xref: /linux/net/ipv4/ip_tunnel.c (revision a81ab36bf52d0ca3a32251a923be1dbced726141)
1 /*
2  * Copyright (c) 2013 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/capability.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/kernel.h>
25 #include <linux/slab.h>
26 #include <linux/uaccess.h>
27 #include <linux/skbuff.h>
28 #include <linux/netdevice.h>
29 #include <linux/in.h>
30 #include <linux/tcp.h>
31 #include <linux/udp.h>
32 #include <linux/if_arp.h>
33 #include <linux/mroute.h>
34 #include <linux/init.h>
35 #include <linux/in6.h>
36 #include <linux/inetdevice.h>
37 #include <linux/igmp.h>
38 #include <linux/netfilter_ipv4.h>
39 #include <linux/etherdevice.h>
40 #include <linux/if_ether.h>
41 #include <linux/if_vlan.h>
42 #include <linux/rculist.h>
43 
44 #include <net/sock.h>
45 #include <net/ip.h>
46 #include <net/icmp.h>
47 #include <net/protocol.h>
48 #include <net/ip_tunnels.h>
49 #include <net/arp.h>
50 #include <net/checksum.h>
51 #include <net/dsfield.h>
52 #include <net/inet_ecn.h>
53 #include <net/xfrm.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
56 #include <net/rtnetlink.h>
57 
58 #if IS_ENABLED(CONFIG_IPV6)
59 #include <net/ipv6.h>
60 #include <net/ip6_fib.h>
61 #include <net/ip6_route.h>
62 #endif
63 
64 static unsigned int ip_tunnel_hash(struct ip_tunnel_net *itn,
65 				   __be32 key, __be32 remote)
66 {
67 	return hash_32((__force u32)key ^ (__force u32)remote,
68 			 IP_TNL_HASH_BITS);
69 }
70 
71 static inline void __tunnel_dst_set(struct ip_tunnel_dst *idst,
72 				    struct dst_entry *dst)
73 {
74 	struct dst_entry *old_dst;
75 
76 	if (dst && (dst->flags & DST_NOCACHE))
77 		dst = NULL;
78 
79 	spin_lock_bh(&idst->lock);
80 	old_dst = rcu_dereference(idst->dst);
81 	rcu_assign_pointer(idst->dst, dst);
82 	dst_release(old_dst);
83 	spin_unlock_bh(&idst->lock);
84 }
85 
86 static inline void tunnel_dst_set(struct ip_tunnel *t, struct dst_entry *dst)
87 {
88 	__tunnel_dst_set(this_cpu_ptr(t->dst_cache), dst);
89 }
90 
91 static inline void tunnel_dst_reset(struct ip_tunnel *t)
92 {
93 	tunnel_dst_set(t, NULL);
94 }
95 
96 static void tunnel_dst_reset_all(struct ip_tunnel *t)
97 {
98 	int i;
99 
100 	for_each_possible_cpu(i)
101 		__tunnel_dst_set(per_cpu_ptr(t->dst_cache, i), NULL);
102 }
103 
104 static inline struct dst_entry *tunnel_dst_get(struct ip_tunnel *t)
105 {
106 	struct dst_entry *dst;
107 
108 	rcu_read_lock();
109 	dst = rcu_dereference(this_cpu_ptr(t->dst_cache)->dst);
110 	if (dst)
111 		dst_hold(dst);
112 	rcu_read_unlock();
113 	return dst;
114 }
115 
116 static struct dst_entry *tunnel_dst_check(struct ip_tunnel *t, u32 cookie)
117 {
118 	struct dst_entry *dst = tunnel_dst_get(t);
119 
120 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
121 		tunnel_dst_reset(t);
122 		return NULL;
123 	}
124 
125 	return dst;
126 }
127 
128 /* Often modified stats are per cpu, other are shared (netdev->stats) */
129 struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev,
130 						struct rtnl_link_stats64 *tot)
131 {
132 	int i;
133 
134 	for_each_possible_cpu(i) {
135 		const struct pcpu_sw_netstats *tstats =
136 						   per_cpu_ptr(dev->tstats, i);
137 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
138 		unsigned int start;
139 
140 		do {
141 			start = u64_stats_fetch_begin_bh(&tstats->syncp);
142 			rx_packets = tstats->rx_packets;
143 			tx_packets = tstats->tx_packets;
144 			rx_bytes = tstats->rx_bytes;
145 			tx_bytes = tstats->tx_bytes;
146 		} while (u64_stats_fetch_retry_bh(&tstats->syncp, start));
147 
148 		tot->rx_packets += rx_packets;
149 		tot->tx_packets += tx_packets;
150 		tot->rx_bytes   += rx_bytes;
151 		tot->tx_bytes   += tx_bytes;
152 	}
153 
