xref: /linux/net/ipv4/ipip.c (revision 606d099cdd1080bbb50ea50dc52d98252f8f10a1)
1 /*
2  *	Linux NET3:	IP/IP protocol decoder.
3  *
4  *	Version: $Id: ipip.c,v 1.50 2001/10/02 02:22:36 davem Exp $
5  *
6  *	Authors:
7  *		Sam Lantinga (slouken@cs.ucdavis.edu)  02/01/95
8  *
9  *	Fixes:
10  *		Alan Cox	:	Merged and made usable non modular (its so tiny its silly as
11  *					a module taking up 2 pages).
12  *		Alan Cox	: 	Fixed bug with 1.3.18 and IPIP not working (now needs to set skb->h.iph)
13  *					to keep ip_forward happy.
14  *		Alan Cox	:	More fixes for 1.3.21, and firewall fix. Maybe this will work soon 8).
15  *		Kai Schulte	:	Fixed #defines for IP_FIREWALL->FIREWALL
16  *              David Woodhouse :       Perform some basic ICMP handling.
17  *                                      IPIP Routing without decapsulation.
18  *              Carlos Picoto   :       GRE over IP support
19  *		Alexey Kuznetsov:	Reworked. Really, now it is truncated version of ipv4/ip_gre.c.
20  *					I do not want to merge them together.
21  *
22  *	This program is free software; you can redistribute it and/or
23  *	modify it under the terms of the GNU General Public License
24  *	as published by the Free Software Foundation; either version
25  *	2 of the License, or (at your option) any later version.
26  *
27  */
28 
29 /* tunnel.c: an IP tunnel driver
30 
31 	The purpose of this driver is to provide an IP tunnel through
32 	which you can tunnel network traffic transparently across subnets.
33 
34 	This was written by looking at Nick Holloway's dummy driver
35 	Thanks for the great code!
36 
37 		-Sam Lantinga	(slouken@cs.ucdavis.edu)  02/01/95
38 
39 	Minor tweaks:
40 		Cleaned up the code a little and added some pre-1.3.0 tweaks.
41 		dev->hard_header/hard_header_len changed to use no headers.
42 		Comments/bracketing tweaked.
43 		Made the tunnels use dev->name not tunnel: when error reporting.
44 		Added tx_dropped stat
45 
46 		-Alan Cox	(Alan.Cox@linux.org) 21 March 95
47 
48 	Reworked:
49 		Changed to tunnel to destination gateway in addition to the
50 			tunnel's pointopoint address
51 		Almost completely rewritten
52 		Note:  There is currently no firewall or ICMP handling done.
53 
54 		-Sam Lantinga	(slouken@cs.ucdavis.edu) 02/13/96
55 
56 */
57 
58 /* Things I wish I had known when writing the tunnel driver:
59 
60 	When the tunnel_xmit() function is called, the skb contains the
61 	packet to be sent (plus a great deal of extra info), and dev
62 	contains the tunnel device that _we_ are.
63 
64 	When we are passed a packet, we are expected to fill in the
65 	source address with our source IP address.
66 
67 	What is the proper way to allocate, copy and free a buffer?
68 	After you allocate it, it is a "0 length" chunk of memory
69 	starting at zero.  If you want to add headers to the buffer
70 	later, you'll have to call "skb_reserve(skb, amount)" with
71 	the amount of memory you want reserved.  Then, you call
72 	"skb_put(skb, amount)" with the amount of space you want in
73 	the buffer.  skb_put() returns a pointer to the top (#0) of
74 	that buffer.  skb->len is set to the amount of space you have
75 	"allocated" with skb_put().  You can then write up to skb->len
76 	bytes to that buffer.  If you need more, you can call skb_put()
77 	again with the additional amount of space you need.  You can
78 	find out how much more space you can allocate by calling
79 	"skb_tailroom(skb)".
80 	Now, to add header space, call "skb_push(skb, header_len)".
81 	This creates space at the beginning of the buffer and returns
82 	a pointer to this new space.  If later you need to strip a
83 	header from a buffer, call "skb_pull(skb, header_len)".
