xref: /openbmc/linux/net/ipv6/ip6_gre.c (revision 12eb4683)
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_MASK (IPV6_FLOWINFO_MASK & ~IPV6_FLOWLABEL_MASK)
65 #define IPV6_TCLASS_SHIFT 20
66 
67 #define HASH_SIZE_SHIFT  5
68 #define HASH_SIZE (1 << HASH_SIZE_SHIFT)
69 
70 static int ip6gre_net_id __read_mostly;
71 struct ip6gre_net {
72 	struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
73 
74 	struct net_device *fb_tunnel_dev;
75 };
76 
77 static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
78 static int ip6gre_tunnel_init(struct net_device *dev);
79 static void ip6gre_tunnel_setup(struct net_device *dev);
80 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
81 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
82 
83 /* Tunnel hash table */
84 
85 /*
86    4 hash tables:
87 
88    3: (remote,local)
89    2: (remote,*)
90    1: (*,local)
91    0: (*,*)
92 
93    We require exact key match i.e. if a key is present in packet
94    it will match only tunnel with the same key; if it is not present,
95    it will match only keyless tunnel.
96 
97    All keysless packets, if not matched configured keyless tunnels
98    will match fallback tunnel.
99  */
100 
101 #define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
102 static u32 HASH_ADDR(const struct in6_addr *addr)
103 {
104 	u32 hash = ipv6_addr_hash(addr);
105 
106 	return hash_32(hash, HASH_SIZE_SHIFT);
107 }
108 
109 #define tunnels_r_l	tunnels[3]
110 #define tunnels_r	tunnels[2]
111 #define tunnels_l	tunnels[1]
112 #define tunnels_wc	tunnels[0]
113 
114 /* Given src, dst and key, find appropriate for input tunnel. */
115 
116 static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
117 		const struct in6_addr *remote, const struct in6_addr *local,
118 		__be32 key, __be16 gre_proto)
119 {
120 	struct net *net = dev_net(dev);
121 	int link = dev->ifindex;
122 	unsigned int h0 = HASH_ADDR(remote);
123 	unsigned int h1 = HASH_KEY(key);
124 	struct ip6_tnl *t, *cand = NULL;
125 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
126 	int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
127 		       ARPHRD_ETHER : ARPHRD_IP6GRE;
128 	int score, cand_score = 4;
129 
130 	for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
131 		if (!ipv6_addr_equal(local, &t->parms.laddr) ||
132 		    !ipv6_addr_equal(remote, &t->parms.raddr) ||
133 		    key != t->parms.i_key ||
134 		    !(t->dev->flags & IFF_UP))
135 			continue;
136 
137 		if (t->dev->type != ARPHRD_IP6GRE &&
138 		    t->dev->type != dev_type)
139 			continue;
140 
141 		score = 0;
142 		if (t->parms.link != link)
143 			score |= 1;
144 		if (t->dev->type != dev_type)
145 			score |= 2;
146 		if (score == 0)
147 			return t;
148 
149 		if (score < cand_score) {
150 			cand = t;
151 			cand_score = score;
152 		}
153 	}
154 
155 	for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
156 		if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
157 		    key != t->parms.i_key ||
158 		    !(t->dev->flags & IFF_UP))
159 			continue;
160 
161 		if (t->dev->type != ARPHRD_IP6GRE &&
162 		    t->dev->type != dev_type)
163 			continue;
164 
165 		score = 0;
166 		if (t->parms.link != link)
167 			score |= 1;
168 		if (t->dev->type != dev_type)
169 			score |= 2;
170 		if (score == 0)
171 			return t;
172 
173 		if (score < cand_score) {
174 			cand = t;
175 			cand_score = score;
176 		}
177 	}
178 
179 	for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
180 		if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
181 			  (!ipv6_addr_equal(local, &t->parms.raddr) ||
182 				 !ipv6_addr_is_multicast(local))) ||
183 		    key != t->parms.i_key ||
184 		    !(t->dev->flags & IFF_UP))
185 			continue;
186 
187 		if (t->dev->type != ARPHRD_IP6GRE &&
188 		    t->dev->type != dev_type)
189 			continue;
190 
191 		score = 0;
192 		if (t->parms.link != link)
193 			score |= 1;
194 		if (t->dev->type != dev_type)
195 			score |= 2;
196 		if (score == 0)
197 			return t;
198 
199 		if (score < cand_score) {
200 			cand = t;
201 			cand_score = score;
202 		}
203 	}
204 
205 	for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
206 		if (t->parms.i_key != key ||
207 		    !(t->dev->flags & IFF_UP))
208 			continue;
209 
210 		if (t->dev->type != ARPHRD_IP6GRE &&
211 		    t->dev->type != dev_type)
212 			continue;
213 
214 		score = 0;
215 		if (t->parms.link != link)
216 			score |= 1;
217 		if (t->dev->type != dev_type)
218 			score |= 2;
219 		if (score == 0)
220 			return t;
221 
222 		if (score < cand_score) {
223 			cand = t;
224 			cand_score = score;
225 		}
226 	}
227 
228 	if (cand != NULL)
229 		return cand;
230 
231 	dev = ign->fb_tunnel_dev;
232 	if (dev->flags & IFF_UP)
233 		return netdev_priv(dev);
234 
235 	return NULL;
236 }
237 
238 static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
239 		const struct __ip6_tnl_parm *p)
240 {
241 	const struct in6_addr *remote = &p->raddr;
242 	const struct in6_addr *local = &p->laddr;
243 	unsigned int h = HASH_KEY(p->i_key);
244 	int prio = 0;
245 
246 	if (!ipv6_addr_any(local))
247 		prio |= 1;
248 	if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
249 		prio |= 2;
250 		h ^= HASH_ADDR(remote);
251 	}
252 
253 	return &ign->tunnels[prio][h];
254 }
255 
256 static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
257 		const struct ip6_tnl *t)
258 {
259 	return __ip6gre_bucket(ign, &t->parms);
260 }
261 
262 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
263 {
264 	struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
265 
266 	rcu_assign_pointer(t->next, rtnl_dereference(*tp));
267 	rcu_assign_pointer(*tp, t);
268 }
269 
270 static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
271 {
272 	struct ip6_tnl __rcu **tp;
273 	struct ip6_tnl *iter;
274 
275 	for (tp = ip6gre_bucket(ign, t);
276 	     (iter = rtnl_dereference(*tp)) != NULL;
277 	     tp = &iter->next) {
278 		if (t == iter) {
279 			rcu_assign_pointer(*tp, t->next);
280 			break;
281 		}
282 	}
283 }
284 
285 static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
286 					   const struct __ip6_tnl_parm *parms,
