xref: /openbmc/linux/net/ipv6/ip6_gre.c (revision 2596e07a)
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/init.h>
28 #include <linux/in6.h>
29 #include <linux/inetdevice.h>
30 #include <linux/igmp.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/etherdevice.h>
33 #include <linux/if_ether.h>
34 #include <linux/hash.h>
35 #include <linux/if_tunnel.h>
36 #include <linux/ip6_tunnel.h>
37 
38 #include <net/sock.h>
39 #include <net/ip.h>
40 #include <net/ip_tunnels.h>
41 #include <net/icmp.h>
42 #include <net/protocol.h>
43 #include <net/addrconf.h>
44 #include <net/arp.h>
45 #include <net/checksum.h>
46 #include <net/dsfield.h>
47 #include <net/inet_ecn.h>
48 #include <net/xfrm.h>
49 #include <net/net_namespace.h>
50 #include <net/netns/generic.h>
51 #include <net/rtnetlink.h>
52 
53 #include <net/ipv6.h>
54 #include <net/ip6_fib.h>
55 #include <net/ip6_route.h>
56 #include <net/ip6_tunnel.h>
57 
58 
59 static bool log_ecn_error = true;
60 module_param(log_ecn_error, bool, 0644);
61 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
62 
63 #define HASH_SIZE_SHIFT  5
64 #define HASH_SIZE (1 << HASH_SIZE_SHIFT)
65 
66 static int ip6gre_net_id __read_mostly;
67 struct ip6gre_net {
68 	struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
69 
70 	struct net_device *fb_tunnel_dev;
71 };
72 
73 static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
74 static struct rtnl_link_ops ip6gre_tap_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)
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 NULL;
318 	if (t || !create)
319 		return t;
320 
321 	if (parms->name[0])
322 		strlcpy(name, parms->name, IFNAMSIZ);
323 	else
324 		strcpy(name, "ip6gre%d");
325 
326 	dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
327 			   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 ip6_tnl *t = netdev_priv(dev);
360 	struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
361 
362 	ip6gre_tunnel_unlink(ign, t);
363 	ip6_tnl_dst_reset(t);
364 	dev_put(dev);
365 }
366 
367 
368 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
369 		u8 type, u8 code, int offset, __be32 info)
370 {
371 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
372 	__be16 *p = (__be16 *)(skb->data + offset);
373 	int grehlen = offset + 4;
374 	struct ip6_tnl *t;
375 	__be16 flags;
376 
377 	flags = p[0];
378 	if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
379 		if (flags&(GRE_VERSION|GRE_ROUTING))
380 			return;
381 		if (flags&GRE_KEY) {
382 			grehlen += 4;
383 			if (flags&GRE_CSUM)
384 				grehlen += 4;
385 		}
386 	}
387 
388 	/* If only 8 bytes returned, keyed message will be dropped here */
389 	if (!pskb_may_pull(skb, grehlen))
390 		return;
391 	ipv6h = (const struct ipv6hdr *)skb->data;
392 	p = (__be16 *)(skb->data + offset);
393 
394 	t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
395 				flags & GRE_KEY ?
396 				*(((__be32 *)p) + (grehlen / 4) - 1) : 0,
397 				p[1]);
398 	if (!t)
399 		return;
400 
401 	switch (type) {
402 		__u32 teli;
403 		struct ipv6_tlv_tnl_enc_lim *tel;
404 		__u32 mtu;
405 	case ICMPV6_DEST_UNREACH:
406 		net_dbg_ratelimited("%s: Path to destination invalid or inactive!\n",
407 				    t->parms.name);
408 		break;
409 	case ICMPV6_TIME_EXCEED:
410 		if (code == ICMPV6_EXC_HOPLIMIT) {
411 			net_dbg_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
412 					    t->parms.name);
413 		}
414 		break;
415 	case ICMPV6_PARAMPROB:
416 		teli = 0;
417 		if (code == ICMPV6_HDR_FIELD)
418 			teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
419 
420 		if (teli && teli == be32_to_cpu(info) - 2) {
421 			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
422 			if (tel->encap_limit == 0) {
423 				net_dbg_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
424 						    t->parms.name);
425 			}
426 		} else {
427 			net_dbg_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
428 					    t->parms.name);
429 		}
430 		break;
431 	case ICMPV6_PKT_TOOBIG:
432 		mtu = be32_to_cpu(info) - offset;
433 		if (mtu < IPV6_MIN_MTU)
434 			mtu = IPV6_MIN_MTU;
435 		t->dev->mtu = mtu;
436 		break;
437 	}
438 
439 	if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
440 		t->err_count++;
441 	else
442 		t->err_count = 1;
443 	t->err_time = jiffies;
444 }
445 
446 static int ip6gre_rcv(struct sk_buff *skb)
447 {
448 	const struct ipv6hdr *ipv6h;
449 	u8     *h;
450 	__be16    flags;
451 	__sum16   csum = 0;
452 	__be32 key = 0;
453 	u32    seqno = 0;
454 	struct ip6_tnl *tunnel;
455 	int    offset = 4;
456 	__be16 gre_proto;
457 	int err;
458 
459 	if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
460 		goto drop;
461 
462 	ipv6h = ipv6_hdr(skb);
463 	h = skb->data;
464 	flags = *(__be16 *)h;
465 
466 	if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
467 		/* - Version must be 0.
468 		   - We do not support routing headers.
