xref: /openbmc/linux/net/ipv6/ip6_gre.c (revision 92a2c6b2)
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 struct rtnl_link_ops ip6gre_tap_ops __read_mostly;
76 static int ip6gre_tunnel_init(struct net_device *dev);
77 static void ip6gre_tunnel_setup(struct net_device *dev);
78 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
79 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
80 
81 /* Tunnel hash table */
82 
83 /*
84    4 hash tables:
85 
86    3: (remote,local)
87    2: (remote,*)
88    1: (*,local)
89    0: (*,*)
90 
91    We require exact key match i.e. if a key is present in packet
92    it will match only tunnel with the same key; if it is not present,
93    it will match only keyless tunnel.
94 
95    All keysless packets, if not matched configured keyless tunnels
96    will match fallback tunnel.
97  */
98 
99 #define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
100 static u32 HASH_ADDR(const struct in6_addr *addr)
101 {
102 	u32 hash = ipv6_addr_hash(addr);
103 
104 	return hash_32(hash, HASH_SIZE_SHIFT);
105 }
106 
107 #define tunnels_r_l	tunnels[3]
108 #define tunnels_r	tunnels[2]
109 #define tunnels_l	tunnels[1]
110 #define tunnels_wc	tunnels[0]
111 
112 /* Given src, dst and key, find appropriate for input tunnel. */
113 
114 static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
115 		const struct in6_addr *remote, const struct in6_addr *local,
116 		__be32 key, __be16 gre_proto)
117 {
118 	struct net *net = dev_net(dev);
119 	int link = dev->ifindex;
120 	unsigned int h0 = HASH_ADDR(remote);
121 	unsigned int h1 = HASH_KEY(key);
122 	struct ip6_tnl *t, *cand = NULL;
123 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
124 	int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
125 		       ARPHRD_ETHER : ARPHRD_IP6GRE;
126 	int score, cand_score = 4;
127 
128 	for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
129 		if (!ipv6_addr_equal(local, &t->parms.laddr) ||
130 		    !ipv6_addr_equal(remote, &t->parms.raddr) ||
131 		    key != t->parms.i_key ||
132 		    !(t->dev->flags & IFF_UP))
133 			continue;
134 
135 		if (t->dev->type != ARPHRD_IP6GRE &&
136 		    t->dev->type != dev_type)
137 			continue;
138 
139 		score = 0;
140 		if (t->parms.link != link)
141 			score |= 1;
142 		if (t->dev->type != dev_type)
143 			score |= 2;
144 		if (score == 0)
145 			return t;
146 
147 		if (score < cand_score) {
148 			cand = t;
149 			cand_score = score;
150 		}
151 	}
152 
153 	for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
154 		if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
155 		    key != t->parms.i_key ||
156 		    !(t->dev->flags & IFF_UP))
157 			continue;
158 
159 		if (t->dev->type != ARPHRD_IP6GRE &&
160 		    t->dev->type != dev_type)
161 			continue;
162 
163 		score = 0;
164 		if (t->parms.link != link)
165 			score |= 1;
166 		if (t->dev->type != dev_type)
167 			score |= 2;
168 		if (score == 0)
169 			return t;
170 
171 		if (score < cand_score) {
172 			cand = t;
173 			cand_score = score;
174 		}
175 	}
176 
177 	for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
178 		if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
179 			  (!ipv6_addr_equal(local, &t->parms.raddr) ||
180 				 !ipv6_addr_is_multicast(local))) ||
181 		    key != t->parms.i_key ||
182 		    !(t->dev->flags & IFF_UP))
183 			continue;
184 
185 		if (t->dev->type != ARPHRD_IP6GRE &&
186 		    t->dev->type != dev_type)
187 			continue;
188 
189 		score = 0;
190 		if (t->parms.link != link)
191 			score |= 1;
192 		if (t->dev->type != dev_type)
193 			score |= 2;
194 		if (score == 0)
195 			return t;
196 
197 		if (score < cand_score) {
198 			cand = t;
199 			cand_score = score;
200 		}
201 	}
202 
203 	for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
204 		if (t->parms.i_key != key ||
205 		    !(t->dev->flags & IFF_UP))
206 			continue;
207 
208 		if (t->dev->type != ARPHRD_IP6GRE &&
209 		    t->dev->type != dev_type)
210 			continue;
211 
212 		score = 0;
213 		if (t->parms.link != link)
214 			score |= 1;
215 		if (t->dev->type != dev_type)
216 			score |= 2;
217 		if (score == 0)
218 			return t;
219 
220 		if (score < cand_score) {
221 			cand = t;
222 			cand_score = score;
223 		}
224 	}
225 
226 	if (cand != NULL)
227 		return cand;
228 
229 	dev = ign->fb_tunnel_dev;
230 	if (dev->flags & IFF_UP)
231 		return netdev_priv(dev);
232 
233 	return NULL;
234 }
235 
236 static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
237 		const struct __ip6_tnl_parm *p)
238 {
239 	const struct in6_addr *remote = &p->raddr;
240 	const struct in6_addr *local = &p->laddr;
241 	unsigned int h = HASH_KEY(p->i_key);
242 	int prio = 0;
243 
244 	if (!ipv6_addr_any(local))
245 		prio |= 1;
246 	if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
247 		prio |= 2;
248 		h ^= HASH_ADDR(remote);
