xref: /openbmc/linux/net/ipv6/sit.c (revision fd589a8f)
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  * Sascha Hlusiak <mail@saschahlusiak.de>:	stateless autoconf for isatap
19  */
20 
21 #include <linux/module.h>
22 #include <linux/capability.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/socket.h>
26 #include <linux/sockios.h>
27 #include <linux/net.h>
28 #include <linux/in6.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/icmp.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_fb_tunnel_init(struct net_device *dev);
67 static void ipip6_tunnel_init(struct net_device *dev);
68 static void ipip6_tunnel_setup(struct net_device *dev);
69 
70 static int sit_net_id;
71 struct sit_net {
72 	struct ip_tunnel *tunnels_r_l[HASH_SIZE];
73 	struct ip_tunnel *tunnels_r[HASH_SIZE];
74 	struct ip_tunnel *tunnels_l[HASH_SIZE];
75 	struct ip_tunnel *tunnels_wc[1];
76 	struct ip_tunnel **tunnels[4];
77 
78 	struct net_device *fb_tunnel_dev;
79 };
80 
81 static DEFINE_RWLOCK(ipip6_lock);
82 
83 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
84 		struct net_device *dev, __be32 remote, __be32 local)
85 {
86 	unsigned h0 = HASH(remote);
87 	unsigned h1 = HASH(local);
88 	struct ip_tunnel *t;
89 	struct sit_net *sitn = net_generic(net, sit_net_id);
90 
91 	for (t = sitn->tunnels_r_l[h0^h1]; t; t = t->next) {
92 		if (local == t->parms.iph.saddr &&
93 		    remote == t->parms.iph.daddr &&
94 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
95 		    (t->dev->flags & IFF_UP))
96 			return t;
97 	}
98 	for (t = sitn->tunnels_r[h0]; t; t = t->next) {
99 		if (remote == t->parms.iph.daddr &&
100 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
101 		    (t->dev->flags & IFF_UP))
102 			return t;
103 	}
104 	for (t = sitn->tunnels_l[h1]; t; t = t->next) {
105 		if (local == t->parms.iph.saddr &&
106 		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
107 		    (t->dev->flags & IFF_UP))
108 			return t;
109 	}
110 	t = sitn->tunnels_wc[0];
111 	if ((t != NULL) && (t->dev->flags & IFF_UP))
112 		return t;
113 	return NULL;
114 }
115 
116 static struct ip_tunnel **__ipip6_bucket(struct sit_net *sitn,
117 		struct ip_tunnel_parm *parms)
118 {
119 	__be32 remote = parms->iph.daddr;
120 	__be32 local = parms->iph.saddr;
121 	unsigned h = 0;
122 	int prio = 0;
123 
124 	if (remote) {
125 		prio |= 2;
126 		h ^= HASH(remote);
127 	}
128 	if (local) {
129 		prio |= 1;
130 		h ^= HASH(local);
131 	}
132 	return &sitn->tunnels[prio][h];
133 }
134 
135 static inline struct ip_tunnel **ipip6_bucket(struct sit_net *sitn,
136 		struct ip_tunnel *t)
137 {
138 	return __ipip6_bucket(sitn, &t->parms);
139 }
140 
141 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
142 {
143 	struct ip_tunnel **tp;
144 
145 	for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
146 		if (t == *tp) {
147 			write_lock_bh(&ipip6_lock);
148 			*tp = t->next;
149 			write_unlock_bh(&ipip6_lock);
150 			break;
151 		}
152 	}
153 }
154 
155 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
156 {
157 	struct ip_tunnel **tp = ipip6_bucket(sitn, t);
158 
159 	t->next = *tp;
160 	write_lock_bh(&ipip6_lock);
161 	*tp = t;
162 	write_unlock_bh(&ipip6_lock);
163 }
164 
165 static struct ip_tunnel * ipip6_tunnel_locate(struct net *net,
166 		struct ip_tunnel_parm *parms, int create)
167 {
168 	__be32 remote = parms->iph.daddr;
169 	__be32 local = parms->iph.saddr;
170 	struct ip_tunnel *t, **tp, *nt;
171 	struct net_device *dev;
172 	char name[IFNAMSIZ];