154 	tot->multicast = dev->stats.multicast;
155 
156 	tot->rx_crc_errors = dev->stats.rx_crc_errors;
157 	tot->rx_fifo_errors = dev->stats.rx_fifo_errors;
158 	tot->rx_length_errors = dev->stats.rx_length_errors;
159 	tot->rx_frame_errors = dev->stats.rx_frame_errors;
160 	tot->rx_errors = dev->stats.rx_errors;
161 
162 	tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
163 	tot->tx_carrier_errors = dev->stats.tx_carrier_errors;
164 	tot->tx_dropped = dev->stats.tx_dropped;
165 	tot->tx_aborted_errors = dev->stats.tx_aborted_errors;
166 	tot->tx_errors = dev->stats.tx_errors;
167 
168 	tot->collisions  = dev->stats.collisions;
169 
170 	return tot;
171 }
172 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
173 
174 static bool ip_tunnel_key_match(const struct ip_tunnel_parm *p,
175 				__be16 flags, __be32 key)
176 {
177 	if (p->i_flags & TUNNEL_KEY) {
178 		if (flags & TUNNEL_KEY)
179 			return key == p->i_key;
180 		else
181 			/* key expected, none present */
182 			return false;
183 	} else
184 		return !(flags & TUNNEL_KEY);
185 }
186 
187 /* Fallback tunnel: no source, no destination, no key, no options
188 
189    Tunnel hash table:
190    We require exact key match i.e. if a key is present in packet
191    it will match only tunnel with the same key; if it is not present,
192    it will match only keyless tunnel.
193 
194    All keysless packets, if not matched configured keyless tunnels
195    will match fallback tunnel.
196    Given src, dst and key, find appropriate for input tunnel.
197 */
198 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
199 				   int link, __be16 flags,
200 				   __be32 remote, __be32 local,
201 				   __be32 key)
202 {
203 	unsigned int hash;
204 	struct ip_tunnel *t, *cand = NULL;
205 	struct hlist_head *head;
206 
207 	hash = ip_tunnel_hash(itn, key, remote);
208 	head = &itn->tunnels[hash];
209 
210 	hlist_for_each_entry_rcu(t, head, hash_node) {
211 		if (local != t->parms.iph.saddr ||
212 		    remote != t->parms.iph.daddr ||
213 		    !(t->dev->flags & IFF_UP))
214 			continue;
215 
216 		if (!ip_tunnel_key_match(&t->parms, flags, key))
217 			continue;
218 
219 		if (t->parms.link == link)
220 			return t;
221 		else
222 			cand = t;
223 	}
224 
225 	hlist_for_each_entry_rcu(t, head, hash_node) {
226 		if (remote != t->parms.iph.daddr ||
227 		    !(t->dev->flags & IFF_UP))
228 			continue;
229 
230 		if (!ip_tunnel_key_match(&t->parms, flags, key))
231 			continue;
232 
233 		if (t->parms.link == link)
234 			return t;
235 		else if (!cand)
236 			cand = t;
237 	}
238 
239 	hash = ip_tunnel_hash(itn, key, 0);
240 	head = &itn->tunnels[hash];
241 
242 	hlist_for_each_entry_rcu(t, head, hash_node) {
243 		if ((local != t->parms.iph.saddr &&
244 		     (local != t->parms.iph.daddr ||
245 		      !ipv4_is_multicast(local))) ||
246 		    !(t->dev->flags & IFF_UP))
247 			continue;
248 
249 		if (!ip_tunnel_key_match(&t->parms, flags, key))
250 			continue;
251 
252 		if (t->parms.link == link)
253 			return t;
254 		else if (!cand)
255 			cand = t;
256 	}
257 
258 	if (flags & TUNNEL_NO_KEY)
259 		goto skip_key_lookup;
260 
261 	hlist_for_each_entry_rcu(t, head, hash_node) {
262 		if (t->parms.i_key != key ||
263 		    !(t->dev->flags & IFF_UP))
264 			continue;
265 
266 		if (t->parms.link == link)
267 			return t;
268 		else if (!cand)
269 			cand = t;
270 	}
271 
272 skip_key_lookup:
273 	if (cand)
274 		return cand;
275 
276 	if (itn->fb_tunnel_dev && itn->fb_tunnel_dev->flags & IFF_UP)
277 		return netdev_priv(itn->fb_tunnel_dev);
278 
279 
280 	return NULL;
281 }
282 EXPORT_SYMBOL_GPL(ip_tunnel_lookup);
283 
284 static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn,
285 				    struct ip_tunnel_parm *parms)
