xref: /openbmc/linux/net/ipv4/ip_gre.c (revision dd5b2498)
1 /*
2  *	Linux NET3:	GRE over IP protocol decoder.
3  *
4  *	Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.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/if_vlan.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 
36 #include <net/sock.h>
37 #include <net/ip.h>
38 #include <net/icmp.h>
39 #include <net/protocol.h>
40 #include <net/ip_tunnels.h>
41 #include <net/arp.h>
42 #include <net/checksum.h>
43 #include <net/dsfield.h>
44 #include <net/inet_ecn.h>
45 #include <net/xfrm.h>
46 #include <net/net_namespace.h>
47 #include <net/netns/generic.h>
48 #include <net/rtnetlink.h>
49 #include <net/gre.h>
50 #include <net/dst_metadata.h>
51 #include <net/erspan.h>
52 
53 /*
54    Problems & solutions
55    --------------------
56 
57    1. The most important issue is detecting local dead loops.
58    They would cause complete host lockup in transmit, which
59    would be "resolved" by stack overflow or, if queueing is enabled,
60    with infinite looping in net_bh.
61 
62    We cannot track such dead loops during route installation,
63    it is infeasible task. The most general solutions would be
64    to keep skb->encapsulation counter (sort of local ttl),
65    and silently drop packet when it expires. It is a good
66    solution, but it supposes maintaining new variable in ALL
67    skb, even if no tunneling is used.
68 
69    Current solution: xmit_recursion breaks dead loops. This is a percpu
70    counter, since when we enter the first ndo_xmit(), cpu migration is
71    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
72 
73    2. Networking dead loops would not kill routers, but would really
74    kill network. IP hop limit plays role of "t->recursion" in this case,
75    if we copy it from packet being encapsulated to upper header.
76    It is very good solution, but it introduces two problems:
77 
78    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
79      do not work over tunnels.
80    - traceroute does not work. I planned to relay ICMP from tunnel,
81      so that this problem would be solved and traceroute output
82      would even more informative. This idea appeared to be wrong:
83      only Linux complies to rfc1812 now (yes, guys, Linux is the only
84      true router now :-)), all routers (at least, in neighbourhood of mine)
85      return only 8 bytes of payload. It is the end.
86 
87    Hence, if we want that OSPF worked or traceroute said something reasonable,
88    we should search for another solution.
89 
90    One of them is to parse packet trying to detect inner encapsulation
91    made by our node. It is difficult or even impossible, especially,
92    taking into account fragmentation. TO be short, ttl is not solution at all.
93 
94    Current solution: The solution was UNEXPECTEDLY SIMPLE.
95    We force DF flag on tunnels with preconfigured hop limit,
96    that is ALL. :-) Well, it does not remove the problem completely,
97    but exponential growth of network traffic is changed to linear
98    (branches, that exceed pmtu are pruned) and tunnel mtu
99    rapidly degrades to value <68, where looping stops.
100    Yes, it is not good if there exists a router in the loop,
101    which does not force DF, even when encapsulating packets have DF set.
102    But it is not our problem! Nobody could accuse us, we made
103    all that we could make. Even if it is your gated who injected
104    fatal route to network, even if it were you who configured
105    fatal static route: you are innocent. :-)
106 
107    Alexey Kuznetsov.
108  */
109 
110 static bool log_ecn_error = true;
111 module_param(log_ecn_error, bool, 0644);
112 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
113 
114 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
115 static int ipgre_tunnel_init(struct net_device *dev);
116 static void erspan_build_header(struct sk_buff *skb,
117 				u32 id, u32 index,
118 				bool truncate, bool is_ipv4);
119 
120 static unsigned int ipgre_net_id __read_mostly;
121 static unsigned int gre_tap_net_id __read_mostly;
122 static unsigned int erspan_net_id __read_mostly;
123 
124 static int ipgre_err(struct sk_buff *skb, u32 info,
125 		     const struct tnl_ptk_info *tpi)
126 {
127 
128 	/* All the routers (except for Linux) return only
129 	   8 bytes of packet payload. It means, that precise relaying of
130 	   ICMP in the real Internet is absolutely infeasible.
131 
132 	   Moreover, Cisco "wise men" put GRE key to the third word
133 	   in GRE header. It makes impossible maintaining even soft
134 	   state for keyed GRE tunnels with enabled checksum. Tell
135 	   them "thank you".
136 
137 	   Well, I wonder, rfc1812 was written by Cisco employee,
138 	   what the hell these idiots break standards established
139 	   by themselves???
140 	   */
141 	struct net *net = dev_net(skb->dev);
142 	struct ip_tunnel_net *itn;
143 	const struct iphdr *iph;
144 	const int type = icmp_hdr(skb)->type;
145 	const int code = icmp_hdr(skb)->code;
146 	unsigned int data_len = 0;
147 	struct ip_tunnel *t;
148 
149 	if (tpi->proto == htons(ETH_P_TEB))
150 		itn = net_generic(net, gre_tap_net_id);
151 	else if (tpi->proto == htons(ETH_P_ERSPAN) ||
152 		 tpi->proto == htons(ETH_P_ERSPAN2))
153 		itn = net_generic(net, erspan_net_id);
154 	else
155 		itn = net_generic(net, ipgre_net_id);
156 
157 	iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
158 	t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
159 			     iph->daddr, iph->saddr, tpi->key);
160 
161 	if (!t)
162 		return -ENOENT;
163 
164 	switch (type) {
165 	default:
166 	case ICMP_PARAMETERPROB:
167 		return 0;
168 
169 	case ICMP_DEST_UNREACH:
170 		switch (code) {
171 		case ICMP_SR_FAILED:
172 		case ICMP_PORT_UNREACH:
173 			/* Impossible event. */
174 			return 0;
175 		default:
176 			/* All others are translated to HOST_UNREACH.
177 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
178 			   I believe they are just ether pollution. --ANK
179 			 */
180 			break;
181 		}
182 		break;
183 
184 	case ICMP_TIME_EXCEEDED:
185 		if (code != ICMP_EXC_TTL)
186 			return 0;
187 		data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
188 		break;
189 
190 	case ICMP_REDIRECT:
191 		break;
192 	}
193 
194 #if IS_ENABLED(CONFIG_IPV6)
195        if (tpi->proto == htons(ETH_P_IPV6) &&
196            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
197 				       type, data_len))
198                return 0;
199 #endif
200 
201 	if (t->parms.iph.daddr == 0 ||
202 	    ipv4_is_multicast(t->parms.iph.daddr))
203 		return 0;
204 
205 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
206 		return 0;
207 
208 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
209 		t->err_count++;
210 	else
211 		t->err_count = 1;
212 	t->err_time = jiffies;
213 
214 	return 0;
215 }
216 
217 static void gre_err(struct sk_buff *skb, u32 info)
218 {
219 	/* All the routers (except for Linux) return only
220 	 * 8 bytes of packet payload. It means, that precise relaying of
221 	 * ICMP in the real Internet is absolutely infeasible.