249 	}
250 
251 	return &ign->tunnels[prio][h];
252 }
253 
254 static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
255 		const struct ip6_tnl *t)
256 {
257 	return __ip6gre_bucket(ign, &t->parms);
258 }
259 
260 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
261 {
262 	struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
263 
264 	rcu_assign_pointer(t->next, rtnl_dereference(*tp));
265 	rcu_assign_pointer(*tp, t);
266 }
267 
268 static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
269 {
270 	struct ip6_tnl __rcu **tp;
271 	struct ip6_tnl *iter;
272 
273 	for (tp = ip6gre_bucket(ign, t);
274 	     (iter = rtnl_dereference(*tp)) != NULL;
275 	     tp = &iter->next) {
276 		if (t == iter) {
277 			rcu_assign_pointer(*tp, t->next);
278 			break;
279 		}
280 	}
281 }
282 
283 static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
284 					   const struct __ip6_tnl_parm *parms,
285 					   int type)
286 {
287 	const struct in6_addr *remote = &parms->raddr;
288 	const struct in6_addr *local = &parms->laddr;
289 	__be32 key = parms->i_key;
290 	int link = parms->link;
291 	struct ip6_tnl *t;
292 	struct ip6_tnl __rcu **tp;
293 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
294 
295 	for (tp = __ip6gre_bucket(ign, parms);
296 	     (t = rtnl_dereference(*tp)) != NULL;
297 	     tp = &t->next)
298 		if (ipv6_addr_equal(local, &t->parms.laddr) &&
299 		    ipv6_addr_equal(remote, &t->parms.raddr) &&
300 		    key == t->parms.i_key &&
301 		    link == t->parms.link &&
302 		    type == t->dev->type)
303 			break;
304 
305 	return t;
306 }
307 
308 static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
309 		const struct __ip6_tnl_parm *parms, int create)
310 {
311 	struct ip6_tnl *t, *nt;
312 	struct net_device *dev;
313 	char name[IFNAMSIZ];
314 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
315 
316 	t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
317 	if (t && create)
318 		return NULL;
319 	if (t || !create)
320 		return t;
321 
322 	if (parms->name[0])
323 		strlcpy(name, parms->name, IFNAMSIZ);
324 	else
325 		strcpy(name, "ip6gre%d");
326 
327 	dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
328 			   ip6gre_tunnel_setup);
329 	if (!dev)
330 		return NULL;
331 
332 	dev_net_set(dev, net);
333 
334 	nt = netdev_priv(dev);
335 	nt->parms = *parms;
336 	dev->rtnl_link_ops = &ip6gre_link_ops;
337 
338 	nt->dev = dev;
339 	nt->net = dev_net(dev);
340 	ip6gre_tnl_link_config(nt, 1);
341 
342 	if (register_netdevice(dev) < 0)
343 		goto failed_free;
344 
345 	/* Can use a lockless transmit, unless we generate output sequences */
346 	if (!(nt->parms.o_flags & GRE_SEQ))
347 		dev->features |= NETIF_F_LLTX;
348 
349 	dev_hold(dev);
350 	ip6gre_tunnel_link(ign, nt);
351 	return nt;
352 
353 failed_free:
354 	free_netdev(dev);
355 	return NULL;
356 }
357 
358 static void ip6gre_tunnel_uninit(struct net_device *dev)
359 {
360 	struct ip6_tnl *t = netdev_priv(dev);
361 	struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
362 
363 	ip6gre_tunnel_unlink(ign, 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 == NULL)
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_warn_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_warn_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_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
424 						     t->parms.name);
425 			}
426 		} else {
427 			net_warn_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_check(tunnel);
637 
638 	if (!dst) {
639 		ndst = ip6_route_output(net, NULL, fl6);
640 
641 		if (ndst->error)
642 			goto tx_err_link_failure;
643 		ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
644 		if (IS_ERR(ndst)) {
645 			err = PTR_ERR(ndst);
646 			ndst = NULL;
647 			goto tx_err_link_failure;
648 		}
649 		dst = ndst;
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) {
705 		skb_dst_set(skb, dst);
706 		ndst = NULL;
707 	} else {
708 		skb_dst_set_noref(skb, dst);
709 	}
710 
711 	proto = NEXTHDR_GRE;
712 	if (encap_limit >= 0) {
713 		init_tel_txopt(&opt, encap_limit);
714 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
715 	}
716 
717 	if (likely(!skb->encapsulation)) {
718 		skb_reset_inner_headers(skb);
719 		skb->encapsulation = 1;
720 	}
721 
722 	skb_push(skb, gre_hlen);
723 	skb_reset_network_header(skb);
724 	skb_set_transport_header(skb, sizeof(*ipv6h));
725 
726 	/*
727 	 *	Push down and install the IP header.
728 	 */
729 	ipv6h = ipv6_hdr(skb);
730 	ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
731 		     ip6_make_flowlabel(net, skb, fl6->flowlabel, false));
732 	ipv6h->hop_limit = tunnel->parms.hop_limit;
733 	ipv6h->nexthdr = proto;
734 	ipv6h->saddr = fl6->saddr;
735 	ipv6h->daddr = fl6->daddr;
736 
737 	((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
738 	protocol = (dev->type == ARPHRD_ETHER) ?
