xref: /openbmc/linux/net/ipv6/ip6_gre.c (revision db181ce0)
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 t;
319 
320 	if (parms->name[0])
321 		strlcpy(name, parms->name, IFNAMSIZ);
322 	else
323 		strcpy(name, "ip6gre%d");
324 
325 	dev = alloc_netdev(sizeof(*t), name, ip6gre_tunnel_setup);
326 	if (!dev)
327 		return NULL;
328 
329 	dev_net_set(dev, net);
330 
331 	nt = netdev_priv(dev);
332 	nt->parms = *parms;
333 	dev->rtnl_link_ops = &ip6gre_link_ops;
334 
335 	nt->dev = dev;
336 	nt->net = dev_net(dev);
337 	ip6gre_tnl_link_config(nt, 1);
338 
339 	if (register_netdevice(dev) < 0)
340 		goto failed_free;
341 
342 	/* Can use a lockless transmit, unless we generate output sequences */
343 	if (!(nt->parms.o_flags & GRE_SEQ))
344 		dev->features |= NETIF_F_LLTX;
345 
346 	dev_hold(dev);
347 	ip6gre_tunnel_link(ign, nt);
348 	return nt;
349 
350 failed_free:
351 	free_netdev(dev);
352 	return NULL;
353 }
354 
355 static void ip6gre_tunnel_uninit(struct net_device *dev)
356 {
357 	struct ip6_tnl *t = netdev_priv(dev);
358 	struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
359 
360 	ip6gre_tunnel_unlink(ign, t);
361 	dev_put(dev);
362 }
363 
364 
365 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
366 		u8 type, u8 code, int offset, __be32 info)
367 {
368 	const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
369 	__be16 *p = (__be16 *)(skb->data + offset);
370 	int grehlen = offset + 4;
371 	struct ip6_tnl *t;
372 	__be16 flags;
373 
374 	flags = p[0];
375 	if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
376 		if (flags&(GRE_VERSION|GRE_ROUTING))
377 			return;
378 		if (flags&GRE_KEY) {
379 			grehlen += 4;
380 			if (flags&GRE_CSUM)
381 				grehlen += 4;
382 		}
383 	}
384 
385 	/* If only 8 bytes returned, keyed message will be dropped here */
386 	if (!pskb_may_pull(skb, grehlen))
387 		return;
388 	ipv6h = (const struct ipv6hdr *)skb->data;
389 	p = (__be16 *)(skb->data + offset);
390 
391 	t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
392 				flags & GRE_KEY ?
393 				*(((__be32 *)p) + (grehlen / 4) - 1) : 0,
394 				p[1]);
395 	if (t == NULL)
396 		return;
397 
398 	switch (type) {
399 		__u32 teli;
400 		struct ipv6_tlv_tnl_enc_lim *tel;
401 		__u32 mtu;
402 	case ICMPV6_DEST_UNREACH:
403 		net_warn_ratelimited("%s: Path to destination invalid or inactive!\n",
404 				     t->parms.name);
405 		break;
406 	case ICMPV6_TIME_EXCEED:
407 		if (code == ICMPV6_EXC_HOPLIMIT) {
408 			net_warn_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
409 					     t->parms.name);
410 		}
411 		break;
412 	case ICMPV6_PARAMPROB:
413 		teli = 0;
414 		if (code == ICMPV6_HDR_FIELD)
415 			teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
416 
417 		if (teli && teli == info - 2) {
418 			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
419 			if (tel->encap_limit == 0) {
420 				net_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
421 						     t->parms.name);
422 			}
423 		} else {
424 			net_warn_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
425 					     t->parms.name);
426 		}
427 		break;
428 	case ICMPV6_PKT_TOOBIG:
429 		mtu = info - offset;
430 		if (mtu < IPV6_MIN_MTU)
431 			mtu = IPV6_MIN_MTU;
432 		t->dev->mtu = mtu;
433 		break;
434 	}
435 
436 	if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
437 		t->err_count++;
438 	else
439 		t->err_count = 1;
440 	t->err_time = jiffies;
441 }
442 
443 static int ip6gre_rcv(struct sk_buff *skb)
444 {
445 	const struct ipv6hdr *ipv6h;
446 	u8     *h;
447 	__be16    flags;
448 	__sum16   csum = 0;
449 	__be32 key = 0;
450 	u32    seqno = 0;
451 	struct ip6_tnl *tunnel;
452 	int    offset = 4;
453 	__be16 gre_proto;
454 	int err;
455 
456 	if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
457 		goto drop;
458 
459 	ipv6h = ipv6_hdr(skb);
460 	h = skb->data;
461 	flags = *(__be16 *)h;
462 
463 	if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
464 		/* - Version must be 0.
465 		   - We do not support routing headers.
466 		 */
467 		if (flags&(GRE_VERSION|GRE_ROUTING))
468 			goto drop;
469 
470 		if (flags&GRE_CSUM) {
471 			csum = skb_checksum_simple_validate(skb);
472 			offset += 4;
473 		}
474 		if (flags&GRE_KEY) {
475 			key = *(__be32 *)(h + offset);
476 			offset += 4;
477 		}
478 		if (flags&GRE_SEQ) {
479 			seqno = ntohl(*(__be32 *)(h + offset));
480 			offset += 4;
481 		}
482 	}
483 
484 	gre_proto = *(__be16 *)(h + 2);
485 
486 	tunnel = ip6gre_tunnel_lookup(skb->dev,
487 					  &ipv6h->saddr, &ipv6h->daddr, key,
488 					  gre_proto);
489 	if (tunnel) {
490 		struct pcpu_sw_netstats *tstats;
491 
492 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
493 			goto drop;
494 
495 		if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
496 			tunnel->dev->stats.rx_dropped++;
497 			goto drop;
498 		}
499 
500 		skb->protocol = gre_proto;
501 		/* WCCP version 1 and 2 protocol decoding.