84 	skb_headroom() will return how much space is left at the top
85 	of the buffer (before the main data).  Remember, this headroom
86 	space must be reserved before the skb_put() function is called.
87 	*/
88 
89 /*
90    This version of net/ipv4/ipip.c is cloned of net/ipv4/ip_gre.c
91 
92    For comments look at net/ipv4/ip_gre.c --ANK
93  */
94 
95 
96 #include <linux/capability.h>
97 #include <linux/module.h>
98 #include <linux/types.h>
99 #include <linux/sched.h>
100 #include <linux/kernel.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <linux/in.h>
105 #include <linux/tcp.h>
106 #include <linux/udp.h>
107 #include <linux/if_arp.h>
108 #include <linux/mroute.h>
109 #include <linux/init.h>
110 #include <linux/netfilter_ipv4.h>
111 #include <linux/if_ether.h>
112 
113 #include <net/sock.h>
114 #include <net/ip.h>
115 #include <net/icmp.h>
116 #include <net/ipip.h>
117 #include <net/inet_ecn.h>
118 #include <net/xfrm.h>
119 
120 #define HASH_SIZE  16
121 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
122 
123 static int ipip_fb_tunnel_init(struct net_device *dev);
124 static int ipip_tunnel_init(struct net_device *dev);
125 static void ipip_tunnel_setup(struct net_device *dev);
126 
127 static struct net_device *ipip_fb_tunnel_dev;
128 
129 static struct ip_tunnel *tunnels_r_l[HASH_SIZE];
130 static struct ip_tunnel *tunnels_r[HASH_SIZE];
131 static struct ip_tunnel *tunnels_l[HASH_SIZE];
132 static struct ip_tunnel *tunnels_wc[1];
133 static struct ip_tunnel **tunnels[4] = { tunnels_wc, tunnels_l, tunnels_r, tunnels_r_l };
134 
135 static DEFINE_RWLOCK(ipip_lock);
136 
137 static struct ip_tunnel * ipip_tunnel_lookup(__be32 remote, __be32 local)
138 {
139 	unsigned h0 = HASH(remote);
140 	unsigned h1 = HASH(local);
141 	struct ip_tunnel *t;
142 
143 	for (t = tunnels_r_l[h0^h1]; t; t = t->next) {
144 		if (local == t->parms.iph.saddr &&
145 		    remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
146 			return t;
147 	}
148 	for (t = tunnels_r[h0]; t; t = t->next) {
149 		if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
150 			return t;
151 	}
152 	for (t = tunnels_l[h1]; t; t = t->next) {
153 		if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
154 			return t;
155 	}
156 	if ((t = tunnels_wc[0]) != NULL && (t->dev->flags&IFF_UP))
157 		return t;
158 	return NULL;
159 }
160 
161 static struct ip_tunnel **ipip_bucket(struct ip_tunnel *t)
162 {
163 	__be32 remote = t->parms.iph.daddr;
164 	__be32 local = t->parms.iph.saddr;
165 	unsigned h = 0;
166 	int prio = 0;
167 
168 	if (remote) {
169 		prio |= 2;
170 		h ^= HASH(remote);
171 	}
172 	if (local) {
173 		prio |= 1;
174 		h ^= HASH(local);
175 	}
176 	return &tunnels[prio][h];
177 }
178 
179 
180 static void ipip_tunnel_unlink(struct ip_tunnel *t)
181 {
182 	struct ip_tunnel **tp;
183 
184 	for (tp = ipip_bucket(t); *tp; tp = &(*tp)->next) {
185 		if (t == *tp) {