739 		    htons(ETH_P_TEB) : skb->protocol;
740 	((__be16 *)(ipv6h + 1))[1] = protocol;
741 
742 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
743 		__be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
744 
745 		if (tunnel->parms.o_flags&GRE_SEQ) {
746 			++tunnel->o_seqno;
747 			*ptr = htonl(tunnel->o_seqno);
748 			ptr--;
749 		}
750 		if (tunnel->parms.o_flags&GRE_KEY) {
751 			*ptr = tunnel->parms.o_key;
752 			ptr--;
753 		}
754 		if (tunnel->parms.o_flags&GRE_CSUM) {
755 			*ptr = 0;
756 			*(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
757 				skb->len - sizeof(struct ipv6hdr));
758 		}
759 	}
760 
761 	skb_set_inner_protocol(skb, protocol);
762 
763 	ip6tunnel_xmit(skb, dev);
764 	if (ndst)
765 		ip6_tnl_dst_store(tunnel, ndst);
766 	return 0;
767 tx_err_link_failure:
768 	stats->tx_carrier_errors++;
769 	dst_link_failure(skb);
770 tx_err_dst_release:
771 	dst_release(ndst);
772 	return err;
773 }
774 
775 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
776 {
777 	struct ip6_tnl *t = netdev_priv(dev);
778 	const struct iphdr  *iph = ip_hdr(skb);
779 	int encap_limit = -1;
780 	struct flowi6 fl6;
781 	__u8 dsfield;
782 	__u32 mtu;
783 	int err;
784 
785 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
786 		encap_limit = t->parms.encap_limit;
787 
788 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
789 	fl6.flowi6_proto = IPPROTO_GRE;
790 
791 	dsfield = ipv4_get_dsfield(iph);
792 
793 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
794 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
795 					  & IPV6_TCLASS_MASK;
796 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
797 		fl6.flowi6_mark = skb->mark;
798 
799 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
800 	if (err != 0) {
801 		/* XXX: send ICMP error even if DF is not set. */
802 		if (err == -EMSGSIZE)
803 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
804 				  htonl(mtu));
805 		return -1;
806 	}
807 
808 	return 0;
809 }
810 
811 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
812 {
813 	struct ip6_tnl *t = netdev_priv(dev);
814 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
815 	int encap_limit = -1;
816 	__u16 offset;
817 	struct flowi6 fl6;
818 	__u8 dsfield;
819 	__u32 mtu;
820 	int err;
821 
822 	if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
823 		return -1;
824 
825 	offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
826 	if (offset > 0) {
827 		struct ipv6_tlv_tnl_enc_lim *tel;
828 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
829 		if (tel->encap_limit == 0) {
830 			icmpv6_send(skb, ICMPV6_PARAMPROB,
831 				    ICMPV6_HDR_FIELD, offset + 2);
832 			return -1;
833 		}
834 		encap_limit = tel->encap_limit - 1;
835 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
836 		encap_limit = t->parms.encap_limit;
837 
838 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
839 	fl6.flowi6_proto = IPPROTO_GRE;
840 
841 	dsfield = ipv6_get_dsfield(ipv6h);
842 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
843 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
844 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
845 		fl6.flowlabel |= ip6_flowlabel(ipv6h);
846 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
847 		fl6.flowi6_mark = skb->mark;
848 
849 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
850 	if (err != 0) {
851 		if (err == -EMSGSIZE)
852 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
853 		return -1;
854 	}
855 
856 	return 0;
857 }
858 
859 /**
860  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
861  *   @t: the outgoing tunnel device
862  *   @hdr: IPv6 header from the incoming packet
863  *
864  * Description:
865  *   Avoid trivial tunneling loop by checking that tunnel exit-point
866  *   doesn't match source of incoming packet.