502 		 * - Change protocol to IP
503 		 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
504 		 */
505 		if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
506 			skb->protocol = htons(ETH_P_IP);
507 			if ((*(h + offset) & 0xF0) != 0x40)
508 				offset += 4;
509 		}
510 
511 		skb->mac_header = skb->network_header;
512 		__pskb_pull(skb, offset);
513 		skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
514 
515 		if (((flags&GRE_CSUM) && csum) ||
516 		    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
517 			tunnel->dev->stats.rx_crc_errors++;
518 			tunnel->dev->stats.rx_errors++;
519 			goto drop;
520 		}
521 		if (tunnel->parms.i_flags&GRE_SEQ) {
522 			if (!(flags&GRE_SEQ) ||
523 			    (tunnel->i_seqno &&
524 					(s32)(seqno - tunnel->i_seqno) < 0)) {
525 				tunnel->dev->stats.rx_fifo_errors++;
526 				tunnel->dev->stats.rx_errors++;
527 				goto drop;
528 			}
529 			tunnel->i_seqno = seqno + 1;
530 		}
531 
532 		/* Warning: All skb pointers will be invalidated! */
533 		if (tunnel->dev->type == ARPHRD_ETHER) {
534 			if (!pskb_may_pull(skb, ETH_HLEN)) {
535 				tunnel->dev->stats.rx_length_errors++;
536 				tunnel->dev->stats.rx_errors++;
537 				goto drop;
538 			}
539 
540 			ipv6h = ipv6_hdr(skb);
541 			skb->protocol = eth_type_trans(skb, tunnel->dev);
542 			skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
543 		}
544 
545 		__skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
546 
547 		skb_reset_network_header(skb);
548 
549 		err = IP6_ECN_decapsulate(ipv6h, skb);
550 		if (unlikely(err)) {
551 			if (log_ecn_error)
552 				net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
553 						     &ipv6h->saddr,
554 						     ipv6_get_dsfield(ipv6h));
555 			if (err > 1) {
556 				++tunnel->dev->stats.rx_frame_errors;
557 				++tunnel->dev->stats.rx_errors;
558 				goto drop;
559 			}
560 		}
561 
562 		tstats = this_cpu_ptr(tunnel->dev->tstats);
563 		u64_stats_update_begin(&tstats->syncp);
564 		tstats->rx_packets++;
565 		tstats->rx_bytes += skb->len;
566 		u64_stats_update_end(&tstats->syncp);
567 
568 		netif_rx(skb);
569 
570 		return 0;
571 	}
572 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
573 
574 drop:
575 	kfree_skb(skb);
576 	return 0;
577 }
578 
579 struct ipv6_tel_txoption {
580 	struct ipv6_txoptions ops;
581 	__u8 dst_opt[8];
582 };
583 
584 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
585 {
586 	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
587 
588 	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
589 	opt->dst_opt[3] = 1;
590 	opt->dst_opt[4] = encap_limit;
591 	opt->dst_opt[5] = IPV6_TLV_PADN;
592 	opt->dst_opt[6] = 1;
593 
594 	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
595 	opt->ops.opt_nflen = 8;
596 }
597 
598 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
599 			 struct net_device *dev,
600 			 __u8 dsfield,
601 			 struct flowi6 *fl6,
602 			 int encap_limit,
603 			 __u32 *pmtu)
604 {
605 	struct ip6_tnl *tunnel = netdev_priv(dev);
606 	struct net *net = tunnel->net;
607 	struct net_device *tdev;    /* Device to other host */
608 	struct ipv6hdr  *ipv6h;     /* Our new IP header */
609 	unsigned int max_headroom = 0; /* The extra header space needed */
610 	int    gre_hlen;
611 	struct ipv6_tel_txoption opt;
612 	int    mtu;
613 	struct dst_entry *dst = NULL, *ndst = NULL;
614 	struct net_device_stats *stats = &tunnel->dev->stats;
615 	int err = -1;
616 	u8 proto;
617 	struct sk_buff *new_skb;
618 
619 	if (dev->type == ARPHRD_ETHER)
620 		IPCB(skb)->flags = 0;
621 
622 	if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
623 		gre_hlen = 0;
624 		ipv6h = (struct ipv6hdr *)skb->data;
625 		fl6->daddr = ipv6h->daddr;
626 	} else {
627 		gre_hlen = tunnel->hlen;
628 		fl6->daddr = tunnel->parms.raddr;
629 	}
630 
631 	if (!fl6->flowi6_mark)
632 		dst = ip6_tnl_dst_check(tunnel);
633 
634 	if (!dst) {
635 		ndst = ip6_route_output(net, NULL, fl6);
636 
637 		if (ndst->error)
638 			goto tx_err_link_failure;
639 		ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
640 		if (IS_ERR(ndst)) {
641 			err = PTR_ERR(ndst);
642 			ndst = NULL;
643 			goto tx_err_link_failure;
644 		}
645 		dst = ndst;
646 	}
647 
648 	tdev = dst->dev;
649 
650 	if (tdev == dev) {
651 		stats->collisions++;
652 		net_warn_ratelimited("%s: Local routing loop detected!\n",
653 				     tunnel->parms.name);
654 		goto tx_err_dst_release;
655 	}
656 
657 	mtu = dst_mtu(dst) - sizeof(*ipv6h);
658 	if (encap_limit >= 0) {
659 		max_headroom += 8;
660 		mtu -= 8;
661 	}
662 	if (mtu < IPV6_MIN_MTU)
663 		mtu = IPV6_MIN_MTU;
664 	if (skb_dst(skb))
665 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
666 	if (skb->len > mtu) {
667 		*pmtu = mtu;
668 		err = -EMSGSIZE;
669 		goto tx_err_dst_release;
670 	}
671 
672 	if (tunnel->err_count > 0) {
673 		if (time_before(jiffies,
674 				tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
675 			tunnel->err_count--;
676 
677 			dst_link_failure(skb);
678 		} else
679 			tunnel->err_count = 0;
680 	}
681 
682 	skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
683 
684 	max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