186 			write_lock_bh(&ipip_lock);
187 			*tp = t->next;
188 			write_unlock_bh(&ipip_lock);
189 			break;
190 		}
191 	}
192 }
193 
194 static void ipip_tunnel_link(struct ip_tunnel *t)
195 {
196 	struct ip_tunnel **tp = ipip_bucket(t);
197 
198 	t->next = *tp;
199 	write_lock_bh(&ipip_lock);
200 	*tp = t;
201 	write_unlock_bh(&ipip_lock);
202 }
203 
204 static struct ip_tunnel * ipip_tunnel_locate(struct ip_tunnel_parm *parms, int create)
205 {
206 	__be32 remote = parms->iph.daddr;
207 	__be32 local = parms->iph.saddr;
208 	struct ip_tunnel *t, **tp, *nt;
209 	struct net_device *dev;
210 	unsigned h = 0;
211 	int prio = 0;
212 	char name[IFNAMSIZ];
213 
214 	if (remote) {
215 		prio |= 2;
216 		h ^= HASH(remote);
217 	}
218 	if (local) {
219 		prio |= 1;
220 		h ^= HASH(local);
221 	}
222 	for (tp = &tunnels[prio][h]; (t = *tp) != NULL; tp = &t->next) {
223 		if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
224 			return t;
225 	}
226 	if (!create)
227 		return NULL;
228 
229 	if (parms->name[0])
230 		strlcpy(name, parms->name, IFNAMSIZ);
231 	else {
232 		int i;
233 		for (i=1; i<100; i++) {
234 			sprintf(name, "tunl%d", i);
235 			if (__dev_get_by_name(name) == NULL)
236 				break;
237 		}
238 		if (i==100)
239 			goto failed;
240 	}
241 
242 	dev = alloc_netdev(sizeof(*t), name, ipip_tunnel_setup);
243 	if (dev == NULL)
244 		return NULL;
245 
246 	nt = netdev_priv(dev);
247 	SET_MODULE_OWNER(dev);
248 	dev->init = ipip_tunnel_init;
249 	nt->parms = *parms;
250 
251 	if (register_netdevice(dev) < 0) {
252 		free_netdev(dev);
253 		goto failed;
254 	}
255 
256 	dev_hold(dev);
257 	ipip_tunnel_link(nt);
258 	return nt;
259 
260 failed:
261 	return NULL;
262 }
263 
264 static void ipip_tunnel_uninit(struct net_device *dev)
265 {
266 	if (dev == ipip_fb_tunnel_dev) {
267 		write_lock_bh(&ipip_lock);
268 		tunnels_wc[0] = NULL;
269 		write_unlock_bh(&ipip_lock);
270 	} else
271 		ipip_tunnel_unlink(netdev_priv(dev));
272 	dev_put(dev);
273 }
274 
275 static int ipip_err(struct sk_buff *skb, u32 info)
276 {
277 #ifndef I_WISH_WORLD_WERE_PERFECT
278 
279 /* It is not :-( All the routers (except for Linux) return only
280    8 bytes of packet payload. It means, that precise relaying of
281    ICMP in the real Internet is absolutely infeasible.
282  */
283 	struct iphdr *iph = (struct iphdr*)skb->data;
284 	int type = skb->h.icmph->type;
285 	int code = skb->h.icmph->code;
286 	struct ip_tunnel *t;
287 	int err;
288 
289 	switch (type) {
290 	default:
291 	case ICMP_PARAMETERPROB:
292 		return 0;
293 
294 	case ICMP_DEST_UNREACH:
295 		switch (code) {
296 		case ICMP_SR_FAILED:
297 		case ICMP_PORT_UNREACH:
298 			/* Impossible event. */
299 			return 0;
300 		case ICMP_FRAG_NEEDED:
301 			/* Soft state for pmtu is maintained by IP core. */
302 			return 0;
303 		default:
304 			/* All others are translated to HOST_UNREACH.