286 {
287 	unsigned int h;
288 	__be32 remote;
289 
290 	if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr))
291 		remote = parms->iph.daddr;
292 	else
293 		remote = 0;
294 
295 	h = ip_tunnel_hash(itn, parms->i_key, remote);
296 	return &itn->tunnels[h];
297 }
298 
299 static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t)
300 {
301 	struct hlist_head *head = ip_bucket(itn, &t->parms);
302 
303 	hlist_add_head_rcu(&t->hash_node, head);
304 }
305 
306 static void ip_tunnel_del(struct ip_tunnel *t)
307 {
308 	hlist_del_init_rcu(&t->hash_node);
309 }
310 
311 static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn,
312 					struct ip_tunnel_parm *parms,
313 					int type)
314 {
315 	__be32 remote = parms->iph.daddr;
316 	__be32 local = parms->iph.saddr;
317 	__be32 key = parms->i_key;
318 	int link = parms->link;
319 	struct ip_tunnel *t = NULL;
320 	struct hlist_head *head = ip_bucket(itn, parms);
321 
322 	hlist_for_each_entry_rcu(t, head, hash_node) {
323 		if (local == t->parms.iph.saddr &&
324 		    remote == t->parms.iph.daddr &&
325 		    key == t->parms.i_key &&
326 		    link == t->parms.link &&
327 		    type == t->dev->type)
328 			break;
329 	}
330 	return t;
331 }
332 
333 static struct net_device *__ip_tunnel_create(struct net *net,
334 					     const struct rtnl_link_ops *ops,
335 					     struct ip_tunnel_parm *parms)
336 {
337 	int err;
338 	struct ip_tunnel *tunnel;
339 	struct net_device *dev;
340 	char name[IFNAMSIZ];
341 
342 	if (parms->name[0])
343 		strlcpy(name, parms->name, IFNAMSIZ);
344 	else {
345 		if (strlen(ops->kind) > (IFNAMSIZ - 3)) {
346 			err = -E2BIG;
347 			goto failed;
348 		}
349 		strlcpy(name, ops->kind, IFNAMSIZ);
350 		strncat(name, "%d", 2);
351 	}
352 
353 	ASSERT_RTNL();
354 	dev = alloc_netdev(ops->priv_size, name, ops->setup);
355 	if (!dev) {
356 		err = -ENOMEM;
357 		goto failed;
358 	}
359 	dev_net_set(dev, net);
360 
361 	dev->rtnl_link_ops = ops;
362 
363 	tunnel = netdev_priv(dev);
364 	tunnel->parms = *parms;
365 	tunnel->net = net;
366 
367 	err = register_netdevice(dev);
368 	if (err)
369 		goto failed_free;
370 
371 	return dev;
372 
373 failed_free:
374 	free_netdev(dev);
375 failed:
376 	return ERR_PTR(err);
377 }
378 
379 static inline void init_tunnel_flow(struct flowi4 *fl4,
380 				    int proto,
381 				    __be32 daddr, __be32 saddr,
382 				    __be32 key, __u8 tos, int oif)
383 {
384 	memset(fl4, 0, sizeof(*fl4));
385 	fl4->flowi4_oif = oif;
386 	fl4->daddr = daddr;
387 	fl4->saddr = saddr;
388 	fl4->flowi4_tos = tos;
389 	fl4->flowi4_proto = proto;
390 	fl4->fl4_gre_key = key;
391 }
392 
393 static int ip_tunnel_bind_dev(struct net_device *dev)
394 {
395 	struct net_device *tdev = NULL;
396 	struct ip_tunnel *tunnel = netdev_priv(dev);
397 	const struct iphdr *iph;
398 	int hlen = LL_MAX_HEADER;
399 	int mtu = ETH_DATA_LEN;
400 	int t_hlen = tunnel->hlen + sizeof(struct iphdr);
401 
402 	iph = &tunnel->parms.iph;
403 
404 	/* Guess output device to choose reasonable mtu and needed_headroom */
405 	if (iph->daddr) {
406 		struct flowi4 fl4;
407 		struct rtable *rt;
408 
409 		init_tunnel_flow(&fl4, iph->protocol, iph->daddr,
410 				 iph->saddr, tunnel->parms.o_key,
411 				 RT_TOS(iph->tos), tunnel->parms.link);
412 		rt = ip_route_output_key(tunnel->net, &fl4);
413 
414 		if (!IS_ERR(rt)) {
415 			tdev = rt->dst.dev;
416 			tunnel_dst_set(tunnel, dst_clone(&rt->dst));
417 			ip_rt_put(rt);
418 		}
419 		if (dev->type != ARPHRD_ETHER)
420 			dev->flags |= IFF_POINTOPOINT;
421 	}
422 
423 	if (!tdev && tunnel->parms.link)
424 		tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link);
425 
426 	if (tdev) {
427 		hlen = tdev->hard_header_len + tdev->needed_headroom;