469 		 */
470 		if (flags&(GRE_VERSION|GRE_ROUTING))
471 			goto drop;
472 
473 		if (flags&GRE_CSUM) {
474 			csum = skb_checksum_simple_validate(skb);
475 			offset += 4;
476 		}
477 		if (flags&GRE_KEY) {
478 			key = *(__be32 *)(h + offset);
479 			offset += 4;
480 		}
481 		if (flags&GRE_SEQ) {
482 			seqno = ntohl(*(__be32 *)(h + offset));
483 			offset += 4;
484 		}
485 	}
486 
487 	gre_proto = *(__be16 *)(h + 2);
488 
489 	tunnel = ip6gre_tunnel_lookup(skb->dev,
490 					  &ipv6h->saddr, &ipv6h->daddr, key,
491 					  gre_proto);
492 	if (tunnel) {
493 		struct pcpu_sw_netstats *tstats;
494 
495 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
496 			goto drop;
497 
498 		if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
499 			tunnel->dev->stats.rx_dropped++;
500 			goto drop;
501 		}
502 
503 		skb->protocol = gre_proto;
504 		/* WCCP version 1 and 2 protocol decoding.
505 		 * - Change protocol to IPv6
506 		 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
507 		 */
508 		if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
509 			skb->protocol = htons(ETH_P_IPV6);
510 			if ((*(h + offset) & 0xF0) != 0x40)
511 				offset += 4;
512 		}
513 
514 		skb->mac_header = skb->network_header;
515 		__pskb_pull(skb, offset);
516 		skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
517 
518 		if (((flags&GRE_CSUM) && csum) ||
519 		    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
520 			tunnel->dev->stats.rx_crc_errors++;
521 			tunnel->dev->stats.rx_errors++;
522 			goto drop;
523 		}
524 		if (tunnel->parms.i_flags&GRE_SEQ) {
525 			if (!(flags&GRE_SEQ) ||
526 			    (tunnel->i_seqno &&
527 					(s32)(seqno - tunnel->i_seqno) < 0)) {
528 				tunnel->dev->stats.rx_fifo_errors++;
529 				tunnel->dev->stats.rx_errors++;
530 				goto drop;
531 			}
532 			tunnel->i_seqno = seqno + 1;
533 		}
534 
535 		/* Warning: All skb pointers will be invalidated! */
536 		if (tunnel->dev->type == ARPHRD_ETHER) {
537 			if (!pskb_may_pull(skb, ETH_HLEN)) {
538 				tunnel->dev->stats.rx_length_errors++;
539 				tunnel->dev->stats.rx_errors++;
540 				goto drop;
541 			}
542 
543 			ipv6h = ipv6_hdr(skb);
544 			skb->protocol = eth_type_trans(skb, tunnel->dev);
545 			skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
546 		}
547 
548 		__skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
549 
550 		skb_reset_network_header(skb);
551 
552 		err = IP6_ECN_decapsulate(ipv6h, skb);
553 		if (unlikely(err)) {
554 			if (log_ecn_error)
555 				net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
556 						     &ipv6h->saddr,
557 						     ipv6_get_dsfield(ipv6h));
558 			if (err > 1) {
559 				++tunnel->dev->stats.rx_frame_errors;
560 				++tunnel->dev->stats.rx_errors;
561 				goto drop;
562 			}
563 		}
564 
565 		tstats = this_cpu_ptr(tunnel->dev->tstats);
566 		u64_stats_update_begin(&tstats->syncp);
567 		tstats->rx_packets++;
568 		tstats->rx_bytes += skb->len;
569 		u64_stats_update_end(&tstats->syncp);
570 
571 		netif_rx(skb);
572 
573 		return 0;
574 	}
575 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
576 
577 drop:
578 	kfree_skb(skb);
579 	return 0;
580 }
581 
582 struct ipv6_tel_txoption {
583 	struct ipv6_txoptions ops;
584 	__u8 dst_opt[8];
585 };
586 
587 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
588 {
589 	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
590 
591 	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
592 	opt->dst_opt[3] = 1;
593 	opt->dst_opt[4] = encap_limit;
594 	opt->dst_opt[5] = IPV6_TLV_PADN;
595 	opt->dst_opt[6] = 1;
596 
597 	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
598 	opt->ops.opt_nflen = 8;
599 }
600 
601 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
602 			 struct net_device *dev,
603 			 __u8 dsfield,
604 			 struct flowi6 *fl6,
605 			 int encap_limit,
606 			 __u32 *pmtu)
607 {
608 	struct ip6_tnl *tunnel = netdev_priv(dev);
609 	struct net *net = tunnel->net;
610 	struct net_device *tdev;    /* Device to other host */
611 	struct ipv6hdr  *ipv6h;     /* Our new IP header */
612 	unsigned int max_headroom = 0; /* The extra header space needed */
613 	int    gre_hlen;
614 	struct ipv6_tel_txoption opt;
615 	int    mtu;
616 	struct dst_entry *dst = NULL, *ndst = NULL;
617 	struct net_device_stats *stats = &tunnel->dev->stats;
618 	int err = -1;
619 	u8 proto;
620 	struct sk_buff *new_skb;
621 	__be16 protocol;
622 
623 	if (dev->type == ARPHRD_ETHER)
624 		IPCB(skb)->flags = 0;
625 
626 	if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
627 		gre_hlen = 0;
628 		ipv6h = (struct ipv6hdr *)skb->data;
629 		fl6->daddr = ipv6h->daddr;
630 	} else {
631 		gre_hlen = tunnel->hlen;
632 		fl6->daddr = tunnel->parms.raddr;
633 	}
634 
635 	if (!fl6->flowi6_mark)
636 		dst = ip6_tnl_dst_get(tunnel);
637 
638 	if (!dst) {
639 		dst = ip6_route_output(net, NULL, fl6);
640 
641 		if (dst->error)
642 			goto tx_err_link_failure;
643 		dst = xfrm_lookup(net, dst, flowi6_to_flowi(fl6), NULL, 0);
644 		if (IS_ERR(dst)) {
645 			err = PTR_ERR(dst);
646 			dst = NULL;
647 			goto tx_err_link_failure;
648 		}
649 		ndst = dst;
650 	}
651 
652 	tdev = dst->dev;
653 
654 	if (tdev == dev) {
655 		stats->collisions++;
656 		net_warn_ratelimited("%s: Local routing loop detected!\n",
657 				     tunnel->parms.name);
658 		goto tx_err_dst_release;
659 	}
660 
661 	mtu = dst_mtu(dst) - sizeof(*ipv6h);
662 	if (encap_limit >= 0) {
663 		max_headroom += 8;
664 		mtu -= 8;
665 	}
666 	if (mtu < IPV6_MIN_MTU)
667 		mtu = IPV6_MIN_MTU;
668 	if (skb_dst(skb))
669 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
670 	if (skb->len > mtu) {
671 		*pmtu = mtu;