305 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
306 			   I believe they are just ether pollution. --ANK
307 			 */
308 			break;
309 		}
310 		break;
311 	case ICMP_TIME_EXCEEDED:
312 		if (code != ICMP_EXC_TTL)
313 			return 0;
314 		break;
315 	}
316 
317 	err = -ENOENT;
318 
319 	read_lock(&ipip_lock);
320 	t = ipip_tunnel_lookup(iph->daddr, iph->saddr);
321 	if (t == NULL || t->parms.iph.daddr == 0)
322 		goto out;
323 
324 	err = 0;
325 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
326 		goto out;
327 
328 	if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
329 		t->err_count++;
330 	else
331 		t->err_count = 1;
332 	t->err_time = jiffies;
333 out:
334 	read_unlock(&ipip_lock);
335 	return err;
336 #else
337 	struct iphdr *iph = (struct iphdr*)dp;
338 	int hlen = iph->ihl<<2;
339 	struct iphdr *eiph;
340 	int type = skb->h.icmph->type;
341 	int code = skb->h.icmph->code;
342 	int rel_type = 0;
343 	int rel_code = 0;
344 	__be32 rel_info = 0;
345 	__u32 n = 0;
346 	struct sk_buff *skb2;
347 	struct flowi fl;
348 	struct rtable *rt;
349 
350 	if (len < hlen + sizeof(struct iphdr))
351 		return 0;
352 	eiph = (struct iphdr*)(dp + hlen);
353 
354 	switch (type) {
355 	default:
356 		return 0;
357 	case ICMP_PARAMETERPROB:
358 		n = ntohl(skb->h.icmph->un.gateway) >> 24;
359 		if (n < hlen)
360 			return 0;
361 
362 		/* So... This guy found something strange INSIDE encapsulated
363 		   packet. Well, he is fool, but what can we do ?
364 		 */
365 		rel_type = ICMP_PARAMETERPROB;
366 		rel_info = htonl((n - hlen) << 24);
367 		break;
368 
369 	case ICMP_DEST_UNREACH:
370 		switch (code) {
371 		case ICMP_SR_FAILED:
372 		case ICMP_PORT_UNREACH:
373 			/* Impossible event. */
374 			return 0;
375 		case ICMP_FRAG_NEEDED:
376 			/* And it is the only really necessary thing :-) */
377 			n = ntohs(skb->h.icmph->un.frag.mtu);
378 			if (n < hlen+68)
379 				return 0;
380 			n -= hlen;
381 			/* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */
382 			if (n > ntohs(eiph->tot_len))
383 				return 0;
384 			rel_info = htonl(n);
385 			break;
386 		default:
387 			/* All others are translated to HOST_UNREACH.
388 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
389 			   I believe, it is just ether pollution. --ANK
390 			 */
391 			rel_type = ICMP_DEST_UNREACH;
392 			rel_code = ICMP_HOST_UNREACH;
393 			break;
394 		}
395 		break;
396 	case ICMP_TIME_EXCEEDED:
397 		if (code != ICMP_EXC_TTL)
398 			return 0;
399 		break;
400 	}
401 
402 	/* Prepare fake skb to feed it to icmp_send */
403 	skb2 = skb_clone(skb, GFP_ATOMIC);
404 	if (skb2 == NULL)
405 		return 0;
406 	dst_release(skb2->dst);
407 	skb2->dst = NULL;
408 	skb_pull(skb2, skb->data - (u8*)eiph);
409 	skb2->nh.raw = skb2->data;
410 
411 	/* Try to guess incoming interface */
412 	memset(&fl, 0, sizeof(fl));
413 	fl.fl4_daddr = eiph->saddr;
414 	fl.fl4_tos = RT_TOS(eiph->tos);
415 	fl.proto = IPPROTO_IPIP;
416 	if (ip_route_output_key(&rt, &key)) {
417 		kfree_skb(skb2);
418 		return 0;
419 	}
420 	skb2->dev = rt->u.dst.dev;
421 
422 	/* route "incoming" packet */
423 	if (rt->rt_flags&RTCF_LOCAL) {
424 		ip_rt_put(rt);
425 		rt = NULL;
426 		fl.fl4_daddr = eiph->daddr;
427 		fl.fl4_src = eiph->saddr;
428 		fl.fl4_tos = eiph->tos;