428 		mtu = tdev->mtu;
429 	}
430 	dev->iflink = tunnel->parms.link;
431 
432 	dev->needed_headroom = t_hlen + hlen;
433 	mtu -= (dev->hard_header_len + t_hlen);
434 
435 	if (mtu < 68)
436 		mtu = 68;
437 
438 	return mtu;
439 }
440 
441 static struct ip_tunnel *ip_tunnel_create(struct net *net,
442 					  struct ip_tunnel_net *itn,
443 					  struct ip_tunnel_parm *parms)
444 {
445 	struct ip_tunnel *nt, *fbt;
446 	struct net_device *dev;
447 
448 	BUG_ON(!itn->fb_tunnel_dev);
449 	fbt = netdev_priv(itn->fb_tunnel_dev);
450 	dev = __ip_tunnel_create(net, itn->fb_tunnel_dev->rtnl_link_ops, parms);
451 	if (IS_ERR(dev))
452 		return NULL;
453 
454 	dev->mtu = ip_tunnel_bind_dev(dev);
455 
456 	nt = netdev_priv(dev);
457 	ip_tunnel_add(itn, nt);
458 	return nt;
459 }
460 
461 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb,
462 		  const struct tnl_ptk_info *tpi, bool log_ecn_error)
463 {
464 	struct pcpu_sw_netstats *tstats;
465 	const struct iphdr *iph = ip_hdr(skb);
466 	int err;
467 
468 #ifdef CONFIG_NET_IPGRE_BROADCAST
469 	if (ipv4_is_multicast(iph->daddr)) {
470 		/* Looped back packet, drop it! */
471 		if (rt_is_output_route(skb_rtable(skb)))
472 			goto drop;
473 		tunnel->dev->stats.multicast++;
474 		skb->pkt_type = PACKET_BROADCAST;
475 	}
476 #endif
477 
478 	if ((!(tpi->flags&TUNNEL_CSUM) &&  (tunnel->parms.i_flags&TUNNEL_CSUM)) ||
479 	     ((tpi->flags&TUNNEL_CSUM) && !(tunnel->parms.i_flags&TUNNEL_CSUM))) {
480 		tunnel->dev->stats.rx_crc_errors++;
481 		tunnel->dev->stats.rx_errors++;
482 		goto drop;
483 	}
484 
485 	if (tunnel->parms.i_flags&TUNNEL_SEQ) {
486 		if (!(tpi->flags&TUNNEL_SEQ) ||
487 		    (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) {
488 			tunnel->dev->stats.rx_fifo_errors++;
489 			tunnel->dev->stats.rx_errors++;
490 			goto drop;
491 		}
492 		tunnel->i_seqno = ntohl(tpi->seq) + 1;
493 	}
494 
495 	err = IP_ECN_decapsulate(iph, skb);
496 	if (unlikely(err)) {
497 		if (log_ecn_error)
498 			net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
499 					&iph->saddr, iph->tos);
500 		if (err > 1) {
501 			++tunnel->dev->stats.rx_frame_errors;
502 			++tunnel->dev->stats.rx_errors;
503 			goto drop;
504 		}
505 	}
506 
507 	tstats = this_cpu_ptr(tunnel->dev->tstats);
508 	u64_stats_update_begin(&tstats->syncp);
509 	tstats->rx_packets++;
510 	tstats->rx_bytes += skb->len;
511 	u64_stats_update_end(&tstats->syncp);
512 
513 	skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev)));
514 
515 	if (tunnel->dev->type == ARPHRD_ETHER) {
516 		skb->protocol = eth_type_trans(skb, tunnel->dev);
517 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
518 	} else {
519 		skb->dev = tunnel->dev;
520 	}
521 
522 	gro_cells_receive(&tunnel->gro_cells, skb);
523 	return 0;
524 
525 drop:
526 	kfree_skb(skb);
527 	return 0;
528 }
529 EXPORT_SYMBOL_GPL(ip_tunnel_rcv);
530 
531 static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb,
532 			    struct rtable *rt, __be16 df)
533 {
534 	struct ip_tunnel *tunnel = netdev_priv(dev);
535 	int pkt_size = skb->len - tunnel->hlen - dev->hard_header_len;
536 	int mtu;
537 
538 	if (df)
539 		mtu = dst_mtu(&rt->dst) - dev->hard_header_len
540 					- sizeof(struct iphdr) - tunnel->hlen;
541 	else
542 		mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
543 
544 	if (skb_dst(skb))
545 		skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
546 
547 	if (skb->protocol == htons(ETH_P_IP)) {
548 		if (!skb_is_gso(skb) &&
549 		    (df & htons(IP_DF)) && mtu < pkt_size) {
550 			memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
551 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
552 			return -E2BIG;
553 		}
554 	}
