xref: /linux/net/ipv6/ip6_gre.c (revision 2d87650a3bf1b80f7d0d150ee1af3f8a89e5b7aa)
1 /*
2  *	GRE over IPv6 protocol decoder.
3  *
4  *	Authors: Dmitry Kozlov (xeb@mail.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/capability.h>
16 #include <linux/module.h>
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/slab.h>
20 #include <linux/uaccess.h>
21 #include <linux/skbuff.h>
22 #include <linux/netdevice.h>
23 #include <linux/in.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/if_arp.h>
27 #include <linux/mroute.h>
28 #include <linux/init.h>
29 #include <linux/in6.h>
30 #include <linux/inetdevice.h>
31 #include <linux/igmp.h>
32 #include <linux/netfilter_ipv4.h>
33 #include <linux/etherdevice.h>
34 #include <linux/if_ether.h>
35 #include <linux/hash.h>
36 #include <linux/if_tunnel.h>
37 #include <linux/ip6_tunnel.h>
38 
39 #include <net/sock.h>
40 #include <net/ip.h>
41 #include <net/ip_tunnels.h>
42 #include <net/icmp.h>
43 #include <net/protocol.h>
44 #include <net/addrconf.h>
45 #include <net/arp.h>
46 #include <net/checksum.h>
47 #include <net/dsfield.h>
48 #include <net/inet_ecn.h>
49 #include <net/xfrm.h>
50 #include <net/net_namespace.h>
51 #include <net/netns/generic.h>
52 #include <net/rtnetlink.h>
53 
54 #include <net/ipv6.h>
55 #include <net/ip6_fib.h>
56 #include <net/ip6_route.h>
57 #include <net/ip6_tunnel.h>
58 
59 
60 static bool log_ecn_error = true;
61 module_param(log_ecn_error, bool, 0644);
62 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
63 
64 #define IPV6_TCLASS_SHIFT 20
65 
66 #define HASH_SIZE_SHIFT  5
67 #define HASH_SIZE (1 << HASH_SIZE_SHIFT)
68 
69 static int ip6gre_net_id __read_mostly;
70 struct ip6gre_net {
71 	struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
72 
73 	struct net_device *fb_tunnel_dev;
74 };
75 
76 static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
77 static int ip6gre_tunnel_init(struct net_device *dev);
78 static void ip6gre_tunnel_setup(struct net_device *dev);
79 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
80 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
81 
82 /* Tunnel hash table */
83 
84 /*
85    4 hash tables:
86 
87    3: (remote,local)
88    2: (remote,*)
89    1: (*,local)
90    0: (*,*)
91 
92    We require exact key match i.e. if a key is present in packet
93    it will match only tunnel with the same key; if it is not present,
94    it will match only keyless tunnel.
95 
96    All keysless packets, if not matched configured keyless tunnels
97    will match fallback tunnel.
98  */
99 
100 #define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
101 static u32 HASH_ADDR(const struct in6_addr *addr)
102 {
103 	u32 hash = ipv6_addr_hash(addr);
104 
105 	return hash_32(hash, HASH_SIZE_SHIFT);
106 }
107 
108 #define tunnels_r_l	tunnels[3]
109 #define tunnels_r	tunnels[2]
110 #define tunnels_l	tunnels[1]
111 #define tunnels_wc	tunnels[0]
112 
113 /* Given src, dst and key, find appropriate for input tunnel. */
114 
115 static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
116 		const struct in6_addr *remote, const struct in6_addr *local,
117 		__be32 key, __be16 gre_proto)
118 {
119 	struct net *net = dev_net(dev);
120 	int link = dev->ifindex;
121 	unsigned int h0 = HASH_ADDR(remote);
122 	unsigned int h1 = HASH_KEY(key);
123 	struct ip6_tnl *t, *cand = NULL;
124 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
125 	int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
126 		       ARPHRD_ETHER : ARPHRD_IP6GRE;
127 	int score, cand_score = 4;
128 
129 	for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
130 		if (!ipv6_addr_equal(local, &t->parms.laddr) ||
131 		    !ipv6_addr_equal(remote, &t->parms.raddr) ||
132 		    key != t->parms.i_key ||
133 		    !(t->dev->flags & IFF_UP))
134 			continue;
135 
136 		if (t->dev->type != ARPHRD_IP6GRE &&
137 		    t->dev->type != dev_type)
138 			continue;
139 
140 		score = 0;
141 		if (t->parms.link != link)
142 			score |= 1;
143 		if (t->dev->type != dev_type)
144 			score |= 2;
145 		if (score == 0)
146 			return t;
147 
148 		if (score < cand_score) {
149 			cand = t;
150 			cand_score = score;
151 		}
152 	}
153 
154 	for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
155 		if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
156 		    key != t->parms.i_key ||
157 		    !(t->dev->flags & IFF_UP))
158 			continue;
159 
160 		if (t->dev->type != ARPHRD_IP6GRE &&
161 		    t->dev->type != dev_type)
162 			continue;
163 
164 		score = 0;
165 		if (t->parms.link != link)
166 			score |= 1;
167 		if (t->dev->type != dev_type)
168 			score |= 2;
169 		if (score == 0)
170 			return t;
171 
172 		if (score < cand_score) {
173 			cand = t;
174 			cand_score = score;
175 		}
176 	}
177 
178 	for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
179 		if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
180 			  (!ipv6_addr_equal(local, &t->parms.raddr) ||
181 				 !ipv6_addr_is_multicast(local))) ||
182 		    key != t->parms.i_key ||
183 		    !(t->dev->flags & IFF_UP))
184 			continue;
185 
186 		if (t->dev->type != ARPHRD_IP6GRE &&
187 		    t->dev->type != dev_type)
188 			continue;
189 
190 		score = 0;
191 		if (t->parms.link != link)
192 			score |= 1;
193 		if (t->dev->type != dev_type)
194 			score |= 2;
195 		if (score == 0)
196 			return t;
197 
198 		if (score < cand_score) {
199 			cand = t;
200 			cand_score = score;
201 		}
202 	}
203 
204 	for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
205 		if (t->parms.i_key != key ||
206 		    !(t->dev->flags & IFF_UP))
207 			continue;
208 
209 		if (t->dev->type != ARPHRD_IP6GRE &&
210 		    t->dev->type != dev_type)
211 			continue;
212 
213 		score = 0;
214 		if (t->parms.link != link)
215 			score |= 1;
216 		if (t->dev->type != dev_type)
217 			score |= 2;
218 		if (score == 0)
219 			return t;
220 
221 		if (score < cand_score) {
222 			cand = t;
223 			cand_score = score;
224 		}
225 	}
226 
227 	if (cand != NULL)
228 		return cand;
229 
230 	dev = ign->fb_tunnel_dev;
231 	if (dev->flags & IFF_UP)
232 		return netdev_priv(dev);
233 
234 	return NULL;
235 }
236 
237 static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
