xref: /linux/net/ipv4/ip_gre.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
1 /*
2  *	Linux NET3:	GRE over IP protocol decoder.
3  *
4  *	Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
5  *
6  *	This program is free software; you can redistribute it and/or
7  *	modify it under the terms of the GNU General Public License
8  *	as published by the Free Software Foundation; either version
9  *	2 of the License, or (at your option) any later version.
10  *
11  */
12 
13 #include <linux/capability.h>
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/slab.h>
18 #include <asm/uaccess.h>
19 #include <linux/skbuff.h>
20 #include <linux/netdevice.h>
21 #include <linux/in.h>
22 #include <linux/tcp.h>
23 #include <linux/udp.h>
24 #include <linux/if_arp.h>
25 #include <linux/mroute.h>
26 #include <linux/init.h>
27 #include <linux/in6.h>
28 #include <linux/inetdevice.h>
29 #include <linux/igmp.h>
30 #include <linux/netfilter_ipv4.h>
31 #include <linux/etherdevice.h>
32 #include <linux/if_ether.h>
33 
34 #include <net/sock.h>
35 #include <net/ip.h>
36 #include <net/icmp.h>
37 #include <net/protocol.h>
38 #include <net/ipip.h>
39 #include <net/arp.h>
40 #include <net/checksum.h>
41 #include <net/dsfield.h>
42 #include <net/inet_ecn.h>
43 #include <net/xfrm.h>
44 #include <net/net_namespace.h>
45 #include <net/netns/generic.h>
46 #include <net/rtnetlink.h>
47 #include <net/gre.h>
48 
49 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
50 #include <net/ipv6.h>
51 #include <net/ip6_fib.h>
52 #include <net/ip6_route.h>
53 #endif
54 
55 /*
56    Problems & solutions
57    --------------------
58 
59    1. The most important issue is detecting local dead loops.
60    They would cause complete host lockup in transmit, which
61    would be "resolved" by stack overflow or, if queueing is enabled,
62    with infinite looping in net_bh.
63 
64    We cannot track such dead loops during route installation,
65    it is infeasible task. The most general solutions would be
66    to keep skb->encapsulation counter (sort of local ttl),
67    and silently drop packet when it expires. It is a good
68    solution, but it supposes maintaing new variable in ALL
69    skb, even if no tunneling is used.
70 
71    Current solution: xmit_recursion breaks dead loops. This is a percpu
72    counter, since when we enter the first ndo_xmit(), cpu migration is
73    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
74 
75    2. Networking dead loops would not kill routers, but would really
76    kill network. IP hop limit plays role of "t->recursion" in this case,
77    if we copy it from packet being encapsulated to upper header.
78    It is very good solution, but it introduces two problems:
79 
80    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
81      do not work over tunnels.
82    - traceroute does not work. I planned to relay ICMP from tunnel,
83      so that this problem would be solved and traceroute output
84      would even more informative. This idea appeared to be wrong:
85      only Linux complies to rfc1812 now (yes, guys, Linux is the only
86      true router now :-)), all routers (at least, in neighbourhood of mine)
87      return only 8 bytes of payload. It is the end.
88 
89    Hence, if we want that OSPF worked or traceroute said something reasonable,
90    we should search for another solution.
91 
92    One of them is to parse packet trying to detect inner encapsulation
93    made by our node. It is difficult or even impossible, especially,
94    taking into account fragmentation. TO be short, tt is not solution at all.
95 
96    Current solution: The solution was UNEXPECTEDLY SIMPLE.
97    We force DF flag on tunnels with preconfigured hop limit,
98    that is ALL. :-) Well, it does not remove the problem completely,
99    but exponential growth of network traffic is changed to linear
100    (branches, that exceed pmtu are pruned) and tunnel mtu
101    fastly degrades to value <68, where looping stops.
102    Yes, it is not good if there exists a router in the loop,
103    which does not force DF, even when encapsulating packets have DF set.
104    But it is not our problem! Nobody could accuse us, we made
105    all that we could make. Even if it is your gated who injected
106    fatal route to network, even if it were you who configured
107    fatal static route: you are innocent. :-)
108 
109 
110 
111    3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain
112    practically identical code. It would be good to glue them
113    together, but it is not very evident, how to make them modular.
114    sit is integral part of IPv6, ipip and gre are naturally modular.
115    We could extract common parts (hash table, ioctl etc)
116    to a separate module (ip_tunnel.c).
117 
118    Alexey Kuznetsov.
119  */
120 
121 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
122 static int ipgre_tunnel_init(struct net_device *dev);
123 static void ipgre_tunnel_setup(struct net_device *dev);
124 static int ipgre_tunnel_bind_dev(struct net_device *dev);
125 
126 /* Fallback tunnel: no source, no destination, no key, no options */
127 
128 #define HASH_SIZE  16
129 
130 static int ipgre_net_id __read_mostly;
131 struct ipgre_net {
132 	struct ip_tunnel __rcu *tunnels[4][HASH_SIZE];
133 
134 	struct net_device *fb_tunnel_dev;
135 };
136 
137 /* Tunnel hash table */
138 
139 /*
140    4 hash tables:
141 
142    3: (remote,local)
143    2: (remote,*)
144    1: (*,local)
145    0: (*,*)
146 
147    We require exact key match i.e. if a key is present in packet
148    it will match only tunnel with the same key; if it is not present,
149    it will match only keyless tunnel.
150 
151    All keysless packets, if not matched configured keyless tunnels
152    will match fallback tunnel.