173 	struct sit_net *sitn = net_generic(net, sit_net_id);
174 
175 	for (tp = __ipip6_bucket(sitn, parms); (t = *tp) != NULL; tp = &t->next) {
176 		if (local == t->parms.iph.saddr &&
177 		    remote == t->parms.iph.daddr &&
178 		    parms->link == t->parms.link) {
179 			if (create)
180 				return NULL;
181 			else
182 				return t;
183 		}
184 	}
185 	if (!create)
186 		goto failed;
187 
188 	if (parms->name[0])
189 		strlcpy(name, parms->name, IFNAMSIZ);
190 	else
191 		sprintf(name, "sit%%d");
192 
193 	dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
194 	if (dev == NULL)
195 		return NULL;
196 
197 	dev_net_set(dev, net);
198 
199 	if (strchr(name, '%')) {
200 		if (dev_alloc_name(dev, name) < 0)
201 			goto failed_free;
202 	}
203 
204 	nt = netdev_priv(dev);
205 
206 	nt->parms = *parms;
207 	ipip6_tunnel_init(dev);
208 
209 	if (parms->i_flags & SIT_ISATAP)
210 		dev->priv_flags |= IFF_ISATAP;
211 
212 	if (register_netdevice(dev) < 0)
213 		goto failed_free;
214 
215 	dev_hold(dev);
216 
217 	ipip6_tunnel_link(sitn, nt);
218 	return nt;
219 
220 failed_free:
221 	free_netdev(dev);
222 failed:
223 	return NULL;
224 }
225 
226 static void ipip6_tunnel_rs_timer(unsigned long data)
227 {
228 	struct ip_tunnel_prl_entry *p = (struct ip_tunnel_prl_entry *) data;
229 	struct inet6_dev *ifp;
230 	struct inet6_ifaddr *addr;
231 
232 	spin_lock(&p->lock);
233 	ifp = __in6_dev_get(p->tunnel->dev);
234 
235 	read_lock_bh(&ifp->lock);
236 	for (addr = ifp->addr_list; addr; addr = addr->if_next) {
237 		struct in6_addr rtr;
238 
239 		if (!(ipv6_addr_type(&addr->addr) & IPV6_ADDR_LINKLOCAL))
240 			continue;
241 
242 		/* Send RS to guessed linklocal address of router
243 		 *
244 		 * Better: send to ff02::2 encapsuled in unicast directly
245 		 * to router-v4 instead of guessing the v6 address.
246 		 *
247 		 * Cisco/Windows seem to not set the u/l bit correctly,
248 		 * so we won't guess right.
249 		 */
250 		ipv6_addr_set(&rtr,  htonl(0xFE800000), 0, 0, 0);
251 		if (!__ipv6_isatap_ifid(rtr.s6_addr + 8,
252 					p->addr)) {
253 			ndisc_send_rs(p->tunnel->dev, &addr->addr, &rtr);
254 		}
255 	}
256 	read_unlock_bh(&ifp->lock);
257 
258 	mod_timer(&p->rs_timer, jiffies + HZ * p->rs_delay);
259 	spin_unlock(&p->lock);
260 
261 	return;
262 }
263 
264 static struct ip_tunnel_prl_entry *
265 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
266 {
267 	struct ip_tunnel_prl_entry *p = (struct ip_tunnel_prl_entry *)NULL;
268 
269 	for (p = t->prl; p; p = p->next)
270 		if (p->addr == addr)
271 			break;
272 	return p;
273 
274 }
275 
276 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
277 				struct ip_tunnel_prl __user *a)
278 {
279 	struct ip_tunnel_prl kprl, *kp;
280 	struct ip_tunnel_prl_entry *prl;
281 	unsigned int cmax, c = 0, ca, len;
282 	int ret = 0;
283 
284 	if (copy_from_user(&kprl, a, sizeof(kprl)))
285 		return -EFAULT;
286 	cmax = kprl.datalen / sizeof(kprl);
287 	if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
288 		cmax = 1;
289 
290 	/* For simple GET or for root users,
291 	 * we try harder to allocate.
292 	 */
293 	kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
294 		kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
295 		NULL;
296 
297 	read_lock(&ipip6_lock);
298 
299 	ca = t->prl_count < cmax ? t->prl_count : cmax;
300 
301 	if (!kp) {
302 		/* We don't try hard to allocate much memory for
303 		 * non-root users.
304 		 * For root users, retry allocating enough memory for
305 		 * the answer.