222 	 *
223 	 * Moreover, Cisco "wise men" put GRE key to the third word
224 	 * in GRE header. It makes impossible maintaining even soft
225 	 * state for keyed
226 	 * GRE tunnels with enabled checksum. Tell them "thank you".
227 	 *
228 	 * Well, I wonder, rfc1812 was written by Cisco employee,
229 	 * what the hell these idiots break standards established
230 	 * by themselves???
231 	 */
232 
233 	const struct iphdr *iph = (struct iphdr *)skb->data;
234 	const int type = icmp_hdr(skb)->type;
235 	const int code = icmp_hdr(skb)->code;
236 	struct tnl_ptk_info tpi;
237 
238 	if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
239 			     iph->ihl * 4) < 0)
240 		return;
241 
242 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
243 		ipv4_update_pmtu(skb, dev_net(skb->dev), info,
244 				 skb->dev->ifindex, IPPROTO_GRE);
245 		return;
246 	}
247 	if (type == ICMP_REDIRECT) {
248 		ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
249 			      IPPROTO_GRE);
250 		return;
251 	}
252 
253 	ipgre_err(skb, info, &tpi);
254 }
255 
256 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
257 		      int gre_hdr_len)
258 {
259 	struct net *net = dev_net(skb->dev);
260 	struct metadata_dst *tun_dst = NULL;
261 	struct erspan_base_hdr *ershdr;
262 	struct ip_tunnel_net *itn;
263 	struct ip_tunnel *tunnel;
264 	const struct iphdr *iph;
265 	struct erspan_md2 *md2;
266 	int ver;
267 	int len;
268 
269 	itn = net_generic(net, erspan_net_id);
270 
271 	iph = ip_hdr(skb);
272 	ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
273 	ver = ershdr->ver;
274 
275 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
276 				  tpi->flags | TUNNEL_KEY,
277 				  iph->saddr, iph->daddr, tpi->key);
278 
279 	if (tunnel) {
280 		len = gre_hdr_len + erspan_hdr_len(ver);
281 		if (unlikely(!pskb_may_pull(skb, len)))
282 			return PACKET_REJECT;
283 
284 		if (__iptunnel_pull_header(skb,
285 					   len,
286 					   htons(ETH_P_TEB),
287 					   false, false) < 0)
288 			goto drop;
289 
290 		if (tunnel->collect_md) {
291 			struct erspan_metadata *pkt_md, *md;
292 			struct ip_tunnel_info *info;
293 			unsigned char *gh;
294 			__be64 tun_id;
295 			__be16 flags;
296 
297 			tpi->flags |= TUNNEL_KEY;
298 			flags = tpi->flags;
299 			tun_id = key32_to_tunnel_id(tpi->key);
300 
301 			tun_dst = ip_tun_rx_dst(skb, flags,
302 						tun_id, sizeof(*md));
303 			if (!tun_dst)
304 				return PACKET_REJECT;
305 
306 			/* skb can be uncloned in __iptunnel_pull_header, so
307 			 * old pkt_md is no longer valid and we need to reset
308 			 * it
309 			 */
310 			gh = skb_network_header(skb) +
311 			     skb_network_header_len(skb);
312 			pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
313 							    sizeof(*ershdr));
314 			md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
315 			md->version = ver;
316 			md2 = &md->u.md2;
317 			memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
318 						       ERSPAN_V2_MDSIZE);
319 
320 			info = &tun_dst->u.tun_info;
321 			info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
322 			info->options_len = sizeof(*md);
323 		}
324 
325 		skb_reset_mac_header(skb);
326 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
327 		return PACKET_RCVD;
328 	}
329 	return PACKET_REJECT;
330 
331 drop:
332 	kfree_skb(skb);
333 	return PACKET_RCVD;
334 }
335 
336 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
337 		       struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
338 {
339 	struct metadata_dst *tun_dst = NULL;
340 	const struct iphdr *iph;
341 	struct ip_tunnel *tunnel;
342 
343 	iph = ip_hdr(skb);
344 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
345 				  iph->saddr, iph->daddr, tpi->key);
346 
347 	if (tunnel) {
348 		if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
349 					   raw_proto, false) < 0)
350 			goto drop;
351 
352 		if (tunnel->dev->type != ARPHRD_NONE)
353 			skb_pop_mac_header(skb);
354 		else
355 			skb_reset_mac_header(skb);
356 		if (tunnel->collect_md) {
357 			__be16 flags;
358 			__be64 tun_id;
359 
360 			flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
361 			tun_id = key32_to_tunnel_id(tpi->key);
362 			tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
363 			if (!tun_dst)
364 				return PACKET_REJECT;
365 		}
366 
367 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
368 		return PACKET_RCVD;
369 	}
370 	return PACKET_NEXT;
371 
372 drop:
373 	kfree_skb(skb);
374 	return PACKET_RCVD;
375 }
376 
377 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
378 		     int hdr_len)
379 {
380 	struct net *net = dev_net(skb->dev);
381 	struct ip_tunnel_net *itn;
382 	int res;
383 
384 	if (tpi->proto == htons(ETH_P_TEB))
385 		itn = net_generic(net, gre_tap_net_id);
386 	else
387 		itn = net_generic(net, ipgre_net_id);
388 
389 	res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
390 	if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
391 		/* ipgre tunnels in collect metadata mode should receive
392 		 * also ETH_P_TEB traffic.