287 					   int type)
288 {
289 	const struct in6_addr *remote = &parms->raddr;
290 	const struct in6_addr *local = &parms->laddr;
291 	__be32 key = parms->i_key;
292 	int link = parms->link;
293 	struct ip6_tnl *t;
294 	struct ip6_tnl __rcu **tp;
295 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
296 
297 	for (tp = __ip6gre_bucket(ign, parms);
298 	     (t = rtnl_dereference(*tp)) != NULL;
299 	     tp = &t->next)
300 		if (ipv6_addr_equal(local, &t->parms.laddr) &&
301 		    ipv6_addr_equal(remote, &t->parms.raddr) &&
302 		    key == t->parms.i_key &&
303 		    link == t->parms.link &&
304 		    type == t->dev->type)
305 			break;
306 
307 	return t;
308 }
309 
310 static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
311 		const struct __ip6_tnl_parm *parms, int create)
312 {
313 	struct ip6_tnl *t, *nt;
314 	struct net_device *dev;
315 	char name[IFNAMSIZ];
316 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
317 
318 	t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
319 	if (t || !create)
320 		return t;
321 
322 	if (parms->name[0])
323 		strlcpy(name, parms->name, IFNAMSIZ);
324 	else
325 		strcpy(name, "ip6gre%d");
326 
327 	dev = alloc_netdev(sizeof(*t), name, ip6gre_tunnel_setup);
328 	if (!dev)
329 		return NULL;
330 
331 	dev_net_set(dev, net);
332 
333 	nt = netdev_priv(dev);
334 	nt->parms = *parms;
335 	dev->rtnl_link_ops = &ip6gre_link_ops;
336 
337 	nt->dev = dev;
338 	nt->net = dev_net(dev);
339 	ip6gre_tnl_link_config(nt, 1);
340 
341 	if (register_netdevice(dev) < 0)
342 		goto failed_free;
343 
344 	/* Can use a lockless transmit, unless we generate output sequences */
345 	if (!(nt->parms.o_flags & GRE_SEQ))
346 		dev->features |= NETIF_F_LLTX;
347 
348 	dev_hold(dev);
349 	ip6gre_tunnel_link(ign, nt);
350 	return nt;
351 
352 failed_free:
353 	free_netdev(dev);
354 	return NULL;
355 }
356 
357 static void ip6gre_tunnel_uninit(struct net_device *dev)
358 {
359 	struct net *net = dev_net(dev);
360 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
361 
362 	ip6gre_tunnel_unlink(ign, netdev_priv(dev));
363 	dev_put(dev);
364 }
365 
366 
367 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
368 		u8 type, u8 code, int offset, __be32 info)
369 {
370 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
371 	__be16 *p = (__be16 *)(skb->data + offset);
372 	int grehlen = offset + 4;
373 	struct ip6_tnl *t;
374 	__be16 flags;
375 
376 	flags = p[0];
377 	if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
378 		if (flags&(GRE_VERSION|GRE_ROUTING))
379 			return;
380 		if (flags&GRE_KEY) {
381 			grehlen += 4;
382 			if (flags&GRE_CSUM)
383 				grehlen += 4;
384 		}
385 	}
386 
387 	/* If only 8 bytes returned, keyed message will be dropped here */
388 	if (!pskb_may_pull(skb, grehlen))
389 		return;
390 	ipv6h = (const struct ipv6hdr *)skb->data;
391 	p = (__be16 *)(skb->data + offset);
392 
393 	t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
394 				flags & GRE_KEY ?
395 				*(((__be32 *)p) + (grehlen / 4) - 1) : 0,
396 				p[1]);
397 	if (t == NULL)
398 		return;
399 
400 	switch (type) {
401 		__u32 teli;
402 		struct ipv6_tlv_tnl_enc_lim *tel;
403 		__u32 mtu;
404 	case ICMPV6_DEST_UNREACH:
405 		net_warn_ratelimited("%s: Path to destination invalid or inactive!\n",
406 				     t->parms.name);
407 		break;
408 	case ICMPV6_TIME_EXCEED:
409 		if (code == ICMPV6_EXC_HOPLIMIT) {
410 			net_warn_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
411 					     t->parms.name);
412 		}
413 		break;
414 	case ICMPV6_PARAMPROB:
415 		teli = 0;
416 		if (code == ICMPV6_HDR_FIELD)
417 			teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
418 
419 		if (teli && teli == info - 2) {
420 			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
421 			if (tel->encap_limit == 0) {
422 				net_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
423 						     t->parms.name);
424 			}
425 		} else {
426 			net_warn_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
427 					     t->parms.name);
428 		}
429 		break;
430 	case ICMPV6_PKT_TOOBIG:
431 		mtu = info - offset;
432 		if (mtu < IPV6_MIN_MTU)
433 			mtu = IPV6_MIN_MTU;
434 		t->dev->mtu = mtu;
435 		break;
436 	}
437 
438 	if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
439 		t->err_count++;
440 	else
441 		t->err_count = 1;
442 	t->err_time = jiffies;
443 }
444 
445 static int ip6gre_rcv(struct sk_buff *skb)
446 {
447 	const struct ipv6hdr *ipv6h;
448 	u8     *h;
449 	__be16    flags;
450 	__sum16   csum = 0;
451 	__be32 key = 0;
452 	u32    seqno = 0;
453 	struct ip6_tnl *tunnel;
454 	int    offset = 4;
455 	__be16 gre_proto;
456 	int err;
457 
458 	if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
459 		goto drop;
460 
461 	ipv6h = ipv6_hdr(skb);
462 	h = skb->data;
463 	flags = *(__be16 *)h;
464 
465 	if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
466 		/* - Version must be 0.
467 		   - We do not support routing headers.
468 		 */
469 		if (flags&(GRE_VERSION|GRE_ROUTING))
470 			goto drop;
471 
472 		if (flags&GRE_CSUM) {
473 			switch (skb->ip_summed) {
474 			case CHECKSUM_COMPLETE:
475 				csum = csum_fold(skb->csum);
476 				if (!csum)
477 					break;
478 				/* fall through */
479 			case CHECKSUM_NONE:
480 				skb->csum = 0;
481 				csum = __skb_checksum_complete(skb);
482 				skb->ip_summed = CHECKSUM_COMPLETE;
483 			}
484 			offset += 4;
485 		}
486 		if (flags&GRE_KEY) {
487 			key = *(__be32 *)(h + offset);
488 			offset += 4;
489 		}
490 		if (flags&GRE_SEQ) {
491 			seqno = ntohl(*(__be32 *)(h + offset));
492 			offset += 4;
493 		}
494 	}
495 
496 	gre_proto = *(__be16 *)(h + 2);
497 
498 	tunnel = ip6gre_tunnel_lookup(skb->dev,
499 					  &ipv6h->saddr, &ipv6h->daddr, key,
500 					  gre_proto);
501 	if (tunnel) {
502 		struct pcpu_tstats *tstats;
503 
504 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
505 			goto drop;
506 
507 		if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
508 			tunnel->dev->stats.rx_dropped++;
509 			goto drop;
510 		}
511 
512 		skb->protocol = gre_proto;
513 		/* WCCP version 1 and 2 protocol decoding.