306 		 */
307 		kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
308 		if (!kp) {
309 			ret = -ENOMEM;
310 			goto out;
311 		}
312 	}
313 
314 	c = 0;
315 	for (prl = t->prl; prl; prl = prl->next) {
316 		if (c > cmax)
317 			break;
318 		if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
319 			continue;
320 		kp[c].addr = prl->addr;
321 		kp[c].flags = prl->flags;
322 		kp[c].rs_delay = prl->rs_delay;
323 		c++;
324 		if (kprl.addr != htonl(INADDR_ANY))
325 			break;
326 	}
327 out:
328 	read_unlock(&ipip6_lock);
329 
330 	len = sizeof(*kp) * c;
331 	ret = 0;
332 	if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
333 		ret = -EFAULT;
334 
335 	kfree(kp);
336 
337 	return ret;
338 }
339 
340 static int
341 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
342 {
343 	struct ip_tunnel_prl_entry *p;
344 	int err = 0;
345 
346 	if (a->addr == htonl(INADDR_ANY))
347 		return -EINVAL;
348 
349 	write_lock(&ipip6_lock);
350 
351 	for (p = t->prl; p; p = p->next) {
352 		if (p->addr == a->addr) {
353 			if (chg)
354 				goto update;
355 			err = -EEXIST;
356 			goto out;
357 		}
358 	}
359 
360 	if (chg) {
361 		err = -ENXIO;
362 		goto out;
363 	}
364 
365 	p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
366 	if (!p) {
367 		err = -ENOBUFS;
368 		goto out;
369 	}
370 
371 	p->next = t->prl;
372 	p->tunnel = t;
373 	t->prl = p;
374 	t->prl_count++;
375 
376 	spin_lock_init(&p->lock);
377 	setup_timer(&p->rs_timer, ipip6_tunnel_rs_timer, (unsigned long) p);
378 update:
379 	p->addr = a->addr;
380 	p->flags = a->flags;
381 	p->rs_delay = a->rs_delay;
382 	if (p->rs_delay == 0)
383 		p->rs_delay = IPTUNNEL_RS_DEFAULT_DELAY;
384 	spin_lock(&p->lock);
385 	del_timer(&p->rs_timer);
386 	if (p->flags & PRL_DEFAULT)
387 		mod_timer(&p->rs_timer, jiffies + 1);
388 	spin_unlock(&p->lock);
389 out:
390 	write_unlock(&ipip6_lock);
391 	return err;
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 	write_lock(&ipip6_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 				spin_lock(&x->lock);
408 				del_timer(&x->rs_timer);
409 				spin_unlock(&x->lock);
410 				kfree(x);
411 				t->prl_count--;
412 				goto out;
413 			}
414 		}
415 		err = -ENXIO;
416 	} else {
417 		while (t->prl) {
418 			x = t->prl;
419 			t->prl = t->prl->next;
420 			spin_lock(&x->lock);
421 			del_timer(&x->rs_timer);
422 			spin_unlock(&x->lock);
423 			kfree(x);
424 			t->prl_count--;
425 		}
426 	}
427 out:
428 	write_unlock(&ipip6_lock);
429 	return err;
430 }
431 
432 static int
433 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
434 {
435 	struct ip_tunnel_prl_entry *p;
436 	int ok = 1;
437 
438 	read_lock(&ipip6_lock);
439 	p = __ipip6_tunnel_locate_prl(t, iph->saddr);
440 	if (p) {
441 		if (p->flags & PRL_DEFAULT)
442 			skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
443 		else
444 			skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
445 	} else {
446 		struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
447 		if (ipv6_addr_is_isatap(addr6) &&
448 		    (addr6->s6_addr32[3] == iph->saddr) &&
449 		    ipv6_chk_prefix(addr6, t->dev))
450 			skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
451 		else
452 			ok = 0;
453 	}
454 	read_unlock(&ipip6_lock);
455 	return ok;
456 }
457 
458 static void ipip6_tunnel_uninit(struct net_device *dev)
459 {
460 	struct net *net = dev_net(dev);
461 	struct sit_net *sitn = net_generic(net, sit_net_id);
462 
463 	if (dev == sitn->fb_tunnel_dev) {
464 		write_lock_bh(&ipip6_lock);
465 		sitn->tunnels_wc[0] = NULL;
466 		write_unlock_bh(&ipip6_lock);
467 		dev_put(dev);
468 	} else {
469 		ipip6_tunnel_unlink(sitn, netdev_priv(dev));
470 		ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
471 		dev_put(dev);
472 	}
473 }
474 
475 
476 static int ipip6_err(struct sk_buff *skb, u32 info)
