xref: /openbmc/linux/net/ipv6/ip6_gre.c (revision 84d517f3)
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 HASH_SIZE_SHIFT  5
65 #define HASH_SIZE (1 << HASH_SIZE_SHIFT)
66 
67 static int ip6gre_net_id __read_mostly;
68 struct ip6gre_net {
69 	struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
70 
71 	struct net_device *fb_tunnel_dev;
72 };
73 
74 static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
75 static int ip6gre_tunnel_init(struct net_device *dev);
76 static void ip6gre_tunnel_setup(struct net_device *dev);
77 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
78 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
79 
80 /* Tunnel hash table */
81 
82 /*
83    4 hash tables:
84 
85    3: (remote,local)
86    2: (remote,*)
87    1: (*,local)
88    0: (*,*)
89 
90    We require exact key match i.e. if a key is present in packet
91    it will match only tunnel with the same key; if it is not present,
92    it will match only keyless tunnel.
93 
94    All keysless packets, if not matched configured keyless tunnels
95    will match fallback tunnel.
96  */
97 
98 #define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
99 static u32 HASH_ADDR(const struct in6_addr *addr)
100 {
101 	u32 hash = ipv6_addr_hash(addr);
102 
103 	return hash_32(hash, HASH_SIZE_SHIFT);
104 }
105 
106 #define tunnels_r_l	tunnels[3]
107 #define tunnels_r	tunnels[2]
108 #define tunnels_l	tunnels[1]
109 #define tunnels_wc	tunnels[0]
110 
111 /* Given src, dst and key, find appropriate for input tunnel. */
112 
113 static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
114 		const struct in6_addr *remote, const struct in6_addr *local,
115 		__be32 key, __be16 gre_proto)
116 {
117 	struct net *net = dev_net(dev);
118 	int link = dev->ifindex;
119 	unsigned int h0 = HASH_ADDR(remote);
120 	unsigned int h1 = HASH_KEY(key);
121 	struct ip6_tnl *t, *cand = NULL;
122 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
123 	int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
124 		       ARPHRD_ETHER : ARPHRD_IP6GRE;
125 	int score, cand_score = 4;
126 
127 	for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
128 		if (!ipv6_addr_equal(local, &t->parms.laddr) ||
129 		    !ipv6_addr_equal(remote, &t->parms.raddr) ||
130 		    key != t->parms.i_key ||
131 		    !(t->dev->flags & IFF_UP))
132 			continue;
133 
134 		if (t->dev->type != ARPHRD_IP6GRE &&
135 		    t->dev->type != dev_type)
136 			continue;
137 
138 		score = 0;
139 		if (t->parms.link != link)
140 			score |= 1;
141 		if (t->dev->type != dev_type)
142 			score |= 2;
143 		if (score == 0)
144 			return t;
145 
146 		if (score < cand_score) {
147 			cand = t;
148 			cand_score = score;
149 		}
150 	}
151 
152 	for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
153 		if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
154 		    key != t->parms.i_key ||
155 		    !(t->dev->flags & IFF_UP))
156 			continue;
157 
158 		if (t->dev->type != ARPHRD_IP6GRE &&
159 		    t->dev->type != dev_type)
160 			continue;
161 
162 		score = 0;
163 		if (t->parms.link != link)
164 			score |= 1;
165 		if (t->dev->type != dev_type)
166 			score |= 2;
167 		if (score == 0)
168 			return t;
169 
170 		if (score < cand_score) {
171 			cand = t;
172 			cand_score = score;
173 		}
174 	}
175 
176 	for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
177 		if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
178 			  (!ipv6_addr_equal(local, &t->parms.raddr) ||
179 				 !ipv6_addr_is_multicast(local))) ||
180 		    key != t->parms.i_key ||
181 		    !(t->dev->flags & IFF_UP))
182 			continue;
183 
184 		if (t->dev->type != ARPHRD_IP6GRE &&
185 		    t->dev->type != dev_type)
186 			continue;
187 
188 		score = 0;
189 		if (t->parms.link != link)
190 			score |= 1;
191 		if (t->dev->type != dev_type)
192 			score |= 2;
193 		if (score == 0)
194 			return t;
195 
196 		if (score < cand_score) {
197 			cand = t;
198 			cand_score = score;
199 		}
200 	}
201 
202 	for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
203 		if (t->parms.i_key != key ||
204 		    !(t->dev->flags & IFF_UP))
205 			continue;
206 
207 		if (t->dev->type != ARPHRD_IP6GRE &&
208 		    t->dev->type != dev_type)
209 			continue;
210 
211 		score = 0;
212 		if (t->parms.link != link)
213 			score |= 1;
214 		if (t->dev->type != dev_type)
215 			score |= 2;
216 		if (score == 0)
217 			return t;
218 
219 		if (score < cand_score) {
220 			cand = t;
221 			cand_score = score;
222 		}
223 	}
224 
225 	if (cand != NULL)
226 		return cand;
227 
228 	dev = ign->fb_tunnel_dev;
229 	if (dev->flags & IFF_UP)
230 		return netdev_priv(dev);
231 
232 	return NULL;
233 }
234 
235 static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
236 		const struct __ip6_tnl_parm *p)
237 {
238 	const struct in6_addr *remote = &p->raddr;
239 	const struct in6_addr *local = &p->laddr;
240 	unsigned int h = HASH_KEY(p->i_key);
241 	int prio = 0;
242 
243 	if (!ipv6_addr_any(local))
244 		prio |= 1;
245 	if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
246 		prio |= 2;
247 		h ^= HASH_ADDR(remote);
248 	}
249 
250 	return &ign->tunnels[prio][h];
251 }
252 
253 static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
254 		const struct ip6_tnl *t)
255 {
256 	return __ip6gre_bucket(ign, &t->parms);
257 }
258 
259 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
260 {
261 	struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
262 