555 #if IS_ENABLED(CONFIG_IPV6)
556 	else if (skb->protocol == htons(ETH_P_IPV6)) {
557 		struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
558 
559 		if (rt6 && mtu < dst_mtu(skb_dst(skb)) &&
560 			   mtu >= IPV6_MIN_MTU) {
561 			if ((tunnel->parms.iph.daddr &&
562 			    !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
563 			    rt6->rt6i_dst.plen == 128) {
564 				rt6->rt6i_flags |= RTF_MODIFIED;
565 				dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
566 			}
567 		}
568 
569 		if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU &&
570 					mtu < pkt_size) {
571 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
572 			return -E2BIG;
573 		}
574 	}
575 #endif
576 	return 0;
577 }
578 
579 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
580 		    const struct iphdr *tnl_params, const u8 protocol)
581 {
582 	struct ip_tunnel *tunnel = netdev_priv(dev);
583 	const struct iphdr *inner_iph;
584 	struct flowi4 fl4;
585 	u8     tos, ttl;
586 	__be16 df;
587 	struct rtable *rt = NULL;	/* Route to the other host */
588 	unsigned int max_headroom;	/* The extra header space needed */
589 	__be32 dst;
590 	int err;
591 	bool connected = true;
592 
593 	inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
594 
595 	dst = tnl_params->daddr;
596 	if (dst == 0) {
597 		/* NBMA tunnel */
598 
599 		if (skb_dst(skb) == NULL) {
600 			dev->stats.tx_fifo_errors++;
601 			goto tx_error;
602 		}
603 
604 		if (skb->protocol == htons(ETH_P_IP)) {
605 			rt = skb_rtable(skb);
606 			dst = rt_nexthop(rt, inner_iph->daddr);
607 		}
608 #if IS_ENABLED(CONFIG_IPV6)
609 		else if (skb->protocol == htons(ETH_P_IPV6)) {
610 			const struct in6_addr *addr6;
611 			struct neighbour *neigh;
612 			bool do_tx_error_icmp;
613 			int addr_type;
614 
615 			neigh = dst_neigh_lookup(skb_dst(skb),
616 						 &ipv6_hdr(skb)->daddr);
617 			if (neigh == NULL)
618 				goto tx_error;
619 
620 			addr6 = (const struct in6_addr *)&neigh->primary_key;
621 			addr_type = ipv6_addr_type(addr6);
622 
623 			if (addr_type == IPV6_ADDR_ANY) {
624 				addr6 = &ipv6_hdr(skb)->daddr;
625 				addr_type = ipv6_addr_type(addr6);
626 			}
627 
628 			if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
629 				do_tx_error_icmp = true;
630 			else {
631 				do_tx_error_icmp = false;
632 				dst = addr6->s6_addr32[3];
633 			}
634 			neigh_release(neigh);
635 			if (do_tx_error_icmp)
636 				goto tx_error_icmp;
637 		}
638 #endif
639 		else
640 			goto tx_error;
641 
642 		connected = false;
643 	}
644 
645 	tos = tnl_params->tos;
646 	if (tos & 0x1) {
647 		tos &= ~0x1;
648 		if (skb->protocol == htons(ETH_P_IP)) {
649 			tos = inner_iph->tos;
650 			connected = false;
651 		} else if (skb->protocol == htons(ETH_P_IPV6)) {
652 			tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
653 			connected = false;
654 		}
655 	}
656 
657 	init_tunnel_flow(&fl4, protocol, dst, tnl_params->saddr,
658 			 tunnel->parms.o_key, RT_TOS(tos), tunnel->parms.link);
659 
660 	if (connected)
661 		rt = (struct rtable *)tunnel_dst_check(tunnel, 0);
662 
663 	if (!rt) {
664 		rt = ip_route_output_key(tunnel->net, &fl4);
665 
666 		if (IS_ERR(rt)) {
667 			dev->stats.tx_carrier_errors++;
668 			goto tx_error;
669 		}
670 		if (connected)
671 			tunnel_dst_set(tunnel, dst_clone(&rt->dst));
672 	}
673 
674 	if (rt->dst.dev == dev) {
675 		ip_rt_put(rt);
676 		dev->stats.collisions++;
677 		goto tx_error;
678 	}
679 
680 	if (tnl_update_pmtu(dev, skb, rt, tnl_params->frag_off)) {
681 		ip_rt_put(rt);
682 		goto tx_error;
683 	}
684 
685 	if (tunnel->err_count > 0) {
686 		if (time_before(jiffies,
687 				tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
688 			tunnel->err_count--;