477 {
478 
479 /* All the routers (except for Linux) return only
480    8 bytes of packet payload. It means, that precise relaying of
481    ICMP in the real Internet is absolutely infeasible.
482  */
483 	struct iphdr *iph = (struct iphdr*)skb->data;
484 	const int type = icmp_hdr(skb)->type;
485 	const int code = icmp_hdr(skb)->code;
486 	struct ip_tunnel *t;
487 	int err;
488 
489 	switch (type) {
490 	default:
491 	case ICMP_PARAMETERPROB:
492 		return 0;
493 
494 	case ICMP_DEST_UNREACH:
495 		switch (code) {
496 		case ICMP_SR_FAILED:
497 		case ICMP_PORT_UNREACH:
498 			/* Impossible event. */
499 			return 0;
500 		case ICMP_FRAG_NEEDED:
501 			/* Soft state for pmtu is maintained by IP core. */
502 			return 0;
503 		default:
504 			/* All others are translated to HOST_UNREACH.
505 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
506 			   I believe they are just ether pollution. --ANK
507 			 */
508 			break;
509 		}
510 		break;
511 	case ICMP_TIME_EXCEEDED:
512 		if (code != ICMP_EXC_TTL)
513 			return 0;
514 		break;
515 	}
516 
517 	err = -ENOENT;
518 
519 	read_lock(&ipip6_lock);
520 	t = ipip6_tunnel_lookup(dev_net(skb->dev),
521 				skb->dev,
522 				iph->daddr,
523 				iph->saddr);
524 	if (t == NULL || t->parms.iph.daddr == 0)
525 		goto out;
526 
527 	err = 0;
528 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
529 		goto out;
530 
531 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
532 		t->err_count++;
533 	else
534 		t->err_count = 1;
535 	t->err_time = jiffies;
536 out:
537 	read_unlock(&ipip6_lock);
538 	return err;
539 }
540 
541 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
542 {
543 	if (INET_ECN_is_ce(iph->tos))
544 		IP6_ECN_set_ce(ipv6_hdr(skb));
545 }
546 
547 static int ipip6_rcv(struct sk_buff *skb)
548 {
549 	struct iphdr *iph;
550 	struct ip_tunnel *tunnel;
551 
552 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
553 		goto out;
554 
555 	iph = ip_hdr(skb);
556 
557 	read_lock(&ipip6_lock);
558 	tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
559 				     iph->saddr, iph->daddr);
560 	if (tunnel != NULL) {
561 		secpath_reset(skb);
562 		skb->mac_header = skb->network_header;
563 		skb_reset_network_header(skb);
564 		IPCB(skb)->flags = 0;
565 		skb->protocol = htons(ETH_P_IPV6);
566 		skb->pkt_type = PACKET_HOST;
567 
568 		if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
569 		    !isatap_chksrc(skb, iph, tunnel)) {
570 			tunnel->dev->stats.rx_errors++;
571 			read_unlock(&ipip6_lock);
572 			kfree_skb(skb);
573 			return 0;
574 		}
575 		tunnel->dev->stats.rx_packets++;
576 		tunnel->dev->stats.rx_bytes += skb->len;
577 		skb->dev = tunnel->dev;
578 		skb_dst_drop(skb);
579 		nf_reset(skb);
580 		ipip6_ecn_decapsulate(iph, skb);
581 		netif_rx(skb);
582 		read_unlock(&ipip6_lock);
583 		return 0;
584 	}
585 
586 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
587 	read_unlock(&ipip6_lock);
588 out:
589 	kfree_skb(skb);
590 	return 0;
591 }
592 
593 /* Returns the embedded IPv4 address if the IPv6 address
594    comes from 6to4 (RFC 3056) addr space */
595 
596 static inline __be32 try_6to4(struct in6_addr *v6dst)
597 {
598 	__be32 dst = 0;
599 
600 	if (v6dst->s6_addr16[0] == htons(0x2002)) {
601 		/* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
602 		memcpy(&dst, &v6dst->s6_addr16[1], 4);
603 	}
604 	return dst;
605 }
606 
607 /*
608  *	This function assumes it is being called from dev_queue_xmit()
609  *	and that skb is filled properly by that function.