429 		if (ip_route_output_key(&rt, &fl) ||
430 		    rt->u.dst.dev->type != ARPHRD_TUNNEL) {
431 			ip_rt_put(rt);
432 			kfree_skb(skb2);
433 			return 0;
434 		}
435 	} else {
436 		ip_rt_put(rt);
437 		if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos, skb2->dev) ||
438 		    skb2->dst->dev->type != ARPHRD_TUNNEL) {
439 			kfree_skb(skb2);
440 			return 0;
441 		}
442 	}
443 
444 	/* change mtu on this route */
445 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
446 		if (n > dst_mtu(skb2->dst)) {
447 			kfree_skb(skb2);
448 			return 0;
449 		}
450 		skb2->dst->ops->update_pmtu(skb2->dst, n);
451 	} else if (type == ICMP_TIME_EXCEEDED) {
452 		struct ip_tunnel *t = netdev_priv(skb2->dev);
453 		if (t->parms.iph.ttl) {
454 			rel_type = ICMP_DEST_UNREACH;
455 			rel_code = ICMP_HOST_UNREACH;
456 		}
457 	}
458 
459 	icmp_send(skb2, rel_type, rel_code, rel_info);
460 	kfree_skb(skb2);
461 	return 0;
462 #endif
463 }
464 
465 static inline void ipip_ecn_decapsulate(struct iphdr *outer_iph, struct sk_buff *skb)
466 {
467 	struct iphdr *inner_iph = skb->nh.iph;
468 
469 	if (INET_ECN_is_ce(outer_iph->tos))
470 		IP_ECN_set_ce(inner_iph);
471 }
472 
473 static int ipip_rcv(struct sk_buff *skb)
474 {
475 	struct iphdr *iph;
476 	struct ip_tunnel *tunnel;
477 
478 	iph = skb->nh.iph;
479 
480 	read_lock(&ipip_lock);
481 	if ((tunnel = ipip_tunnel_lookup(iph->saddr, iph->daddr)) != NULL) {
482 		if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
483 			read_unlock(&ipip_lock);
484 			kfree_skb(skb);
485 			return 0;
486 		}
487 
488 		secpath_reset(skb);
489 
490 		skb->mac.raw = skb->nh.raw;
491 		skb->nh.raw = skb->data;
492 		skb->protocol = htons(ETH_P_IP);
493 		skb->pkt_type = PACKET_HOST;
494 
495 		tunnel->stat.rx_packets++;
496 		tunnel->stat.rx_bytes += skb->len;
497 		skb->dev = tunnel->dev;
498 		dst_release(skb->dst);
499 		skb->dst = NULL;
500 		nf_reset(skb);
501 		ipip_ecn_decapsulate(iph, skb);
502 		netif_rx(skb);
503 		read_unlock(&ipip_lock);
504 		return 0;
505 	}
506 	read_unlock(&ipip_lock);
507 
508 	return -1;
509 }
510 
511 /*
512  *	This function assumes it is being called from dev_queue_xmit()
513  *	and that skb is filled properly by that function.
514  */
515 
516 static int ipip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
517 {
518 	struct ip_tunnel *tunnel = netdev_priv(dev);
519 	struct net_device_stats *stats = &tunnel->stat;
520 	struct iphdr  *tiph = &tunnel->parms.iph;
521 	u8     tos = tunnel->parms.iph.tos;
522 	__be16 df = tiph->frag_off;
523 	struct rtable *rt;     			/* Route to the other host */
524 	struct net_device *tdev;			/* Device to other host */
525 	struct iphdr  *old_iph = skb->nh.iph;
526 	struct iphdr  *iph;			/* Our new IP header */
527 	int    max_headroom;			/* The extra header space needed */
528 	__be32 dst = tiph->daddr;
529 	int    mtu;
530 
531 	if (tunnel->recursion++) {
532 		tunnel->stat.collisions++;
533 		goto tx_error;
534 	}
535 
536 	if (skb->protocol != htons(ETH_P_IP))
537 		goto tx_error;
538 
539 	if (tos&1)
540 		tos = old_iph->tos;
541 
542 	if (!dst) {
543 		/* NBMA tunnel */
544 		if ((rt = (struct rtable*)skb->dst) == NULL) {
545 			tunnel->stat.tx_fifo_errors++;
546 			goto tx_error;
547 		}
548 		if ((dst = rt->rt_gateway) == 0)