867  *
868  * Return:
869  *   1 if conflict,
870  *   0 else
871  **/
872 
873 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
874 	const struct ipv6hdr *hdr)
875 {
876 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
877 }
878 
879 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
880 {
881 	struct ip6_tnl *t = netdev_priv(dev);
882 	int encap_limit = -1;
883 	struct flowi6 fl6;
884 	__u32 mtu;
885 	int err;
886 
887 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
888 		encap_limit = t->parms.encap_limit;
889 
890 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
891 	fl6.flowi6_proto = skb->protocol;
892 
893 	err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
894 
895 	return err;
896 }
897 
898 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
899 	struct net_device *dev)
900 {
901 	struct ip6_tnl *t = netdev_priv(dev);
902 	struct net_device_stats *stats = &t->dev->stats;
903 	int ret;
904 
905 	if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
906 		goto tx_err;
907 
908 	switch (skb->protocol) {
909 	case htons(ETH_P_IP):
910 		ret = ip6gre_xmit_ipv4(skb, dev);
911 		break;
912 	case htons(ETH_P_IPV6):
913 		ret = ip6gre_xmit_ipv6(skb, dev);
914 		break;
915 	default:
916 		ret = ip6gre_xmit_other(skb, dev);
917 		break;
918 	}
919 
920 	if (ret < 0)
921 		goto tx_err;
922 
923 	return NETDEV_TX_OK;
924 
925 tx_err:
926 	stats->tx_errors++;
927 	stats->tx_dropped++;
928 	kfree_skb(skb);
929 	return NETDEV_TX_OK;
930 }
931 
932 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
933 {
934 	struct net_device *dev = t->dev;
935 	struct __ip6_tnl_parm *p = &t->parms;
936 	struct flowi6 *fl6 = &t->fl.u.ip6;
937 	int addend = sizeof(struct ipv6hdr) + 4;
938 
939 	if (dev->type != ARPHRD_ETHER) {
940 		memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
941 		memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
942 	}
943 
944 	/* Set up flowi template */
945 	fl6->saddr = p->laddr;
946 	fl6->daddr = p->raddr;
947 	fl6->flowi6_oif = p->link;
948 	fl6->flowlabel = 0;
949 
950 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
951 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
952 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
953 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
954 
955 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
956 	p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
957 
958 	if (p->flags&IP6_TNL_F_CAP_XMIT &&
959 			p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
960 		dev->flags |= IFF_POINTOPOINT;
961 	else
962 		dev->flags &= ~IFF_POINTOPOINT;
963 
964 	/* Precalculate GRE options length */
965 	if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
966 		if (t->parms.o_flags&GRE_CSUM)
967 			addend += 4;
968 		if (t->parms.o_flags&GRE_KEY)
969 			addend += 4;
970 		if (t->parms.o_flags&GRE_SEQ)
971 			addend += 4;
972 	}
973 	t->hlen = addend;
974 
975 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
976 		int strict = (ipv6_addr_type(&p->raddr) &
977 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
978 
979 		struct rt6_info *rt = rt6_lookup(t->net,
980 						 &p->raddr, &p->laddr,
981 						 p->link, strict);
982 
983 		if (rt == NULL)
984 			return;
985 
986 		if (rt->dst.dev) {
987 			dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
988 
989 			if (set_mtu) {
990 				dev->mtu = rt->dst.dev->mtu - addend;
991 				if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
992 					dev->mtu -= 8;
993 
994 				if (dev->mtu < IPV6_MIN_MTU)
995 					dev->mtu = IPV6_MIN_MTU;
996 			}
997 		}
998 		ip6_rt_put(rt);
999 	}
1000 }
1001 
1002 static int ip6gre_tnl_change(struct ip6_tnl *t,
1003 	const struct __ip6_tnl_parm *p, int set_mtu)
1004 {
1005 	t->parms.laddr = p->laddr;
1006 	t->parms.raddr = p->raddr;
1007 	t->parms.flags = p->flags;
1008 	t->parms.hop_limit = p->hop_limit;
1009 	t->parms.encap_limit = p->encap_limit;
1010 	t->parms.flowinfo = p->flowinfo;
1011 	t->parms.link = p->link;
1012 	t->parms.proto = p->proto;
1013 	t->parms.i_key = p->i_key;
1014 	t->parms.o_key = p->o_key;
1015 	t->parms.i_flags = p->i_flags;
1016 	t->parms.o_flags = p->o_flags;
1017 	ip6_tnl_dst_reset(t);
1018 	ip6gre_tnl_link_config(t, set_mtu);
1019 	return 0;
1020 }
1021 
1022 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1023 	const struct ip6_tnl_parm2 *u)
1024 {
1025 	p->laddr = u->laddr;
1026 	p->raddr = u->raddr;
1027 	p->flags = u->flags;
1028 	p->hop_limit = u->hop_limit;
1029 	p->encap_limit = u->encap_limit;
1030 	p->flowinfo = u->flowinfo;
1031 	p->link = u->link;
1032 	p->i_key = u->i_key;
1033 	p->o_key = u->o_key;
1034 	p->i_flags = u->i_flags;
1035 	p->o_flags = u->o_flags;
1036 	memcpy(p->name, u->name, sizeof(u->name));
1037 }
1038 
1039 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1040 	const struct __ip6_tnl_parm *p)
1041 {
1042 	u->proto = IPPROTO_GRE;