685 
686 	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
687 	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
688 		new_skb = skb_realloc_headroom(skb, max_headroom);
689 		if (max_headroom > dev->needed_headroom)
690 			dev->needed_headroom = max_headroom;
691 		if (!new_skb)
692 			goto tx_err_dst_release;
693 
694 		if (skb->sk)
695 			skb_set_owner_w(new_skb, skb->sk);
696 		consume_skb(skb);
697 		skb = new_skb;
698 	}
699 
700 	if (fl6->flowi6_mark) {
701 		skb_dst_set(skb, dst);
702 		ndst = NULL;
703 	} else {
704 		skb_dst_set_noref(skb, dst);
705 	}
706 
707 	proto = NEXTHDR_GRE;
708 	if (encap_limit >= 0) {
709 		init_tel_txopt(&opt, encap_limit);
710 		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
711 	}
712 
713 	if (likely(!skb->encapsulation)) {
714 		skb_reset_inner_headers(skb);
715 		skb->encapsulation = 1;
716 	}
717 
718 	skb_push(skb, gre_hlen);
719 	skb_reset_network_header(skb);
720 	skb_set_transport_header(skb, sizeof(*ipv6h));
721 
722 	/*
723 	 *	Push down and install the IP header.
724 	 */
725 	ipv6h = ipv6_hdr(skb);
726 	ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);
727 	ipv6h->hop_limit = tunnel->parms.hop_limit;
728 	ipv6h->nexthdr = proto;
729 	ipv6h->saddr = fl6->saddr;
730 	ipv6h->daddr = fl6->daddr;
731 
732 	((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
733 	((__be16 *)(ipv6h + 1))[1] = (dev->type == ARPHRD_ETHER) ?
734 				   htons(ETH_P_TEB) : skb->protocol;
735 
736 	if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
737 		__be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
738 
739 		if (tunnel->parms.o_flags&GRE_SEQ) {
740 			++tunnel->o_seqno;
741 			*ptr = htonl(tunnel->o_seqno);
742 			ptr--;
743 		}
744 		if (tunnel->parms.o_flags&GRE_KEY) {
745 			*ptr = tunnel->parms.o_key;
746 			ptr--;
747 		}
748 		if (tunnel->parms.o_flags&GRE_CSUM) {
749 			*ptr = 0;
750 			*(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
751 				skb->len - sizeof(struct ipv6hdr));
752 		}
753 	}
754 
755 	ip6tunnel_xmit(skb, dev);
756 	if (ndst)
757 		ip6_tnl_dst_store(tunnel, ndst);
758 	return 0;
759 tx_err_link_failure:
760 	stats->tx_carrier_errors++;
761 	dst_link_failure(skb);
762 tx_err_dst_release:
763 	dst_release(ndst);
764 	return err;
765 }
766 
767 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
768 {
769 	struct ip6_tnl *t = netdev_priv(dev);
770 	const struct iphdr  *iph = ip_hdr(skb);
771 	int encap_limit = -1;
772 	struct flowi6 fl6;
773 	__u8 dsfield;
774 	__u32 mtu;
775 	int err;
776 
777 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
778 		encap_limit = t->parms.encap_limit;
779 
780 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
781 	fl6.flowi6_proto = IPPROTO_IPIP;
782 
783 	dsfield = ipv4_get_dsfield(iph);
784 
785 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
786 		fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
787 					  & IPV6_TCLASS_MASK;
788 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
789 		fl6.flowi6_mark = skb->mark;
790 
791 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
792 	if (err != 0) {
793 		/* XXX: send ICMP error even if DF is not set. */
794 		if (err == -EMSGSIZE)
795 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
796 				  htonl(mtu));
797 		return -1;
798 	}
799 
800 	return 0;
801 }
802 
803 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
804 {
805 	struct ip6_tnl *t = netdev_priv(dev);
806 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
807 	int encap_limit = -1;
808 	__u16 offset;
809 	struct flowi6 fl6;
810 	__u8 dsfield;
811 	__u32 mtu;
812 	int err;
813 
814 	if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
815 		return -1;
816 
817 	offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
818 	if (offset > 0) {
819 		struct ipv6_tlv_tnl_enc_lim *tel;
820 		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
821 		if (tel->encap_limit == 0) {
822 			icmpv6_send(skb, ICMPV6_PARAMPROB,
823 				    ICMPV6_HDR_FIELD, offset + 2);
824 			return -1;
825 		}
826 		encap_limit = tel->encap_limit - 1;
827 	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
828 		encap_limit = t->parms.encap_limit;
829 
830 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
831 	fl6.flowi6_proto = IPPROTO_IPV6;
832 
833 	dsfield = ipv6_get_dsfield(ipv6h);
834 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
835 		fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
836 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
837 		fl6.flowlabel |= ip6_flowlabel(ipv6h);
838 	if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
839 		fl6.flowi6_mark = skb->mark;
840 
841 	err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
842 	if (err != 0) {
843 		if (err == -EMSGSIZE)
844 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
845 		return -1;
846 	}
847 
848 	return 0;
849 }
850 
851 /**
852  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
853  *   @t: the outgoing tunnel device
854  *   @hdr: IPv6 header from the incoming packet
855  *
856  * Description:
857  *   Avoid trivial tunneling loop by checking that tunnel exit-point
858  *   doesn't match source of incoming packet.