263 	rcu_assign_pointer(t->next, rtnl_dereference(*tp));
264 	rcu_assign_pointer(*tp, t);
265 }
266 
267 static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
268 {
269 	struct ip6_tnl __rcu **tp;
270 	struct ip6_tnl *iter;
271 
272 	for (tp = ip6gre_bucket(ign, t);
273 	     (iter = rtnl_dereference(*tp)) != NULL;
274 	     tp = &iter->next) {
275 		if (t == iter) {
276 			rcu_assign_pointer(*tp, t->next);
277 			break;
278 		}
279 	}
280 }
281 
282 static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
283 					   const struct __ip6_tnl_parm *parms,
284 					   int type)
285 {
286 	const struct in6_addr *remote = &parms->raddr;
287 	const struct in6_addr *local = &parms->laddr;
288 	__be32 key = parms->i_key;
289 	int link = parms->link;
290 	struct ip6_tnl *t;
291 	struct ip6_tnl __rcu **tp;
292 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
293 
294 	for (tp = __ip6gre_bucket(ign, parms);
295 	     (t = rtnl_dereference(*tp)) != NULL;
296 	     tp = &t->next)
297 		if (ipv6_addr_equal(local, &t->parms.laddr) &&
298 		    ipv6_addr_equal(remote, &t->parms.raddr) &&
299 		    key == t->parms.i_key &&
300 		    link == t->parms.link &&
301 		    type == t->dev->type)
302 			break;
303 
304 	return t;
305 }
306 
307 static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
308 		const struct __ip6_tnl_parm *parms, int create)
309 {
310 	struct ip6_tnl *t, *nt;
311 	struct net_device *dev;
312 	char name[IFNAMSIZ];
313 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
314 
315 	t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
316 	if (t || !create)
317 		return t;
318 
319 	if (parms->name[0])
320 		strlcpy(name, parms->name, IFNAMSIZ);
321 	else
322 		strcpy(name, "ip6gre%d");
323 
324 	dev = alloc_netdev(sizeof(*t), name, ip6gre_tunnel_setup);
325 	if (!dev)
326 		return NULL;
327 
328 	dev_net_set(dev, net);
329 
330 	nt = netdev_priv(dev);
331 	nt->parms = *parms;
332 	dev->rtnl_link_ops = &ip6gre_link_ops;
333 
334 	nt->dev = dev;
335 	nt->net = dev_net(dev);
336 	ip6gre_tnl_link_config(nt, 1);
337 
338 	if (register_netdevice(dev) < 0)
339 		goto failed_free;
340 
341 	/* Can use a lockless transmit, unless we generate output sequences */
342 	if (!(nt->parms.o_flags & GRE_SEQ))
343 		dev->features |= NETIF_F_LLTX;
344 
345 	dev_hold(dev);
346 	ip6gre_tunnel_link(ign, nt);
347 	return nt;
348 
349 failed_free:
350 	free_netdev(dev);
351 	return NULL;
352 }
353 
354 static void ip6gre_tunnel_uninit(struct net_device *dev)
355 {
356 	struct net *net = dev_net(dev);
357 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
358 
359 	ip6gre_tunnel_unlink(ign, netdev_priv(dev));
360 	dev_put(dev);
361 }
362 
363 
364 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
365 		u8 type, u8 code, int offset, __be32 info)
366 {
367 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
368 	__be16 *p = (__be16 *)(skb->data + offset);
369 	int grehlen = offset + 4;
370 	struct ip6_tnl *t;
371 	__be16 flags;
372 
373 	flags = p[0];
374 	if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
375 		if (flags&(GRE_VERSION|GRE_ROUTING))
376 			return;
377 		if (flags&GRE_KEY) {
378 			grehlen += 4;
379 			if (flags&GRE_CSUM)
380 				grehlen += 4;
381 		}
382 	}
383 
384 	/* If only 8 bytes returned, keyed message will be dropped here */
385 	if (!pskb_may_pull(skb, grehlen))
386 		return;
387 	ipv6h = (const struct ipv6hdr *)skb->data;
388 	p = (__be16 *)(skb->data + offset);
389 
390 	t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
391 				flags & GRE_KEY ?
392 				*(((__be32 *)p) + (grehlen / 4) - 1) : 0,
393 				p[1]);
394 	if (t == NULL)
395 		return;
396 
397 	switch (type) {
398 		__u32 teli;
399 		struct ipv6_tlv_tnl_enc_lim *tel;
400 		__u32 mtu;
401 	case ICMPV6_DEST_UNREACH:
402 		net_warn_ratelimited("%s: Path to destination invalid or inactive!\n",
403 				     t->parms.name);
404 		break;
405 	case ICMPV6_TIME_EXCEED:
406 		if (code == ICMPV6_EXC_HOPLIMIT) {
407 			net_warn_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
408 					     t->parms.name);
409 		}
410 		break;
411 	case ICMPV6_PARAMPROB:
412 		teli = 0;
413 		if (code == ICMPV6_HDR_FIELD)
414 			teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
415 
416 		if (teli && teli == info - 2) {
417 			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
418 			if (tel->encap_limit == 0) {
419 				net_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
420 						     t->parms.name);
421 			}
422 		} else {
423 			net_warn_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
424 					     t->parms.name);
425 		}
426 		break;
427 	case ICMPV6_PKT_TOOBIG:
428 		mtu = info - offset;
429 		if (mtu < IPV6_MIN_MTU)
430 			mtu = IPV6_MIN_MTU;
431 		t->dev->mtu = mtu;
432 		break;
433 	}
434 
435 	if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
436 		t->err_count++;
437 	else
438 		t->err_count = 1;
439 	t->err_time = jiffies;
440 }
441 
442 static int ip6gre_rcv(struct sk_buff *skb)
443 {
444 	const struct ipv6hdr *ipv6h;
445 	u8     *h;
446 	__be16    flags;
447 	__sum16   csum = 0;
448 	__be32 key = 0;
449 	u32    seqno = 0;
450 	struct ip6_tnl *tunnel;
451 	int    offset = 4;
452 	__be16 gre_proto;
453 	int err;
454 
455 	if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
456 		goto drop;
457 
458 	ipv6h = ipv6_hdr(skb);
459 	h = skb->data;
460 	flags = *(__be16 *)h;
461 
462 	if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
463 		/* - Version must be 0.