1043 	u->laddr = p->laddr;
1044 	u->raddr = p->raddr;
1045 	u->flags = p->flags;
1046 	u->hop_limit = p->hop_limit;
1047 	u->encap_limit = p->encap_limit;
1048 	u->flowinfo = p->flowinfo;
1049 	u->link = p->link;
1050 	u->i_key = p->i_key;
1051 	u->o_key = p->o_key;
1052 	u->i_flags = p->i_flags;
1053 	u->o_flags = p->o_flags;
1054 	memcpy(u->name, p->name, sizeof(u->name));
1055 }
1056 
1057 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1058 	struct ifreq *ifr, int cmd)
1059 {
1060 	int err = 0;
1061 	struct ip6_tnl_parm2 p;
1062 	struct __ip6_tnl_parm p1;
1063 	struct ip6_tnl *t = netdev_priv(dev);
1064 	struct net *net = t->net;
1065 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1066 
1067 	switch (cmd) {
1068 	case SIOCGETTUNNEL:
1069 		if (dev == ign->fb_tunnel_dev) {
1070 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1071 				err = -EFAULT;
1072 				break;
1073 			}
1074 			ip6gre_tnl_parm_from_user(&p1, &p);
1075 			t = ip6gre_tunnel_locate(net, &p1, 0);
1076 			if (t == NULL)
1077 				t = netdev_priv(dev);
1078 		}
1079 		memset(&p, 0, sizeof(p));
1080 		ip6gre_tnl_parm_to_user(&p, &t->parms);
1081 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1082 			err = -EFAULT;
1083 		break;
1084 
1085 	case SIOCADDTUNNEL:
1086 	case SIOCCHGTUNNEL:
1087 		err = -EPERM;
1088 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1089 			goto done;
1090 
1091 		err = -EFAULT;
1092 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1093 			goto done;
1094 
1095 		err = -EINVAL;
1096 		if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1097 			goto done;
1098 
1099 		if (!(p.i_flags&GRE_KEY))
1100 			p.i_key = 0;
1101 		if (!(p.o_flags&GRE_KEY))
1102 			p.o_key = 0;
1103 
1104 		ip6gre_tnl_parm_from_user(&p1, &p);
1105 		t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1106 
1107 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1108 			if (t != NULL) {
1109 				if (t->dev != dev) {
1110 					err = -EEXIST;
1111 					break;
1112 				}
1113 			} else {
1114 				t = netdev_priv(dev);
1115 
1116 				ip6gre_tunnel_unlink(ign, t);
1117 				synchronize_net();
1118 				ip6gre_tnl_change(t, &p1, 1);
1119 				ip6gre_tunnel_link(ign, t);
1120 				netdev_state_change(dev);
1121 			}
1122 		}
1123 
1124 		if (t) {
1125 			err = 0;
1126 
1127 			memset(&p, 0, sizeof(p));
1128 			ip6gre_tnl_parm_to_user(&p, &t->parms);
1129 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1130 				err = -EFAULT;
1131 		} else
1132 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1133 		break;
1134 
1135 	case SIOCDELTUNNEL:
1136 		err = -EPERM;
1137 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1138 			goto done;
1139 
1140 		if (dev == ign->fb_tunnel_dev) {
1141 			err = -EFAULT;
1142 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1143 				goto done;
1144 			err = -ENOENT;
1145 			ip6gre_tnl_parm_from_user(&p1, &p);
1146 			t = ip6gre_tunnel_locate(net, &p1, 0);
1147 			if (t == NULL)
1148 				goto done;
1149 			err = -EPERM;
1150 			if (t == netdev_priv(ign->fb_tunnel_dev))
1151 				goto done;
1152 			dev = t->dev;
1153 		}
1154 		unregister_netdevice(dev);
1155 		err = 0;
1156 		break;
1157 
1158 	default:
1159 		err = -EINVAL;
1160 	}
1161 
1162 done:
1163 	return err;
1164 }
1165 
1166 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1167 {
1168 	if (new_mtu < 68 ||
1169 	    new_mtu > 0xFFF8 - dev->hard_header_len)
1170 		return -EINVAL;
1171 	dev->mtu = new_mtu;
1172 	return 0;
1173 }
1174 
1175 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1176 			unsigned short type,
1177 			const void *daddr, const void *saddr, unsigned int len)
1178 {
1179 	struct ip6_tnl *t = netdev_priv(dev);
1180 	struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1181 	__be16 *p = (__be16 *)(ipv6h+1);
1182 
1183 	ip6_flow_hdr(ipv6h, 0,
1184 		     ip6_make_flowlabel(dev_net(dev), skb,
1185 					t->fl.u.ip6.flowlabel, false));
1186 	ipv6h->hop_limit = t->parms.hop_limit;
1187 	ipv6h->nexthdr = NEXTHDR_GRE;
1188 	ipv6h->saddr = t->parms.laddr;
1189 	ipv6h->daddr = t->parms.raddr;
1190 
1191 	p[0]		= t->parms.o_flags;
1192 	p[1]		= htons(type);
1193 
1194 	/*
1195 	 *	Set the source hardware address.
1196 	 */
1197 
1198 	if (saddr)
1199 		memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1200 	if (daddr)
1201 		memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1202 	if (!ipv6_addr_any(&ipv6h->daddr))
1203 		return t->hlen;
1204 
1205 	return -t->hlen;
1206 }
1207 
1208 static const struct header_ops ip6gre_header_ops = {
1209 	.create	= ip6gre_header,
1210 };
1211 
1212 static const struct net_device_ops ip6gre_netdev_ops = {
1213 	.ndo_init		= ip6gre_tunnel_init,
1214 	.ndo_uninit		= ip6gre_tunnel_uninit,
1215 	.ndo_start_xmit		= ip6gre_tunnel_xmit,
1216 	.ndo_do_ioctl		= ip6gre_tunnel_ioctl,
1217 	.ndo_change_mtu		= ip6gre_tunnel_change_mtu,
1218 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1219 };