238 		const struct __ip6_tnl_parm *p)
239 {
240 	const struct in6_addr *remote = &p->raddr;
241 	const struct in6_addr *local = &p->laddr;
242 	unsigned int h = HASH_KEY(p->i_key);
243 	int prio = 0;
244 
245 	if (!ipv6_addr_any(local))
246 		prio |= 1;
247 	if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
248 		prio |= 2;
249 		h ^= HASH_ADDR(remote);
250 	}
251 
252 	return &ign->tunnels[prio][h];
253 }
254 
255 static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
256 		const struct ip6_tnl *t)
257 {
258 	return __ip6gre_bucket(ign, &t->parms);
259 }
260 
261 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
262 {
263 	struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
264 
265 	rcu_assign_pointer(t->next, rtnl_dereference(*tp));
266 	rcu_assign_pointer(*tp, t);
267 }
268 
269 static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
270 {
271 	struct ip6_tnl __rcu **tp;
272 	struct ip6_tnl *iter;
273 
274 	for (tp = ip6gre_bucket(ign, t);
275 	     (iter = rtnl_dereference(*tp)) != NULL;
276 	     tp = &iter->next) {
277 		if (t == iter) {
278 			rcu_assign_pointer(*tp, t->next);
279 			break;
280 		}
281 	}
282 }
283 
284 static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
285 					   const struct __ip6_tnl_parm *parms,
286 					   int type)
287 {
288 	const struct in6_addr *remote = &parms->raddr;
289 	const struct in6_addr *local = &parms->laddr;
290 	__be32 key = parms->i_key;
291 	int link = parms->link;
292 	struct ip6_tnl *t;
293 	struct ip6_tnl __rcu **tp;
294 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
295 
296 	for (tp = __ip6gre_bucket(ign, parms);
297 	     (t = rtnl_dereference(*tp)) != NULL;
298 	     tp = &t->next)
299 		if (ipv6_addr_equal(local, &t->parms.laddr) &&
300 		    ipv6_addr_equal(remote, &t->parms.raddr) &&
301 		    key == t->parms.i_key &&
302 		    link == t->parms.link &&
303 		    type == t->dev->type)
304 			break;
305 
306 	return t;
307 }
308 
309 static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
310 		const struct __ip6_tnl_parm *parms, int create)
311 {
312 	struct ip6_tnl *t, *nt;
313 	struct net_device *dev;
314 	char name[IFNAMSIZ];
315 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
316 
317 	t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
318 	if (t || !create)
319 		return t;
320 
321 	if (parms->name[0])
322 		strlcpy(name, parms->name, IFNAMSIZ);
323 	else
324 		strcpy(name, "ip6gre%d");
325 
326 	dev = alloc_netdev(sizeof(*t), name, ip6gre_tunnel_setup);
327 	if (!dev)
328 		return NULL;
329 
330 	dev_net_set(dev, net);
331 
332 	nt = netdev_priv(dev);
333 	nt->parms = *parms;
334 	dev->rtnl_link_ops = &ip6gre_link_ops;
335 
336 	nt->dev = dev;
337 	nt->net = dev_net(dev);
338 	ip6gre_tnl_link_config(nt, 1);
339 
340 	if (register_netdevice(dev) < 0)
341 		goto failed_free;
342 
343 	/* Can use a lockless transmit, unless we generate output sequences */
344 	if (!(nt->parms.o_flags & GRE_SEQ))
345 		dev->features |= NETIF_F_LLTX;
346 
347 	dev_hold(dev);
348 	ip6gre_tunnel_link(ign, nt);
349 	return nt;
350 
351 failed_free:
352 	free_netdev(dev);
353 	return NULL;
354 }
355 
356 static void ip6gre_tunnel_uninit(struct net_device *dev)
357 {
358 	struct net *net = dev_net(dev);
359 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
360 
361 	ip6gre_tunnel_unlink(ign, netdev_priv(dev));
362 	dev_put(dev);
363 }
364 
365 
366 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
367 		u8 type, u8 code, int offset, __be32 info)
368 {
369 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
370 	__be16 *p = (__be16 *)(skb->data + offset);
371 	int grehlen = offset + 4;
372 	struct ip6_tnl *t;
373 	__be16 flags;
374 
375 	flags = p[0];
376 	if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
377 		if (flags&(GRE_VERSION|GRE_ROUTING))
378 			return;
379 		if (flags&GRE_KEY) {
380 			grehlen += 4;
381 			if (flags&GRE_CSUM)
382 				grehlen += 4;
383 		}
384 	}
385 
386 	/* If only 8 bytes returned, keyed message will be dropped here */
387 	if (!pskb_may_pull(skb, grehlen))
388 		return;
389 	ipv6h = (const struct ipv6hdr *)skb->data;
390 	p = (__be16 *)(skb->data + offset);
391 
392 	t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
393 				flags & GRE_KEY ?
394 				*(((__be32 *)p) + (grehlen / 4) - 1) : 0,
395 				p[1]);
396 	if (t == NULL)
397 		return;
398 
399 	switch (type) {
400 		__u32 teli;
401 		struct ipv6_tlv_tnl_enc_lim *tel;
402 		__u32 mtu;
403 	case ICMPV6_DEST_UNREACH:
404 		net_warn_ratelimited("%s: Path to destination invalid or inactive!\n",
405 				     t->parms.name);
406 		break;
407 	case ICMPV6_TIME_EXCEED:
408 		if (code == ICMPV6_EXC_HOPLIMIT) {
409 			net_warn_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
410 					     t->parms.name);
411 		}
412 		break;
413 	case ICMPV6_PARAMPROB:
414 		teli = 0;
415 		if (code == ICMPV6_HDR_FIELD)
416 			teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
417 
418 		if (teli && teli == info - 2) {
419 			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
420 			if (tel->encap_limit == 0) {
421 				net_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
422 						     t->parms.name);
423 			}
424 		} else {
425 			net_warn_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
426 					     t->parms.name);
427 		}
428 		break;
429 	case ICMPV6_PKT_TOOBIG:
430 		mtu = info - offset;
431 		if (mtu < IPV6_MIN_MTU)
432 			mtu = IPV6_MIN_MTU;
433 		t->dev->mtu = mtu;
434 		break;
435 	}
436 
437 	if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
438 		t->err_count++;
439 	else
440 		t->err_count = 1;
441 	t->err_time = jiffies;
442 }
443 
444 static int ip6gre_rcv(struct sk_buff *skb)
445 {
446 	const struct ipv6hdr *ipv6h;
447 	u8     *h;
448 	__be16    flags;
449 	__sum16   csum = 0;
450 	__be32 key = 0;
451 	u32    seqno = 0;
452 	struct ip6_tnl *tunnel;
453 	int    offset = 4;
454 	__be16 gre_proto;
455 	int err;
456 
457 	if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
458 		goto drop;
459 
460 	ipv6h = ipv6_hdr(skb);
461 	h = skb->data;
462 	flags = *(__be16 *)h;
463 
464 	if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
465 		/* - Version must be 0.