859  *
860  * Return:
861  *   1 if conflict,
862  *   0 else
863  **/
864 
865 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
866 	const struct ipv6hdr *hdr)
867 {
868 	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
869 }
870 
871 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
872 {
873 	struct ip6_tnl *t = netdev_priv(dev);
874 	int encap_limit = -1;
875 	struct flowi6 fl6;
876 	__u32 mtu;
877 	int err;
878 
879 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
880 		encap_limit = t->parms.encap_limit;
881 
882 	memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
883 	fl6.flowi6_proto = skb->protocol;
884 
885 	err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
886 
887 	return err;
888 }
889 
890 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
891 	struct net_device *dev)
892 {
893 	struct ip6_tnl *t = netdev_priv(dev);
894 	struct net_device_stats *stats = &t->dev->stats;
895 	int ret;
896 
897 	if (!ip6_tnl_xmit_ctl(t))
898 		goto tx_err;
899 
900 	switch (skb->protocol) {
901 	case htons(ETH_P_IP):
902 		ret = ip6gre_xmit_ipv4(skb, dev);
903 		break;
904 	case htons(ETH_P_IPV6):
905 		ret = ip6gre_xmit_ipv6(skb, dev);
906 		break;
907 	default:
908 		ret = ip6gre_xmit_other(skb, dev);
909 		break;
910 	}
911 
912 	if (ret < 0)
913 		goto tx_err;
914 
915 	return NETDEV_TX_OK;
916 
917 tx_err:
918 	stats->tx_errors++;
919 	stats->tx_dropped++;
920 	kfree_skb(skb);
921 	return NETDEV_TX_OK;
922 }
923 
924 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
925 {
926 	struct net_device *dev = t->dev;
927 	struct __ip6_tnl_parm *p = &t->parms;
928 	struct flowi6 *fl6 = &t->fl.u.ip6;
929 	int addend = sizeof(struct ipv6hdr) + 4;
930 
931 	if (dev->type != ARPHRD_ETHER) {
932 		memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
933 		memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
934 	}
935 
936 	/* Set up flowi template */
937 	fl6->saddr = p->laddr;
938 	fl6->daddr = p->raddr;
939 	fl6->flowi6_oif = p->link;
940 	fl6->flowlabel = 0;
941 
942 	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
943 		fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
944 	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
945 		fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
946 
947 	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
948 	p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
949 
950 	if (p->flags&IP6_TNL_F_CAP_XMIT &&
951 			p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
952 		dev->flags |= IFF_POINTOPOINT;
953 	else
954 		dev->flags &= ~IFF_POINTOPOINT;
955 
956 	dev->iflink = p->link;
957 
958 	/* Precalculate GRE options length */
959 	if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
960 		if (t->parms.o_flags&GRE_CSUM)
961 			addend += 4;
962 		if (t->parms.o_flags&GRE_KEY)
963 			addend += 4;
964 		if (t->parms.o_flags&GRE_SEQ)
965 			addend += 4;
966 	}
967 	t->hlen = addend;
968 
969 	if (p->flags & IP6_TNL_F_CAP_XMIT) {
970 		int strict = (ipv6_addr_type(&p->raddr) &
971 			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
972 
973 		struct rt6_info *rt = rt6_lookup(t->net,
974 						 &p->raddr, &p->laddr,
975 						 p->link, strict);
976 
977 		if (rt == NULL)
978 			return;
979 
980 		if (rt->dst.dev) {
981 			dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
982 
983 			if (set_mtu) {
984 				dev->mtu = rt->dst.dev->mtu - addend;
985 				if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
986 					dev->mtu -= 8;
987 
988 				if (dev->mtu < IPV6_MIN_MTU)
989 					dev->mtu = IPV6_MIN_MTU;
990 			}
991 		}
992 		ip6_rt_put(rt);
993 	}
994 }
995 
996 static int ip6gre_tnl_change(struct ip6_tnl *t,
997 	const struct __ip6_tnl_parm *p, int set_mtu)
998 {
999 	t->parms.laddr = p->laddr;
1000 	t->parms.raddr = p->raddr;
1001 	t->parms.flags = p->flags;
1002 	t->parms.hop_limit = p->hop_limit;
1003 	t->parms.encap_limit = p->encap_limit;
1004 	t->parms.flowinfo = p->flowinfo;
1005 	t->parms.link = p->link;
1006 	t->parms.proto = p->proto;
1007 	t->parms.i_key = p->i_key;
1008 	t->parms.o_key = p->o_key;
1009 	t->parms.i_flags = p->i_flags;
1010 	t->parms.o_flags = p->o_flags;
1011 	ip6_tnl_dst_reset(t);
1012 	ip6gre_tnl_link_config(t, set_mtu);
1013 	return 0;
1014 }
1015 
1016 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1017 	const struct ip6_tnl_parm2 *u)
1018 {
1019 	p->laddr = u->laddr;
1020 	p->raddr = u->raddr;
1021 	p->flags = u->flags;
1022 	p->hop_limit = u->hop_limit;
1023 	p->encap_limit = u->encap_limit;