1220 
1221 static void ip6gre_dev_free(struct net_device *dev)
1222 {
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->iflink = 0;
1242 	dev->addr_len = sizeof(struct in6_addr);
1243 	netif_keep_dst(dev);
1244 }
1245 
1246 static int ip6gre_tunnel_init(struct net_device *dev)
1247 {
1248 	struct ip6_tnl *tunnel;
1249 	int i;
1250 
1251 	tunnel = netdev_priv(dev);
1252 
1253 	tunnel->dev = dev;
1254 	tunnel->net = dev_net(dev);
1255 	strcpy(tunnel->parms.name, dev->name);
1256 
1257 	memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1258 	memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1259 
1260 	if (ipv6_addr_any(&tunnel->parms.raddr))
1261 		dev->header_ops = &ip6gre_header_ops;
1262 
1263 	dev->tstats = alloc_percpu(struct pcpu_sw_netstats);
1264 	if (!dev->tstats)
1265 		return -ENOMEM;
1266 
1267 	for_each_possible_cpu(i) {
1268 		struct pcpu_sw_netstats *ip6gre_tunnel_stats;
1269 		ip6gre_tunnel_stats = per_cpu_ptr(dev->tstats, i);
1270 		u64_stats_init(&ip6gre_tunnel_stats->syncp);
1271 	}
1272 
1273 	dev->iflink = tunnel->parms.link;
1274 
1275 	return 0;
1276 }
1277 
1278 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1279 {
1280 	struct ip6_tnl *tunnel = netdev_priv(dev);
1281 
1282 	tunnel->dev = dev;
1283 	tunnel->net = dev_net(dev);
1284 	strcpy(tunnel->parms.name, dev->name);
1285 
1286 	tunnel->hlen		= sizeof(struct ipv6hdr) + 4;
1287 
1288 	dev_hold(dev);
1289 }
1290 
1291 
1292 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1293 	.handler     = ip6gre_rcv,
1294 	.err_handler = ip6gre_err,
1295 	.flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1296 };
1297 
1298 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1299 {
1300 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1301 	struct net_device *dev, *aux;
1302 	int prio;
1303 
1304 	for_each_netdev_safe(net, dev, aux)
1305 		if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1306 		    dev->rtnl_link_ops == &ip6gre_tap_ops)
1307 			unregister_netdevice_queue(dev, head);
1308 
1309 	for (prio = 0; prio < 4; prio++) {
1310 		int h;
1311 		for (h = 0; h < HASH_SIZE; h++) {
1312 			struct ip6_tnl *t;
1313 
1314 			t = rtnl_dereference(ign->tunnels[prio][h]);
1315 
1316 			while (t != NULL) {
1317 				/* If dev is in the same netns, it has already
1318 				 * been added to the list by the previous loop.
1319 				 */
1320 				if (!net_eq(dev_net(t->dev), net))
1321 					unregister_netdevice_queue(t->dev,
1322 								   head);
1323 				t = rtnl_dereference(t->next);
1324 			}
1325 		}
1326 	}
1327 }
1328 
1329 static int __net_init ip6gre_init_net(struct net *net)
1330 {
1331 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1332 	int err;
1333 
1334 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1335 					  NET_NAME_UNKNOWN,
1336 					  ip6gre_tunnel_setup);
1337 	if (!ign->fb_tunnel_dev) {
1338 		err = -ENOMEM;
1339 		goto err_alloc_dev;
1340 	}
1341 	dev_net_set(ign->fb_tunnel_dev, net);
1342 	/* FB netdevice is special: we have one, and only one per netns.
1343 	 * Allowing to move it to another netns is clearly unsafe.
1344 	 */
1345 	ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1346 
1347 
1348 	ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1349 	ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1350 
1351 	err = register_netdev(ign->fb_tunnel_dev);
1352 	if (err)
1353 		goto err_reg_dev;
1354 
1355 	rcu_assign_pointer(ign->tunnels_wc[0],
1356 			   netdev_priv(ign->fb_tunnel_dev));
1357 	return 0;
1358 
1359 err_reg_dev:
1360 	ip6gre_dev_free(ign->fb_tunnel_dev);
1361 err_alloc_dev:
1362 	return err;
1363 }
1364 
1365 static void __net_exit ip6gre_exit_net(struct net *net)
1366 {
1367 	LIST_HEAD(list);
1368 
1369 	rtnl_lock();
1370 	ip6gre_destroy_tunnels(net, &list);
1371 	unregister_netdevice_many(&list);
1372 	rtnl_unlock();
1373 }
1374 
1375 static struct pernet_operations ip6gre_net_ops = {
1376 	.init = ip6gre_init_net,
1377 	.exit = ip6gre_exit_net,
1378 	.id   = &ip6gre_net_id,
1379 	.size = sizeof(struct ip6gre_net),
1380 };
1381 
1382 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1383 {
1384 	__be16 flags;
1385 
1386 	if (!data)
1387 		return 0;
1388 
1389 	flags = 0;
1390 	if (data[IFLA_GRE_IFLAGS])
1391 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1392 	if (data[IFLA_GRE_OFLAGS])
1393 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1394 	if (flags & (GRE_VERSION|GRE_ROUTING))
1395 		return -EINVAL;
1396 
1397 	return 0;
1398 }
1399 
1400 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1401 {
1402 	struct in6_addr daddr;
1403 
1404 	if (tb[IFLA_ADDRESS]) {
1405 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1406 			return -EINVAL;
1407 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1408 			return -EADDRNOTAVAIL;
1409 	}
1410 
1411 	if (!data)
1412 		goto out;
1413 
1414 	if (data[IFLA_GRE_REMOTE]) {