466 		   - We do not support routing headers.
467 		 */
468 		if (flags&(GRE_VERSION|GRE_ROUTING))
469 			goto drop;
470 
471 		if (flags&GRE_CSUM) {
472 			switch (skb->ip_summed) {
473 			case CHECKSUM_COMPLETE:
474 				csum = csum_fold(skb->csum);
475 				if (!csum)
476 					break;
477 				/* fall through */
478 			case CHECKSUM_NONE:
479 				skb->csum = 0;
480 				csum = __skb_checksum_complete(skb);
481 				skb->ip_summed = CHECKSUM_COMPLETE;
482 			}
483 			offset += 4;
484 		}
485 		if (flags&GRE_KEY) {
486 			key = *(__be32 *)(h + offset);
487 			offset += 4;
488 		}
489 		if (flags&GRE_SEQ) {
490 			seqno = ntohl(*(__be32 *)(h + offset));
491 			offset += 4;
492 		}
493 	}
494 
495 	gre_proto = *(__be16 *)(h + 2);
496 
497 	tunnel = ip6gre_tunnel_lookup(skb->dev,
498 					  &ipv6h->saddr, &ipv6h->daddr, key,
499 					  gre_proto);
500 	if (tunnel) {
501 		struct pcpu_tstats *tstats;
502 
503 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
504 			goto drop;
505 
506 		if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
507 			tunnel->dev->stats.rx_dropped++;
508 			goto drop;
509 		}
510 
511 		skb->protocol = gre_proto;
512 		/* WCCP version 1 and 2 protocol decoding.
513 		 * - Change protocol to IP
514 		 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
515 		 */
516 		if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
517 			skb->protocol = htons(ETH_P_IP);
518 			if ((*(h + offset) & 0xF0) != 0x40)
519 				offset += 4;
520 		}
521 
522 		skb->mac_header = skb->network_header;
523 		__pskb_pull(skb, offset);
524 		skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
525 
526 		if (((flags&GRE_CSUM) && csum) ||
527 		    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
528 			tunnel->dev->stats.rx_crc_errors++;
529 			tunnel->dev->stats.rx_errors++;
530 			goto drop;
531 		}
532 		if (tunnel->parms.i_flags&GRE_SEQ) {
533 			if (!(flags&GRE_SEQ) ||
534 			    (tunnel->i_seqno &&
535 					(s32)(seqno - tunnel->i_seqno) < 0)) {
536 				tunnel->dev->stats.rx_fifo_errors++;
537 				tunnel->dev->stats.rx_errors++;
538 				goto drop;
539 			}
540 			tunnel->i_seqno = seqno + 1;
541 		}
542 
543 		/* Warning: All skb pointers will be invalidated! */
544 		if (tunnel->dev->type == ARPHRD_ETHER) {
545 			if (!pskb_may_pull(skb, ETH_HLEN)) {
546 				tunnel->dev->stats.rx_length_errors++;
547 				tunnel->dev->stats.rx_errors++;
548 				goto drop;
549 			}
550 
551 			ipv6h = ipv6_hdr(skb);
552 			skb->protocol = eth_type_trans(skb, tunnel->dev);
553 			skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
554 		}
555 
556 		__skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
557 
558 		skb_reset_network_header(skb);
559 
560 		err = IP6_ECN_decapsulate(ipv6h, skb);
561 		if (unlikely(err)) {
562 			if (log_ecn_error)
563 				net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
564 						     &ipv6h->saddr,
565 						     ipv6_get_dsfield(ipv6h));
566 			if (err > 1) {
567 				++tunnel->dev->stats.rx_frame_errors;
568 				++tunnel->dev->stats.rx_errors;
569 				goto drop;
570 			}
571 		}
572 
573 		tstats = this_cpu_ptr(tunnel->dev->tstats);
574 		u64_stats_update_begin(&tstats->syncp);
575 		tstats->rx_packets++;
576 		tstats->rx_bytes += skb->len;
577 		u64_stats_update_end(&tstats->syncp);
578 
579 		netif_rx(skb);
580 
581 		return 0;
582 	}
583 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
584 
585 drop:
586 	kfree_skb(skb);
587 	return 0;
588 }
589 
590 struct ipv6_tel_txoption {
591 	struct ipv6_txoptions ops;
592 	__u8 dst_opt[8];
593 };
594 
595 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
596 {
597 	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
598 
599 	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
600 	opt->dst_opt[3] = 1;
601 	opt->dst_opt[4] = encap_limit;
602 	opt->dst_opt[5] = IPV6_TLV_PADN;
603 	opt->dst_opt[6] = 1;
604 
605 	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
606 	opt->ops.opt_nflen = 8;
607 }
608 
609 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
610 			 struct net_device *dev,
611 			 __u8 dsfield,
612 			 struct flowi6 *fl6,
613 			 int encap_limit,
614 			 __u32 *pmtu)
615 {
616 	struct net *net = dev_net(dev);
617 	struct ip6_tnl *tunnel = netdev_priv(dev);
618 	struct net_device *tdev;    /* Device to other host */
619 	struct ipv6hdr  *ipv6h;     /* Our new IP header */
620 	unsigned int max_headroom = 0; /* The extra header space needed */
621 	int    gre_hlen;
622 	struct ipv6_tel_txoption opt;
623 	int    mtu;
624 	struct dst_entry *dst = NULL, *ndst = NULL;
625 	struct net_device_stats *stats = &tunnel->dev->stats;
626 	int err = -1;
627 	u8 proto;
628 	struct sk_buff *new_skb;
629 
630 	if (dev->type == ARPHRD_ETHER)
631 		IPCB(skb)->flags = 0;
632 
633 	if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
634 		gre_hlen = 0;
635 		ipv6h = (struct ipv6hdr *)skb->data;
636 		fl6->daddr = ipv6h->daddr;
637 	} else {
638 		gre_hlen = tunnel->hlen;
639 		fl6->daddr = tunnel->parms.raddr;
640 	}
641 
642 	if (!fl6->flowi6_mark)
643 		dst = ip6_tnl_dst_check(tunnel);
644 
645 	if (!dst) {
646 		ndst = ip6_route_output(net, NULL, fl6);
647 
648 		if (ndst->error)
649 			goto tx_err_link_failure;
650 		ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
651 		if (IS_ERR(ndst)) {
652 			err = PTR_ERR(ndst);
653 			ndst = NULL;
654 			goto tx_err_link_failure;
655 		}
656 		dst = ndst;
657 	}
658 
659 	tdev = dst->dev;
660 
661 	if (tdev == dev) {
662 		stats->collisions++;
663 		net_warn_ratelimited("%s: Local routing loop detected!\n",
664 				     tunnel->parms.name);
665 		goto tx_err_dst_release;
666 	}
667 
668 	mtu = dst_mtu(dst) - sizeof(*ipv6h);
669 	if (encap_limit >= 0) {
670 		max_headroom += 8;
671 		mtu -= 8;
672 	}