672 		err = -EMSGSIZE;
673 		goto tx_err_dst_release;
674 	}
675 
676 	if (tunnel->err_count > 0) {
677 		if (time_before(jiffies,
678 				tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
679 			tunnel->err_count--;
680 
681 			dst_link_failure(skb);
682 		} else
683 			tunnel->err_count = 0;
684 	}
685 
686 	skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
687 
688 	max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
689 
690 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
691 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
692 		new_skb = skb_realloc_headroom(skb, max_headroom);
693 		if (max_headroom > dev->needed_headroom)
694 			dev->needed_headroom = max_headroom;
695 		if (!new_skb)
696 			goto tx_err_dst_release;
697 
698 		if (skb->sk)
699 			skb_set_owner_w(new_skb, skb->sk);
700 		consume_skb(skb);
701 		skb = new_skb;
702 	}
703 
704 	if (!fl6->flowi6_mark && ndst)
705 		ip6_tnl_dst_set(tunnel, ndst);
706 	skb_dst_set(skb, dst);
707 
708 	proto = NEXTHDR_GRE;
709 	if (encap_limit >= 0) {
710 		init_tel_txopt(&opt, encap_limit);
711 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
712 	}
713 
714 	if (likely(!skb->encapsulation)) {
715 		skb_reset_inner_headers(skb);
716 		skb->encapsulation = 1;
717 	}
718 
719 	skb_push(skb, gre_hlen);
720 	skb_reset_network_header(skb);
721 	skb_set_transport_header(skb, sizeof(*ipv6h));
722 
723 	/*
724 	 *	Push down and install the IP header.
725 	 */
726 	ipv6h = ipv6_hdr(skb);
727 	ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
728 		     ip6_make_flowlabel(net, skb, fl6->flowlabel, true, fl6));
729 	ipv6h->hop_limit = tunnel->parms.hop_limit;
730 	ipv6h->nexthdr = proto;
731 	ipv6h->saddr = fl6->saddr;
732 	ipv6h->daddr = fl6->daddr;
733 
734 	((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
735 	protocol = (dev->type == ARPHRD_ETHER) ?
736 		    htons(ETH_P_TEB) : skb->protocol;
737 	((__be16 *)(ipv6h + 1))[1] = protocol;
738 
739 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
740 		__be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
741 
742 		if (tunnel->parms.o_flags&GRE_SEQ) {
743 			++tunnel->o_seqno;
744 			*ptr = htonl(tunnel->o_seqno);
745 			ptr--;
746 		}
747 		if (tunnel->parms.o_flags&GRE_KEY) {
748 			*ptr = tunnel->parms.o_key;
749 			ptr--;
750 		}
751 		if (tunnel->parms.o_flags&GRE_CSUM) {
752 			*ptr = 0;
753 			*(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
754 				skb->len - sizeof(struct ipv6hdr));
755 		}
756 	}
757 
758 	skb_set_inner_protocol(skb, protocol);
759 
760 	ip6tunnel_xmit(NULL, skb, dev);
761 	return 0;
762 tx_err_link_failure:
763 	stats->tx_carrier_errors++;
764 	dst_link_failure(skb);
765 tx_err_dst_release:
766 	dst_release(dst);
767 	return err;
768 }
769 
770 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
771 {
772 	struct ip6_tnl *t = netdev_priv(dev);
773 	const struct iphdr  *iph = ip_hdr(skb);
774 	int encap_limit = -1;
775 	struct flowi6 fl6;
776 	__u8 dsfield;
777 	__u32 mtu;
778 	int err;
779 
780 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
781 		encap_limit = t->parms.encap_limit;
782 
783 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
784 	fl6.flowi6_proto = IPPROTO_GRE;
785 
786 	dsfield = ipv4_get_dsfield(iph);
787 
788 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
789 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
790 					  & IPV6_TCLASS_MASK;
791 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
792 		fl6.flowi6_mark = skb->mark;
793 
794 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
795 	if (err != 0) {
796 		/* XXX: send ICMP error even if DF is not set. */
797 		if (err == -EMSGSIZE)
798 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
799 				  htonl(mtu));
800 		return -1;
801 	}
802 
803 	return 0;
804 }
805 
806 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
807 {
808 	struct ip6_tnl *t = netdev_priv(dev);
809 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
810 	int encap_limit = -1;
811 	__u16 offset;
812 	struct flowi6 fl6;
813 	__u8 dsfield;
814 	__u32 mtu;
815 	int err;
816 
817 	if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
818 		return -1;
819 
820 	offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
821 	if (offset > 0) {
822 		struct ipv6_tlv_tnl_enc_lim *tel;
823 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
824 		if (tel->encap_limit == 0) {
825 			icmpv6_send(skb, ICMPV6_PARAMPROB,
826 				    ICMPV6_HDR_FIELD, offset + 2);
827 			return -1;
828 		}
829 		encap_limit = tel->encap_limit - 1;
830 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
831 		encap_limit = t->parms.encap_limit;
832 
833 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
834 	fl6.flowi6_proto = IPPROTO_GRE;
835 
836 	dsfield = ipv6_get_dsfield(ipv6h);
837 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
838 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
839 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
840 		fl6.flowlabel |= ip6_flowlabel(ipv6h);
841 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
842 		fl6.flowi6_mark = skb->mark;
843 
844 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
845 	if (err != 0) {
846 		if (err == -EMSGSIZE)
847 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
848 		return -1;
849 	}
850 
851 	return 0;
852 }
853 
854 /**
855  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
856  *   @t: the outgoing tunnel device
857  *   @hdr: IPv6 header from the incoming packet
858  *
859  * Description:
860  *   Avoid trivial tunneling loop by checking that tunnel exit-point
861  *   doesn't match source of incoming packet.