514 		 * - Change protocol to IP
515 		 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
516 		 */
517 		if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
518 			skb->protocol = htons(ETH_P_IP);
519 			if ((*(h + offset) & 0xF0) != 0x40)
520 				offset += 4;
521 		}
522 
523 		skb->mac_header = skb->network_header;
524 		__pskb_pull(skb, offset);
525 		skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
526 
527 		if (((flags&GRE_CSUM) && csum) ||
528 		    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
529 			tunnel->dev->stats.rx_crc_errors++;
530 			tunnel->dev->stats.rx_errors++;
531 			goto drop;
532 		}
533 		if (tunnel->parms.i_flags&GRE_SEQ) {
534 			if (!(flags&GRE_SEQ) ||
535 			    (tunnel->i_seqno &&
536 					(s32)(seqno - tunnel->i_seqno) < 0)) {
537 				tunnel->dev->stats.rx_fifo_errors++;
538 				tunnel->dev->stats.rx_errors++;
539 				goto drop;
540 			}
541 			tunnel->i_seqno = seqno + 1;
542 		}
543 
544 		/* Warning: All skb pointers will be invalidated! */
545 		if (tunnel->dev->type == ARPHRD_ETHER) {
546 			if (!pskb_may_pull(skb, ETH_HLEN)) {
547 				tunnel->dev->stats.rx_length_errors++;
548 				tunnel->dev->stats.rx_errors++;
549 				goto drop;
550 			}
551 
552 			ipv6h = ipv6_hdr(skb);
553 			skb->protocol = eth_type_trans(skb, tunnel->dev);
554 			skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
555 		}
556 
557 		__skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
558 
559 		skb_reset_network_header(skb);
560 
561 		err = IP6_ECN_decapsulate(ipv6h, skb);
562 		if (unlikely(err)) {
563 			if (log_ecn_error)
564 				net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
565 						     &ipv6h->saddr,
566 						     ipv6_get_dsfield(ipv6h));
567 			if (err > 1) {
568 				++tunnel->dev->stats.rx_frame_errors;
569 				++tunnel->dev->stats.rx_errors;
570 				goto drop;
571 			}
572 		}
573 
574 		tstats = this_cpu_ptr(tunnel->dev->tstats);
575 		u64_stats_update_begin(&tstats->syncp);
576 		tstats->rx_packets++;
577 		tstats->rx_bytes += skb->len;
578 		u64_stats_update_end(&tstats->syncp);
579 
580 		netif_rx(skb);
581 
582 		return 0;
583 	}
584 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
585 
586 drop:
587 	kfree_skb(skb);
588 	return 0;
589 }
590 
591 struct ipv6_tel_txoption {
592 	struct ipv6_txoptions ops;
593 	__u8 dst_opt[8];
594 };
595 
596 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
597 {
598 	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
599 
600 	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
601 	opt->dst_opt[3] = 1;
602 	opt->dst_opt[4] = encap_limit;
603 	opt->dst_opt[5] = IPV6_TLV_PADN;
604 	opt->dst_opt[6] = 1;
605 
606 	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
607 	opt->ops.opt_nflen = 8;
608 }
609 
610 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
611 			 struct net_device *dev,
612 			 __u8 dsfield,
613 			 struct flowi6 *fl6,
614 			 int encap_limit,
615 			 __u32 *pmtu)
616 {
617 	struct net *net = dev_net(dev);
618 	struct ip6_tnl *tunnel = netdev_priv(dev);
619 	struct net_device *tdev;    /* Device to other host */
620 	struct ipv6hdr  *ipv6h;     /* Our new IP header */
621 	unsigned int max_headroom = 0; /* The extra header space needed */
622 	int    gre_hlen;
623 	struct ipv6_tel_txoption opt;
624 	int    mtu;
625 	struct dst_entry *dst = NULL, *ndst = NULL;
626 	struct net_device_stats *stats = &tunnel->dev->stats;
627 	int err = -1;
628 	u8 proto;
629 	struct sk_buff *new_skb;
630 
631 	if (dev->type == ARPHRD_ETHER)
632 		IPCB(skb)->flags = 0;
633 
634 	if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
635 		gre_hlen = 0;
636 		ipv6h = (struct ipv6hdr *)skb->data;
637 		fl6->daddr = ipv6h->daddr;
638 	} else {
639 		gre_hlen = tunnel->hlen;
640 		fl6->daddr = tunnel->parms.raddr;
641 	}
642 
643 	if (!fl6->flowi6_mark)
644 		dst = ip6_tnl_dst_check(tunnel);
645 
646 	if (!dst) {
647 		ndst = ip6_route_output(net, NULL, fl6);
648 
649 		if (ndst->error)
650 			goto tx_err_link_failure;
651 		ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
652 		if (IS_ERR(ndst)) {
653 			err = PTR_ERR(ndst);
654 			ndst = NULL;
655 			goto tx_err_link_failure;
656 		}
657 		dst = ndst;
658 	}
659 
660 	tdev = dst->dev;
661 
662 	if (tdev == dev) {
663 		stats->collisions++;
664 		net_warn_ratelimited("%s: Local routing loop detected!\n",
665 				     tunnel->parms.name);
666 		goto tx_err_dst_release;
667 	}
668 
669 	mtu = dst_mtu(dst) - sizeof(*ipv6h);
670 	if (encap_limit >= 0) {
671 		max_headroom += 8;
672 		mtu -= 8;
673 	}
674 	if (mtu < IPV6_MIN_MTU)
675 		mtu = IPV6_MIN_MTU;
676 	if (skb_dst(skb))
677 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
678 	if (skb->len > mtu) {
679 		*pmtu = mtu;
680 		err = -EMSGSIZE;
681 		goto tx_err_dst_release;
682 	}
683 
684 	if (tunnel->err_count > 0) {
685 		if (time_before(jiffies,
686 				tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
687 			tunnel->err_count--;
688 
689 			dst_link_failure(skb);
690 		} else
691 			tunnel->err_count = 0;
692 	}
693 
694 	skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
695 
696 	max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
697 
698 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
699 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
700 		new_skb = skb_realloc_headroom(skb, max_headroom);
701 		if (max_headroom > dev->needed_headroom)
702 			dev->needed_headroom = max_headroom;
703 		if (!new_skb)
704 			goto tx_err_dst_release;
705 
706 		if (skb->sk)
707 			skb_set_owner_w(new_skb, skb->sk);
708 		consume_skb(skb);
709 		skb = new_skb;
710 	}
711 
712 	if (fl6->flowi6_mark) {
713 		skb_dst_set(skb, dst);
714 		ndst = NULL;
715 	} else {
716 		skb_dst_set_noref(skb, dst);
717 	}
718 
719 	proto = NEXTHDR_GRE;
720 	if (encap_limit >= 0) {
721 		init_tel_txopt(&opt, encap_limit);
722 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
723 	}
724 
725 	if (likely(!skb->encapsulation)) {
726 		skb_reset_inner_headers(skb);
727 		skb->encapsulation = 1;
728 	}
729 
730 	skb_push(skb, gre_hlen);
731 	skb_reset_network_header(skb);
732 	skb_set_transport_header(skb, sizeof(*ipv6h));
733 
734 	/*
735 	 *	Push down and install the IP header.