673 	if (mtu < IPV6_MIN_MTU)
674 		mtu = IPV6_MIN_MTU;
675 	if (skb_dst(skb))
676 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
677 	if (skb->len > mtu) {
678 		*pmtu = mtu;
679 		err = -EMSGSIZE;
680 		goto tx_err_dst_release;
681 	}
682 
683 	if (tunnel->err_count > 0) {
684 		if (time_before(jiffies,
685 				tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
686 			tunnel->err_count--;
687 
688 			dst_link_failure(skb);
689 		} else
690 			tunnel->err_count = 0;
691 	}
692 
693 	skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
694 
695 	max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
696 
697 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
698 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
699 		new_skb = skb_realloc_headroom(skb, max_headroom);
700 		if (max_headroom > dev->needed_headroom)
701 			dev->needed_headroom = max_headroom;
702 		if (!new_skb)
703 			goto tx_err_dst_release;
704 
705 		if (skb->sk)
706 			skb_set_owner_w(new_skb, skb->sk);
707 		consume_skb(skb);
708 		skb = new_skb;
709 	}
710 
711 	if (fl6->flowi6_mark) {
712 		skb_dst_set(skb, dst);
713 		ndst = NULL;
714 	} else {
715 		skb_dst_set_noref(skb, dst);
716 	}
717 
718 	proto = NEXTHDR_GRE;
719 	if (encap_limit >= 0) {
720 		init_tel_txopt(&opt, encap_limit);
721 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
722 	}
723 
724 	if (likely(!skb->encapsulation)) {
725 		skb_reset_inner_headers(skb);
726 		skb->encapsulation = 1;
727 	}
728 
729 	skb_push(skb, gre_hlen);
730 	skb_reset_network_header(skb);
731 	skb_set_transport_header(skb, sizeof(*ipv6h));
732 
733 	/*
734 	 *	Push down and install the IP header.
735 	 */
736 	ipv6h = ipv6_hdr(skb);
737 	ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);
738 	ipv6h->hop_limit = tunnel->parms.hop_limit;
739 	ipv6h->nexthdr = proto;
740 	ipv6h->saddr = fl6->saddr;
741 	ipv6h->daddr = fl6->daddr;
742 
743 	((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
744 	((__be16 *)(ipv6h + 1))[1] = (dev->type == ARPHRD_ETHER) ?
745 				   htons(ETH_P_TEB) : skb->protocol;
746 
747 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
748 		__be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
749 
750 		if (tunnel->parms.o_flags&GRE_SEQ) {
751 			++tunnel->o_seqno;
752 			*ptr = htonl(tunnel->o_seqno);
753 			ptr--;
754 		}
755 		if (tunnel->parms.o_flags&GRE_KEY) {
756 			*ptr = tunnel->parms.o_key;
757 			ptr--;
758 		}
759 		if (tunnel->parms.o_flags&GRE_CSUM) {
760 			*ptr = 0;
761 			*(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
762 				skb->len - sizeof(struct ipv6hdr));
763 		}
764 	}
765 
766 	ip6tunnel_xmit(skb, dev);
767 	if (ndst)
768 		ip6_tnl_dst_store(tunnel, ndst);
769 	return 0;
770 tx_err_link_failure:
771 	stats->tx_carrier_errors++;
772 	dst_link_failure(skb);
773 tx_err_dst_release:
774 	dst_release(ndst);
775 	return err;
776 }
777 
778 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
779 {
780 	struct ip6_tnl *t = netdev_priv(dev);
781 	const struct iphdr  *iph = ip_hdr(skb);
782 	int encap_limit = -1;
783 	struct flowi6 fl6;
784 	__u8 dsfield;
785 	__u32 mtu;
786 	int err;
787 
788 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
789 		encap_limit = t->parms.encap_limit;
790 
791 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
792 	fl6.flowi6_proto = IPPROTO_IPIP;
793 
794 	dsfield = ipv4_get_dsfield(iph);
795 
796 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
797 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
798 					  & IPV6_TCLASS_MASK;
799 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
800 		fl6.flowi6_mark = skb->mark;
801 
802 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
803 	if (err != 0) {
804 		/* XXX: send ICMP error even if DF is not set. */
805 		if (err == -EMSGSIZE)
806 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
807 				  htonl(mtu));
808 		return -1;
809 	}
810 
811 	return 0;
812 }
813 
814 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
815 {
816 	struct ip6_tnl *t = netdev_priv(dev);
817 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
818 	int encap_limit = -1;
819 	__u16 offset;
820 	struct flowi6 fl6;
821 	__u8 dsfield;
822 	__u32 mtu;
823 	int err;
824 
825 	if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
826 		return -1;
827 
828 	offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
829 	if (offset > 0) {
830 		struct ipv6_tlv_tnl_enc_lim *tel;
831 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
832 		if (tel->encap_limit == 0) {
833 			icmpv6_send(skb, ICMPV6_PARAMPROB,
834 				    ICMPV6_HDR_FIELD, offset + 2);
835 			return -1;
836 		}
837 		encap_limit = tel->encap_limit - 1;
838 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
839 		encap_limit = t->parms.encap_limit;
840 
841 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
842 	fl6.flowi6_proto = IPPROTO_IPV6;
843 
844 	dsfield = ipv6_get_dsfield(ipv6h);
845 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
846 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
847 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
848 		fl6.flowlabel |= ip6_flowlabel(ipv6h);
849 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
850 		fl6.flowi6_mark = skb->mark;
851 
852 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
853 	if (err != 0) {
854 		if (err == -EMSGSIZE)
855 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
856 		return -1;
857 	}
858 
859 	return 0;
860 }
861 
862 /**
863  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
864  *   @t: the outgoing tunnel device
865  *   @hdr: IPv6 header from the incoming packet
866  *
867  * Description:
868  *   Avoid trivial tunneling loop by checking that tunnel exit-point
869  *   doesn't match source of incoming packet.