464 		   - We do not support routing headers.
465 		 */
466 		if (flags&(GRE_VERSION|GRE_ROUTING))
467 			goto drop;
468 
469 		if (flags&GRE_CSUM) {
470 			switch (skb->ip_summed) {
471 			case CHECKSUM_COMPLETE:
472 				csum = csum_fold(skb->csum);
473 				if (!csum)
474 					break;
475 				/* fall through */
476 			case CHECKSUM_NONE:
477 				skb->csum = 0;
478 				csum = __skb_checksum_complete(skb);
479 				skb->ip_summed = CHECKSUM_COMPLETE;
480 			}
481 			offset += 4;
482 		}
483 		if (flags&GRE_KEY) {
484 			key = *(__be32 *)(h + offset);
485 			offset += 4;
486 		}
487 		if (flags&GRE_SEQ) {
488 			seqno = ntohl(*(__be32 *)(h + offset));
489 			offset += 4;
490 		}
491 	}
492 
493 	gre_proto = *(__be16 *)(h + 2);
494 
495 	tunnel = ip6gre_tunnel_lookup(skb->dev,
496 					  &ipv6h->saddr, &ipv6h->daddr, key,
497 					  gre_proto);
498 	if (tunnel) {
499 		struct pcpu_sw_netstats *tstats;
500 
501 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
502 			goto drop;
503 
504 		if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
505 			tunnel->dev->stats.rx_dropped++;
506 			goto drop;
507 		}
508 
509 		skb->protocol = gre_proto;
510 		/* WCCP version 1 and 2 protocol decoding.
511 		 * - Change protocol to IP
512 		 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
513 		 */
514 		if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
515 			skb->protocol = htons(ETH_P_IP);
516 			if ((*(h + offset) & 0xF0) != 0x40)
517 				offset += 4;
518 		}
519 
520 		skb->mac_header = skb->network_header;
521 		__pskb_pull(skb, offset);
522 		skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
523 
524 		if (((flags&GRE_CSUM) && csum) ||
525 		    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
526 			tunnel->dev->stats.rx_crc_errors++;
527 			tunnel->dev->stats.rx_errors++;
528 			goto drop;
529 		}
530 		if (tunnel->parms.i_flags&GRE_SEQ) {
531 			if (!(flags&GRE_SEQ) ||
532 			    (tunnel->i_seqno &&
533 					(s32)(seqno - tunnel->i_seqno) < 0)) {
534 				tunnel->dev->stats.rx_fifo_errors++;
535 				tunnel->dev->stats.rx_errors++;
536 				goto drop;
537 			}
538 			tunnel->i_seqno = seqno + 1;
539 		}
540 
541 		/* Warning: All skb pointers will be invalidated! */
542 		if (tunnel->dev->type == ARPHRD_ETHER) {
543 			if (!pskb_may_pull(skb, ETH_HLEN)) {
544 				tunnel->dev->stats.rx_length_errors++;
545 				tunnel->dev->stats.rx_errors++;
546 				goto drop;
547 			}
548 
549 			ipv6h = ipv6_hdr(skb);
550 			skb->protocol = eth_type_trans(skb, tunnel->dev);
551 			skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
552 		}
553 
554 		__skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
555 
556 		skb_reset_network_header(skb);
557 
558 		err = IP6_ECN_decapsulate(ipv6h, skb);
559 		if (unlikely(err)) {
560 			if (log_ecn_error)
561 				net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
562 						     &ipv6h->saddr,
563 						     ipv6_get_dsfield(ipv6h));
564 			if (err > 1) {
565 				++tunnel->dev->stats.rx_frame_errors;
566 				++tunnel->dev->stats.rx_errors;
567 				goto drop;
568 			}
569 		}
570 
571 		tstats = this_cpu_ptr(tunnel->dev->tstats);
572 		u64_stats_update_begin(&tstats->syncp);
573 		tstats->rx_packets++;
574 		tstats->rx_bytes += skb->len;
575 		u64_stats_update_end(&tstats->syncp);
576 
577 		netif_rx(skb);
578 
579 		return 0;
580 	}
581 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
582 
583 drop:
584 	kfree_skb(skb);
585 	return 0;
586 }
587 
588 struct ipv6_tel_txoption {
589 	struct ipv6_txoptions ops;
590 	__u8 dst_opt[8];
591 };
592 
593 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
594 {
595 	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
596 
597 	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
598 	opt->dst_opt[3] = 1;
599 	opt->dst_opt[4] = encap_limit;
600 	opt->dst_opt[5] = IPV6_TLV_PADN;
601 	opt->dst_opt[6] = 1;
602 
603 	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
604 	opt->ops.opt_nflen = 8;
605 }
606 
607 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
608 			 struct net_device *dev,
609 			 __u8 dsfield,
610 			 struct flowi6 *fl6,
611 			 int encap_limit,
612 			 __u32 *pmtu)
613 {
614 	struct net *net = dev_net(dev);
615 	struct ip6_tnl *tunnel = netdev_priv(dev);
616 	struct net_device *tdev;    /* Device to other host */
617 	struct ipv6hdr  *ipv6h;     /* Our new IP header */
618 	unsigned int max_headroom = 0; /* The extra header space needed */
619 	int    gre_hlen;
620 	struct ipv6_tel_txoption opt;
621 	int    mtu;
622 	struct dst_entry *dst = NULL, *ndst = NULL;
623 	struct net_device_stats *stats = &tunnel->dev->stats;
624 	int err = -1;
625 	u8 proto;
626 	struct sk_buff *new_skb;
627 
628 	if (dev->type == ARPHRD_ETHER)
629 		IPCB(skb)->flags = 0;
630 
631 	if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
632 		gre_hlen = 0;
633 		ipv6h = (struct ipv6hdr *)skb->data;
634 		fl6->daddr = ipv6h->daddr;
635 	} else {
636 		gre_hlen = tunnel->hlen;
637 		fl6->daddr = tunnel->parms.raddr;
638 	}
639 
640 	if (!fl6->flowi6_mark)
641 		dst = ip6_tnl_dst_check(tunnel);
642 
643 	if (!dst) {
644 		ndst = ip6_route_output(net, NULL, fl6);
645 
646 		if (ndst->error)
647 			goto tx_err_link_failure;
648 		ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
649 		if (IS_ERR(ndst)) {
650 			err = PTR_ERR(ndst);
651 			ndst = NULL;
652 			goto tx_err_link_failure;
653 		}
654 		dst = ndst;
655 	}
656 
657 	tdev = dst->dev;
658 
659 	if (tdev == dev) {
660 		stats->collisions++;
661 		net_warn_ratelimited("%s: Local routing loop detected!\n",
662 				     tunnel->parms.name);
663 		goto tx_err_dst_release;
664 	}
665 
666 	mtu = dst_mtu(dst) - sizeof(*ipv6h);
667 	if (encap_limit >= 0) {
668 		max_headroom += 8;
669 		mtu -= 8;
670 	}