1415 		nla_memcpy(&daddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1416 		if (ipv6_addr_any(&daddr))
1417 			return -EINVAL;
1418 	}
1419 
1420 out:
1421 	return ip6gre_tunnel_validate(tb, data);
1422 }
1423 
1424 
1425 static void ip6gre_netlink_parms(struct nlattr *data[],
1426 				struct __ip6_tnl_parm *parms)
1427 {
1428 	memset(parms, 0, sizeof(*parms));
1429 
1430 	if (!data)
1431 		return;
1432 
1433 	if (data[IFLA_GRE_LINK])
1434 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1435 
1436 	if (data[IFLA_GRE_IFLAGS])
1437 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1438 
1439 	if (data[IFLA_GRE_OFLAGS])
1440 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1441 
1442 	if (data[IFLA_GRE_IKEY])
1443 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1444 
1445 	if (data[IFLA_GRE_OKEY])
1446 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1447 
1448 	if (data[IFLA_GRE_LOCAL])
1449 		nla_memcpy(&parms->laddr, data[IFLA_GRE_LOCAL], sizeof(struct in6_addr));
1450 
1451 	if (data[IFLA_GRE_REMOTE])
1452 		nla_memcpy(&parms->raddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1453 
1454 	if (data[IFLA_GRE_TTL])
1455 		parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1456 
1457 	if (data[IFLA_GRE_ENCAP_LIMIT])
1458 		parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1459 
1460 	if (data[IFLA_GRE_FLOWINFO])
1461 		parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1462 
1463 	if (data[IFLA_GRE_FLAGS])
1464 		parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1465 }
1466 
1467 static int ip6gre_tap_init(struct net_device *dev)
1468 {
1469 	struct ip6_tnl *tunnel;
1470 
1471 	tunnel = netdev_priv(dev);
1472 
1473 	tunnel->dev = dev;
1474 	tunnel->net = dev_net(dev);
1475 	strcpy(tunnel->parms.name, dev->name);
1476 
1477 	ip6gre_tnl_link_config(tunnel, 1);
1478 
1479 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1480 	if (!dev->tstats)
1481 		return -ENOMEM;
1482 
1483 	dev->iflink = tunnel->parms.link;
1484 
1485 	return 0;
1486 }
1487 
1488 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1489 	.ndo_init = ip6gre_tap_init,
1490 	.ndo_uninit = ip6gre_tunnel_uninit,
1491 	.ndo_start_xmit = ip6gre_tunnel_xmit,
1492 	.ndo_set_mac_address = eth_mac_addr,
1493 	.ndo_validate_addr = eth_validate_addr,
1494 	.ndo_change_mtu = ip6gre_tunnel_change_mtu,
1495 	.ndo_get_stats64 = ip_tunnel_get_stats64,
1496 };
1497 
1498 static void ip6gre_tap_setup(struct net_device *dev)
1499 {
1500 
1501 	ether_setup(dev);
1502 
1503 	dev->netdev_ops = &ip6gre_tap_netdev_ops;
1504 	dev->destructor = ip6gre_dev_free;
1505 
1506 	dev->iflink = 0;
1507 	dev->features |= NETIF_F_NETNS_LOCAL;
1508 }
1509 
1510 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1511 	struct nlattr *tb[], struct nlattr *data[])
1512 {
1513 	struct ip6_tnl *nt;
1514 	struct net *net = dev_net(dev);
1515 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1516 	int err;
1517 
1518 	nt = netdev_priv(dev);
1519 	ip6gre_netlink_parms(data, &nt->parms);
1520 
1521 	if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1522 		return -EEXIST;
1523 
1524 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1525 		eth_hw_addr_random(dev);
1526 
1527 	nt->dev = dev;
1528 	nt->net = dev_net(dev);
1529 	ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1530 
1531 	/* Can use a lockless transmit, unless we generate output sequences */
1532 	if (!(nt->parms.o_flags & GRE_SEQ))
1533 		dev->features |= NETIF_F_LLTX;
1534 
1535 	err = register_netdevice(dev);
1536 	if (err)
1537 		goto out;
1538 
1539 	dev_hold(dev);
1540 	ip6gre_tunnel_link(ign, nt);
1541 
1542 out:
1543 	return err;
1544 }
1545 
1546 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1547 			    struct nlattr *data[])
1548 {
1549 	struct ip6_tnl *t, *nt = netdev_priv(dev);
1550 	struct net *net = nt->net;
1551 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1552 	struct __ip6_tnl_parm p;
1553 
1554 	if (dev == ign->fb_tunnel_dev)
1555 		return -EINVAL;
1556 
1557 	ip6gre_netlink_parms(data, &p);
1558 
1559 	t = ip6gre_tunnel_locate(net, &p, 0);
1560 
1561 	if (t) {
1562 		if (t->dev != dev)
1563 			return -EEXIST;
1564 	} else {
1565 		t = nt;
1566 
1567 		ip6gre_tunnel_unlink(ign, t);
1568 		ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1569 		ip6gre_tunnel_link(ign, t);
1570 		netdev_state_change(dev);
1571 	}
1572 
1573 	return 0;
1574 }
1575 
1576 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1577 {
1578 	struct net *net = dev_net(dev);
1579 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1580 
1581 	if (dev != ign->fb_tunnel_dev)
1582 		unregister_netdevice_queue(dev, head);
1583 }
1584 
1585 static size_t ip6gre_get_size(const struct net_device *dev)
1586 {
1587 	return
1588 		/* IFLA_GRE_LINK */
1589 		nla_total_size(4) +
1590 		/* IFLA_GRE_IFLAGS */
1591 		nla_total_size(2) +
1592 		/* IFLA_GRE_OFLAGS */