870  *
871  * Return:
872  *   1 if conflict,
873  *   0 else
874  **/
875 
876 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
877 	const struct ipv6hdr *hdr)
878 {
879 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
880 }
881 
882 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
883 {
884 	struct ip6_tnl *t = netdev_priv(dev);
885 	int encap_limit = -1;
886 	struct flowi6 fl6;
887 	__u32 mtu;
888 	int err;
889 
890 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
891 		encap_limit = t->parms.encap_limit;
892 
893 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
894 	fl6.flowi6_proto = skb->protocol;
895 
896 	err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
897 
898 	return err;
899 }
900 
901 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
902 	struct net_device *dev)
903 {
904 	struct ip6_tnl *t = netdev_priv(dev);
905 	struct net_device_stats *stats = &t->dev->stats;
906 	int ret;
907 
908 	if (!ip6_tnl_xmit_ctl(t))
909 		goto tx_err;
910 
911 	switch (skb->protocol) {
912 	case htons(ETH_P_IP):
913 		ret = ip6gre_xmit_ipv4(skb, dev);
914 		break;
915 	case htons(ETH_P_IPV6):
916 		ret = ip6gre_xmit_ipv6(skb, dev);
917 		break;
918 	default:
919 		ret = ip6gre_xmit_other(skb, dev);
920 		break;
921 	}
922 
923 	if (ret < 0)
924 		goto tx_err;
925 
926 	return NETDEV_TX_OK;
927 
928 tx_err:
929 	stats->tx_errors++;
930 	stats->tx_dropped++;
931 	kfree_skb(skb);
932 	return NETDEV_TX_OK;
933 }
934 
935 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
936 {
937 	struct net_device *dev = t->dev;
938 	struct __ip6_tnl_parm *p = &t->parms;
939 	struct flowi6 *fl6 = &t->fl.u.ip6;
940 	int addend = sizeof(struct ipv6hdr) + 4;
941 
942 	if (dev->type != ARPHRD_ETHER) {
943 		memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
944 		memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
945 	}
946 
947 	/* Set up flowi template */
948 	fl6->saddr = p->laddr;
949 	fl6->daddr = p->raddr;
950 	fl6->flowi6_oif = p->link;
951 	fl6->flowlabel = 0;
952 
953 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
954 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
955 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
956 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
957 
958 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
959 	p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
960 
961 	if (p->flags&IP6_TNL_F_CAP_XMIT &&
962 			p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
963 		dev->flags |= IFF_POINTOPOINT;
964 	else
965 		dev->flags &= ~IFF_POINTOPOINT;
966 
967 	dev->iflink = p->link;
968 
969 	/* Precalculate GRE options length */
970 	if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
971 		if (t->parms.o_flags&GRE_CSUM)
972 			addend += 4;
973 		if (t->parms.o_flags&GRE_KEY)
974 			addend += 4;
975 		if (t->parms.o_flags&GRE_SEQ)
976 			addend += 4;
977 	}
978 	t->hlen = addend;
979 
980 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
981 		int strict = (ipv6_addr_type(&p->raddr) &
982 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
983 
984 		struct rt6_info *rt = rt6_lookup(dev_net(dev),
985 						 &p->raddr, &p->laddr,
986 						 p->link, strict);
987 
988 		if (rt == NULL)
989 			return;
990 
991 		if (rt->dst.dev) {
992 			dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
993 
994 			if (set_mtu) {
995 				dev->mtu = rt->dst.dev->mtu - addend;
996 				if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
997 					dev->mtu -= 8;
998 
999 				if (dev->mtu < IPV6_MIN_MTU)
1000 					dev->mtu = IPV6_MIN_MTU;
1001 			}
1002 		}
1003 		ip6_rt_put(rt);
1004 	}
1005 }
1006 
1007 static int ip6gre_tnl_change(struct ip6_tnl *t,
1008 	const struct __ip6_tnl_parm *p, int set_mtu)
1009 {
1010 	t->parms.laddr = p->laddr;
1011 	t->parms.raddr = p->raddr;
1012 	t->parms.flags = p->flags;
1013 	t->parms.hop_limit = p->hop_limit;
1014 	t->parms.encap_limit = p->encap_limit;
1015 	t->parms.flowinfo = p->flowinfo;
1016 	t->parms.link = p->link;
1017 	t->parms.proto = p->proto;
1018 	t->parms.i_key = p->i_key;
1019 	t->parms.o_key = p->o_key;
1020 	t->parms.i_flags = p->i_flags;
1021 	t->parms.o_flags = p->o_flags;
1022 	ip6_tnl_dst_reset(t);
1023 	ip6gre_tnl_link_config(t, set_mtu);
1024 	return 0;
1025 }
1026 
1027 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1028 	const struct ip6_tnl_parm2 *u)
1029 {
1030 	p->laddr = u->laddr;
1031 	p->raddr = u->raddr;
1032 	p->flags = u->flags;
1033 	p->hop_limit = u->hop_limit;
1034 	p->encap_limit = u->encap_limit;
1035 	p->flowinfo = u->flowinfo;
1036 	p->link = u->link;
1037 	p->i_key = u->i_key;
1038 	p->o_key = u->o_key;
1039 	p->i_flags = u->i_flags;
1040 	p->o_flags = u->o_flags;
1041 	memcpy(p->name, u->name, sizeof(u->name));
1042 }
1043 
1044 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1045 	const struct __ip6_tnl_parm *p)
1046 {
1047 	u->proto = IPPROTO_GRE;
1048 	u->laddr = p->laddr;
1049 	u->raddr = p->raddr;
1050 	u->flags = p->flags;
1051 	u->hop_limit = p->hop_limit;
1052 	u->encap_limit = p->encap_limit;
1053 	u->flowinfo = p->flowinfo;
1054 	u->link = p->link;
1055 	u->i_key = p->i_key;
1056 	u->o_key = p->o_key;
1057 	u->i_flags = p->i_flags;
1058 	u->o_flags = p->o_flags;
1059 	memcpy(u->name, p->name, sizeof(u->name));
1060 }
1061 
1062 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1063 	struct ifreq *ifr, int cmd)
1064 {
1065 	int err = 0;
1066 	struct ip6_tnl_parm2 p;
1067 	struct __ip6_tnl_parm p1;
1068 	struct ip6_tnl *t;
1069 	struct net *net = dev_net(dev);
1070 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1071 
1072 	switch (cmd) {
1073 	case SIOCGETTUNNEL:
1074 		t = NULL;
1075 		if (dev == ign->fb_tunnel_dev) {
1076 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1077 				err = -EFAULT;
1078 				break;
1079 			}
1080 			ip6gre_tnl_parm_from_user(&p1, &p);
1081 			t = ip6gre_tunnel_locate(net, &p1, 0);
1082 		}
1083 		if (t == NULL)
1084 			t = netdev_priv(dev);
1085 		memset(&p, 0, sizeof(p));
1086 		ip6gre_tnl_parm_to_user(&p, &t->parms);