1024 	p->flowinfo = u->flowinfo;
1025 	p->link = u->link;
1026 	p->i_key = u->i_key;
1027 	p->o_key = u->o_key;
1028 	p->i_flags = u->i_flags;
1029 	p->o_flags = u->o_flags;
1030 	memcpy(p->name, u->name, sizeof(u->name));
1031 }
1032 
1033 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1034 	const struct __ip6_tnl_parm *p)
1035 {
1036 	u->proto = IPPROTO_GRE;
1037 	u->laddr = p->laddr;
1038 	u->raddr = p->raddr;
1039 	u->flags = p->flags;
1040 	u->hop_limit = p->hop_limit;
1041 	u->encap_limit = p->encap_limit;
1042 	u->flowinfo = p->flowinfo;
1043 	u->link = p->link;
1044 	u->i_key = p->i_key;
1045 	u->o_key = p->o_key;
1046 	u->i_flags = p->i_flags;
1047 	u->o_flags = p->o_flags;
1048 	memcpy(u->name, p->name, sizeof(u->name));
1049 }
1050 
1051 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1052 	struct ifreq *ifr, int cmd)
1053 {
1054 	int err = 0;
1055 	struct ip6_tnl_parm2 p;
1056 	struct __ip6_tnl_parm p1;
1057 	struct ip6_tnl *t = netdev_priv(dev);
1058 	struct net *net = t->net;
1059 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1060 
1061 	switch (cmd) {
1062 	case SIOCGETTUNNEL:
1063 		if (dev == ign->fb_tunnel_dev) {
1064 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1065 				err = -EFAULT;
1066 				break;
1067 			}
1068 			ip6gre_tnl_parm_from_user(&p1, &p);
1069 			t = ip6gre_tunnel_locate(net, &p1, 0);
1070 			if (t == NULL)
1071 				t = netdev_priv(dev);
1072 		}
1073 		memset(&p, 0, sizeof(p));
1074 		ip6gre_tnl_parm_to_user(&p, &t->parms);
1075 		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1076 			err = -EFAULT;
1077 		break;
1078 
1079 	case SIOCADDTUNNEL:
1080 	case SIOCCHGTUNNEL:
1081 		err = -EPERM;
1082 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1083 			goto done;
1084 
1085 		err = -EFAULT;
1086 		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1087 			goto done;
1088 
1089 		err = -EINVAL;
1090 		if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1091 			goto done;
1092 
1093 		if (!(p.i_flags&GRE_KEY))
1094 			p.i_key = 0;
1095 		if (!(p.o_flags&GRE_KEY))
1096 			p.o_key = 0;
1097 
1098 		ip6gre_tnl_parm_from_user(&p1, &p);
1099 		t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1100 
1101 		if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1102 			if (t != NULL) {
1103 				if (t->dev != dev) {
1104 					err = -EEXIST;
1105 					break;
1106 				}
1107 			} else {
1108 				t = netdev_priv(dev);
1109 
1110 				ip6gre_tunnel_unlink(ign, t);
1111 				synchronize_net();
1112 				ip6gre_tnl_change(t, &p1, 1);
1113 				ip6gre_tunnel_link(ign, t);
1114 				netdev_state_change(dev);
1115 			}
1116 		}
1117 
1118 		if (t) {
1119 			err = 0;
1120 
1121 			memset(&p, 0, sizeof(p));
1122 			ip6gre_tnl_parm_to_user(&p, &t->parms);
1123 			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1124 				err = -EFAULT;
1125 		} else
1126 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1127 		break;
1128 
1129 	case SIOCDELTUNNEL:
1130 		err = -EPERM;
1131 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1132 			goto done;
1133 
1134 		if (dev == ign->fb_tunnel_dev) {
1135 			err = -EFAULT;
1136 			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1137 				goto done;
1138 			err = -ENOENT;
1139 			ip6gre_tnl_parm_from_user(&p1, &p);
1140 			t = ip6gre_tunnel_locate(net, &p1, 0);
1141 			if (t == NULL)
1142 				goto done;
1143 			err = -EPERM;
1144 			if (t == netdev_priv(ign->fb_tunnel_dev))
1145 				goto done;
1146 			dev = t->dev;
1147 		}
1148 		unregister_netdevice(dev);
1149 		err = 0;
1150 		break;
1151 
1152 	default:
1153 		err = -EINVAL;
1154 	}
1155 
1156 done:
1157 	return err;
1158 }
1159 
1160 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1161 {
1162 	if (new_mtu < 68 ||
1163 	    new_mtu > 0xFFF8 - dev->hard_header_len)
1164 		return -EINVAL;
1165 	dev->mtu = new_mtu;
1166 	return 0;
1167 }
1168 
1169 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1170 			unsigned short type,
1171 			const void *daddr, const void *saddr, unsigned int len)
1172 {
1173 	struct ip6_tnl *t = netdev_priv(dev);
1174 	struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1175 	__be16 *p = (__be16 *)(ipv6h+1);
1176 
1177 	ip6_flow_hdr(ipv6h, 0, t->fl.u.ip6.flowlabel);
1178 	ipv6h->hop_limit = t->parms.hop_limit;
1179 	ipv6h->nexthdr = NEXTHDR_GRE;
1180 	ipv6h->saddr = t->parms.laddr;
1181 	ipv6h->daddr = t->parms.raddr;
1182 
1183 	p[0]		= t->parms.o_flags;
1184 	p[1]		= htons(type);
1185 
1186 	/*
1187 	 *	Set the source hardware address.