736 	 */
737 	ipv6h = ipv6_hdr(skb);
738 	ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);
739 	ipv6h->hop_limit = tunnel->parms.hop_limit;
740 	ipv6h->nexthdr = proto;
741 	ipv6h->saddr = fl6->saddr;
742 	ipv6h->daddr = fl6->daddr;
743 
744 	((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
745 	((__be16 *)(ipv6h + 1))[1] = (dev->type == ARPHRD_ETHER) ?
746 				   htons(ETH_P_TEB) : skb->protocol;
747 
748 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
749 		__be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
750 
751 		if (tunnel->parms.o_flags&GRE_SEQ) {
752 			++tunnel->o_seqno;
753 			*ptr = htonl(tunnel->o_seqno);
754 			ptr--;
755 		}
756 		if (tunnel->parms.o_flags&GRE_KEY) {
757 			*ptr = tunnel->parms.o_key;
758 			ptr--;
759 		}
760 		if (tunnel->parms.o_flags&GRE_CSUM) {
761 			*ptr = 0;
762 			*(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
763 				skb->len - sizeof(struct ipv6hdr));
764 		}
765 	}
766 
767 	ip6tunnel_xmit(skb, dev);
768 	if (ndst)
769 		ip6_tnl_dst_store(tunnel, ndst);
770 	return 0;
771 tx_err_link_failure:
772 	stats->tx_carrier_errors++;
773 	dst_link_failure(skb);
774 tx_err_dst_release:
775 	dst_release(ndst);
776 	return err;
777 }
778 
779 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
780 {
781 	struct ip6_tnl *t = netdev_priv(dev);
782 	const struct iphdr  *iph = ip_hdr(skb);
783 	int encap_limit = -1;
784 	struct flowi6 fl6;
785 	__u8 dsfield;
786 	__u32 mtu;
787 	int err;
788 
789 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
790 		encap_limit = t->parms.encap_limit;
791 
792 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
793 	fl6.flowi6_proto = IPPROTO_IPIP;
794 
795 	dsfield = ipv4_get_dsfield(iph);
796 
797 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
798 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
799 					  & IPV6_TCLASS_MASK;
800 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
801 		fl6.flowi6_mark = skb->mark;
802 
803 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
804 	if (err != 0) {
805 		/* XXX: send ICMP error even if DF is not set. */
806 		if (err == -EMSGSIZE)
807 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
808 				  htonl(mtu));
809 		return -1;
810 	}
811 
812 	return 0;
813 }
814 
815 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
816 {
817 	struct ip6_tnl *t = netdev_priv(dev);
818 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
819 	int encap_limit = -1;
820 	__u16 offset;
821 	struct flowi6 fl6;
822 	__u8 dsfield;
823 	__u32 mtu;
824 	int err;
825 
826 	if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
827 		return -1;
828 
829 	offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
830 	if (offset > 0) {
831 		struct ipv6_tlv_tnl_enc_lim *tel;
832 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
833 		if (tel->encap_limit == 0) {
834 			icmpv6_send(skb, ICMPV6_PARAMPROB,
835 				    ICMPV6_HDR_FIELD, offset + 2);
836 			return -1;
837 		}
838 		encap_limit = tel->encap_limit - 1;
839 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
840 		encap_limit = t->parms.encap_limit;
841 
842 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
843 	fl6.flowi6_proto = IPPROTO_IPV6;
844 
845 	dsfield = ipv6_get_dsfield(ipv6h);
846 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
847 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
848 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
849 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_FLOWLABEL_MASK);
850 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
851 		fl6.flowi6_mark = skb->mark;
852 
853 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
854 	if (err != 0) {
855 		if (err == -EMSGSIZE)
856 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
857 		return -1;
858 	}
859 
860 	return 0;
861 }
862 
863 /**
864  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
865  *   @t: the outgoing tunnel device
866  *   @hdr: IPv6 header from the incoming packet
867  *
868  * Description:
869  *   Avoid trivial tunneling loop by checking that tunnel exit-point
870  *   doesn't match source of incoming packet.