1087 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1088 			err = -EFAULT;
1089 		break;
1090 
1091 	case SIOCADDTUNNEL:
1092 	case SIOCCHGTUNNEL:
1093 		err = -EPERM;
1094 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1095 			goto done;
1096 
1097 		err = -EFAULT;
1098 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1099 			goto done;
1100 
1101 		err = -EINVAL;
1102 		if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1103 			goto done;
1104 
1105 		if (!(p.i_flags&GRE_KEY))
1106 			p.i_key = 0;
1107 		if (!(p.o_flags&GRE_KEY))
1108 			p.o_key = 0;
1109 
1110 		ip6gre_tnl_parm_from_user(&p1, &p);
1111 		t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1112 
1113 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1114 			if (t != NULL) {
1115 				if (t->dev != dev) {
1116 					err = -EEXIST;
1117 					break;
1118 				}
1119 			} else {
1120 				t = netdev_priv(dev);
1121 
1122 				ip6gre_tunnel_unlink(ign, t);
1123 				synchronize_net();
1124 				ip6gre_tnl_change(t, &p1, 1);
1125 				ip6gre_tunnel_link(ign, t);
1126 				netdev_state_change(dev);
1127 			}
1128 		}
1129 
1130 		if (t) {
1131 			err = 0;
1132 
1133 			memset(&p, 0, sizeof(p));
1134 			ip6gre_tnl_parm_to_user(&p, &t->parms);
1135 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1136 				err = -EFAULT;
1137 		} else
1138 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1139 		break;
1140 
1141 	case SIOCDELTUNNEL:
1142 		err = -EPERM;
1143 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1144 			goto done;
1145 
1146 		if (dev == ign->fb_tunnel_dev) {
1147 			err = -EFAULT;
1148 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1149 				goto done;
1150 			err = -ENOENT;
1151 			ip6gre_tnl_parm_from_user(&p1, &p);
1152 			t = ip6gre_tunnel_locate(net, &p1, 0);
1153 			if (t == NULL)
1154 				goto done;
1155 			err = -EPERM;
1156 			if (t == netdev_priv(ign->fb_tunnel_dev))
1157 				goto done;
1158 			dev = t->dev;
1159 		}
1160 		unregister_netdevice(dev);
1161 		err = 0;
1162 		break;
1163 
1164 	default:
1165 		err = -EINVAL;
1166 	}
1167 
1168 done:
1169 	return err;
1170 }
1171 
1172 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1173 {
1174 	if (new_mtu < 68 ||
1175 	    new_mtu > 0xFFF8 - dev->hard_header_len)
1176 		return -EINVAL;
1177 	dev->mtu = new_mtu;
1178 	return 0;
1179 }
1180 
1181 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1182 			unsigned short type,
1183 			const void *daddr, const void *saddr, unsigned int len)
1184 {
1185 	struct ip6_tnl *t = netdev_priv(dev);
1186 	struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1187 	__be16 *p = (__be16 *)(ipv6h+1);
1188 
1189 	ip6_flow_hdr(ipv6h, 0, t->fl.u.ip6.flowlabel);
1190 	ipv6h->hop_limit = t->parms.hop_limit;
1191 	ipv6h->nexthdr = NEXTHDR_GRE;
1192 	ipv6h->saddr = t->parms.laddr;
1193 	ipv6h->daddr = t->parms.raddr;
1194 
1195 	p[0]		= t->parms.o_flags;
1196 	p[1]		= htons(type);
1197 
1198 	/*
1199 	 *	Set the source hardware address.
1200 	 */
1201 
1202 	if (saddr)
1203 		memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1204 	if (daddr)
1205 		memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1206 	if (!ipv6_addr_any(&ipv6h->daddr))
1207 		return t->hlen;
1208 
1209 	return -t->hlen;
1210 }
1211 
1212 static const struct header_ops ip6gre_header_ops = {
1213 	.create	= ip6gre_header,
1214 };
1215 
1216 static const struct net_device_ops ip6gre_netdev_ops = {
1217 	.ndo_init		= ip6gre_tunnel_init,
1218 	.ndo_uninit		= ip6gre_tunnel_uninit,
1219 	.ndo_start_xmit		= ip6gre_tunnel_xmit,
1220 	.ndo_do_ioctl		= ip6gre_tunnel_ioctl,
1221 	.ndo_change_mtu		= ip6gre_tunnel_change_mtu,
1222 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1223 };
1224 
1225 static void ip6gre_dev_free(struct net_device *dev)
1226 {
1227 	free_percpu(dev->tstats);
1228 	free_netdev(dev);
1229 }
1230 
1231 static void ip6gre_tunnel_setup(struct net_device *dev)
1232 {
1233 	struct ip6_tnl *t;
1234 
1235 	dev->netdev_ops = &ip6gre_netdev_ops;
1236 	dev->destructor = ip6gre_dev_free;
1237 
1238 	dev->type = ARPHRD_IP6GRE;
1239 	dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1240 	dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1241 	t = netdev_priv(dev);
1242 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1243 		dev->mtu -= 8;
1244 	dev->flags |= IFF_NOARP;
1245 	dev->iflink = 0;
1246 	dev->addr_len = sizeof(struct in6_addr);
1247 	dev->features |= NETIF_F_NETNS_LOCAL;
1248 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1249 }
1250 
1251 static int ip6gre_tunnel_init(struct net_device *dev)
1252 {
1253 	struct ip6_tnl *tunnel;
1254 	int i;
1255 
1256 	tunnel = netdev_priv(dev);
1257 
1258 	tunnel->dev = dev;
1259 	tunnel->net = dev_net(dev);
1260 	strcpy(tunnel->parms.name, dev->name);
1261 
1262 	memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1263 	memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1264 
1265 	if (ipv6_addr_any(&tunnel->parms.raddr))
1266 		dev->header_ops = &ip6gre_header_ops;
1267 
1268 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1269 	if (!dev->tstats)
1270 		return -ENOMEM;
1271 
1272 	for_each_possible_cpu(i) {
1273 		struct pcpu_tstats *ip6gre_tunnel_stats;
1274 		ip6gre_tunnel_stats = per_cpu_ptr(dev->tstats, i);
1275 		u64_stats_init(&ip6gre_tunnel_stats->syncp);
1276 	}
1277 
1278 
1279 	return 0;
1280 }
1281 
1282 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1283 {
1284 	struct ip6_tnl *tunnel = netdev_priv(dev);
1285 
1286 	tunnel->dev = dev;
1287 	tunnel->net = dev_net(dev);
1288 	strcpy(tunnel->parms.name, dev->name);
1289 
1290 	tunnel->hlen		= sizeof(struct ipv6hdr) + 4;
1291 
1292 	dev_hold(dev);
1293 }
1294 
1295 
1296 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1297 	.handler     = ip6gre_rcv,
1298 	.err_handler = ip6gre_err,
1299 	.flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1300 };
1301 
1302 static void ip6gre_destroy_tunnels(struct ip6gre_net *ign,
1303 	struct list_head *head)
1304 {
1305 	int prio;