610  */
611 
612 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
613 				     struct net_device *dev)
614 {
615 	struct ip_tunnel *tunnel = netdev_priv(dev);
616 	struct net_device_stats *stats = &tunnel->dev->stats;
617 	struct iphdr  *tiph = &tunnel->parms.iph;
618 	struct ipv6hdr *iph6 = ipv6_hdr(skb);
619 	u8     tos = tunnel->parms.iph.tos;
620 	struct rtable *rt;     			/* Route to the other host */
621 	struct net_device *tdev;			/* Device to other host */
622 	struct iphdr  *iph;			/* Our new IP header */
623 	unsigned int max_headroom;		/* The extra header space needed */
624 	__be32 dst = tiph->daddr;
625 	int    mtu;
626 	struct in6_addr *addr6;
627 	int addr_type;
628 
629 	if (tunnel->recursion++) {
630 		stats->collisions++;
631 		goto tx_error;
632 	}
633 
634 	if (skb->protocol != htons(ETH_P_IPV6))
635 		goto tx_error;
636 
637 	/* ISATAP (RFC4214) - must come before 6to4 */
638 	if (dev->priv_flags & IFF_ISATAP) {
639 		struct neighbour *neigh = NULL;
640 
641 		if (skb_dst(skb))
642 			neigh = skb_dst(skb)->neighbour;
643 
644 		if (neigh == NULL) {
645 			if (net_ratelimit())
646 				printk(KERN_DEBUG "sit: nexthop == NULL\n");
647 			goto tx_error;
648 		}
649 
650 		addr6 = (struct in6_addr*)&neigh->primary_key;
651 		addr_type = ipv6_addr_type(addr6);
652 
653 		if ((addr_type & IPV6_ADDR_UNICAST) &&
654 		     ipv6_addr_is_isatap(addr6))
655 			dst = addr6->s6_addr32[3];
656 		else
657 			goto tx_error;
658 	}
659 
660 	if (!dst)
661 		dst = try_6to4(&iph6->daddr);
662 
663 	if (!dst) {
664 		struct neighbour *neigh = NULL;
665 
666 		if (skb_dst(skb))
667 			neigh = skb_dst(skb)->neighbour;
668 
669 		if (neigh == NULL) {
670 			if (net_ratelimit())
671 				printk(KERN_DEBUG "sit: nexthop == NULL\n");
672 			goto tx_error;
673 		}
674 
675 		addr6 = (struct in6_addr*)&neigh->primary_key;
676 		addr_type = ipv6_addr_type(addr6);
677 
678 		if (addr_type == IPV6_ADDR_ANY) {
679 			addr6 = &ipv6_hdr(skb)->daddr;
680 			addr_type = ipv6_addr_type(addr6);
681 		}
682 
683 		if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
684 			goto tx_error_icmp;
685 
686 		dst = addr6->s6_addr32[3];
687 	}
688 
689 	{
690 		struct flowi fl = { .nl_u = { .ip4_u =
691 					      { .daddr = dst,
692 						.saddr = tiph->saddr,
693 						.tos = RT_TOS(tos) } },
694 				    .oif = tunnel->parms.link,
695 				    .proto = IPPROTO_IPV6 };
696 		if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
697 			stats->tx_carrier_errors++;
698 			goto tx_error_icmp;
699 		}
700 	}
701 	if (rt->rt_type != RTN_UNICAST) {
702 		ip_rt_put(rt);
703 		stats->tx_carrier_errors++;
704 		goto tx_error_icmp;
705 	}
706 	tdev = rt->u.dst.dev;
707 
708 	if (tdev == dev) {
709 		ip_rt_put(rt);
710 		stats->collisions++;
711 		goto tx_error;
712 	}
713 
714 	if (tiph->frag_off)
715 		mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
716 	else
717 		mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
718 
719 	if (mtu < 68) {
720 		stats->collisions++;
721 		ip_rt_put(rt);
722 		goto tx_error;
723 	}
724 	if (mtu < IPV6_MIN_MTU)
725 		mtu = IPV6_MIN_MTU;
726 	if (tunnel->parms.iph.daddr && skb_dst(skb))
727 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
728 
729 	if (skb->len > mtu) {
730 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
731 		ip_rt_put(rt);
732 		goto tx_error;
733 	}
734 
735 	if (tunnel->err_count > 0) {
736 		if (time_before(jiffies,
737 				tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
738 			tunnel->err_count--;
739 			dst_link_failure(skb);
740 		} else
741 			tunnel->err_count = 0;
742 	}
743 
744 	/*
745 	 * Okay, now see if we can stuff it in the buffer as-is.
746 	 */
747 	max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
748 
749 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
750 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
751 		struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
752 		if (!new_skb) {
753 			ip_rt_put(rt);
754 			stats->tx_dropped++;
755 			dev_kfree_skb(skb);
756 			tunnel->recursion--;
757 			return NETDEV_TX_OK;
758 		}
759 		if (skb->sk)
760 			skb_set_owner_w(new_skb, skb->sk);
761 		dev_kfree_skb(skb);
762 		skb = new_skb;
763 		iph6 = ipv6_hdr(skb);
764 	}
765 
766 	skb->transport_header = skb->network_header;
767 	skb_push(skb, sizeof(struct iphdr));
768 	skb_reset_network_header(skb);
769 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
770 	IPCB(skb)->flags = 0;
771 	skb_dst_drop(skb);
772 	skb_dst_set(skb, &rt->u.dst);
773 
774 	/*
775 	 *	Push down and install the IPIP header.