671 	if (mtu < IPV6_MIN_MTU)
672 		mtu = IPV6_MIN_MTU;
673 	if (skb_dst(skb))
674 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
675 	if (skb->len > mtu) {
676 		*pmtu = mtu;
677 		err = -EMSGSIZE;
678 		goto tx_err_dst_release;
679 	}
680 
681 	if (tunnel->err_count > 0) {
682 		if (time_before(jiffies,
683 				tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
684 			tunnel->err_count--;
685 
686 			dst_link_failure(skb);
687 		} else
688 			tunnel->err_count = 0;
689 	}
690 
691 	skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
692 
693 	max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
694 
695 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
696 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
697 		new_skb = skb_realloc_headroom(skb, max_headroom);
698 		if (max_headroom > dev->needed_headroom)
699 			dev->needed_headroom = max_headroom;
700 		if (!new_skb)
701 			goto tx_err_dst_release;
702 
703 		if (skb->sk)
704 			skb_set_owner_w(new_skb, skb->sk);
705 		consume_skb(skb);
706 		skb = new_skb;
707 	}
708 
709 	if (fl6->flowi6_mark) {
710 		skb_dst_set(skb, dst);
711 		ndst = NULL;
712 	} else {
713 		skb_dst_set_noref(skb, dst);
714 	}
715 
716 	proto = NEXTHDR_GRE;
717 	if (encap_limit >= 0) {
718 		init_tel_txopt(&opt, encap_limit);
719 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
720 	}
721 
722 	if (likely(!skb->encapsulation)) {
723 		skb_reset_inner_headers(skb);
724 		skb->encapsulation = 1;
725 	}
726 
727 	skb_push(skb, gre_hlen);
728 	skb_reset_network_header(skb);
729 	skb_set_transport_header(skb, sizeof(*ipv6h));
730 
731 	/*
732 	 *	Push down and install the IP header.
733 	 */
734 	ipv6h = ipv6_hdr(skb);
735 	ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);
736 	ipv6h->hop_limit = tunnel->parms.hop_limit;
737 	ipv6h->nexthdr = proto;
738 	ipv6h->saddr = fl6->saddr;
739 	ipv6h->daddr = fl6->daddr;
740 
741 	((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
742 	((__be16 *)(ipv6h + 1))[1] = (dev->type == ARPHRD_ETHER) ?
743 				   htons(ETH_P_TEB) : skb->protocol;
744 
745 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
746 		__be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
747 
748 		if (tunnel->parms.o_flags&GRE_SEQ) {
749 			++tunnel->o_seqno;
750 			*ptr = htonl(tunnel->o_seqno);
751 			ptr--;
752 		}
753 		if (tunnel->parms.o_flags&GRE_KEY) {
754 			*ptr = tunnel->parms.o_key;
755 			ptr--;
756 		}
757 		if (tunnel->parms.o_flags&GRE_CSUM) {
758 			*ptr = 0;
759 			*(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
760 				skb->len - sizeof(struct ipv6hdr));
761 		}
762 	}
763 
764 	ip6tunnel_xmit(skb, dev);
765 	if (ndst)
766 		ip6_tnl_dst_store(tunnel, ndst);
767 	return 0;
768 tx_err_link_failure:
769 	stats->tx_carrier_errors++;
770 	dst_link_failure(skb);
771 tx_err_dst_release:
772 	dst_release(ndst);
773 	return err;
774 }
775 
776 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
777 {
778 	struct ip6_tnl *t = netdev_priv(dev);
779 	const struct iphdr  *iph = ip_hdr(skb);
780 	int encap_limit = -1;
781 	struct flowi6 fl6;
782 	__u8 dsfield;
783 	__u32 mtu;
784 	int err;
785 
786 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
787 		encap_limit = t->parms.encap_limit;
788 
789 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
790 	fl6.flowi6_proto = IPPROTO_IPIP;
791 
792 	dsfield = ipv4_get_dsfield(iph);
793 
794 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
795 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
796 					  & IPV6_TCLASS_MASK;
797 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
798 		fl6.flowi6_mark = skb->mark;
799 
800 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
801 	if (err != 0) {
802 		/* XXX: send ICMP error even if DF is not set. */
803 		if (err == -EMSGSIZE)
804 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
805 				  htonl(mtu));
806 		return -1;
807 	}
808 
809 	return 0;
810 }
811 
812 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
813 {
814 	struct ip6_tnl *t = netdev_priv(dev);
815 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
816 	int encap_limit = -1;
817 	__u16 offset;
818 	struct flowi6 fl6;
819 	__u8 dsfield;
820 	__u32 mtu;
821 	int err;
822 
823 	if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
824 		return -1;
825 
826 	offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
827 	if (offset > 0) {
828 		struct ipv6_tlv_tnl_enc_lim *tel;
829 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
830 		if (tel->encap_limit == 0) {
831 			icmpv6_send(skb, ICMPV6_PARAMPROB,
832 				    ICMPV6_HDR_FIELD, offset + 2);
833 			return -1;
834 		}
835 		encap_limit = tel->encap_limit - 1;
836 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
837 		encap_limit = t->parms.encap_limit;
838 
839 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
840 	fl6.flowi6_proto = IPPROTO_IPV6;
841 
842 	dsfield = ipv6_get_dsfield(ipv6h);
843 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
844 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
845 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
846 		fl6.flowlabel |= ip6_flowlabel(ipv6h);
847 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
848 		fl6.flowi6_mark = skb->mark;
849 
850 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
851 	if (err != 0) {
852 		if (err == -EMSGSIZE)
853 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
854 		return -1;
855 	}
856 
857 	return 0;
858 }
859 
860 /**
861  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
862  *   @t: the outgoing tunnel device
863  *   @hdr: IPv6 header from the incoming packet
864  *
865  * Description:
866  *   Avoid trivial tunneling loop by checking that tunnel exit-point
867  *   doesn't match source of incoming packet.