153  */
154 
155 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
156 
157 #define tunnels_r_l	tunnels[3]
158 #define tunnels_r	tunnels[2]
159 #define tunnels_l	tunnels[1]
160 #define tunnels_wc	tunnels[0]
161 /*
162  * Locking : hash tables are protected by RCU and RTNL
163  */
164 
165 #define for_each_ip_tunnel_rcu(start) \
166 	for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
167 
168 /* often modified stats are per cpu, other are shared (netdev->stats) */
169 struct pcpu_tstats {
170 	unsigned long	rx_packets;
171 	unsigned long	rx_bytes;
172 	unsigned long	tx_packets;
173 	unsigned long	tx_bytes;
174 };
175 
176 static struct net_device_stats *ipgre_get_stats(struct net_device *dev)
177 {
178 	struct pcpu_tstats sum = { 0 };
179 	int i;
180 
181 	for_each_possible_cpu(i) {
182 		const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
183 
184 		sum.rx_packets += tstats->rx_packets;
185 		sum.rx_bytes   += tstats->rx_bytes;
186 		sum.tx_packets += tstats->tx_packets;
187 		sum.tx_bytes   += tstats->tx_bytes;
188 	}
189 	dev->stats.rx_packets = sum.rx_packets;
190 	dev->stats.rx_bytes   = sum.rx_bytes;
191 	dev->stats.tx_packets = sum.tx_packets;
192 	dev->stats.tx_bytes   = sum.tx_bytes;
193 	return &dev->stats;
194 }
195 
196 /* Given src, dst and key, find appropriate for input tunnel. */
197 
198 static struct ip_tunnel * ipgre_tunnel_lookup(struct net_device *dev,
199 					      __be32 remote, __be32 local,
200 					      __be32 key, __be16 gre_proto)
201 {
202 	struct net *net = dev_net(dev);
203 	int link = dev->ifindex;
204 	unsigned int h0 = HASH(remote);
205 	unsigned int h1 = HASH(key);
206 	struct ip_tunnel *t, *cand = NULL;
207 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
208 	int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
209 		       ARPHRD_ETHER : ARPHRD_IPGRE;
210 	int score, cand_score = 4;
211 
212 	for_each_ip_tunnel_rcu(ign->tunnels_r_l[h0 ^ h1]) {
213 		if (local != t->parms.iph.saddr ||
214 		    remote != t->parms.iph.daddr ||
215 		    key != t->parms.i_key ||
216 		    !(t->dev->flags & IFF_UP))
217 			continue;
218 
219 		if (t->dev->type != ARPHRD_IPGRE &&
220 		    t->dev->type != dev_type)
221 			continue;
222 
223 		score = 0;
224 		if (t->parms.link != link)
225 			score |= 1;
226 		if (t->dev->type != dev_type)
227 			score |= 2;
228 		if (score == 0)
229 			return t;
230 
231 		if (score < cand_score) {
232 			cand = t;
233 			cand_score = score;
234 		}
235 	}
236 
237 	for_each_ip_tunnel_rcu(ign->tunnels_r[h0 ^ h1]) {
238 		if (remote != t->parms.iph.daddr ||
239 		    key != t->parms.i_key ||
240 		    !(t->dev->flags & IFF_UP))
241 			continue;
242 
243 		if (t->dev->type != ARPHRD_IPGRE &&
244 		    t->dev->type != dev_type)
245 			continue;
246 
247 		score = 0;
248 		if (t->parms.link != link)
249 			score |= 1;
250 		if (t->dev->type != dev_type)
251 			score |= 2;
252 		if (score == 0)
253 			return t;
254 
255 		if (score < cand_score) {
256 			cand = t;
257 			cand_score = score;
258 		}
259 	}
260 
261 	for_each_ip_tunnel_rcu(ign->tunnels_l[h1]) {
262 		if ((local != t->parms.iph.saddr &&
263 		     (local != t->parms.iph.daddr ||
264 		      !ipv4_is_multicast(local))) ||
265 		    key != t->parms.i_key ||
266 		    !(t->dev->flags & IFF_UP))
267 			continue;
268 
269 		if (t->dev->type != ARPHRD_IPGRE &&
270 		    t->dev->type != dev_type)
271 			continue;
272 
273 		score = 0;
274 		if (t->parms.link != link)
275 			score |= 1;
276 		if (t->dev->type != dev_type)
277 			score |= 2;
278 		if (score == 0)
279 			return t;
280 
281 		if (score < cand_score) {
282 			cand = t;
283 			cand_score = score;
284 		}
285 	}
286 
287 	for_each_ip_tunnel_rcu(ign->tunnels_wc[h1]) {
288 		if (t->parms.i_key != key ||
289 		    !(t->dev->flags & IFF_UP))
290 			continue;
291 
292 		if (t->dev->type != ARPHRD_IPGRE &&
293 		    t->dev->type != dev_type)
294 			continue;
295 
296 		score = 0;
297 		if (t->parms.link != link)
298 			score |= 1;
299 		if (t->dev->type != dev_type)
300 			score |= 2;
301 		if (score == 0)
302 			return t;
303 
304 		if (score < cand_score) {
305 			cand = t;
306 			cand_score = score;
307 		}
308 	}
309 
310 	if (cand != NULL)
311 		return cand;
312 
313 	dev = ign->fb_tunnel_dev;
314 	if (dev->flags & IFF_UP)
315 		return netdev_priv(dev);
316 
317 	return NULL;
318 }
319 
320 static struct ip_tunnel __rcu **__ipgre_bucket(struct ipgre_net *ign,
321 		struct ip_tunnel_parm *parms)
322 {
323 	__be32 remote = parms->iph.daddr;
324 	__be32 local = parms->iph.saddr;
325 	__be32 key = parms->i_key;
326 	unsigned int h = HASH(key);
327 	int prio = 0;
328 
329 	if (local)
330 		prio |= 1;
331 	if (remote && !ipv4_is_multicast(remote)) {
332 		prio |= 2;
333 		h ^= HASH(remote);
334 	}
335 
336 	return &ign->tunnels[prio][h];
337 }
338 
339 static inline struct ip_tunnel __rcu **ipgre_bucket(struct ipgre_net *ign,
340 		struct ip_tunnel *t)
341 {
342 	return __ipgre_bucket(ign, &t->parms);
343 }
344 
345 static void ipgre_tunnel_link(struct ipgre_net *ign, struct ip_tunnel *t)
346 {
347 	struct ip_tunnel __rcu **tp = ipgre_bucket(ign, t);
348 
349 	rcu_assign_pointer(t->next, rtnl_dereference(*tp));
350 	rcu_assign_pointer(*tp, t);
351 }
352 
353 static void ipgre_tunnel_unlink(struct ipgre_net *ign, struct ip_tunnel *t)
354 {
355 	struct ip_tunnel __rcu **tp;
356 	struct ip_tunnel *iter;
357 
358 	for (tp = ipgre_bucket(ign, t);
359 	     (iter = rtnl_dereference(*tp)) != NULL;
360 	     tp = &iter->next) {
361 		if (t == iter) {
362 			rcu_assign_pointer(*tp, t->next);
363 			break;
364 		}
365 	}
366 }
367 
368 static struct ip_tunnel *ipgre_tunnel_find(struct net *net,
369 					   struct ip_tunnel_parm *parms,
370 					   int type)
371 {
372 	__be32 remote = parms->iph.daddr;
373 	__be32 local = parms->iph.saddr;
374 	__be32 key = parms->i_key;
375 	int link = parms->link;
376 	struct ip_tunnel *t;
377 	struct ip_tunnel __rcu **tp;
378 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
379 
380 	for (tp = __ipgre_bucket(ign, parms);
381 	     (t = rtnl_dereference(*tp)) != NULL;
382 	     tp = &t->next)
383 		if (local == t->parms.iph.saddr &&
384 		    remote == t->parms.iph.daddr &&
385 		    key == t->parms.i_key &&
386 		    link == t->parms.link &&
387 		    type == t->dev->type)
388 			break;
389 
390 	return t;
391 }
392 
393 static struct ip_tunnel *ipgre_tunnel_locate(struct net *net,
394 		struct ip_tunnel_parm *parms, int create)
395 {
396 	struct ip_tunnel *t, *nt;
397 	struct net_device *dev;
398 	char name[IFNAMSIZ];
399 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
400 
401 	t = ipgre_tunnel_find(net, parms, ARPHRD_IPGRE);
402 	if (t || !create)
403 		return t;
404 
405 	if (parms->name[0])
406 		strlcpy(name, parms->name, IFNAMSIZ);
407 	else
408 		strcpy(name, "gre%d");
409 
410 	dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
411 	if (!dev)
412 		return NULL;
413 
414 	dev_net_set(dev, net);
415 
416 	nt = netdev_priv(dev);
417 	nt->parms = *parms;
418 	dev->rtnl_link_ops = &ipgre_link_ops;
419 
420 	dev->mtu = ipgre_tunnel_bind_dev(dev);
421 
422 	if (register_netdevice(dev) < 0)
423 		goto failed_free;
424 
425 	dev_hold(dev);
426 	ipgre_tunnel_link(ign, nt);
427 	return nt;
428 
429 failed_free:
430 	free_netdev(dev);
431 	return NULL;
432 }
433 
434 static void ipgre_tunnel_uninit(struct net_device *dev)
435 {
436 	struct net *net = dev_net(dev);
437 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
438 
439 	ipgre_tunnel_unlink(ign, netdev_priv(dev));
440 	dev_put(dev);
441 }
442 
443 
444 static void ipgre_err(struct sk_buff *skb, u32 info)
445 {
446 
447 /* All the routers (except for Linux) return only
448    8 bytes of packet payload. It means, that precise relaying of
449    ICMP in the real Internet is absolutely infeasible.
450 
451    Moreover, Cisco "wise men" put GRE key to the third word
452    in GRE header. It makes impossible maintaining even soft state for keyed
453    GRE tunnels with enabled checksum. Tell them "thank you".
454 
455    Well, I wonder, rfc1812 was written by Cisco employee,
456    what the hell these idiots break standrads established
457    by themself???