1188 	 */
1189 
1190 	if (saddr)
1191 		memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1192 	if (daddr)
1193 		memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1194 	if (!ipv6_addr_any(&ipv6h->daddr))
1195 		return t->hlen;
1196 
1197 	return -t->hlen;
1198 }
1199 
1200 static const struct header_ops ip6gre_header_ops = {
1201 	.create	= ip6gre_header,
1202 };
1203 
1204 static const struct net_device_ops ip6gre_netdev_ops = {
1205 	.ndo_init		= ip6gre_tunnel_init,
1206 	.ndo_uninit		= ip6gre_tunnel_uninit,
1207 	.ndo_start_xmit		= ip6gre_tunnel_xmit,
1208 	.ndo_do_ioctl		= ip6gre_tunnel_ioctl,
1209 	.ndo_change_mtu		= ip6gre_tunnel_change_mtu,
1210 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1211 };
1212 
1213 static void ip6gre_dev_free(struct net_device *dev)
1214 {
1215 	free_percpu(dev->tstats);
1216 	free_netdev(dev);
1217 }
1218 
1219 static void ip6gre_tunnel_setup(struct net_device *dev)
1220 {
1221 	struct ip6_tnl *t;
1222 
1223 	dev->netdev_ops = &ip6gre_netdev_ops;
1224 	dev->destructor = ip6gre_dev_free;
1225 
1226 	dev->type = ARPHRD_IP6GRE;
1227 	dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1228 	dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1229 	t = netdev_priv(dev);
1230 	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1231 		dev->mtu -= 8;
1232 	dev->flags |= IFF_NOARP;
1233 	dev->iflink = 0;
1234 	dev->addr_len = sizeof(struct in6_addr);
1235 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1236 }
1237 
1238 static int ip6gre_tunnel_init(struct net_device *dev)
1239 {
1240 	struct ip6_tnl *tunnel;
1241 	int i;
1242 
1243 	tunnel = netdev_priv(dev);
1244 
1245 	tunnel->dev = dev;
1246 	tunnel->net = dev_net(dev);
1247 	strcpy(tunnel->parms.name, dev->name);
1248 
1249 	memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1250 	memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1251 
1252 	if (ipv6_addr_any(&tunnel->parms.raddr))
1253 		dev->header_ops = &ip6gre_header_ops;
1254 
1255 	dev->tstats = alloc_percpu(struct pcpu_sw_netstats);
1256 	if (!dev->tstats)
1257 		return -ENOMEM;
1258 
1259 	for_each_possible_cpu(i) {
1260 		struct pcpu_sw_netstats *ip6gre_tunnel_stats;
1261 		ip6gre_tunnel_stats = per_cpu_ptr(dev->tstats, i);
1262 		u64_stats_init(&ip6gre_tunnel_stats->syncp);
1263 	}
1264 
1265 
1266 	return 0;
1267 }
1268 
1269 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1270 {
1271 	struct ip6_tnl *tunnel = netdev_priv(dev);
1272 
1273 	tunnel->dev = dev;
1274 	tunnel->net = dev_net(dev);
1275 	strcpy(tunnel->parms.name, dev->name);
1276 
1277 	tunnel->hlen		= sizeof(struct ipv6hdr) + 4;
1278 
1279 	dev_hold(dev);
1280 }
1281 
1282 
1283 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1284 	.handler     = ip6gre_rcv,
1285 	.err_handler = ip6gre_err,
1286 	.flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1287 };
1288 
1289 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1290 {
1291 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1292 	struct net_device *dev, *aux;
1293 	int prio;
1294 
1295 	for_each_netdev_safe(net, dev, aux)
1296 		if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1297 		    dev->rtnl_link_ops == &ip6gre_tap_ops)
1298 			unregister_netdevice_queue(dev, head);
1299 
1300 	for (prio = 0; prio < 4; prio++) {
1301 		int h;
1302 		for (h = 0; h < HASH_SIZE; h++) {
1303 			struct ip6_tnl *t;
1304 
1305 			t = rtnl_dereference(ign->tunnels[prio][h]);
1306 
1307 			while (t != NULL) {
1308 				/* If dev is in the same netns, it has already
1309 				 * been added to the list by the previous loop.
1310 				 */
1311 				if (!net_eq(dev_net(t->dev), net))
1312 					unregister_netdevice_queue(t->dev,
1313 								   head);
1314 				t = rtnl_dereference(t->next);
1315 			}
1316 		}
1317 	}
1318 }
1319 
1320 static int __net_init ip6gre_init_net(struct net *net)
1321 {
1322 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1323 	int err;
1324 
1325 	ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1326 					   ip6gre_tunnel_setup);
1327 	if (!ign->fb_tunnel_dev) {
1328 		err = -ENOMEM;
1329 		goto err_alloc_dev;
1330 	}
1331 	dev_net_set(ign->fb_tunnel_dev, net);
1332 	/* FB netdevice is special: we have one, and only one per netns.
1333 	 * Allowing to move it to another netns is clearly unsafe.