1306 
1307 	for (prio = 0; prio < 4; prio++) {
1308 		int h;
1309 		for (h = 0; h < HASH_SIZE; h++) {
1310 			struct ip6_tnl *t;
1311 
1312 			t = rtnl_dereference(ign->tunnels[prio][h]);
1313 
1314 			while (t != NULL) {
1315 				unregister_netdevice_queue(t->dev, head);
1316 				t = rtnl_dereference(t->next);
1317 			}
1318 		}
1319 	}
1320 }
1321 
1322 static int __net_init ip6gre_init_net(struct net *net)
1323 {
1324 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1325 	int err;
1326 
1327 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1328 					   ip6gre_tunnel_setup);
1329 	if (!ign->fb_tunnel_dev) {
1330 		err = -ENOMEM;
1331 		goto err_alloc_dev;
1332 	}
1333 	dev_net_set(ign->fb_tunnel_dev, net);
1334 
1335 	ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1336 	ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1337 
1338 	err = register_netdev(ign->fb_tunnel_dev);
1339 	if (err)
1340 		goto err_reg_dev;
1341 
1342 	rcu_assign_pointer(ign->tunnels_wc[0],
1343 			   netdev_priv(ign->fb_tunnel_dev));
1344 	return 0;
1345 
1346 err_reg_dev:
1347 	ip6gre_dev_free(ign->fb_tunnel_dev);
1348 err_alloc_dev:
1349 	return err;
1350 }
1351 
1352 static void __net_exit ip6gre_exit_net(struct net *net)
1353 {
1354 	struct ip6gre_net *ign;
1355 	LIST_HEAD(list);
1356 
1357 	ign = net_generic(net, ip6gre_net_id);
1358 	rtnl_lock();
1359 	ip6gre_destroy_tunnels(ign, &list);
1360 	unregister_netdevice_many(&list);
1361 	rtnl_unlock();
1362 }
1363 
1364 static struct pernet_operations ip6gre_net_ops = {
1365 	.init = ip6gre_init_net,
1366 	.exit = ip6gre_exit_net,
1367 	.id   = &ip6gre_net_id,
1368 	.size = sizeof(struct ip6gre_net),
1369 };
1370 
1371 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1372 {
1373 	__be16 flags;
1374 
1375 	if (!data)
1376 		return 0;
1377 
1378 	flags = 0;
1379 	if (data[IFLA_GRE_IFLAGS])
1380 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1381 	if (data[IFLA_GRE_OFLAGS])
1382 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1383 	if (flags & (GRE_VERSION|GRE_ROUTING))
1384 		return -EINVAL;
1385 
1386 	return 0;
1387 }
1388 
1389 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1390 {
1391 	struct in6_addr daddr;
1392 
1393 	if (tb[IFLA_ADDRESS]) {
1394 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1395 			return -EINVAL;
1396 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1397 			return -EADDRNOTAVAIL;
1398 	}
1399 
1400 	if (!data)
1401 		goto out;
1402 
1403 	if (data[IFLA_GRE_REMOTE]) {
1404 		nla_memcpy(&daddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1405 		if (ipv6_addr_any(&daddr))
1406 			return -EINVAL;
1407 	}
1408 
1409 out:
1410 	return ip6gre_tunnel_validate(tb, data);
1411 }
1412 
1413 
1414 static void ip6gre_netlink_parms(struct nlattr *data[],
1415 				struct __ip6_tnl_parm *parms)
1416 {
1417 	memset(parms, 0, sizeof(*parms));
1418 
1419 	if (!data)
1420 		return;
1421 
1422 	if (data[IFLA_GRE_LINK])
1423 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1424 
1425 	if (data[IFLA_GRE_IFLAGS])
1426 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1427 
1428 	if (data[IFLA_GRE_OFLAGS])
1429 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1430 
1431 	if (data[IFLA_GRE_IKEY])
1432 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1433 
1434 	if (data[IFLA_GRE_OKEY])
1435 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1436 
1437 	if (data[IFLA_GRE_LOCAL])
1438 		nla_memcpy(&parms->laddr, data[IFLA_GRE_LOCAL], sizeof(struct in6_addr));
1439 
1440 	if (data[IFLA_GRE_REMOTE])
1441 		nla_memcpy(&parms->raddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1442 
1443 	if (data[IFLA_GRE_TTL])
1444 		parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1445 
1446 	if (data[IFLA_GRE_ENCAP_LIMIT])
1447 		parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1448 
1449 	if (data[IFLA_GRE_FLOWINFO])
1450 		parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1451 
1452 	if (data[IFLA_GRE_FLAGS])
1453 		parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1454 }
1455 
1456 static int ip6gre_tap_init(struct net_device *dev)
1457 {
1458 	struct ip6_tnl *tunnel;
1459 	int i;
1460 
1461 	tunnel = netdev_priv(dev);
1462 
1463 	tunnel->dev = dev;
1464 	tunnel->net = dev_net(dev);
1465 	strcpy(tunnel->parms.name, dev->name);
1466 
1467 	ip6gre_tnl_link_config(tunnel, 1);
1468 
1469 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1470 	if (!dev->tstats)
1471 		return -ENOMEM;
1472 
1473 	for_each_possible_cpu(i) {
1474 		struct pcpu_tstats *ip6gre_tap_stats;
1475 		ip6gre_tap_stats = per_cpu_ptr(dev->tstats, i);
1476 		u64_stats_init(&ip6gre_tap_stats->syncp);
1477 	}
1478 
1479 	return 0;
1480 }
1481 
1482 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1483 	.ndo_init = ip6gre_tap_init,
1484 	.ndo_uninit = ip6gre_tunnel_uninit,
1485 	.ndo_start_xmit = ip6gre_tunnel_xmit,
1486 	.ndo_set_mac_address = eth_mac_addr,
1487 	.ndo_validate_addr = eth_validate_addr,
1488 	.ndo_change_mtu = ip6gre_tunnel_change_mtu,
1489 	.ndo_get_stats64 = ip_tunnel_get_stats64,
1490 };
1491 
1492 static void ip6gre_tap_setup(struct net_device *dev)
1493 {
1494 
1495 	ether_setup(dev);
1496 
1497 	dev->netdev_ops = &ip6gre_tap_netdev_ops;
1498 	dev->destructor = ip6gre_dev_free;
1499 
1500 	dev->iflink = 0;
1501 	dev->features |= NETIF_F_NETNS_LOCAL;
1502 }
1503 
1504 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1505 	struct nlattr *tb[], struct nlattr *data[])
1506 {
1507 	struct ip6_tnl *nt;
1508 	struct net *net = dev_net(dev);
1509 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1510 	int err;
1511 
1512 	nt = netdev_priv(dev);
1513 	ip6gre_netlink_parms(data, &nt->parms);
1514 
1515 	if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1516 		return -EEXIST;
1517 
1518 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1519 		eth_hw_addr_random(dev);
1520 
1521 	nt->dev = dev;
1522 	nt->net = dev_net(dev);
1523 	ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1524 