862  *
863  * Return:
864  *   1 if conflict,
865  *   0 else
866  **/
867 
868 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
869 	const struct ipv6hdr *hdr)
870 {
871 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
872 }
873 
874 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
875 {
876 	struct ip6_tnl *t = netdev_priv(dev);
877 	int encap_limit = -1;
878 	struct flowi6 fl6;
879 	__u32 mtu;
880 	int err;
881 
882 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
883 		encap_limit = t->parms.encap_limit;
884 
885 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
886 	fl6.flowi6_proto = skb->protocol;
887 
888 	err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
889 
890 	return err;
891 }
892 
893 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
894 	struct net_device *dev)
895 {
896 	struct ip6_tnl *t = netdev_priv(dev);
897 	struct net_device_stats *stats = &t->dev->stats;
898 	int ret;
899 
900 	if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
901 		goto tx_err;
902 
903 	switch (skb->protocol) {
904 	case htons(ETH_P_IP):
905 		ret = ip6gre_xmit_ipv4(skb, dev);
906 		break;
907 	case htons(ETH_P_IPV6):
908 		ret = ip6gre_xmit_ipv6(skb, dev);
909 		break;
910 	default:
911 		ret = ip6gre_xmit_other(skb, dev);
912 		break;
913 	}
914 
915 	if (ret < 0)
916 		goto tx_err;
917 
918 	return NETDEV_TX_OK;
919 
920 tx_err:
921 	stats->tx_errors++;
922 	stats->tx_dropped++;
923 	kfree_skb(skb);
924 	return NETDEV_TX_OK;
925 }
926 
927 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
928 {
929 	struct net_device *dev = t->dev;
930 	struct __ip6_tnl_parm *p = &t->parms;
931 	struct flowi6 *fl6 = &t->fl.u.ip6;
932 	int addend = sizeof(struct ipv6hdr) + 4;
933 
934 	if (dev->type != ARPHRD_ETHER) {
935 		memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
936 		memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
937 	}
938 
939 	/* Set up flowi template */
940 	fl6->saddr = p->laddr;
941 	fl6->daddr = p->raddr;
942 	fl6->flowi6_oif = p->link;
943 	fl6->flowlabel = 0;
944 
945 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
946 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
947 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
948 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
949 
950 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
951 	p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
952 
953 	if (p->flags&IP6_TNL_F_CAP_XMIT &&
954 			p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
955 		dev->flags |= IFF_POINTOPOINT;
956 	else
957 		dev->flags &= ~IFF_POINTOPOINT;
958 
959 	/* Precalculate GRE options length */
960 	if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
961 		if (t->parms.o_flags&GRE_CSUM)
962 			addend += 4;
963 		if (t->parms.o_flags&GRE_KEY)
964 			addend += 4;
965 		if (t->parms.o_flags&GRE_SEQ)
966 			addend += 4;
967 	}
968 	t->hlen = addend;
969 
970 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
971 		int strict = (ipv6_addr_type(&p->raddr) &
972 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
973 
974 		struct rt6_info *rt = rt6_lookup(t->net,
975 						 &p->raddr, &p->laddr,
976 						 p->link, strict);
977 
978 		if (!rt)
979 			return;
980 
981 		if (rt->dst.dev) {
982 			dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
983 
984 			if (set_mtu) {
985 				dev->mtu = rt->dst.dev->mtu - addend;
986 				if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
987 					dev->mtu -= 8;
988 
989 				if (dev->mtu < IPV6_MIN_MTU)
990 					dev->mtu = IPV6_MIN_MTU;
991 			}
992 		}
993 		ip6_rt_put(rt);
994 	}
995 }
996 
997 static int ip6gre_tnl_change(struct ip6_tnl *t,
998 	const struct __ip6_tnl_parm *p, int set_mtu)
999 {
1000 	t->parms.laddr = p->laddr;
1001 	t->parms.raddr = p->raddr;
1002 	t->parms.flags = p->flags;
1003 	t->parms.hop_limit = p->hop_limit;
1004 	t->parms.encap_limit = p->encap_limit;
1005 	t->parms.flowinfo = p->flowinfo;
1006 	t->parms.link = p->link;
1007 	t->parms.proto = p->proto;
1008 	t->parms.i_key = p->i_key;
1009 	t->parms.o_key = p->o_key;
1010 	t->parms.i_flags = p->i_flags;
1011 	t->parms.o_flags = p->o_flags;
1012 	ip6_tnl_dst_reset(t);
1013 	ip6gre_tnl_link_config(t, set_mtu);
1014 	return 0;
1015 }
1016 
1017 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1018 	const struct ip6_tnl_parm2 *u)
1019 {
1020 	p->laddr = u->laddr;
1021 	p->raddr = u->raddr;
1022 	p->flags = u->flags;
1023 	p->hop_limit = u->hop_limit;
1024 	p->encap_limit = u->encap_limit;
1025 	p->flowinfo = u->flowinfo;