776 	 */
777 
778 	iph 			=	ip_hdr(skb);
779 	iph->version		=	4;
780 	iph->ihl		=	sizeof(struct iphdr)>>2;
781 	if (mtu > IPV6_MIN_MTU)
782 		iph->frag_off	=	tiph->frag_off;
783 	else
784 		iph->frag_off	=	0;
785 
786 	iph->protocol		=	IPPROTO_IPV6;
787 	iph->tos		=	INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
788 	iph->daddr		=	rt->rt_dst;
789 	iph->saddr		=	rt->rt_src;
790 
791 	if ((iph->ttl = tiph->ttl) == 0)
792 		iph->ttl	=	iph6->hop_limit;
793 
794 	nf_reset(skb);
795 
796 	IPTUNNEL_XMIT();
797 	tunnel->recursion--;
798 	return NETDEV_TX_OK;
799 
800 tx_error_icmp:
801 	dst_link_failure(skb);
802 tx_error:
803 	stats->tx_errors++;
804 	dev_kfree_skb(skb);
805 	tunnel->recursion--;
806 	return NETDEV_TX_OK;
807 }
808 
809 static void ipip6_tunnel_bind_dev(struct net_device *dev)
810 {
811 	struct net_device *tdev = NULL;
812 	struct ip_tunnel *tunnel;
813 	struct iphdr *iph;
814 
815 	tunnel = netdev_priv(dev);
816 	iph = &tunnel->parms.iph;
817 
818 	if (iph->daddr) {
819 		struct flowi fl = { .nl_u = { .ip4_u =
820 					      { .daddr = iph->daddr,
821 						.saddr = iph->saddr,
822 						.tos = RT_TOS(iph->tos) } },
823 				    .oif = tunnel->parms.link,
824 				    .proto = IPPROTO_IPV6 };
825 		struct rtable *rt;
826 		if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
827 			tdev = rt->u.dst.dev;
828 			ip_rt_put(rt);
829 		}
830 		dev->flags |= IFF_POINTOPOINT;
831 	}
832 
833 	if (!tdev && tunnel->parms.link)
834 		tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
835 
836 	if (tdev) {
837 		dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
838 		dev->mtu = tdev->mtu - sizeof(struct iphdr);
839 		if (dev->mtu < IPV6_MIN_MTU)
840 			dev->mtu = IPV6_MIN_MTU;
841 	}
842 	dev->iflink = tunnel->parms.link;
843 }
844 
845 static int
846 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
847 {
848 	int err = 0;
849 	struct ip_tunnel_parm p;
850 	struct ip_tunnel_prl prl;
851 	struct ip_tunnel *t;
852 	struct net *net = dev_net(dev);
853 	struct sit_net *sitn = net_generic(net, sit_net_id);
854 
855 	switch (cmd) {
856 	case SIOCGETTUNNEL:
857 		t = NULL;
858 		if (dev == sitn->fb_tunnel_dev) {
859 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
860 				err = -EFAULT;
861 				break;
862 			}
863 			t = ipip6_tunnel_locate(net, &p, 0);
864 		}
865 		if (t == NULL)
866 			t = netdev_priv(dev);
867 		memcpy(&p, &t->parms, sizeof(p));
868 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
869 			err = -EFAULT;
870 		break;
871 
872 	case SIOCADDTUNNEL:
873 	case SIOCCHGTUNNEL:
874 		err = -EPERM;
875 		if (!capable(CAP_NET_ADMIN))
876 			goto done;
877 
878 		err = -EFAULT;
879 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
880 			goto done;
881 
882 		err = -EINVAL;
883 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
884 		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
885 			goto done;
886 		if (p.iph.ttl)
887 			p.iph.frag_off |= htons(IP_DF);
888 
889 		t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
890 
891 		if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
892 			if (t != NULL) {
893 				if (t->dev != dev) {
894 					err = -EEXIST;
895 					break;
896 				}
897 			} else {
898 				if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
899 				    (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
900 					err = -EINVAL;
901 					break;
902 				}
903 				t = netdev_priv(dev);
904 				ipip6_tunnel_unlink(sitn, t);