393 		 */
394 		itn = net_generic(net, ipgre_net_id);
395 		res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
396 	}
397 	return res;
398 }
399 
400 static int gre_rcv(struct sk_buff *skb)
401 {
402 	struct tnl_ptk_info tpi;
403 	bool csum_err = false;
404 	int hdr_len;
405 
406 #ifdef CONFIG_NET_IPGRE_BROADCAST
407 	if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
408 		/* Looped back packet, drop it! */
409 		if (rt_is_output_route(skb_rtable(skb)))
410 			goto drop;
411 	}
412 #endif
413 
414 	hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
415 	if (hdr_len < 0)
416 		goto drop;
417 
418 	if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
419 		     tpi.proto == htons(ETH_P_ERSPAN2))) {
420 		if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
421 			return 0;
422 		goto out;
423 	}
424 
425 	if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
426 		return 0;
427 
428 out:
429 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
430 drop:
431 	kfree_skb(skb);
432 	return 0;
433 }
434 
435 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
436 		       const struct iphdr *tnl_params,
437 		       __be16 proto)
438 {
439 	struct ip_tunnel *tunnel = netdev_priv(dev);
440 
441 	if (tunnel->parms.o_flags & TUNNEL_SEQ)
442 		tunnel->o_seqno++;
443 
444 	/* Push GRE header. */
445 	gre_build_header(skb, tunnel->tun_hlen,
446 			 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
447 			 htonl(tunnel->o_seqno));
448 
449 	ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
450 }
451 
452 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
453 {
454 	return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
455 }
456 
457 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
458 			__be16 proto)
459 {
460 	struct ip_tunnel *tunnel = netdev_priv(dev);
461 	struct ip_tunnel_info *tun_info;
462 	const struct ip_tunnel_key *key;
463 	int tunnel_hlen;
464 	__be16 flags;
465 
466 	tun_info = skb_tunnel_info(skb);
467 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
468 		     ip_tunnel_info_af(tun_info) != AF_INET))
469 		goto err_free_skb;
470 
471 	key = &tun_info->key;
472 	tunnel_hlen = gre_calc_hlen(key->tun_flags);
473 
474 	if (skb_cow_head(skb, dev->needed_headroom))
475 		goto err_free_skb;
476 
477 	/* Push Tunnel header. */
478 	if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
479 		goto err_free_skb;
480 
481 	flags = tun_info->key.tun_flags &
482 		(TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
483 	gre_build_header(skb, tunnel_hlen, flags, proto,
484 			 tunnel_id_to_key32(tun_info->key.tun_id),
485 			 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0);
486 
487 	ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
488 
489 	return;
490 
491 err_free_skb:
492 	kfree_skb(skb);
493 	dev->stats.tx_dropped++;
494 }
495 
496 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
497 {
498 	struct ip_tunnel *tunnel = netdev_priv(dev);
499 	struct ip_tunnel_info *tun_info;
500 	const struct ip_tunnel_key *key;
501 	struct erspan_metadata *md;
502 	bool truncate = false;
503 	__be16 proto;
504 	int tunnel_hlen;
505 	int version;
506 	int nhoff;
507 	int thoff;
508 
509 	tun_info = skb_tunnel_info(skb);
510 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
511 		     ip_tunnel_info_af(tun_info) != AF_INET))
512 		goto err_free_skb;
513 
514 	key = &tun_info->key;
515 	if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT))
516 		goto err_free_skb;
517 	md = ip_tunnel_info_opts(tun_info);
518 	if (!md)
519 		goto err_free_skb;
520 
521 	/* ERSPAN has fixed 8 byte GRE header */
522 	version = md->version;
523 	tunnel_hlen = 8 + erspan_hdr_len(version);
524 
525 	if (skb_cow_head(skb, dev->needed_headroom))
526 		goto err_free_skb;
527 
528 	if (gre_handle_offloads(skb, false))
529 		goto err_free_skb;
530 
531 	if (skb->len > dev->mtu + dev->hard_header_len) {
532 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
533 		truncate = true;
534 	}
535 
536 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
537 	if (skb->protocol == htons(ETH_P_IP) &&
538 	    (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
539 		truncate = true;
540 
541 	thoff = skb_transport_header(skb) - skb_mac_header(skb);
542 	if (skb->protocol == htons(ETH_P_IPV6) &&
543 	    (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff))
544 		truncate = true;
545 
546 	if (version == 1) {
547 		erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
548 				    ntohl(md->u.index), truncate, true);
549 		proto = htons(ETH_P_ERSPAN);
550 	} else if (version == 2) {
551 		erspan_build_header_v2(skb,
552 				       ntohl(tunnel_id_to_key32(key->tun_id)),
553 				       md->u.md2.dir,
554 				       get_hwid(&md->u.md2),
555 				       truncate, true);
556 		proto = htons(ETH_P_ERSPAN2);
557 	} else {
558 		goto err_free_skb;
559 	}
560 
561 	gre_build_header(skb, 8, TUNNEL_SEQ,
562 			 proto, 0, htonl(tunnel->o_seqno++));
563 
564 	ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
565 
566 	return;
567 
568 err_free_skb:
569 	kfree_skb(skb);
570 	dev->stats.tx_dropped++;
571 }
572 
573 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
574 {
575 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
576 	const struct ip_tunnel_key *key;
577 	struct rtable *rt;
578 	struct flowi4 fl4;
579 
580 	if (ip_tunnel_info_af(info) != AF_INET)
581 		return -EINVAL;
582 
583 	key = &info->key;
584 	ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
585 			    tunnel_id_to_key32(key->tun_id), key->tos, 0,
586 			    skb->mark, skb_get_hash(skb));
587 	rt = ip_route_output_key(dev_net(dev), &fl4);
588 	if (IS_ERR(rt))
589 		return PTR_ERR(rt);
590 
591 	ip_rt_put(rt);
592 	info->key.u.ipv4.src = fl4.saddr;
593 	return 0;
594 }
595 
596 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
597 			      struct net_device *dev)
598 {
599 	struct ip_tunnel *tunnel = netdev_priv(dev);
600 	const struct iphdr *tnl_params;
601 
602 	if (!pskb_inet_may_pull(skb))
603 		goto free_skb;
604 
605 	if (tunnel->collect_md) {
606 		gre_fb_xmit(skb, dev, skb->protocol);
607 		return NETDEV_TX_OK;
608 	}
609 
610 	if (dev->header_ops) {
611 		/* Need space for new headers */
612 		if (skb_cow_head(skb, dev->needed_headroom -
613 				      (tunnel->hlen + sizeof(struct iphdr))))
614 			goto free_skb;
615 
616 		tnl_params = (const struct iphdr *)skb->data;
617 
618 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
619 		 * to gre header.