1026 	p->link = u->link;
1027 	p->i_key = u->i_key;
1028 	p->o_key = u->o_key;
1029 	p->i_flags = u->i_flags;
1030 	p->o_flags = u->o_flags;
1031 	memcpy(p->name, u->name, sizeof(u->name));
1032 }
1033 
1034 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1035 	const struct __ip6_tnl_parm *p)
1036 {
1037 	u->proto = IPPROTO_GRE;
1038 	u->laddr = p->laddr;
1039 	u->raddr = p->raddr;
1040 	u->flags = p->flags;
1041 	u->hop_limit = p->hop_limit;
1042 	u->encap_limit = p->encap_limit;
1043 	u->flowinfo = p->flowinfo;
1044 	u->link = p->link;
1045 	u->i_key = p->i_key;
1046 	u->o_key = p->o_key;
1047 	u->i_flags = p->i_flags;
1048 	u->o_flags = p->o_flags;
1049 	memcpy(u->name, p->name, sizeof(u->name));
1050 }
1051 
1052 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1053 	struct ifreq *ifr, int cmd)
1054 {
1055 	int err = 0;
1056 	struct ip6_tnl_parm2 p;
1057 	struct __ip6_tnl_parm p1;
1058 	struct ip6_tnl *t = netdev_priv(dev);
1059 	struct net *net = t->net;
1060 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1061 
1062 	switch (cmd) {
1063 	case SIOCGETTUNNEL:
1064 		if (dev == ign->fb_tunnel_dev) {
1065 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1066 				err = -EFAULT;
1067 				break;
1068 			}
1069 			ip6gre_tnl_parm_from_user(&p1, &p);
1070 			t = ip6gre_tunnel_locate(net, &p1, 0);
1071 			if (!t)
1072 				t = netdev_priv(dev);
1073 		}
1074 		memset(&p, 0, sizeof(p));
1075 		ip6gre_tnl_parm_to_user(&p, &t->parms);
1076 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1077 			err = -EFAULT;
1078 		break;
1079 
1080 	case SIOCADDTUNNEL:
1081 	case SIOCCHGTUNNEL:
1082 		err = -EPERM;
1083 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1084 			goto done;
1085 
1086 		err = -EFAULT;
1087 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1088 			goto done;
1089 
1090 		err = -EINVAL;
1091 		if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1092 			goto done;
1093 
1094 		if (!(p.i_flags&GRE_KEY))
1095 			p.i_key = 0;
1096 		if (!(p.o_flags&GRE_KEY))
1097 			p.o_key = 0;
1098 
1099 		ip6gre_tnl_parm_from_user(&p1, &p);
1100 		t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1101 
1102 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1103 			if (t) {
1104 				if (t->dev != dev) {
1105 					err = -EEXIST;
1106 					break;
1107 				}
1108 			} else {
1109 				t = netdev_priv(dev);
1110 
1111 				ip6gre_tunnel_unlink(ign, t);
1112 				synchronize_net();
1113 				ip6gre_tnl_change(t, &p1, 1);
1114 				ip6gre_tunnel_link(ign, t);
1115 				netdev_state_change(dev);
1116 			}
1117 		}
1118 
1119 		if (t) {
1120 			err = 0;
1121 
1122 			memset(&p, 0, sizeof(p));
1123 			ip6gre_tnl_parm_to_user(&p, &t->parms);
1124 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1125 				err = -EFAULT;
1126 		} else
1127 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1128 		break;
1129 
1130 	case SIOCDELTUNNEL:
1131 		err = -EPERM;
1132 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1133 			goto done;
1134 
1135 		if (dev == ign->fb_tunnel_dev) {
1136 			err = -EFAULT;
1137 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1138 				goto done;
1139 			err = -ENOENT;
1140 			ip6gre_tnl_parm_from_user(&p1, &p);
1141 			t = ip6gre_tunnel_locate(net, &p1, 0);
1142 			if (!t)
1143 				goto done;
1144 			err = -EPERM;
1145 			if (t == netdev_priv(ign->fb_tunnel_dev))
1146 				goto done;
1147 			dev = t->dev;
1148 		}
1149 		unregister_netdevice(dev);
1150 		err = 0;
1151 		break;
1152 
1153 	default:
1154 		err = -EINVAL;
1155 	}
1156 
1157 done:
1158 	return err;
1159 }
1160 
1161 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1162 {
1163 	if (new_mtu < 68 ||
1164 	    new_mtu > 0xFFF8 - dev->hard_header_len)
1165 		return -EINVAL;
1166 	dev->mtu = new_mtu;
1167 	return 0;
1168 }
1169 
1170 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1171 			unsigned short type,
1172 			const void *daddr, const void *saddr, unsigned int len)
1173 {
1174 	struct ip6_tnl *t = netdev_priv(dev);
1175 	struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1176 	__be16 *p = (__be16 *)(ipv6h+1);
1177 
1178 	ip6_flow_hdr(ipv6h, 0,
1179 		     ip6_make_flowlabel(dev_net(dev), skb,
1180 					t->fl.u.ip6.flowlabel, true,
1181 					&t->fl.u.ip6));
1182 	ipv6h->hop_limit = t->parms.hop_limit;
1183 	ipv6h->nexthdr = NEXTHDR_GRE;
1184 	ipv6h->saddr = t->parms.laddr;
1185 	ipv6h->daddr = t->parms.raddr;
1186 
1187 	p[0]		= t->parms.o_flags;
1188 	p[1]		= htons(type);
1189 
1190 	/*
1191 	 *	Set the source hardware address.