549 			goto tx_error_icmp;
550 	}
551 
552 	{
553 		struct flowi fl = { .oif = tunnel->parms.link,
554 				    .nl_u = { .ip4_u =
555 					      { .daddr = dst,
556 						.saddr = tiph->saddr,
557 						.tos = RT_TOS(tos) } },
558 				    .proto = IPPROTO_IPIP };
559 		if (ip_route_output_key(&rt, &fl)) {
560 			tunnel->stat.tx_carrier_errors++;
561 			goto tx_error_icmp;
562 		}
563 	}
564 	tdev = rt->u.dst.dev;
565 
566 	if (tdev == dev) {
567 		ip_rt_put(rt);
568 		tunnel->stat.collisions++;
569 		goto tx_error;
570 	}
571 
572 	if (tiph->frag_off)
573 		mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
574 	else
575 		mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
576 
577 	if (mtu < 68) {
578 		tunnel->stat.collisions++;
579 		ip_rt_put(rt);
580 		goto tx_error;
581 	}
582 	if (skb->dst)
583 		skb->dst->ops->update_pmtu(skb->dst, mtu);
584 
585 	df |= (old_iph->frag_off&htons(IP_DF));
586 
587 	if ((old_iph->frag_off&htons(IP_DF)) && mtu < ntohs(old_iph->tot_len)) {
588 		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
589 		ip_rt_put(rt);
590 		goto tx_error;
591 	}
592 
593 	if (tunnel->err_count > 0) {
594 		if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
595 			tunnel->err_count--;
596 			dst_link_failure(skb);
597 		} else
598 			tunnel->err_count = 0;
599 	}
600 
601 	/*
602 	 * Okay, now see if we can stuff it in the buffer as-is.
603 	 */
604 	max_headroom = (LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr));
605 
606 	if (skb_headroom(skb) < max_headroom || skb_cloned(skb) || skb_shared(skb)) {
607 		struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
608 		if (!new_skb) {
609 			ip_rt_put(rt);
610   			stats->tx_dropped++;
611 			dev_kfree_skb(skb);
612 			tunnel->recursion--;
613 			return 0;
614 		}
615 		if (skb->sk)
616 			skb_set_owner_w(new_skb, skb->sk);
617 		dev_kfree_skb(skb);
618 		skb = new_skb;
619 		old_iph = skb->nh.iph;
620 	}
621 
622 	skb->h.raw = skb->nh.raw;
623 	skb->nh.raw = skb_push(skb, sizeof(struct iphdr));
624 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
625 	IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
626 			      IPSKB_REROUTED);
627 	dst_release(skb->dst);
628 	skb->dst = &rt->u.dst;
629 
630 	/*
631 	 *	Push down and install the IPIP header.
632 	 */
633 
634 	iph 			=	skb->nh.iph;
635 	iph->version		=	4;
636 	iph->ihl		=	sizeof(struct iphdr)>>2;
637 	iph->frag_off		=	df;
638 	iph->protocol		=	IPPROTO_IPIP;
639 	iph->tos		=	INET_ECN_encapsulate(tos, old_iph->tos);
640 	iph->daddr		=	rt->rt_dst;
641 	iph->saddr		=	rt->rt_src;
642 
643 	if ((iph->ttl = tiph->ttl) == 0)
644 		iph->ttl	=	old_iph->ttl;
645 
646 	nf_reset(skb);
647 
648 	IPTUNNEL_XMIT();
649 	tunnel->recursion--;
650 	return 0;
651 
652 tx_error_icmp:
653 	dst_link_failure(skb);
654 tx_error:
655 	stats->tx_errors++;
656 	dev_kfree_skb(skb);
657 	tunnel->recursion--;
658 	return 0;
659 }
660 
661 static int
662 ipip_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
663 {
664 	int err = 0;
665 	struct ip_tunnel_parm p;
666 	struct ip_tunnel *t;
667 
668 	switch (cmd) {
669 	case SIOCGETTUNNEL:
670 		t = NULL;
671 		if (dev == ipip_fb_tunnel_dev) {
672 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
673 				err = -EFAULT;
674 				break;
675 			}