458  */
459 
460 	const struct iphdr *iph = (const struct iphdr *)skb->data;
461 	__be16	     *p = (__be16*)(skb->data+(iph->ihl<<2));
462 	int grehlen = (iph->ihl<<2) + 4;
463 	const int type = icmp_hdr(skb)->type;
464 	const int code = icmp_hdr(skb)->code;
465 	struct ip_tunnel *t;
466 	__be16 flags;
467 
468 	flags = p[0];
469 	if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
470 		if (flags&(GRE_VERSION|GRE_ROUTING))
471 			return;
472 		if (flags&GRE_KEY) {
473 			grehlen += 4;
474 			if (flags&GRE_CSUM)
475 				grehlen += 4;
476 		}
477 	}
478 
479 	/* If only 8 bytes returned, keyed message will be dropped here */
480 	if (skb_headlen(skb) < grehlen)
481 		return;
482 
483 	switch (type) {
484 	default:
485 	case ICMP_PARAMETERPROB:
486 		return;
487 
488 	case ICMP_DEST_UNREACH:
489 		switch (code) {
490 		case ICMP_SR_FAILED:
491 		case ICMP_PORT_UNREACH:
492 			/* Impossible event. */
493 			return;
494 		case ICMP_FRAG_NEEDED:
495 			/* Soft state for pmtu is maintained by IP core. */
496 			return;
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;
508 		break;
509 	}
510 
511 	rcu_read_lock();
512 	t = ipgre_tunnel_lookup(skb->dev, iph->daddr, iph->saddr,
513 				flags & GRE_KEY ?
514 				*(((__be32 *)p) + (grehlen / 4) - 1) : 0,
515 				p[1]);
516 	if (t == NULL || t->parms.iph.daddr == 0 ||
517 	    ipv4_is_multicast(t->parms.iph.daddr))
518 		goto out;
519 
520 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
521 		goto out;
522 
523 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
524 		t->err_count++;
525 	else
526 		t->err_count = 1;
527 	t->err_time = jiffies;
528 out:
529 	rcu_read_unlock();
530 }
531 
532 static inline void ipgre_ecn_decapsulate(const struct iphdr *iph, struct sk_buff *skb)
533 {
534 	if (INET_ECN_is_ce(iph->tos)) {
535 		if (skb->protocol == htons(ETH_P_IP)) {
536 			IP_ECN_set_ce(ip_hdr(skb));
537 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
538 			IP6_ECN_set_ce(ipv6_hdr(skb));
539 		}
540 	}
541 }
542 
543 static inline u8
544 ipgre_ecn_encapsulate(u8 tos, const struct iphdr *old_iph, struct sk_buff *skb)
545 {
546 	u8 inner = 0;
547 	if (skb->protocol == htons(ETH_P_IP))
548 		inner = old_iph->tos;
549 	else if (skb->protocol == htons(ETH_P_IPV6))
550 		inner = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
551 	return INET_ECN_encapsulate(tos, inner);
552 }
553 
554 static int ipgre_rcv(struct sk_buff *skb)
555 {
556 	const struct iphdr *iph;
557 	u8     *h;
558 	__be16    flags;
559 	__sum16   csum = 0;
560 	__be32 key = 0;
561 	u32    seqno = 0;
562 	struct ip_tunnel *tunnel;
563 	int    offset = 4;
564 	__be16 gre_proto;
565 
566 	if (!pskb_may_pull(skb, 16))
567 		goto drop_nolock;
568 
569 	iph = ip_hdr(skb);
570 	h = skb->data;
571 	flags = *(__be16*)h;
572 
573 	if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
574 		/* - Version must be 0.
575 		   - We do not support routing headers.
576 		 */
577 		if (flags&(GRE_VERSION|GRE_ROUTING))
578 			goto drop_nolock;
579 
580 		if (flags&GRE_CSUM) {
581 			switch (skb->ip_summed) {
582 			case CHECKSUM_COMPLETE:
583 				csum = csum_fold(skb->csum);
584 				if (!csum)
585 					break;
586 				/* fall through */
587 			case CHECKSUM_NONE:
588 				skb->csum = 0;
589 				csum = __skb_checksum_complete(skb);
590 				skb->ip_summed = CHECKSUM_COMPLETE;
591 			}
592 			offset += 4;
593 		}
594 		if (flags&GRE_KEY) {
595 			key = *(__be32*)(h + offset);
596 			offset += 4;
597 		}
598 		if (flags&GRE_SEQ) {
599 			seqno = ntohl(*(__be32*)(h + offset));
600 			offset += 4;
601 		}
602 	}
603 
604 	gre_proto = *(__be16 *)(h + 2);
605 
606 	rcu_read_lock();
607 	if ((tunnel = ipgre_tunnel_lookup(skb->dev,
608 					  iph->saddr, iph->daddr, key,
609 					  gre_proto))) {
610 		struct pcpu_tstats *tstats;
611 
612 		secpath_reset(skb);
613 
614 		skb->protocol = gre_proto;
615 		/* WCCP version 1 and 2 protocol decoding.