868  *
869  * Return:
870  *   1 if conflict,
871  *   0 else
872  **/
873 
874 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
875 	const struct ipv6hdr *hdr)
876 {
877 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
878 }
879 
880 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
881 {
882 	struct ip6_tnl *t = netdev_priv(dev);
883 	int encap_limit = -1;
884 	struct flowi6 fl6;
885 	__u32 mtu;
886 	int err;
887 
888 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
889 		encap_limit = t->parms.encap_limit;
890 
891 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
892 	fl6.flowi6_proto = skb->protocol;
893 
894 	err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
895 
896 	return err;
897 }
898 
899 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
900 	struct net_device *dev)
901 {
902 	struct ip6_tnl *t = netdev_priv(dev);
903 	struct net_device_stats *stats = &t->dev->stats;
904 	int ret;
905 
906 	if (!ip6_tnl_xmit_ctl(t))
907 		goto tx_err;
908 
909 	switch (skb->protocol) {
910 	case htons(ETH_P_IP):
911 		ret = ip6gre_xmit_ipv4(skb, dev);
912 		break;
913 	case htons(ETH_P_IPV6):
914 		ret = ip6gre_xmit_ipv6(skb, dev);
915 		break;
916 	default:
917 		ret = ip6gre_xmit_other(skb, dev);
918 		break;
919 	}
920 
921 	if (ret < 0)
922 		goto tx_err;
923 
924 	return NETDEV_TX_OK;
925 
926 tx_err:
927 	stats->tx_errors++;
928 	stats->tx_dropped++;
929 	kfree_skb(skb);
930 	return NETDEV_TX_OK;
931 }
932 
933 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
934 {
935 	struct net_device *dev = t->dev;
936 	struct __ip6_tnl_parm *p = &t->parms;
937 	struct flowi6 *fl6 = &t->fl.u.ip6;
938 	int addend = sizeof(struct ipv6hdr) + 4;
939 
940 	if (dev->type != ARPHRD_ETHER) {
941 		memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
942 		memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
943 	}
944 
945 	/* Set up flowi template */
946 	fl6->saddr = p->laddr;
947 	fl6->daddr = p->raddr;
948 	fl6->flowi6_oif = p->link;
949 	fl6->flowlabel = 0;
950 
951 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
952 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
953 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
954 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
955 
956 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
957 	p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
958 
959 	if (p->flags&IP6_TNL_F_CAP_XMIT &&
960 			p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
961 		dev->flags |= IFF_POINTOPOINT;
962 	else
963 		dev->flags &= ~IFF_POINTOPOINT;
964 
965 	dev->iflink = p->link;
966 
967 	/* Precalculate GRE options length */
968 	if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
969 		if (t->parms.o_flags&GRE_CSUM)
970 			addend += 4;
971 		if (t->parms.o_flags&GRE_KEY)
972 			addend += 4;
973 		if (t->parms.o_flags&GRE_SEQ)
974 			addend += 4;
975 	}
976 	t->hlen = addend;
977 
978 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
979 		int strict = (ipv6_addr_type(&p->raddr) &
980 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
981 
982 		struct rt6_info *rt = rt6_lookup(dev_net(dev),
983 						 &p->raddr, &p->laddr,
984 						 p->link, strict);
985 
986 		if (rt == NULL)
987 			return;
988 
989 		if (rt->dst.dev) {
990 			dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
991 
992 			if (set_mtu) {
993 				dev->mtu = rt->dst.dev->mtu - addend;
994 				if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
995 					dev->mtu -= 8;
996 
997 				if (dev->mtu < IPV6_MIN_MTU)
998 					dev->mtu = IPV6_MIN_MTU;
999 			}
1000 		}
1001 		ip6_rt_put(rt);
1002 	}
1003 }
1004 
1005 static int ip6gre_tnl_change(struct ip6_tnl *t,
1006 	const struct __ip6_tnl_parm *p, int set_mtu)
1007 {
1008 	t->parms.laddr = p->laddr;
1009 	t->parms.raddr = p->raddr;
1010 	t->parms.flags = p->flags;
1011 	t->parms.hop_limit = p->hop_limit;
1012 	t->parms.encap_limit = p->encap_limit;
1013 	t->parms.flowinfo = p->flowinfo;
1014 	t->parms.link = p->link;
1015 	t->parms.proto = p->proto;
1016 	t->parms.i_key = p->i_key;
1017 	t->parms.o_key = p->o_key;
1018 	t->parms.i_flags = p->i_flags;
1019 	t->parms.o_flags = p->o_flags;
1020 	ip6_tnl_dst_reset(t);
1021 	ip6gre_tnl_link_config(t, set_mtu);
1022 	return 0;
1023 }
1024 
1025 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1026 	const struct ip6_tnl_parm2 *u)
1027 {
1028 	p->laddr = u->laddr;
1029 	p->raddr = u->raddr;
1030 	p->flags = u->flags;
1031 	p->hop_limit = u->hop_limit;
1032 	p->encap_limit = u->encap_limit;