689 
690 			dst_link_failure(skb);
691 		} else
692 			tunnel->err_count = 0;
693 	}
694 
695 	tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
696 	ttl = tnl_params->ttl;
697 	if (ttl == 0) {
698 		if (skb->protocol == htons(ETH_P_IP))
699 			ttl = inner_iph->ttl;
700 #if IS_ENABLED(CONFIG_IPV6)
701 		else if (skb->protocol == htons(ETH_P_IPV6))
702 			ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
703 #endif
704 		else
705 			ttl = ip4_dst_hoplimit(&rt->dst);
706 	}
707 
708 	df = tnl_params->frag_off;
709 	if (skb->protocol == htons(ETH_P_IP))
710 		df |= (inner_iph->frag_off&htons(IP_DF));
711 
712 	max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr)
713 			+ rt->dst.header_len;
714 	if (max_headroom > dev->needed_headroom)
715 		dev->needed_headroom = max_headroom;
716 
717 	if (skb_cow_head(skb, dev->needed_headroom)) {
718 		dev->stats.tx_dropped++;
719 		dev_kfree_skb(skb);
720 		return;
721 	}
722 
723 	err = iptunnel_xmit(rt, skb, fl4.saddr, fl4.daddr, protocol,
724 			    tos, ttl, df, !net_eq(tunnel->net, dev_net(dev)));
725 	iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
726 
727 	return;
728 
729 #if IS_ENABLED(CONFIG_IPV6)
730 tx_error_icmp:
731 	dst_link_failure(skb);
732 #endif
733 tx_error:
734 	dev->stats.tx_errors++;
735 	dev_kfree_skb(skb);
736 }
737 EXPORT_SYMBOL_GPL(ip_tunnel_xmit);
738 
739 static void ip_tunnel_update(struct ip_tunnel_net *itn,
740 			     struct ip_tunnel *t,
741 			     struct net_device *dev,
742 			     struct ip_tunnel_parm *p,
743 			     bool set_mtu)
744 {
745 	ip_tunnel_del(t);
746 	t->parms.iph.saddr = p->iph.saddr;
747 	t->parms.iph.daddr = p->iph.daddr;
748 	t->parms.i_key = p->i_key;
749 	t->parms.o_key = p->o_key;
750 	if (dev->type != ARPHRD_ETHER) {
751 		memcpy(dev->dev_addr, &p->iph.saddr, 4);
752 		memcpy(dev->broadcast, &p->iph.daddr, 4);
753 	}
754 	ip_tunnel_add(itn, t);
755 
756 	t->parms.iph.ttl = p->iph.ttl;
757 	t->parms.iph.tos = p->iph.tos;
758 	t->parms.iph.frag_off = p->iph.frag_off;
759 
760 	if (t->parms.link != p->link) {
761 		int mtu;
762 
763 		t->parms.link = p->link;
764 		mtu = ip_tunnel_bind_dev(dev);
765 		if (set_mtu)
766 			dev->mtu = mtu;
767 	}
768 	tunnel_dst_reset_all(t);
769 	netdev_state_change(dev);
770 }
771 
772 int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd)
773 {
774 	int err = 0;
775 	struct ip_tunnel *t;
776 	struct net *net = dev_net(dev);
777 	struct ip_tunnel *tunnel = netdev_priv(dev);
778 	struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id);
779 
780 	BUG_ON(!itn->fb_tunnel_dev);
781 	switch (cmd) {
782 	case SIOCGETTUNNEL:
783 		t = NULL;
784 		if (dev == itn->fb_tunnel_dev)
785 			t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
786 		if (t == NULL)
787 			t = netdev_priv(dev);
788 		memcpy(p, &t->parms, sizeof(*p));
789 		break;
790 
791 	case SIOCADDTUNNEL:
792 	case SIOCCHGTUNNEL:
793 		err = -EPERM;
794 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
795 			goto done;
796 		if (p->iph.ttl)
797 			p->iph.frag_off |= htons(IP_DF);
798 		if (!(p->i_flags&TUNNEL_KEY))
799 			p->i_key = 0;
800 		if (!(p->o_flags&TUNNEL_KEY))
801 			p->o_key = 0;
802 
803 		t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
804 
805 		if (!t && (cmd == SIOCADDTUNNEL))
806 			t = ip_tunnel_create(net, itn, p);
807 
808 		if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
809 			if (t != NULL) {
810 				if (t->dev != dev) {
811 					err = -EEXIST;
812 					break;
813 				}
814 			} else {
815 				unsigned int nflags = 0;
816 
817 				if (ipv4_is_multicast(p->iph.daddr))
818 					nflags = IFF_BROADCAST;
819 				else if (p->iph.daddr)
820 					nflags = IFF_POINTOPOINT;
821 