676 			t = ipip_tunnel_locate(&p, 0);
677 		}
678 		if (t == NULL)
679 			t = netdev_priv(dev);
680 		memcpy(&p, &t->parms, sizeof(p));
681 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
682 			err = -EFAULT;
683 		break;
684 
685 	case SIOCADDTUNNEL:
686 	case SIOCCHGTUNNEL:
687 		err = -EPERM;
688 		if (!capable(CAP_NET_ADMIN))
689 			goto done;
690 
691 		err = -EFAULT;
692 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
693 			goto done;
694 
695 		err = -EINVAL;
696 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
697 		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
698 			goto done;
699 		if (p.iph.ttl)
700 			p.iph.frag_off |= htons(IP_DF);
701 
702 		t = ipip_tunnel_locate(&p, cmd == SIOCADDTUNNEL);
703 
704 		if (dev != ipip_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
705 			if (t != NULL) {
706 				if (t->dev != dev) {
707 					err = -EEXIST;
708 					break;
709 				}
710 			} else {
711 				if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
712 				    (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
713 					err = -EINVAL;
714 					break;
715 				}
716 				t = netdev_priv(dev);
717 				ipip_tunnel_unlink(t);
718 				t->parms.iph.saddr = p.iph.saddr;
719 				t->parms.iph.daddr = p.iph.daddr;
720 				memcpy(dev->dev_addr, &p.iph.saddr, 4);
721 				memcpy(dev->broadcast, &p.iph.daddr, 4);
722 				ipip_tunnel_link(t);
723 				netdev_state_change(dev);
724 			}
725 		}
726 
727 		if (t) {
728 			err = 0;
729 			if (cmd == SIOCCHGTUNNEL) {
730 				t->parms.iph.ttl = p.iph.ttl;
731 				t->parms.iph.tos = p.iph.tos;
732 				t->parms.iph.frag_off = p.iph.frag_off;
733 			}
734 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
735 				err = -EFAULT;
736 		} else
737 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
738 		break;
739 
740 	case SIOCDELTUNNEL:
741 		err = -EPERM;
742 		if (!capable(CAP_NET_ADMIN))
743 			goto done;
744 
745 		if (dev == ipip_fb_tunnel_dev) {
746 			err = -EFAULT;
747 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
748 				goto done;
749 			err = -ENOENT;
750 			if ((t = ipip_tunnel_locate(&p, 0)) == NULL)
751 				goto done;
752 			err = -EPERM;
753 			if (t->dev == ipip_fb_tunnel_dev)
754 				goto done;
755 			dev = t->dev;
756 		}
757 		err = unregister_netdevice(dev);
758 		break;
759 
760 	default:
761 		err = -EINVAL;
762 	}
763 
764 done:
765 	return err;
766 }
767 
768 static struct net_device_stats *ipip_tunnel_get_stats(struct net_device *dev)
769 {
770 	return &(((struct ip_tunnel*)netdev_priv(dev))->stat);
771 }
772 
773 static int ipip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
774 {
775 	if (new_mtu < 68 || new_mtu > 0xFFF8 - sizeof(struct iphdr))
776 		return -EINVAL;
777 	dev->mtu = new_mtu;
778 	return 0;
779 }
780 
781 static void ipip_tunnel_setup(struct net_device *dev)
782 {
783 	SET_MODULE_OWNER(dev);
784 	dev->uninit		= ipip_tunnel_uninit;
785 	dev->hard_start_xmit	= ipip_tunnel_xmit;
786 	dev->get_stats		= ipip_tunnel_get_stats;
787 	dev->do_ioctl		= ipip_tunnel_ioctl;
788 	dev->change_mtu		= ipip_tunnel_change_mtu;
789 	dev->destructor		= free_netdev;
790 
791 	dev->type		= ARPHRD_TUNNEL;
792 	dev->hard_header_len 	= LL_MAX_HEADER + sizeof(struct iphdr);
793 	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr);
