
1078 
1079 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1080 {
1081 	if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1082 		return -EINVAL;
1083 	dev->mtu = new_mtu;
1084 	return 0;
1085 }
1086 
1087 static const struct net_device_ops ipip6_netdev_ops = {
1088 	.ndo_uninit	= ipip6_tunnel_uninit,
1089 	.ndo_start_xmit	= ipip6_tunnel_xmit,
1090 	.ndo_do_ioctl	= ipip6_tunnel_ioctl,
1091 	.ndo_change_mtu	= ipip6_tunnel_change_mtu,
1092 };
1093 
1094 static void ipip6_tunnel_setup(struct net_device *dev)
1095 {
1096 	dev->netdev_ops		= &ipip6_netdev_ops;
1097 	dev->destructor 	= free_netdev;
1098 
1099 	dev->type		= ARPHRD_SIT;
1100 	dev->hard_header_len 	= LL_MAX_HEADER + sizeof(struct iphdr);
1101 	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr);
1102 	dev->flags		= IFF_NOARP;
1103 	dev->priv_flags	       &= ~IFF_XMIT_DST_RELEASE;
1104 	dev->iflink		= 0;
1105 	dev->addr_len		= 4;
1106 	dev->features		|= NETIF_F_NETNS_LOCAL;
1107 }
1108 
1109 static void ipip6_tunnel_init(struct net_device *dev)
1110 {
1111 	struct ip_tunnel *tunnel = netdev_priv(dev);
1112 
1113 	tunnel->dev = dev;
1114 	strcpy(tunnel->parms.name, dev->name);
1115 
1116 	memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1117 	memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1118 
1119 	ipip6_tunnel_bind_dev(dev);
1120 }
1121 
1122 static void __net_init ipip6_fb_tunnel_init(struct net_device *dev)
1123 {
1124 	struct ip_tunnel *tunnel = netdev_priv(dev);
1125 	struct iphdr *iph = &tunnel->parms.iph;
1126 	struct net *net = dev_net(dev);
1127 	struct sit_net *sitn = net_generic(net, sit_net_id);
1128 
1129 	tunnel->dev = dev;
1130 	strcpy(tunnel->parms.name, dev->name);
1131 
1132 	iph->version		= 4;
1133 	iph->protocol		= IPPROTO_IPV6;
1134 	iph->ihl		= 5;
1135 	iph->ttl		= 64;
1136 
1137 	dev_hold(dev);
1138 	sitn->tunnels_wc[0]	= tunnel;
1139 }
1140 
1141 static struct xfrm_tunnel sit_handler = {
1142 	.handler	=	ipip6_rcv,
1143 	.err_handler	=	ipip6_err,
1144 	.priority	=	1,
1145 };
1146 
1147 static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1148 {
1149 	int prio;
1150 
1151 	for (prio = 1; prio < 4; prio++) {
1152 		int h;
1153 		for (h = 0; h < HASH_SIZE; h++) {
1154 			struct ip_tunnel *t = sitn->tunnels[prio][h];
1155 
1156 			while (t != NULL) {
1157 				unregister_netdevice_queue(t->dev, head);
1158 				t = t->next;
1159 			}
1160 		}
1161 	}
1162 }
1163 
1164 static int __net_init sit_init_net(struct net *net)
1165 {
1166 	struct sit_net *sitn = net_generic(net, sit_net_id);
1167 	int err;
1168 
1169 	sitn->tunnels[0] = sitn->tunnels_wc;
1170 	sitn->tunnels[1] = sitn->tunnels_l;
1171 	sitn->tunnels[2] = sitn->tunnels_r;
1172 	sitn->tunnels[3] = sitn->tunnels_r_l;
1173 
1174 	sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1175 					   ipip6_tunnel_setup);
1176 	if (!sitn->fb_tunnel_dev) {
1177 		err = -ENOMEM;
1178 		goto err_alloc_dev;
1179 	}
1180 	dev_net_set(sitn->fb_tunnel_dev, net);
1181 
1182 	ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1183 	ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1184 
1185 	if ((err = register_netdev(sitn->fb_tunnel_dev)))
1186 		goto err_reg_dev;
1187 
1188 	return 0;
1189 
1190 err_reg_dev:
1191 	dev_put(sitn->fb_tunnel_dev);
1192 	free_netdev(sitn->fb_tunnel_dev);
1193 err_alloc_dev:
1194 	return err;
1195 }
1196 
1197 static void __net_exit sit_exit_net(struct net *net)
1198 {
1199 	struct sit_net *sitn = net_generic(net, sit_net_id);
1200 	LIST_HEAD(list);
1201 
1202 	rtnl_lock();
1203 	sit_destroy_tunnels(sitn, &list);
1204 	unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
1205 	unregister_netdevice_many(&list);
1206 	rtnl_unlock();
1207 }
1208 
1209 static struct pernet_operations sit_net_ops = {
1210 	.init = sit_init_net,
1211 	.exit = sit_exit_net,
1212 	.id   = &sit_net_id,
1213 	.size = sizeof(struct sit_net),
1214 };
1215 
1216 static void __exit sit_cleanup(void)
1217 {
1218 	xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1219 
1220 	unregister_pernet_device(&sit_net_ops);
1221 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
1222 }
1223 
1224 static int __init sit_init(void)
1225 {
1226 	int err;
1227 
1228 	printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1229 
1230 	err = register_pernet_device(&sit_net_ops);
1231 	if (err < 0)
1232 		return err;
1233 	err = xfrm4_tunnel_register(&sit_handler, AF_INET6);
1234 	if (err < 0) {
1235 		unregister_pernet_device(&sit_net_ops);
1236 		printk(KERN_INFO "sit init: Can't add protocol\n");
1237 	}
1238 	return err;
1239 }
1240 
1241 module_init(sit_init);
1242 module_exit(sit_cleanup);
1243 MODULE_LICENSE("GPL");
1244 MODULE_ALIAS("sit0");
1245