616 		 * - Change protocol to IP
617 		 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
618 		 */
619 		if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
620 			skb->protocol = htons(ETH_P_IP);
621 			if ((*(h + offset) & 0xF0) != 0x40)
622 				offset += 4;
623 		}
624 
625 		skb->mac_header = skb->network_header;
626 		__pskb_pull(skb, offset);
627 		skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
628 		skb->pkt_type = PACKET_HOST;
629 #ifdef CONFIG_NET_IPGRE_BROADCAST
630 		if (ipv4_is_multicast(iph->daddr)) {
631 			/* Looped back packet, drop it! */
632 			if (rt_is_output_route(skb_rtable(skb)))
633 				goto drop;
634 			tunnel->dev->stats.multicast++;
635 			skb->pkt_type = PACKET_BROADCAST;
636 		}
637 #endif
638 
639 		if (((flags&GRE_CSUM) && csum) ||
640 		    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
641 			tunnel->dev->stats.rx_crc_errors++;
642 			tunnel->dev->stats.rx_errors++;
643 			goto drop;
644 		}
645 		if (tunnel->parms.i_flags&GRE_SEQ) {
646 			if (!(flags&GRE_SEQ) ||
647 			    (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
648 				tunnel->dev->stats.rx_fifo_errors++;
649 				tunnel->dev->stats.rx_errors++;
650 				goto drop;
651 			}
652 			tunnel->i_seqno = seqno + 1;
653 		}
654 
655 		/* Warning: All skb pointers will be invalidated! */
656 		if (tunnel->dev->type == ARPHRD_ETHER) {
657 			if (!pskb_may_pull(skb, ETH_HLEN)) {
658 				tunnel->dev->stats.rx_length_errors++;
659 				tunnel->dev->stats.rx_errors++;
660 				goto drop;
661 			}
662 
663 			iph = ip_hdr(skb);
664 			skb->protocol = eth_type_trans(skb, tunnel->dev);
665 			skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
666 		}
667 
668 		tstats = this_cpu_ptr(tunnel->dev->tstats);
669 		tstats->rx_packets++;
670 		tstats->rx_bytes += skb->len;
671 
672 		__skb_tunnel_rx(skb, tunnel->dev);
673 
674 		skb_reset_network_header(skb);
675 		ipgre_ecn_decapsulate(iph, skb);
676 
677 		netif_rx(skb);
678 
679 		rcu_read_unlock();
680 		return 0;
681 	}
682 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
683 
684 drop:
685 	rcu_read_unlock();
686 drop_nolock:
687 	kfree_skb(skb);
688 	return 0;
689 }
690 
691 static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
692 {
693 	struct ip_tunnel *tunnel = netdev_priv(dev);
694 	struct pcpu_tstats *tstats;
695 	const struct iphdr  *old_iph = ip_hdr(skb);
696 	const struct iphdr  *tiph;
697 	struct flowi4 fl4;
698 	u8     tos;
699 	__be16 df;
700 	struct rtable *rt;     			/* Route to the other host */
701 	struct net_device *tdev;		/* Device to other host */
702 	struct iphdr  *iph;			/* Our new IP header */
703 	unsigned int max_headroom;		/* The extra header space needed */
704 	int    gre_hlen;
705 	__be32 dst;
706 	int    mtu;
707 
708 	if (dev->type == ARPHRD_ETHER)
709 		IPCB(skb)->flags = 0;
710 
711 	if (dev->header_ops && dev->type == ARPHRD_IPGRE) {
712 		gre_hlen = 0;
713 		tiph = (const struct iphdr *)skb->data;
714 	} else {
715 		gre_hlen = tunnel->hlen;
716 		tiph = &tunnel->parms.iph;
717 	}
718 
719 	if ((dst = tiph->daddr) == 0) {
720 		/* NBMA tunnel */
721 
722 		if (skb_dst(skb) == NULL) {
723 			dev->stats.tx_fifo_errors++;
724 			goto tx_error;
725 		}
726 
727 		if (skb->protocol == htons(ETH_P_IP)) {
728 			rt = skb_rtable(skb);
729 			if ((dst = rt->rt_gateway) == 0)
730 				goto tx_error_icmp;
731 		}
732 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
733 		else if (skb->protocol == htons(ETH_P_IPV6)) {
734 			struct neighbour *neigh = dst_get_neighbour(skb_dst(skb));
735 			const struct in6_addr *addr6;
736 			int addr_type;
737 
738 			if (neigh == NULL)
739 				goto tx_error;
740 
741 			addr6 = (const struct in6_addr *)&neigh->primary_key;
742 			addr_type = ipv6_addr_type(addr6);
743 
744 			if (addr_type == IPV6_ADDR_ANY) {
745 				addr6 = &ipv6_hdr(skb)->daddr;
746 				addr_type = ipv6_addr_type(addr6);
747 			}
748 
749 			if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
750 				goto tx_error_icmp;
751 
752 			dst = addr6->s6_addr32[3];
753 		}
754 #endif
755 		else
756 			goto tx_error;
757 	}
758 
759 	tos = tiph->tos;
760 	if (tos == 1) {
761 		tos = 0;
762 		if (skb->protocol == htons(ETH_P_IP))
763 			tos = old_iph->tos;
764 		else if (skb->protocol == htons(ETH_P_IPV6))
765 			tos = ipv6_get_dsfield((const struct ipv6hdr *)old_iph);
766 	}
767 
768 	rt = ip_route_output_gre(dev_net(dev), &fl4, dst, tiph->saddr,
769 				 tunnel->parms.o_key, RT_TOS(tos),
770 				 tunnel->parms.link);
771 	if (IS_ERR(rt)) {
772 		dev->stats.tx_carrier_errors++;
773 		goto tx_error;
774 	}
775 	tdev = rt->dst.dev;
776 
777 	if (tdev == dev) {
778 		ip_rt_put(rt);
779 		dev->stats.collisions++;
780 		goto tx_error;
781 	}
782 
783 	df = tiph->frag_off;
784 	if (df)
785 		mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen;
786 	else
787 		mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
788 
789 	if (skb_dst(skb))
790 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
791 
792 	if (skb->protocol == htons(ETH_P_IP)) {
793 		df |= (old_iph->frag_off&htons(IP_DF));
794 
795 		if ((old_iph->frag_off&htons(IP_DF)) &&
796 		    mtu < ntohs(old_iph->tot_len)) {
797 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
798 			ip_rt_put(rt);
799 			goto tx_error;
800 		}
801 	}
802 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
803 	else if (skb->protocol == htons(ETH_P_IPV6)) {
804 		struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
805 
806 		if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) {
807 			if ((tunnel->parms.iph.daddr &&
808 			     !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
809 			    rt6->rt6i_dst.plen == 128) {
810 				rt6->rt6i_flags |= RTF_MODIFIED;
811 				dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
812 			}
813 		}
814 
815 		if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
816 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
817 			ip_rt_put(rt);
818 			goto tx_error;
819 		}
820 	}
821 #endif
822 
823 	if (tunnel->err_count > 0) {
824 		if (time_before(jiffies,
825 				tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
826 			tunnel->err_count--;
827 
828 			dst_link_failure(skb);
829 		} else
830 			tunnel->err_count = 0;
831 	}
832 
833 	max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len;
834 
835 	if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
836 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
837 		struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
838 		if (!new_skb) {
839 			ip_rt_put(rt);
840 			dev->stats.tx_dropped++;
841 			dev_kfree_skb(skb);
842 			return NETDEV_TX_OK;
843 		}
844 		if (skb->sk)
845 			skb_set_owner_w(new_skb, skb->sk);
846 		dev_kfree_skb(skb);
847 		skb = new_skb;
848 		old_iph = ip_hdr(skb);
849 	}
850 
851 	skb_reset_transport_header(skb);
852 	skb_push(skb, gre_hlen);
853 	skb_reset_network_header(skb);
854 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
855 	IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
856 			      IPSKB_REROUTED);
857 	skb_dst_drop(skb);
858 	skb_dst_set(skb, &rt->dst);
859 
860 	/*
861 	 *	Push down and install the IPIP header.
862 	 */
863 
864 	iph 			=	ip_hdr(skb);
865 	iph->version		=	4;
866 	iph->ihl		=	sizeof(struct iphdr) >> 2;
867 	iph->frag_off		=	df;
868 	iph->protocol		=	IPPROTO_GRE;
869 	iph->tos		=	ipgre_ecn_encapsulate(tos, old_iph, skb);
870 	iph->daddr		=	fl4.daddr;
871 	iph->saddr		=	fl4.saddr;
872 
873 	if ((iph->ttl = tiph->ttl) == 0) {
874 		if (skb->protocol == htons(ETH_P_IP))
875 			iph->ttl = old_iph->ttl;
876 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
877 		else if (skb->protocol == htons(ETH_P_IPV6))
878 			iph->ttl = ((const struct ipv6hdr *)old_iph)->hop_limit;
879 #endif
880 		else
881 			iph->ttl = ip4_dst_hoplimit(&rt->dst);
882 	}
883 
884 	((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags;
885 	((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ?