794 	dev->flags		= IFF_NOARP;
795 	dev->iflink		= 0;
796 	dev->addr_len		= 4;
797 }
798 
799 static int ipip_tunnel_init(struct net_device *dev)
800 {
801 	struct net_device *tdev = NULL;
802 	struct ip_tunnel *tunnel;
803 	struct iphdr *iph;
804 
805 	tunnel = netdev_priv(dev);
806 	iph = &tunnel->parms.iph;
807 
808 	tunnel->dev = dev;
809 	strcpy(tunnel->parms.name, dev->name);
810 
811 	memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
812 	memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
813 
814 	if (iph->daddr) {
815 		struct flowi fl = { .oif = tunnel->parms.link,
816 				    .nl_u = { .ip4_u =
817 					      { .daddr = iph->daddr,
818 						.saddr = iph->saddr,
819 						.tos = RT_TOS(iph->tos) } },
820 				    .proto = IPPROTO_IPIP };
821 		struct rtable *rt;
822 		if (!ip_route_output_key(&rt, &fl)) {
823 			tdev = rt->u.dst.dev;
824 			ip_rt_put(rt);
825 		}
826 		dev->flags |= IFF_POINTOPOINT;
827 	}
828 
829 	if (!tdev && tunnel->parms.link)
830 		tdev = __dev_get_by_index(tunnel->parms.link);
831 
832 	if (tdev) {
833 		dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
834 		dev->mtu = tdev->mtu - sizeof(struct iphdr);
835 	}
836 	dev->iflink = tunnel->parms.link;
837 
838 	return 0;
839 }
840 
841 static int __init ipip_fb_tunnel_init(struct net_device *dev)
842 {
843 	struct ip_tunnel *tunnel = netdev_priv(dev);
844 	struct iphdr *iph = &tunnel->parms.iph;
845 
846 	tunnel->dev = dev;
847 	strcpy(tunnel->parms.name, dev->name);
848 
849 	iph->version		= 4;
850 	iph->protocol		= IPPROTO_IPIP;
851 	iph->ihl		= 5;
852 
853 	dev_hold(dev);
854 	tunnels_wc[0]		= tunnel;
855 	return 0;
856 }
857 
858 static struct xfrm_tunnel ipip_handler = {
859 	.handler	=	ipip_rcv,
860 	.err_handler	=	ipip_err,
861 	.priority	=	1,
862 };
863 
864 static char banner[] __initdata =
865 	KERN_INFO "IPv4 over IPv4 tunneling driver\n";
866 
867 static int __init ipip_init(void)
868 {
869 	int err;
870 
871 	printk(banner);
872 
873 	if (xfrm4_tunnel_register(&ipip_handler)) {
874 		printk(KERN_INFO "ipip init: can't register tunnel\n");
875 		return -EAGAIN;
876 	}
877 
878 	ipip_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
879 					   "tunl0",
880 					   ipip_tunnel_setup);
881 	if (!ipip_fb_tunnel_dev) {
882 		err = -ENOMEM;
883 		goto err1;
884 	}
885 
886 	ipip_fb_tunnel_dev->init = ipip_fb_tunnel_init;
887 
888 	if ((err = register_netdev(ipip_fb_tunnel_dev)))
889 		goto err2;
890  out:
891 	return err;
892  err2:
893 	free_netdev(ipip_fb_tunnel_dev);
894  err1:
895 	xfrm4_tunnel_deregister(&ipip_handler);
896 	goto out;
897 }
898 
899 static void __exit ipip_destroy_tunnels(void)
900 {
901 	int prio;
902 
903 	for (prio = 1; prio < 4; prio++) {
904 		int h;
905 		for (h = 0; h < HASH_SIZE; h++) {
906 			struct ip_tunnel *t;
907 			while ((t = tunnels[prio][h]) != NULL)
908 				unregister_netdevice(t->dev);
909 		}
910 	}
911 }
912 
913 static void __exit ipip_fini(void)
914 {
915 	if (xfrm4_tunnel_deregister(&ipip_handler))
916 		printk(KERN_INFO "ipip close: can't deregister tunnel\n");
917 
918 	rtnl_lock();
919 	ipip_destroy_tunnels();
920 	unregister_netdevice(ipip_fb_tunnel_dev);
921 	rtnl_unlock();
922 }
923 
924 module_init(ipip_init);
925 module_exit(ipip_fini);
926 MODULE_LICENSE("GPL");
927