871  *
872  * Return:
873  *   1 if conflict,
874  *   0 else
875  **/
876 
877 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
878 	const struct ipv6hdr *hdr)
879 {
880 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
881 }
882 
883 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
884 {
885 	struct ip6_tnl *t = netdev_priv(dev);
886 	int encap_limit = -1;
887 	struct flowi6 fl6;
888 	__u32 mtu;
889 	int err;
890 
891 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
892 		encap_limit = t->parms.encap_limit;
893 
894 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
895 	fl6.flowi6_proto = skb->protocol;
896 
897 	err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
898 
899 	return err;
900 }
901 
902 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
903 	struct net_device *dev)
904 {
905 	struct ip6_tnl *t = netdev_priv(dev);
906 	struct net_device_stats *stats = &t->dev->stats;
907 	int ret;
908 
909 	if (!ip6_tnl_xmit_ctl(t))
910 		goto tx_err;
911 
912 	switch (skb->protocol) {
913 	case htons(ETH_P_IP):
914 		ret = ip6gre_xmit_ipv4(skb, dev);
915 		break;
916 	case htons(ETH_P_IPV6):
917 		ret = ip6gre_xmit_ipv6(skb, dev);
918 		break;
919 	default:
920 		ret = ip6gre_xmit_other(skb, dev);
921 		break;
922 	}
923 
924 	if (ret < 0)
925 		goto tx_err;
926 
927 	return NETDEV_TX_OK;
928 
929 tx_err:
930 	stats->tx_errors++;
931 	stats->tx_dropped++;
932 	kfree_skb(skb);
933 	return NETDEV_TX_OK;
934 }
935 
936 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
937 {
938 	struct net_device *dev = t->dev;
939 	struct __ip6_tnl_parm *p = &t->parms;
940 	struct flowi6 *fl6 = &t->fl.u.ip6;
941 	int addend = sizeof(struct ipv6hdr) + 4;
942 
943 	if (dev->type != ARPHRD_ETHER) {
944 		memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
945 		memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
946 	}
947 
948 	/* Set up flowi template */
949 	fl6->saddr = p->laddr;
950 	fl6->daddr = p->raddr;
951 	fl6->flowi6_oif = p->link;
952 	fl6->flowlabel = 0;
953 
954 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
955 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
956 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
957 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
958 
959 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
960 	p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
961 
962 	if (p->flags&IP6_TNL_F_CAP_XMIT &&
963 			p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
964 		dev->flags |= IFF_POINTOPOINT;
965 	else
966 		dev->flags &= ~IFF_POINTOPOINT;
967 
968 	dev->iflink = p->link;
969 
970 	/* Precalculate GRE options length */
971 	if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
972 		if (t->parms.o_flags&GRE_CSUM)
973 			addend += 4;
974 		if (t->parms.o_flags&GRE_KEY)
975 			addend += 4;
976 		if (t->parms.o_flags&GRE_SEQ)
977 			addend += 4;
978 	}
979 	t->hlen = addend;
980 
981 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
982 		int strict = (ipv6_addr_type(&p->raddr) &
983 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
984 
985 		struct rt6_info *rt = rt6_lookup(dev_net(dev),
986 						 &p->raddr, &p->laddr,
987 						 p->link, strict);
988 
989 		if (rt == NULL)
990 			return;
991 
992 		if (rt->dst.dev) {
993 			dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
994 
995 			if (set_mtu) {
996 				dev->mtu = rt->dst.dev->mtu - addend;
997 				if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
998 					dev->mtu -= 8;
999 
1000 				if (dev->mtu < IPV6_MIN_MTU)
1001 					dev->mtu = IPV6_MIN_MTU;
1002 			}
1003 		}
1004 		ip6_rt_put(rt);
1005 	}
1006 }
1007 
1008 static int ip6gre_tnl_change(struct ip6_tnl *t,
1009 	const struct __ip6_tnl_parm *p, int set_mtu)
1010 {
1011 	t->parms.laddr = p->laddr;
1012 	t->parms.raddr = p->raddr;
1013 	t->parms.flags = p->flags;
1014 	t->parms.hop_limit = p->hop_limit;
1015 	t->parms.encap_limit = p->encap_limit;
1016 	t->parms.flowinfo = p->flowinfo;
1017 	t->parms.link = p->link;
1018 	t->parms.proto = p->proto;
1019 	t->parms.i_key = p->i_key;
1020 	t->parms.o_key = p->o_key;
1021 	t->parms.i_flags = p->i_flags;
1022 	t->parms.o_flags = p->o_flags;
1023 	ip6_tnl_dst_reset(t);
1024 	ip6gre_tnl_link_config(t, set_mtu);
1025 	return 0;
1026 }
1027 
1028 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1029 	const struct ip6_tnl_parm2 *u)
1030 {
1031 	p->laddr = u->laddr;
1032 	p->raddr = u->raddr;
1033 	p->flags = u->flags;
1034 	p->hop_limit = u->hop_limit;
1035 	p->encap_limit = u->encap_limit;
1036 	p->flowinfo = u->flowinfo;
1037 	p->link = u->link;
1038 	p->i_key = u->i_key;
1039 	p->o_key = u->o_key;
1040 	p->i_flags = u->i_flags;
1041 	p->o_flags = u->o_flags;
1042 	memcpy(p->name, u->name, sizeof(u->name));
1043 }
1044 
1045 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1046 	const struct __ip6_tnl_parm *p)
1047 {
1048 	u->proto = IPPROTO_GRE;
1049 	u->laddr = p->laddr;
1050 	u->raddr = p->raddr;
1051 	u->flags = p->flags;
1052 	u->hop_limit = p->hop_limit;
1053 	u->encap_limit = p->encap_limit;
1054 	u->flowinfo = p->flowinfo;
1055 	u->link = p->link;
1056 	u->i_key = p->i_key;
1057 	u->o_key = p->o_key;
1058 	u->i_flags = p->i_flags;
1059 	u->o_flags = p->o_flags;
1060 	memcpy(u->name, p->name, sizeof(u->name));
1061 }
1062 
1063 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1064 	struct ifreq *ifr, int cmd)
1065 {
1066 	int err = 0;
1067 	struct ip6_tnl_parm2 p;
1068 	struct __ip6_tnl_parm p1;
1069 	struct ip6_tnl *t;
1070 	struct net *net = dev_net(dev);
1071 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1072 
1073 	switch (cmd) {
1074 	case SIOCGETTUNNEL:
1075 		t = NULL;
1076 		if (dev == ign->fb_tunnel_dev) {
1077 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1078 				err = -EFAULT;
1079 				break;
1080 			}
1081 			ip6gre_tnl_parm_from_user(&p1, &p);
1082 			t = ip6gre_tunnel_locate(net, &p1, 0);
1083 		}
1084 		if (t == NULL)
1085 			t = netdev_priv(dev);
1086 		memset(&p, 0, sizeof(p));
1087 		ip6gre_tnl_parm_to_user(&p, &t->parms);