1192 	 */
1193 
1194 	if (saddr)
1195 		memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1196 	if (daddr)
1197 		memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1198 	if (!ipv6_addr_any(&ipv6h->daddr))
1199 		return t->hlen;
1200 
1201 	return -t->hlen;
1202 }
1203 
1204 static const struct header_ops ip6gre_header_ops = {
1205 	.create	= ip6gre_header,
1206 };
1207 
1208 static const struct net_device_ops ip6gre_netdev_ops = {
1209 	.ndo_init		= ip6gre_tunnel_init,
1210 	.ndo_uninit		= ip6gre_tunnel_uninit,
1211 	.ndo_start_xmit		= ip6gre_tunnel_xmit,
1212 	.ndo_do_ioctl		= ip6gre_tunnel_ioctl,
1213 	.ndo_change_mtu		= ip6gre_tunnel_change_mtu,
1214 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1215 	.ndo_get_iflink		= ip6_tnl_get_iflink,
1216 };
1217 
1218 static void ip6gre_dev_free(struct net_device *dev)
1219 {
1220 	struct ip6_tnl *t = netdev_priv(dev);
1221 
1222 	ip6_tnl_dst_destroy(t);
1223 	free_percpu(dev->tstats);
1224 	free_netdev(dev);
1225 }
1226 
1227 static void ip6gre_tunnel_setup(struct net_device *dev)
1228 {
1229 	struct ip6_tnl *t;
1230 
1231 	dev->netdev_ops = &ip6gre_netdev_ops;
1232 	dev->destructor = ip6gre_dev_free;
1233 
1234 	dev->type = ARPHRD_IP6GRE;
1235 	dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1236 	dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1237 	t = netdev_priv(dev);
1238 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1239 		dev->mtu -= 8;
1240 	dev->flags |= IFF_NOARP;
1241 	dev->addr_len = sizeof(struct in6_addr);
1242 	netif_keep_dst(dev);
1243 }
1244 
1245 static int ip6gre_tunnel_init_common(struct net_device *dev)
1246 {
1247 	struct ip6_tnl *tunnel;
1248 	int ret;
1249 
1250 	tunnel = netdev_priv(dev);
1251 
1252 	tunnel->dev = dev;
1253 	tunnel->net = dev_net(dev);
1254 	strcpy(tunnel->parms.name, dev->name);
1255 
1256 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1257 	if (!dev->tstats)
1258 		return -ENOMEM;
1259 
1260 	ret = ip6_tnl_dst_init(tunnel);
1261 	if (ret) {
1262 		free_percpu(dev->tstats);
1263 		dev->tstats = NULL;
1264 		return ret;
1265 	}
1266 
1267 	return 0;
1268 }
1269 
1270 static int ip6gre_tunnel_init(struct net_device *dev)
1271 {
1272 	struct ip6_tnl *tunnel;
1273 	int ret;
1274 
1275 	ret = ip6gre_tunnel_init_common(dev);
1276 	if (ret)
1277 		return ret;
1278 
1279 	tunnel = netdev_priv(dev);
1280 
1281 	memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1282 	memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1283 
1284 	if (ipv6_addr_any(&tunnel->parms.raddr))
1285 		dev->header_ops = &ip6gre_header_ops;
1286 
1287 	return 0;
1288 }
1289 
1290 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1291 {
1292 	struct ip6_tnl *tunnel = netdev_priv(dev);
1293 
1294 	tunnel->dev = dev;
1295 	tunnel->net = dev_net(dev);
1296 	strcpy(tunnel->parms.name, dev->name);
1297 
1298 	tunnel->hlen		= sizeof(struct ipv6hdr) + 4;
1299 
1300 	dev_hold(dev);
1301 }
1302 
1303 
1304 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1305 	.handler     = ip6gre_rcv,
1306 	.err_handler = ip6gre_err,
1307 	.flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1308 };
1309 
1310 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1311 {
1312 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1313 	struct net_device *dev, *aux;
1314 	int prio;
1315 
1316 	for_each_netdev_safe(net, dev, aux)
1317 		if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1318 		    dev->rtnl_link_ops == &ip6gre_tap_ops)
1319 			unregister_netdevice_queue(dev, head);
1320 
1321 	for (prio = 0; prio < 4; prio++) {
1322 		int h;
1323 		for (h = 0; h < HASH_SIZE; h++) {
1324 			struct ip6_tnl *t;
1325 
1326 			t = rtnl_dereference(ign->tunnels[prio][h]);
1327 
1328 			while (t) {
1329 				/* If dev is in the same netns, it has already
1330 				 * been added to the list by the previous loop.
1331 				 */
1332 				if (!net_eq(dev_net(t->dev), net))
1333 					unregister_netdevice_queue(t->dev,
1334 								   head);
1335 				t = rtnl_dereference(t->next);
1336 			}
1337 		}
1338 	}
1339 }
1340 
1341 static int __net_init ip6gre_init_net(struct net *net)
1342 {
1343 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1344 	int err;
1345 
1346 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1347 					  NET_NAME_UNKNOWN,
1348 					  ip6gre_tunnel_setup);
1349 	if (!ign->fb_tunnel_dev) {
1350 		err = -ENOMEM;
1351 		goto err_alloc_dev;
1352 	}
1353 	dev_net_set(ign->fb_tunnel_dev, net);
1354 	/* FB netdevice is special: we have one, and only one per netns.
1355 	 * Allowing to move it to another netns is clearly unsafe.