620 		 */
621 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
622 		skb_reset_mac_header(skb);
623 	} else {
624 		if (skb_cow_head(skb, dev->needed_headroom))
625 			goto free_skb;
626 
627 		tnl_params = &tunnel->parms.iph;
628 	}
629 
630 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
631 		goto free_skb;
632 
633 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
634 	return NETDEV_TX_OK;
635 
636 free_skb:
637 	kfree_skb(skb);
638 	dev->stats.tx_dropped++;
639 	return NETDEV_TX_OK;
640 }
641 
642 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
643 			       struct net_device *dev)
644 {
645 	struct ip_tunnel *tunnel = netdev_priv(dev);
646 	bool truncate = false;
647 	__be16 proto;
648 
649 	if (!pskb_inet_may_pull(skb))
650 		goto free_skb;
651 
652 	if (tunnel->collect_md) {
653 		erspan_fb_xmit(skb, dev);
654 		return NETDEV_TX_OK;
655 	}
656 
657 	if (gre_handle_offloads(skb, false))
658 		goto free_skb;
659 
660 	if (skb_cow_head(skb, dev->needed_headroom))
661 		goto free_skb;
662 
663 	if (skb->len > dev->mtu + dev->hard_header_len) {
664 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
665 		truncate = true;
666 	}
667 
668 	/* Push ERSPAN header */
669 	if (tunnel->erspan_ver == 1) {
670 		erspan_build_header(skb, ntohl(tunnel->parms.o_key),
671 				    tunnel->index,
672 				    truncate, true);
673 		proto = htons(ETH_P_ERSPAN);
674 	} else if (tunnel->erspan_ver == 2) {
675 		erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
676 				       tunnel->dir, tunnel->hwid,
677 				       truncate, true);
678 		proto = htons(ETH_P_ERSPAN2);
679 	} else {
680 		goto free_skb;
681 	}
682 
683 	tunnel->parms.o_flags &= ~TUNNEL_KEY;
684 	__gre_xmit(skb, dev, &tunnel->parms.iph, proto);
685 	return NETDEV_TX_OK;
686 
687 free_skb:
688 	kfree_skb(skb);
689 	dev->stats.tx_dropped++;
690 	return NETDEV_TX_OK;
691 }
692 
693 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
694 				struct net_device *dev)
695 {
696 	struct ip_tunnel *tunnel = netdev_priv(dev);
697 
698 	if (!pskb_inet_may_pull(skb))
699 		goto free_skb;
700 
701 	if (tunnel->collect_md) {
702 		gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
703 		return NETDEV_TX_OK;
704 	}
705 
706 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
707 		goto free_skb;
708 
709 	if (skb_cow_head(skb, dev->needed_headroom))
710 		goto free_skb;
711 
712 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
713 	return NETDEV_TX_OK;
714 
715 free_skb:
716 	kfree_skb(skb);
717 	dev->stats.tx_dropped++;
718 	return NETDEV_TX_OK;
719 }
720 
721 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
722 {
723 	struct ip_tunnel *tunnel = netdev_priv(dev);
724 	int len;
725 
726 	len = tunnel->tun_hlen;
727 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
728 	len = tunnel->tun_hlen - len;
729 	tunnel->hlen = tunnel->hlen + len;
730 
731 	dev->needed_headroom = dev->needed_headroom + len;
732 	if (set_mtu)
733 		dev->mtu = max_t(int, dev->mtu - len, 68);
734 
735 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
736 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
737 		    tunnel->encap.type == TUNNEL_ENCAP_NONE) {
738 			dev->features |= NETIF_F_GSO_SOFTWARE;
739 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
740 		} else {
741 			dev->features &= ~NETIF_F_GSO_SOFTWARE;
742 			dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
743 		}
744 		dev->features |= NETIF_F_LLTX;
745 	} else {
746 		dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
747 		dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
748 	}
749 }
750 
751 static int ipgre_tunnel_ioctl(struct net_device *dev,
752 			      struct ifreq *ifr, int cmd)
753 {
754 	struct ip_tunnel_parm p;
755 	int err;
756 
757 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
758 		return -EFAULT;
759 
760 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
761 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
762 		    p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
763 		    ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
764 			return -EINVAL;
765 	}
766 
767 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
768 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
769 
770 	err = ip_tunnel_ioctl(dev, &p, cmd);
771 	if (err)
772 		return err;
773 
774 	if (cmd == SIOCCHGTUNNEL) {
775 		struct ip_tunnel *t = netdev_priv(dev);
776 
777 		t->parms.i_flags = p.i_flags;
778 		t->parms.o_flags = p.o_flags;
779 
780 		if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
781 			ipgre_link_update(dev, true);
782 	}
783 
784 	p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
785 	p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
786 
787 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
788 		return -EFAULT;
789 
790 	return 0;
791 }
792 
793 /* Nice toy. Unfortunately, useless in real life :-)
794    It allows to construct virtual multiprotocol broadcast "LAN"
795    over the Internet, provided multicast routing is tuned.
796 
797 
798    I have no idea was this bicycle invented before me,
799    so that I had to set ARPHRD_IPGRE to a random value.
800    I have an impression, that Cisco could make something similar,
801    but this feature is apparently missing in IOS<=11.2(8).
802 
803    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
804    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
805 
806    ping -t 255 224.66.66.66
807 
808    If nobody answers, mbone does not work.
809 
810    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
811    ip addr add 10.66.66.<somewhat>/24 dev Universe
812    ifconfig Universe up
813    ifconfig Universe add fe80::<Your_real_addr>/10
814    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
815    ftp 10.66.66.66
816    ...
817    ftp fec0:6666:6666::193.233.7.65
818    ...
819  */
820 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
821 			unsigned short type,
822 			const void *daddr, const void *saddr, unsigned int len)
823 {
824 	struct ip_tunnel *t = netdev_priv(dev);
825 	struct iphdr *iph;
826 	struct gre_base_hdr *greh;
827 
828 	iph = skb_push(skb, t->hlen + sizeof(*iph));
829 	greh = (struct gre_base_hdr *)(iph+1);
830 	greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
831 	greh->protocol = htons(type);
832 
833 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
834 
835 	/* Set the source hardware address. */
836 	if (saddr)
837 		memcpy(&iph->saddr, saddr, 4);
838 	if (daddr)
839 		memcpy(&iph->daddr, daddr, 4);
840 	if (iph->daddr)
841 		return t->hlen + sizeof(*iph);
842 
843 	return -(t->hlen + sizeof(*iph));
844 }
845 