1593 		nla_total_size(2) +
1594 		/* IFLA_GRE_IKEY */
1595 		nla_total_size(4) +
1596 		/* IFLA_GRE_OKEY */
1597 		nla_total_size(4) +
1598 		/* IFLA_GRE_LOCAL */
1599 		nla_total_size(sizeof(struct in6_addr)) +
1600 		/* IFLA_GRE_REMOTE */
1601 		nla_total_size(sizeof(struct in6_addr)) +
1602 		/* IFLA_GRE_TTL */
1603 		nla_total_size(1) +
1604 		/* IFLA_GRE_TOS */
1605 		nla_total_size(1) +
1606 		/* IFLA_GRE_ENCAP_LIMIT */
1607 		nla_total_size(1) +
1608 		/* IFLA_GRE_FLOWINFO */
1609 		nla_total_size(4) +
1610 		/* IFLA_GRE_FLAGS */
1611 		nla_total_size(4) +
1612 		0;
1613 }
1614 
1615 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1616 {
1617 	struct ip6_tnl *t = netdev_priv(dev);
1618 	struct __ip6_tnl_parm *p = &t->parms;
1619 
1620 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1621 	    nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1622 	    nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1623 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1624 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1625 	    nla_put(skb, IFLA_GRE_LOCAL, sizeof(struct in6_addr), &p->laddr) ||
1626 	    nla_put(skb, IFLA_GRE_REMOTE, sizeof(struct in6_addr), &p->raddr) ||
1627 	    nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1628 	    /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1629 	    nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1630 	    nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1631 	    nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1632 		goto nla_put_failure;
1633 	return 0;
1634 
1635 nla_put_failure:
1636 	return -EMSGSIZE;
1637 }
1638 
1639 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1640 	[IFLA_GRE_LINK]        = { .type = NLA_U32 },
1641 	[IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1642 	[IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1643 	[IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1644 	[IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1645 	[IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1646 	[IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1647 	[IFLA_GRE_TTL]         = { .type = NLA_U8 },
1648 	[IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1649 	[IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1650 	[IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1651 };
1652 
1653 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1654 	.kind		= "ip6gre",
1655 	.maxtype	= IFLA_GRE_MAX,
1656 	.policy		= ip6gre_policy,
1657 	.priv_size	= sizeof(struct ip6_tnl),
1658 	.setup		= ip6gre_tunnel_setup,
1659 	.validate	= ip6gre_tunnel_validate,
1660 	.newlink	= ip6gre_newlink,
1661 	.changelink	= ip6gre_changelink,
1662 	.dellink	= ip6gre_dellink,
1663 	.get_size	= ip6gre_get_size,
1664 	.fill_info	= ip6gre_fill_info,
1665 	.get_link_net	= ip6_tnl_get_link_net,
1666 };
1667 
1668 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1669 	.kind		= "ip6gretap",
1670 	.maxtype	= IFLA_GRE_MAX,
1671 	.policy		= ip6gre_policy,
1672 	.priv_size	= sizeof(struct ip6_tnl),
1673 	.setup		= ip6gre_tap_setup,
1674 	.validate	= ip6gre_tap_validate,
1675 	.newlink	= ip6gre_newlink,
1676 	.changelink	= ip6gre_changelink,
1677 	.get_size	= ip6gre_get_size,
1678 	.fill_info	= ip6gre_fill_info,
1679 	.get_link_net	= ip6_tnl_get_link_net,
1680 };
1681 
1682 /*
1683  *	And now the modules code and kernel interface.
1684  */
1685 
1686 static int __init ip6gre_init(void)
1687 {
1688 	int err;
1689 
1690 	pr_info("GRE over IPv6 tunneling driver\n");
1691 
1692 	err = register_pernet_device(&ip6gre_net_ops);
1693 	if (err < 0)
1694 		return err;
1695 
1696 	err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1697 	if (err < 0) {
1698 		pr_info("%s: can't add protocol\n", __func__);
1699 		goto add_proto_failed;
1700 	}
1701 
1702 	err = rtnl_link_register(&ip6gre_link_ops);
1703 	if (err < 0)
1704 		goto rtnl_link_failed;
1705 
1706 	err = rtnl_link_register(&ip6gre_tap_ops);
1707 	if (err < 0)
1708 		goto tap_ops_failed;
1709 
1710 out:
1711 	return err;
1712 
1713 tap_ops_failed:
1714 	rtnl_link_unregister(&ip6gre_link_ops);
1715 rtnl_link_failed:
1716 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1717 add_proto_failed:
1718 	unregister_pernet_device(&ip6gre_net_ops);
1719 	goto out;
1720 }
1721 
1722 static void __exit ip6gre_fini(void)
1723 {
1724 	rtnl_link_unregister(&ip6gre_tap_ops);
1725 	rtnl_link_unregister(&ip6gre_link_ops);
1726 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1727 	unregister_pernet_device(&ip6gre_net_ops);
1728 }
1729 
1730 module_init(ip6gre_init);
1731 module_exit(ip6gre_fini);
1732 MODULE_LICENSE("GPL");
1733 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1734 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1735 MODULE_ALIAS_RTNL_LINK("ip6gre");
1736 MODULE_ALIAS_RTNL_LINK("ip6gretap");
1737 MODULE_ALIAS_NETDEV("ip6gre0");
1738