1033 	p->flowinfo = u->flowinfo;
1034 	p->link = u->link;
1035 	p->i_key = u->i_key;
1036 	p->o_key = u->o_key;
1037 	p->i_flags = u->i_flags;
1038 	p->o_flags = u->o_flags;
1039 	memcpy(p->name, u->name, sizeof(u->name));
1040 }
1041 
1042 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1043 	const struct __ip6_tnl_parm *p)
1044 {
1045 	u->proto = IPPROTO_GRE;
1046 	u->laddr = p->laddr;
1047 	u->raddr = p->raddr;
1048 	u->flags = p->flags;
1049 	u->hop_limit = p->hop_limit;
1050 	u->encap_limit = p->encap_limit;
1051 	u->flowinfo = p->flowinfo;
1052 	u->link = p->link;
1053 	u->i_key = p->i_key;
1054 	u->o_key = p->o_key;
1055 	u->i_flags = p->i_flags;
1056 	u->o_flags = p->o_flags;
1057 	memcpy(u->name, p->name, sizeof(u->name));
1058 }
1059 
1060 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1061 	struct ifreq *ifr, int cmd)
1062 {
1063 	int err = 0;
1064 	struct ip6_tnl_parm2 p;
1065 	struct __ip6_tnl_parm p1;
1066 	struct ip6_tnl *t;
1067 	struct net *net = dev_net(dev);
1068 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1069 
1070 	switch (cmd) {
1071 	case SIOCGETTUNNEL:
1072 		t = NULL;
1073 		if (dev == ign->fb_tunnel_dev) {
1074 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1075 				err = -EFAULT;
1076 				break;
1077 			}
1078 			ip6gre_tnl_parm_from_user(&p1, &p);
1079 			t = ip6gre_tunnel_locate(net, &p1, 0);
1080 		}
1081 		if (t == NULL)
1082 			t = netdev_priv(dev);
1083 		memset(&p, 0, sizeof(p));
1084 		ip6gre_tnl_parm_to_user(&p, &t->parms);
1085 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1086 			err = -EFAULT;
1087 		break;
1088 
1089 	case SIOCADDTUNNEL:
1090 	case SIOCCHGTUNNEL:
1091 		err = -EPERM;
1092 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1093 			goto done;
1094 
1095 		err = -EFAULT;
1096 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1097 			goto done;
1098 
1099 		err = -EINVAL;
1100 		if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1101 			goto done;
1102 
1103 		if (!(p.i_flags&GRE_KEY))
1104 			p.i_key = 0;
1105 		if (!(p.o_flags&GRE_KEY))
1106 			p.o_key = 0;
1107 
1108 		ip6gre_tnl_parm_from_user(&p1, &p);
1109 		t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1110 
1111 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1112 			if (t != NULL) {
1113 				if (t->dev != dev) {
1114 					err = -EEXIST;
1115 					break;
1116 				}
1117 			} else {
1118 				t = netdev_priv(dev);
1119 
1120 				ip6gre_tunnel_unlink(ign, t);
1121 				synchronize_net();
1122 				ip6gre_tnl_change(t, &p1, 1);
1123 				ip6gre_tunnel_link(ign, t);
1124 				netdev_state_change(dev);
1125 			}
1126 		}
1127 
1128 		if (t) {
1129 			err = 0;
1130 
1131 			memset(&p, 0, sizeof(p));
1132 			ip6gre_tnl_parm_to_user(&p, &t->parms);
1133 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1134 				err = -EFAULT;
1135 		} else
1136 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1137 		break;
1138 
1139 	case SIOCDELTUNNEL:
1140 		err = -EPERM;
1141 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1142 			goto done;
1143 
1144 		if (dev == ign->fb_tunnel_dev) {
1145 			err = -EFAULT;
1146 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1147 				goto done;
1148 			err = -ENOENT;
1149 			ip6gre_tnl_parm_from_user(&p1, &p);
1150 			t = ip6gre_tunnel_locate(net, &p1, 0);
1151 			if (t == NULL)
1152 				goto done;
1153 			err = -EPERM;
1154 			if (t == netdev_priv(ign->fb_tunnel_dev))
1155 				goto done;
1156 			dev = t->dev;
1157 		}
1158 		unregister_netdevice(dev);
1159 		err = 0;
1160 		break;
1161 
1162 	default:
1163 		err = -EINVAL;
1164 	}
1165 
1166 done:
1167 	return err;
1168 }
1169 
1170 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1171 {
1172 	if (new_mtu < 68 ||
1173 	    new_mtu > 0xFFF8 - dev->hard_header_len)
1174 		return -EINVAL;
1175 	dev->mtu = new_mtu;
1176 	return 0;
1177 }
1178 
1179 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1180 			unsigned short type,
1181 			const void *daddr, const void *saddr, unsigned int len)
1182 {
1183 	struct ip6_tnl *t = netdev_priv(dev);
1184 	struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1185 	__be16 *p = (__be16 *)(ipv6h+1);
1186 
1187 	ip6_flow_hdr(ipv6h, 0, t->fl.u.ip6.flowlabel);
1188 	ipv6h->hop_limit = t->parms.hop_limit;
1189 	ipv6h->nexthdr = NEXTHDR_GRE;
1190 	ipv6h->saddr = t->parms.laddr;
1191 	ipv6h->daddr = t->parms.raddr;
1192 
1193 	p[0]		= t->parms.o_flags;
1194 	p[1]		= htons(type);
1195 
1196 	/*
1197 	 *	Set the source hardware address.