822 				if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
823 					err = -EINVAL;
824 					break;
825 				}
826 
827 				t = netdev_priv(dev);
828 			}
829 		}
830 
831 		if (t) {
832 			err = 0;
833 			ip_tunnel_update(itn, t, dev, p, true);
834 		} else
835 			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
836 		break;
837 
838 	case SIOCDELTUNNEL:
839 		err = -EPERM;
840 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
841 			goto done;
842 
843 		if (dev == itn->fb_tunnel_dev) {
844 			err = -ENOENT;
845 			t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
846 			if (t == NULL)
847 				goto done;
848 			err = -EPERM;
849 			if (t == netdev_priv(itn->fb_tunnel_dev))
850 				goto done;
851 			dev = t->dev;
852 		}
853 		unregister_netdevice(dev);
854 		err = 0;
855 		break;
856 
857 	default:
858 		err = -EINVAL;
859 	}
860 
861 done:
862 	return err;
863 }
864 EXPORT_SYMBOL_GPL(ip_tunnel_ioctl);
865 
866 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
867 {
868 	struct ip_tunnel *tunnel = netdev_priv(dev);
869 	int t_hlen = tunnel->hlen + sizeof(struct iphdr);
870 
871 	if (new_mtu < 68 ||
872 	    new_mtu > 0xFFF8 - dev->hard_header_len - t_hlen)
873 		return -EINVAL;
874 	dev->mtu = new_mtu;
875 	return 0;
876 }
877 EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu);
878 
879 static void ip_tunnel_dev_free(struct net_device *dev)
880 {
881 	struct ip_tunnel *tunnel = netdev_priv(dev);
882 
883 	gro_cells_destroy(&tunnel->gro_cells);
884 	free_percpu(tunnel->dst_cache);
885 	free_percpu(dev->tstats);
886 	free_netdev(dev);
887 }
888 
889 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head)
890 {
891 	struct ip_tunnel *tunnel = netdev_priv(dev);
892 	struct ip_tunnel_net *itn;
893 
894 	itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id);
895 
896 	if (itn->fb_tunnel_dev != dev) {
897 		ip_tunnel_del(netdev_priv(dev));
898 		unregister_netdevice_queue(dev, head);
899 	}
900 }
901 EXPORT_SYMBOL_GPL(ip_tunnel_dellink);
902 
903 int ip_tunnel_init_net(struct net *net, int ip_tnl_net_id,
904 				  struct rtnl_link_ops *ops, char *devname)
905 {
906 	struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id);
907 	struct ip_tunnel_parm parms;
908 	unsigned int i;
909 
910 	for (i = 0; i < IP_TNL_HASH_SIZE; i++)
911 		INIT_HLIST_HEAD(&itn->tunnels[i]);
912 
913 	if (!ops) {
914 		itn->fb_tunnel_dev = NULL;
915 		return 0;
916 	}
917 
918 	memset(&parms, 0, sizeof(parms));
919 	if (devname)
920 		strlcpy(parms.name, devname, IFNAMSIZ);
921 
922 	rtnl_lock();
923 	itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms);
924 	/* FB netdevice is special: we have one, and only one per netns.
925 	 * Allowing to move it to another netns is clearly unsafe.
926 	 */
927 	if (!IS_ERR(itn->fb_tunnel_dev)) {
928 		itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
929 		ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev));
930 	}
931 	rtnl_unlock();
932 
933 	return PTR_RET(itn->fb_tunnel_dev);
934 }
935 EXPORT_SYMBOL_GPL(ip_tunnel_init_net);
936 
937 static void ip_tunnel_destroy(struct ip_tunnel_net *itn, struct list_head *head,
938 			      struct rtnl_link_ops *ops)
939 {
940 	struct net *net = dev_net(itn->fb_tunnel_dev);
941 	struct net_device *dev, *aux;
942 	int h;
943 
944 	for_each_netdev_safe(net, dev, aux)
945 		if (dev->rtnl_link_ops == ops)
946 			unregister_netdevice_queue(dev, head);
947 
948 	for (h = 0; h < IP_TNL_HASH_SIZE; h++) {
949 		struct ip_tunnel *t;
950 		struct hlist_node *n;
951 		struct hlist_head *thead = &itn->tunnels[h];
952 
953 		hlist_for_each_entry_safe(t, n, thead, hash_node)
954 			/* If dev is in the same netns, it has already
955 			 * been added to the list by the previous loop.