886 				   htons(ETH_P_TEB) : skb->protocol;
887 
888 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
889 		__be32 *ptr = (__be32*)(((u8*)iph) + tunnel->hlen - 4);
890 
891 		if (tunnel->parms.o_flags&GRE_SEQ) {
892 			++tunnel->o_seqno;
893 			*ptr = htonl(tunnel->o_seqno);
894 			ptr--;
895 		}
896 		if (tunnel->parms.o_flags&GRE_KEY) {
897 			*ptr = tunnel->parms.o_key;
898 			ptr--;
899 		}
900 		if (tunnel->parms.o_flags&GRE_CSUM) {
901 			*ptr = 0;
902 			*(__sum16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr));
903 		}
904 	}
905 
906 	nf_reset(skb);
907 	tstats = this_cpu_ptr(dev->tstats);
908 	__IPTUNNEL_XMIT(tstats, &dev->stats);
909 	return NETDEV_TX_OK;
910 
911 tx_error_icmp:
912 	dst_link_failure(skb);
913 
914 tx_error:
915 	dev->stats.tx_errors++;
916 	dev_kfree_skb(skb);
917 	return NETDEV_TX_OK;
918 }
919 
920 static int ipgre_tunnel_bind_dev(struct net_device *dev)
921 {
922 	struct net_device *tdev = NULL;
923 	struct ip_tunnel *tunnel;
924 	const struct iphdr *iph;
925 	int hlen = LL_MAX_HEADER;
926 	int mtu = ETH_DATA_LEN;
927 	int addend = sizeof(struct iphdr) + 4;
928 
929 	tunnel = netdev_priv(dev);
930 	iph = &tunnel->parms.iph;
931 
932 	/* Guess output device to choose reasonable mtu and needed_headroom */
933 
934 	if (iph->daddr) {
935 		struct flowi4 fl4;
936 		struct rtable *rt;
937 
938 		rt = ip_route_output_gre(dev_net(dev), &fl4,
939 					 iph->daddr, iph->saddr,
940 					 tunnel->parms.o_key,
941 					 RT_TOS(iph->tos),
942 					 tunnel->parms.link);
943 		if (!IS_ERR(rt)) {
944 			tdev = rt->dst.dev;
945 			ip_rt_put(rt);
946 		}
947 
948 		if (dev->type != ARPHRD_ETHER)
949 			dev->flags |= IFF_POINTOPOINT;
950 	}
951 
952 	if (!tdev && tunnel->parms.link)
953 		tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
954 
955 	if (tdev) {
956 		hlen = tdev->hard_header_len + tdev->needed_headroom;
957 		mtu = tdev->mtu;
958 	}
959 	dev->iflink = tunnel->parms.link;
960 
961 	/* Precalculate GRE options length */
962 	if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
963 		if (tunnel->parms.o_flags&GRE_CSUM)
964 			addend += 4;
965 		if (tunnel->parms.o_flags&GRE_KEY)
966 			addend += 4;
967 		if (tunnel->parms.o_flags&GRE_SEQ)
968 			addend += 4;
969 	}
970 	dev->needed_headroom = addend + hlen;
971 	mtu -= dev->hard_header_len + addend;
972 
973 	if (mtu < 68)
974 		mtu = 68;
975 
976 	tunnel->hlen = addend;
977 
978 	return mtu;
979 }
980 
981 static int
982 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
983 {
984 	int err = 0;
985 	struct ip_tunnel_parm p;
986 	struct ip_tunnel *t;
987 	struct net *net = dev_net(dev);
988 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
989 
990 	switch (cmd) {
991 	case SIOCGETTUNNEL:
992 		t = NULL;
993 		if (dev == ign->fb_tunnel_dev) {
994 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
995 				err = -EFAULT;
996 				break;
997 			}
998 			t = ipgre_tunnel_locate(net, &p, 0);
999 		}
1000 		if (t == NULL)
1001 			t = netdev_priv(dev);
1002 		memcpy(&p, &t->parms, sizeof(p));
1003 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1004 			err = -EFAULT;
1005 		break;
1006 
1007 	case SIOCADDTUNNEL:
1008 	case SIOCCHGTUNNEL:
1009 		err = -EPERM;
1010 		if (!capable(CAP_NET_ADMIN))
1011 			goto done;
1012 
1013 		err = -EFAULT;
1014 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1015 			goto done;
1016 
1017 		err = -EINVAL;
1018 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
1019 		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
1020 		    ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
1021 			goto done;
1022 		if (p.iph.ttl)
1023 			p.iph.frag_off |= htons(IP_DF);
1024 
1025 		if (!(p.i_flags&GRE_KEY))
1026 			p.i_key = 0;
1027 		if (!(p.o_flags&GRE_KEY))
1028 			p.o_key = 0;
1029 
1030 		t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
1031 
1032 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1033 			if (t != NULL) {
1034 				if (t->dev != dev) {
1035 					err = -EEXIST;
1036 					break;
1037 				}
1038 			} else {
1039 				unsigned int nflags = 0;
1040 
1041 				t = netdev_priv(dev);
1042 
1043 				if (ipv4_is_multicast(p.iph.daddr))
1044 					nflags = IFF_BROADCAST;
1045 				else if (p.iph.daddr)
1046 					nflags = IFF_POINTOPOINT;
1047 
1048 				if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
1049 					err = -EINVAL;
1050 					break;
1051 				}
1052 				ipgre_tunnel_unlink(ign, t);
1053 				synchronize_net();
1054 				t->parms.iph.saddr = p.iph.saddr;
1055 				t->parms.iph.daddr = p.iph.daddr;
1056 				t->parms.i_key = p.i_key;
1057 				t->parms.o_key = p.o_key;
1058 				memcpy(dev->dev_addr, &p.iph.saddr, 4);
1059 				memcpy(dev->broadcast, &p.iph.daddr, 4);
1060 				ipgre_tunnel_link(ign, t);
1061 				netdev_state_change(dev);
1062 			}
1063 		}
1064 
1065 		if (t) {
1066 			err = 0;
1067 			if (cmd == SIOCCHGTUNNEL) {
1068 				t->parms.iph.ttl = p.iph.ttl;
1069 				t->parms.iph.tos = p.iph.tos;
1070 				t->parms.iph.frag_off = p.iph.frag_off;
1071 				if (t->parms.link != p.link) {
1072 					t->parms.link = p.link;
1073 					dev->mtu = ipgre_tunnel_bind_dev(dev);
1074 					netdev_state_change(dev);
1075 				}
1076 			}
1077 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
1078 				err = -EFAULT;
1079 		} else
1080 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1081 		break;
1082 
1083 	case SIOCDELTUNNEL:
1084 		err = -EPERM;
1085 		if (!capable(CAP_NET_ADMIN))
1086 			goto done;
1087 
1088 		if (dev == ign->fb_tunnel_dev) {
1089 			err = -EFAULT;
1090 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1091 				goto done;
1092 			err = -ENOENT;
1093 			if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL)
1094 				goto done;
1095 			err = -EPERM;
1096 			if (t == netdev_priv(ign->fb_tunnel_dev))
1097 				goto done;
1098 			dev = t->dev;
1099 		}
1100 		unregister_netdevice(dev);
1101 		err = 0;
1102 		break;
1103 
1104 	default:
1105 		err = -EINVAL;
1106 	}
1107 
1108 done:
1109 	return err;
1110 }
1111 
1112 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1113 {
1114 	struct ip_tunnel *tunnel = netdev_priv(dev);
1115 	if (new_mtu < 68 ||
1116 	    new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen)
1117 		return -EINVAL;
1118 	dev->mtu = new_mtu;
1119 	return 0;
1120 }
1121 
1122 /* Nice toy. Unfortunately, useless in real life :-)
1123    It allows to construct virtual multiprotocol broadcast "LAN"
1124    over the Internet, provided multicast routing is tuned.