1334 	 */
1335 	ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1336 
1337 
1338 	ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1339 	ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1340 
1341 	err = register_netdev(ign->fb_tunnel_dev);
1342 	if (err)
1343 		goto err_reg_dev;
1344 
1345 	rcu_assign_pointer(ign->tunnels_wc[0],
1346 			   netdev_priv(ign->fb_tunnel_dev));
1347 	return 0;
1348 
1349 err_reg_dev:
1350 	ip6gre_dev_free(ign->fb_tunnel_dev);
1351 err_alloc_dev:
1352 	return err;
1353 }
1354 
1355 static void __net_exit ip6gre_exit_net(struct net *net)
1356 {
1357 	LIST_HEAD(list);
1358 
1359 	rtnl_lock();
1360 	ip6gre_destroy_tunnels(net, &list);
1361 	unregister_netdevice_many(&list);
1362 	rtnl_unlock();
1363 }
1364 
1365 static struct pernet_operations ip6gre_net_ops = {
1366 	.init = ip6gre_init_net,
1367 	.exit = ip6gre_exit_net,
1368 	.id   = &ip6gre_net_id,
1369 	.size = sizeof(struct ip6gre_net),
1370 };
1371 
1372 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1373 {
1374 	__be16 flags;
1375 
1376 	if (!data)
1377 		return 0;
1378 
1379 	flags = 0;
1380 	if (data[IFLA_GRE_IFLAGS])
1381 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1382 	if (data[IFLA_GRE_OFLAGS])
1383 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1384 	if (flags & (GRE_VERSION|GRE_ROUTING))
1385 		return -EINVAL;
1386 
1387 	return 0;
1388 }
1389 
1390 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1391 {
1392 	struct in6_addr daddr;
1393 
1394 	if (tb[IFLA_ADDRESS]) {
1395 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1396 			return -EINVAL;
1397 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1398 			return -EADDRNOTAVAIL;
1399 	}
1400 
1401 	if (!data)
1402 		goto out;
1403 
1404 	if (data[IFLA_GRE_REMOTE]) {
1405 		nla_memcpy(&daddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1406 		if (ipv6_addr_any(&daddr))
1407 			return -EINVAL;
1408 	}
1409 
1410 out:
1411 	return ip6gre_tunnel_validate(tb, data);
1412 }
1413 
1414 
1415 static void ip6gre_netlink_parms(struct nlattr *data[],
1416 				struct __ip6_tnl_parm *parms)
1417 {
1418 	memset(parms, 0, sizeof(*parms));
1419 
1420 	if (!data)
1421 		return;
1422 
1423 	if (data[IFLA_GRE_LINK])
1424 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1425 
1426 	if (data[IFLA_GRE_IFLAGS])
1427 		parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1428 
1429 	if (data[IFLA_GRE_OFLAGS])
1430 		parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1431 
1432 	if (data[IFLA_GRE_IKEY])
1433 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1434 
1435 	if (data[IFLA_GRE_OKEY])
1436 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1437 
1438 	if (data[IFLA_GRE_LOCAL])
1439 		nla_memcpy(&parms->laddr, data[IFLA_GRE_LOCAL], sizeof(struct in6_addr));
1440 
1441 	if (data[IFLA_GRE_REMOTE])
1442 		nla_memcpy(&parms->raddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1443 
1444 	if (data[IFLA_GRE_TTL])
1445 		parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1446 
1447 	if (data[IFLA_GRE_ENCAP_LIMIT])
1448 		parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1449 
1450 	if (data[IFLA_GRE_FLOWINFO])
1451 		parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1452 
1453 	if (data[IFLA_GRE_FLAGS])
1454 		parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1455 }
1456 
1457 static int ip6gre_tap_init(struct net_device *dev)
1458 {
1459 	struct ip6_tnl *tunnel;
1460 
1461 	tunnel = netdev_priv(dev);
1462 
1463 	tunnel->dev = dev;
1464 	tunnel->net = dev_net(dev);
1465 	strcpy(tunnel->parms.name, dev->name);
1466 
1467 	ip6gre_tnl_link_config(tunnel, 1);
1468 
1469 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1470 	if (!dev->tstats)
1471 		return -ENOMEM;
1472 
1473 	return 0;
1474 }
1475 
1476 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1477 	.ndo_init = ip6gre_tap_init,
1478 	.ndo_uninit = ip6gre_tunnel_uninit,
1479 	.ndo_start_xmit = ip6gre_tunnel_xmit,
1480 	.ndo_set_mac_address = eth_mac_addr,
1481 	.ndo_validate_addr = eth_validate_addr,
1482 	.ndo_change_mtu = ip6gre_tunnel_change_mtu,
1483 	.ndo_get_stats64 = ip_tunnel_get_stats64,
1484 };
1485 
1486 static void ip6gre_tap_setup(struct net_device *dev)
1487 {
1488 
1489 	ether_setup(dev);
1490 
1491 	dev->netdev_ops = &ip6gre_tap_netdev_ops;
1492 	dev->destructor = ip6gre_dev_free;
1493 
1494 	dev->iflink = 0;
1495 	dev->features |= NETIF_F_NETNS_LOCAL;
1496 }
1497 
1498 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1499 	struct nlattr *tb[], struct nlattr *data[])
1500 {
1501 	struct ip6_tnl *nt;
1502 	struct net *net = dev_net(dev);
1503 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1504 	int err;
1505 
1506 	nt = netdev_priv(dev);
1507 	ip6gre_netlink_parms(data, &nt->parms);
1508 
1509 	if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1510 		return -EEXIST;
1511 
1512 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1513 		eth_hw_addr_random(dev);
1514 
1515 	nt->dev = dev;
1516 	nt->net = dev_net(dev);
1517 	ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1518 
1519 	/* Can use a lockless transmit, unless we generate output sequences */
1520 	if (!(nt->parms.o_flags & GRE_SEQ))
1521 		dev->features |= NETIF_F_LLTX;
1522 
1523 	err = register_netdevice(dev);
1524 	if (err)
1525 		goto out;
1526 
1527 	dev_hold(dev);
1528 	ip6gre_tunnel_link(ign, nt);
1529 
1530 out:
1531 	return err;
1532 }
1533 
1534 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1535 			    struct nlattr *data[])
1536 {
1537 	struct ip6_tnl *t, *nt = netdev_priv(dev);
1538 	struct net *net = nt->net;
1539 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1540 	struct __ip6_tnl_parm p;
1541 