846 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
847 {
848 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
849 	memcpy(haddr, &iph->saddr, 4);
850 	return 4;
851 }
852 
853 static const struct header_ops ipgre_header_ops = {
854 	.create	= ipgre_header,
855 	.parse	= ipgre_header_parse,
856 };
857 
858 #ifdef CONFIG_NET_IPGRE_BROADCAST
859 static int ipgre_open(struct net_device *dev)
860 {
861 	struct ip_tunnel *t = netdev_priv(dev);
862 
863 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
864 		struct flowi4 fl4;
865 		struct rtable *rt;
866 
867 		rt = ip_route_output_gre(t->net, &fl4,
868 					 t->parms.iph.daddr,
869 					 t->parms.iph.saddr,
870 					 t->parms.o_key,
871 					 RT_TOS(t->parms.iph.tos),
872 					 t->parms.link);
873 		if (IS_ERR(rt))
874 			return -EADDRNOTAVAIL;
875 		dev = rt->dst.dev;
876 		ip_rt_put(rt);
877 		if (!__in_dev_get_rtnl(dev))
878 			return -EADDRNOTAVAIL;
879 		t->mlink = dev->ifindex;
880 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
881 	}
882 	return 0;
883 }
884 
885 static int ipgre_close(struct net_device *dev)
886 {
887 	struct ip_tunnel *t = netdev_priv(dev);
888 
889 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
890 		struct in_device *in_dev;
891 		in_dev = inetdev_by_index(t->net, t->mlink);
892 		if (in_dev)
893 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
894 	}
895 	return 0;
896 }
897 #endif
898 
899 static const struct net_device_ops ipgre_netdev_ops = {
900 	.ndo_init		= ipgre_tunnel_init,
901 	.ndo_uninit		= ip_tunnel_uninit,
902 #ifdef CONFIG_NET_IPGRE_BROADCAST
903 	.ndo_open		= ipgre_open,
904 	.ndo_stop		= ipgre_close,
905 #endif
906 	.ndo_start_xmit		= ipgre_xmit,
907 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
908 	.ndo_change_mtu		= ip_tunnel_change_mtu,
909 	.ndo_get_stats64	= ip_tunnel_get_stats64,
910 	.ndo_get_iflink		= ip_tunnel_get_iflink,
911 };
912 
913 #define GRE_FEATURES (NETIF_F_SG |		\
914 		      NETIF_F_FRAGLIST |	\
915 		      NETIF_F_HIGHDMA |		\
916 		      NETIF_F_HW_CSUM)
917 
918 static void ipgre_tunnel_setup(struct net_device *dev)
919 {
920 	dev->netdev_ops		= &ipgre_netdev_ops;
921 	dev->type		= ARPHRD_IPGRE;
922 	ip_tunnel_setup(dev, ipgre_net_id);
923 }
924 
925 static void __gre_tunnel_init(struct net_device *dev)
926 {
927 	struct ip_tunnel *tunnel;
928 
929 	tunnel = netdev_priv(dev);
930 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
931 	tunnel->parms.iph.protocol = IPPROTO_GRE;
932 
933 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
934 
935 	dev->features		|= GRE_FEATURES;
936 	dev->hw_features	|= GRE_FEATURES;
937 
938 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
939 		/* TCP offload with GRE SEQ is not supported, nor
940 		 * can we support 2 levels of outer headers requiring
941 		 * an update.
942 		 */
943 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
944 		    (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
945 			dev->features    |= NETIF_F_GSO_SOFTWARE;
946 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
947 		}
948 
949 		/* Can use a lockless transmit, unless we generate
950 		 * output sequences
951 		 */
952 		dev->features |= NETIF_F_LLTX;
953 	}
954 }
955 
956 static int ipgre_tunnel_init(struct net_device *dev)
957 {
958 	struct ip_tunnel *tunnel = netdev_priv(dev);
959 	struct iphdr *iph = &tunnel->parms.iph;
960 
961 	__gre_tunnel_init(dev);
962 
963 	memcpy(dev->dev_addr, &iph->saddr, 4);
964 	memcpy(dev->broadcast, &iph->daddr, 4);
965 
966 	dev->flags		= IFF_NOARP;
967 	netif_keep_dst(dev);
968 	dev->addr_len		= 4;
969 
970 	if (iph->daddr && !tunnel->collect_md) {
971 #ifdef CONFIG_NET_IPGRE_BROADCAST
972 		if (ipv4_is_multicast(iph->daddr)) {
973 			if (!iph->saddr)
974 				return -EINVAL;
975 			dev->flags = IFF_BROADCAST;
976 			dev->header_ops = &ipgre_header_ops;
977 		}
978 #endif
979 	} else if (!tunnel->collect_md) {
980 		dev->header_ops = &ipgre_header_ops;
981 	}
982 
983 	return ip_tunnel_init(dev);
984 }
985 
986 static const struct gre_protocol ipgre_protocol = {
987 	.handler     = gre_rcv,
988 	.err_handler = gre_err,
989 };
990 
991 static int __net_init ipgre_init_net(struct net *net)
992 {
993 	return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
994 }
995 
996 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
997 {
998 	ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
999 }
1000 
1001 static struct pernet_operations ipgre_net_ops = {
1002 	.init = ipgre_init_net,
1003 	.exit_batch = ipgre_exit_batch_net,
1004 	.id   = &ipgre_net_id,
1005 	.size = sizeof(struct ip_tunnel_net),
1006 };
1007 
1008 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1009 				 struct netlink_ext_ack *extack)
1010 {
1011 	__be16 flags;
1012 
1013 	if (!data)
1014 		return 0;
1015 
1016 	flags = 0;
1017 	if (data[IFLA_GRE_IFLAGS])
1018 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1019 	if (data[IFLA_GRE_OFLAGS])
1020 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1021 	if (flags & (GRE_VERSION|GRE_ROUTING))
1022 		return -EINVAL;
1023 
1024 	if (data[IFLA_GRE_COLLECT_METADATA] &&
1025 	    data[IFLA_GRE_ENCAP_TYPE] &&
1026 	    nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1027 		return -EINVAL;
1028 
1029 	return 0;
1030 }
1031 
1032 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1033 			      struct netlink_ext_ack *extack)
1034 {
1035 	__be32 daddr;
1036 
1037 	if (tb[IFLA_ADDRESS]) {
1038 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1039 			return -EINVAL;
1040 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1041 			return -EADDRNOTAVAIL;
1042 	}
1043 
1044 	if (!data)
1045 		goto out;
1046 
1047 	if (data[IFLA_GRE_REMOTE]) {
1048 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1049 		if (!daddr)
1050 			return -EINVAL;
1051 	}
1052 
1053 out:
1054 	return ipgre_tunnel_validate(tb, data, extack);
1055 }
1056 
1057 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1058 			   struct netlink_ext_ack *extack)
1059 {
1060 	__be16 flags = 0;
1061 	int ret;
1062 
1063 	if (!data)
1064 		return 0;
1065 
1066 	ret = ipgre_tap_validate(tb, data, extack);
1067 	if (ret)
1068 		return ret;
1069 
1070 	/* ERSPAN should only have GRE sequence and key flag */
1071 	if (data[IFLA_GRE_OFLAGS])
1072 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1073 	if (data[IFLA_GRE_IFLAGS])
1074 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1075 	if (!data[IFLA_GRE_COLLECT_METADATA] &&
1076 	    flags != (GRE_SEQ | GRE_KEY))
1077 		return -EINVAL;
1078 
1079 	/* ERSPAN Session ID only has 10-bit. Since we reuse
1080 	 * 32-bit key field as ID, check it's range.