1088 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1089 			err = -EFAULT;
1090 		break;
1091 
1092 	case SIOCADDTUNNEL:
1093 	case SIOCCHGTUNNEL:
1094 		err = -EPERM;
1095 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1096 			goto done;
1097 
1098 		err = -EFAULT;
1099 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1100 			goto done;
1101 
1102 		err = -EINVAL;
1103 		if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1104 			goto done;
1105 
1106 		if (!(p.i_flags&GRE_KEY))
1107 			p.i_key = 0;
1108 		if (!(p.o_flags&GRE_KEY))
1109 			p.o_key = 0;
1110 
1111 		ip6gre_tnl_parm_from_user(&p1, &p);
1112 		t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1113 
1114 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1115 			if (t != NULL) {
1116 				if (t->dev != dev) {
1117 					err = -EEXIST;
1118 					break;
1119 				}
1120 			} else {
1121 				t = netdev_priv(dev);
1122 
1123 				ip6gre_tunnel_unlink(ign, t);
1124 				synchronize_net();
1125 				ip6gre_tnl_change(t, &p1, 1);
1126 				ip6gre_tunnel_link(ign, t);
1127 				netdev_state_change(dev);
1128 			}
1129 		}
1130 
1131 		if (t) {
1132 			err = 0;
1133 
1134 			memset(&p, 0, sizeof(p));
1135 			ip6gre_tnl_parm_to_user(&p, &t->parms);
1136 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1137 				err = -EFAULT;
1138 		} else
1139 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1140 		break;
1141 
1142 	case SIOCDELTUNNEL:
1143 		err = -EPERM;
1144 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1145 			goto done;
1146 
1147 		if (dev == ign->fb_tunnel_dev) {
1148 			err = -EFAULT;
1149 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1150 				goto done;
1151 			err = -ENOENT;
1152 			ip6gre_tnl_parm_from_user(&p1, &p);
1153 			t = ip6gre_tunnel_locate(net, &p1, 0);
1154 			if (t == NULL)
1155 				goto done;
1156 			err = -EPERM;
1157 			if (t == netdev_priv(ign->fb_tunnel_dev))
1158 				goto done;
1159 			dev = t->dev;
1160 		}
1161 		unregister_netdevice(dev);
1162 		err = 0;
1163 		break;
1164 
1165 	default:
1166 		err = -EINVAL;
1167 	}
1168 
1169 done:
1170 	return err;
1171 }
1172 
1173 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1174 {
1175 	if (new_mtu < 68 ||
1176 	    new_mtu > 0xFFF8 - dev->hard_header_len)
1177 		return -EINVAL;
1178 	dev->mtu = new_mtu;
1179 	return 0;
1180 }
1181 
1182 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1183 			unsigned short type,
1184 			const void *daddr, const void *saddr, unsigned int len)
1185 {
1186 	struct ip6_tnl *t = netdev_priv(dev);
1187 	struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1188 	__be16 *p = (__be16 *)(ipv6h+1);
1189 
1190 	ip6_flow_hdr(ipv6h, 0, t->fl.u.ip6.flowlabel);
1191 	ipv6h->hop_limit = t->parms.hop_limit;
1192 	ipv6h->nexthdr = NEXTHDR_GRE;
1193 	ipv6h->saddr = t->parms.laddr;
1194 	ipv6h->daddr = t->parms.raddr;
1195 
1196 	p[0]		= t->parms.o_flags;
1197 	p[1]		= htons(type);
1198 
1199 	/*
1200 	 *	Set the source hardware address.
1201 	 */
1202 
1203 	if (saddr)
1204 		memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1205 	if (daddr)
1206 		memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1207 	if (!ipv6_addr_any(&ipv6h->daddr))
1208 		return t->hlen;
1209 
1210 	return -t->hlen;
1211 }
1212 
1213 static const struct header_ops ip6gre_header_ops = {
1214 	.create	= ip6gre_header,
1215 };
1216 
1217 static const struct net_device_ops ip6gre_netdev_ops = {
1218 	.ndo_init		= ip6gre_tunnel_init,
1219 	.ndo_uninit		= ip6gre_tunnel_uninit,
1220 	.ndo_start_xmit		= ip6gre_tunnel_xmit,
1221 	.ndo_do_ioctl		= ip6gre_tunnel_ioctl,
1222 	.ndo_change_mtu		= ip6gre_tunnel_change_mtu,
1223 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1224 };
1225 
1226 static void ip6gre_dev_free(struct net_device *dev)
1227 {
1228 	free_percpu(dev->tstats);
1229 	free_netdev(dev);
1230 }
1231 
1232 static void ip6gre_tunnel_setup(struct net_device *dev)
1233 {
1234 	struct ip6_tnl *t;
1235 
1236 	dev->netdev_ops = &ip6gre_netdev_ops;
1237 	dev->destructor = ip6gre_dev_free;
1238 
1239 	dev->type = ARPHRD_IP6GRE;
1240 	dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1241 	dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1242 	t = netdev_priv(dev);
1243 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1244 		dev->mtu -= 8;
1245 	dev->flags |= IFF_NOARP;
1246 	dev->iflink = 0;
1247 	dev->addr_len = sizeof(struct in6_addr);
1248 	dev->features |= NETIF_F_NETNS_LOCAL;
1249 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1250 }
1251 
1252 static int ip6gre_tunnel_init(struct net_device *dev)
1253 {
1254 	struct ip6_tnl *tunnel;
1255 	int i;
1256 
1257 	tunnel = netdev_priv(dev);
1258 
1259 	tunnel->dev = dev;
1260 	tunnel->net = dev_net(dev);
1261 	strcpy(tunnel->parms.name, dev->name);
1262 
1263 	memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1264 	memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1265 
1266 	if (ipv6_addr_any(&tunnel->parms.raddr))
1267 		dev->header_ops = &ip6gre_header_ops;
1268 
1269 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1270 	if (!dev->tstats)
1271 		return -ENOMEM;
1272 
1273 	for_each_possible_cpu(i) {
1274 		struct pcpu_tstats *ip6gre_tunnel_stats;
1275 		ip6gre_tunnel_stats = per_cpu_ptr(dev->tstats, i);
1276 		u64_stats_init(&ip6gre_tunnel_stats->syncp);
1277 	}
1278 
1279 
1280 	return 0;
1281 }
1282 
1283 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1284 {
1285 	struct ip6_tnl *tunnel = netdev_priv(dev);
1286 
1287 	tunnel->dev = dev;
1288 	tunnel->net = dev_net(dev);
1289 	strcpy(tunnel->parms.name, dev->name);
1290 
1291 	tunnel->hlen		= sizeof(struct ipv6hdr) + 4;
1292 
1293 	dev_hold(dev);
1294 }
1295 
1296 
1297 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1298 	.handler     = ip6gre_rcv,
1299 	.err_handler = ip6gre_err,
1300 	.flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1301 };
1302 
1303 static void ip6gre_destroy_tunnels(struct ip6gre_net *ign,
1304 	struct list_head *head)
1305 {
1306 	int prio;