1542 	if (dev == ign->fb_tunnel_dev)
1543 		return -EINVAL;
1544 
1545 	ip6gre_netlink_parms(data, &p);
1546 
1547 	t = ip6gre_tunnel_locate(net, &p, 0);
1548 
1549 	if (t) {
1550 		if (t->dev != dev)
1551 			return -EEXIST;
1552 	} else {
1553 		t = nt;
1554 
1555 		ip6gre_tunnel_unlink(ign, t);
1556 		ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1557 		ip6gre_tunnel_link(ign, t);
1558 		netdev_state_change(dev);
1559 	}
1560 
1561 	return 0;
1562 }
1563 
1564 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1565 {
1566 	struct net *net = dev_net(dev);
1567 	struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1568 
1569 	if (dev != ign->fb_tunnel_dev)
1570 		unregister_netdevice_queue(dev, head);
1571 }
1572 
1573 static size_t ip6gre_get_size(const struct net_device *dev)
1574 {
1575 	return
1576 		/* IFLA_GRE_LINK */
1577 		nla_total_size(4) +
1578 		/* IFLA_GRE_IFLAGS */
1579 		nla_total_size(2) +
1580 		/* IFLA_GRE_OFLAGS */
1581 		nla_total_size(2) +
1582 		/* IFLA_GRE_IKEY */
1583 		nla_total_size(4) +
1584 		/* IFLA_GRE_OKEY */
1585 		nla_total_size(4) +
1586 		/* IFLA_GRE_LOCAL */
1587 		nla_total_size(sizeof(struct in6_addr)) +
1588 		/* IFLA_GRE_REMOTE */
1589 		nla_total_size(sizeof(struct in6_addr)) +
1590 		/* IFLA_GRE_TTL */
1591 		nla_total_size(1) +
1592 		/* IFLA_GRE_TOS */
1593 		nla_total_size(1) +
1594 		/* IFLA_GRE_ENCAP_LIMIT */
1595 		nla_total_size(1) +
1596 		/* IFLA_GRE_FLOWINFO */
1597 		nla_total_size(4) +
1598 		/* IFLA_GRE_FLAGS */
1599 		nla_total_size(4) +
1600 		0;
1601 }
1602 
1603 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1604 {
1605 	struct ip6_tnl *t = netdev_priv(dev);
1606 	struct __ip6_tnl_parm *p = &t->parms;
1607 
1608 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1609 	    nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1610 	    nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1611 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1612 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1613 	    nla_put(skb, IFLA_GRE_LOCAL, sizeof(struct in6_addr), &p->laddr) ||
1614 	    nla_put(skb, IFLA_GRE_REMOTE, sizeof(struct in6_addr), &p->raddr) ||
1615 	    nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1616 	    /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1617 	    nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1618 	    nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1619 	    nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1620 		goto nla_put_failure;
1621 	return 0;
1622 
1623 nla_put_failure:
1624 	return -EMSGSIZE;
1625 }
1626 
1627 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1628 	[IFLA_GRE_LINK]        = { .type = NLA_U32 },
1629 	[IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1630 	[IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1631 	[IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1632 	[IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1633 	[IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1634 	[IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1635 	[IFLA_GRE_TTL]         = { .type = NLA_U8 },
1636 	[IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1637 	[IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1638 	[IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1639 };
1640 
1641 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1642 	.kind		= "ip6gre",
1643 	.maxtype	= IFLA_GRE_MAX,
1644 	.policy		= ip6gre_policy,
1645 	.priv_size	= sizeof(struct ip6_tnl),
1646 	.setup		= ip6gre_tunnel_setup,
1647 	.validate	= ip6gre_tunnel_validate,
1648 	.newlink	= ip6gre_newlink,
1649 	.changelink	= ip6gre_changelink,
1650 	.dellink	= ip6gre_dellink,
1651 	.get_size	= ip6gre_get_size,
1652 	.fill_info	= ip6gre_fill_info,
1653 };
1654 
1655 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1656 	.kind		= "ip6gretap",
1657 	.maxtype	= IFLA_GRE_MAX,
1658 	.policy		= ip6gre_policy,
1659 	.priv_size	= sizeof(struct ip6_tnl),
1660 	.setup		= ip6gre_tap_setup,
1661 	.validate	= ip6gre_tap_validate,
1662 	.newlink	= ip6gre_newlink,
1663 	.changelink	= ip6gre_changelink,
1664 	.get_size	= ip6gre_get_size,
1665 	.fill_info	= ip6gre_fill_info,
1666 };
1667 
1668 /*
1669  *	And now the modules code and kernel interface.
1670  */
1671 
1672 static int __init ip6gre_init(void)
1673 {
1674 	int err;
1675 
1676 	pr_info("GRE over IPv6 tunneling driver\n");
1677 
1678 	err = register_pernet_device(&ip6gre_net_ops);
1679 	if (err < 0)
1680 		return err;
1681 
1682 	err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1683 	if (err < 0) {
1684 		pr_info("%s: can't add protocol\n", __func__);
1685 		goto add_proto_failed;
1686 	}
1687 
1688 	err = rtnl_link_register(&ip6gre_link_ops);
1689 	if (err < 0)
1690 		goto rtnl_link_failed;
1691 
1692 	err = rtnl_link_register(&ip6gre_tap_ops);
1693 	if (err < 0)
1694 		goto tap_ops_failed;
1695 
1696 out:
1697 	return err;
1698 
1699 tap_ops_failed:
1700 	rtnl_link_unregister(&ip6gre_link_ops);
1701 rtnl_link_failed:
1702 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1703 add_proto_failed:
1704 	unregister_pernet_device(&ip6gre_net_ops);
1705 	goto out;
1706 }
1707 
1708 static void __exit ip6gre_fini(void)
1709 {
1710 	rtnl_link_unregister(&ip6gre_tap_ops);
1711 	rtnl_link_unregister(&ip6gre_link_ops);
1712 	inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1713 	unregister_pernet_device(&ip6gre_net_ops);
1714 }
1715 
1716 module_init(ip6gre_init);
1717 module_exit(ip6gre_fini);
1718 MODULE_LICENSE("GPL");
1719 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1720 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1721 MODULE_ALIAS_RTNL_LINK("ip6gre");
1722 MODULE_ALIAS_NETDEV("ip6gre0");
1723