1125 
1126 
1127    I have no idea was this bicycle invented before me,
1128    so that I had to set ARPHRD_IPGRE to a random value.
1129    I have an impression, that Cisco could make something similar,
1130    but this feature is apparently missing in IOS<=11.2(8).
1131 
1132    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1133    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1134 
1135    ping -t 255 224.66.66.66
1136 
1137    If nobody answers, mbone does not work.
1138 
1139    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1140    ip addr add 10.66.66.<somewhat>/24 dev Universe
1141    ifconfig Universe up
1142    ifconfig Universe add fe80::<Your_real_addr>/10
1143    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1144    ftp 10.66.66.66
1145    ...
1146    ftp fec0:6666:6666::193.233.7.65
1147    ...
1148 
1149  */
1150 
1151 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
1152 			unsigned short type,
1153 			const void *daddr, const void *saddr, unsigned int len)
1154 {
1155 	struct ip_tunnel *t = netdev_priv(dev);
1156 	struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1157 	__be16 *p = (__be16*)(iph+1);
1158 
1159 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1160 	p[0]		= t->parms.o_flags;
1161 	p[1]		= htons(type);
1162 
1163 	/*
1164 	 *	Set the source hardware address.
1165 	 */
1166 
1167 	if (saddr)
1168 		memcpy(&iph->saddr, saddr, 4);
1169 	if (daddr)
1170 		memcpy(&iph->daddr, daddr, 4);
1171 	if (iph->daddr)
1172 		return t->hlen;
1173 
1174 	return -t->hlen;
1175 }
1176 
1177 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
1178 {
1179 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
1180 	memcpy(haddr, &iph->saddr, 4);
1181 	return 4;
1182 }
1183 
1184 static const struct header_ops ipgre_header_ops = {
1185 	.create	= ipgre_header,
1186 	.parse	= ipgre_header_parse,
1187 };
1188 
1189 #ifdef CONFIG_NET_IPGRE_BROADCAST
1190 static int ipgre_open(struct net_device *dev)
1191 {
1192 	struct ip_tunnel *t = netdev_priv(dev);
1193 
1194 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
1195 		struct flowi4 fl4;
1196 		struct rtable *rt;
1197 
1198 		rt = ip_route_output_gre(dev_net(dev), &fl4,
1199 					 t->parms.iph.daddr,
1200 					 t->parms.iph.saddr,
1201 					 t->parms.o_key,
1202 					 RT_TOS(t->parms.iph.tos),
1203 					 t->parms.link);
1204 		if (IS_ERR(rt))
1205 			return -EADDRNOTAVAIL;
1206 		dev = rt->dst.dev;
1207 		ip_rt_put(rt);
1208 		if (__in_dev_get_rtnl(dev) == NULL)
1209 			return -EADDRNOTAVAIL;
1210 		t->mlink = dev->ifindex;
1211 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1212 	}
1213 	return 0;
1214 }
1215 
1216 static int ipgre_close(struct net_device *dev)
1217 {
1218 	struct ip_tunnel *t = netdev_priv(dev);
1219 
1220 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
1221 		struct in_device *in_dev;
1222 		in_dev = inetdev_by_index(dev_net(dev), t->mlink);
1223 		if (in_dev)
1224 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1225 	}
1226 	return 0;
1227 }
1228 
1229 #endif
1230 
1231 static const struct net_device_ops ipgre_netdev_ops = {
1232 	.ndo_init		= ipgre_tunnel_init,
1233 	.ndo_uninit		= ipgre_tunnel_uninit,
1234 #ifdef CONFIG_NET_IPGRE_BROADCAST
1235 	.ndo_open		= ipgre_open,
1236 	.ndo_stop		= ipgre_close,
1237 #endif
1238 	.ndo_start_xmit		= ipgre_tunnel_xmit,
1239 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
1240 	.ndo_change_mtu		= ipgre_tunnel_change_mtu,
1241 	.ndo_get_stats		= ipgre_get_stats,
1242 };
1243 
1244 static void ipgre_dev_free(struct net_device *dev)
1245 {
1246 	free_percpu(dev->tstats);
1247 	free_netdev(dev);
1248 }
1249 
1250 static void ipgre_tunnel_setup(struct net_device *dev)
1251 {
1252 	dev->netdev_ops		= &ipgre_netdev_ops;
1253 	dev->destructor 	= ipgre_dev_free;
1254 
1255 	dev->type		= ARPHRD_IPGRE;
1256 	dev->needed_headroom 	= LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1257 	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1258 	dev->flags		= IFF_NOARP;
1259 	dev->iflink		= 0;
1260 	dev->addr_len		= 4;
1261 	dev->features		|= NETIF_F_NETNS_LOCAL;
1262 	dev->priv_flags		&= ~IFF_XMIT_DST_RELEASE;
1263 }
1264 
1265 static int ipgre_tunnel_init(struct net_device *dev)
1266 {
1267 	struct ip_tunnel *tunnel;
1268 	struct iphdr *iph;
1269 
1270 	tunnel = netdev_priv(dev);
1271 	iph = &tunnel->parms.iph;
1272 
1273 	tunnel->dev = dev;
1274 	strcpy(tunnel->parms.name, dev->name);
1275 
1276 	memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1277 	memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1278 
1279 	if (iph->daddr) {
1280 #ifdef CONFIG_NET_IPGRE_BROADCAST
1281 		if (ipv4_is_multicast(iph->daddr)) {
1282 			if (!iph->saddr)
1283 				return -EINVAL;
1284 			dev->flags = IFF_BROADCAST;
1285 			dev->header_ops = &ipgre_header_ops;
1286 		}
1287 #endif
1288 	} else
1289 		dev->header_ops = &ipgre_header_ops;
1290 
1291 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1292 	if (!dev->tstats)
1293 		return -ENOMEM;
1294 
1295 	return 0;
1296 }
1297 
1298 static void ipgre_fb_tunnel_init(struct net_device *dev)
1299 {
1300 	struct ip_tunnel *tunnel = netdev_priv(dev);
1301 	struct iphdr *iph = &tunnel->parms.iph;
1302 
1303 	tunnel->dev = dev;
1304 	strcpy(tunnel->parms.name, dev->name);
1305 
1306 	iph->version		= 4;
1307 	iph->protocol		= IPPROTO_GRE;
1308 	iph->ihl		= 5;
1309 	tunnel->hlen		= sizeof(struct iphdr) + 4;
1310 
1311 	dev_hold(dev);
1312 }
1313 
1314 
1315 static const struct gre_protocol ipgre_protocol = {
1316 	.handler     = ipgre_rcv,
1317 	.err_handler = ipgre_err,
1318 };
1319 
1320 static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head)
1321 {
1322 	int prio;
1323 
1324 	for (prio = 0; prio < 4; prio++) {
1325 		int h;
1326 		for (h = 0; h < HASH_SIZE; h++) {
1327 			struct ip_tunnel *t;
1328 
1329 			t = rtnl_dereference(ign->tunnels[prio][h]);
1330 
1331 			while (t != NULL) {