956 			 */
957 			if (!net_eq(dev_net(t->dev), net))
958 				unregister_netdevice_queue(t->dev, head);
959 	}
960 }
961 
962 void ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops)
963 {
964 	LIST_HEAD(list);
965 
966 	rtnl_lock();
967 	ip_tunnel_destroy(itn, &list, ops);
968 	unregister_netdevice_many(&list);
969 	rtnl_unlock();
970 }
971 EXPORT_SYMBOL_GPL(ip_tunnel_delete_net);
972 
973 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
974 		      struct ip_tunnel_parm *p)
975 {
976 	struct ip_tunnel *nt;
977 	struct net *net = dev_net(dev);
978 	struct ip_tunnel_net *itn;
979 	int mtu;
980 	int err;
981 
982 	nt = netdev_priv(dev);
983 	itn = net_generic(net, nt->ip_tnl_net_id);
984 
985 	if (ip_tunnel_find(itn, p, dev->type))
986 		return -EEXIST;
987 
988 	nt->net = net;
989 	nt->parms = *p;
990 	err = register_netdevice(dev);
991 	if (err)
992 		goto out;
993 
994 	if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
995 		eth_hw_addr_random(dev);
996 
997 	mtu = ip_tunnel_bind_dev(dev);
998 	if (!tb[IFLA_MTU])
999 		dev->mtu = mtu;
1000 
1001 	ip_tunnel_add(itn, nt);
1002 
1003 out:
1004 	return err;
1005 }
1006 EXPORT_SYMBOL_GPL(ip_tunnel_newlink);
1007 
1008 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[],
1009 			 struct ip_tunnel_parm *p)
1010 {
1011 	struct ip_tunnel *t;
1012 	struct ip_tunnel *tunnel = netdev_priv(dev);
1013 	struct net *net = tunnel->net;
1014 	struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id);
1015 
1016 	if (dev == itn->fb_tunnel_dev)
1017 		return -EINVAL;
1018 
1019 	t = ip_tunnel_find(itn, p, dev->type);
1020 
1021 	if (t) {
1022 		if (t->dev != dev)
1023 			return -EEXIST;
1024 	} else {
1025 		t = tunnel;
1026 
1027 		if (dev->type != ARPHRD_ETHER) {
1028 			unsigned int nflags = 0;
1029 
1030 			if (ipv4_is_multicast(p->iph.daddr))
1031 				nflags = IFF_BROADCAST;
1032 			else if (p->iph.daddr)
1033 				nflags = IFF_POINTOPOINT;
1034 
1035 			if ((dev->flags ^ nflags) &
1036 			    (IFF_POINTOPOINT | IFF_BROADCAST))
1037 				return -EINVAL;
1038 		}
1039 	}
1040 
1041 	ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU]);
1042 	return 0;
1043 }
1044 EXPORT_SYMBOL_GPL(ip_tunnel_changelink);
1045 
1046 int ip_tunnel_init(struct net_device *dev)
1047 {
1048 	struct ip_tunnel *tunnel = netdev_priv(dev);
1049 	struct iphdr *iph = &tunnel->parms.iph;
1050 	int i, err;
1051 
1052 	dev->destructor	= ip_tunnel_dev_free;
1053 	dev->tstats = alloc_percpu(struct pcpu_sw_netstats);
1054 	if (!dev->tstats)
1055 		return -ENOMEM;
1056 
1057 	for_each_possible_cpu(i) {
1058 		struct pcpu_sw_netstats *ipt_stats;
1059 		ipt_stats = per_cpu_ptr(dev->tstats, i);
1060 		u64_stats_init(&ipt_stats->syncp);
1061 	}
1062 
1063 	tunnel->dst_cache = alloc_percpu(struct ip_tunnel_dst);
1064 	if (!tunnel->dst_cache) {
1065 		free_percpu(dev->tstats);
1066 		return -ENOMEM;
1067 	}
1068 
1069 	for_each_possible_cpu(i) {
1070 		struct ip_tunnel_dst *idst = per_cpu_ptr(tunnel->dst_cache, i);
1071 		idst-> dst = NULL;
1072 		spin_lock_init(&idst->lock);
1073 	}
1074 
1075 	err = gro_cells_init(&tunnel->gro_cells, dev);
1076 	if (err) {
1077 		free_percpu(tunnel->dst_cache);
1078 		free_percpu(dev->tstats);
1079 		return err;
1080 	}
1081 
1082 	tunnel->dev = dev;
1083 	tunnel->net = dev_net(dev);
1084 	strcpy(tunnel->parms.name, dev->name);
1085 	iph->version		= 4;
1086 	iph->ihl		= 5;
1087 
1088 	return 0;
1089 }
1090 EXPORT_SYMBOL_GPL(ip_tunnel_init);
1091 
1092 void ip_tunnel_uninit(struct net_device *dev)
1093 {
1094 	struct ip_tunnel *tunnel = netdev_priv(dev);
1095 	struct net *net = tunnel->net;
1096 	struct ip_tunnel_net *itn;
1097 
1098 	itn = net_generic(net, tunnel->ip_tnl_net_id);
1099 	/* fb_tunnel_dev will be unregisted in net-exit call. */
1100 	if (itn->fb_tunnel_dev != dev)
1101 		ip_tunnel_del(netdev_priv(dev));
1102 
1103 	tunnel_dst_reset_all(tunnel);
1104 }
1105 EXPORT_SYMBOL_GPL(ip_tunnel_uninit);
1106 
1107 /* Do least required initialization, rest of init is done in tunnel_init call */
1108 void ip_tunnel_setup(struct net_device *dev, int net_id)
1109 {
1110 	struct ip_tunnel *tunnel = netdev_priv(dev);
1111 	tunnel->ip_tnl_net_id = net_id;
1112 }
1113 EXPORT_SYMBOL_GPL(ip_tunnel_setup);
1114 
1115 MODULE_LICENSE("GPL");
1116