1356 	 */
1357 	ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1358 
1359 
1360 	ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1361 	ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1362 
1363 	err = register_netdev(ign->fb_tunnel_dev);
1364 	if (err)
1365 		goto err_reg_dev;
1366 
1367 	rcu_assign_pointer(ign->tunnels_wc[0],
1368 			   netdev_priv(ign->fb_tunnel_dev));
1369 	return 0;
1370 
1371 err_reg_dev:
1372 	ip6gre_dev_free(ign->fb_tunnel_dev);
1373 err_alloc_dev:
1374 	return err;
1375 }
1376 
1377 static void __net_exit ip6gre_exit_net(struct net *net)
1378 {
1379 	LIST_HEAD(list);
1380 
1381 	rtnl_lock();
1382 	ip6gre_destroy_tunnels(net, &list);
1383 	unregister_netdevice_many(&list);
1384 	rtnl_unlock();
1385 }
1386 
1387 static struct pernet_operations ip6gre_net_ops = {
1388 	.init = ip6gre_init_net,
1389 	.exit = ip6gre_exit_net,
1390 	.id   = &ip6gre_net_id,
1391 	.size = sizeof(struct ip6gre_net),
1392 };
1393 
1394 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1395 {
1396 	__be16 flags;
1397 
1398 	if (!data)
1399 		return 0;
1400 
1401 	flags = 0;
1402 	if (data[IFLA_GRE_IFLAGS])
1403 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1404 	if (data[IFLA_GRE_OFLAGS])
1405 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1406 	if (flags & (GRE_VERSION|GRE_ROUTING))
1407 		return -EINVAL;
1408 
1409 	return 0;
1410 }
1411 
1412 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1413 {
1414 	struct in6_addr daddr;
1415 
1416 	if (tb[IFLA_ADDRESS]) {
1417 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1418 			return -EINVAL;
1419 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1420 			return -EADDRNOTAVAIL;
1421 	}
1422 
1423 	if (!data)
1424 		goto out;
1425 
1426 	if (data[IFLA_GRE_REMOTE]) {
1427 		daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1428 		if (ipv6_addr_any(&daddr))
1429 			return -EINVAL;
1430 	}
1431 
1432 out:
1433 	return ip6gre_tunnel_validate(tb, data);
1434 }
1435 
1436 
1437 static void ip6gre_netlink_parms(struct nlattr *data[],
1438 				struct __ip6_tnl_parm *parms)
1439 {
1440 	memset(parms, 0, sizeof(*parms));
1441 
1442 	if (!data)
1443 		return;
1444 
1445 	if (data[IFLA_GRE_LINK])
1446 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1447 
1448 	if (data[IFLA_GRE_IFLAGS])
1449 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1450 
1451 	if (data[IFLA_GRE_OFLAGS])
1452 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1453 
1454 	if (data[IFLA_GRE_IKEY])
1455 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1456 
1457 	if (data[IFLA_GRE_OKEY])
1458 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1459 
1460 	if (data[IFLA_GRE_LOCAL])
1461 		parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
1462 
1463 	if (data[IFLA_GRE_REMOTE])
1464 		parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1465 
1466 	if (data[IFLA_GRE_TTL])
1467 		parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1468 
1469 	if (data[IFLA_GRE_ENCAP_LIMIT])
1470 		parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1471 
1472 	if (data[IFLA_GRE_FLOWINFO])
1473 		parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1474 
1475 	if (data[IFLA_GRE_FLAGS])
1476 		parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1477 }
1478 
1479 static int ip6gre_tap_init(struct net_device *dev)
1480 {
1481 	struct ip6_tnl *tunnel;
1482 	int ret;
1483 
1484 	ret = ip6gre_tunnel_init_common(dev);
1485 	if (ret)
1486 		return ret;
1487 
1488 	tunnel = netdev_priv(dev);
1489 
1490 	ip6gre_tnl_link_config(tunnel, 1);
1491 
1492 	return 0;
1493 }
1494 
1495 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1496 	.ndo_init = ip6gre_tap_init,
1497 	.ndo_uninit = ip6gre_tunnel_uninit,
1498 	.ndo_start_xmit = ip6gre_tunnel_xmit,
1499 	.ndo_set_mac_address = eth_mac_addr,
1500 	.ndo_validate_addr = eth_validate_addr,
1501 	.ndo_change_mtu = ip6gre_tunnel_change_mtu,
1502 	.ndo_get_stats64 = ip_tunnel_get_stats64,
1503 	.ndo_get_iflink = ip6_tnl_get_iflink,
1504 };
1505 
1506 static void ip6gre_tap_setup(struct net_device *dev)
1507 {
1508 
1509 	ether_setup(dev);
1510 
1511 	dev->netdev_ops = &ip6gre_tap_netdev_ops;
1512 	dev->destructor = ip6gre_dev_free;
1513 
1514 	dev->features |= NETIF_F_NETNS_LOCAL;
1515 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1516 }
1517 
1518 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1519 	struct nlattr *tb[], struct nlattr *data[])
1520 {
1521 	struct ip6_tnl *nt;
1522 	struct net *net = dev_net(dev);
1523 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1524 	int err;
1525 
1526 	nt = netdev_priv(dev);
1527 	ip6gre_netlink_parms(data, &nt->parms);
1528 
1529 	if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1530 		return -EEXIST;
1531 
1532 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1533 		eth_hw_addr_random(dev);
1534 
1535 	nt->dev = dev;
1536 	nt->net = dev_net(dev);
1537 	ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1538 
1539 	/* Can use a lockless transmit, unless we generate output sequences */
1540 	if (!(nt->parms.o_flags & GRE_SEQ))
1541 		dev->features |= NETIF_F_LLTX;
1542 
1543 	err = register_netdevice(dev);
1544 	if (err)
1545 		goto out;
1546 
1547 	dev_hold(dev);
1548 	ip6gre_tunnel_link(ign, nt);
1549 
1550 out:
1551 	return err;
1552 }
1553 
1554 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1555 			    struct nlattr *data[])
1556 {
1557 	struct ip6_tnl *t, *nt = netdev_priv(dev);
1558 	struct net *net = nt->net;
1559 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1560 	struct __ip6_tnl_parm p;
1561 
1562 	if (dev == ign->fb_tunnel_dev)
1563 		return -EINVAL;
1564 
1565 	ip6gre_netlink_parms(data, &p);