1332 				unregister_netdevice_queue(t->dev, head);
1333 				t = rtnl_dereference(t->next);
1334 			}
1335 		}
1336 	}
1337 }
1338 
1339 static int __net_init ipgre_init_net(struct net *net)
1340 {
1341 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1342 	int err;
1343 
1344 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1345 					   ipgre_tunnel_setup);
1346 	if (!ign->fb_tunnel_dev) {
1347 		err = -ENOMEM;
1348 		goto err_alloc_dev;
1349 	}
1350 	dev_net_set(ign->fb_tunnel_dev, net);
1351 
1352 	ipgre_fb_tunnel_init(ign->fb_tunnel_dev);
1353 	ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops;
1354 
1355 	if ((err = register_netdev(ign->fb_tunnel_dev)))
1356 		goto err_reg_dev;
1357 
1358 	rcu_assign_pointer(ign->tunnels_wc[0],
1359 			   netdev_priv(ign->fb_tunnel_dev));
1360 	return 0;
1361 
1362 err_reg_dev:
1363 	ipgre_dev_free(ign->fb_tunnel_dev);
1364 err_alloc_dev:
1365 	return err;
1366 }
1367 
1368 static void __net_exit ipgre_exit_net(struct net *net)
1369 {
1370 	struct ipgre_net *ign;
1371 	LIST_HEAD(list);
1372 
1373 	ign = net_generic(net, ipgre_net_id);
1374 	rtnl_lock();
1375 	ipgre_destroy_tunnels(ign, &list);
1376 	unregister_netdevice_many(&list);
1377 	rtnl_unlock();
1378 }
1379 
1380 static struct pernet_operations ipgre_net_ops = {
1381 	.init = ipgre_init_net,
1382 	.exit = ipgre_exit_net,
1383 	.id   = &ipgre_net_id,
1384 	.size = sizeof(struct ipgre_net),
1385 };
1386 
1387 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1388 {
1389 	__be16 flags;
1390 
1391 	if (!data)
1392 		return 0;
1393 
1394 	flags = 0;
1395 	if (data[IFLA_GRE_IFLAGS])
1396 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1397 	if (data[IFLA_GRE_OFLAGS])
1398 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1399 	if (flags & (GRE_VERSION|GRE_ROUTING))
1400 		return -EINVAL;
1401 
1402 	return 0;
1403 }
1404 
1405 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1406 {
1407 	__be32 daddr;
1408 
1409 	if (tb[IFLA_ADDRESS]) {
1410 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1411 			return -EINVAL;
1412 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1413 			return -EADDRNOTAVAIL;
1414 	}
1415 
1416 	if (!data)
1417 		goto out;
1418 
1419 	if (data[IFLA_GRE_REMOTE]) {
1420 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1421 		if (!daddr)
1422 			return -EINVAL;
1423 	}
1424 
1425 out:
1426 	return ipgre_tunnel_validate(tb, data);
1427 }
1428 
1429 static void ipgre_netlink_parms(struct nlattr *data[],
1430 				struct ip_tunnel_parm *parms)
1431 {
1432 	memset(parms, 0, sizeof(*parms));
1433 
1434 	parms->iph.protocol = IPPROTO_GRE;
1435 
1436 	if (!data)
1437 		return;
1438 
1439 	if (data[IFLA_GRE_LINK])
1440 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1441 
1442 	if (data[IFLA_GRE_IFLAGS])
1443 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1444 
1445 	if (data[IFLA_GRE_OFLAGS])
1446 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1447 
1448 	if (data[IFLA_GRE_IKEY])
1449 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1450 
1451 	if (data[IFLA_GRE_OKEY])
1452 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1453 
1454 	if (data[IFLA_GRE_LOCAL])
1455 		parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
1456 
1457 	if (data[IFLA_GRE_REMOTE])
1458 		parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
1459 
1460 	if (data[IFLA_GRE_TTL])
1461 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1462 
1463 	if (data[IFLA_GRE_TOS])
1464 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1465 
1466 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
1467 		parms->iph.frag_off = htons(IP_DF);
1468 }
1469 
1470 static int ipgre_tap_init(struct net_device *dev)
1471 {
1472 	struct ip_tunnel *tunnel;
1473 
1474 	tunnel = netdev_priv(dev);
1475 
1476 	tunnel->dev = dev;
1477 	strcpy(tunnel->parms.name, dev->name);
1478 
1479 	ipgre_tunnel_bind_dev(dev);
1480 
1481 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1482 	if (!dev->tstats)
1483 		return -ENOMEM;
1484 
1485 	return 0;
1486 }
1487 
1488 static const struct net_device_ops ipgre_tap_netdev_ops = {
1489 	.ndo_init		= ipgre_tap_init,
1490 	.ndo_uninit		= ipgre_tunnel_uninit,
1491 	.ndo_start_xmit		= ipgre_tunnel_xmit,
1492 	.ndo_set_mac_address 	= eth_mac_addr,
1493 	.ndo_validate_addr	= eth_validate_addr,
1494 	.ndo_change_mtu		= ipgre_tunnel_change_mtu,
1495 	.ndo_get_stats		= ipgre_get_stats,
1496 };
1497 
1498 static void ipgre_tap_setup(struct net_device *dev)
1499 {
1500 
1501 	ether_setup(dev);
1502 
1503 	dev->netdev_ops		= &ipgre_tap_netdev_ops;
1504 	dev->destructor 	= ipgre_dev_free;
1505 
1506 	dev->iflink		= 0;
1507 	dev->features		|= NETIF_F_NETNS_LOCAL;
1508 }
1509 
1510 static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[],
1511 			 struct nlattr *data[])
1512 {
1513 	struct ip_tunnel *nt;
1514 	struct net *net = dev_net(dev);
1515 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1516 	int mtu;
1517 	int err;
1518 
1519 	nt = netdev_priv(dev);
1520 	ipgre_netlink_parms(data, &nt->parms);
1521 
1522 	if (ipgre_tunnel_find(net, &nt->parms, dev->type))
1523 		return -EEXIST;
1524 
1525 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1526 		random_ether_addr(dev->dev_addr);
1527 
1528 	mtu = ipgre_tunnel_bind_dev(dev);
1529 	if (!tb[IFLA_MTU])
1530 		dev->mtu = mtu;
1531 
1532 	/* Can use a lockless transmit, unless we generate output sequences */
1533 	if (!(nt->parms.o_flags & GRE_SEQ))
1534 		dev->features |= NETIF_F_LLTX;
1535 
1536 	err = register_netdevice(dev);
1537 	if (err)
1538 		goto out;
1539 
1540 	dev_hold(dev);
1541 	ipgre_tunnel_link(ign, nt);
1542 
1543 out:
1544 	return err;
1545 }
1546 
1547 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1548 			    struct nlattr *data[])
1549 {
1550 	struct ip_tunnel *t, *nt;
1551 	struct net *net = dev_net(dev);
1552 	struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1553 	struct ip_tunnel_parm p;