1198 	 */
1199 
1200 	if (saddr)
1201 		memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1202 	if (daddr)
1203 		memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1204 	if (!ipv6_addr_any(&ipv6h->daddr))
1205 		return t->hlen;
1206 
1207 	return -t->hlen;
1208 }
1209 
1210 static const struct header_ops ip6gre_header_ops = {
1211 	.create	= ip6gre_header,
1212 };
1213 
1214 static const struct net_device_ops ip6gre_netdev_ops = {
1215 	.ndo_init		= ip6gre_tunnel_init,
1216 	.ndo_uninit		= ip6gre_tunnel_uninit,
1217 	.ndo_start_xmit		= ip6gre_tunnel_xmit,
1218 	.ndo_do_ioctl		= ip6gre_tunnel_ioctl,
1219 	.ndo_change_mtu		= ip6gre_tunnel_change_mtu,
1220 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1221 };
1222 
1223 static void ip6gre_dev_free(struct net_device *dev)
1224 {
1225 	free_percpu(dev->tstats);
1226 	free_netdev(dev);
1227 }
1228 
1229 static void ip6gre_tunnel_setup(struct net_device *dev)
1230 {
1231 	struct ip6_tnl *t;
1232 
1233 	dev->netdev_ops = &ip6gre_netdev_ops;
1234 	dev->destructor = ip6gre_dev_free;
1235 
1236 	dev->type = ARPHRD_IP6GRE;
1237 	dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1238 	dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1239 	t = netdev_priv(dev);
1240 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1241 		dev->mtu -= 8;
1242 	dev->flags |= IFF_NOARP;
1243 	dev->iflink = 0;
1244 	dev->addr_len = sizeof(struct in6_addr);
1245 	dev->features |= NETIF_F_NETNS_LOCAL;
1246 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1247 }
1248 
1249 static int ip6gre_tunnel_init(struct net_device *dev)
1250 {
1251 	struct ip6_tnl *tunnel;
1252 	int i;
1253 
1254 	tunnel = netdev_priv(dev);
1255 
1256 	tunnel->dev = dev;
1257 	tunnel->net = dev_net(dev);
1258 	strcpy(tunnel->parms.name, dev->name);
1259 
1260 	memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1261 	memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1262 
1263 	if (ipv6_addr_any(&tunnel->parms.raddr))
1264 		dev->header_ops = &ip6gre_header_ops;
1265 
1266 	dev->tstats = alloc_percpu(struct pcpu_sw_netstats);
1267 	if (!dev->tstats)
1268 		return -ENOMEM;
1269 
1270 	for_each_possible_cpu(i) {
1271 		struct pcpu_sw_netstats *ip6gre_tunnel_stats;
1272 		ip6gre_tunnel_stats = per_cpu_ptr(dev->tstats, i);
1273 		u64_stats_init(&ip6gre_tunnel_stats->syncp);
1274 	}
1275 
1276 
1277 	return 0;
1278 }
1279 
1280 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1281 {
1282 	struct ip6_tnl *tunnel = netdev_priv(dev);
1283 
1284 	tunnel->dev = dev;
1285 	tunnel->net = dev_net(dev);
1286 	strcpy(tunnel->parms.name, dev->name);
1287 
1288 	tunnel->hlen		= sizeof(struct ipv6hdr) + 4;
1289 
1290 	dev_hold(dev);
1291 }
1292 
1293 
1294 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1295 	.handler     = ip6gre_rcv,
1296 	.err_handler = ip6gre_err,
1297 	.flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1298 };
1299 
1300 static void ip6gre_destroy_tunnels(struct ip6gre_net *ign,
1301 	struct list_head *head)
1302 {
1303 	int prio;
1304 
1305 	for (prio = 0; prio < 4; prio++) {
1306 		int h;
1307 		for (h = 0; h < HASH_SIZE; h++) {
1308 			struct ip6_tnl *t;
1309 
1310 			t = rtnl_dereference(ign->tunnels[prio][h]);
1311 
1312 			while (t != NULL) {
1313 				unregister_netdevice_queue(t->dev, head);
1314 				t = rtnl_dereference(t->next);
1315 			}
1316 		}
1317 	}
1318 }
1319 
1320 static int __net_init ip6gre_init_net(struct net *net)
1321 {
1322 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1323 	int err;
1324 
1325 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1326 					   ip6gre_tunnel_setup);
1327 	if (!ign->fb_tunnel_dev) {
1328 		err = -ENOMEM;
1329 		goto err_alloc_dev;
1330 	}
1331 	dev_net_set(ign->fb_tunnel_dev, net);
1332 
1333 	ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1334 	ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1335 
1336 	err = register_netdev(ign->fb_tunnel_dev);
1337 	if (err)
1338 		goto err_reg_dev;
1339 
1340 	rcu_assign_pointer(ign->tunnels_wc[0],
1341 			   netdev_priv(ign->fb_tunnel_dev));
1342 	return 0;
1343 
1344 err_reg_dev:
1345 	ip6gre_dev_free(ign->fb_tunnel_dev);
1346 err_alloc_dev:
1347 	return err;
1348 }
1349 
1350 static void __net_exit ip6gre_exit_net(struct net *net)
1351 {
1352 	struct ip6gre_net *ign;
1353 	LIST_HEAD(list);
1354 
1355 	ign = net_generic(net, ip6gre_net_id);
1356 	rtnl_lock();
1357 	ip6gre_destroy_tunnels(ign, &list);
1358 	unregister_netdevice_many(&list);
1359 	rtnl_unlock();
1360 }
1361 
1362 static struct pernet_operations ip6gre_net_ops = {
1363 	.init = ip6gre_init_net,
1364 	.exit = ip6gre_exit_net,
1365 	.id   = &ip6gre_net_id,
1366 	.size = sizeof(struct ip6gre_net),
1367 };
1368 
1369 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1370 {
1371 	__be16 flags;
1372 
1373 	if (!data)
1374 		return 0;
1375 
1376 	flags = 0;
1377 	if (data[IFLA_GRE_IFLAGS])
1378 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1379 	if (data[IFLA_GRE_OFLAGS])
1380 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1381 	if (flags & (GRE_VERSION|GRE_ROUTING))
1382 		return -EINVAL;
1383 
1384 	return 0;
1385 }
1386 