1566 
1567 	t = ip6gre_tunnel_locate(net, &p, 0);
1568 
1569 	if (t) {
1570 		if (t->dev != dev)
1571 			return -EEXIST;
1572 	} else {
1573 		t = nt;
1574 	}
1575 
1576 	ip6gre_tunnel_unlink(ign, t);
1577 	ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1578 	ip6gre_tunnel_link(ign, t);
1579 	return 0;
1580 }
1581 
1582 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1583 {
1584 	struct net *net = dev_net(dev);
1585 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1586 
1587 	if (dev != ign->fb_tunnel_dev)
1588 		unregister_netdevice_queue(dev, head);
1589 }
1590 
1591 static size_t ip6gre_get_size(const struct net_device *dev)
1592 {
1593 	return
1594 		/* IFLA_GRE_LINK */
1595 		nla_total_size(4) +
1596 		/* IFLA_GRE_IFLAGS */
1597 		nla_total_size(2) +
1598 		/* IFLA_GRE_OFLAGS */
1599 		nla_total_size(2) +
1600 		/* IFLA_GRE_IKEY */
1601 		nla_total_size(4) +
1602 		/* IFLA_GRE_OKEY */
1603 		nla_total_size(4) +
1604 		/* IFLA_GRE_LOCAL */
1605 		nla_total_size(sizeof(struct in6_addr)) +
1606 		/* IFLA_GRE_REMOTE */
1607 		nla_total_size(sizeof(struct in6_addr)) +
1608 		/* IFLA_GRE_TTL */
1609 		nla_total_size(1) +
1610 		/* IFLA_GRE_TOS */
1611 		nla_total_size(1) +
1612 		/* IFLA_GRE_ENCAP_LIMIT */
1613 		nla_total_size(1) +
1614 		/* IFLA_GRE_FLOWINFO */
1615 		nla_total_size(4) +
1616 		/* IFLA_GRE_FLAGS */
1617 		nla_total_size(4) +
1618 		0;
1619 }
1620 
1621 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1622 {
1623 	struct ip6_tnl *t = netdev_priv(dev);
1624 	struct __ip6_tnl_parm *p = &t->parms;
1625 
1626 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1627 	    nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1628 	    nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1629 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1630 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1631 	    nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1632 	    nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
1633 	    nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1634 	    /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1635 	    nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1636 	    nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1637 	    nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1638 		goto nla_put_failure;
1639 	return 0;
1640 
1641 nla_put_failure:
1642 	return -EMSGSIZE;
1643 }
1644 
1645 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1646 	[IFLA_GRE_LINK]        = { .type = NLA_U32 },
1647 	[IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1648 	[IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1649 	[IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1650 	[IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1651 	[IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1652 	[IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1653 	[IFLA_GRE_TTL]         = { .type = NLA_U8 },
1654 	[IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1655 	[IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1656 	[IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1657 };
1658 
1659 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1660 	.kind		= "ip6gre",
1661 	.maxtype	= IFLA_GRE_MAX,
1662 	.policy		= ip6gre_policy,
1663 	.priv_size	= sizeof(struct ip6_tnl),
1664 	.setup		= ip6gre_tunnel_setup,
1665 	.validate	= ip6gre_tunnel_validate,
1666 	.newlink	= ip6gre_newlink,
1667 	.changelink	= ip6gre_changelink,
1668 	.dellink	= ip6gre_dellink,
1669 	.get_size	= ip6gre_get_size,
1670 	.fill_info	= ip6gre_fill_info,
1671 	.get_link_net	= ip6_tnl_get_link_net,
1672 };
1673 
1674 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1675 	.kind		= "ip6gretap",
1676 	.maxtype	= IFLA_GRE_MAX,
1677 	.policy		= ip6gre_policy,
1678 	.priv_size	= sizeof(struct ip6_tnl),
1679 	.setup		= ip6gre_tap_setup,
1680 	.validate	= ip6gre_tap_validate,
1681 	.newlink	= ip6gre_newlink,
1682 	.changelink	= ip6gre_changelink,
1683 	.get_size	= ip6gre_get_size,
1684 	.fill_info	= ip6gre_fill_info,
1685 	.get_link_net	= ip6_tnl_get_link_net,
1686 };
1687 
1688 /*
1689  *	And now the modules code and kernel interface.
1690  */
1691 
1692 static int __init ip6gre_init(void)
1693 {
1694 	int err;
1695 
1696 	pr_info("GRE over IPv6 tunneling driver\n");
1697 
1698 	err = register_pernet_device(&ip6gre_net_ops);
1699 	if (err < 0)
1700 		return err;
1701 
1702 	err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1703 	if (err < 0) {
1704 		pr_info("%s: can't add protocol\n", __func__);
1705 		goto add_proto_failed;
1706 	}
1707 
1708 	err = rtnl_link_register(&ip6gre_link_ops);
1709 	if (err < 0)
1710 		goto rtnl_link_failed;
1711 
1712 	err = rtnl_link_register(&ip6gre_tap_ops);
1713 	if (err < 0)
1714 		goto tap_ops_failed;
1715 
1716 out:
1717 	return err;
1718 
1719 tap_ops_failed:
1720 	rtnl_link_unregister(&ip6gre_link_ops);
1721 rtnl_link_failed:
1722 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1723 add_proto_failed:
1724 	unregister_pernet_device(&ip6gre_net_ops);
1725 	goto out;
1726 }
1727 
1728 static void __exit ip6gre_fini(void)
1729 {
1730 	rtnl_link_unregister(&ip6gre_tap_ops);
1731 	rtnl_link_unregister(&ip6gre_link_ops);
1732 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1733 	unregister_pernet_device(&ip6gre_net_ops);
1734 }
1735 
1736 module_init(ip6gre_init);
1737 module_exit(ip6gre_fini);
1738 MODULE_LICENSE("GPL");
1739 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1740 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1741 MODULE_ALIAS_RTNL_LINK("ip6gre");
1742 MODULE_ALIAS_RTNL_LINK("ip6gretap");
1743 MODULE_ALIAS_NETDEV("ip6gre0");
1744