1081 	 */
1082 	if (data[IFLA_GRE_IKEY] &&
1083 	    (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1084 		return -EINVAL;
1085 
1086 	if (data[IFLA_GRE_OKEY] &&
1087 	    (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1088 		return -EINVAL;
1089 
1090 	return 0;
1091 }
1092 
1093 static int ipgre_netlink_parms(struct net_device *dev,
1094 				struct nlattr *data[],
1095 				struct nlattr *tb[],
1096 				struct ip_tunnel_parm *parms,
1097 				__u32 *fwmark)
1098 {
1099 	struct ip_tunnel *t = netdev_priv(dev);
1100 
1101 	memset(parms, 0, sizeof(*parms));
1102 
1103 	parms->iph.protocol = IPPROTO_GRE;
1104 
1105 	if (!data)
1106 		return 0;
1107 
1108 	if (data[IFLA_GRE_LINK])
1109 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1110 
1111 	if (data[IFLA_GRE_IFLAGS])
1112 		parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1113 
1114 	if (data[IFLA_GRE_OFLAGS])
1115 		parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1116 
1117 	if (data[IFLA_GRE_IKEY])
1118 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1119 
1120 	if (data[IFLA_GRE_OKEY])
1121 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1122 
1123 	if (data[IFLA_GRE_LOCAL])
1124 		parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1125 
1126 	if (data[IFLA_GRE_REMOTE])
1127 		parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1128 
1129 	if (data[IFLA_GRE_TTL])
1130 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1131 
1132 	if (data[IFLA_GRE_TOS])
1133 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1134 
1135 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1136 		if (t->ignore_df)
1137 			return -EINVAL;
1138 		parms->iph.frag_off = htons(IP_DF);
1139 	}
1140 
1141 	if (data[IFLA_GRE_COLLECT_METADATA]) {
1142 		t->collect_md = true;
1143 		if (dev->type == ARPHRD_IPGRE)
1144 			dev->type = ARPHRD_NONE;
1145 	}
1146 
1147 	if (data[IFLA_GRE_IGNORE_DF]) {
1148 		if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1149 		  && (parms->iph.frag_off & htons(IP_DF)))
1150 			return -EINVAL;
1151 		t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1152 	}
1153 
1154 	if (data[IFLA_GRE_FWMARK])
1155 		*fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1156 
1157 	if (data[IFLA_GRE_ERSPAN_VER]) {
1158 		t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1159 
1160 		if (t->erspan_ver != 1 && t->erspan_ver != 2)
1161 			return -EINVAL;
1162 	}
1163 
1164 	if (t->erspan_ver == 1) {
1165 		if (data[IFLA_GRE_ERSPAN_INDEX]) {
1166 			t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1167 			if (t->index & ~INDEX_MASK)
1168 				return -EINVAL;
1169 		}
1170 	} else if (t->erspan_ver == 2) {
1171 		if (data[IFLA_GRE_ERSPAN_DIR]) {
1172 			t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1173 			if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1174 				return -EINVAL;
1175 		}
1176 		if (data[IFLA_GRE_ERSPAN_HWID]) {
1177 			t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1178 			if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1179 				return -EINVAL;
1180 		}
1181 	}
1182 
1183 	return 0;
1184 }
1185 
1186 /* This function returns true when ENCAP attributes are present in the nl msg */
1187 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1188 				      struct ip_tunnel_encap *ipencap)
1189 {
1190 	bool ret = false;
1191 
1192 	memset(ipencap, 0, sizeof(*ipencap));
1193 
1194 	if (!data)
1195 		return ret;
1196 
1197 	if (data[IFLA_GRE_ENCAP_TYPE]) {
1198 		ret = true;
1199 		ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1200 	}
1201 
1202 	if (data[IFLA_GRE_ENCAP_FLAGS]) {
1203 		ret = true;
1204 		ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1205 	}
1206 
1207 	if (data[IFLA_GRE_ENCAP_SPORT]) {
1208 		ret = true;
1209 		ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1210 	}
1211 
1212 	if (data[IFLA_GRE_ENCAP_DPORT]) {
1213 		ret = true;
1214 		ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1215 	}
1216 
1217 	return ret;
1218 }
1219 
1220 static int gre_tap_init(struct net_device *dev)
1221 {
1222 	__gre_tunnel_init(dev);
1223 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1224 	netif_keep_dst(dev);
1225 
1226 	return ip_tunnel_init(dev);
1227 }
1228 
1229 static const struct net_device_ops gre_tap_netdev_ops = {
1230 	.ndo_init		= gre_tap_init,
1231 	.ndo_uninit		= ip_tunnel_uninit,
1232 	.ndo_start_xmit		= gre_tap_xmit,
1233 	.ndo_set_mac_address 	= eth_mac_addr,
1234 	.ndo_validate_addr	= eth_validate_addr,
1235 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1236 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1237 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1238 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1239 };
1240 
1241 static int erspan_tunnel_init(struct net_device *dev)
1242 {
1243 	struct ip_tunnel *tunnel = netdev_priv(dev);
1244 
1245 	tunnel->tun_hlen = 8;
1246 	tunnel->parms.iph.protocol = IPPROTO_GRE;
1247 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1248 		       erspan_hdr_len(tunnel->erspan_ver);
1249 
1250 	dev->features		|= GRE_FEATURES;
1251 	dev->hw_features	|= GRE_FEATURES;
1252 	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE;
1253 	netif_keep_dst(dev);
1254 
1255 	return ip_tunnel_init(dev);
1256 }
1257 
1258 static const struct net_device_ops erspan_netdev_ops = {
1259 	.ndo_init		= erspan_tunnel_init,
1260 	.ndo_uninit		= ip_tunnel_uninit,
1261 	.ndo_start_xmit		= erspan_xmit,
1262 	.ndo_set_mac_address	= eth_mac_addr,
1263 	.ndo_validate_addr	= eth_validate_addr,
1264 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1265 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1266 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1267 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1268 };
1269 
1270 static void ipgre_tap_setup(struct net_device *dev)
1271 {
1272 	ether_setup(dev);
1273 	dev->max_mtu = 0;
1274 	dev->netdev_ops	= &gre_tap_netdev_ops;
1275 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1276 	dev->priv_flags	|= IFF_LIVE_ADDR_CHANGE;
1277 	ip_tunnel_setup(dev, gre_tap_net_id);
1278 }
1279 
1280 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1281 			 struct nlattr *tb[], struct nlattr *data[],
1282 			 struct netlink_ext_ack *extack)
1283 {