1554 	int mtu;
1555 
1556 	if (dev == ign->fb_tunnel_dev)
1557 		return -EINVAL;
1558 
1559 	nt = netdev_priv(dev);
1560 	ipgre_netlink_parms(data, &p);
1561 
1562 	t = ipgre_tunnel_locate(net, &p, 0);
1563 
1564 	if (t) {
1565 		if (t->dev != dev)
1566 			return -EEXIST;
1567 	} else {
1568 		t = nt;
1569 
1570 		if (dev->type != ARPHRD_ETHER) {
1571 			unsigned int nflags = 0;
1572 
1573 			if (ipv4_is_multicast(p.iph.daddr))
1574 				nflags = IFF_BROADCAST;
1575 			else if (p.iph.daddr)
1576 				nflags = IFF_POINTOPOINT;
1577 
1578 			if ((dev->flags ^ nflags) &
1579 			    (IFF_POINTOPOINT | IFF_BROADCAST))
1580 				return -EINVAL;
1581 		}
1582 
1583 		ipgre_tunnel_unlink(ign, t);
1584 		t->parms.iph.saddr = p.iph.saddr;
1585 		t->parms.iph.daddr = p.iph.daddr;
1586 		t->parms.i_key = p.i_key;
1587 		if (dev->type != ARPHRD_ETHER) {
1588 			memcpy(dev->dev_addr, &p.iph.saddr, 4);
1589 			memcpy(dev->broadcast, &p.iph.daddr, 4);
1590 		}
1591 		ipgre_tunnel_link(ign, t);
1592 		netdev_state_change(dev);
1593 	}
1594 
1595 	t->parms.o_key = p.o_key;
1596 	t->parms.iph.ttl = p.iph.ttl;
1597 	t->parms.iph.tos = p.iph.tos;
1598 	t->parms.iph.frag_off = p.iph.frag_off;
1599 
1600 	if (t->parms.link != p.link) {
1601 		t->parms.link = p.link;
1602 		mtu = ipgre_tunnel_bind_dev(dev);
1603 		if (!tb[IFLA_MTU])
1604 			dev->mtu = mtu;
1605 		netdev_state_change(dev);
1606 	}
1607 
1608 	return 0;
1609 }
1610 
1611 static size_t ipgre_get_size(const struct net_device *dev)
1612 {
1613 	return
1614 		/* IFLA_GRE_LINK */
1615 		nla_total_size(4) +
1616 		/* IFLA_GRE_IFLAGS */
1617 		nla_total_size(2) +
1618 		/* IFLA_GRE_OFLAGS */
1619 		nla_total_size(2) +
1620 		/* IFLA_GRE_IKEY */
1621 		nla_total_size(4) +
1622 		/* IFLA_GRE_OKEY */
1623 		nla_total_size(4) +
1624 		/* IFLA_GRE_LOCAL */
1625 		nla_total_size(4) +
1626 		/* IFLA_GRE_REMOTE */
1627 		nla_total_size(4) +
1628 		/* IFLA_GRE_TTL */
1629 		nla_total_size(1) +
1630 		/* IFLA_GRE_TOS */
1631 		nla_total_size(1) +
1632 		/* IFLA_GRE_PMTUDISC */
1633 		nla_total_size(1) +
1634 		0;
1635 }
1636 
1637 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1638 {
1639 	struct ip_tunnel *t = netdev_priv(dev);
1640 	struct ip_tunnel_parm *p = &t->parms;
1641 
1642 	NLA_PUT_U32(skb, IFLA_GRE_LINK, p->link);
1643 	NLA_PUT_BE16(skb, IFLA_GRE_IFLAGS, p->i_flags);
1644 	NLA_PUT_BE16(skb, IFLA_GRE_OFLAGS, p->o_flags);
1645 	NLA_PUT_BE32(skb, IFLA_GRE_IKEY, p->i_key);
1646 	NLA_PUT_BE32(skb, IFLA_GRE_OKEY, p->o_key);
1647 	NLA_PUT_BE32(skb, IFLA_GRE_LOCAL, p->iph.saddr);
1648 	NLA_PUT_BE32(skb, IFLA_GRE_REMOTE, p->iph.daddr);
1649 	NLA_PUT_U8(skb, IFLA_GRE_TTL, p->iph.ttl);
1650 	NLA_PUT_U8(skb, IFLA_GRE_TOS, p->iph.tos);
1651 	NLA_PUT_U8(skb, IFLA_GRE_PMTUDISC, !!(p->iph.frag_off & htons(IP_DF)));
1652 
1653 	return 0;
1654 
1655 nla_put_failure:
1656 	return -EMSGSIZE;
1657 }
1658 
1659 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1660 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1661 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1662 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1663 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1664 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1665 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1666 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1667 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1668 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1669 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1670 };
1671 
1672 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1673 	.kind		= "gre",
1674 	.maxtype	= IFLA_GRE_MAX,
1675 	.policy		= ipgre_policy,
1676 	.priv_size	= sizeof(struct ip_tunnel),
1677 	.setup		= ipgre_tunnel_setup,
1678 	.validate	= ipgre_tunnel_validate,
1679 	.newlink	= ipgre_newlink,
1680 	.changelink	= ipgre_changelink,
1681 	.get_size	= ipgre_get_size,
1682 	.fill_info	= ipgre_fill_info,
1683 };
1684 
1685 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1686 	.kind		= "gretap",
1687 	.maxtype	= IFLA_GRE_MAX,
1688 	.policy		= ipgre_policy,
1689 	.priv_size	= sizeof(struct ip_tunnel),
1690 	.setup		= ipgre_tap_setup,
1691 	.validate	= ipgre_tap_validate,
1692 	.newlink	= ipgre_newlink,
1693 	.changelink	= ipgre_changelink,
1694 	.get_size	= ipgre_get_size,
1695 	.fill_info	= ipgre_fill_info,
1696 };
1697 
1698 /*
1699  *	And now the modules code and kernel interface.
1700  */
1701 
1702 static int __init ipgre_init(void)
1703 {
1704 	int err;
1705 
1706 	printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1707 
1708 	err = register_pernet_device(&ipgre_net_ops);
1709 	if (err < 0)
1710 		return err;
1711 
1712 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1713 	if (err < 0) {
1714 		printk(KERN_INFO "ipgre init: can't add protocol\n");
1715 		goto add_proto_failed;
1716 	}
1717 
1718 	err = rtnl_link_register(&ipgre_link_ops);
1719 	if (err < 0)
1720 		goto rtnl_link_failed;
1721 
1722 	err = rtnl_link_register(&ipgre_tap_ops);
1723 	if (err < 0)
1724 		goto tap_ops_failed;
1725 
1726 out:
1727 	return err;
1728 
1729 tap_ops_failed:
1730 	rtnl_link_unregister(&ipgre_link_ops);
1731 rtnl_link_failed:
1732 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1733 add_proto_failed:
1734 	unregister_pernet_device(&ipgre_net_ops);
1735 	goto out;
1736 }
1737 
1738 static void __exit ipgre_fini(void)
1739 {
1740 	rtnl_link_unregister(&ipgre_tap_ops);
1741 	rtnl_link_unregister(&ipgre_link_ops);
1742 	if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0)
1743 		printk(KERN_INFO "ipgre close: can't remove protocol\n");
1744 	unregister_pernet_device(&ipgre_net_ops);
1745 }
1746 
1747 module_init(ipgre_init);
1748 module_exit(ipgre_fini);
1749 MODULE_LICENSE("GPL");
1750 MODULE_ALIAS_RTNL_LINK("gre");
1751 MODULE_ALIAS_RTNL_LINK("gretap");
1752 MODULE_ALIAS_NETDEV("gre0");
1753