905 				t->parms.iph.saddr = p.iph.saddr;
906 				t->parms.iph.daddr = p.iph.daddr;
907 				memcpy(dev->dev_addr, &p.iph.saddr, 4);
908 				memcpy(dev->broadcast, &p.iph.daddr, 4);
909 				ipip6_tunnel_link(sitn, t);
910 				netdev_state_change(dev);
911 			}
912 		}
913 
914 		if (t) {
915 			err = 0;
916 			if (cmd == SIOCCHGTUNNEL) {
917 				t->parms.iph.ttl = p.iph.ttl;
918 				t->parms.iph.tos = p.iph.tos;
919 				if (t->parms.link != p.link) {
920 					t->parms.link = p.link;
921 					ipip6_tunnel_bind_dev(dev);
922 					netdev_state_change(dev);
923 				}
924 			}
925 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
926 				err = -EFAULT;
927 		} else
928 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
929 		break;
930 
931 	case SIOCDELTUNNEL:
932 		err = -EPERM;
933 		if (!capable(CAP_NET_ADMIN))
934 			goto done;
935 
936 		if (dev == sitn->fb_tunnel_dev) {
937 			err = -EFAULT;
938 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
939 				goto done;
940 			err = -ENOENT;
941 			if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
942 				goto done;
943 			err = -EPERM;
944 			if (t == netdev_priv(sitn->fb_tunnel_dev))
945 				goto done;
946 			dev = t->dev;
947 		}
948 		unregister_netdevice(dev);
949 		err = 0;
950 		break;
951 
952 	case SIOCGETPRL:
953 		err = -EINVAL;
954 		if (dev == sitn->fb_tunnel_dev)
955 			goto done;
956 		err = -ENOENT;
957 		if (!(t = netdev_priv(dev)))
958 			goto done;
959 		err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
960 		break;
961 
962 	case SIOCADDPRL:
963 	case SIOCDELPRL:
964 	case SIOCCHGPRL:
965 		err = -EPERM;
966 		if (!capable(CAP_NET_ADMIN))
967 			goto done;
968 		err = -EINVAL;
969 		if (dev == sitn->fb_tunnel_dev)
970 			goto done;
971 		err = -EFAULT;
972 		if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
973 			goto done;
974 		err = -ENOENT;
975 		if (!(t = netdev_priv(dev)))
976 			goto done;
977 
978 		switch (cmd) {
979 		case SIOCDELPRL:
980 			err = ipip6_tunnel_del_prl(t, &prl);
981 			break;
982 		case SIOCADDPRL:
983 		case SIOCCHGPRL:
984 			err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
985 			break;
986 		}
987 		netdev_state_change(dev);
988 		break;
989 
990 	default:
991 		err = -EINVAL;
992 	}
993 
994 done:
995 	return err;
996 }
997 
998 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
999 {
1000 	if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1001 		return -EINVAL;
1002 	dev->mtu = new_mtu;
1003 	return 0;
1004 }
1005 
1006 static const struct net_device_ops ipip6_netdev_ops = {
1007 	.ndo_uninit	= ipip6_tunnel_uninit,
1008 	.ndo_start_xmit	= ipip6_tunnel_xmit,
1009 	.ndo_do_ioctl	= ipip6_tunnel_ioctl,
1010 	.ndo_change_mtu	= ipip6_tunnel_change_mtu,
1011 };
1012 
1013 static void ipip6_tunnel_setup(struct net_device *dev)
1014 {
1015 	dev->netdev_ops		= &ipip6_netdev_ops;
1016 	dev->destructor 	= free_netdev;
1017 
1018 	dev->type		= ARPHRD_SIT;
1019 	dev->hard_header_len 	= LL_MAX_HEADER + sizeof(struct iphdr);
1020 	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr);
1021 	dev->flags		= IFF_NOARP;
1022 	dev->priv_flags	       &= ~IFF_XMIT_DST_RELEASE;
1023 	dev->iflink		= 0;
1024 	dev->addr_len		= 4;
1025 	dev->features		|= NETIF_F_NETNS_LOCAL;
1026 }
1027 
1028 static void ipip6_tunnel_init(struct net_device *dev)
1029 {
1030 	struct ip_tunnel *tunnel = netdev_priv(dev);
1031 
1032 	tunnel->dev = dev;
1033 	strcpy(tunnel->parms.name, dev->name);
1034 