1307 
1308 	for (prio = 0; prio < 4; prio++) {
1309 		int h;
1310 		for (h = 0; h < HASH_SIZE; h++) {
1311 			struct ip6_tnl *t;
1312 
1313 			t = rtnl_dereference(ign->tunnels[prio][h]);
1314 
1315 			while (t != NULL) {
1316 				unregister_netdevice_queue(t->dev, head);
1317 				t = rtnl_dereference(t->next);
1318 			}
1319 		}
1320 	}
1321 }
1322 
1323 static int __net_init ip6gre_init_net(struct net *net)
1324 {
1325 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1326 	int err;
1327 
1328 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1329 					   ip6gre_tunnel_setup);
1330 	if (!ign->fb_tunnel_dev) {
1331 		err = -ENOMEM;
1332 		goto err_alloc_dev;
1333 	}
1334 	dev_net_set(ign->fb_tunnel_dev, net);
1335 
1336 	ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1337 	ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1338 
1339 	err = register_netdev(ign->fb_tunnel_dev);
1340 	if (err)
1341 		goto err_reg_dev;
1342 
1343 	rcu_assign_pointer(ign->tunnels_wc[0],
1344 			   netdev_priv(ign->fb_tunnel_dev));
1345 	return 0;
1346 
1347 err_reg_dev:
1348 	ip6gre_dev_free(ign->fb_tunnel_dev);
1349 err_alloc_dev:
1350 	return err;
1351 }
1352 
1353 static void __net_exit ip6gre_exit_net(struct net *net)
1354 {
1355 	struct ip6gre_net *ign;
1356 	LIST_HEAD(list);
1357 
1358 	ign = net_generic(net, ip6gre_net_id);
1359 	rtnl_lock();
1360 	ip6gre_destroy_tunnels(ign, &list);
1361 	unregister_netdevice_many(&list);
1362 	rtnl_unlock();
1363 }
1364 
1365 static struct pernet_operations ip6gre_net_ops = {
1366 	.init = ip6gre_init_net,
1367 	.exit = ip6gre_exit_net,
1368 	.id   = &ip6gre_net_id,
1369 	.size = sizeof(struct ip6gre_net),
1370 };
1371 
1372 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1373 {
1374 	__be16 flags;
1375 
1376 	if (!data)
1377 		return 0;
1378 
1379 	flags = 0;
1380 	if (data[IFLA_GRE_IFLAGS])
1381 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1382 	if (data[IFLA_GRE_OFLAGS])
1383 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1384 	if (flags & (GRE_VERSION|GRE_ROUTING))
1385 		return -EINVAL;
1386 
1387 	return 0;
1388 }
1389 
1390 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1391 {
1392 	struct in6_addr daddr;
1393 
1394 	if (tb[IFLA_ADDRESS]) {
1395 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1396 			return -EINVAL;
1397 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1398 			return -EADDRNOTAVAIL;
1399 	}
1400 
1401 	if (!data)
1402 		goto out;
1403 
1404 	if (data[IFLA_GRE_REMOTE]) {
1405 		nla_memcpy(&daddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1406 		if (ipv6_addr_any(&daddr))
1407 			return -EINVAL;
1408 	}
1409 
1410 out:
1411 	return ip6gre_tunnel_validate(tb, data);
1412 }
1413 
1414 
1415 static void ip6gre_netlink_parms(struct nlattr *data[],
1416 				struct __ip6_tnl_parm *parms)
1417 {
1418 	memset(parms, 0, sizeof(*parms));
1419 
1420 	if (!data)
1421 		return;
1422 
1423 	if (data[IFLA_GRE_LINK])
1424 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1425 
1426 	if (data[IFLA_GRE_IFLAGS])
1427 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1428 
1429 	if (data[IFLA_GRE_OFLAGS])
1430 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1431 
1432 	if (data[IFLA_GRE_IKEY])
1433 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1434 
1435 	if (data[IFLA_GRE_OKEY])
1436 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1437 
1438 	if (data[IFLA_GRE_LOCAL])
1439 		nla_memcpy(&parms->laddr, data[IFLA_GRE_LOCAL], sizeof(struct in6_addr));
1440 
1441 	if (data[IFLA_GRE_REMOTE])
1442 		nla_memcpy(&parms->raddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1443 
1444 	if (data[IFLA_GRE_TTL])
1445 		parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1446 
1447 	if (data[IFLA_GRE_ENCAP_LIMIT])
1448 		parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1449 
1450 	if (data[IFLA_GRE_FLOWINFO])
1451 		parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1452 
1453 	if (data[IFLA_GRE_FLAGS])
1454 		parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1455 }
1456 
1457 static int ip6gre_tap_init(struct net_device *dev)
1458 {
1459 	struct ip6_tnl *tunnel;
1460 	int i;
1461 
1462 	tunnel = netdev_priv(dev);
1463 
1464 	tunnel->dev = dev;
1465 	tunnel->net = dev_net(dev);
1466 	strcpy(tunnel->parms.name, dev->name);
1467 
1468 	ip6gre_tnl_link_config(tunnel, 1);
1469 
1470 	dev->tstats = alloc_percpu(struct pcpu_tstats);
1471 	if (!dev->tstats)
1472 		return -ENOMEM;
1473 
1474 	for_each_possible_cpu(i) {
1475 		struct pcpu_tstats *ip6gre_tap_stats;
1476 		ip6gre_tap_stats = per_cpu_ptr(dev->tstats, i);
1477 		u64_stats_init(&ip6gre_tap_stats->syncp);
1478 	}
1479 
1480 	return 0;
1481 }
1482 
1483 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1484 	.ndo_init = ip6gre_tap_init,
1485 	.ndo_uninit = ip6gre_tunnel_uninit,
1486 	.ndo_start_xmit = ip6gre_tunnel_xmit,
1487 	.ndo_set_mac_address = eth_mac_addr,
1488 	.ndo_validate_addr = eth_validate_addr,
1489 	.ndo_change_mtu = ip6gre_tunnel_change_mtu,
1490 	.ndo_get_stats64 = ip_tunnel_get_stats64,
1491 };
1492 
1493 static void ip6gre_tap_setup(struct net_device *dev)
1494 {
1495 
1496 	ether_setup(dev);
1497 
1498 	dev->netdev_ops = &ip6gre_tap_netdev_ops;
1499 	dev->destructor = ip6gre_dev_free;
1500 
1501 	dev->iflink = 0;
1502 	dev->features |= NETIF_F_NETNS_LOCAL;
1503 }
1504 
1505 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1506 	struct nlattr *tb[], struct nlattr *data[])
1507 {
1508 	struct ip6_tnl *nt;
1509 	struct net *net = dev_net(dev);
1510 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1511 	int err;
1512 
1513 	nt = netdev_priv(dev);
1514 	ip6gre_netlink_parms(data, &nt->parms);
1515 
1516 	if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1517 		return -EEXIST;
1518 
1519 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1520 		eth_hw_addr_random(dev);
1521 
1522 	nt->dev = dev;
1523 	nt->net = dev_net(dev);
1524 	ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1525 