1284 	struct ip_tunnel_parm p;
1285 	struct ip_tunnel_encap ipencap;
1286 	__u32 fwmark = 0;
1287 	int err;
1288 
1289 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1290 		struct ip_tunnel *t = netdev_priv(dev);
1291 		err = ip_tunnel_encap_setup(t, &ipencap);
1292 
1293 		if (err < 0)
1294 			return err;
1295 	}
1296 
1297 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1298 	if (err < 0)
1299 		return err;
1300 	return ip_tunnel_newlink(dev, tb, &p, fwmark);
1301 }
1302 
1303 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1304 			    struct nlattr *data[],
1305 			    struct netlink_ext_ack *extack)
1306 {
1307 	struct ip_tunnel *t = netdev_priv(dev);
1308 	struct ip_tunnel_encap ipencap;
1309 	__u32 fwmark = t->fwmark;
1310 	struct ip_tunnel_parm p;
1311 	int err;
1312 
1313 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1314 		err = ip_tunnel_encap_setup(t, &ipencap);
1315 
1316 		if (err < 0)
1317 			return err;
1318 	}
1319 
1320 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1321 	if (err < 0)
1322 		return err;
1323 
1324 	err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1325 	if (err < 0)
1326 		return err;
1327 
1328 	t->parms.i_flags = p.i_flags;
1329 	t->parms.o_flags = p.o_flags;
1330 
1331 	if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1332 		ipgre_link_update(dev, !tb[IFLA_MTU]);
1333 
1334 	return 0;
1335 }
1336 
1337 static size_t ipgre_get_size(const struct net_device *dev)
1338 {
1339 	return
1340 		/* IFLA_GRE_LINK */
1341 		nla_total_size(4) +
1342 		/* IFLA_GRE_IFLAGS */
1343 		nla_total_size(2) +
1344 		/* IFLA_GRE_OFLAGS */
1345 		nla_total_size(2) +
1346 		/* IFLA_GRE_IKEY */
1347 		nla_total_size(4) +
1348 		/* IFLA_GRE_OKEY */
1349 		nla_total_size(4) +
1350 		/* IFLA_GRE_LOCAL */
1351 		nla_total_size(4) +
1352 		/* IFLA_GRE_REMOTE */
1353 		nla_total_size(4) +
1354 		/* IFLA_GRE_TTL */
1355 		nla_total_size(1) +
1356 		/* IFLA_GRE_TOS */
1357 		nla_total_size(1) +
1358 		/* IFLA_GRE_PMTUDISC */
1359 		nla_total_size(1) +
1360 		/* IFLA_GRE_ENCAP_TYPE */
1361 		nla_total_size(2) +
1362 		/* IFLA_GRE_ENCAP_FLAGS */
1363 		nla_total_size(2) +
1364 		/* IFLA_GRE_ENCAP_SPORT */
1365 		nla_total_size(2) +
1366 		/* IFLA_GRE_ENCAP_DPORT */
1367 		nla_total_size(2) +
1368 		/* IFLA_GRE_COLLECT_METADATA */
1369 		nla_total_size(0) +
1370 		/* IFLA_GRE_IGNORE_DF */
1371 		nla_total_size(1) +
1372 		/* IFLA_GRE_FWMARK */
1373 		nla_total_size(4) +
1374 		/* IFLA_GRE_ERSPAN_INDEX */
1375 		nla_total_size(4) +
1376 		/* IFLA_GRE_ERSPAN_VER */
1377 		nla_total_size(1) +
1378 		/* IFLA_GRE_ERSPAN_DIR */
1379 		nla_total_size(1) +
1380 		/* IFLA_GRE_ERSPAN_HWID */
1381 		nla_total_size(2) +
1382 		0;
1383 }
1384 
1385 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1386 {
1387 	struct ip_tunnel *t = netdev_priv(dev);
1388 	struct ip_tunnel_parm *p = &t->parms;
1389 	__be16 o_flags = p->o_flags;
1390 
1391 	if (t->erspan_ver == 1 || t->erspan_ver == 2) {
1392 		if (!t->collect_md)
1393 			o_flags |= TUNNEL_KEY;
1394 
1395 		if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1396 			goto nla_put_failure;
1397 
1398 		if (t->erspan_ver == 1) {
1399 			if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1400 				goto nla_put_failure;
1401 		} else {
1402 			if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1403 				goto nla_put_failure;
1404 			if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1405 				goto nla_put_failure;
1406 		}
1407 	}
1408 
1409 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1410 	    nla_put_be16(skb, IFLA_GRE_IFLAGS,
1411 			 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1412 	    nla_put_be16(skb, IFLA_GRE_OFLAGS,
1413 			 gre_tnl_flags_to_gre_flags(o_flags)) ||
1414 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1415 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1416 	    nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1417 	    nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1418 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1419 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1420 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1421 		       !!(p->iph.frag_off & htons(IP_DF))) ||
1422 	    nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1423 		goto nla_put_failure;
1424 
1425 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1426 			t->encap.type) ||
1427 	    nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1428 			 t->encap.sport) ||
1429 	    nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1430 			 t->encap.dport) ||
1431 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1432 			t->encap.flags))
1433 		goto nla_put_failure;
1434 
1435 	if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1436 		goto nla_put_failure;
1437 
1438 	if (t->collect_md) {
1439 		if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1440 			goto nla_put_failure;
1441 	}
1442 
1443 	return 0;
1444 
1445 nla_put_failure:
1446 	return -EMSGSIZE;
1447 }
1448 
1449 static void erspan_setup(struct net_device *dev)
1450 {
1451 	struct ip_tunnel *t = netdev_priv(dev);
1452 
1453 	ether_setup(dev);
1454 	dev->netdev_ops = &erspan_netdev_ops;
1455 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1456 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1457 	ip_tunnel_setup(dev, erspan_net_id);
1458 	t->erspan_ver = 1;
1459 }
1460 
1461 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1462 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1463 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1464 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1465 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1466 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1467 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1468 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1469 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1470 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1471 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1472 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
1473 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
1474 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
1475 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