1035 	memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1036 	memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1037 
1038 	ipip6_tunnel_bind_dev(dev);
1039 }
1040 
1041 static void ipip6_fb_tunnel_init(struct net_device *dev)
1042 {
1043 	struct ip_tunnel *tunnel = netdev_priv(dev);
1044 	struct iphdr *iph = &tunnel->parms.iph;
1045 	struct net *net = dev_net(dev);
1046 	struct sit_net *sitn = net_generic(net, sit_net_id);
1047 
1048 	tunnel->dev = dev;
1049 	strcpy(tunnel->parms.name, dev->name);
1050 
1051 	iph->version		= 4;
1052 	iph->protocol		= IPPROTO_IPV6;
1053 	iph->ihl		= 5;
1054 	iph->ttl		= 64;
1055 
1056 	dev_hold(dev);
1057 	sitn->tunnels_wc[0]	= tunnel;
1058 }
1059 
1060 static struct xfrm_tunnel sit_handler = {
1061 	.handler	=	ipip6_rcv,
1062 	.err_handler	=	ipip6_err,
1063 	.priority	=	1,
1064 };
1065 
1066 static void sit_destroy_tunnels(struct sit_net *sitn)
1067 {
1068 	int prio;
1069 
1070 	for (prio = 1; prio < 4; prio++) {
1071 		int h;
1072 		for (h = 0; h < HASH_SIZE; h++) {
1073 			struct ip_tunnel *t;
1074 			while ((t = sitn->tunnels[prio][h]) != NULL)
1075 				unregister_netdevice(t->dev);
1076 		}
1077 	}
1078 }
1079 
1080 static int sit_init_net(struct net *net)
1081 {
1082 	int err;
1083 	struct sit_net *sitn;
1084 
1085 	err = -ENOMEM;
1086 	sitn = kzalloc(sizeof(struct sit_net), GFP_KERNEL);
1087 	if (sitn == NULL)
1088 		goto err_alloc;
1089 
1090 	err = net_assign_generic(net, sit_net_id, sitn);
1091 	if (err < 0)
1092 		goto err_assign;
1093 
1094 	sitn->tunnels[0] = sitn->tunnels_wc;
1095 	sitn->tunnels[1] = sitn->tunnels_l;
1096 	sitn->tunnels[2] = sitn->tunnels_r;
1097 	sitn->tunnels[3] = sitn->tunnels_r_l;
1098 
1099 	sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1100 					   ipip6_tunnel_setup);
1101 	if (!sitn->fb_tunnel_dev) {
1102 		err = -ENOMEM;
1103 		goto err_alloc_dev;
1104 	}
1105 	dev_net_set(sitn->fb_tunnel_dev, net);
1106 
1107 	ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1108 
1109 	if ((err = register_netdev(sitn->fb_tunnel_dev)))
1110 		goto err_reg_dev;
1111 
1112 	return 0;
1113 
1114 err_reg_dev:
1115 	dev_put(sitn->fb_tunnel_dev);
1116 	free_netdev(sitn->fb_tunnel_dev);
1117 err_alloc_dev:
1118 	/* nothing */
1119 err_assign:
1120 	kfree(sitn);
1121 err_alloc:
1122 	return err;
1123 }
1124 
1125 static void sit_exit_net(struct net *net)
1126 {
1127 	struct sit_net *sitn;
1128 
1129 	sitn = net_generic(net, sit_net_id);
1130 	rtnl_lock();
1131 	sit_destroy_tunnels(sitn);
1132 	unregister_netdevice(sitn->fb_tunnel_dev);
1133 	rtnl_unlock();
1134 	kfree(sitn);
1135 }
1136 
1137 static struct pernet_operations sit_net_ops = {
1138 	.init = sit_init_net,
1139 	.exit = sit_exit_net,
1140 };
1141 
1142 static void __exit sit_cleanup(void)
1143 {
1144 	xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1145 
1146 	unregister_pernet_gen_device(sit_net_id, &sit_net_ops);
1147 }
1148 
1149 static int __init sit_init(void)
1150 {
1151 	int err;
1152 
1153 	printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1154 
1155 	if (xfrm4_tunnel_register(&sit_handler, AF_INET6) < 0) {
1156 		printk(KERN_INFO "sit init: Can't add protocol\n");
1157 		return -EAGAIN;
1158 	}
1159 
1160 	err = register_pernet_gen_device(&sit_net_id, &sit_net_ops);
1161 	if (err < 0)
1162 		xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1163 
1164 	return err;
1165 }
1166 
1167 module_init(sit_init);
1168 module_exit(sit_cleanup);
1169 MODULE_LICENSE("GPL");
1170 MODULE_ALIAS("sit0");
1171