1526 	/* Can use a lockless transmit, unless we generate output sequences */
1527 	if (!(nt->parms.o_flags & GRE_SEQ))
1528 		dev->features |= NETIF_F_LLTX;
1529 
1530 	err = register_netdevice(dev);
1531 	if (err)
1532 		goto out;
1533 
1534 	dev_hold(dev);
1535 	ip6gre_tunnel_link(ign, nt);
1536 
1537 out:
1538 	return err;
1539 }
1540 
1541 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1542 			    struct nlattr *data[])
1543 {
1544 	struct ip6_tnl *t, *nt;
1545 	struct net *net = dev_net(dev);
1546 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1547 	struct __ip6_tnl_parm p;
1548 
1549 	if (dev == ign->fb_tunnel_dev)
1550 		return -EINVAL;
1551 
1552 	nt = netdev_priv(dev);
1553 	ip6gre_netlink_parms(data, &p);
1554 
1555 	t = ip6gre_tunnel_locate(net, &p, 0);
1556 
1557 	if (t) {
1558 		if (t->dev != dev)
1559 			return -EEXIST;
1560 	} else {
1561 		t = nt;
1562 
1563 		ip6gre_tunnel_unlink(ign, t);
1564 		ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1565 		ip6gre_tunnel_link(ign, t);
1566 		netdev_state_change(dev);
1567 	}
1568 
1569 	return 0;
1570 }
1571 
1572 static size_t ip6gre_get_size(const struct net_device *dev)
1573 {
1574 	return
1575 		/* IFLA_GRE_LINK */
1576 		nla_total_size(4) +
1577 		/* IFLA_GRE_IFLAGS */
1578 		nla_total_size(2) +
1579 		/* IFLA_GRE_OFLAGS */
1580 		nla_total_size(2) +
1581 		/* IFLA_GRE_IKEY */
1582 		nla_total_size(4) +
1583 		/* IFLA_GRE_OKEY */
1584 		nla_total_size(4) +
1585 		/* IFLA_GRE_LOCAL */
1586 		nla_total_size(sizeof(struct in6_addr)) +
1587 		/* IFLA_GRE_REMOTE */
1588 		nla_total_size(sizeof(struct in6_addr)) +
1589 		/* IFLA_GRE_TTL */
1590 		nla_total_size(1) +
1591 		/* IFLA_GRE_TOS */
1592 		nla_total_size(1) +
1593 		/* IFLA_GRE_ENCAP_LIMIT */
1594 		nla_total_size(1) +
1595 		/* IFLA_GRE_FLOWINFO */
1596 		nla_total_size(4) +
1597 		/* IFLA_GRE_FLAGS */
1598 		nla_total_size(4) +
1599 		0;
1600 }
1601 
1602 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1603 {
1604 	struct ip6_tnl *t = netdev_priv(dev);
1605 	struct __ip6_tnl_parm *p = &t->parms;
1606 
1607 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1608 	    nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1609 	    nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1610 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1611 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1612 	    nla_put(skb, IFLA_GRE_LOCAL, sizeof(struct in6_addr), &p->laddr) ||
1613 	    nla_put(skb, IFLA_GRE_REMOTE, sizeof(struct in6_addr), &p->raddr) ||
1614 	    nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1615 	    /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1616 	    nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1617 	    nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1618 	    nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1619 		goto nla_put_failure;
1620 	return 0;
1621 
1622 nla_put_failure:
1623 	return -EMSGSIZE;
1624 }
1625 
1626 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1627 	[IFLA_GRE_LINK]        = { .type = NLA_U32 },
1628 	[IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1629 	[IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1630 	[IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1631 	[IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1632 	[IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1633 	[IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1634 	[IFLA_GRE_TTL]         = { .type = NLA_U8 },
1635 	[IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1636 	[IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1637 	[IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1638 };
1639 
1640 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1641 	.kind		= "ip6gre",
1642 	.maxtype	= IFLA_GRE_MAX,
1643 	.policy		= ip6gre_policy,
1644 	.priv_size	= sizeof(struct ip6_tnl),
1645 	.setup		= ip6gre_tunnel_setup,
1646 	.validate	= ip6gre_tunnel_validate,
1647 	.newlink	= ip6gre_newlink,
1648 	.changelink	= ip6gre_changelink,
1649 	.get_size	= ip6gre_get_size,
1650 	.fill_info	= ip6gre_fill_info,
1651 };
1652 
1653 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1654 	.kind		= "ip6gretap",
1655 	.maxtype	= IFLA_GRE_MAX,
1656 	.policy		= ip6gre_policy,
1657 	.priv_size	= sizeof(struct ip6_tnl),
1658 	.setup		= ip6gre_tap_setup,
1659 	.validate	= ip6gre_tap_validate,
1660 	.newlink	= ip6gre_newlink,
1661 	.changelink	= ip6gre_changelink,
1662 	.get_size	= ip6gre_get_size,
1663 	.fill_info	= ip6gre_fill_info,
1664 };
1665 
1666 /*
1667  *	And now the modules code and kernel interface.
1668  */
1669 
1670 static int __init ip6gre_init(void)
1671 {
1672 	int err;
1673 
1674 	pr_info("GRE over IPv6 tunneling driver\n");
1675 
1676 	err = register_pernet_device(&ip6gre_net_ops);
1677 	if (err < 0)
1678 		return err;
1679 
1680 	err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1681 	if (err < 0) {
1682 		pr_info("%s: can't add protocol\n", __func__);
1683 		goto add_proto_failed;
1684 	}
1685 
1686 	err = rtnl_link_register(&ip6gre_link_ops);
1687 	if (err < 0)
1688 		goto rtnl_link_failed;
1689 
1690 	err = rtnl_link_register(&ip6gre_tap_ops);
1691 	if (err < 0)
1692 		goto tap_ops_failed;
1693 
1694 out:
1695 	return err;
1696 
1697 tap_ops_failed:
1698 	rtnl_link_unregister(&ip6gre_link_ops);
1699 rtnl_link_failed:
1700 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1701 add_proto_failed:
1702 	unregister_pernet_device(&ip6gre_net_ops);
1703 	goto out;
1704 }
1705 
1706 static void __exit ip6gre_fini(void)
1707 {
1708 	rtnl_link_unregister(&ip6gre_tap_ops);
1709 	rtnl_link_unregister(&ip6gre_link_ops);
1710 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1711 	unregister_pernet_device(&ip6gre_net_ops);
1712 }
1713 
1714 module_init(ip6gre_init);
1715 module_exit(ip6gre_fini);
1716 MODULE_LICENSE("GPL");
1717 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1718 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1719 MODULE_ALIAS_RTNL_LINK("ip6gre");
1720 MODULE_ALIAS_NETDEV("ip6gre0");
1721