1525 	/* Can use a lockless transmit, unless we generate output sequences */
1526 	if (!(nt->parms.o_flags & GRE_SEQ))
1527 		dev->features |= NETIF_F_LLTX;
1528 
1529 	err = register_netdevice(dev);
1530 	if (err)
1531 		goto out;
1532 
1533 	dev_hold(dev);
1534 	ip6gre_tunnel_link(ign, nt);
1535 
1536 out:
1537 	return err;
1538 }
1539 
1540 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1541 			    struct nlattr *data[])
1542 {
1543 	struct ip6_tnl *t, *nt;
1544 	struct net *net = dev_net(dev);
1545 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1546 	struct __ip6_tnl_parm p;
1547 
1548 	if (dev == ign->fb_tunnel_dev)
1549 		return -EINVAL;
1550 
1551 	nt = netdev_priv(dev);
1552 	ip6gre_netlink_parms(data, &p);
1553 
1554 	t = ip6gre_tunnel_locate(net, &p, 0);
1555 
1556 	if (t) {
1557 		if (t->dev != dev)
1558 			return -EEXIST;
1559 	} else {
1560 		t = nt;
1561 
1562 		ip6gre_tunnel_unlink(ign, t);
1563 		ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1564 		ip6gre_tunnel_link(ign, t);
1565 		netdev_state_change(dev);
1566 	}
1567 
1568 	return 0;
1569 }
1570 
1571 static size_t ip6gre_get_size(const struct net_device *dev)
1572 {
1573 	return
1574 		/* IFLA_GRE_LINK */
1575 		nla_total_size(4) +
1576 		/* IFLA_GRE_IFLAGS */
1577 		nla_total_size(2) +
1578 		/* IFLA_GRE_OFLAGS */
1579 		nla_total_size(2) +
1580 		/* IFLA_GRE_IKEY */
1581 		nla_total_size(4) +
1582 		/* IFLA_GRE_OKEY */
1583 		nla_total_size(4) +
1584 		/* IFLA_GRE_LOCAL */
1585 		nla_total_size(sizeof(struct in6_addr)) +
1586 		/* IFLA_GRE_REMOTE */
1587 		nla_total_size(sizeof(struct in6_addr)) +
1588 		/* IFLA_GRE_TTL */
1589 		nla_total_size(1) +
1590 		/* IFLA_GRE_TOS */
1591 		nla_total_size(1) +
1592 		/* IFLA_GRE_ENCAP_LIMIT */
1593 		nla_total_size(1) +
1594 		/* IFLA_GRE_FLOWINFO */
1595 		nla_total_size(4) +
1596 		/* IFLA_GRE_FLAGS */
1597 		nla_total_size(4) +
1598 		0;
1599 }
1600 
1601 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1602 {
1603 	struct ip6_tnl *t = netdev_priv(dev);
1604 	struct __ip6_tnl_parm *p = &t->parms;
1605 
1606 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1607 	    nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1608 	    nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1609 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1610 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1611 	    nla_put(skb, IFLA_GRE_LOCAL, sizeof(struct in6_addr), &p->laddr) ||
1612 	    nla_put(skb, IFLA_GRE_REMOTE, sizeof(struct in6_addr), &p->raddr) ||
1613 	    nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1614 	    /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1615 	    nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1616 	    nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1617 	    nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1618 		goto nla_put_failure;
1619 	return 0;
1620 
1621 nla_put_failure:
1622 	return -EMSGSIZE;
1623 }
1624 
1625 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1626 	[IFLA_GRE_LINK]        = { .type = NLA_U32 },
1627 	[IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1628 	[IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1629 	[IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1630 	[IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1631 	[IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1632 	[IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1633 	[IFLA_GRE_TTL]         = { .type = NLA_U8 },
1634 	[IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1635 	[IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1636 	[IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1637 };
1638 
1639 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1640 	.kind		= "ip6gre",
1641 	.maxtype	= IFLA_GRE_MAX,
1642 	.policy		= ip6gre_policy,
1643 	.priv_size	= sizeof(struct ip6_tnl),
1644 	.setup		= ip6gre_tunnel_setup,
1645 	.validate	= ip6gre_tunnel_validate,
1646 	.newlink	= ip6gre_newlink,
1647 	.changelink	= ip6gre_changelink,
1648 	.get_size	= ip6gre_get_size,
1649 	.fill_info	= ip6gre_fill_info,
1650 };
1651 
1652 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1653 	.kind		= "ip6gretap",
1654 	.maxtype	= IFLA_GRE_MAX,
1655 	.policy		= ip6gre_policy,
1656 	.priv_size	= sizeof(struct ip6_tnl),
1657 	.setup		= ip6gre_tap_setup,
1658 	.validate	= ip6gre_tap_validate,
1659 	.newlink	= ip6gre_newlink,
1660 	.changelink	= ip6gre_changelink,
1661 	.get_size	= ip6gre_get_size,
1662 	.fill_info	= ip6gre_fill_info,
1663 };
1664 
1665 /*
1666  *	And now the modules code and kernel interface.
1667  */
1668 
1669 static int __init ip6gre_init(void)
1670 {
1671 	int err;
1672 
1673 	pr_info("GRE over IPv6 tunneling driver\n");
1674 
1675 	err = register_pernet_device(&ip6gre_net_ops);
1676 	if (err < 0)
1677 		return err;
1678 
1679 	err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1680 	if (err < 0) {
1681 		pr_info("%s: can't add protocol\n", __func__);
1682 		goto add_proto_failed;
1683 	}
1684 
1685 	err = rtnl_link_register(&ip6gre_link_ops);
1686 	if (err < 0)
1687 		goto rtnl_link_failed;
1688 
1689 	err = rtnl_link_register(&ip6gre_tap_ops);
1690 	if (err < 0)
1691 		goto tap_ops_failed;
1692 
1693 out:
1694 	return err;
1695 
1696 tap_ops_failed:
1697 	rtnl_link_unregister(&ip6gre_link_ops);
1698 rtnl_link_failed:
1699 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1700 add_proto_failed:
1701 	unregister_pernet_device(&ip6gre_net_ops);
1702 	goto out;
1703 }
1704 
1705 static void __exit ip6gre_fini(void)
1706 {
1707 	rtnl_link_unregister(&ip6gre_tap_ops);
1708 	rtnl_link_unregister(&ip6gre_link_ops);
1709 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1710 	unregister_pernet_device(&ip6gre_net_ops);
1711 }
1712 
1713 module_init(ip6gre_init);
1714 module_exit(ip6gre_fini);
1715 MODULE_LICENSE("GPL");
1716 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1717 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1718 MODULE_ALIAS_RTNL_LINK("ip6gre");
1719 MODULE_ALIAS_NETDEV("ip6gre0");
1720