1476 	[IFLA_GRE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1477 	[IFLA_GRE_IGNORE_DF]	= { .type = NLA_U8 },
1478 	[IFLA_GRE_FWMARK]	= { .type = NLA_U32 },
1479 	[IFLA_GRE_ERSPAN_INDEX]	= { .type = NLA_U32 },
1480 	[IFLA_GRE_ERSPAN_VER]	= { .type = NLA_U8 },
1481 	[IFLA_GRE_ERSPAN_DIR]	= { .type = NLA_U8 },
1482 	[IFLA_GRE_ERSPAN_HWID]	= { .type = NLA_U16 },
1483 };
1484 
1485 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1486 	.kind		= "gre",
1487 	.maxtype	= IFLA_GRE_MAX,
1488 	.policy		= ipgre_policy,
1489 	.priv_size	= sizeof(struct ip_tunnel),
1490 	.setup		= ipgre_tunnel_setup,
1491 	.validate	= ipgre_tunnel_validate,
1492 	.newlink	= ipgre_newlink,
1493 	.changelink	= ipgre_changelink,
1494 	.dellink	= ip_tunnel_dellink,
1495 	.get_size	= ipgre_get_size,
1496 	.fill_info	= ipgre_fill_info,
1497 	.get_link_net	= ip_tunnel_get_link_net,
1498 };
1499 
1500 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1501 	.kind		= "gretap",
1502 	.maxtype	= IFLA_GRE_MAX,
1503 	.policy		= ipgre_policy,
1504 	.priv_size	= sizeof(struct ip_tunnel),
1505 	.setup		= ipgre_tap_setup,
1506 	.validate	= ipgre_tap_validate,
1507 	.newlink	= ipgre_newlink,
1508 	.changelink	= ipgre_changelink,
1509 	.dellink	= ip_tunnel_dellink,
1510 	.get_size	= ipgre_get_size,
1511 	.fill_info	= ipgre_fill_info,
1512 	.get_link_net	= ip_tunnel_get_link_net,
1513 };
1514 
1515 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1516 	.kind		= "erspan",
1517 	.maxtype	= IFLA_GRE_MAX,
1518 	.policy		= ipgre_policy,
1519 	.priv_size	= sizeof(struct ip_tunnel),
1520 	.setup		= erspan_setup,
1521 	.validate	= erspan_validate,
1522 	.newlink	= ipgre_newlink,
1523 	.changelink	= ipgre_changelink,
1524 	.dellink	= ip_tunnel_dellink,
1525 	.get_size	= ipgre_get_size,
1526 	.fill_info	= ipgre_fill_info,
1527 	.get_link_net	= ip_tunnel_get_link_net,
1528 };
1529 
1530 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1531 					u8 name_assign_type)
1532 {
1533 	struct nlattr *tb[IFLA_MAX + 1];
1534 	struct net_device *dev;
1535 	LIST_HEAD(list_kill);
1536 	struct ip_tunnel *t;
1537 	int err;
1538 
1539 	memset(&tb, 0, sizeof(tb));
1540 
1541 	dev = rtnl_create_link(net, name, name_assign_type,
1542 			       &ipgre_tap_ops, tb, NULL);
1543 	if (IS_ERR(dev))
1544 		return dev;
1545 
1546 	/* Configure flow based GRE device. */
1547 	t = netdev_priv(dev);
1548 	t->collect_md = true;
1549 
1550 	err = ipgre_newlink(net, dev, tb, NULL, NULL);
1551 	if (err < 0) {
1552 		free_netdev(dev);
1553 		return ERR_PTR(err);
1554 	}
1555 
1556 	/* openvswitch users expect packet sizes to be unrestricted,
1557 	 * so set the largest MTU we can.
1558 	 */
1559 	err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1560 	if (err)
1561 		goto out;
1562 
1563 	err = rtnl_configure_link(dev, NULL);
1564 	if (err < 0)
1565 		goto out;
1566 
1567 	return dev;
1568 out:
1569 	ip_tunnel_dellink(dev, &list_kill);
1570 	unregister_netdevice_many(&list_kill);
1571 	return ERR_PTR(err);
1572 }
1573 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1574 
1575 static int __net_init ipgre_tap_init_net(struct net *net)
1576 {
1577 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1578 }
1579 
1580 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1581 {
1582 	ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1583 }
1584 
1585 static struct pernet_operations ipgre_tap_net_ops = {
1586 	.init = ipgre_tap_init_net,
1587 	.exit_batch = ipgre_tap_exit_batch_net,
1588 	.id   = &gre_tap_net_id,
1589 	.size = sizeof(struct ip_tunnel_net),
1590 };
1591 
1592 static int __net_init erspan_init_net(struct net *net)
1593 {
1594 	return ip_tunnel_init_net(net, erspan_net_id,
1595 				  &erspan_link_ops, "erspan0");
1596 }
1597 
1598 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1599 {
1600 	ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1601 }
1602 
1603 static struct pernet_operations erspan_net_ops = {
1604 	.init = erspan_init_net,
1605 	.exit_batch = erspan_exit_batch_net,
1606 	.id   = &erspan_net_id,
1607 	.size = sizeof(struct ip_tunnel_net),
1608 };
1609 
1610 static int __init ipgre_init(void)
1611 {
1612 	int err;
1613 
1614 	pr_info("GRE over IPv4 tunneling driver\n");
1615 
1616 	err = register_pernet_device(&ipgre_net_ops);
1617 	if (err < 0)
1618 		return err;
1619 
1620 	err = register_pernet_device(&ipgre_tap_net_ops);
1621 	if (err < 0)
1622 		goto pnet_tap_failed;
1623 
1624 	err = register_pernet_device(&erspan_net_ops);
1625 	if (err < 0)
1626 		goto pnet_erspan_failed;
1627 
1628 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1629 	if (err < 0) {
1630 		pr_info("%s: can't add protocol\n", __func__);
1631 		goto add_proto_failed;
1632 	}
1633 
1634 	err = rtnl_link_register(&ipgre_link_ops);
1635 	if (err < 0)
1636 		goto rtnl_link_failed;
1637 
1638 	err = rtnl_link_register(&ipgre_tap_ops);
1639 	if (err < 0)
1640 		goto tap_ops_failed;
1641 
1642 	err = rtnl_link_register(&erspan_link_ops);
1643 	if (err < 0)
1644 		goto erspan_link_failed;
1645 
1646 	return 0;
1647 
1648 erspan_link_failed:
1649 	rtnl_link_unregister(&ipgre_tap_ops);
1650 tap_ops_failed:
1651 	rtnl_link_unregister(&ipgre_link_ops);
1652 rtnl_link_failed:
1653 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1654 add_proto_failed:
1655 	unregister_pernet_device(&erspan_net_ops);
1656 pnet_erspan_failed:
1657 	unregister_pernet_device(&ipgre_tap_net_ops);
1658 pnet_tap_failed:
1659 	unregister_pernet_device(&ipgre_net_ops);
1660 	return err;
1661 }
1662 
1663 static void __exit ipgre_fini(void)
1664 {
1665 	rtnl_link_unregister(&ipgre_tap_ops);
1666 	rtnl_link_unregister(&ipgre_link_ops);
1667 	rtnl_link_unregister(&erspan_link_ops);
1668 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1669 	unregister_pernet_device(&ipgre_tap_net_ops);
1670 	unregister_pernet_device(&ipgre_net_ops);
1671 	unregister_pernet_device(&erspan_net_ops);
1672 }
1673 
1674 module_init(ipgre_init);
1675 module_exit(ipgre_fini);
1676 MODULE_LICENSE("GPL");
1677 MODULE_ALIAS_RTNL_LINK("gre");
1678 MODULE_ALIAS_RTNL_LINK("gretap");
1679 MODULE_ALIAS_RTNL_LINK("erspan");
1680 MODULE_ALIAS_NETDEV("gre0");
1681 MODULE_ALIAS_NETDEV("gretap0");
1682 MODULE_ALIAS_NETDEV("erspan0");
1683