1387 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1388 {
1389 	struct in6_addr daddr;
1390 
1391 	if (tb[IFLA_ADDRESS]) {
1392 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1393 			return -EINVAL;
1394 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1395 			return -EADDRNOTAVAIL;
1396 	}
1397 
1398 	if (!data)
1399 		goto out;
1400 
1401 	if (data[IFLA_GRE_REMOTE]) {
1402 		nla_memcpy(&daddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1403 		if (ipv6_addr_any(&daddr))
1404 			return -EINVAL;
1405 	}
1406 
1407 out:
1408 	return ip6gre_tunnel_validate(tb, data);
1409 }
1410 
1411 
1412 static void ip6gre_netlink_parms(struct nlattr *data[],
1413 				struct __ip6_tnl_parm *parms)
1414 {
1415 	memset(parms, 0, sizeof(*parms));
1416 
1417 	if (!data)
1418 		return;
1419 
1420 	if (data[IFLA_GRE_LINK])
1421 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1422 
1423 	if (data[IFLA_GRE_IFLAGS])
1424 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1425 
1426 	if (data[IFLA_GRE_OFLAGS])
1427 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1428 
1429 	if (data[IFLA_GRE_IKEY])
1430 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1431 
1432 	if (data[IFLA_GRE_OKEY])
1433 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1434 
1435 	if (data[IFLA_GRE_LOCAL])
1436 		nla_memcpy(&parms->laddr, data[IFLA_GRE_LOCAL], sizeof(struct in6_addr));
1437 
1438 	if (data[IFLA_GRE_REMOTE])
1439 		nla_memcpy(&parms->raddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1440 
1441 	if (data[IFLA_GRE_TTL])
1442 		parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1443 
1444 	if (data[IFLA_GRE_ENCAP_LIMIT])
1445 		parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1446 
1447 	if (data[IFLA_GRE_FLOWINFO])
1448 		parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1449 
1450 	if (data[IFLA_GRE_FLAGS])
1451 		parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1452 }
1453 
1454 static int ip6gre_tap_init(struct net_device *dev)
1455 {
1456 	struct ip6_tnl *tunnel;
1457 
1458 	tunnel = netdev_priv(dev);
1459 
1460 	tunnel->dev = dev;
1461 	tunnel->net = dev_net(dev);
1462 	strcpy(tunnel->parms.name, dev->name);
1463 
1464 	ip6gre_tnl_link_config(tunnel, 1);
1465 
1466 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1467 	if (!dev->tstats)
1468 		return -ENOMEM;
1469 
1470 	return 0;
1471 }
1472 
1473 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1474 	.ndo_init = ip6gre_tap_init,
1475 	.ndo_uninit = ip6gre_tunnel_uninit,
1476 	.ndo_start_xmit = ip6gre_tunnel_xmit,
1477 	.ndo_set_mac_address = eth_mac_addr,
1478 	.ndo_validate_addr = eth_validate_addr,
1479 	.ndo_change_mtu = ip6gre_tunnel_change_mtu,
1480 	.ndo_get_stats64 = ip_tunnel_get_stats64,
1481 };
1482 
1483 static void ip6gre_tap_setup(struct net_device *dev)
1484 {
1485 
1486 	ether_setup(dev);
1487 
1488 	dev->netdev_ops = &ip6gre_tap_netdev_ops;
1489 	dev->destructor = ip6gre_dev_free;
1490 
1491 	dev->iflink = 0;
1492 	dev->features |= NETIF_F_NETNS_LOCAL;
1493 }
1494 
1495 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1496 	struct nlattr *tb[], struct nlattr *data[])
1497 {
1498 	struct ip6_tnl *nt;
1499 	struct net *net = dev_net(dev);
1500 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1501 	int err;
1502 
1503 	nt = netdev_priv(dev);
1504 	ip6gre_netlink_parms(data, &nt->parms);
1505 
1506 	if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1507 		return -EEXIST;
1508 
1509 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1510 		eth_hw_addr_random(dev);
1511 
1512 	nt->dev = dev;
1513 	nt->net = dev_net(dev);
1514 	ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1515 
1516 	/* Can use a lockless transmit, unless we generate output sequences */
1517 	if (!(nt->parms.o_flags & GRE_SEQ))
1518 		dev->features |= NETIF_F_LLTX;
1519 
1520 	err = register_netdevice(dev);
1521 	if (err)
1522 		goto out;
1523 
1524 	dev_hold(dev);
1525 	ip6gre_tunnel_link(ign, nt);
1526 
1527 out:
1528 	return err;
1529 }
1530 
1531 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1532 			    struct nlattr *data[])
1533 {
1534 	struct ip6_tnl *t, *nt;
1535 	struct net *net = dev_net(dev);
1536 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1537 	struct __ip6_tnl_parm p;
1538 
1539 	if (dev == ign->fb_tunnel_dev)
1540 		return -EINVAL;
1541 
1542 	nt = netdev_priv(dev);
1543 	ip6gre_netlink_parms(data, &p);
1544 
1545 	t = ip6gre_tunnel_locate(net, &p, 0);
1546 
1547 	if (t) {
1548 		if (t->dev != dev)
1549 			return -EEXIST;
1550 	} else {
1551 		t = nt;
1552 
1553 		ip6gre_tunnel_unlink(ign, t);
1554 		ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1555 		ip6gre_tunnel_link(ign, t);
1556 		netdev_state_change(dev);
1557 	}
1558 
1559 	return 0;
1560 }
1561 
1562 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1563 {
1564 	struct net *net = dev_net(dev);
1565 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1566 
1567 	if (dev != ign->fb_tunnel_dev)
1568 		unregister_netdevice_queue(dev, head);
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 	.dellink	= ip6gre_dellink,
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